Investigation of Forensic Handwriting Examiners’

Skill in Detecting Unnatural

Handwriting Processes

Submitted by
Carolyne Lindsay Bird
BTech (Forensic & Analytical Chemistry), BSc (Hons)

A thesis submitted in total fulfilment of the requirements for the degree of

Doctor of Philosophy

Department of Human Biosciences

Faculty of Health Sciences
La Trobe University
Victoria 3086
Australia

March 2012

 
 Table of Contents
Summary .......................................................................................................................................iii

Statement of Authorship............................................................................................................iv

Publications ...................................................................................................................................v

Acknowledgements ....................................................................................................................vii

Chapter 1 Introduction ..........................................................................................................1

Chapter 2 Forensic Handwriting Examiners’ Opinions on the

Process of Production of Disguised and Simulated Signatures...................17

Chapter 3 Forensic Handwriting Examiners’ Skill in Distinguishing Between

Natural and Disguised Behaviour from Handwritten Text Pairs................30

Chapter 4 Forensic Handwriting Examiners’ Skill in Detecting

Disguise Behaviour from Individual Handwritten Text Samples ...............40

Chapter 5 Forensic Handwriting Examiners’ Skill in Detecting

Disguise and Simulation Behaviour from Handwritten Text ......................49

Chapter 6 Skill Characteristics of Forensic Handwriting Examiners

Associated with Simulated Handwritten Text................................................66

Chapter 7 Dynamic Features of Naturally Written, Disguised and

Simulated Handwritten Text.............................................................................75

Chapter 8 Disguise Strategies............................................................................................101

Chapter 9 Predictors of Disguised and Simulated Handwritten Text ........................123

Chapter 10 General Discussion ..........................................................................................128

Appendix A Natural and Disguised Handwriting Provider Information.......................135

Appendix B Example Page from Disguise and Simulation Trial Sample Booklet .......136

Appendix C Extract of Disguise and Simulation Trial

Information and Answer Booklet..................................................................137

Appendix D Disguised Handwriting Provider Questionnaire .........................................143

Appendix E Extract of Declared Pairs Trial Answer Booklet ............................................144

Appendix F Extract of Individual Samples Trial Answer Booklet......................................146

References .................................................................................................................................148

ii
 
 Summary

Over the past 15 years, and following academic criticisms, there has been a move to address
the issue of characterising and validating the skill of forensic handwriting examiners (FHEs).
Past studies have compared FHEs’ opinions with those of lay people and found that FHEs
do possess expertise in relation to expressing opinions of authorship of questioned
handwriting and signatures. However, recent research has identified that simulated and
disguised writings, which result from unnatural handwriting processes, are problematic for
FHEs.

Little empirical data currently exists on the skill of FHEs in identifying the process of
production of handwriting or signatures, and how the perception of this process impacts on
authorship opinions expressed. This skill, should it exist, is an important step in assessing
the origin of the writing being examined. It is proposed that FHEs express a first stage
opinion that writings are either naturally or unnaturally written, which will allow the
examiner to proceed to the second stage (authorship) opinion with proper caution, and
should serve to reduce any potential error rate in casework.

The research covered in this thesis has attempted to address this issue by determining the
level of skill and expertise of FHEs in distinguishing between natural and disguised text-
based handwritings, and to compare this with the ability in a group of lay people. FHEs’ skill
in detecting disguise and simulation behaviour from handwritten text samples has also been
investigated, and an attempt made to identify predictors of natural versus unnatural writing
processes that may be incorporated into forensic handwriting methodology. Furthermore,
some dynamic features of natural, disguised and simulated writings have been compared to
determine whether these may potentially be used by FHEs to aid in their assessment of
handwriting behaviour.

iii
 
 Statement of Authorship

Except where reference is made in the text of the thesis, this thesis contains no material
published elsewhere or extracted in whole or in part from a thesis submitted for the award
of any other degree or diploma.
No other person’s work has been used without due acknowledgement in the main text
of the thesis.
This thesis has not been submitted for the award of any degree or diploma in any
other tertiary institution.
Processing of statistical data presented in Chapter 6 was provided by Dr Reinoud
Stoel (Team Leader, Forensic Statistics, Netherlands Forensic Institute), in Chapter 7 by Dr
Kaye Ballantyne (Senior Research and Development Officer, Biology Division, Victoria
Police Forensic Services Department), and in Chapter 9 by Dr Siew-Pang Chan (Senior
Lecturer, Mathematical Sciences, La Trobe University). Kaye Ballantyne also provided
assistance with statistical analyses in Chapter 2.
All research procedures reported in this thesis were approved by the La Trobe
University Faculty of Health Sciences Human Ethics Committee (FEHC 06/142).

8 March 2012 __________________________________

Carolyne Lindsay Bird

iv
 
 Publications
Chapter Corresponding Publications

Chapters in Books

1 Bird, C. (2012). Handwriting. In J. Siegel, G. Knupfer, & P. Saukko (Eds.),

Encyclopedia of Forensic Sciences (2nd ed., in press). Academic Press.

Refereed Journal Articles

2 Bird, C., Found, B., Ballantyne, K. & Rogers, D. (2010). Forensic handwriting
examiners’ opinions on the process of production of disguised and simulated
signatures. Forensic Science International, 195: 103–107.
3 Bird, C., Found, B. & Rogers, D. (2010). Forensic handwriting examiners' skill
in distinguishing between natural and disguised handwriting behaviours.
Journal of Forensic Sciences, 55: 1291–1295.
4 Bird, C., Found, B. & Rogers, D. (accepted, 2011). Forensic handwriting
examiners’ skill in detecting disguise behaviour from handwritten text
samples. Journal of Forensic Document Examination.
6 Bird, C., Stoel, R., Found, B. & Rogers, D. (2011). Skill characteristics of
forensic handwriting examiners associated with simulated handwritten text.
Journal of the American Society of Questioned Document Examiners, 14(2): 29-34.

Conference Proceedings and Presentations

2 Bird, C., Found, B., Ballantyne, K. & Rogers, D. (2007). Forensic handwriting
examiners’ opinions on the process of production of disguised and simulated
signatures. In J.G. Phillips, D. Rogers & R.P. Ogeil (Eds.), Proceedings of the 13th
Conference of the International Graphonomics Society (pp. 171 – 174). Melbourne,
Australia, November 11-14. Clayton, Australia: Multimedia Services, Monash
University.
3 Bird, C., Found, B. & Rogers, D. (2008). Forensic handwriting examiners' skill
in distinguishing between natural and disguised handwriting behaviours.
Presented at the 19th International Symposium on the Forensic Sciences, Melbourne,
Australia, October 6-9.
4 Bird, C., Found, B. & Rogers, D. (2009). Forensic handwriting examiners' skill
in detecting disguise behaviour from handwritten text samples. Presented at the
Conference of the European Network of Forensic Handwriting Experts, Cracow,
Poland, October 8-10.

v
 
 5 Bird, C., Found, B. & Rogers, D. (2010). Forensic handwriting examiners’
skill in detecting disguise and simulation behaviour from handwritten text
samples. Presented at the 20th International Symposium on the Forensic Sciences, Sydney,
Australia, September 5-9.
3, 4, 5 Bird, C., Found, B. & Rogers, D. (2010). Un-natural writing behaviours: a
closer look workshop. Presented at the 20thInternational Symposium on the Forensic
Sciences, Sydney, Australia, September 5-9.
6 Bird, C., Found, B. & Rogers, D. (2010). The impact of practice and use of a
direct model for simulation on forensic handwriting examiners’ ability to
detect simulated text. Proceedings of the American Society of Questioned Document
Examiners 68th Annual Meeting (pp.89-93). Victoria, B.C., Canada, August 28 -
September 2.
7 Bird, C., Found, B., Ballantyne, K. & Rogers, D. (2011). Dynamic features of
(Section 7.2) naturally written, disguised and forged handwritten text. In E. Grassi& J.L.
Contreras-Vidal (Eds.), Proceedings of the 15th Conference of the International
Graphonomics Society (pp.181-184). Cancun, Mexico, June 12 – 15. College Park,
MD: University of Maryland.
7 Bird, C., Found, B., Ballantyne, K. & Rogers, D. (2012). Dynamic features of
(Section 7.3) naturally written, disguised and simulated handwritten text. To be presented at the
NIJ/FBI Impression and Pattern Evidence Symposium, Clearwater Beach, Florida,
USA, 6 – 9 August 2012.
8 Bird, C., Found, B. & Rogers, D. (2012). A comparison of writer-reported
and forensic handwriting experts’ perceptions of disguise strategies in
handwritten text. To be presented at the 21stInternational Symposium on the Forensic
Sciences, Hobart, Australia, 23 – 27 September 2012.

vi
 
 Acknowledgements
I could not, and most likely would not, have undertaken this research without the input,
guidance and support of my supervisors Dr Doug Rogers and Dr Bryan Found. They have
challenged and extended my knowledge since our first meeting, and their belief in me has
fuelled my motivation to get through this. I am honoured to have had the opportunity to
work with such passionate and respected scientists. For all of that, I thank them deeply.
My colleagues in the Document Examination Group at Forensic Science SA: Glyn
Smith, Sharon Birchall and Liz Ockleshaw, have encouraged my research from the start and
enabled me to spend the time I’ve needed on it without complaint or making me feel guilty.
I couldn’t have asked for a lovelier group of people to work with. Thanks also to Forensic
Science SA management who have supported my pursuit of this degree over the past six
years.
I thank the volunteer handwriting providers, forensic handwriting experts and lay
people who each gave up their valuable time to take part in my research, and without whom
this work could not have been undertaken.
Thank you to Tahnee Dewhurst (Victoria Police Forensic Services Department and La
Trobe University), who provided initial instruction in the use of the MovAlyzeR software.
As I progressed with my experiments and data acquisition, invaluable advice was provided
by Hans-Leo Teulings (Neuroscript), who always gave prompt replies to my questions
regarding the workings and processing of data using the software.
Three people have provided me with advice and services in relation to statistical
analyses which go way beyond my skill set, and for which I am extremely grateful. Dr Kaye
Ballantyne (Senior Research and Development Officer, Biology Division, Victoria Police
Forensic Services Department) has undertaken some heavy duty analysis of data included in
Chapters 2 and 7, and has a way of explaining the methods, measures and results in an easy-
to-understand way. Dr Reinoud Stoel (Team Leader, Forensic Statistics, Netherlands
Forensic Institute) took an early and ongoing interest in my research and has provided
analysis on data presented in Chapter 6 as well as feedback on other aspects of this thesis.
Dr Siew-Pang Chan (Senior Lecturer, Mathematical Sciences, La Trobe University)
transformed a data set containing descriptive terms as well as numbers into the results dealt
with in Chapter 9.
Thanks also to my family: Anthony Bird, Bernadette Bird and Rachael Wake, who
have supported me in all my endeavours. I hope I have made you proud.
And finally, my heartfelt thanks to Wes Jeffries, my best friend, confidante, and
tempering agent. I’m sure making it through to the end of this process would have been
much more difficult without you in my life.
vii
 
 Chapter 1
Introduction

Handwriting opinion evidence has been admitted in courts of law for nearly one hundred
and fifty years (Dillon, 1993; Matley, 1999). Whereas many other sciences utilised in the legal
system have had their beginnings in academia and then found their way into judicial
environments, handwriting examination is a field that did not develop in an academic
environment but rather out of judicial need. The potential value of forensic handwriting
comparisons is in its perceived ability to provide evidence as to authorship of a line trace, or
in regard to the writing process itself (e.g. whether a signature is genuine or simulated). In
relation to other ‘identification sciences’ such as DNA profiling, fingerprints or tool mark
comparisons, handwriting evidence stands alone as the examinations are based solely on
movement outcomes. The overarching basis for the belief that handwriting can be used to
identify and eliminate individuals from a writing event is that handwriting features
embedded within the writing trace are believed to be characteristic of an individual. Since
handwriting is a learnt behaviour, utilising the complex machinery of the human movement
system, it is found that the features of writing vary both within and between writing events.
Handwriting can, however, be purposefully distorted, or can be copied by others. The
probative value of this type of evidence is reliant upon the relationship between the features
observed in the movement outcome and the extent to which those features characterise an
individual or behaviour. Despite clothing itself with the “trappings of science” (US v.
Starzecpyzel, 1995), and claiming a scientific approach to the examination of questioned
handwriting, early texts on the subject of forensic document examination rely on
predominantly anecdotal evidence to support the field’s claims to validity (Ellen, 1989;
Hilton, 1982; Osborn, 1929).
In 1993 the Daubert v. Merrell Dow Pharmaceuticals, Inc case set a new standard, in
American courts, for the admissibility of scientific evidence. The Daubert guidelines under
Rule 702 of the Federal Rules of Evidence were intended to ensure that an expert’s
testimony is both reliable and relevant by considering, in relation to the field in question:
1. Whether a method consists of a testable hypothesis;
2. Whether the method has been subject to peer review;
3. The method’s known or potential error rate;
4. The existence and maintenance of standards controlling the technique’s operation;
and
5. Whether the method is generally accepted.

1
Later, Kumho Tire Co., Ltd. v. Carmichael (1999) extended the Daubert ruling from exclusively
scientific evidence to cover all types of expert evidence1.
Although handwriting identification evidence has been accepted in courts for well
over a century, the technique has been found lacking in relation to all of the Daubert
guidelines listed above. This has resulted in a number of published challenges to the field
(Risinger, Denbeaux & Saks, 1989; Risinger, Denbeaux & Saks, 1998; Saks & Koehler, 2005;
Saks & Risinger, 1996; Saks & VanderHaar, 2005). These commentators focussed their
criticism on the lack of empirical testing of the theoretical basis underlying authorship
opinions along with outlining perceived shortfalls in available data which may support or
refute forensic handwriting examiners’ (FHEs) claims to validity. They suggest that there is
some question as to whether the expertise associated with the varying tasks FHEs carry out
existed at all. A key United States of America hearing investigating the validity of forensic
handwriting evidence concluded that forensic document examination “does not rest on
carefully articulated postulates, does not employ rigorous methodology, and has not
convincingly documented the accuracy of its determinations” (US v Starzecpyzel, 1995).
However, the Starzecpyzel court acknowledged that “forensic document examination does
involve true expertise, which may prove helpful to a fact finder”. It deemed the expertise is
not properly characterised as scientific, but as practical in nature. Indeed, there has been a
dearth of scientific research conducted on forensic handwriting examinations specifically,
until relatively recent years (Found & Rogers, 2003; Found, Rogers & Herkt, 2001; Found,
Sita & Rogers, 1999; Galbraith, Galbraith & Galbraith, 1995; Kam ca. 2010; Kam, Fielding
& Conn, 1997; Kam, Gummadidala, Fielding & Conn, 2001; Kam, Wetstein & Conn, 1994;
Sita, Found & Rogers, 2002; Srihari, Cha, Arora & Lee, 2002). More recently, the findings of
the National Research Council report (2009) solidly supported not only the critics’ views
regarding the shortfalls in forensic handwriting science, but that the criticisms are equally
relevant to many of the other pattern evidence sciences.
The work for this thesis was undertaken in light of the need to strengthen the
scientific basis for handwriting comparisons. An aim of this thesis is to provide evidence
that will characterise FHEs’ expertise for determining whether handwritten text is naturally
or unnaturally written. For the purpose of the thesis, natural writing is the writing a person
makes as a result of their learnt writing behaviour without paying particular attention to the
process, and unnatural writing refers to writing that is either disguised or simulated. The
thesis further investigates FHEs’ skill and expertise for determining whether handwritten
text is the product of disguise versus simulation behaviour. In addition, exploration is made

1Zlotnick & Lin (2001) provides an overview of admissibility standards applied to handwriting
evidence in American courts up to 1999, including some cases following the Kumho opinion.
2
of a range of dynamic and static features of handwriting to determine their potential
usefulness as predictors of handwriting behaviours, in particular to discriminate between
disguise and simulation.

1.1 Handwriting as a Complex Psycho-motor Skill

Handwriting is a complex learned behaviour that is the product of cognitive, psychomotor
and biomechanical processes directing the movement of the arm, hand and fingers in order
to manipulate the writing implement in three dimensional space. This process results in a
visual trace on the writing surface. As with any class of learnt complex movement, the
process of skill acquisition is thought to be comprised of three relatively distinct phases
(Fitts, 1964; Fitts & Posner, 1967). During the cognitive phase, the learner writer is concerned
with how to construct letters and symbols. This usually requires a high degree of cognition
as different motor strategies are trialled, unsuccessful strategies abandoned and successful
strategies preserved. As each step in the movement progression is considered and
controlled, the writing produced during this phase usually displays poor fluency. The
associative phase arrives when the writer has determined an effective way of constructing
characters and is focussed on employing subtle adjustments to the movements in order to
improve output. The movement, and therefore the handwriting trace, becomes more
consistent and controlled during this period. The associative phase for handwriting
movement acquisition can extend over years. When the writer reaches a stage where
movement production becomes ‘automatic’, they have reached the autonomous stage. Now,
their handwriting can be carried out seemingly with minimal conscious attention to the
process.
Individuals travel though handwriting movement acquisition in different ways to
ultimately achieve similar goals. Although the goal of the movements is to produce
handwriting that can be read, the movement style is thought to become characteristic to
some extent. The formation of letters is initially dependent on the copybook style learnt by
the writer, however, as the skill develops, variations may be introduced due to differences in
teaching methods, muscular control, sequence of movements employed, aesthetic
preferences, exposure to writings of others and frequency of writing. This results in inter-
writer variation in features (Figure 1.1).

3
Figure 1.1. The word ‘community’ written by twelve different people, illustrating inter-
writer feature variation.

In the progression from the cognitive to the autonomous stage of handwriting

movement acquisition, the abstract motor programs controlling the muscles responsible for
the movement are generated by stringing together smaller programmed units of behaviour,
which are eventually controlled as a single unit (Figure 1.2). These motor programs are
found not to be muscle specific; the movements required to produce a character, word or
signature are, to some extent, differentially transferable between muscle sets not normally
associated with writing tasks (Railbert, 1977; van Galen, 1980). There are numerous spatial
and temporal features that are preserved in the movement patterns (Summers & Anson,
2009; Teulings, 1996; Teulings, Thomassen & van Galen, 1986), resulting in a constancy of
form production within a writer, in spite of influences such as body position, arm and hand
position and writing surface. The time course and velocity profile of handwriting written at
different sizes is relatively constant. When the size of writing is kept constant and the writing
speed is manipulated the gross characteristics of the velocity profile are retained.

Figure 1.2. Phases of handwriting movement acquisition (adapted from Keele & Summers,
1976).

In spite of the kinematic order observed in skilled writing, no person writes so

consistently that every character is exactly the same each time it is written. There is found to

4
be a small range of variation in features such as proportions, size, slope and spacing. This
intra-writer variation arises from a combination of the writer’s motor output varying to
different extents due to “the non-muscle specific nature of the movement’s representation
in the brain” (van Galen, 1980) as well as features associated with particular letter
combinations, position in a word, style, format of the document or the writer’s mood. The
handwriting of an individual may also be modified by external factors such as the writing
environment (paper surface, writing implement, writing surface), or intoxication by drugs or
alcohol, and of course by introducing deliberate changes. Kapoor, Kapoor and Sharma
(1985) report that the extent and form of natural variation may also change over time,
though their study is limited by a small sample size.

1.2 Handwriting as Motor Skill and its Application to Forensic

Individualisation
It is generally believed that the psycho-motor aspects of handwriting production,
responsible for inter and intra-writer variation, allows for it to be a useful form of ‘opinion
identification evidence’ in the forensic sciences. In general, the following beliefs are
embraced within the practical application of handwriting movement science to forensic
practice;
• Given a sufficient amount of handwriting for comparison, the likelihood of two
writers sharing identical handwriting features is very low;
• The natural handwriting of individual writers varies, but to a lower degree than inter-
writer variation;
• Individuals find it difficult to disguise extended bodies of their writing without
leaving evidence of the process;
• Individuals find it difficult to accurately simulate the writing characteristics of others
without leaving evidence of the process.

The beliefs regarding inter-writer variation are supported not simply by observing the
variation encountered in everyday life, but by various empirical studies. Srihari, Cha, Arora
and Lee (2002) used computer algorithms to extract features, based on some of those used
by forensic document examiners, from a database of scanned images of handwriting.
Individuality was established quantitatively on the basis of feature characteristics (with a 98%
confidence), by means of a machine learning approach. A further statistical study of
numerals written by 187 subjects found that no two individuals exhibited the same set of
characteristic codes (which represented their profile of writing habits) for all ten numerals 0
– 9 (Li, Poon, Fung & Yang, 2005). Statistical analysis of the handwriting of twins (Srihari,
5
Huang & Srinivasan, 2008) along with numerous studies undertaken by FHEs (Beacom,
1960; Boot, 1998; Gamble, 1980; Ramsey Lines & Franck, 2003) have shown that even this
handwriting, which may be expected to be the most similar of all pairs of writers, is able to
be discriminated. The frequencies of occurrence of, and variation in, handwriting
characteristics in different populations have been explored over many years (Chapran,
Fairhurst, Guest & Ujam, 2008; Crane, 1999; Eldridge, Nimmo-Smith, Wing & Totty, 1984;
Horton, 1996; King, 2003; Leung, Chung, Tsui, Cheung, Cheung & Mok, 1987;
Muehlberger, Newman, Regent & Wichmann, 1977; Seaman Kelly, 2002; Simner, 1998;
Turnbull, Jones, & Allen, 2010; Wing & Nimmo-Smith, 1987). Muehlberger et al. (1977)
recorded the variation in 200 writers’ forms of the th combination, and claim several
significant feature correlations within writers. Eldridge et al. (1984) developed a classification
system that assigns a Discrimination Index (DI) to letters or features, which may then be
used to discriminate between writers. Of the six cursive letters they examined, the letter k
was found to have the highest DI over all features, offering the most discriminating power.
The investigation of a spatial characteristic of the combination el and letter d by Wing and
Nimmo-Smith (1987) suggests that while these formations display similar degrees of within-
writer consistency, there is considerably more between-writer variation in the d proportion.
Vastrick (1998) has referenced other empirical studies relevant to the body of evidence on
inter and intra-writer variation, both published and unpublished.
Further, there are a number of large-scale databases, for example the Forensic
Information System for Handwriting (FISH) in Germany (Eiserman & Hecker, 1986), which
classify similarities and differences within handwriting. There have yet to be reports of two
people sharing exactly the same combination of handwriting features.

1.3 Forensic Handwriting Examination Theory

Aside from developments in our understanding of the psycho-motor properties of
handwriting, the general approach to practical forensic handwriting examination remains
little changed since the publication of early textbooks on the subject (Conway, 1959, pp. 31-
57; Osborn, 1929, pp. 225, 237-248). These early works assert that handwriting examinations
must be based on the general qualities of the handwriting (class characteristics) in
conjunction with its individual or rare characteristics. Historically, forensic handwriting
examination has been based on the underlying belief that training and experience enables the
handwriting examiner to distinguish between class and individual characteristics. Class
characteristics are those that are derived from the general model or copybook system taught,
and individual characteristics are deviations from the copybook form, introduced into the
writing or developed over time either consciously or unconsciously. Class characteristics may

6
be shared with many other people taught the same system of handwriting, while individual
characteristics are the primary features that a forensic handwriting examiner relies upon to
differentiate the writings of different people (Osborn, 1929, pp. 249-269; Simner, 1998).
Osborn (1929, p. 250) contends that the characteristics with the highest differentiating
power are those that are most divergent from the regular or national features of the
handwriting being examined; and that class features alone are insufficient to base any
identification on. However, while this theory seems logical, examining it in light of everyday
forensic handwriting examinations, a number of problems with the theory have been
articulated (Found & Rogers, 1995). Examples of these are:
1. There is no evidence that FHE experience, once trained, increases the validity of
their findings;
2. When pressed FHEs do not claim that they can reliably identify the class and
individual characteristics of the samples that they examine;
3. FHEs can subscribe to the theory yet express opinions as to the authorship of
foreign writings where the class system is unknown to them.

In addition, Simner (1998) concluded that class characteristics might not actually be
dependent on the copybook pattern taught, so determining the ‘regular’ features of a group
of writers may be difficult to achieve.
More recently, feature detection and complexity theory has been put forward to better
explain the existence of identification expertise (Found & Rogers, 1995, 1998). Feature
detection relates to the comparison process. FHEs compare the features between two
populations of writings (either known to questioned or questioned to questioned) and form
an opinion as to whether the feature characteristics associated with one population is similar
or dissimilar to the comparison population of writings. This forms the ‘first stage’ opinion.
Although this opinion is not related to authorship it does focus the examiner on the relevant
set of alternative propositions which may ultimately result in an authorship opinion being
formed. With this approach there is no attempt to assign the feature characteristics as class
or individual. ‘Complexity theory’ describes the foundation upon which an authorship
opinion can be rationalised, where explanations are proposed in terms of authorship to
account for the observed similarity or dissimilarity in the comparison population of writings
being compared.
Complexity theory, whilst embracing class and class-divergent properties of writings,
proposes an inverse relationship between the complexity and amount of handwriting in the
samples examined, and the ease of successful simulation of that handwriting. In addition, as
the amount of skilled writing increases, the likelihood of a chance match also decreases

7
(Figure 1.3, from Found & Rogers, 1999). This theory explains the common ground
between text based and signature examinations and informs the assessment of whether
sufficient material is available on which to express a valid opinion.
Complexity theory has been empirically examined and, although the exact shape of the
curves remains in question, generally the relationships hold true (Alewijnse, van den Heuvel
& Stoel, 2011; Mattijssen, van den Heuvel & Stoel, 2011).

Figure 1.3. Generalised complexity relationships between the amount of skilled writing
available in a sample, the difficulty with which the sample may be simulated and the
likelihood of a chance match with more than one writer.

Using the above theoretical framework, FHEs form first stage opinions as to feature
similarity and dissimilarities to focus on the appropriate propositions, which may then be
translated into authorship opinions. If the population of features associated with two groups
of writings is similar then three propositions can be considered to explain the observations:
1. One population of writing was written by the writer of the comparison population of
writing;
2. One population of writing is a simulation based on the handwriting features
associated with the writer of the comparison population of writing;
3. One population of writing was not written by the writer of the comparison
population of writing and the similarities have occurred by chance.

If the population of features associated with two groups of writings is dissimilar then
five propositions can be considered to explain the observations:
1. One population of writing was naturally written, but not written by the writer of the
comparison population of writing;
2. One population of writing was not written by the writer of the comparison
population of writing but was simulated;
8
 3. One population of writing was written by the writer of the comparison population of
writing but one or both was disguised;
4. One population of writing was written by the writer of the comparison population of
writing but one or both of the writings did not capture the full range of variation in
the writer;
5. One population of writing was written by the writer of the comparison population
however one or both of the writings was affected by unknown internal or
environmental factors (see McAlexander & Maguire, 1991).

In some instances, FHEs may find it difficult to support any one proposition over one
or more of the others. The relationship between complexity, amount of handwriting and
ease of simulation can be used to aid in the process of forming an opinion. For example, if
the questioned writing is similar to the specimen writing, but the writing is not complex and
restricted to a few words (or elements in a signature), the possibility that the writing has
been simulated or that a chance match has occurred cannot be practically excluded and an
inconclusive or qualified opinion may be expressed. However, if the questioned writing is
similar to the specimen writing and the writing is complex and comprises a paragraph or
more, the possibility that the writing has been simulated or that a chance match has occurred
may be able to be set aside and an opinion that there is very strong support for the
proposition that the writer of one population of writings wrote the comparison population
of writings may be expressed.
Difficulties in handwriting comparisons arise in characterising what constitutes
dissimilarities, particularly given intra-writer variation and limited comparison samples, and
determining what they can be attributed to. If significant differences are found in structural
features, this usually results in an opinion that there is evidence that the two writings are
unlikely to have been written by the same writer.

1.4 Validating Forensic Handwriting Examiners’ Expertise

Given the above overview of the theoretical underpinnings of forensic handwriting
examination practice we should turn our attention to the assessment of the validity of the
approach. The error rate of a method (or individual) can provide a good measure for
determining the probative value of evidence being presented in court. However, to test the
validity of the method(s) used in handwriting identification, a whole array of empirical tests
needs to be carried out in order to capture the range of tasks involved. Proficiency tests
undertaken by practitioners in accredited laboratories do provide some measure of validity;
however, obtaining statistically significant results is problematic since a proficiency test

9
generally examines only one of the many possible propositions that are mounted to explain
similarities and dissimilarities evident between populations of writings. In addition, the level
of difficulty associated with the samples varies, dependent on the specific subtask employed.
Collaborative Testing Services, Inc (2010), an international forensic proficiency test
provider, recently issued a statement on the unsuitability of proficiency test data for error
rate determination due to:
1. The broad range of participants and lack of categorised demographic data;
2. Responses provided as agreeing or disagreeing with consensus, which may or may
not reflect correct or incorrect responses;
3. The tests being designed primarily to meet laboratory accreditation demands not
provide examiners with ‘real world’ case-work-like samples – test design considers
the overall laboratory process and requires the use of artefact samples.

Following the emergence of academic critics of forensic handwriting examination a

number of studies were undertaken to address issues of the accuracy, reliability and validity
of expert opinion in handwriting examination tasks. In 1994 Kam, Wetstein and Conn
published the results of their test of professional document examiners’ proficiency versus
that of a control group of lay people. Their finding that handwriting identification expertise
does exist has been supported by a number of further studies (Found, Sita & Rogers, 1999;
Kam, ca. 2010; Kam, Fielding & Conn, 1997; Sita, Found & Rogers, 2002). These studies
report that handwriting expertise is characterised by what may be described as ‘conservatism’
of the professional group compared to the control group. Using blind questioned to known
sample studies, it has been found that both FHE and lay groups report a similar number of
correct associations of questioned writings with known writings. The difference between the
FHE and lay groups is in the significantly lower error rate for FHEs. This has been
attributed to FHEs expressing more inconclusive opinions, where the control group express
a greater number of erroneous false matches.
The skill of FHEs has also been tested in relation to the different types of
examinations conducted, and the associated error rates calculated (Found & Rogers, 2003;
Found, Rogers & Herkt, 2001; Sita, Found & Rogers, 2002). However, in many instances,
these rates are reported as the combined error rates of all participants, therefore using them
to assess the validity of an individual FHE who did not participate in the testing is
questionable. An added complication is the variability in error rate which has been found to
be different according to the handwriting behaviour being compared (natural, disguised or
simulated writings). Data collected from these validation trials undertaken by a range of
FHEs have shown that there is a low or non-existent error rate associated with questioned

10
samples that are normal writings either by the specimen writer or another writer. Markedly
higher error rates are associated with questioned writings that are simulated or disguised
(Found & Rogers, 2005a, 2008). Along with higher error rates, simulated and disguised
writings are found to attract high percentages of inconclusive opinions (Found, Rogers &
Herkt, 2001). This is thought to result from combinations of similarities and dissimilarities
present in the questioned writings. All empirical evidence to date indicates that disguised and
simulated writings are problematic for FHEs.
Further validation of the work of FHEs may be found in the allied area of automatic
writer identification research. Chapran, Fairhurst, Guest and Ujam (2008) provide a good
synopsis of the computational methods within this field, which offers “technologically
advanced methods to facilitate, enhance and validate the work of a document examiner” (p.
76). In spite of reports of this type, forensic handwriting examination remains qualitative in
nature. Advances in computational methods are, however, likely to result in more objective
comparison techniques being available for FHEs in the future.

1.5 Determining Writing Processes

Large scale blind validation trials have provided evidence that examiners are less likely to
express erroneous opinions on naturally written handwriting samples as opposed to those
that are unnatural (Found & Rogers, 2005a, 2008). The combined results for a group of six
FHEs (Found, Rogers & Herkt, 2001) show an error rate for handwriting samples simulated
by the specimen writer of 2.5% (called error rate of 13.5% when inconclusive opinions are
ignored). The error rate for all other writing types was found to be <1%. Despite the studies
carried out on unnatural handwriting and methods of disguise, prior to the research
described in this thesis, no published empirical research of FHEs’ capacity or reliability to
determine the difference between natural and unnatural handwritten text could be sourced.
Given that disguise and simulation behaviours attract high error scores amongst
FHEs, it is thought that investigations into the identification of the writing process (natural
as compared to simulated or disguised) might enable the source of the authorship error to
be diagnosed. That is, if FHEs can identify a piece of questioned writing as the product of
an unnatural (disguised or simulated) writing process, this may provide a guide as to whether
an authorship opinion might be reliably expressed. The aim of this thesis is to investigate the
skill and expertise associated with FHEs in determining the predictors of unnatural versus
natural handwriting, including the processes by which this discrimination might be
accomplished. Any findings are envisaged to be fed back into a revised method for
approaching text-based handwriting comparisons.

11
1.6 Unnatural Handwriting Behaviours
Unnatural handwriting may be considered that which is disguised, simulated, or modified by
internal or external factors such as the writing environment, illness, medication or
intoxication by drugs or alcohol. However, only disguise and simulation are considered in
this thesis. For the purpose of this thesis, disguise behaviour is an attempt by a writer to
purposefully alter their writing in order to avoid being identified or for them to deny having
created the writing. Simulation behaviour is an attempt by the writer to imitate the writing
characteristics of another person’s writing (auto-simulation is not considered in this thesis).
The features of disguised and simulated writings can be quite different from natural writing
due to differences in their psycho-motor properties.
Disguise in writing has been reported in numerous studies (Alford, 1970; Harrison,
1958, pp. 349-372; Herkt, 1986; Hull, 1993; Keckler, 1997; Osborn, 1929, pp.407-408;
including one on Swedish handwriting: Konstantinidis, 1987; and two on Chinese writing:
Leung, Chung, Tsui & Cheung, 1988; Tsui, 1997). However, of the studies only Alford
(1970), Konstantinidis (1987), Hull (1993) and Keckler (1997) deal with somewhat extended
handwritten entries. There is general consensus that disguise incorporates changes to that
which is obvious and superficial, while the less conspicuous features remain unchanged
(Huber & Headrick, 1999, p.279). Commonly reported beliefs regarding disguise include
(Harrison, 1958, pp.349-372; Cain, 1998; Morris, 2000, pp.171-173):
• Disguised handwriting exhibits less fluency and poorer rhythm than natural writing;
• Most disguises are relatively simple;
• The internal consistency of disguised writing is poor; any disguise is rarely
maintained over more than a few lines;
• Certain features, such as word and line spacing, are rarely disguised.

When attempting disguise, a writer has to suppress their normal writing behaviour and
introduce new features that they consider significantly different. However, some writers may
not realise what characteristics are significant (Webb, 1977), or find the task of disguise so
difficult, that their attempts are unsuccessful and an examination may lead the examiner to
express an opinion that the writing is genuine.
Publications dealing with the simulation of extended writings are few, and are
predominately related to high profile cases such as the Howard Hughes autobiography
(Cabanne, 1974), and the diaries of Adolph Hitler (Michel & Baier, 1985) and Jack the
Ripper (Nickell, 1997). Freehand simulations (the only type of simulations to be considered
in this research) are ‘drawn’ either based on a physical model or a mental image. For this
activity, the writer is required to imitate another person’s complex mechanisms for
12
controlling muscle contractions whilst suppressing the characteristics of their own system.
The same areas of the brain will be used as for natural handwriting, but differently
configured, with more input from sensory areas responsible for the visual input of the
model and the monitoring of the progress and success of the simulated writing. This results
in a much higher cognitive load for the simulator which is thought to result in disturbance to
the movement outcome. The movements associated with simulation behaviour will generally
be slower and less fluent than natural handwriting movements. This is a result of the higher
cognitive load, the inherent nature of the closed-loop mode of movement control which is
dependent on feedback, as well as the writer stopping frequently to check letter construction
and progress against the model. This writing behaviour is likely to be reflected in the static
handwriting trace in terms of features we can readily observe: a lack of fluency
(incorporating such elements as tremor, awkward or interrupted movements), unusual pen
lifts, pauses, retouches and internal inconsistencies in pen direction, letter construction or
connectivity. Indeed it is these features that are generally accepted to be indicative of
simulation behaviour. Generally, simulations may capture one or the other of the pictorial or
line quality features of the genuine writing, but very rarely both. Analogous to disguised
writings, and expectedly, the writer focuses on the most obvious characteristics of the model
writing such as slant and letterforms, while the more subtle features, for example
proportions and spacing, may be overlooked (Harrison, 1958, p. 365; Leung, Cheng, Fung &
Poon, 1993; Muehlberger, 1990).
It is difficult for writers to maintain either type of unnatural writing behaviour, so with
extended writing some of the writers’ normal characteristics are likely to appear. This results
in the presence of both similarities and dissimilarities with the comparison writing. Recent
research provides limited empirical data for the difficulty that FHEs have in distinguishing
between disguised and simulated writing behaviours (Found & Rogers, 2005a, 2008). This is
investigated further in Chapters 2 (relating to disguised and simulated signatures) and 5 (in
relation to disguised and simulated text-based writing).
In addition to the mixed signal of similarities and differences arising from the
unnatural writing process, writer skill can complicate the task of determining writing process
as some people’s natural writing may exhibit poor line quality and/or the wide degree of
variation sometimes observed in unnatural writings. The ability of FHEs to determine
disguised and natural writing samples is explored in Chapters 3 and 4. The trials from which
these data come were created from the same handwriting samples, but in Chapter 3 the
participants examined pairs of writing where one was disguised and the other naturally
written, and the FHE was required to give an opinion as to which of the pair was disguised.

13
In Chapter 4, the handwriting samples were presented as individual samples that were either
naturally written or disguised.
Poor writer skill in natural writing is not the only complicating factor in an assessment
of writing behaviour. Writers may be sufficiently skilled to be able to simulate model writing
relatively fluently, without introducing an abundance of their own handwriting
characteristics. This is supported by a study on the effects of practice on the dynamics of
handwriting (Portier, van Galen & Meulenbroek, 1990, quoted in van Gemmert and van
Galen, 1996) which found that dynamic measures such as movement time and dysfluencies
can change as a result of practice, so even the dynamic features may not be a good indication
of forgery if the writer has practised. A primary conclusion of this kinematic study is that the
static script alone cannot be used to distinguish simulated from genuine writings. Chapter 6
deals with the skill characteristics of FHEs associated with simulated handwritten text,
taking into account simulator practice and the use of a direct model for simulation.
Furthermore, some of the generally accepted features of the static image of disguised
and simulated writings are supported by empirical research investigating dynamic features.
Van Galen & van Gemmert (1996) investigated the kinematic and dynamic features of
forging another person’s writing and reported slower movement velocities and more
dysfluencies in forged samples compared to the simulators’ own normal handwriting. They
found the writing was produced by more frequent, but smaller force pulses, and the stiffness
of the writing limb is greater. This is due to the natural ‘automatic’ behaviour being
suppressed while the writer attempts to copy, draw or otherwise alter their normal writing.
Limb stiffness can also lead to a smaller range of variability in spatial parameters such as size
and slant. Pen pressure was also reported to increase under stress conditions (such as could
be expected when disguising or forging a handwriting sample), and was less variable
throughout the writing trace, although Sita & Rogers (1999) found that some simulators
decreased their writing pressure toward that of the model writer. Findings similar to van
Galen and van Gemmert (1996) relating to decreased movement velocities and increased
pen pressure have been reported for disguised writings (van Gemmert, van Galen, Hardy &
Thomassen, 1996).
Dynamic features of natural, disguised and simulated handwriting samples are
explored in Chapter 7, in order to determine whether any of the five measured dynamic
features can be used to distinguish between disguise and simulation writing behaviours, and
thus potentially serve as a predictor of handwriting behaviour in the static trace.
The relationship between writers’ opinions on their disguise strategies and FHEs
perceptions of those same disguise strategies (observed in the static trace of the sample) is
investigated in Chapter 8. FHEs perceptions of the hallmarks of disguise and simulation are

14
probed in Chapter 9, based upon their verbal statements as to what feature(s) of a specific
handwriting sample made them arrive at the conclusion that it was either disguised or
simulated.

1.7 Summary and Scope of the Thesis

Ongoing empirical research in the area of forensic handwriting examination is needed to
improve the scientific foundations of the field. An overall aim of this thesis is to investigate
the scientific validity of some of the principles underpinning forensic handwriting
examination, and obtain empirical data on the skill of examiners. Understanding the
rationale used and logic applied to certain handwriting examination problems can then
potentially feed back into method development, and inform the decision-making process.
This thesis specifically investigates the following:
• FHEs’ opinions on the process of production of disguised and simulated signatures,
to test whether the high misleading and inconclusive rates observed for authorship
opinions given on these types of signatures is due to a failure to detect the unnatural
writing process, or a misinterpretation of the unnatural process (Chapter 2).
• Whether or not FHEs have a skill for determining disguised and natural text-based
writings, when the samples are presented as a pair of natural and disguised writings
(Chapter 3) or as individual samples that were either naturally written or disguised
(Chapter 4).
• Whether or not FHEs have a skill for determining disguised and simulated text-
based writings, when the questioned unnaturally written sample was presented along
with an associated natural comparison writing sample (Chapter 5).
• The skill characteristics of FHEs associated with simulated samples that were
executed with or without practice, and with or without the use of a direct model.
This study investigated the hypotheses that practice will improve the success of
simulation, and that the increased cognitive load required to assemble a simulated
text from course of business writings will have an adverse effect on the success of
the simulation compared to writings simulated from a direct model, which in turn
would affect FHEs’ opinions (Chapter 6).
• The dynamic character of features associated with naturally written, disguised and
simulated text-based writings, exploring whether any of the measured features may
be used as a predictor of writing behaviour. Further consideration was made of those
dynamic features which may be inferred from the static image (Chapter 7).
• The relationship between writers’ opinions on disguise strategies they applied to a
text-based sample and FHEs’ perceptions of the disguise strategies they observed in
15
 the static image of those samples, to determine the degree of association between the
two (Chapter 8).
• The success of disguise strategies employed by writers, as assessed by two FHEs,
based on a comparison between the naturally written and disguised samples (Chapter
8).
• The relationship between the FHE groups’ verbal statements of features within
handwriting samples that they deemed to be indicative of either disguise or
simulation, and their scores for correctly determining the disguise or simulation
behaviour (Chapter 9).

16
 Chapter 2
Forensic Handwriting Examiners’ Opinions on the
Process of Production of Disguised and Simulated
Signatures

2.1 Introduction
Experts have been offering opinions on the authorship of questioned signatures in courts
for over 100 years (Huber & Headrick, 1999 pp. 4-5). Contemporary forensic handwriting
examiners learn the skills required to examine and compare handwriting features primarily
from mentor training, educational workshops, traditional text books (Conway, 1959; Ellen,
1989; Harrison, 1958; Hilton, 1982; Huber & Headrick, 1999; Osborn, 1929), journal
articles, proficiency tests and practical experience examining signatures. However, it is only
in the past 20 years, since the publication of a series of articles (Risinger et al., 1989; Saks &
Risinger, 1996) questioning the tenets of the field, that there has been a move towards
characterising the purported skill of forensic handwriting examiners. Past studies (Dyer,
Found & Rogers, 2006, 2008; Found et al., 1999; Kam et al., 2001; Sita et al., 2002) have
compared FHEs’ opinions with those of lay people and found that FHEs do possess
expertise in relation to expressing opinions of authorship of questioned signatures. The
nature of the expertise lies in the significantly lower rate of misleading (erroneous) opinions
expressed by FHEs as compared to lay people. However, large-scale blind testing has
identified some problem areas in FHEs’ skill, primarily associated with simulation and
disguised writing behaviours (Found & Rogers, 2005a, 2008). Found and Rogers (2008)
show that FHEs are markedly more conservative expressing authorship opinions on
questioned disguised and simulated signatures compared to opinions expressed on
questioned genuine signatures. In addition, authorship opinions on questioned disguised and
simulated signatures attracted higher inconclusive rates and misleading scores than did the
questioned genuine signature group. Thus the greatest source of FHEs’ misleading and
inconclusive authorship opinions were associated with questioned signatures where it would
normally be expected that a combination of similar and dissimilar features would exist, when
compared to the specimen signature group.
The misleading opinions in the study reported by Found and Rogers (2008) could
have arisen from two sources in terms of the cognitive comparison processes. Firstly, the
FHEs may have detected that there were dissimilar features associated with the disguise
and/or simulated signatures and attributed the dissimilarities to the incorrect writing
17
behaviour (i.e. simulation for disguised, and vice versa). For example, an FHE may have
detected feature dissimilarities associated with a questioned disguised signature and
attributed these to simulation behaviour and thereby conclude that the signature was not by
the specimen writer. Alternately, an FHE may have detected feature dissimilarities associated
with a questioned simulated signature and attributed these to disguise. This would result in
the erroneous opinion that the signature was written by the specimen writer. Both of these
decisions would result in misleading authorship opinions. Secondly, an FHE may not detect
features in the questioned signature that were dissimilar to the specimen material, or even
may consider that any dissimilarity is the product of natural variation in the signature rather
than indicative of a disguise or simulation process. This approach would result in correct
authorship opinions for questioned disguised signatures but misleading authorship opinions
for questioned simulated signatures.
In this chapter the potential source of the misleading authorship opinions for the
disguised and simulated signatures is explored by relating FHEs’ opinions on authorship
with their opinions on the process of production of each questioned signature. These
process opinions can be described as ‘first stage’ opinions, which inform the ‘second stage’
authorship opinion. The process opinions were requested of FHEs when offering opinions
regarding the authorship of questioned signatures on validation trials. These validation trials
comprised sets of materials comprising a known group of signatures from an individual
(specimens) and a comparison set of questioned signatures that contained a randomised
mixed of genuine, disguised and simulated signatures. In each trial set the actual authorship
and process of production was known to the experimenters but not to the examiners. The
task of the FHE was to express an opinion as to the process of production and the
authorship of each of the questioned signatures within the trials. FHEs were provided with
three process opinion options; 1. There is evidence that the questioned signature under
examination was naturally written, 2. There is evidence that the questioned signature is the
product of a disguise or simulation process, or 3. The evidence did not clearly support either
of these two propositions (inconclusive opinion). These ‘first stage’ opinions could then be
compared to the misleading authorship opinions for the questioned disguised and simulated
signatures. The source of the misleading authorship opinions could then be attributed to
either FHEs detecting and misinterpreting the differences in writer behaviour, or to them
not detecting any differences in the writer’s behaviour in comparison to the specimen
material. In this way the misleading authorship opinions can be further characterised on the
basis of the FHEs’ assessments of the first stage process of production of each of the
questioned disguised and simulated signatures.

18
 An analysis of the process opinions also allowed further exploration as to whether the
high inconclusive rates for authorship opinions on questioned disguised and simulated
signatures resulted from examiners not detecting the indicators of the disguise/simulation
process or whether they did detect the signs but were not prepared to opine whether the
process was produced by the author of the specimens or by somebody else. This is
important as a correct opinion that a questioned simulated signature is the product of either
a disguise or simulation process (i.e. support for the proposition that a signature is
something other than genuine) does provide potentially useful information in terms of the
investigation of a case. That the FHE does not go on to form an authorship opinion
regarding the signature does not detract from the evidential importance of the first stage
opinion.

2.2 Methods
The following trials were administered and raw data collected prior to commencement of
research for this degree. Trial data and other relevant information were made available by
the experimenters, Bryan Found and Doug Rogers. Work undertaken by the author
commenced at the Analysis stage of this chapter.

2.2.1 Participants
Results reported here are opinions of FHEs that were self-declared ‘qualified’ to give
evidence regarding their opinions on the authorship of questioned handwriting and
signatures. One trial was administered to participants each year for a four-year period. Every
participant received the same material at approximately the same time. The participants in
any given year were not necessarily the same as in other trial years. Participants’ responses
submitted for assessment were a mixture of peer-reviewed (agreed opinions of two
examiners) and individual responses. The number of peer reviewed and individual responses
provided for each of the four trials is presented in Table 2.1. One examiner giving single
responses in each of the 2004 and 2005 trials did not give any opinions on the process of
production of the questioned signatures. The results of this participant were excluded from
the data for this analysis.

19
Table 2.1. Number and type of responses submitted along with the number of specimen
and questioned signatures and total number of opinions expressed by the group for the 2002
to 2005 trials.
Peer Total #
Analysed Individual Specimen Questioned
Year Reviewed of
Responses Responses Signatures Signatures
Responses Opinions
2002 34 13 21 9 200 6800
2003 39 13 26 5 199 7761
2004 19 6 13 16 100 1900
2005 29 5 24 15 100 2900

2.2.2 Materials
The trials were constructed according to the accepted process of comparing specimen
(known) writings with a questioned writing sample. The rationale for the structure of the
trial material, where repetitions of a single person’s specimen signature is compared to
multiple questioned signatures has been described by Found & Rogers (2003). The writer of
each of the questioned samples was known to the experimenters but not to the participants.
For each trial, the questioned samples were a random mixture of normal signatures written
by the specimen writer, disguised signatures written by the specimen writer, and simulated
signatures written by forgers freehand copying the feature characteristics of the specimen
writer’s signature. Within each trial, all writings were made using the same make of ballpoint
pen and the same make of paper. The signatures were scanned at 600dpi. For the 2002 to
2004 trials the scanned signatures were inkjet printed into a booklet. In 2005 the scanned
signatures were reproduced photographically. Table 2.1 gives the number of specimen
signatures and the total number of questioned signatures for each trial. The number of
questioned genuine, disguised and simulated signatures for each of the trials is given in Table
2.2.

Table 2.2. Number of questioned genuine, disguised and simulated signatures used in each
of the 2002 to 2005 trials.

Year Genuine Signatures Disguised Signatures Simulated Signatures

2002 76 20 104
2003 120 22 57
2004 50 8 42
2005 20 9 71

20
2.2.3 Procedure for obtaining opinions
FHEs were provided with either a sample booklet (with PDF files of the images on a CD
also included) or photographs, and an answer-recording booklet. Examiners were informed
that the questioned samples were genuine, disguised or simulated and that the date range
over which the specimen signatures were taken was around the time that the questioned
samples were written. They were asked to compare each questioned signature with the
specimen material and give their opinion on whether or not the questioned samples were
written by the specimen writer (or whether they were unable to say) and whether each
questioned signature was written naturally, simulated/disguised or whether the examiner was
unable to say (i.e. a process opinion) by entering the appropriate code (a digit that was 1, 2,
3, 4 or 5 for authorship; and 1, 2 or 3 for process) in the answer booklet comprising boxes
corresponding to each one of the questioned samples. FHEs were provided with the
following definition of the opinion levels:
Authorship
The first digit in the answer code for each of the questioned signatures refers to the
authorship opinion. This digit was a 1, 2, 3, 4 or 5. The levels examiners had to choose from
were:
1. The questioned signature was written by the writer of the signature specimens. This
is an opinion of ‘moral certainty’ on the part of the examiner. For some FHEs this
opinion level translates into ‘very strong support for the proposition that the
questioned signature was written by the writer of the signature specimens’.
2. There are indications that the questioned signature was written by the writer of the
signature specimens. For some FHEs this opinion translates to statements such as
‘moderate support for the proposition that the questioned signature was written by
the writer of the signature specimens’ or ‘it is probable that the questioned
signature was written by the writer of the signature specimens’.
3. No opinion can be expressed as to whether or not the questioned signature was
written by the writer of the signature specimens. This is an inconclusive opinion.
4. There are indications that the questioned signature was not written by the writer of
the signature specimens. For some FHEs this opinion translates to statements such
as ‘moderate support for the proposition that the questioned signature was not
written by the writer of the signature specimens’ or ‘it is probable that the
questioned signature was not written by the writer of the signature specimens’.
5. The questioned signature was not written by the writer of the signature specimens.
This is an opinion of ‘moral certainty’ on the part of the examiner. For some FHEs

21
 this opinion level translates into ‘very strong support for the proposition that the
questioned signature was not written by the writer of the signature specimens’.
Process
The second digit in the answer code for each of the questioned signatures refers to the
opinion regarding the process by which the questioned signature was produced. This digit
was a 1, 2 or 3. The strength of the process opinion was not required.
1. There is evidence that the questioned signature was written naturally (that is, there
is no evidence of a disguise and/or a simulation process).
2. There is evidence that the questioned signature was written using a disguise and/or
a simulation process.
3. No opinion can be expressed as to the process by which the questioned signature
was produced.
For the purpose of the trials, participants were provided with the following definitions of
disguise and simulation processes;
Disguise process: an attempt by a writer to purposefully alter their signature in order to
avoid being identified or for them to deny writing the signature.
Simulation process: an attempt by the writer to imitate the signature characteristics of a
visual or mental model. This imitation behaviour can either be the specimen writer
imitating their own signature characteristics (as a form of disguise or resulting from
acute internal or external influences) or someone other than the specimen writer
‘forging’ the specimen writer’s signature characteristics.

2.2.4 Analysis
Participants’ authorship and process responses were marked independently as correct,
misleading or inconclusive. For disguised and genuine signatures, authorship opinion codes
1 and 2 were marked correct while codes 4 and 5 were marked misleading. The opposite was
true for the simulated signatures. These marks were then analysed to produce scores for
each of the different questioned signature types (genuine, disguised and simulated) and the
relationship between authorship and process opinions was investigated. For this chapter,
only the results for disguised and simulated signatures have been considered. The process
opinion was combined with the authorship opinion and analysed for disguised and simulated
signatures where there was a misleading or inconclusive authorship opinion. Table 2.3 gives
the combined opinion codes of interest, which are listed with the authorship opinion first,
followed by the process opinion. These may each be Correct, Misleading or Inconclusive.
Therefore, M/C corresponds to a misleading (incorrect) authorship opinion with a correct
process opinion while I/C represents an inconclusive authorship opinion with a correct

22
process opinion. The corresponding perceived signature types that the authorship/process
opinion combinations represent are also shown in Table 2.3.

Table 2.3. Perceived signature type for authorship/process opinion combinations

compared to actual signature type (M = Misleading, C = Correct and I = Inconclusive).

Author/Process Actual Disguised Signature Actual Simulated Signature

Opinion
M/C Simulated Disguised
M/M Natural by other writer Genuine
By other writer (process By specimen writer (process
M/I
unknown) unknown)
I/C Unnatural (author unknown) Unnatural (author unknown)
I/M Natural (author unknown) Natural (author unknown)
I/I Don’t know Don’t know

2.3 Results and Discussion

Over the four trials the group expressed 19361 authorship/process opinions. A summary of
the authorship opinions alone is given in Table 2.4. The opinion scores of interest in this
chapter are the misleading and inconclusive scores relating to the disguised and simulated
signature types (highlighted in grey).

Table 2.4. Authorship opinion rates over the four trial years for genuine, disguised and
simulated signatures.
Score Genuine Disguised Simulated
% Correct 83.6 24.3 46.6
% Misleading 2.3 12.3 2.9
% Inconclusive 14.2 63.3 50.5

Table 2.5 sets out the combined authorship and process opinions by questioned
signature type for disguised and simulated signatures where there were misleading and
inconclusive authorship opinions. The opinion codes are listed with the authorship opinion
first, followed by the process opinion. The overall scores for the group varied from trial to
trial. For example, the M/C (misleading authorship with correct process) opinions for
disguised signatures ranged from 5.0% (34 opinions out of 680 opinions on disguised
signatures in 2002) to 13.0% (34 opinions out of 261 opinions on disguised signatures in
2005). Variations of this type may be the result of differences between the trials with respect
to participant skill, the relative complexities of the signatures being examined, the extent to

23
which the specimen signatures covered the writer’s range of natural variation, the skill
characteristics of the simulator groups and the effectiveness of the disguise strategies
employed by the specimen writers. Statistical comparisons of the categories year by year
cannot be done due to sample size and multiple testing error.

Table 2.5. Number of combined authorship and process opinions for disguised and
simulated questioned signatures for each of the 2002 to 2005 trials.
Author/Process
2002 2003 2004 2005 Totals
Opinion
Disguised
M/C 34 91 8 34 167
M/M 2 29 1 0 32
M/I 1 25 4 12 42
I/C 290 235 49 133 707
I/M 1 87 2 0 90
I/I 86 319 27 12 444

Simulated
M/C 19 43 4 33 99
M/M 4 2 2 136 144
M/I 0 0 3 5 8
I/C 1542 1184 426 686 3838
I/M 17 8 1 11 37
I/I 214 168 61 50 493

For the total misleading authorship opinions related to disguised signatures (final
column of Table 2.5), the majority of process opinions were correct (167 compared with 32
misleading process opinions and 42 inconclusive process opinions ). This suggests that the
major cause of the misleading authorship opinions is the examiner misinterpreting the
disguise/simulation process as indicative of a different writer rather than failure to detect a
process different to a genuine writing process. Figure 2.1 gives two examples of such
questioned signatures, along with genuine signatures from their respective trial years.
For the total inconclusive authorship opinions related to disguised signatures, again
the majority of process opinions were correct (707 compared with 90 misleading and 444
inconclusive opinions). The difference between the correct and misleading process opinions
for these signatures was significant (p = 0.0008), with the difference between correct and
inconclusive opinions approaching significance (p = 0.055). This result suggests that in many
instances examiners are detecting an unnatural writing process but are unwilling to opine as
to the author.

24
 a

Figure 2.1. (a) Genuine signature, and (b) disguised signature thought simulated by at least
one of the participants in the 2003 trial. (c) Genuine signature, and (d) disguised signature
thought simulated by at least one participant in the 2004 trial.

When the 4368 inconclusive authorship opinions for simulated signatures are
considered it is clear that in the vast majority of instances (3838) FHEs detected the
simulation/disguise process. In only 11.3% of instances (493 of 4368) did FHEs provide an
inconclusive process opinion and in only 0.85% of instances (37 of 4368) were their process
opinions misleading. There was a significantly greater number of correct process opinions
compared with inconclusive process opinions (p = 0.005) and misleading opinions (p <
0.0001). These results provide strong support for the usefulness of the first stage process
opinion as FHEs’ inconclusive authorship opinions are generally the result of detecting the
signs of a disguise/simulation process and declining to opine on the author. In casework,
the opinion that a signature is not the result of a natural writing process may be of high
evidential value particularly as in many instances signature disguise can be excluded as an
alternate hypothesis by the courts on the basis of other evidence.
The majority of the total misleading authorship opinions relating to simulated
signatures were associated with incorrect process opinions (144 compared with 107 for the
inconclusive and correct scores combined), however, the differences between the process

25
opinions for these signatures are not statistically significant (H = 2.96, p = 0.23). The results
of the 2005 trial (with 136 of the 144 opinions of this type) are very different from the
previous years. This shows that for the 2005 trial, the main cause of the misleading
authorship opinions is examiners not identifying the simulation process and deeming
signatures to be genuine. This trial incorporated skilled simulations from a simulator who
was asked to practice the signature many hundreds of times. Under this practice regime it
was reasonably expected that the resulting simulations could be of higher quality than for a
routine simulation collection regime. It was therefore not surprising that the opinions
reported were more conservative or misleading than for a low practice condition trial. An
example of one of these signatures, along with a corresponding specimen signature, is given
in Figure 2.2. Apart from the spike in the 2005 results, it is clear that in the majority of cases
FHEs detected the simulation/disguise process but misinterpreted the observed
inconsistencies as an attempt by the genuine writer to disguise their signature. The variation
between trial years may be indicative of the skill of the simulator(s) and/or FHE
participants.

a b

Figure 2.2. (a) Specimen signature and (b) simulated signature thought genuine by at least
one of the participants in the 2005 trial.

If we disregard the authorship opinions and compare the total misleading and
inconclusive process opinions for disguised and simulated signatures (given in Table 2.6 and
2.7 respectively), it is apparent that there is more examiner uncertainty associated with the
disguised signatures. For the process opinions on disguised signatures 37.8% were
misleading or inconclusive while for simulated signatures 14.1% were misleading or
inconclusive. This result is significant (χ2(1, N=10567) = 590, p< 0.0005), and the ratio of
misleading to inconclusive opinions is substantially higher for disguised signatures than for
simulated signatures (χ2(1, N=1955) = 94.4, p< 0.0005). This result can be explained in part
by the observation of Michel (1978) that a large proportion of people are unable to
successfully disguise their signature, creating disguised signatures that predominately fall
within the normal range of variation of their genuine signatures (Figure 2.3), perhaps

26
changing only obvious features such as letter shapes or slant, but retaining attributes of the
finer structure.

Table 2.6. Process opinion rates for disguised signature types over the four trial years.
2002 2003 2004 2005 Total
% Correct 77.1 43.8 50.7 90.8 62.2
% Misleading 3.4 13.5 15.1 0 8.3
% Inconclusive 19.6 42.7 34.2 9.2 29.5

Table 2.7. Process opinion rates for simulated signature types over the four trial years.
2002 2003 2004 2005 Total
% Correct 87.6 90.9 89.0 76.3 85.9
% Misleading 4.7 0.9 2.0 16.1 6.2
% Inconclusive 7.7 8.3 9.0 7.7 8.0

Figure 2.3. (a) Specimen signature and (b) disguised signature thought genuine by at least
one of the participants in the 2002 trial.

The year-to-year breakdown of scores reveals that there is a wide range of variation in
the scores for each signature type. We can explore the possible explanations for this. In
order to form an opinion on the process used to create a questioned signature FHEs will
27
examine the signature for evidence of simulation/disguise in conjunction with the usual
method of feature comparison to determine authorship. The features considered may relate
to line quality, spatio-temporal features or the extent of variation in the formation being
examined.
Freehand simulations (the only type of simulation used in these trials) are likely to
contain similarities with the specimen signatures in gross features, along with discrepancies
in the details of formation, as well as poor line quality and fluency. These are the result of
the simulator usually only being able to focus attention on one of the two major
requirements of achieving a good simulation: accuracy in formation (size, proportion,
shape); or fluency of the line trace. In most cases, the signs of a simulation process (poor
line quality, differences in size, proportion, slope and/or formation) will be clearly
observable and identified as such, as evidenced by the FHEs overall low inconclusive and
misleading scores for this signature type. However, some simulated signatures may not bear
a resemblance to the specimen signature at all and in some of these cases the questioned
simulated signature may display evidence of having been fluently written and therefore be
deemed by the FHE to have “no evidence of a simulation/disguise process”. This is an
artefact of the ambiguity of the process term “naturally written”, and it being equated to
“fluently written”. In these cases, the misleading process opinion will typically not result in
an incorrect authorship opinion.
Disguised signatures are also expected to exhibit similarities and differences with the
specimen signatures. Often, the differences will be in the more obvious features, while the
finer details of construction and fluency remain intact. Some disguised signatures may not
resemble the specimen signature at all, perhaps being written in the writer’s natural
handwriting rather than their usual signature form. These may display similar characteristics
to the “fluently written” simulated signatures described above. In some cases, an individual
disguising their signature may employ the techniques of a simulator and copy their genuine
signature slowly. However, a large proportion of people are unable to successfully disguise
their signature, creating signatures that predominately fall within the normal range of
variation of their genuine signatures (Michel, 1978). This may explain the relatively higher
misleading scores associated with this signature type, as the lack of significant differences
may lead FHEs to opine that the questioned disguised signatures are the result of a natural
writing process.
The impact of misleading process (first stage) opinions is revealed when considered in
conjunction with the second stage authorship opinion. A potentially misleading process
opinion does not lead to a misleading error when the corresponding authorship opinion is
correct or inconclusive. The only cases where a misleading process opinion leads to a

28
misleading authorship opinion are the M/M combinations in Table 2.5. For simulated
signatures, these errors reflect a perceived genuine signature and are dependent at least in
part upon the skill characteristics of the simulator group. For disguised signatures, the error
indicates perceived naturally written signatures by another writer. In this case we would
expect the signature to exhibit differences from the genuine signature along with fluency of
the line trace. These real errors varied markedly from year to year as described above. If the
misleading rates are averaged over the four years of trials they show that when an opinion of
a natural writing process is expressed, it leads to an error in just over 1.4% of authorship
opinions (the error associated with disguised and simulated signatures combined).

2.4 Conclusion
FHEs as a group exhibited a high level of skill in detecting the unnatural process of
production of simulated signatures and the high inconclusive rate for authorship opinions
on signatures of this type is due to the difficulty in ascribing an author to the process. This
difficulty is highlighted by the observation that the majority of erroneous authorship
opinions for the disguised and simulated signatures occurred when FHEs detected the
unnatural writing process but incorrectly excluded or identified the genuine writer. This is
consistent with, for example, Harrison (1958, p. 349) who notes that “where disguise is
confidently expected (the FHE) will run the very real risk of making an incorrect
identification of the writer by lightly dismissing as disguise any consistent dissimilarities
which otherwise he would have attributed to different authorship.”
These results provide a foundation for the usefulness of FHEs offering a first stage
simulation/disguise process opinion without going on to form an opinion on authorship, as
support for the proposition that a signature is something other than genuine may be of
strong evidential value. It is then up to the judiciary to ascribe weight to the alternate
propositions based on the other evidence available to it, if any. The error associated with
authorship opinions could be greatly reduced in most cases by stopping short of expressing
an opinion if a simulation/disguise process is identified. This situation may change should
reliable predictors of unnatural writing types be identified. The remainder of this thesis is
oriented toward investigating the skill of FHEs relating to process opinions on questioned
handwritten text, and identifying predictors of disguise and simulation handwriting
behaviours.

29
 Chapter 3
Forensic Handwriting Examiners’ Skill in
Distinguishing Between Natural and Disguised
Behaviour from Handwritten Text Pairs

3.1 Introduction
In the past two decades a number of studies (Kam, ca. 2010; Kam et al., 1994, 1997) have
compared examiners’ opinions with those of lay people and found that FHEs do have
expertise in relation to opinions on questioned handwritten text. However, these expertise
studies were not intended to characterise more subtle aspects of the purported expertise and
in none of the reported studies were participants required to provide authorship opinions on
unnatural questioned writings that were a mixture of disguised handwritten text by the
genuine writer and simulated writings by someone other than the genuine writer. Similar to
the findings for signatures (considered in Chapter 2), recent research involving only FHEs as
participants has identified disguised and simulated handwritten text as problem areas for
forensic handwriting examiners when the questioned writings have included a mixture of
disguised and simulated writings (Found & Rogers, 2005a). The study has shown that these
writing types attract higher incorrect and inconclusive rates on authorship opinions than
genuine writing does.
Of interest here are authorship opinions formed by FHEs on questioned disguised
and simulated writing samples. Given that FHEs have been shown to have a higher error
rate for unnatural handwriting than normal writing, if FHEs can identify a piece of
questioned writing as the product of an unnatural (disguised or simulated) writing process,
they can express this ‘first stage’ opinion without necessarily proceeding on to express an
opinion on authorship (a ‘second stage’ opinion). Conservatism with regard to second stage
opinions should help to reduce any potential errors in casework. However, there is currently
no published empirical data on FHEs’ capacity to distinguish between natural and disguised
writing behaviour in a blinded fashion. In this chapter we attempt to examine whether
forensic handwriting examiners have a skill in determining which of a pair of handwriting
samples was written using a disguise process. In order to do this, their opinions on the
process of production of 140 pairs of natural and disguised handwriting samples were
examined and compared to the opinions of a control group of lay people.

30
3.2 Methods
3.2.1 Participants
Results reported here are from two groups of participants (forensic handwriting experts and
lay people) who provided independent opinions on the process of production of the
supplied handwriting samples.
Eleven FHEs from two countries took part in the trial, six from the United States of
America and five from Australia. All participants had a minimum of two years full-time
government laboratory training under a qualified examiner and have worked in government
laboratories. Their training meets the standard set out in the ‘ASTM E2388-05 Standard
Guide for Minimum Training Requirements for Forensic Document Examiners’. All of the
participants have been authorised by their employers to release opinions regarding the
authorship of questioned handwriting and signatures. At the time of data collection, ten
FHEs were working in government laboratories and one was working privately.
None of the ten lay people had training in forensic handwriting and signature
examinations; however two reported having some knowledge of the discipline. The self-
reported knowledge of forensic handwriting examination came from a participant who
works within a forensic science laboratory in an unrelated discipline, and a student who had
undertaken a short work experience placement within a forensic handwriting examination
laboratory.
The author is aware that the number of participating FHEs is small. However, with
the relatively small number of FHEs worldwide available (and willing) to be tested, and this
trial comprising one third of the larger project requiring the involvement of FHEs, a
significantly larger sample size could not practically be obtained.

3.2.2 Materials
The trial consisted of 140 pairs of handwritten text, each pair written by one writer, with 70
writers in total. One of the samples in the handwriting pairs was the writer’s normal
handwriting, in either upper or lower case print, and the other was a disguised handwriting
sample in the same case. No directions were given to the handwriting providers as to how to
disguise their handwriting, aside from imagining that they do not want the reader of the text
to recognise them as the author (see Appendix A for the full instructions given to
handwriting providers). All writings were made using the same make of ballpoint pen and
the same make of white paper. The samples were scanned at 600dpi and inkjet printed into a
booklet and converted into high resolution PDF files.

31
3.2.3 Procedure
The FHE group was provided with a sample booklet of inkjet-printed images of the
handwriting samples, while the lay people were provided with a CD containing PDF files of
the samples. The different mode of application for the trials is a result of the large number
of samples, and time constraints involved in distributing a single sample folder to all
participants. The author notes that the different form of material examined by lay people
and FHEs is a limitation of this study. However, anecdotal evidence indicates that FHEs are
equally effective examining digital representations as hardcopies, and it could be argued that
digital files allow easier magnification of the fine features compared with a printed copy of
the original. Empirical evidence of the success of FHEs examining handwriting on-screen
can be found in Holmes, Ostrum and Barton (2011).
Every participant was given an answer-recording booklet and informed that each
sample pair was written by one writer, and consisted of one naturally written and one
disguised handwriting sample. They were asked to make an assessment as to which of the
sample pair (A or B) was disguised by entering the appropriate code (A, B or I (for an
inconclusive opinion)) in the answer booklet comprising boxes corresponding to each one
of the questioned sample pairs. If an inconclusive opinion was given, the participant was
required to record a forced opinion (A or B) for the handwriting pair. The forced and
unforced opinions of the participants were analysed separately and compared. In addition,
the disguise strategy or strategies the participant believed the writer used was recorded in the
answer booklet by marking the appropriate box(es) corresponding to writer-reported
disguise strategies and those commonly mentioned in the literature (Alford, 1970; Harrison,
1958 pp. 349-372; Keckler, 1997; Konstantinidis, 1987; Regent, 1977; Totty, 1991). An
analysis of this aspect of the trial is reported in Chapter 8.

3.2.4 Ethics approval

Approval for this study was obtained from the La Trobe University Human Ethics
Committee on the basis that the handwriting providers and test participants gave full
consent for samples of their handwriting, or research data provided by them respectively, to
be included in published material, on the condition that their name and any other identifying
information remain confidential.

3.2.5 Analysis
Participants’ responses were marked as correct, misleading or inconclusive, and analysed to
produce scores for the groups of FHEs and lay people. In addition, the forced responses
following an inconclusive opinion were further marked as forced correct or forced
32
misleading. To determine whether there were differences in scores between the examiner
group and the lay group a series of planned comparisons were undertaken using
independent sample t-tests (two-tail).

3.3 Results and Discussion

When the overall results for the forensic handwriting examiners and lay people are
compared, it is clear that there is a difference in the response profiles between the two
groups. Table 3.1 provides a summary of the correct, misleading and inconclusive responses
for the group of FHEs and lay people tested.

Table 3.1. Scores of grouped responses for FHEs and lay people.
Score FHEs Lay people
% Correct 73.38 80.07
% Misleading 3.38 11.43
% Inconclusive 23.25 8.43

Previous studies (Dyer et al., 2006, 2008; Found et al., 1999; Kam et al., 1994, 1997,
2001; Sita et al., 2002) have found that when compared to lay people, FHEs do have a
greater skill in relation to expressing authorship opinions on questioned handwriting and
signature samples. The results reported in this chapter are aligned with the trend evident in
these previous findings, where FHEs exhibit a greater skill in determining which of a pair of
handwriting samples was disguised. Furthermore, as in the research reported by Sita et al.
(2002) FHEs were found to be more conservative than their untrained counterparts, with an
inconclusive rate more than twice that of lay people (23.25% versus 8.43%, t = 3.43, df =
19, p = 0.003). Although the raw scores indicate that lay people have a higher correct rate
than FHEs, this difference is not significant (t = 1.81, df = 19, p = 0.086); however, the
error rate of FHEs is less than a third of the error rate of lay people (3.38% compared with
11.43%, t = 4.08, df = 19, p < 0.001). This means that when participants were willing to give
an opinion on which of a pair of handwriting samples was disguised, the FHEs have a
significantly higher correct rate than lay people (95.66% correct called compared to 87.84%,
t = 3.86, df = 19, p = 0.001). This is where the skill of FHEs rests: when an opinion is
given, it is more likely to be correct than when a lay person gives an opinion. Table 3.2 gives
the correct and misleading called rates for the two groups.

33
Table 3.2. Scores of grouped called responses (excluding inconclusive responses) for FHEs
and lay people.
Score FHEs Lay people
% Correct Called * 95.66 87.84
% Misleading Called 4.34 12.16
* #Correct*100 / (#Correct + #Misleading)

The high inconclusive rate for FHEs may be the result of unexpected similarities in
the handwriting sample pairs, where either the writer is not very adept at introducing or
maintaining a disguise, so the two samples appear very similar; or where the natural writing
of the writer is messy, internally inconsistent and/or lacks fluency, which are some of the
traditional hallmarks of unnatural writing (Alford, 1970; Harrison, 1958 pp. 349-372;
Keckler, 1997; Konstantinidis, 1987; Muehlberger, 1990; Regent, 1977; Totty, 1991). In
these cases where the disguised sample is not clear, the FHE, having been trained to be
conservative, will decline to opine as to which sample is disguised, while a lay person is more
likely to hazard a guess. An example of a sample pair with an inadequate disguise strategy is
given in Figure 3.1. In this case, six of the eleven FHEs gave an inconclusive opinion, while
the remaining examiners correctly identified the disguised sample. Conversely, only two of
the ten lay people declined to offer an opinion, with six correctly identifying the disguised
sample and the remaining two providing an erroneous opinion.
The second scenario is illustrated in Figure 3.2, where both the writer’s normal and
disguised handwriting samples appear internally inconsistent, with some tremor in the line
trace. Here, eight of the eleven FHEs preferred not to opine as to which was the disguised
sample and gave an inconclusive opinion. The three FHEs who did offer an opinion were
correct. The lay people had a 70% correct rate for this sample pair, however only one lay
person offered an inconclusive opinion, and two were incorrect.

34
 a b
Figure 3.1. Portion of a handwriting sample pair written by one person, where the disguised
sample was not clearly identified by some FHEs due to a subtle disguise strategy. (a)
Normal handwriting sample and (b) disguised handwriting sample.

a b
Figure 3.2. Portion of a handwriting sample pair written by one person, where the disguised
sample was not clearly identified by some FHEs due to a lack of fluency in the normal
writing. (a) Normal handwriting sample and (b) disguised handwriting sample.

If we examine the individual responses of participants in the two groups (Figure 3.3)
the trend touched on above becomes apparent and distinct examiner and lay people profiles
emerge. The typical examiner profile shows an inconclusive rate of between 14.3 – 37.9%,
with a misleading rate of 1.4 – 4.3%, while the typical lay person profile exhibits an
inconclusive rate between 0 and 14.9% and misleading rate of 6.4 – 20.7%. Only two of the
lay people (L04 and L15) exhibit the “examiner profile”, and one examiner (E08) displays a
“lay person profile”. The lay people with examiner-like profiles were the only two lay people
who self-reported knowledge of forensic handwriting examination at some level.

35
Figure 3.3. Individual participants’ correct, misleading and inconclusive responses for
handwriting sample pairs. Note: E denotes a forensic handwriting examiner, L denotes a lay
person.

In line with previous research (Sita et al., 2002; Found et al., 2001), there is no
correlation between FHEs’ years of experience and their skill for determining which of a
pair of handwriting samples was written using a disguise process. Regression statistics on
the% Correct for examiners were R = 0.32, p = 0.340 and for% Correct Called, R = 0.07, p
= 0.836.
Table 3.3 shows the percentage of forced correct and forced misleading opinions
expressed by FHEs and lay people. The difference between FHEs and lay people with
regard to their forced percentage correct (and misleading) scores is not significant (t = 1.26,
df = 15, p = 0.227). For both participant groups, the forced percentage correct scores are
less than 1% different when compared to the unforced correct scores, however, the forced
misleading scores are much higher than the unforced misleading scores. This difference is
significant for the FHE group (t = 12.29, df = 19, p < 0.0002), but is not so for lay people (t
= 2.05, df = 15, p = 0.058). This indicates that the decision to give an inconclusive opinion
is a good one (in particular, for FHEs), and does reduce the number of errors that an FHE
might attract should they be forced to proffer an opinion on the process of production.

36
Table 3.3. Scores of grouped forced responses for FHEs and lay people (i.e. the opinion
finally given when an initial inconclusive opinion was recorded).
Score FHEs Lay people
Forced% Correct 73.96 79.47
Forced% Misleading 26.24 20.53

As intuitively expected, many disguised handwriting samples are easily identified by

their stilted, unnatural appearance (Figure 3.4). These types of sample pairs did not pose a
problem for either the FHE or lay person group.

a b

Figure 3.4. Portion of a handwriting sample pair written by one person, where the disguised
sample was clearly identified by participants. (a) Normal handwriting sample and (b)
disguised handwriting sample.

However, other writers were able to employ a reasonably consistent, fluently executed
disguise which both FHEs and lay people had trouble identifying. The sample pair given in
Figure 3.5 attracted high erroneous opinions across both groups, with all of the examiners
who offered an opinion (72.7% of examiners) identifying the naturally written sample as
disguised. The remaining three FHEs gave an inconclusive opinion. The lay people fared
little better, with a 90% error rate and one person correctly identifying the disguised sample.
Interestingly, there were a few sample pairs where FHEs had no problem or
uncertainty in identifying the disguised sample, whereas lay people attracted an error. Figure
3.6 is one such example where FHEs had a 100% correct rate, while lay people had a 30%
error rate. Further investigation of the disguise strategies participants believed were
employed in creating these samples may help to identify predictors of disguised handwriting,
and further characterise the skill of FHEs undertaking this task.

37
 a b

Figure 3.5. Portion of a handwriting sample pair written by one person, where the disguised
sample was thought by some FHEs to be naturally written. (a) Normal handwriting sample
and (b) disguised handwriting sample.

a b

Figure 3.6. Portion of a handwriting sample pair written by one person, where the disguised
sample posed no problem to FHEs but attracted an error from lay people. (a) Normal
handwriting sample and (b) disguised handwriting sample.

3.4 Conclusion
As a group, FHEs do possess a skill in determining which of a pair of handwriting samples,
written by one writer, is disguised. The reported results show that FHEs have what would
be considered acceptably low misleading responses (≤ 4.3%). On a practical level this skill is
used by FHEs routinely in casework examination when assessing the likelihood of a sample
of questioned handwriting, or in some instances the comparison handwriting, being the
product of disguise behaviour. If a comparison sample is considered to be the result of an
38
unnatural writing process, further comparison material may be required. Alternately, if a
questioned handwriting sample is opined to be unnaturally written, this should lead the FHE
to carefully consider whether any authorship opinion should be formed given the reported
high misleading scores associated with this category of questioned writing (Found & Rogers,
2005a; 2008). Even without an opinion regarding authorship, the first stage process opinion
that a sample of handwriting is something other than genuine may be of strong evidential
value.
The high inconclusive rate of FHEs may be due in part to the presence of minimal
observed differences between natural and disguised samples. For example, when a writer is
poor at introducing or maintaining a disguise, there will appear to be many similarities
between the two samples, generally causing FHEs to decline to offer an opinion on which of
the pair is disguised. If the writer’s natural handwriting is not highly skilled, it may exhibit
the hallmarks of unnatural writing, which we would also expect to see in the disguised
handwriting sample. The lack of a clear difference in fluency causes the FHE to be
conservative. In other cases, the handwriting providers were able to successfully create a
disguised sample that appeared fluently written and different to their natural writing. The
combination of comparable fluency but different features and forms also brings about
conservatism in FHEs. The success of disguise and how this may impact on the skill of
FHEs will be investigated in Chapter 8.

39
 Chapter 4
Forensic Handwriting Examiners’ Skill in Detecting
Disguise Behaviour from Individual Handwritten
Text Samples

4.1 Introduction
Chapter 3 reported that forensic handwriting experts have a skill compared to lay people in
determining which of a pair of handwriting samples was written using a disguise versus a
natural handwriting process. However, when initially assessing the quality of a sample of
specimen or questioned handwritten text, each sample is inspected with minimal regard to
the other samples received for examination. Together with other factors, such as the style
and quantity of writing features, consideration is given to whether or not the characters and
text appear to be naturally written. In terms of contemporary approaches to method
development this may result in the formation of a first stage “process” opinion. In this
regard, being able to accurately determine whether an individual handwriting sample is
naturally written or otherwise may impact on the second stage (authorship) opinion. For a
questioned sample of heavily disguised writing where the disguise process is not detected,
the outcome may be an erroneous exclusionary authorship opinion. For a questioned
naturally written sample where, on comparison to a specimen sample, the dissimilar features
are erroneously associated with a disguise process this could result in an erroneous
identification authorship opinion. Clearly the determination of predictor features of disguise
behaviour is integral to the work of a forensic handwriting examiner.
The results reported in this chapter investigate the skill of FHEs and lay people at
discriminating between disguised and naturally written handwritten text based on 200
individual handwriting samples.

4.2 Methods
4.2.1 Participants
Results reported here are from two groups of participants (FHEs and lay people) who
provided independent opinions on the process of production of handwriting samples. Ten
FHEs from three countries (Australia, New Zealand and The Netherlands) took part in the
trial. All participants had a minimum of two years full-time government laboratory training
under a qualified examiner and each met the standard set out in the ‘ASTM E2388-05

40
Standard Guide for Minimum Training Requirements for Forensic Document Examiners’.
All FHEs were authorised by their employers to release opinions regarding the authorship of
questioned handwriting and signatures. At the time of data collection, all of the FHEs were
working in government laboratories.
None of the 10 lay people had training in forensic handwriting and signature
examinations; however, three reported having some knowledge of the discipline. The self-
reported knowledge of forensic handwriting examination came from one participant who
works within a forensic science laboratory in an unrelated (non-scientific) discipline, and two
students who had undertaken a short work experience placement within a forensic
handwriting examination laboratory.
The author is aware that the number of participating FHEs is small. However, with
the relatively small number of FHEs available to be tested, and this trial comprising one
third of the larger project requiring the involvement of FHEs, a larger sample size was not
able to be obtained.

4.2.2 Materials
The trial consisted of 200 individual samples of a specific paragraph of handwritten text.
Ninety eight of the samples were naturally written and the remaining 102 samples were
disguised. A total of 70 subjects volunteered to provide a handwriting sample in their normal
upper and/or lower case print style, and in a disguised upper and/or lower case print style.
Each writer was provided with the paragraph printed on a document (a handwritten copy of
the paragraph is provided in Figure 4.1). No directions were given to the handwriting
providers as to how to disguise their handwriting. All writings were made using the same
make of ballpoint pen and the same make of white paper. Each of the samples was scanned
at 600 dpi, and inkjet-printed into a booklet. The samples were also converted into high-
resolution PDF files.

Figure 4.1. One writer’s handwriting sample, showing the specific text requested in full.
41
4.2.3 Procedure
The FHE group was provided with a sample booklet of inkjet-printed images of the
handwriting samples, while the lay people were provided with either a sample booklet of
inkjet-printed images, or a CD containing PDF files of the samples. The different mode of
application for the trials is a result of the large number of samples and time constraints
involved in distributing a single sample folder to all participants. Since there is some
evidence indicating that FHEs are equally effective examining photocopies as originals, it is
thought that there would be little difference between lay persons’ opinions on digital versus
inkjet representations (Found et al., 2001; Found & Rogers, 2005b). The author notes that
the different form of material examined by lay people and FHEs is a limitation of this study.
Each participant was given an answer-recording booklet and informed that each
sample was either naturally written or disguised. They were asked to form an opinion as to
whether each sample was naturally written or disguised, or if they could not say. They
recorded their response for each of the 200 samples in the answer-recording booklet by
entering a code (1 = naturally written, 2 = disguised, or 3 = inconclusive). For those samples
where an inconclusive opinion was recorded, the participant was required to provide an
additional forced opinion (1 or 2) for the handwriting sample. Both the forced and unforced
opinions of the participants are considered in this chapter. As in the trial discussed in
Chapter 3, in addition to the process opinion (when participants gave an unforced opinion
of disguise) they were asked to also record the disguise strategy or strategies they believed
the writer used. This was done by marking the appropriate box(es) in the answer booklet.

4.2.4 Ethics approval

Approval for this study was obtained from the La Trobe University Human Ethics
Committee on the basis that the handwriting providers and test participants gave full
consent for samples of their handwriting, or research data provided by them, respectively, to
be included in published material, on the condition that their name and any other identifying
information remain confidential.

4.2.5 Analysis
Participants’ responses were marked as correct, misleading (erroneous), or inconclusive and
analysed to produce scores for the groups of FHEs and lay people. In addition, the forced
responses following an inconclusive opinion were marked as forced correct or forced
misleading. To determine whether there were differences in scores between the examiner
group and the lay group, a series of planned comparisons were undertaken using
independent sample t-tests (two-tail).
42
4.3 Results and Discussion
The results of the grouped responses across natural and disguised handwriting types are
given in Table 4.1.

Table 4.1. Scores of grouped responses for FHEs and lay people across both natural and
disguised handwriting samples.
All samples FHEs Lay people
% Correct 47.2 55.5
% Misleading 18.9 22.0
% Inconclusive 34.0 22.6

Both FHEs and lay people have a correct rate that is close to that of chance (47.2%
and 55.5% respectively) and both attract high misleading scores (18.9% and 22.0%
respectively). This provides evidence that the determination of writing process for an
individual sample is a difficult task for FHEs and lay persons alike. Although the raw scores
indicate that lay people have a higher correct rate than FHEs, this difference is not
significant (t = 1.17, df = 18, p = 0.256). Nor are the differences in misleading (t = 0.57, df
= 18, p = 0.573) or inconclusive rates (t = 0.97, df = 18, p = 0.347). However, previous
research investigating FHEs’ authorship opinions has found that disguised (and simulated)
handwriting types attract higher misleading and inconclusive rates than does natural
handwriting (Found & Rogers, 2005a, 2008). In order to determine if the different
handwriting processes also have an effect on the scores for process opinions, each
handwriting type was examined separately for the FHE and lay groups. Tables 4.2 and 4.3
give the grouped responses of FHEs and lay people for natural and disguised handwriting
samples respectively.

Table 4.2. Scores of grouped responses for FHEs and lay people for natural handwriting
samples.
Natural handwriting FHEs Lay people
% Correct 65.1 65.7
% Misleading 5.5 13.9
% Inconclusive 29.4 20.4

43
Table 4.3. Scores of grouped responses for FHEs and lay people for disguised handwriting
samples.
Disguised handwriting FHEs Lay people
% Correct 29.9 45.6
% Misleading 31.7 29.7
% Inconclusive 38.4 24.7

Again, although the raw scores indicate that lay people have a higher correct rate than
FHEs for both handwriting types, these differences are not significant. Furthermore, the
differences between FHEs’ and lay persons’ misleading and inconclusive scores for each of
the handwriting processes are not significant (see Table 4.4).

Table 4.4. t-Test results for FHE versus lay people scores for natural and disguised writing
types.
Natural Disguised
% Correct t = 0.05, df = 18, p = 0.958 t = 1.70, df = 18, p = 0.107
% Misleading t = 1.83, df = 18, p = 0.084 t = 0.19, df = 18, p = 0.850
% Inconclusive t = 0.75, df = 18, p = 0.465 t = 1.11, df = 18, p = 0.281

However, both groups of participants show a difference in the correct rate for natural
versus disguised writing samples, with natural handwriting behaviours significantly easier to
correctly identify than disguised writing behaviours2. There is a corresponding significant
difference in the misleading rate for natural versus disguised samples but not for the
inconclusive rate between the two handwriting types for either of the participant groups (see
Table 4.5).

2 Although it may appear attractive here to consider the results within a likelihood ratio (LR)
framework which provides a measure of the probative value of the FHEs and lay persons opinions
on natural and disguised handwriting, this was not calculated. In order to calculate the relevant
likelihood ratios (in this case the probability of the FHE or lay person opining that the sample was
naturally written when this hypothesis was true, divided by the probability of the FHE or lay person
opining that the sample was naturally written when the hypothesis was not true) there is a
requirement that the two hypotheses are independent. As is discussed in this chapter, disguised
samples may exhibit no evidence of disguise and therefore appear naturally written. The hypotheses
are therefore not independent. It is because of this that LR results have not been calculated.
44
Table 4.5. t-Test results for natural versus disguised writing for FHEs and lay people.
FHEs Lay people
% Correct t = 2.78, df = 18, p = 0.012 t = 2.69, df = 18, p = 0.015
% Misleading t = 2.61, df = 18, p = 0.018 t = 3.13, df = 18, p = 0.006
% Inconclusive t = 0.66, df = 18, p = 0.516 t = 0.41, df = 18, p = 0.688

Table 4.6 shows the percentage of correct forced opinions expressed by FHEs and lay
people for natural and disguised samples. Although for each handwriting type the percentage
of forced responses that were correct is higher for the FHE group than the lay people, the
difference between the groups with regard to their forced percentage correct (and
misleading) scores is not significant for natural (t = 0.21, df = 15, p = 0.837) or disguised
samples (t = 1.78, df = 15, p = 0.096).

Table 4.6. Scores of grouped correct forced responses for FHEs and lay people.
% Correct of Forced FHEs Lay people
Natural handwriting 51.1 48.2
Disguised handwriting 70.3 55.5

For the FHE group, the forced percentage correct scores are significantly higher than
the unforced correct scores for disguised handwriting samples (t = 4.34, df = 17, p =
0.0004), while the forced and unforced misleading scores are not significantly different (t =
0.18, df = 17, p = 0.86). For natural samples, the findings for FHEs were reversed. That is,
the difference between the forced and unforced correct scores was not significant (t = 0.98,
df = 16, p = 0.34) but the difference between the forced and unforced misleading scores
was (t = 4.99, df = 16, p = 0.0001).
The lay person group also showed a higher forced correct than unforced correct score
for disguised samples, but neither this difference (t = 1.16, df = 16, p = 0.26) nor the
difference between the forced correct and unforced correct scores for natural samples were
significant (t = 1.57, df = 17, p = 0.14). However, the forced misleading scores are
significantly higher than the unforced misleading scores for both disguised (t = 2.16, df =
16, p = 0.05) and natural handwriting samples (t = 3.56, df = 17, p = 0.002).
As these findings vary depending on handwriting type, and are particularly disparate
for FHEs, the decision to give an inconclusive opinion is a good one. Although the
combined unforced and forced correct rate taken over all handwriting samples (61.3%) is
higher than the unforced correct rate alone, the combined misleading rate rises to 38.7%
from less than 20%. Therefore, by declining to offer an opinion on some samples as
45
presented in this trial, a reduction is achieved in the number of errors that may have been
made by an FHE.
Ninety percent of individual handwriting samples which attracted high (≥70%)
unforced erroneous rates were disguised handwriting samples. Closer inspection of the
disguised samples where high errors were detected revealed that these samples bore a strong
resemblance to the writer’s natural writing behaviour. Although in this trial participants
examined samples individually, the similarity between disguised and natural samples means
that the disguise strategies employed by these writers were so subtle or inadequate (e.g. a
change in slope, connectivity, or the formation of a few letters), that the expected hallmarks
of disguise (decreased fluency, line quality disturbance, internal inconsistency (Cain, 1998;
Harrison, 1958 pp. 349-372; Konstantinidis, 1987)), were not clear to either the FHE or lay
group. Interestingly, Webb (1977) claims that some writers apparently find the task of
disguise so difficult that none is attempted at all. Alford (1970) reports in his survey of
methods of disguise, that most attempts are neither consistent nor successful.
Of the disguised samples that attracted high error rates, the majority (70.6%) were
lower case samples. If the results are further examined based on handwriting case as well as
writing type there appears to be a slightly higher correct rate for natural lower case writings
(FHE 66%, lay 67%) compared to natural upper case writings (FHE 64%, lay 64%).This
trend is reversed for disguised samples (lower case: FHE 25.6%, lay 39.1%; upper case: FHE
34.8%, lay 52.9%), with a significant difference between the correct rates of lower and upper
case disguised writings for lay people (t = 2.49, df = 100, p = 0.014).
These observations perhaps indicate that the exposure we have to lower case
handwriting leads to a very large pool of writing that we accept as natural. Sciacca, Langlois-
Peter, Gilhodes, Margot and Velay (2008) in their study on handwriting variability under
different postural conditions reported that the mean variability was higher with words
written in lower case than those written in upper case. Furthermore, Sciacca, Langlois-Peter,
Gilhodes, Margot and Velay (2009) and Sciacca, Langlois-Peter, Margot and Velay (2011)
found that the ratio of between-subject variability to within-subject variability was
considerably higher for lower case writings (3.8) than for upper case writings (1.5) for all
writing postures and tablet positions tested, except ‘lying/vertical tablet'. From this research
it can be surmised that upper case writings exhibit less variability (as evidenced by the low
values of both within- and between-subject variability) than lower case writings.
Furthermore, lower case writings display low within-subject variability, but high between-
subject variability. This wide range of variation in naturally written lower case words seems
to support the proposition that we, generally, and FHEs in particular, are primed for a
broader variety of lower case than upper case writings. This might explain why at least some
46
of the attempts at disguise in the current study were thought to fall within the range of
natural writing rather than outside it.
As expected, some samples were easily correctly identified, with no conservatism or
errors made by any of the participants. Figure 4.2 shows a disguised and a naturally written
sample with this response profile.

a b

Figure 4.2. (a) A disguised sample and (b) a naturally written sample that each attracted
100% correct responses from both FHE and lay people groups.

However, the wide range of people’s normal handwriting skill and line quality
complicates the process of determining whether an isolated sample is the result of a natural
or disguise writing process. Figure 4.3 shows two samples of natural writing which display
poor internal consistency, which may be interpreted as evidence of an unnatural writing
process. The handwriting sample in Figure 4.3a attracted a 20% correct response from
FHEs and no correct responses from lay people. The sample depicted in Figure 4.3b also
yielded no correct responses from lay people, while FHEs had a 30% correct rate.

47
Figure 4.3. Two naturally written samples displaying poor internal consistency, which
attracted much lower than the average correct responses compared to all natural handwriting
samples.

4.4 Conclusion
When examining an isolated handwritten text sample, FHEs are not significantly different
from lay people at the task of identifying naturally written and disguised samples. For both
groups the majority of the error detected is associated with opining that disguised samples
are naturally written. This type of error is likely to have resulted from the sample writers’
ineffective disguise strategies and should not be interpreted as an error in the application of
FHEs skills. An analysis of disguise success and its relationship to FHEs’ response rates is
made in Chapter 8. In terms of casework application, this error in process opinions would
not result in an erroneous authorship opinion. Complications may arise, however, if FHEs
have similar problems with detecting simulation behaviour. Their skill in determining
disguise and simulation behaviour in handwritten text samples is covered in the following
chapter.

48
 Chapter 5
Forensic Handwriting Examiners’ Skill in Detecting
Disguise and Simulation Behaviour from
Handwritten Text

5.1 Introduction
Handwriting disguise techniques and the simulation of signatures are discussed in detail in
the standard texts on handwriting examination (Ellen, 1989 pp. 37-45; Harrison, 1958, pp.
349-376, 379-385, 399-407; Hilton, 1982, pp. 168-171, 182-186; Huber & Headrick, 1999,
pp. 279-295; Osborn, 1929, pp. 139-140, 270-308, 396-408). Huber & Headrick (1999, p.
279) characterise the features of disguised handwriting as changes to the pictorial
appearance, while the inconspicuous characteristics are left unchanged. Webb (1977) goes
further and states that most writers will not even realise what characteristics are significant
and which are not; and that some writers apparently find the task of disguise so difficult that
none is attempted at all.
A number of authors have documented the features of disguised writings and how to
establish the naturalness of writing (Cain, 1998; Harrison, 1958; Morris, 2000). The general
principles they suggest to assist in the recognition of disguise behaviour are:
• Disguised handwriting exhibits less fluency and poorer rhythm than the natural
writing, with poor line quality a feature;
• Most disguises are relatively simple (e.g. change in size or slant), and originality in
disguise is rare;
• The internal consistency of the writing is disturbed by the introduction of disguise;
any disguise is rarely consistent over more than a few lines, due to the writer
becoming tired or losing concentration;
• Certain features are rarely disguised. For example word and line spacing, as well as
arrangement on the page usually remain unchanged.

The issue of simulated handwritings, as opposed to signatures, is not well represented

in the literature. However, the key feature of simulated writings is generally accepted to be a
lack of fluency of the line trace, incorporating such things as tremor, awkward or interrupted
movements, pen lifts and blunt endings (Ellen, 1989, p. 50; Harrison, 1958, p. 400; Hilton,
1982, p. 185; Huber & Headrick, 1999, pp. 273, 288, 289; Osborn, 1929, pp. 273-274). As

49
with disguised writings, and expectedly, the writer focuses on the most obvious
characteristics of the model writing such as slant and letterforms, while the more subtle
features, for example proportions and spacing, are usually overlooked (Harrison, 1958, pp.
386, 425; Leung et al., 1993; Muehlberger, 1990).
Both types of un-natural writing, disguised and simulated, are expected to display
similarities with, and differences to, the corresponding genuine natural writing. In the case of
disguised writings (by the genuine writer), the mixed signals arise from forms ingrained in
motor memory, combined with purposefully altered structures associated with the disguise
behaviour. With simulated writings (by someone other than the genuine writer), the
similarities occur from the copying process, while the differences are reflective of the
difficulty in accurately capturing all of characteristics of the subject writing. The hallmark
features noted for simulation relate primarily to the parallel difficulty in suppressing the
learnt motor behaviour for natural handwriting production while attempting to maintain an
unfamiliar writing form, including good registration of structural elements as well as fluency.
Many of the resulting features, including decreased fluency, line quality disturbance and
increased internal inconsistency are said to be present in both simulated (Leung et al., 1993;
Muehlberger, 1990; Osborn, 1929, pp. 273-274) and disguised writings (Cain, 1998;
Harrison, 1958, pp. 352-353, 357-359; Konstantinidis, 1987) which may be reasonably
assumed to impact on the ability of FHEs to detect disguised as compared to simulated
writings.
The aim of the study reported in this chapter was to determine whether participant
FHEs were able to apply their specialised knowledge and skills to a handwriting comparison
task aimed at determining whether a sample of questioned handwriting was the product of
disguise or simulation behaviour. While FHEs’ skill in discriminating between disguised and
forged signatures has been touched upon (Dyer et al., 2008), to the author’s knowledge, the
same has not been undertaken for handwritten text. In this study, FHEs were provided with
100 questioned handwriting samples paired with a naturally written comparison sample to
determine whether the questioned writings were disguised by the comparison writer or
written by another writer attempting to simulate the comparison writer’s handwriting
features.

5.2 Methods
5.2.1 Participants
Results reported here are from FHEs who provided independent opinions on the process of
production of the supplied handwriting samples. Thirty-five FHEs from 14 countries took
part in the trial (seven of these countries do not have English as an official language). All of
50
the participants were authorised by their employers to release opinions regarding the
authorship of questioned handwriting and signatures. At the time of data collection, 30
FHEs were working in government laboratories and five were working privately.

5.2.2 Materials
The trial consisted of 100 pairs of handwritten text, each with a naturally written comparison
sample (100 different writers) and a questioned sample that was either disguised by the
comparison writer or written by another writer attempting to simulate the comparison
writer’s handwriting features. The writings, all of a specific text, were made using one of two
makes of ballpoint pen3 and the same make of white paper. The samples were scanned at
600dpi and laser-printed into a booklet as well as converted into high-resolution PDF files.
For the disguised writings, no directions were given to the handwriting providers as to how
to disguise their handwriting. Fifty-nine writers provided the sixty-six simulated samples
used in the trial. Simulators were provided with target writing of the comparison writer that
was either the same text to be simulated (direct model), or collected/course of business
writings from which they had to create the standard text. In the case of the latter, simulators
were provided with the paragraph to be written laser-printed on a document. Simulated
samples based on these two groups of target writings were either made with or without
practice. A period of one week was allowed for practice. No conditions were set for the
number of attempts to be made prior to providing the final sample. Tracing was not
permitted. Although the models provided to some simulators contained the exact text
included in the trial, the models were not used as comparison samples in the trial. An
analysis of the skill characteristics of FHEs associated with these different simulation
conditions is covered in Chapter 6.

5.2.3 Procedure
Participants were provided with a sample booklet of laser-printed images of the handwriting
samples (see Appendix B for an example of a page from the sample booklet) and a DVD
containing PDF files of the samples. Each participant was given an answer-recording
booklet (see Appendix C for an extract of this) and informed that each sample pair consisted

3 All of the natural and disguised sample pairs were written using the same make and model of

ballpoint pen, since these were collected for the initial trial dealing only with naturally written and
disguised samples (see Chapters 3 and 4). Around half of the natural and simulated sample pairs were
written with the same ballpoint pen, as these were collected on the digitising tablet and required a
proprietary inking pen to be used. The other half of the natural and simulated sample pairs were
written using two makes of ballpoint pen; the natural sample being written in the same make of pen
as the natural and disguised sample pairs and the simulated sample being written in the inking pen
for use with the digitising tablet. The author notes that this is a limitation of this trial.
51
of a naturally written comparison sample and a questioned sample that was either disguised
by the comparison writer or simulated by a different writer attempting to copy the
handwriting features of the comparison writer. Participants were asked to make an
assessment as to whether the questioned sample was disguised or simulated, or if they could
not say. They recorded their response for each of the questioned samples in the answer-
recording booklet by entering a code (D = disguised, S = simulated, or I = inconclusive). If
an inconclusive opinion was given, the participant was required to record an additional
forced opinion (D or S) for the handwriting sample; however, not all participants did so.
The unforced and forced opinions are considered separately in this chapter. In addition to
the process opinion, participants were instructed to record in the answer booklet the
feature(s) they observed which indicated the process when they gave an unforced opinion.
Recording this observation was optional in the case of a forced opinion. A comparison of
the FHE groups’ verbal statements and their scores for disguised and simulated samples is
made in Chapter 9.

5.2.4 Ethics approval

Approval for this study was obtained from the La Trobe University Human Ethics
Committee on the basis that the handwriting providers and test participants gave full
consent for samples of their handwriting or research data provided by them, respectively, to
be included in published material on the condition that their name and any other identifying
information remain confidential.

5.2.5 Analysis
Participants’ responses were marked as correct, misleading, or inconclusive and analysed to
produce scores for each participant and for examiners combined (group scores). In the first
group analysis the initial opinions of all examiners were considered. As four of the
participants responded with only inconclusive opinions this analysis was undertaken to
determine the overall inconclusive rate for the group of examiners (n = 35). In the second
analysis four participants were excluded to provide scores for the group of participants
prepared to offer a decisive opinion on at least one sample. The scores were determined for
this group (n = 31) across all handwriting samples, and for the disguised and simulated
samples independently. In addition, the forced responses following an inconclusive opinion
for the whole group of participants were marked as forced correct or forced misleading. For
the second and third analyses, a series of planned comparisons were undertaken using
independent sample t-tests (two-tail) to determine whether there were differences in scores
between the two handwriting types. Likelihood ratios were calculated for the unforced
52
opinions of the group to determine the probative value of their responses. An additional
analysis of scores was undertaken for the forced responses of the group of four participants
who initially gave only inconclusive opinions on all samples.
In addition, a number of participants provided unsolicited comments and feedback
on aspects of the trial. Due to the insights offered amongst these, selected comments are
discussed.

5.3 Results and Discussion

Table 5.1 provides a summary of the percentage correct, misleading, inconclusive, correct
called and misleading called responses across the group of FHEs who were prepared to
offer an initial opinion other than inconclusive on at least one sample. The scores are
provided for disguised and simulated handwriting samples separately and for all samples
combined (“Overall”). Called scores are calculated by removing the inconclusive opinions
and expressing the score as a percentage.

Table 5.1. Grouped scores for questioned samples overall, and disguised and simulated
samples.
Overall Disguised Simulated
% Correct 47.6 67.8 37.1
% Misleading 21.5 7.2 28.8
% Inconclusive 30.9 25.0 34.0
% Correct Called 68.9 90.4 56.3
% Misleading Called 31.1 9.6 43.7

Overall these examiners were not very confident at the task of determining whether
the questioned samples were disguised or simulated, as evidenced by the high overall
inconclusive rate (38.8%). When they were willing to make a call, they were correct about
the process almost 70% of the time. However, more meaningful results can be gained if we
examine the results for each of the questioned handwriting types separately.
The inconclusive rate for each of the processes is not significantly different (t = 1.40,
df = 60, p = 0.167), but the percentage misleading and correct rates are (for misleading: t =
4.95, df = 60, p < 0.0001; for correct: t = 5.19, df = 60, p < 0.0005). This leads to
significantly different percent correct called and percent misleading called rates (where the
inconclusive responses are removed from the analysis, t = 6.45, df = 60, p < 0.0001).
Although the high percent correct called rate for disguised handwriting samples makes
it appear that FHEs have a skill at detecting disguise, the corresponding high misleading

53
called rate for simulated writings suggests a different story. The high misleading rate for the
simulated samples means that FHEs were often wrongly opining that the sample was
disguised. Thus while the rate for identifying disguised samples, when they were disguised,
was high, the rate for calling a sample disguised when it was not in fact disguised was also
relatively high. If we consider the results in the context of the Bayesian framework (Evett,
1998; Robertson & Vignaux, 1995) and treat the FHE as an instrument, we can determine
the likelihood ratios (and hence probative value) for the group of FHEs’ evidence when
their opinions were that the writing samples were disguised and when their opinions were
that the samples were simulations, respectively.
The likelihood ratios have been calculated below for when FHEs gave an opinion
other than inconclusive (i.e. a called opinion). The author considers this is justified as it can
be argued that it is only in these instances that the FHEs would be proffering evidence that
updates the prior odds of the evidence before the courts.
The likelihood ratio is defined as the probability of observing the evidence given the
proposition is true, divided by the probability of observing the evidence given that a stated
alternative proposition is true. Therefore, the likelihood ratio for the FHE groups’ opinion
that a sample is disguised is given by:
Probability of FHEs’ opinion that the sample is disguised, given that the sample is disguised
/ Probability of FHEs’ opinion that the sample is disguised, given that the sample is
simulated.
This translates to:
% Correct Called for disguised sample / % Misleading Called for simulated samples
= 90.4 / 43.7 = 2.

The likelihood ratio for the FHE groups’ opinion that a sample is simulated is given by:
Probability of FHEs’ opinion that the sample is simulated, given that the sample is simulated
/ Probability of FHEs’ opinion that the sample is simulated, given that the sample
is disguised.
This translates to:
= % Correct Called for simulated samples / % Misleading Called for disguised
samples
= 56.3 / 9.6 = 6.
These likelihood ratios suggest that the strength of the evidence for this group of
FHEs is greater when their opinion is that the sample is simulated than it is when their
opinion is that the sample is disguised. However, both ratios are small, indicating the
strength of the evidence is low.
54
 These results assume that all examiners are behaving similarly. Figure 5.1 shows the
distribution of individual examiner percentage correct called results for disguised and
simulated samples. The scatter plot does not show a clear trend with respect to clustering of
examiners’ responses. However, of the nine examiners who had 100% correct called
responses for disguised samples, five were the poorest performers with respect to simulated
samples. This shows that the tendency for calling questioned samples disguised in this trial is
marked for certain FHEs. In fact, only one of the nine FHEs had a correct called rate above
the 56% overall average for simulated samples. There were four examiners with an average
of about 90% for both writing types. The best performing FHE had greater than 96%
correct called scores for both disguised and simulated samples, with only a slightly higher
than average inconclusive rate of 34%.

Figure 5.1. Scatter plot of individual examiner percentage correct called scores for disguised
and simulated handwriting samples.

The distribution of unforced scores for each disguised and simulated sample is shown
in Figures 5.2 and 5.3 respectively. As can be seen, and gleaned from Table 5.1, the majority
of disguised samples attracted a large number of correct responses, and a small number of
misleading responses. However, the simulated samples do not display such stratification of
responses across the group. Some simulated samples (such as the example shown in Figure
5.6, represented as sample 65 and indicated by a dashed line in the Figure 5.3 scatter plot)

55
have a response profile similar to the majority of the disguised samples with a high correct
response rate and a small number of misleading responses. Others, such as samples 16 and
52 (indicated by dash and dot lines in Figure 5.3), exhibit a reversed response profile.
Among the simulated samples, there are some that have attracted the same, or almost the
same, number of correct and misleading responses. This indicates that the predictor features
that FHEs use to recognise simulated writings may be flawed.

Figure 5.2. Distribution of correct (C), misleading (M) and inconclusive (I) scores for each
disguised sample.

The analysis above is based on traditional comparison methods and suggests that
FHEs, as a group, do not have a particular skill in determining disguised versus simulated
samples. However, a likelihood ratio approach may be taken to assess the probative value of
individual examiners’ opinions. Figure 5.4 gives the likelihood ratio of each examiner for
both disguised and simulated samples, calculated in the same way as the likelihood ratio for
the group (above). Note for those examiners with a percent misleading called rate of zero,
1% was substituted as the denominator for the likelihood ratio calculation.

56
Figure 5.3. Distribution of correct (C), misleading (M) and inconclusive (I) scores for each
simulated sample.

Figure 5.4. Likelihood ratios per examiner for disguised (LR DIS) and simulated (LR SIM)
samples. Note that examiner numbers are arbitrary and do not correspond to the examiner
numbers shown in Figure 5.1.

It is interesting that while the strength of the evidence of the majority of examiners is
low (likelihood ratios of 10 or less), there are some examiners whose opinions provide
evidence of relatively high probative value. This suggests that some examiners do have a skill
57
at determining between disguised and simulated handwriting behaviours. Identification of
these examiners and the specific methods they employ could result in the development of a
method to more reliably detect disguise and simulation behaviour, increasing the value of
FHEs’ evidence of this type.
With the aim to see whether reducing FHEs’ conservatism improved correct rates,
forced responses were analysed. Table 5.2 shows the percentage of forced opinions
expressed that were correct and misleading, for disguised and simulated samples. Note that
these do not add up to 100% as for some samples some participants did not provide a
forced opinion as instructed.

Table 5.2. Grouped correct and misleading forced scores for disguised and simulated
samples.
Disguised Simulated
% Forced Correct 77.8 58.7
% Forced Misleading 19.5 38.8

As with the unforced opinions, significant differences were found between disguised
and simulated samples with regard to the percentage of forced opinions that were correct (t
= 5.68, df = 98, p < 0.0001) and misleading (t = 5.74, df = 98, p < 0.0001). The disguised
samples again attracted a higher percentage of correct opinions. Interestingly, while both
handwriting types display a significantly higher percentage of forced correct responses than
unforced correct responses, the misleading responses are also significantly higher for the
forced opinions compared to unforced opinions (see Table 5.3).

Table 5.3. t-Test results for unforced versus forced correct and misleading scores for
disguised and simulated writing types.
Unforced versus forced scores Disguised Simulated
Correct t = 5.28, df = 66, p < 0.0001 t = 10.63, df = 130, < 0.0001
Misleading t = 4.81, df = 66, p < 0.0001 t = 5.62, df = 130, < 0.0001

The combined unforced and forced correct score taken across the participant group is
significantly higher than the unforced correct score alone for both the disguised and
simulated samples. However, the combined unforced and forced misleading score is also
significantly higher than the unforced misleading score alone (see Table 5.4; the unforced
scores given in this table are different to those given in Table 5.1 due to the inclusion here
of the four participants whose initial responses were wholly inconclusive). This means that
58
by declining to give an opinion, FHEs are obtaining a smaller misleading rate than they
otherwise might.

Table 5.4. Grouped correct and misleading unforced and combined forced and unforced
scores, and t-test results for these scores, for disguised and simulated samples.
Disguised Simulated
% Correct unforced 60.1 32.9
% Correct combined 85.5 57.0
t = 9.72, df = 66, p < 0.0001 t = 10.55, df = 130, p < 0.0001
% Misleading unforced 6.4 25.5
% Misleading combined 13.5 42.0
t = 4.02, df = 66, p < 0.0001 t = 7.62, df = 130, p < 0.0001

In order to determine if the FHEs who initially offered only inconclusive opinions
performed differently, when they were forced to give an opinion, than the other FHEs, an
analysis was made of their grouped forced scores. Table 5.5 shows the percentage forced
opinions that were correct and misleading for disguised and simulated samples for both the
group of four FHEs who gave wholly inconclusive opinions to begin with and the
remainder of the FHE group. Note that these do not add up to 100% as some participants
did not provide a forced opinion for every sample as instructed. No significant differences
were found between the ‘all inconclusive’ group and the remainder of the FHE group for
the percentage forced correct or misleading scores for either writing type. However, this test
for significance may be affected by the difference in size of the two groups being compared.
Statistical analysis of the percentage forced correct and forced misleading scores of the ‘all
inconclusive’ group against the percentage correct and misleading scores of the remainder of
the FHE group (given in Table 5.1) again shows no significant difference between the two
groups of FHEs and writing types (see Table 5.6), despite all percentage forced responses
being higher than the corresponding unforced responses.
The results reported here must reflect on the assumptions that FHEs make regarding
what these un-natural writing types should look like. But what are FHE assumptions based
on? As already mentioned, the literature reported features of these writing types overlap but
there is little description of how they may vary between writing behaviours. In addition, in
the author’s experience, and as borne out by the relative scarcity of literature references,
casework may not often present text-based writings that are thought to be simulated,
whereas samples that are disguised, particularly request comparison samples, are more
common.

59
Table 5.5. Grouped forced correct and forced misleading scores for disguised and simulated
writing types, for participants who initially gave all inconclusive opinions and the remainder
of the FHE group, and the corresponding t-test results.
Disguised Simulated
% Correct % Misleading % Correct % Misleading
Forced Forced Forced Forced
All
83.1 15.4 54.9 43.6
Inconclusive
Other FHEs 75.8 23.6 54.4 45.2
t = 0.93, df = 27, t = 0.99, df = 27, t = 0.05, df = 28, t = 0.15, df = 28,
t-Test Result p = 0.359 p = 0.331 p = 0.961 p = 0.881

Table 5.6. t-Test results for forced scores of FHEs initially offering only inconclusive
opinions versus unforced scores of FHEs initially willing to offer an opinion on at least one
sample, for disguised and simulated writing types.
Forced ‘all inconclusive’ versus
Disguised Simulated
unforced ‘other FHEs’ scores
Correct t = 1.31, df = 33, p = 0.199 t = 1.46, df = 33, p = 0.155
Misleading t = 1.70, df = 33, p =0.098 t = 1.27, df = 33, p = 0.212

It may be surmised that if a questioned handwriting displays poor fluency of the line
trace with subtle or inconspicuous similarities to the natural writing, then the questioned
writing may be disguised. If a lack of fluency is associated with differences in subtle features,
then the questioned writing may be simulated. Figure 5.5 shows portions of a handwriting
sample pair where the questioned sample fits well within the expected features of disguise,
with similarities in formation and connectivity, and a change in slant that tended toward the
natural slope as writing progressed. This questioned disguised sample posed comparatively
little problem for FHEs, attracting no errors, 87% correct and 13% inconclusive responses.
Similarly, Figure 5.6 illustrates portions of a handwriting sample pair where the
questioned sample fits well within the expected features of simulation: imitation of obvious
features along with subtle differences in the execution or consistency of these, signs of
dysfluency, and differences in proportions and spacing. This questioned simulated sample
attracted 61% correct, 6% misleading (the smallest of all the simulated samples) and 32%
inconclusive responses.

60
 a b

Figure 5.5. Portions of handwriting sample pair where the questioned disguised sample
exhibits some of the expected features of disguised writing. (a) Natural handwriting and (b)
disguised handwriting.

a b

Figure 5.6. Portions of handwriting sample pair where the questioned simulated sample
exhibits some of the expected features of simulated writing. (a) Natural handwriting and (b)
simulated handwriting.

The assessment of unnatural writing processes is not always so straightforward,

however. Due to the dynamic and skilled nature of handwriting production, other issues
must be taken into consideration, including: the skill of the comparison writer for disguising
their handwriting, the complexity of the target handwriting, the skill of the simulator for
successfully copying the features of the target writer, the effect of practising, the impact of
having (or not having) the exact text to copy, and the associated differences in variation
when comparison writing comes from course of business writings.
Muehlberger (1990) asserts that “to be considered a simulation, a questioned signature
must bear a striking resemblance to a genuine signature…” If this can be applied to
simulated handwritten text also, this criterion may mean that simulated writings were not
identified as simulations due to their divergence from the target writing. Figure 5.7 shows a
section of a handwriting pair where the questioned sample is simulated. It displays obvious
dysfluency, but does not pictorially resemble the comparison sample so if using the

61
reasoning of Muehlberger’s for signatures it may not be considered simulated. However, in
addition to the obvious differences, there are subtle differences (rather than similarities) too,
in size, spacing and formation of almost every letter, which may lead the FHE to the
opinion that the sample is in fact a ‘simulation’. The pictorial and subtle differences between
the comparison and questioned samples made FHEs uncertain about the writing process,
with this simulated sample drawing 52% inconclusive responses and a lower than average
correct score (26%).

a b

Figure 5.7. Portions of handwriting sample pair where the questioned simulated sample
does not pictorially resemble the target (normal) handwriting. (a) Natural handwriting and
(b) simulated handwriting.

In addition to the difficulties in accurately assessing the process of un-natural

handwriting arising from aspects related to handwriting production, there are other factors
that may have impacted on the results of the trial. The time FHE participants took to
complete the task might have been less than the time they would normally spend in forming
opinions on writings in actual casework. The arduous task of examining one hundred pairs
of handwriting may have resulted in fatigue or changed the normal approach taken by
examiners. We have no evidence that these were factors influencing the results but they
cannot be ruled out. Familiarity of FHEs with copybook models and indeed with
handwritten English words could have been an issue for some examiners whose first
language is not English and who work in countries where English is not the official
language. However, for the group of thirteen such examiners their overall correct called rate
(73.9%), misleading called rate (26.1%) and inconclusive rate (26%) were comparable if not
better than the full group of examiners.
A combination of factors, including the skill of the simulator for the task, and FHEs
preconceptions about the appearance of un-natural writing types, may have led to the high
rate of error (71% misleading called, with a lower than average inconclusive rate: 23%)
62
associated with the questioned simulated sample illustrated in Figure 5.8. Many examiners
commented on the close similarities between the comparison and questioned samples, but
thought they were indicative of a poor disguise rather than a good simulation, despite the
presence of subtle differences in some letter forms (for example the length of the tail of e,
direction of the tick at the base of T, proportions of w).

a b

Figure 5.8. Portions of handwriting sample pair where the questioned simulated sample was
mistaken for disguised writing by 71% of participants who chose to give an opinion. (a)
Natural handwriting and (b) simulated handwriting.

These results indicate that FHEs’ expectations of the predictor features of simulated
and disguised writings may be flawed. Insight into this may be obtained after considering
some of the questions and comments received by the researcher from participants, who
presumably felt that the answers would assist in their deliberations regarding the handwriting
behaviour used to create the questioned samples. Questions included:
• What was the education of the writers? Hull (1993) has provided some evidence of a
positive correlation between formal education of the writer and the number of
changes introduced in disguised handwriting. Hull’s study also found that certain
disguise techniques, including changes to special characters, speed and use of space,
increased in frequency with increasing years of formal education. Participants may
have felt that the success of disguise (and simulation) techniques would be increased,
and therefore potentially more difficult to determine, if the writers possessed a high
education level.
• What instructions were given to the writers and what were the writing conditions? It is generally
accepted (Alford, 1970; Cain, 1998; Harrison, 1958; Morris, 2000) that most writers
employ relatively simple disguise methods. If writers in this study were given
instructions about how to successfully disguise their writing, participants might have
perceived the task at hand as being of greater difficulty. With regard to simulation
writing conditions, the use of a direct model versus collected writings of the target
63
 writer were employed for different forgers. As the effect that these conditions may
have on FHEs’ opinions was one of the areas being investigated (see Chapter 6),
participants were not provided with any information relating to these conditions.
• Was practice allowed and if so, how were the trial samples selected? Although some simulators
were allowed to practise while others were not, the effect of this was again being
explored (reported in Chapter 6) and therefore details were not disclosed to
participants. However, a study of practised signature simulations (Radley, 2005)
found that practice resulted in only slight improvements in the quality of
simulations, and in many cases the practice only served to establish incorrect features
in the forger’s simulations. Furthermore, in relation to practised disguised signatures,
Michel (1978) found no relevant differences between first disguised signatures and
those made after practice.
• What data was hoped to be collected: FHEs ability to identify simulation vs. disguise; or the general
characteristics of each type of writing?

This last question is interesting, as in at least one participant’s mind, the approach they
used would be different for each. They stated that the first task requires detailed analysis and
comparison, while the second would be a first-impression/classic indicators examination.
The participant wrote “There were instances during my trial when I thought that the writing
was classically indicative of simulation yet contained elements of disguise.” This indicates
that the ‘classic indicators’ of disguise and simulation may not relate well to the actual
process used by the writer, and that there is some doubt that those features referred to in the
literature will lead to the correct process, and therefore authorship, opinion.
Comments from another participant showed that at least one participant approached
the task in the opposite way than the authors suggest examinations should be undertaken,
with a first stage process opinion leading to the second stage authorship opinion. In
unsolicited feedback from this participant, they stated that they first determined if the
questioned sample was written by the comparison writer; if evidence supporting that
proposition was observed then they gave the opinion that the sample was disguised. If
evidence of a different writer was observed they gave the opinion that the sample was
simulated. However, the exact process by which they made the determination of authorship
is not clear, so perhaps if subtle similarities were observed this was seen as evidence
supporting the proposition that the two samples were written by a common writer, and
hence the questioned sample was opined to be disguised.

64
5.4 Conclusion
This chapter does not take into account questioned natural handwritings. It was found in
Chapter 4 that when faced with pairs of handwriting by an individual, with one sample
naturally written and the other disguised, FHEs offering an opinion as to the process were
correct 95.7% of the time. The results of this chapter, however, show that forensic
handwriting examiners have a tendency to call questioned un-natural handwriting samples
disguised (as opposed to simulated), which indicates that their expectations of the predictor
features of these un-natural writing behaviours is not accurate. Exposure to known disguised
and simulated writings will broaden FHEs’ knowledge of features associated with each of
these writings types, adding to the theoretical aspects of un-natural handwriting behaviours.
However, due to the questionable skill of FHEs in detecting disguise versus simulated
handwriting, if an un-natural handwriting process is suspected, it would be advisable to
proceed to an authorship opinion with caution.

65
 Chapter 6
Skill Characteristics of Forensic Handwriting
Examiners Associated with Simulated Handwritten Text

6.1 Introduction
In the previous chapter, some questions received from FHE participants were briefly
considered. These questions related to writing conditions for obtaining the disguised and
simulated writing samples used in the task that they undertook. A presumption can be made
that by requesting information about certain conditions, FHEs might believe that the
variation of these conditions could affect their ability to successfully discriminate between
disguised and simulated handwriting samples.
There was no variation in the conditions for collecting disguised samples (aside from
the pen, as noted in Chapter 5 footnote 3). However, two bimodal conditions were
introduced for the simulated writings: practice and a direct model. This chapter directly
addresses the impact of practice and the use of a direct model for simulation on FHEs’
ability to detect simulated handwriting. Although the issue of practice has been explored in
relation to signature simulation (Radley, 2005) and disguise (Michel, 1978), to the author’s
knowledge there is no published empirical data on the effect of simulation practice on the
success of extended handwritten text. Neither have there been reported analyses on the use
of a direct model versus collected handwritings as the target for simulated writings.
The purpose of this study was to further investigate examiners’ skill at assessing the
process of production of unnatural writings and focuses on any effects that might arise
when FHEs are attempting to discriminate between disguise and simulation behaviours
resulting from the strategy used by simulators in copying writing samples.

6.2 Methods
The participants, materials, procedure and ethics approval for this study were the same as
those given in Chapter 5. The results reported here are based on a different analysis of the
same data dealt with in the previous chapter.

6.2.1 Analysis
Participants’ responses were marked as correct, misleading or inconclusive, and analysed to
produce scores. Four of the participants responded with only inconclusive opinions and
their data was excluded from this analysis. For all participants, the forced responses

66
following an inconclusive opinion were further marked as forced correct or forced
misleading. Since the same samples were examined by the group of FHEs, and since we are
interested here in aspects of the handwriting itself, we aggregated over examiners. To
determine whether there were differences in scores between the different conditions 2x2
analyses of variance (ANOVA; IBM SPSS Statistics version 19) were performed. In addition,
for each questioned handwriting sample the proportion of correct called and inconclusive
opinions was computed, and these were used as dependent variables in the ANOVA.
Similar analyses were also performed using a Generalised Mixed Model approach that takes
into account the dichotomous nature of the responses, and the clustering of FHEs’ opinions
within each sample4. These analyses resulted in similar conclusions as the 2x2 ANOVA.
It is noted that assistance with statistical analysis in this chapter was provided by Dr
Reinoud Stoel, Team Leader, Forensic Statistics, Netherlands Forensic Institute.

6.3 Results and Discussion

There were a total of 66 simulated samples in the trial. These fell into four groups based on
the conditions under which they were created:
1. Direct model of the target handwriting with practice time. The direct model
condition is one where the simulator was provided with handwritten text from the
target writer that was exactly the text that they were to simulate;
2. Direct model of target handwriting without practice time;
3. No direct model of target handwriting with practice time. Collected samples of
normal handwritten text written by the target writer during their normal life activities
were provided to the simulator (and were not therefore exactly the same text as the
simulator was required to generate);
4. No direct model of target handwriting without practice time.

Table 6.1 gives the number of samples, number of FHEs, and the number of conclusive
opinions expressed in each of the simulation condition groups.

Table 6.1. Number of samples, FHEs and conclusive opinions expressed in the different
simulation condition groups.
Simulation
Practice No practice
condition
Model nsamples=18; nFHE=31; nopinions=387 nsamples=17; nFHE=29; nopinions=352
No Model nsamples=13 nFHE=31; nopinions=252 nsamples=18; nFHE=30; nopinions=359

4Tabachnick and Fidell (2007a, 2007b) and Gelman and Hill (2007) can be referred to for detailed
descriptions of the applied techniques.
67
 An overview of the percentage correct called responses (calculated after having
removed the inconclusive opinions) in each of the four conditions is provided in Table 6.2.
The percentages of inconclusive opinions for all examiners are presented in Table 6.3.

Table 6.2. Percentage correct called opinions for each condition.

Simulation condition Practice No practice Total
Model 55.7 52.7 54.3
No Model 49.8 63.4 57.7
Total 53.2 58.2 55.9

Table 6.3. Percentage inconclusive opinions for each condition.

Simulation condition Practice No practice Total
Model 30.7 33.2 31.9
No Model 37.5 35.7 36.4
Total 33.5 34.5 34.0

With respect to FHE opinions on the handwriting samples, the results show a small,
though significant interaction between model and practice conditions (F62, 1 = 4.28, p <
0.05). Although the differences are relatively small, the highest correct scores are obtained in
the ‘no practice, no model’ condition, while the lowest correct scores are obtained in the
‘practice, no model’ condition. It could be hypothesised that the ‘no practice, no model’
condition, giving the simulator no time to become accustomed to the features of the target
writing and requiring they search out each individual letter one at a time to build the
simulated sample, would result in the highest cognitive load. Of all the task conditions, this
would be the most difficult and result in the poorest simulation attempts. This would enable
the simulation behaviour to be easily recognised and lead to the highest correct scores
amongst the FHEs. However, by this logic, the ‘practice, model’ condition ought to be the
easiest simulation task and we would expect that condition to attract the lowest correct
scores, rather than the observed ‘practice, no model’ condition. With respect to the
proportion of inconclusive opinions the results show a small but significant main effect for
model (F62, 1 = 4.97, p < 0.05), with the highest proportion of inconclusive opinions obtained
in the ‘no model’ condition. This may be explained by the difficulty of the task of locating
the relevant letter in the ‘no model’ condition, which may lead to instances of certain letters
that do not closely resemble the target writing. The mixed signals of similarities and
dissimilarities in the sample writing can be expected to increase the uncertainty of examiners’
opinions.

68
 While the differences between the conditions are relatively small, the results do clearly
show that, as a group, FHEs’ are not skilled in the task of identifying whether a questioned
sample was simulated rather than disguised. This is evidenced by the high percentage
misleading called scores (in total 44%). The high percentage inconclusive scores (in total
34%) also show that FHEs, in this trial, were not confident in carrying out the task.
Examiners’ forced opinions may reveal whether their conservatism was well-founded.
The percentage forced correct and misleading opinions for each of the conditions are
provided in Tables 6.4 and 6.5 respectively. These tables are dependent, but it should be
noted that, in some instances, these do not add up to 100%. This is due to some participants
neglecting to provide a forced opinion for some handwriting samples. The differences in
percentages forced correct (and forced misleading) between the different conditions are not
significant. There is no significant interaction between model and practice (F62, 1 = 0.001, p =
0.98), and no main effects for model and practice (F62, 1 = 0.41, p = 0.52 and F62, 1 = 0.59, p =
0.45, respectively).

Table 6.4. Percentage forced correct opinions for each condition.

Simulation condition Practice No practice Total
Model 58.3 61.5 59.8
No Model 55.8 58.8 57.5
Total 57.2 60.1 58.7

Table 6.5. Percentage forced misleading opinions for each condition.

Simulation condition Practice No practice Total
Model 38.6 36.1 37.3
No Model 42.0 39.2 40.4
Total 40.0 37.7 38.8

It appears that when some FHEs are forced to offer an opinion on handwriting samples
where they were initially unwilling to, they were able to correctly associate handwriting
features with simulation behaviour rather than disguise (see Figures 6.1 and 6. 2 for an
illustration of the interaction between model, practice and forced/unforced responses, F62, 1
= 5.61, p < 0.02). While the error rates for forced opinions are lower than the unforced
error rates in three of the four simulation condition pairs, the occurrence of any error in
forced responses results in an overall increased error rate. The initial conservatism of the
FHE group was therefore justified.

69
Figure 6.1. Interaction of unforced correct called and forced correct responses and the
practice conditions for when model was used.

Figure 6.2. Interaction of unforced correct called and forced correct responses and the
practice conditions for when no model was used.

Further insight into the nature of FHE skill may be gained by investigating individual
examiner profiles and taking into account the ‘quality’ of the simulations. This is touched
upon later in this section, and a treatment along these lines for disguised writings is
undertaken in Chapter 8.
The task of copying the exact text of the target handwriting in the form of a direct
model could be assumed to be a far more straightforward task than having to search out the
individual letterforms or words from a collection of course of business writings. Simulators,
having to construct text by searching for appropriate characters or words amongst collected
70
writings, may produce samples that are different to those produced under the direct model
condition. Writings of this type may provide features to FHEs that may or may not assist
them in making the determination that the sample is indeed simulated (rather than
disguised). However, regardless of simulation strategy, FHEs performance remained poor as
characterised by the high percentage misleading called and percentage inconclusive scores.
The slightly lower inconclusive scores for simulations executed using a direct model suggest
that FHEs were more confident in expressing opinions on these samples. The lack of
differences between the correct and misleading called scores shows that this confidence does
not translate into an increased skill. On the contrary, it implies that the predictor features
FHEs used to form opinions on the un-natural writing samples were not effective.
The uncertainty and error associated with simulated samples across all simulation
conditions may stem from a number of sources. Poorly performed simulations that do not
resemble the comparison handwriting may lead the FHE towards the opinion that the
questioned sample was disguised. However, poorly and skilfully performed simulations alike
are liable to exhibit subtle dissimilarities, which will tend to point towards a writer other than
the comparison writer (i.e. simulation). Simulations that display both gross and subtle
dissimilarities provide a mixed signal and presumably result in inconclusive opinions.
For the purpose of illustrating different profiles of FHE opinions, Figures 6.3 and 6.4
depict portions of two of the trial sample pairs. The questioned simulated sample (Figure
6.3b), was created without a direct model but with practice and displays both gross and
subtle dissimilarities. This sample attracted high inconclusive (54.8%) and misleading called
(71.4%) scores. Figure 6.4b shows a simulated sample that was written using a direct model
but no practice. The presence of similarities in gross features with subtle dissimilarities lead
to a high correct called score (90.5%), with a moderate inconclusive score (32.3%).

71
 a
b

Figure 6.3. (a) Naturally written comparison sample and (b) questioned simulated sample
(no model, practice) that attracted high inconclusive (54.8%) and misleading called (71.4%)
scores.

a
b

Figure 6.4. (a) Naturally written comparison sample and (b) questioned simulated sample
(model, no practice) that attracted a moderate inconclusive score (32.3%) and a high correct
called score (90.5%).

72
 The performance of FHEs was further investigated by undertaking exploratory
analyses using an Item-Response Model (e.g. Embretson & Reise, 2000). Figure 6.5 shows
the results of this analysis, where each line represents a different expert. The ‘difficulty’ of
the handwriting sample (as determined by the proportion of correct responses per sample) is
given on the x- axis while the y-axis represents the probability of giving a correct response.
As can be seen, for some experts the lines increase toward the right, showing that
these FHEs have a tendency to provide correct opinions as the samples become more
difficult, while for other FHEs the lines decrease, showing that these FHEs tend to provide
correct opinions only when the samples are easy. So, there appear to be different groups of
experts that, within group, respond in a similar way. This could be the topic of future
research into whether subgroups of FHEs can be identified that have a similar response
pattern, and how the skill characteristics of subgroups might be explained. The results of
this could be valuable information for casework, since it could be used to select FHEs for
specific cases (i.e. for selecting FHEs that have complementary expertise) or potentially be
used in FHE training.

Figure 6.5. Probability, for each participant, of providing a correct opinion on questioned
simulated handwriting samples versus the ‘difficulty’ of the questioned sample.

None of the FHEs tested have a profile showing they are equally good at providing
correct responses regardless of the difficultly level of the questioned simulated sample. This
suggests that none of the methods employed by FHEs for determining whether a
questioned unnaturally written sample is disguised or simulated works well for both easy and
difficult simulated samples.

73
6.4 Conclusion
The results of this study provide further evidence that unnatural writing behaviours are
problematic for FHEs. Their ability to discriminate between simulated and disguised
writings using traditional methodological approaches is poor as characterised by high
inconclusive and misleading scores. Given the results of prior research (Found & Rogers,
2005a, 2008) and the findings reported in Chapters 3, 4 and 5, the data presented here
provides further evidence that when it is opined that a questioned handwriting sample is
unnaturally written, and where both disguise and simulation propositions are being
considered, there is still no empirical evidence that would support a second-stage authorship
opinion being expressed. It is the case that research of the type described here does largely
rely on data from FHEs that is grouped. It may be that some individual FHEs have superior
skills to their colleagues at the tasks investigated in this study. The challenge is, however,
how to best identify who these FHEs are and what the character of their superior skill might
be.

74
 Chapter 7
Dynamic Features of Naturally Written, Disguised and
Simulated Handwritten Text

7.1 Introduction
In recent years, attempts have been to increase the empirical basis of forensic handwriting
examination by bringing together allied fields such as human motor behaviour (e.g. Found,
1997; Hecker, 1993; Phillips, Noutsis, Hughes & Rogers, 2000; Thomassen & van Galen,
1997; van Galen & van Gemmert, 1996; van Gemmert & van Galen, 1996; van Gemmert et
al., 1996) and automatic handwriting analysis (e.g. Chapran et al., 2008; Estabrooks, 2000;
Franke & Grube, 1998; Franke, Schomaker, Vuurpijl & Giesler, 2003; Guest, Fairhurst,
Abreu & Linnell, 2011; Liwicki et al., 2011; Srihari & Leedham, 2003). Thomassen & van
Galen (1997) have outlined how dynamic features can be used to discriminate between
writers, and although forensic handwriting examinations are made on static images, dynamic
information, such as pen pressure, fluency, and velocity, may be inferred and can impact on
opinions regarding handwriting processes. Assessing the process of production of
handwriting (naturally written, disguised or simulated) is an important step in forensic
examination methodology. As mentioned in Chapter 1, recent research on disguised and
simulated writings has identified these as problem writing types for FHEs (Found and
Rogers, 2005a, 2008). These unnatural writing types are reported to attract higher incorrect
and inconclusive rates on authorship opinions than genuine writing does. If FHEs can
identify a sample of questioned writing as the product of an unnatural (disguised or
simulated) writing process, this can be reported as a ‘‘first-stage’’ opinion. The validity of
proceeding on to express an opinion as to authorship (a ‘‘second-stage’’ opinion) in the
absence of validated predictors remains questionable. In Chapter 3 it was found that FHEs
do have a skill over lay people in distinguishing between natural and disguised handwriting
behaviours when presented as handwriting sample pairs (Bird, Found & Rogers, 2010).
However, to date, there has been no published empirical data on the skill of FHEs in
discriminating between simulated and disguised handwritten text. This chapter investigates
dynamic differences between naturally written, disguised and simulated handwritten text
made by eleven sets of writers, with the aim to determine whether any objective dynamic
measures exist which may be inferred from static images of handwriting, and hence used to
discriminate between disguised and simulated handwriting behaviour.

75
 Harrison (1958, pp. 349-372) and Cain (1998) have studied the features of disguised
writings, and how the naturalness of handwriting might be established. They conclude that
most disguised handwriting is characterised by changes to gross features such as slope, size
and letter design while subtle features such as word and line spacing remain unchanged. The
additional cognitive load required to suppress the natural abstract motor programs also
results in disguised handwriting that may be written less fluently and with poorer line quality
than natural handwriting. Publications dealing with the simulation of extended writings are
few, and are predominately related to high profile cases (Cabanne, 1974; Michel & Baier,
1985; Nickell, 1997). However, the key feature of simulated writings is generally accepted to
be the lack of fluency of the line trace, with a slow, drawn quality and interrupted
movements. Analogous to disguised writings, the writer focuses on the most obvious
features of the model, such as size, slant and some letter forms while the subtle features,
such as proportions and spacing, are usually overlooked (Muehlberger, 1990; Leung et al.,
1993; Harrison, 1958, p. 365).
Some of the generally accepted features of the static image of disguised and simulated
writings are supported by empirical research investigating dynamic features. Van Galen and
van Gemmert (1996) reported slower movement velocities and more dysfluencies in
simulated samples, comprising a six-word sentence or individual words, compared to the
simulators’ own normal handwriting. They also report that pen pressure is increased under
stress conditions (such as could be expected when disguising or simulating a handwriting
sample), and less variable in the writing trace, but Sita and Rogers (1999) found that
simulators changed their writing pressure toward that of their model writer. Slower
movement velocities have also been reported for disguised writings (van Gemmert et al.,
1996). However, to the author’s knowledge, there is currently no published comparison of
the dynamic features of disguised versus simulated handwritten text.
The results reported here attempt to bridge that gap, investigating whether
significant differences exist between naturally written, disguised and simulated handwritten
text in terms of the dynamic features of velocity, jerk (a measure of dysfluency)5, pen
pressure, duration and pen down duration. In the forensic environment velocity, fluency and
pen pressure may be estimated from the static images. Therefore, any empirical differences
observed between naturally written, disguised and simulated handwritten text, in addition to
being relevant to dynamic automatic handwriting analysis, may potentially be used in static
analysis by automated methods, as well as by FHEs to aid in their assessment of the
handwriting behaviour employed when a piece of writing was created.

5
Jerk is the third derivative of movement coordinates i.e. rate of change of acceleration.
Units: cm/s3 (SI units m/s3).
76
7.2 Methods
7.2.1 Participants and procedure
Eleven subjects (64% female, 91% right-handed, with a mean age of 30 years) volunteered
to provide natural and disguised handwriting samples of specific text, consisting of 87
words, in a lowercase style. No directions were given to the handwriting providers as to
how to disguise their handwriting. A portion of one writer’s natural (a) and disguised (b)
samples are shown in Figure 7.1.

a b

c
Figure 7.1. Portion of one writer’s (a) natural handwriting sample and (b) disguised
handwriting sample. Portion of the (c) simulated sample of the ‘a’ target writer’s
handwriting.

Another eleven subjects (36% female, 82% right-handed, with a mean age of 36 years)
volunteered to provide a simulation of one of the initial writers’ handwriting features, again
writing the same specific text. When natural writing was requested, the specific text was
provided in a laser printed form. For simulators, the model to be imitated was presented
either as a direct model (i.e. the same text written as was to be simulated), or as
collected/course of business writings from which they had to create the standard text, with
the text to write provided in a laser printed form. Tracing was not permitted. None of the
simulators simulated the handwriting of a person they were familiar with. A portion of the
simulated sample of the Figure 1 target writer’s natural handwriting is provided in Figure
7.1(c).

77
 Two natural samples were collected from ten of the eleven writers, with the remaining
writer providing one natural sample. One disguised and one simulated sample were
collected for each writer. The same brand of white 80gsm copy paper and the same Wacom
Intuos3 inking pen (model ZP-130) were used for sample collection. Each sample was taken
with the paper positioned over a Wacom Intuos3 digitising tablet (model PTZ-630)
sampling at 200Hz and providing 0.0005cm resolution. The tablet was placed on a
horizontal table and was able to be repositioned by the writer to provide a comfortable
writing position.

7.2.2 Ethics approval

Ethics approval for this study was obtained from the La Trobe University Human Ethics
Committee on the basis that the handwriting providers gave full consent for samples of
their handwriting to be included in published material, on the condition that neither their
name nor any other identifying information be used.

7.2.3 Data analysis

Dynamic data from each handwriting sample was collected using MovAlyzeR® software
v5.9. The data from the MovAlyzeR® software were exported to Microsoft Excel and
analysed statistically with SPSS v17.0 (SPSS Inc). Data sets were grouped by natural writer,
i.e. writer 1’s natural sample(s), writer 1’s disguised sample and writer A’s simulation attempt
of writer 1’s natural handwriting features.
In Section 7.3, the data were averaged over each handwriting sample in full, and
compared. A 4-way multivariate ANOVA was performed, using duration, relative pen down
duration, average absolute velocity, and average pen pressure as dependent variables,
handwriting type as a fixed factor (three levels – Natural, Disguised and Simulated) and the
individual writers as covariates.
In Section 7.4, the data relating to individual words repeated a number of times
throughout each handwriting sample was analysed and compared. A linear mixed-effect
maximum likelihood model was applied, using duration, relative pen down duration,
average absolute velocity, normalised jerk and average pen pressure as dependent variables,
handwriting type as a fixed factor (three levels – Natural, Disguised and Simulated) and the
individual writers as covariates. This type of analysis was undertaken to take into account the
unbalanced data set (natural writing had twice as much data as disguised and simulated
writing types), and random factors associated with how the data was collected, including the
different words, the order the words were written, and subject/word interaction.

78
 It is noted that assistance with statistical analyses in this chapter was provided by Dr
Kaye Ballantyne, Senior Research and Development Officer, Biology Division, Victoria
Police Forensic Services Department.

7.3 Data Averaged Across Handwriting Samples in Full

For the following statistical analyses, dynamic data relating to four features6 (duration,
relative pen down duration, average absolute velocity, and average pen pressure) of all words
within the individual handwriting samples were averaged over the entire handwriting sample.
The MovAlyzeR® software can extract movements of multiple-word trials into individual
words. It does this by testing for a new word every time there is a pen-up event. When this
occurs, a number of word extraction parameters are evaluated to determine if a new word
has started. The settings used for these parameters are given in Table 7.1. Since the values of
these parameters may be expected to vary with different handwriting behaviours, if the
values are kept constant for all handwriting samples the accuracy of the word extraction will
be affected. Generally, the simulated and disguised samples were written more slowly than
the natural samples, so a different set of values was used for natural versus disguised and
simulated samples, in order to optimise word extraction accuracy across the groups.

Table 7.1. MovAlyzeR® word extraction settings used.

Setting
Parameter Setting (Natural) (Disguised & Units
Simulated)
Minimum word width 0.3 0.1 cm
Minimum leftward distance between
0.3 0.5 cm
words
Minimum leftward penlift movement
0.5 0.5 cm
between words
Minimum downward distance between
1 5 cm
word beginnings
Minimum duration of penlifts between
0.6 1.5 s
words
Maximum duration of penlifts within a
0.1 0.15 s
word

7.3.1 Results
Handwriting type and writer set factors were examined to determine their effect, if any, on
the dynamic features of interest. Using Pillai’s trace with partial η2, it was found that both
the handwriting type (F12, 11703 = 109.7, p < 1x10-36, η2 = 0.101) and writer (F4, 3901 = 9.778, p

6Jerk was not included in the analysis in this section as the relevant data was not able to be
meaningfully extracted to allow this treatment.
79
= 1.59x10-5, η2 = 0.010) contribute a substantial amount of variation to the overall model,
and there is significant interaction between them (F8, 7800 = 30.04, p = 1.09x10-46, η2 = 0.030).
This indicates that different writers have different dynamic profiles for handwriting types.
However, the effect of handwriting type is much greater than that of the individual writer
on the outcomes measured.
A full MANOVA test model reveals that three of the features (duration, relative pen
down duration and average pen pressure) show a significant difference for both the
handwriting type, the individual writer, and the interaction between the two (all p values <
4.55x10-4). For the average absolute velocity however, while there were significant
differences between the handwriting types (F3 = 97.1, p = 1.34x10-60), and the interaction
between the individual and handwriting type (F2 = 21.1 p = 1.44x10-11), there was little
difference observed between individuals (F1 = 0.11, p = 0.741). Of the four dynamic
features measured, the relative pen down duration showed the largest effect within the
model, contributing 22% to the variation observed between the different handwriting types.
As the goal of the research was to determine which measures may be used to assess
the handwriting behaviour used to create an unknown sample, the nature of the differences
between the dynamic features due to handwriting types was investigated in more detail.
Table 7.2 shows the mean value and SEM7 for each measurement, taken across all writers
within the study. The SEM gives an indication of the variation of the measure within each
handwriting type. As this is very small for relative pen down duration it can be deduced that
the majority of the variation is between handwriting types rather than within the
handwriting types.

Table 7.2. Mean value (and SEM) for each measurement relative to handwriting type across
the writer group.
Relative pen Average pen
Duration Average absolute
down duration pressure
(seconds) velocity (cm/s)
(seconds) (arbitrary units)
Natural 1.45 (0.02) 0.79 (0.003) 3.02 (0.02) 385 (2.42)
Disguised 1.92 (0.05) 0.74 (0.004) 2.88 (0.03) 374 (3.72)
Simulated 3.00 (0.09) 0.73 (0.006) 1.59 (0.03) 303 (3.87)

To examine the significance of the differences within each characteristic measured,

post-hoc Bonferroni adjusted comparisons were carried out between each of the
handwriting types for each measurement. Where two natural samples were collected from

7Standard error of the mean. A measure of how well the sample mean reflects the true population
mean.
80
writers, there were no significant differences between the dynamic features of the two
samples, except for average pen pressure (p = 0.001). Although average pen pressure was
found to differ significantly between disguised and simulated writings (p = 2.34x10-43), this
difference should be taken with caution due to the small significant difference between the
two natural samples, which suggests that pen pressure may differ between handwriting
samples regardless of handwriting behaviour. This finding is in contradiction with those
reported by Mohammed, Found, Caligiuri, and Rogers (2009) and Tytell (1998) where for
signatures, pen pressure patterns were found to be consistent over time and therefore could
be used as a discriminatory feature between genuine and simulated signatures, or for
automatic signature verification (Teulings, 1996). However, handwriting and signature
production are different motor tasks and hence could display different dynamic features.
This hypothesis is supported by Chapran et al. (2008). They report, in relationship to pen
pressure, that there were greater variations in pen pressure while subjects wrote a large
amount of text compared to very low pen pressure variation when writing a signature.
Van Galen and van Gemmert’s (1996) finding that pen pressure is increased under
stress conditions, such as disguising handwriting, are also not borne out by this research,
where no significant difference in pressure was observed between naturally written and
disguised samples.
Duration and average absolute velocity were significantly different for all handwriting
type comparisons (natural to disguised, natural to simulated and disguised to simulated, all p
values < 0.001). However, for relative pen down duration, although natural to disguised and
natural to simulated comparisons displayed significant differences, disguised and simulated
did not. This would indicate that altering writing behaviour from a writer’s natural style
decreases the relative time the pen is down, but that disguise and simulation behaviour are
similar in this regard. As far as the authors are aware, this particular feature has not been
reported in the literature, but may be related to pause durations (between letters, and letter
strokes, either on or off the page) which have been reported by van Gemmert et al. (1996)
as being increased for disguised compared to natural writing.
However, the non-significant difference for relative pen down duration between
disguised and simulated writing, based on post-hoc Bonferroni adjusted comparisons,
differs from the result shown in the contrast table (Table 7.3). This gives the contrast results
for disguised versus simulated writing types, which indicates the change in the dynamic
measurements between the disguised writing and simulated writing types. Significant
differences are observed between these writing types for all four variables measured. The
difference in results occurs because the post-hoc tests are based on the variance in the data
and test whether there are significant amounts of variation between each individual data

81
point and all other data points in the same group, compared to all other points in all other
groups. The contrast tables give the differences between the means of the group, with the
95% confidence interval taking into account the error observed within the data. The post-
hoc Bonferroni adjusted comparisons and the contract results should usually be about the
same, unless there are differences in the amount of variation in the mean. Skewed data
points lead to different results for the two tests, which is the case for relative pen down
duration.

Table 7.3. Contrast results (K matrix) for disguised versus simulated handwriting types.
Duration RelPenDown AveAbsVel AvPenPress
Contrast
-1.180* .021* 1.284* 79.426*
Estimate
95% CI (-1.323 – -1.038) (0.010 – 0.033) (1.232 – 1.337) (71.352 – 87.5)
*Statistically significant at 5%.

Box plots for each feature as a function of handwriting type are shown in Figure 7.2.
This illustrates the variability of the data within and between handwriting types. As can be
seen, simulated writings showed considerable variation in duration and relative pen down
duration, and lower median values for average absolute velocity and average pen pressure,
when compared to natural and disguised writing.

Figure 7.2. Box plot for each characteristic measured as a function of handwriting type.
Each box represents the variation between the 25th and 75th quartiles, with significant
outliers indicated with asterisks.

82
The variation between the 25th and 75th quartiles and the significant outliers presented in the
box plots indicate the between writer variation. This can be more easily observed in Figures
7.3 – 7.6, where the estimated marginal means for the dynamic measures are given per
writer for each of the handwriting types.

Figure 7.3. Estimated marginal means of per word duration for each writer and writing
type (non-estimable means are not plotted).

Figure 7.3 shows that, with respect to duration, there is little variation between
individuals for the naturally written samples, with an increased amount of variation for
disguised writings, and a much wider range of variation in the simulated writings. It can also
be seen that for the majority of writer sets, the duration per word increases from genuine to
disguised, and from disguised to simulated writings. This is likely due to the increased
cognitive load required for the un-natural writing processes, and is linked to the velocity
measure also.
The variation in average absolute velocity (Figure 7.4) is approximately equal between
individuals across all writing types. The means for simulated writings are considerably lower
than those for natural and disguised writing types. For all writer sets, the simulated writings
have a lower average absolute velocity than the target writer’s natural handwriting, and in
the majority of writer sets this is also lower than the disguised writing sample.

83
Figure 7.4. Estimated marginal means of average absolute velocity for each writer and
writing type (non-estimable means are not plotted).

Figure 7.5. Estimated marginal means of relative pen down duration for each writer and
writing type (non-estimable means are not plotted).

84
For pen down duration (Figure 7.5) there are approximately equal amounts of variation
between individuals for natural and simulated writing types, and less for disguised writings.
However, the overall mean for simulated samples is less, revealing that for this handwriting
behaviour writers tended to hold the pen above the page for longer than their
corresponding target writers did (when executing both natural and disguised writings). This
is not unexpected, as it could be anticipated that simulators spend significantly more time
‘in-air’ while working out what the next character should look like and how to achieve that.

Figure 7.6. Estimated marginal means of average pen pressure for each writer and writing
type (non-estimable means are not plotted).

The variation in average pen pressure (Figure 7.6) is approximately equal between
individuals across all handwriting types. There are no clear trends for individual writers
between writing types, however the majority of simulators exhibited a lower average pen
pressure than their corresponding target writer did in their natural writing. How this relates
to Sita and Rogers’ (1999) finding that simulators tend to modify their writing pressure to be
more like that of their target writer, or van Galen and van Gemmert’s (1996) finding that
pen pressure is increased under stress conditions is unknown, as data on the simulators’
natural handwriting dynamics were not collected. The pressure data here may also be
affected by factors relating to the accuracy and range of collection through the tablet and

85
software. Pressure was measured in arbitrary units with pressure divided linearly into bins
between zero and 1023 but was not calibrated for gram weight.

7.3.2 Discussion
Not surprisingly, measures of duration, relative pen down duration and average absolute
velocity showed little difference between the two sets of natural samples which illustrates
relative constancy in normal writing behaviour. Between natural and disguised writings
however, all characteristics measured showed significant differences. Taking into account
variation between individual writers, it would appear that the disguise process decreases the
velocity of pen movements, the time the pen is down, and the pressure exerted, but
increases the overall time taken. While this is not constant for all individuals, as a group,
disguised behaviour did significantly alter these parameters. When considering simulated
writing, the same patterns were observed as for disguised behaviour, but with the magnitude
of differences substantially increased compared to natural writing.
How then, do the disguised and simulated samples compare? There was substantially
more variation between individuals for simulated samples, possibly indicating that some
individuals within the experiment were more capable than others at simulation behaviour,
while all were similarly able to disguise their writing. This is supported by the significant
interaction between the individual and handwriting type within the MANOVA model.
Regardless, clear differences were seen between disguised and simulated writing, with large
increases in the overall time taken to write and decreases in writing speed.
These results suggest that velocity may be used as a predictor of disguise versus
simulated handwriting. However, the practical use of this measure will depend on forensic
handwriting examiners’ skill in reliably estimating the relative magnitude of velocity of a
static handwriting sample in the order of the differences observed between disguised and
simulated writings. This could potentially be achieved by assessing static features such as
tremor, hesitations, pen drag and tapering in the line trace. However, the significant overlap
of the range of variation in this measure will mean that some samples will not be able to be
accurately differentiated. Further discussion on this point is made in Section 7.4.2.

7.4 Data Averaged Over Individual Repeated Words Within Samples

The statistical analyses reported in this section were based on data relating to individual
words repeated a number of times throughout each handwriting sample. Table 7.4 shows
the words used and their number of repetitions in the specific text.

86
Table 7.4. Individual words used in analysis for this section, and number of their repetitions
in the specific text.
Word that you with will and
# of repetitions 5 5 5 4 4

The dynamic features analysed were duration, relative pen down duration, average
absolute velocity, normalised jerk and average pen pressure, for the three handwriting types
(natural, disguised and simulated). Following word extraction, further processing was
undertaken to obtain data for individual strokes. The MovAlyzeR® software can segment
movements into strokes by zero crossings of the vertical velocity after peak velocity. The
writing is thus segmented into successive up and down strokes (Figure 7.7). For an x-y data
pattern as given in Figure 7.7a the vertical (y) velocity data chart is as shown in Figure 7.7b.

Figure 7.7. (a) x-y data pattern of the written word know. Red circles show vertical velocity
zero crossings (strokes). (b) Vertical velocity data chart for x-y data pattern 7.7a.

87
 The settings for word extraction and segmentation resulted in less than 100% accurate
data. Some words were split further than they ought to have been, and others were run into
other words, letters or elements (e.g. full stops or commas). The data were manually cleaned
so that each condition related to only one word and an entire word where possible. In some
cases, however, some words or letters of interest were not written, and in one case a letter
form was written but not recorded. These are limitations of the study. The data for each
stroke within every repetition of a word was subjected to the linear mixed-effect maximum
likelihood model.

7.4.1 Results
Fixed effects tests show significant differences between the three conditions of natural,
disguised and simulated handwriting for the measures of duration (F2, 24 = 7.379, p = 0.003),
average absolute velocity (F2, 24 = 6.680, p = 0.005) and normalised jerk (F2, 24 = 5.017, p =
0.015).
The average pen pressure data for this analysis included some zero measures (where
the pen was off the page, or axial pressure was below the minimum threshold – set at zero
N, but not calibrated) while other measures were in the hundreds. This vast range of
variation within each category means that no significance can be observed between them.
One of the hallmarks of unnatural writing is said to be blunt beginnings and endings, as
opposed to natural writing’s flying starts and finishes (Cain, 1998; Ellen, 1989, p. 50;
Harrison, 1958, pp. 375, 400; Hilton, 1982, p. 185; Huber & Headrick, 1999, pp. 113, 273,
288, 289; Morris, 2000; Osborn, 1929, pp. 273-274). In addition, van Galen and van
Gemmert (1996) report that writing under stress conditions results in increased but less
variable pen pressure than natural writing so it could be postulated that writing of different
types (natural, disguised or simulated), may display different pressure profiles. Conversely, it
could be hypothesised that the first stroke of a word may be executed with heavier pressure
than subsequent strokes of the word, as the writer adjusts to the writing surface and
implement, following a pen lift. Although it has been reported (Kao, 1983 in Teulings,
1996) that there is a uniform, systematic increase of pen pressure towards the end of a
word. Plots of axial pressure versus time for individual words extracted from a handwriting
sample set (Figure 7.8a – f) reveal that the first stroke of a word is not generally heavier than
following strokes, that the pen pressure does not generally increase towards the end of the
word, and that the pressure profiles for different writing types are similar. However, there
may be unknown effects related to copying specific text rather than writing freely,
particularly for natural writing.

88
Figure 7.8.8 (a) Axial pressure versus time for word that (inset) in its first instance in a
naturally written sample.

Figure 7.8. (b) Axial pressure versus time for word that (inset) in its second instance in a
naturally written sample.

8 Note that the inset words that in Figure 7.8a-f show both pen down and pen up strokes. The
strokes made with pen up off the page correspond to the plotted sections of zero axial pressure.
89
Figure 7.8. (c) Axial pressure versus time for word that (inset) in its first instance in a
disguised sample.

Figure 7.8. (d) Axial pressure versus time for word that (inset) in its second instance in a
disguised sample.

90
Figure 7.8. (e) Axial pressure versus time for word that (inset) in its first instance in a
simulated sample.

Figure 7.8. (f) Axial pressure versus time for word that (inset) in its second instance in a
simulated sample.

Significant differences were also obtained between words for relative pen down
duration (F4, 106 = 25.79, p = 6.88x10-15), velocity (F4, 135 = 10.87, p = 1.138x10-7) and average
pen pressure (F4, 130 = 15.08, p = 3.758x10-10), and between word order for velocity (F19, 9823
= 2.790, p = 4.827x10-5) and pressure (F4, 9812 = 2.217, p = 0.002).

91
 The significant difference between words for relative pen down duration may be the
result of the different number of pen lifts expected within the words. For example, in a
hand printed form, we would normally expect the word with to be written with five pen lifts
(and therefore a lower relative pen down duration) while the word and would be expected to
be written with just two pen lifts, and hence a higher relative pen down duration.
It was found that the later in the sample a word was written, the higher its average
velocity. This could be the result of the writer taking less care as the sample progressed, or
becoming more comfortable with the experimental set-up.
The test model reveals significant covariance of writers for all the dynamic measures,
and the interaction between writers and the word written for four of the five measures, as
given in Table 7.5. This shows that the different writers had significantly different values for
all of the dynamic features measured, and that for all of the measures except jerk, the
magnitude of the difference between the writers differed depending on what word was
being written.

Table 7.5. Significance of inter-individual differences in writers and writer/word

interactions for each of the five dynamic parameters.
Parameter Writer Effects Writer/Word Interactions
Duration Z = 3.459, p = 0.001 Z = 3.369, p = 0.001
Relative Pen Down Duration Z = 2.999, p = 0.003 Z = 2.803, p = 0.005
Average Absolute Velocity Z = 3.416, p = 0.001 Z = 4.243, p = 0.026
Normalised Jerk Z = 3.218, p = 0.001 Z = 1.684, p = 0.092
Average Pen Pressure Z = 3.331, p = 0.001 Z = 3.525, p = 0.0001

Box plots for each feature as a function of handwriting type are shown in Figure 7.9,
illustrating the estimated marginal mean and 95% confidence interval for each of the writing
types and dynamic measures.
To examine the significance of the differences within each characteristic measured,
post-hoc Bonferroni adjusted comparisons were carried out between each of the
handwriting types for each measurement. No significant differences were observed between
natural and disguised writing types for any of the measures (p > 0.3), nor between the other
writing type comparisons (natural versus simulated, disguised versus simulated) for relative
pen down duration (p = 0.097; p > 0.5) and average pen pressure (p = 0.272; p = 0.613).
Again, this latter result suggests that pen pressure may vary between handwriting samples
regardless of handwriting behaviour.

92
Figure 7.9. Box plot for each characteristic measured as a function of handwriting type.
Each box represents the variation between the 25th and 75th quartiles, with significant
outliers indicated with asterisks.

Significant differences were found between natural and simulated, and disguised and
simulated writings for duration (p = 0.004; p = 0.020), velocity (p = 0.011; p = 0.015) and
normalised jerk (p = 0.024; p = 0.049). Plots of absolute velocity versus time and absolute
jerk versus time for a writer set are given in Figures 7.10 – 7.12.

93
 a

Figure 7.10. (a) Absolute velocity and (b) absolute jerk versus time for word knees (inset) in
a naturally written sample.

94
 a

Figure 7.11. (a) Absolute velocity and (b) absolute jerk versus time for word knees (inset) in
a disguised sample.

95
 a

Figure 7.12. (a) Absolute velocity and (b) absolute jerk versus time for word knees (inset) in
a simulated sample.

7.4.2 Discussion
As in the previous section, analyses have revealed significant differences between naturally
written and simulated, and disguised and simulated writings in the dynamic features of
duration and average absolute velocity, as well as normalised jerk. The results in relation to
natural and simulated writings are in line with anecdotal literature, and empirical research
reported by van Galen and van Gemmert (1996), with simulated writings being more slowly

96
written and displaying decreased fluency of the line trace. Research relating to simulated
signatures has come to similar conclusions, with writing speed influencing line quality (as
well as spatial correspondence; Halder-Sinn & Funsch, 1998; Phillips et al., 2000).
In terms of translating these results to features that may be observed in a static image
of handwriting, the time taken to write a sample is not able to be assessed, but fluency (jerk)
and velocity may be able to be inferred from features visible in the written image, such as
tremor, hesitations, pen drag and tapering. Some studies have been undertaken dealing with
the inference of dynamic properties from the static image. Halder-Sinn and Funsch (1998)
report on indicators of velocity change present in the static trace of a handwritten word
anlagen, finding that visible tremor, the extent of deformation of the letter n, and the length
of retraced lines are reliable within a limited range of velocities. More recently, Guest et al.
(2011) note that in their study involving nine FHEs examining genuine or forged signatures
the experts were able to make accurate, albeit non-quantifiable, observations of pen speed
with respect to the ground truth. However, to our knowledge the skill of FHEs for the task
of estimating the relative magnitude of dynamic features has not been empirically tested and
reported in the literature, so should be further investigated. Notwithstanding this, Dick,
Found, Lacey and Rogers (2004) presented findings relating to the cognitive ability of FHEs
to assess relative writing speed using common word comparisons9. Their test of 31 FHEs
included 80 sets of image pairs (a single word written in uppercase print), and found that
there was a 97% accuracy rate in categorising the images with respect to average pen speed
(second word written more slowly, or faster than the first word, or could not say). The study
found that over the group of experts tested, the mean speed difference between the sample
pairs increases, the number of correct opinions regarding relative speed increase, the
number of inconclusive opinions decrease and the number of misleading opinions does not
change significantly. Based on data contained within the Revision and Corrective Action
Package for this speed trial (Table 7.6; Found, Rogers, Dick & Lacey, 2002), we can assess
the magnitude of the difference in speed in the word pairs and how the reported skill of
FHEs relates to the differences in velocities observed in the data covered in this chapter.
When there was no difference in qualitative speed between the pair of writing samples
(i.e. the level of difference is 0) there is generally a high number of inconclusive opinions –
usually more than half of the opinions expressed – and a higher number of misleading
opinions than observed for pairs where there was a qualitative difference in speed. The
velocity difference for these sample pairs ranged from 0.21 to 0.94, with one at 1.69 cm/s
(this pair written ‘fast’). Three sample pairs with a difference in qualitative speed level of one

97
(n/f or n/s) had differences in speed within this range (0.61, 1.27 & 1.58 cm/s). The
corresponding inconclusive rates were generally less than the previously mentioned pairs and
no errors were expressed.

Table 7.6. Scores for handwriting speed validation trial, with corresponding difference in
speed between word pairs (adapted from Found, Rogers, Dick & Lacey, 2002).
Qualitative
Difference Level of
Sample speed1 Correct Inconclusive Misleading
(speed) cm/s difference2
A B
11 f f 0.208 4 25 2 0
6 s s 0.259 1 30 0
10 n n 0.477 3 27 1 0
12 n n 0.477 4 26 1 0
21 n f 0.61 22 9 1
3 s s 0.898 23 8 0
18 f f 0.938 2 19 10 0
15 n f 1.27 23 8 1
7 n s 1.582 28 3 1
1 f f 1.686 11 18 2 0
14 f n 1.941 31 1
22 f n 1.995 24 7 1
24 n f 2.046 22 9 1
26 s n 2.246 27 4 1
13 n s 2.697 31 1
8 s n 3.137 30 1 1
4 n f 3.152 14 17 1
25 n s 3.168 31 1
9 n f 3.378 28 3 1
2 s f 4.002 31 2
16 s f 4.211 31 2
20 f s 4.663 30 1 2
5 f s 4.752 31 2
23 s f 4.927 31 2
17 s f 4.935 30 1 2
19 s f 5.599 31 2
1
Qualitative speed as executed by the writer for each word, A and B, in the sample pairs; f =
fast, n = normal, s = slow.
2
Between qualitative speeds for each sample pair.

The majority of pairs with a 1-level difference in speed exhibited quantitative

differences between 1.94 and 3.38 cm/s. With the exception of one sample pair, these had
inconclusive rates of less than a third of the opinions expressed (one had inconclusive rate
of just over half).

9Will (2011) reports findings from a similar study, but despite expressing the aim of exploring FHEs
ability to infer handwriting speed from the static trace, refers to durations of the writings (and quotes
values in seconds) rather than velocities.
98
 Sample pairs with a 2-level difference in speed (s/f) show quantitative speed
differences of 4.00 to 5.60 cm/s. Two of these samples had a single inconclusive or
misleading opinion expressed.
These data indicate that if the relative speed of two samples is ‘the same’ (up to 1 cm/s
difference in mean tangential velocity), the skill of FHEs in detecting the difference in speed
is relatively poor. The skill of FHEs in detecting the difference in speed increases with
increasing difference once the relative speed of two samples is greater than 1 cm/s. At a
difference of 4 cm/s the error and inconclusive rates are small. The mean tangential speed
for slowly written words was 1.7 – 3.4 cm/s, for normal writing speed: 4.6 – 5.9 cm/s and
for fast: 5.7 – 8.6 cm/s.
Mean average absolute velocity averaged across handwriting samples from the
MovAlyzeR® data was 2.16 – 4.41 cm/s for normal writing, 1.75 – 4.72 cm/s for disguised
writing, and 0.76 – 2.50 cm/s for simulated writing. A possible reason that these speeds for
normal handwriting are lower than those recorded from the 2002 handwriting speed trial
data may be that in the earlier trial, a single word, easily memorised by the writer, was
written. In the current trial, writers had to copy an extended text (87 words), and the act of
having to look forward to see what they were to write next may have slowed down the
writing process. This hypothesis is supported by research on temporal effects reported by
Thomassen and van Galen (1997) and also Teulings (1996), where activities that require
higher levels of information processing result in a slowing of the pen in its trajectory.
Data plotted in Figure 7.2 shows that at least 75% of simulated extended text
samples had average absolute velocity of less than 2 cm/s (mean 1.59 cm/s), while at least
75% of disguised (and natural) samples had average absolute velocity greater than 2 cm/s
(mean 2.88 cm/s). These results suggest that, if a questioned unnatural sample is compared
to a known natural sample and the velocity difference is detectable (slower in the unnatural
sample), it is more likely to be a forged sample than a disguised sample. Conversely, if the
writing speed of the unnatural sample is not noticeably different from that of the natural
sample, it is more likely to be disguised than simulated. However, these findings would
benefit from more rigorous testing via a specifically designed study.

7.5 Conclusion
This study investigated the dynamic features of duration, relative pen down duration,
average absolute velocity, normalised jerk and average pen pressure in naturally written,
disguised and simulated handwritten text made by 11 sets of writers. Significant differences
were found between natural and simulated, and disguised and simulated writings in the
features of duration, normalised jerk and average absolute velocity. Duration and jerk

99
measures were found to be larger for simulated writings, while velocity was decreased in
simulated writings compared with both naturally written and disguised samples.
Velocity and jerk may be used as predictors to discriminate between disguised and
simulated handwriting behaviour. However, the practical use of these measures will depend
on FHEs’ skill in reliably estimating the relative magnitude of them in a static handwriting
sample in the order of the differences observed between disguised and simulated writings.
Furthermore, different writers were found to have significantly different values for all of the
dynamic features measured, with the inter-writer variation causing overlap between values in
different categories (writing types), which will decrease the accuracy of differentiation,
should the skill exist.

100
 Chapter 8
Disguise Strategies

8.1 Introduction
Numerous studies have been carried out over the past 60 years investigating the nature of
disguise behaviour in handwritten text, and reporting the most commonly encountered
disguise strategies (Alford, 1970; Harrison, 1958, pp. 349-372; Keckler, 1997; Konstantinidis,
1987; Regent, 1977; Totty, 1991). Huber and Headrick (1999, pp. 280-281) have distilled
from these studies a list of changes that may occur in disguised writings. The features
reported as common amongst these are:
• Slant
• Letter design
• Commencement or terminating strokes
• Size
• Style
• Hand used
• Spacing
• Ascenders and descenders
• Speed or fluency
• Greater angularity
• Grotesque letter forms
• Flourishes or embellishments
• Pen stops and pen lifts
• Connections

These changes may, in themselves, be intentional or simply a by-product of the

intentional alterations and thus may occur alone or in combinations. In fact, Morris (2000, p.
169) notes that ‘no single characteristic of writing can be changed without also affecting
other characteristics’. So, when slant changes, the pictorial appearance of writing also
changes, perhaps along with the writer’s relative speed and pressure habits, letter styles,
construction and connecting strokes. When style is changed, the speed and fluency of
writing may also be affected resulting from the increased cognitive load associated with
writers’ diverging from their natural writing style. This implies that the disguise strategy

101
intentionally employed by a writer may not be directly associated with the features that an
FHE observes as predicting the behaviour.
This chapter investigates the relationship between writer-reported disguise strategies
and FHEs’ observations of those disguise strategies. This is achieved by comparing three
sets of data relating to experts’ perceptions with the disguise strategies that were self-
reported. The results reported are based on the disguise strategies recorded by participants
in the answer booklets for the trials with process opinions (reported in Chapters 3 and 4). In
addition, a further set of results is included, from FHEs who were only asked to make an
assessment of disguise strategies. The trial for this set of results was termed ‘Declared pairs
trial’.
The combined results provide further empirical support for previously reported
disguise behaviour and enable FHEs to determine the degree to which their perceptions
align with writer-reported strategies.
Further analysis compares the success of disguise (as assessed by two FHEs) with
FHEs’ ability to correctly determine whether a sample is disguised or natural.

8.2 Disguise Strategies

8.2.1 Methods
8.2.1.1 Participants
Results reported here are from forensic handwriting examiners who provided independent
opinions on the disguise strategies employed in the production of handwriting samples.
These opinions were then compared with the writers’ self-reported disguise strategies.
The results reported here are based on participants’ opinions on three different trials.
These were the ‘Declared pairs’ trial where six FHEs from two countries took part, the
‘Sample pairs’ trial where eleven FHEs from two countries took part, and the ‘Individual
sample’ trial where ten FHEs from three countries participated.
All participants had a minimum of two years full-time government laboratory training
under a qualified examiner and each met the standard set out in the ‘ASTM E2388-05
Standard Guide for Minimum Training Requirements for Forensic Document Examiners’.
All FHEs were authorised by their employers to release opinions regarding the authorship of
questioned handwriting and signatures. At the time of data collection, all but one of the
FHEs was working in a government laboratory. The remaining FHE was working privately.

102
8.2.1.2 Materials
Seventy writers volunteered to provide one naturally written and one disguised sample in
both an upper case and lower case print style. The ‘print’ specification was not strictly
enforced and some writers provided some samples in what could be considered a mixed
style. These were still included in the analysis. For each disguised sample, the writer was
instructed to note the disguise strategy or strategies that they employed to create that
sample. No information was given to the handwriting providers about how to disguise their
handwriting, and the sheet where they were to record the strategies they used did not
contain any ‘prompts’, i.e. there was no list of disguise strategies provided; they had to write
in free text the strategy they used (see Appendix D).
The handwriting samples consisted of a specific paragraph, and were executed on the
same make of white paper and using the same make and model of ballpoint pen. Each of the
samples was scanned at 600 dpi, and inkjet-printed into a booklet. The samples were also
converted into high-resolution PDF files. Three separate trials were constructed using the
materials gathered from the handwriting providers, and provided to the three sets of
participants.
Declared pairs and sample pairs trials. These two trials consisted of 140 pairs of equivalent
handwritten text, each pair written by one writer, with 70 writers in total. One of the
samples in the handwriting pairs was the writer’s normal handwriting, in either upper or
lower case print, and the other was a disguised handwriting sample in the same case.
Individual sample trial. This trial consisted of 200 individual samples of handwritten text,
which were either naturally written or disguised. Ninety eight samples were naturally written
and the remaining 102 samples were disguised.

8.2.1.3 Procedure
Declared pairs trial. Participants undertaking this trial were told which sample in each pair was
naturally written and which was disguised.
Sample pairs trial. In this trial, participants were asked to make an assessment as to which of
the sample pair (A or B) was disguised by entering the appropriate code (A, B or I (for an
inconclusive opinion)) in the answer booklet, as described in Chapter 3.
Individual sample trial. Participants were asked to form an opinion as to whether each sample
was naturally written or disguised, or if they couldn’t say, and record their response in the
answer-recording booklet by entering the appropriate code (N = naturally written, D =
disguised, or I = inconclusive).
For the declared pairs trial, and if an opinion other than inconclusive was given in the
sample pairs and individual sample trials, the participant was required to record in the
103
answer booklet the disguise strategy or strategies they believed the writer used, by marking
the appropriate box(es) corresponding to writer-reported disguise strategies and those
commonly mentioned in the literature (Alford, 1970; Harrison, 1958; Keckler, 1997;
Konstantinidis, 1987; Regent, 1977; Totty, 1991). See Appendix E for an extract of the
declared pairs trial answer booklet, showing the disguise strategies listed. These were
included in a slightly different way for the individual sample trial (see Appendix F), as the
two ‘pairs’ trials had features listed in a comparative way (e.g. faster writing speed, slower
writing speed) while the individual sample trial necessitated the inclusion of strategies in an
absolute sense (e.g. internal differences in writing speed).

8.2.1.4 Ethics approval

Approval for this study, and that reported in Section 8.3 below, was obtained from the La
Trobe University Human Ethics Committee on the basis that the handwriting providers and
test participants gave full consent for samples of their handwriting, or research data
provided by them respectively, to be included in published material, on the condition that
their name and any other identifying information remain confidential.

8.2.1.5 Analysis
In this section, a qualitative exploration was performed on the free text descriptions of
disguise strategies employed by handwriting providers and the disguise strategies that were
nominated by FHEs.

8.2.2 Results
8.2.2.1 Writers’ reported strategies
The free text descriptions of disguise strategies employed by the handwriting providers were
collated and tabulated to assess the disguise strategies most commonly used by this group of
writers. The data is given in Figure 8.1 for both upper and lower case disguised writings. It
should be noted that many of the listed strategies can result in changes in two directions, e.g.
angularity, neatness, letter size and writing speed may be increased or decreased. The
direction of these changes are indicated in the figure by a positive (increase) or negative
(decrease) value.
Although the ‘other’ category tops the frequency list, it contains a wide variety of
disguise techinques with none highly frequent compared to the listed strategies. These
approaches to disguise included writing with the non-preferred hand (five upper case
writings, one lower case writing), altering paper orientation (two upper case writings, four
lower case writings), and introducing spelling errors (two each upper case and lower case
104
writings). The most commonly employed strategies were found to be altered letter
construction, slope and angularity (with almost equal numbers of writers changing to more
angular forms as more rounded).
The frequency of most disguise strategies is similar for both upper and lower case
writing styles, with the exception of altered letter size, proportions and stroke order, which
were more common in lower case writings; and altered format of text and introduction of
tremor, which were only observed in upper case writings. However, slightly more writers
introduced a larger number of disguise strategies into their lower case writings than their
upper case writings (34 versus 25 writers, Figure 8.2). Eleven writers reported they employed
the same number of strategies into both of their handwriting samples. For some writers, the
strategies they utilised for both upper and lower case writings were the same (particularly
when altered letter design or altered slope were employed), but this was not a trend across
the group.

105
Figure 8.1. Frequency of writer self-reported disguise strategies for upper case and lower case handwritten text.
Figure 8.2. Number of disguise strategies employed by each writer for upper case and lower
case writings.

8.2.2.2 FHEs’ perceived strategies

The frequency of observed features that FHEs deemed were the disguise strategies
employed by writers, when examining a pair of handwritten samples, are given in Figure 8.3.
This includes the responses for the trial where the disguised samples were declared, and
where the FHE participants had to assess which of the sample pair was disguised and which
was naturally written. A limitation of this analysis is that the correctness of the response for
the sample pairs trial was not considered for inclusion of the perceived disguise strategies. This
means that for those naturally written samples that were considered disguised (less than 4%
of responses in this trial were erroneous), the perceived disguise strategies are included in
this analysis.
When examined in a pair with a naturally written counterpart, disguised writings were
perceived by FHEs to most frequently have altered letter construction, speed and angularity
(Figure 8.3). As can be seen, the observations made by FHEs on both declared and
undeclared pairs are similar in terms of the relative frequency of strategies. In conjunction
with the finding reported in Chapter 3 that FHEs have a skill in determining which of a pair
of handwriting samples, written by one writer, was disguised, this suggests that regardless of
whether or not disguised writing is declared, when presented in a pair with natural
handwriting FHEs are able to consistently identify features that are indicative of the
unnatural handwriting behaviour.

107
Figure 8.3. Frequency of disguise strategies perceived by FHEs when examining pairs of
handwriting samples where the natural and disguised samples were declared or had to be
assessed by the participants.

The frequency of disguise techniques perceived by FHEs when examining individual

writing samples are shown in Figure 8.4. As there was an error rate associated with FHEs
determining if each questioned individual sample was disguised or naturally written, some of
the perceived disguise strategies actually relate to naturally written samples. These are
indicated in Figure 8.4 with red bars.
Disguised writings examined as an individual sample by FHEs were thought to most
commonly display evidence of altered letter construction, tremor, or an altered slope (Figure
8.4). The ‘other’ category (the fourth most frequently mentioned) included many instances
of observed slow writing speed and overall even or heavy writing pressure10, as well as
overwriting.

10The ‘writing speed’ and ‘average writing pressure’ categories shown in Figure 8.4 refer to instances
of observed internal differences in these features, rather than the specified slow writing speed, and
even or heavy writing pressure articulated in the ‘other’ category.
108
 Interestingly, altered letter construction and tremor, along with letter height, were also
the most commonly perceived ‘disguise strategies’ when FHEs mistakenly deemed a
naturally written sample to have been written using a disguise process.

Figure 8.4. Frequency of disguise strategies perceived by FHEs when examining individual
handwriting samples (natural or disguised), where the disguised writing had to be assessed by
them.

The differences in the number of responses between the pairs (Figure 8.3) and
individual sample (Figure 8.4) trials is not explicable by the difference in the number of
FHEs in the trial groups, as the declared pairs trial had less participants than the individual
sample trial but has a much higher response rate. This may be explained by the relative ease
with which differences between a pair of writings may be observed, and then that difference
given some weight in terms of its function as a predictor of disguise behaviour, compared
with the difficulty in assessing the features of an individual handwritten sample as being
indicative of an unnatural writing process. In this way, some features not evident as
particularly unusual in a lone handwriting sample may have been picked up more readily in a
pair of writings.

109
8.2.2.3 Comparison of self-reported strategies and FHEs’ perceptions
Comparing self-reported disguise strategies with those observed by FHEs in disguised
samples reveals that two of the three most frequently employed habits are also amongst the
three features most commonly detected by FHEs. Altered letter construction was reported
by both writers and experts most often. Alterations to angularity often introduced by writers
was picked up by FHEs examining sample pairs, but not observed as a change by FHEs
who had no comparison sample. Conversely, the repeatedly utilsed alteration to slope was
not observed to be a disguise strategy by FHEs exposed to handwriting sample pairs, but
this strategy was perceived by experts looking at lone samples.
Table 8.1 shows the main writer-reported strategies for disguised samples that were
rated successfully disguised (rating 4 or 5, see Section 8.3) and the number of FHEs who
detected that main strategy (or strategies) when examining each declared disguised sample
(i.e. results of FHEs who undertook the declared pairs trial).

Table 8.1. Writer-reported disguise strategy for successfully disguised samples, and
proportion of FHEs who perceived the disguised strategy, when examining declared sample
pairs (disguised and natural).
Success Proportion of FHEs who
Sample£ Writer-reported disguise strategy
rating† detected disguise strategy
4l 4 Smaller, altered slope 5/6, 4/6
5l 4 Altered letter construction 5/6
5u 4 Altered letter construction 5/6
9l 4 Altered writing position 0/6
9u 4 Altered writing position 0/6
‡
10 u 4 Disconnected letter forms 4/6
11 l 4 Altered letter construction, altered slope 4/6, 1/6
14 l 4 Altered letter construction 5/6
17 l 4 Larger, more rounded 5/6, 2/6
20 u 4 Heavier pressure, smaller 4/6, 1/6
24 l 4 Larger, more rounded 4/6, 6/6
24 u 4 Larger, more rounded 4/6, 5/6
26 l 4 Altered pen grip, more angular 0/6, 0/6
26 u 4 Altered letter construction 5/6
27 l 4 More rounded 1/6
27 u 4 Smaller, more angular 6/6, 2/6
29 l 4 Slower, neater 5/6, 1/6
29 u 4 Non-preferred hand 3/6

110
 Success Proportion of FHEs who
Sample£ Writer-reported disguise strategy
rating† detected disguise strategy
30 l 4 Non-preferred hand 0/6
0/6 (4/6 altered
32 l 4 Altered stroke direction
construction)
0/6 (3/6 altered
32 u 4 Altered stroke direction
construction)
Altered stroke direction, disconnected
33 l 4 0/6, 3/6
letter forms‡
1/6, 0/6 (6/6 altered
35 l 4 Wider, altered stroke direction
construction)
2/6, 0/6 (6/6 altered
35 u 4 Wider, altered stroke direction
construction)
Larger, more rounded, altered letter
39 l 4 5/6, 3/6, 3/6
construction
41 l 4 Smaller, altered slope 5/6, 3/6
42 l 4 Altered slope, lighter pressure 6/6, 5/6
47 l 4 Non-preferred hand 1/6
47 u 4 Larger, altered letter construction 5/6, 5/6
49 u 4 Disconnected letter forms‡ 5/6
51 l 4 Altered slope, more angular 1/6, 3/6
51 u 4 Altered letter construction, altered slope5/6, 1/6
55 l 4 Shorter, altered slope 0/6, 3/6
0/6 (2/6 altered
56 l 4 Altered stroke direction
construction)
0/6 (4/6 altered
56 u 4 Altered stroke direction, taller
construction), 2/6
59 l 4 Larger, neater 2/6, 1/6
59 u 4 Altered letter construction 5/6
65 l 4 Altered pen grip, shorter/fatter 0/6, 1/6
70 l 4 Altered slope, smaller 5/6, 2/6
71 l 4 Untidier 4/6
7u 5 Non-preferred hand 2/6
8l 5 Altered slope 3/6
8u 5 Letter size 6/6
11 u 5 Untidy, altered slope 4/6, 5/6
12 l 5 Altered slope 5/6
14 u 5 Altered letter construction 6/6
16 l 5 Larger, altered letter construction 4/6, 4/6
16 u 5 Wider 5/6
18 l 5 Introduction of tremor 6/6
‡
18 u 5 Disconnected letter forms 6/6
21 l 5 More angular, altered letter construction 6/6, 5/6
21 u 5 More angular, altered letter construction 6/6, 4/6

111
 Success Proportion of FHEs who
Sample£ Writer-reported disguise strategy
rating† detected disguise strategy
22 l 5 Larger 5/6
22 u 5 Larger, more rounded 4/6, 3/6
23 u 5 Altered letter construction 2/6
‡
25 u 5 Larger, disconnected letter forms 5/6, 4/6
30 u 5 Non-preferred hand 0/6
40 l 5 Altered slope, more rounded 4/6, 5/6
40 u 5 Altered slope, altered spacing 6/6, 2/6
41 u 5 More rounded, larger 6/6, 4/6
58 l 5 Altered slope, altered letter construction 4/6, 3/6
68 u 5 Taller, introduction of tremor 4/6, 4/6
70 u 5 More angular, untidier 4/6, 2/6
71 u 5 Altered letter construction 4/6
£
Where l indicates a sample written in lower case, and u indicates a sample written in upper
case.
†
Where 1 = unsuccessfully disguised and 5 = successfully disguised. See Section 8.3 for
further explanation.
‡
Proportion includes perceived strategy of altered letter construction

8.2.3 Discussion
Harrison (1958, pp. 352-359, 372) states, and there is general agreement in the literature that:
• Disguised writing exhibits less fluency and poorer rhythm than natural writing,
• Changes introduced to slope are rarely constant,
• Disguised writing often contains altered letter designs,
• Internal consistency is affected by disguise,
• Features such as spacing, margin and baseline habits, and other format arrangements
are rarely disguised.

These observations are generally supported by the results of this study.

Altered letter construction was the strategy most often employed by writers in this
study (introduced into 57 of the 140 samples), and was also the strategy most commonly
perceived by FHE participants. Fluency and rhythm are related to writing speed, with
tremor negatively correlated. Although changes to writing speed and the (deliberate)
introduction of tremor were not often employed as disguise strategies, these features were
frequently observed by FHEs. This provides additional support for the theory that the
difficult act of introducing a disguise strategy will often impact on the speed and fluency of
execution of the required movements. Empirical evidence of this was reported in Chapter 7,

112
where disguised writing displayed significantly slower movement velocities and higher jerk
measures than natural writing.
Alteration to slope was a frequently used disguise strategy (attempted in 44 of the
140 samples) which was not observed by FHEs examining handwriting pairs as often as may
have been expected. This may be due to the difficulty in maintaining the changed slope such
that the writer’s normal slope habits were revealed as the writing progressed. This hypothesis
is supported by the observation that altered slope was the third ranked disguise strategy
observed by FHEs who had only the disguise sample and no comparison handwriting; the
FHEs examining these individual samples were more likely attuned to internal
inconsistencies in particular features and would have picked up the variation in slope
between the beginning and end of the written paragraph.
Amongst the least applied disguise strategies were altered format, baseline,
punctuation and proportions (combined, used in 8 of the 140 samples). These are what can
be considered inconspicuous features of handwriting, which are reported to be rarely
disguised. Spacing as a feature of disguise was employed in only 18 of the 140 samples, one
of the few findings that does not align well with the literature.

8.3 Success of Disguise

The results reported in the previous section, and in Chapters 3 and 4 examining the skill of
FHEs at discriminating between disguised and naturally written handwritten text samples (in
pairs or individually), may be affected by the skill of the writer for disguising their
handwriting. If a writer employs an ineffective disguise, such that their disguised handwriting
sample does not appreciably differ from their natural writing sample, an FHE not detecting
the disguise (for example in the individual sample trial) will give a misleading process opinion,
however, this will not result in an erroneous authorship opinion, and is unlikely to materially
affect a real-world examination. The success of disguise and its relationship to FHEs’
opinions on the process of production will be explored in the following section.

8.3.1 Methods
8.3.1.1 Participants
Two experts, who did not undertake any of the previously mentioned disguised handwriting
trials, were asked to inspect pairs of handwriting samples, one being declared disguised and
the other declared natural, and arrive at a consensus opinion on the success of the disguise
strategy for each disguised sample.

113
 The training of these FHEs was in line with that of the FHE participants for the
Disguise Strategies section of this chapter (see 8.2.1.1) and both were working in
government laboratories at the time of data collection.

8.3.1.2 Materials
The trial consisted of the same 140 pairs of scanned and printed handwritten text samples as
was provided to participants of the declared pairs trial, and once again these writing behaviours
were declared to the participants.

8.3.1.3 Procedure
The two FHEs were provided with the sample booklet of inkjet-printed images of the
handwriting samples, and an answer sheet. They were asked to record their consensus
opinion on the success of the disguise strategy or strategies employed by the writer of each
sample, by using a scale of 1 to 5, where 1 was an unsuccessful disguise, and 5 a very
successful disguise. They were instructed that a successful disguise is one where features of
the natural writer are not evident in the resultant handwriting sample. An unsuccessful
disguise, therefore, is one where many of the natural features of the writer are observable in
the resultant sample.

8.3.1.4 Analysis
In this section, descriptive statistics for FHEs’ opinions according to the success of disguise
scale were calculated and where relevant, independent sample t-tests (two-tail) were
undertaken.

8.3.2 Results and Discussion

The mean percentage correct, inconclusive and correct called rates, corresponding to FHEs
opinions offered on individual disguised handwriting samples (individual sample trial), for each
disguise rating are given in Table 8.2.
It may be hypothesized that, when FHEs were asked to give an opinion on whether a
questioned individual sample was disguised or natural, samples utilising an unsuccessful
disguise will have attracted a lower correct rate compared to samples with a successful
disguise. This is because the unsuccessfully disguised samples display similarities to the
natural handwriting of the writer, and thus may be easily mistaken for natural handwriting
samples. Although to be rated a successful disguise a sample needed to display dissimilarities
(rather than similarities) to the naturally written sample, it could be expected that samples
where a successful disguise strategy was employed would attract a higher correct rate, as
114
these may be expected to exhibit hallmark features of an unnatural writing process, such as
stilted, laboured writing containing unnecessary or unexpected pen lifts, hesitations, tremors,
retouches and deliberate or forced features. Difficulty in assessing the disguised nature of a
handwriting sample may arise where the writer is able to significantly, consistently and
fluently alter their natural handwriting such that the disguised writing appears naturally
written. These same hypotheses hold for results of FHEs being asked to given an opinion
on which of a pair of handwriting samples is disguised (the other being naturally written).

Table 8.2. Mean percentage inconclusive, correct and correct called (obtained by removing
the inconclusive opinions) rates grouped by disguise success rating, for FHEs giving
opinions on individual handwriting samples.
Disguise %I %C %CC
Unsuccessful 1 46.52 13.90 28.85
2 44.95 17.17 29.84
3 25.97 25.97 40.83
4 43.43 33.00 56.18
Successful 5 25.84 57.89 71.83

As expected, the samples deemed successfully disguised have the highest correct rate
and lowest inconclusive rate, while the unsuccessfully disguised samples have the lowest
correct rate and the highest inconclusive rate. A statistically significant difference was
observed between the successfully (rating 5) and unsuccessfully (rating 1) disguised samples
in both correct (df = 34, t = 4.526, p = 7.0 x10-5) and inconclusive rates (df = 34, t = 3.430,
p = 0.002). This also resulted in a significant difference in correct called responses between
the two ratings (1 and 5, df = 34, t = 3.955, p = 0.0004) despite the variation in responses
on the successfully disguised samples ranging from 0% correct called for some samples, to
100% correct called for other samples.
The range in the correct responses for each of the levels of disguise was broad (see
Figure 8.5); the samples deemed unsuccessfully disguised showed the smallest range, while
the successfully disguised samples had the widest range.

115
Figure 8.5. Average percent correct rate for each sample with a given disguise success rating
(1 = unsuccessful, 5 = successful), when samples presented individually.

The same hypothesis regarding unsuccessful disguises leading to increased

inconclusive and/or erroneous process opinions could be considered for samples examined
in a pair (sample pairs trial). Once again this hypothesis is supported by the results of this
study (Table 8.3) with significant differences between the successfully and unsuccessfully
disguised samples in both correct (df = 48, t = 3.13, p = 0.003) and inconclusive rates (df =
48, t = 3.51, p = 0.001). However, since the percent correct called rates are similar across the
highest and lowest success ratings, there is no significant difference between these (df = 48,
t = 1.22, p = 0.229).

Table 8.3. Mean percentage inconclusive, correct and correct called (obtained by removing
the inconclusive opinions) rates grouped by disguise success rating, for FHEs giving
opinions on pairs of handwriting samples.
Disguise %I %C %CC
Unsuccessful 1 41.96 53.15 85.49
2 36.36 58.68 87.50
3 20.45 78.25 98.25
4 9.55 87.05 93.67
Successful 5 15.91 80.30 94.38

The range in correct responses for paired handwriting samples within each success
rating (Figure 8.6) is broader than that for the individual handwriting samples (Figure 8.5),
with a larger number of samples, in all disguise success categories, attracting 100% correct

116
scores. This reflects the relative ease of the task of determining which of a pair of
handwriting samples is disguised (or naturally written) compared to determining if an
individual sample is disguised or naturally written.

Figure 8.6. Average percent correct rate for each sample with a given disguise success rating
(1 = unsuccessful, 5 = successful), when samples presented in a pair (naturally written and
disguised).

Insight into the reasons for the broad range in responses within each disguise success
level may be revealed by examining examples of some handwriting samples of the specific
ratings. An example of a sample for which the disguise strategies employed (self-reported to
be altered letter construction, slope and pen grip, and writing less neatly) were determined to
be unsuccessful is given in Figure 8.7. It can be seen that it bears many of the same
handwriting characteristics of its naturally written pair, appearing only to have been written
with less care. This sample, when presented in a pair, attracted a high inconclusive rate
(90.1%) and a low correct response rate (9.1%). When examined by FHEs as an individual
sample, none of the responses were correct, with only 18% inconclusive, meaning that the
vast majority of participants thought the sample was naturally written.

117
 a

Figure 8.7. Portion of a sample pair in which the disguised sample (a) was deemed
unsuccessfully disguised (rated 1). Comparison to the naturally written sample (b) reveals
similarities in many handwriting features, with little in the disguised sample to indicate it has
not been naturally written.

The example shown in Figure 8.8 is a disguised sample that was also deemed to have
an unsuccessful disguise (writer-reported as altered pen grip), with proportions and letter
construction similar to the naturally written sample. However, the disguised sample bears
the hallmark dysfluency and poor line quality of unnatural writing, which lead all FHEs to
give a correct response when examining this sample pair. Although 63.6% of FHEs gave an
inconclusive opinion when looking at this disguised sample alone, 75% of respondents
willing to offer an opinion were correct. The high inconclusive rate may be due to experts
being uncertain as to whether the writer had a low skill level, or if the sample was disguised.

118
 a

Figure 8.8. Sample pair in which the disguised sample (a) was deemed unsuccessfully
disguised (rated 1). Comparison to the naturally written sample (b) reveals similarities in
many handwriting features; however the disguised sample displays poor fluency and line
quality.

The writer of the disguised sample in Figure 8.9 employed altered letter construction
that successfully masked many of their natural writing characteristics (disguise success rated
5), while also appearing obviously unnatural. This disguised sample attracted 100% correct
responses from FHEs regardless of whether it was presented alone or in a pair.
The disguise strategies used in the sample shown in Figure 8.10 were alteration to the
spacing, slope and roundedness of characters. These were assessed as successful. The
fluency of the unnatural writing and its distinct pictorial difference to the natural sample
(probably brought about primarily by the change in slope and connectivity) caused
difficulties for FHEs. Presented in a pair, none of the responses were correct, with a
relatively small amount of inconclusive opinions (18.2%). FHEs examining the individual
disguised sample were less certain, with an inconclusive rate of 63.6% but a correct rate of
27.3%. This sample represents one of the few samples where the writer was able to
successfully implement a disguise that appeared to be fluently and skilfully written.

119
 a

b
Figure 8.9. Portion of a sample pair in which the disguised sample (a) was deemed
successfully disguised (rated 5). Comparison to the naturally written sample (b) reveals most
natural handwriting features are masked, with the disguised sample also appearing obviously
unnatural.

120
 a

Figure 8.10. Portion of a sample pair in which the disguised sample (a) was deemed
successfully disguised (rated 5). Comparison to the naturally written sample (b) reveals most
natural handwriting features are masked, with no evidence of an unnatural writing process in
the disguised sample.

8.4 Conclusion
These results suggest that the features observed by FHEs in disguised writing samples
(whether compared against a naturally written sample or examined alone) are largely aligned
to writer-reported strategies. The commonly reported strategies and observations in this
study compare well with those mentioned in the literature. In addition, these results provide
further evidence that any disguise strategy is likely to bring about other unintentional
observable changes.
When considering the success of the disguise strategies employed, there is a positive
relationship between this and the certainty and ability of FHEs for determining disguise.
That is, the more successful the disguise (where success is related to the absence of the
writer’s natural writing characteristics), the lower the inconclusive rate, and higher the
correct rate obtained by the group of FHEs.
This relationship can provide a method for minimising erroneous authorship opinions
in cases of suspected disguised writing. If the disguise is unsuccessful, the writing will
contain many of the features of the writer’s normal handwriting. It may not be identified as
having been written using a disguise process, however, observations made would be
121
expected to lead to a correct authorship opinion. Conversely, if the disguise is successful
there will often be evidence of the unnatural writing process (as seen in Figures 8.8 and 8.9).
This should flag the FHE to proceed to an authorship opinion with caution, given the
results on disguised and simulated writing processes reported in Chapter 5. Difficulties will
arise if the writer is able to execute a disguise fluently, such as in Figure 8.10.

122
 Chapter 9
Predictors of Disguised and Simulated
Handwritten Text

9.1 Introduction
Assessing the authorship of writings created using disguise and simulation behaviour is
problematic for FHEs. Previous research has reported that authorship opinions on these
writing types attract higher inconclusive and misleading rates than natural writing does
(Found & Rogers, 2005a, 2008). While in Chapter 3 it was found that FHEs have a skill over
laypeople in correctly determining the writing process of pairs of naturally written and
disguised writings, there does not appear to be a skill (in either FHEs or laypersons)
associated with determining the process of production of individual samples that are either
naturally written or disguised (Chapter 4). Likewise, Chapter 5 explored the skill of FHEs in
discriminating between disguised and simulated writing samples and found that while the
strength of the evidence of an FHE was greater when they gave an opinion that a sample
was simulated rather than disguised, the probative value of the evidence for both disguised
and simulated writings is low.
The task of discriminating between disguised and naturally written samples was found,
in Chapter 8, to be affected by the success of the disguise strategy employed (those disguised
writings that display many of the writer’s natural writing characteristics were often mistaken
for being naturally written). So when examining questioned natural and disguised samples,
FHEs had a tendency to call disguised samples natural. Furthermore, when examining
questioned disguised and simulated samples there was a propensity for calling simulated
samples disguised. These results suggest that FHEs’ expectations of the predictor features of
these unnatural writing types are not accurate.
This chapter investigates the relationship between FHEs’ responses on the process of
production of questioned disguised and simulated handwriting samples and their verbal
statements relating to the features they observed as indicative of the particular unnatural
writing behaviour. Although, in Chapter 7, it was found that velocity and jerk may be
suitable features to differentiate disguised versus simulated writings, the application of these
measures to casework-like samples is largely unknown. A clear relationship between FHEs’
verbal statements and correct responses will, however, enable elucidation of predictor
features of disguised versus simulated writings that are present in the static images.

123
9.2 Methods
The participants, materials, procedure and ethics approval for this study were the same as
those given in Chapter 5. The results reported in this chapter are based on the observations
participants recorded in the answer booklet regarding the feature(s) they observed which
indicated the process (disguise or simulation) when they gave an unforced opinion.
Observations of 29 FHEs who provided both unforced opinions and observations are
considered.

9.2.1 Analysis
The observations recorded by FHEs regarding the features they observed as indicative of
either disguise or simulation behaviour were free text. The groups’ observations were
reviewed so that suitable identifiers could be developed. This led to the collation of a list,
aligned to the features used in the natural and disguised writing studies (Chapters 3, 4 and 8).
This list was further refined by grouping features together which relate to a common theme,
to facilitate statistical analysis. The groupings (factors, in the statistical analysis) are given in
Table 9.1.
To ascertain how the identified factors are associated with FHEs’ responses
(ANSWER; 1: correct, 0: incorrect), Population-Average Generalised Estimating Equations
(PA-GEE) with binomial distribution and LogitLink was applied (Liang & Zeger, 1986).
The identified factors, as shown in Table 9.1, are SPEED, CONSTRUCTION, SIZE,
SPACING, NEATNESS, SLOPE, PRESSURE, FORMAT and TREMOR. The binary
outcome, ANSWER, was constructed by comparing the FHEs’ responses and the CODE
(disguised or simulated) for each questioned handwriting sample. There were 29 subjects in
the sample set, but each subject contributed 100 repeated observations (one ‘observation’
for each of the 100 questioned samples). As the repeated observations are nested within the
separate subjects, the data set exhibits a typical two-level structure. The PA-GEE was
applied because of the multi-level data structure. However, there was no real advantage to
apply other hierarchical or multi-level models such as the Generalised Linear Mixed Model
(GLMM), where the fixed and random effects must be specified. The repeated observations
are assumed to be correlated in an exchangeable pattern in PA-GEE. Theoretically
speaking, this specification is most appropriate and it has been proved that the results would
remain consistent if the working correlation matrix is mis-specified. Analysed with Stata 11.0
(Stata Corp, Texas, USA), all statistical tests were conducted at 5% level of significance.
Tests were undertaken using the unforced responses of 29 participants. The model was
analysed to determine if there were significant differences in the factors associated with

124
ANSWER when taking into account the CODE i.e. if different factors are associated with
obtaining a correct answer for disguised versus simulated samples.
It is noted that assistance with statistical analysis in this chapter was provided by Dr
Siew-Pang Chan, Senior Lecturer, Mathematical Sciences, La Trobe University.

Table 9.1. Groupings of observed features into factors.

Factor Observed features
Speed Writing speed
Construction Altered letter construction/style
Disconnected letter forms
Altered ascenders &/or descenders
Altered initial or terminal strokes
Angularity
Altered stroke order
Unusual/awkward letter forms/embellished
Proportions
No attempt to simulate
Connectivity
Inconspicuous significant dissimilarities
Diacritic
Similarities in letter forms/details
Imitation of obvious features
Size Letter size
Spacing Spacing
Neatness Neatness
Retracing/overwriting
Corrections
Lack of internal consistency
Slope Altered slope
Pressure Apparent writing pressure
Format Altered baseline
Altered punctuation
Spelling
Margin habits
Tremor Complexity/skill level
Fluency

9.3 Results
Table 9.2 gives the results for the model, where examiners were willing to offer an opinion
as to whether a questioned sample was disguised or simulated. It should be noted that where
a significant Adjusted Odds Ratio (AOR) is above unity for a feature there was a higher
chance of obtaining a correct answer for the CODE. Conversely, where a significant AOR is
below unity there is a higher chance of obtaining an incorrect answer for the CODE.

125
Table 9.2. PA-GEE analysis with binomial distribution, logit link and exchangeable working
for unforced model, including CODE.
CODE = 1 (Disguised) CODE = 2 (Simulated)
Factor Adjusted Odds Ratio (AOR) Adjusted Odds Ratio (AOR)
SPEED 2.82 0.82
CONSTRUCTION 1.17 1.42*
SIZE 2.55 0.48*
SPACING 0.51 0.98
NEATNESS 1.00 1.01
SLOPE 5.33* 0.36*
PRESSURE 5.87 0.80
FORMAT 0.48 1.70
TREMOR 0.78 1.64*
*Statistically significant at 5%.

For disguised samples, the only factor significantly related to ANSWER is SLOPE,
which has a positive relationship. Conversely, for simulated samples, there is a higher chance
of having a correct answer without SLOPE. SIZE is also significantly, negatively associated
with ANSWER for simulated samples, while CONSTRUCTION and TREMOR are
significantly, positively associated with ANSWER.

9.4 Discussion
These results suggest that the identification of altered slope in a questioned sample when
compared to a naturally written sample may be a predictor of disguise behaviour. This is
aligned with observations made in Chapter 8 where altered slope was the second most
commonly employed disguise strategy. If altered slope is not observed as a feature in the
comparison of questioned and naturally written samples this may be a predictor of
simulation behaviour.
Slope and size are both relatively obvious features in handwriting as opposed to stroke
order, stroke direction or writing pressure for example. It may thus be supposed that these
obvious features are copied reasonably well by simulators; hence the greater chance of a
correct answer in the absence of their observation by FHEs in questioned simulated
samples.
Interestingly, in the previous chapter it was found that altered letter construction was
the most commonly employed disguise strategy used by writers in that study, but the
CONSTRUCTION factor was not found to be significantly associated with a correct answer
for disguised samples.

126
 The positive association of letter construction and tremor features with a correct
answer for simulated samples, along with these features not being significantly associated
with answer for disguised samples, means that they may be used as predictors of simulation
behaviour in questioned samples.
A limitation of the analysis undertaken revolves around the diversity in disguise
strategies. In Chapter 8 we saw that while some disguise strategies were commonly used, the
success of disguise, even when the same strategy was employed, varied widely from writer to
writer. Any single predictor feature will only be relevant in cases where the particular
strategy (or by-product) has been employed (or is observable), so will not apply to all
disguised samples.
The factors identified as associated with correct answers for disguised and simulated
writings in this analysis may be used to create a predictor model e.g. using discriminant
function analysis. Such an approach was described by Found, Rogers, Rowe, and Dick
(1998) for modeling experts’ perceptions of signature complexity. Interestingly, these
authors found that the most successful approach for modeling was to identify a ‘concordant’
group of experts who showed similar profiles in their assessments of simulation difficulties
and only model this group. In this way the noise associated with inter-expert variation in
perceptions could be reduced and a cleaner statistical model achieved. The data generated
within this thesis would allow for the selection of participants that behaved similarly on
responses to trial material and, although the subject number would be small compared to the
number in this chapter, it may be that a clearer signal regarding the codes would emerge.
This should be the subject for future research into the discrimination of disguised and
simulated writings.

9.5 Conclusion
These results suggest that observations of altered slope, differences in features of
construction and size and the presence of tremor may be used as predictors to discriminate
between disguise and simulation behaviour. Observing an alteration in slope was found to
be significantly associated with correct answers for disguised samples while observations
relating to construction and tremor were significantly associated with correct answers for
simulated samples. The absence of observations relating to size is also significantly
associated with correct answers for simulated samples.
A predictor model could be created based on these results and would require testing
on a validation set.

127
 Chapter 10
General Discussion

10.1 Introduction
The need for empirical validation of forensic handwriting evidence has been expounded
since 1989 with the first of a series of publications criticising the lack of testing of the
varying tasks FHEs routinely carry out (Risinger et al., 1989). Since that time the critics have
continued to suggest that the scientific rigor of forensic handwriting identification needs to
be improved, particularly with respect to articulation of the underlying theoretical basis of
the discipline and empirical characterisation of potential error rates for different types of
examinations (Risinger et al., 1998; Saks & Koehler, 2005; Saks & Risinger, 1996; Saks &
VanderHaar, 2005). Some twenty years on, and the same concerns and criticisms have been
levelled at this field, and other pattern evidence disciplines, in the USA through the
publication of the National Research Council report (2009).
Although in all reported research the expertise of FHEs compared to laypeople has
been found to exist (Found et al., 1999; Kam, ca. 2010; Kam et al., 1994, 1997; Sita et al.,
2002) these studies were not intended to characterise more subtle aspects of the purported
expertise. Of interest in this research were authorship opinions formed by FHEs on
questioned disguised and simulated writing samples as these types of questioned writing
behaviours attract significantly higher misleading and inconclusive scores on blind trials than
do writings that were naturally written (Found & Rogers, 2005a, 2008). To explore this
further in this thesis it was thought that investigations into the identification of the writing
process (natural as compared to simulated or disguised) might enable the source of
authorship error to be diagnosed. This in turn may provide a guide as to whether an
authorship opinion might be reliably expressed. Dynamic measures of these unnatural
writing behaviours, as well as features extracted from static images by FHEs were
investigated. Comparison of features observed by FHEs with their opinions on the process
of production of disguised and simulated writing was undertaken. This approach was taken
such that the results of these analyses might inform a framework within which FHEs could
reliably express authorship opinions on these problematic writing types. The major novel
outcomes were:
• The provision of evidence regarding FHEs’ expertise for determining natural versus
disguised handwritten text;

128
 • The characterisation of FHEs’ expertise for determining disguised versus simulated
handwritten text;
• The elucidation of relevant dynamic features of handwriting as potential aids for
FHEs in assessing handwriting behaviour;
• The identification of predictors of disguise and simulation writing processes that
may be incorporated into forensic handwriting methodology.

10.2 Summary of Thesis

10.2.1 Evidence for examiner skill and expertise
Previous research supports the proposition that FHEs have expertise over laypeople for
expressing authorship opinions on questioned handwriting samples. Building on that body
of work, the findings of this research provides considerable evidence that FHEs’ opinions
on the process of production of questioned signatures (for which they gave an inconclusive
authorship opinion) only attract a small percentage of error. For the total misleading
authorship opinions relating to disguised signatures, the majority of the process opinions
were correct (69.3%) indicating that a disguise/simulation process was detected, but
misinterpreted as being by another writer. These results show the usefulness of FHEs
offering a first stage simulation/disguise process opinion even in the absence of an opinion
regarding authorship. The unnatural process opinion itself may be of strong evidential value,
with the court able to draw inferences from other evidence or circumstances which may lead
to support for one of the unnatural writing processes over the other.
It was also found that FHEs have a skill over laypeople for determining the writing
process of 140 pairs of natural and disguised text based writings. Analogous to previous
studies, this skill is best characterised by FHEs’ conservatism, where experts express a
higher proportion of inconclusive opinions (23.1% for FHEs compared to 8.4% for the
control group). This leads to the FHEs expressing a smaller percentage of misleading
responses when calling writings as either naturally written or disguised (4.3% for FHEs
compared with 12.2% for the control group). The expertise appears to lay in FHEs having a
greater understanding of the limitations of the examination.
Although a level of expertise was revealed for examinations of pairs of naturally
written and disguised samples, there was no evidence of expertise for examinations of
individual naturally written or disguised handwritings. Based on 200 individual handwritten
samples provided to both an FHE and lay group, no significant difference was found with
respect to their opinion profiles on whether samples were the product of natural or disguise
behaviour. For both subject groups, within-group analysis of opinions shows a difference in

129
the correct rate for natural versus disguised writing samples (FHEs: 65.1% compared to
29.9%; laypeople: 65.7% compared to 45.6%). Natural handwriting behaviours appear
significantly easier for subjects to correctly identify than disguised writing behaviours, with
90% of those samples that showed high (≥70%) error rates being the product of disguise.
Inspection of high error disguised samples revealed that these bore a strong resemblance to
the writer’s natural writing behaviour. This indicates that the disguise strategies employed by
these writers were so subtle or inadequate that the predictors of disguise were not clear to
either the FHEs or lay group. In terms of casework methodology, the errors for disguised
samples reported here would not result in erroneous authorship opinions.
Investigating the skill of FHEs at differentiating between disguised and simulated
writing samples, where the predictive features of these handwriting behaviours may overlap,
revealed that FHEs were not confident in (inconclusive response score = 30.9%), nor
particularly good at, undertaking the task. FHEs were invited to express opinions regarding
100 pairs of handwriting samples with a naturally written comparison sample, and a
questioned sample that was either disguised by the comparison writer, or written by another
writer attempting to simulate the comparison writer’s handwriting features. There was a
significant difference between the correct called responses for disguised and simulated
handwritings. Treatment of the data using likelihood ratios showed that the strength of the
evidence for this group of FHEs was greater when their opinion was that the sample was
simulated than it was when their opinion was that the sample was disguised. However, the
ratios associated with the simulated and disguised data were small indicating that the
probative value of FHEs’ opinions is low.
In addition, the simulation conditions of practice or no practice and the use of a direct
model or not were explored in relation to FHEs’ process opinions. Analysis revealed no
significant differences between the correct or misleading (erroneous) scores of any of the
combinations of simulated writing types. Simulated handwritten text taken under all the
specified conditions attracted high misleading and inconclusive scores. This indicates that
the predictor features that FHEs, on average, used to form opinions on the unnatural
writing samples in this study were not effective.
In contrast, the success of the disguise strategy employed was found to affect the
correct and inconclusive scores of FHEs when giving opinions on the process of production
of individual handwriting samples, or those presented in pairs (naturally written and
disguised). The more successful the disguise, the higher the correct scores, and lower the
inconclusive scores.

130
10.2.2 Predictors of disguised and simulated handwriting
An investigation was made of the relationship between 70 writers’ self-reported disguise
strategies, and 27 FHEs’ observations of disguise strategies in the same handwriting samples.
This was carried out to determine the degree to which the disguise strategy intentionally
employed by a writer is reflected in the static image and observable as feature(s) that a FHE
deems as the most telling of the disguise behaviour. The most commonly attempted disguise
strategies were found to be altered letter construction, slope and angularity. When examined
in a pair, with a naturally written counterpart, disguised writings were perceived by FHEs to
most frequently display altered letter construction, speed and angularity. Conversely, when
examined as a lone sample by FHEs, disguised writings were thought to most commonly
display evidence of altered letter construction, tremor, and an altered slope.
These results suggest that the features observed by FHEs in disguised writing samples
(whether compared against a naturally written sample or examined alone) are largely aligned
to writer-reported strategies; however, in line with other literature, any disguise stratgey is
likely to bring about other unintentional observable changes which were also detected by
FHEs.
The additional cognitive load required to suppress the natural abstract motor programs
when executing both disguised and simulated handwriting results in writing that is generally
accepted to be written slower and less fluently than natural handwriting, such that both of
these unnatural behaviours may exhibit similar features. Investigations were carried out into
whether significant differences exist between naturally written, disguised and simulated
handwritten text in terms of the dynamic features of velocity, jerk and pen pressure, as well
as duration and pen down duration. Analysis based on data averaged across the handwriting
samples in full revealed significant differences between all handwriting types for three of the
four dynamic features measured (duration, relative pen down duration and average absolute
velocity). The strongest effect, when inter-individual variation was discounted, was found to
be associated with the velocity measure. When data was averaged over individual repeated
words within samples, significant differences were found between all handwriting types for
the dynamic features of duration, velocity and jerk. Again, when inter-individual variation
was removed, the strongest effect was associated with the velocity and jerk measures. These
results support anecdotal and limited empirical evidence that, with reliable static estimations,
indicators of fluency (velocity and jerk) may be used to discriminate between disguised and
simulated handwriting behaviour. At present, the ability of FHEs for this task is largely
unknown.
The probative value of FHEs’ opinions on the process of production of disguised and
simulated handwriting samples was found to be low, suggesting that, on average, for the
131
FHE group tested, the predictor features used to discriminate between these unnatural
writing types are ineffective. The relationship of FHEs’ responses on the process of
production of questioned disguised and simulated handwriting samples, and their verbal
statements relating to features they observed as indicative of the unnatural writing behaviour
was explored. The model, based on grouped responses and not taking into account the
success of the disguise or simulation, indicated that altered slope may be a predictor of
disguise behaviour, while features associated with letter construction and tremor may be
predictors of simulation behaviour. Given that tremor is a measure of fluency (along with the
dynamic measures of velocity and jerk, which were found to be significantly different in
disguised and simulated samples), fluency may be one of the most useful predictors to
distinguish between disguise and simulation behaviour.

10.3 Theoretical and Practical Implications

This thesis has provided empirical evidence of the difficulty associated with assessing
whether a handwriting sample is disguised or simulated. While FHEs may be able to detect a
successful unnatural writing process, determining whether the unnatural features are due to
simulation or disguise is not a strong skill. Exposure to known disguised and simulated
handwritten text samples may broaden the practical knowledge of FHEs. This is important
as, based on the results of this research, the current predictor features that the participant
FHEs, on average, use to distinguish disguised and simulated writings appear to be largely
inaccurate.
In order to minimise errors in authorship opinions in case work, the results of this
thesis support a method where, if an unnatural handwriting process is identified, the FHE
does not proceed to proffer an authorship opinion. The expert providing an opinion of
support for the proposition that a signature is something other than genuine may be, in
itself, of strong evidential value.

10.4 Limitations of Current Research

A number of limitations were associated with the research described in this thesis. These
include:
• The relative size of the handwriting trials. Consisting of 100 or 140 handwriting
comparisons, or 200 individual examinations, these trials were large and cognitively
demanding. The care and approach taken by FHE participants may not have been
well aligned with case work and may have affected outcomes;

132
 • The different modes of application used to form the trial materials (inkjet prints, laser
prints and digital images). In addition, no original samples were examined. Some
features (for example stroke order, stroke direction, and apparent pressure) may not
have been readily observed by participants;
• The likely greater incentive for FHE participants to perform as compared to
laypeople in the natural and disguised studies;
• A relatively small sample size of FHEs involved in testing;
• The unknown impact of FHEs involved in testing for whom English is not their first
language and who may not be used to examining handwriting written in English. This
group were as good as the group as a whole but they may have had a superior
performance with writings in their own languages;
• Less than optimized parameters employed for some of the MovAlyzeR® software
functions;
• The unknown incentive and ability of writers to successfully disguise their writing
and/or simulate someone else’s writing;
• The assessment of the success of disguise being made by a small number of FHEs
subjectively;
• The presence of some incomplete participant data sets (i.e. not all forced opinions
were given, and not all participants gave verbal statements regarding what they
deemed were predictor features)
• The grouping of data of participants’ responses and handwriting samples, which may
mask individual FHE skill and ‘muddy’ the data for modelling.

10.5 Directions for Future Research

The findings of this thesis provide additional empirical characterisation to an aspect of
forensic handwriting examination involving the identification and discrimination of
questioned disguised and simulated handwritten text. However, arising from this research
are some issues that remain unaddressed.
Analysis throughout this thesis was based on grouped handwriting samples. Although
the effect of simulation success was touched upon in Chapter 6, and disguise success was
investigated in Chapter 8, further exploration could be undertaken into these issues and their
impact on FHEs’ responses, for example, the use of expert penmen could be considered
along the lines of Dewhurst, Found & Rogers’ (2008) study with signatures.
In addition, data was largely based on grouped data of participants’ responses. It may
be that individual or subgroups of FHEs are better at certain tasks than other FHEs.

133
Identification of these individuals, their specific skill characteristics, and how these might be
explained could provide valuable information for casework and FHE training. If a subgroup
of high performing FHEs can be identified for discriminating between disguised and
simulated writings, a predictor model could also be created based on their correct process
answers and verbal statements relating to observed features.
The analysis of additional kinematic features of unnatural compared to natural
handwriting would enable a better understanding of the dynamics of these processes, which
may then in turn be related to the static image usually encountered in the forensic casework
environment. Further, focussed investigation could be undertaken into whether FHEs can
reliably estimate dynamic features of fluency and velocity from static images.
The work presented in this thesis deals with a small subset of tasks undertaken by
FHEs. Continued validation and skill testing of other aspects of the field should be carried
out in order to solidly address the criticisms raised over the past 25 years. As forensic
handwriting examination did not grow from an academic foundation, it is up to forensic
handwriting examiners ourselves to engage in the scientific process.

134
 Appendix A
Natural and Disguised Handwriting Provider
Information

Forensic Expertise Profiling Laboratory

La Trobe University
Handwriting sample collection for research project: Expertise of forensic handwriting examiners
for determining text-based natural and un-natural writing processes

Dear Sir/Madam:

Subject: Request for samples of your handwriting

The identification of handwriting requires the expert to compare questioned (or unknown)
handwriting to handwriting taken from a suspect (specimen handwriting). Our test will require the
expert to examine samples of questioned handwriting, which are either normal or disguised, and
then express an opinion as to the process used to create each questioned sample. For each test
we require a large number of handwriting samples. This is where we ask for your assistance. You
may, of course, decline to participate, in which case we ask you to hand the kit back. Please note
that you will not, at any stage, be able to be identified from your handwriting sample and the
samples will only be used for the purpose of academic research and training.

Please follow these instructions:

Handwriting collection
1 Sit comfortably with the supplied folder of material at hand.
2 Place the page bearing the printed text commencing ‘NOTE TO BE COPIED’ in front of
you.
3 Separate the page bearing the heading ‘NORMAL 1’ from the other sheets, place in front
of you and write out the note in your normal lowercase print handwriting. Please write
within the box and please do not place the sheet on top of other sheets in the package
while you are writing.
4 Write out the note again, this time disguising your writing (imagine that you do not want
the recipient of the letter to recognise you as the author). Remember the sample is to be
in lowercase print writing. This sample is to be written on the page bearing the heading
‘DISGUISE 1’. Again write within the box and please do not place the sheet on top of
other sheets in the package while you are writing.
5 On the page bearing the heading ‘NORMAL 2’, write out the note again within the box in
your normal uppercase handwriting with the sheet separated from the other sheets
6 Write out this passage again in uppercase, this time disguising your writing. This
sample is to be written within the box on the page bearing the heading ‘DISGUISE 2’.
Make sure the sheet is separate from the other sheets while writing.

I thank you greatly for your participation in this trial. Should you have any questions or concerns
please feel free to contact me by email: bird.carolyne@saugov.sa.gov.au or by phone on 0417
864 463.
Respectfully yours,

Carolyne Bird
Postgraduate Student
Forensic Expertise Profiling Laboratory

135
 Appendix B
Example Page from Disguise and Simulation Trial
Sample Booklet

136
 Appendix C
Extract of Disguise and Simulation Trial
Information and Answer Booklet

137
138
139
140
141
142
 Appendix D
Disguised Handwriting Provider Questionnaire

Forensic Expertise Profiling Laboratory

La Trobe University

Research project: Expertise of forensic handwriting examiners for determining

text-based natural and un-natural writing processes

PARTICIPANT QUESTIONNAIRE FOR HANDWRITING SAMPLE

COLLECTION

Please respond to the following:

Your Gender: Male Female (Please circle)

Your Age: 18 – 25 26 – 35 36 – 45 46 – 55 56 –65 66+ (Please circle)

Country where you first were taught to write: ....................................................................

First / primary language:....................................................................................................

Would you like to be informed of the results of this project? ………………………………

If yes please provide an address (preferably email) to which they can be sent

……………………………………………………………………………………………………..

For your handwriting sample ‘DISGUISE 1’ (lowercase print writing), what changes did
you introduce into your handwriting?

..........................................................................................................................................

..........................................................................................................................................

..........................................................................................................................................

For your handwriting sample ‘DISGUISE 2’ (uppercase print writing), what changes did
you introduce into your handwriting?

..........................................................................................................................................

..........................................................................................................................................

..........................................................................................................................................

143
 Appendix E
Extract of Declared Pairs Trial Answer Booklet

Expertise of Forensic Handwriting Examiners for determining text-

based natural and un-natural writing processes

ANSWER BOOKLET

Number of years since commencing forensic handwriting examination training:....

If none, do you have any knowledge of forensic handwriting examination? ...........

Length of handwriting training program (to the nearest year): ................................

Are you authorised by your laboratory, or have you released opinions regarding
forensic handwriting examination? ........................................................................

Would you like to be informed of the results of this project? ..................................

If yes, please provide an address (preferably email) to which they can be sent

...............................................................................................................................

Instructions

Please place a tick in the appropriate disguise strategy box/es for each
handwriting sample pair (001 - 140).

If you believe a disguise strategy that is not listed has been used, write it in the
"Other" box for that handwriting sample.

144
 Disguise strategy 001 002 003 004 005 006 007 008 009 010
slower writing speed
faster writing speed
altered letter
construction
disconnected letter
form
altered ascenders
&/or descenders
altered terminal
strokes (loops, ticks)
more angular
more rounded
altered stroke order
altered stroke
direction
altered proportions
shorter letters
taller letters
thinner letter forms
wider letter forms
decreased spacing
increased spacing
less tidy
neater
altered slope
altered baseline
altered punctuation
heavier average
writing pressure
lighter average
writing pressure
increased
connectivity
decreased
connectivity
introduction of
tremor
altered format of text
other (please
comment)

145
 Appendix F
Extract of Individual Samples Trial Answer Booklet

Expertise of Forensic Handwriting Examiners for determining text-

based natural and un-natural writing processes

ANSWER BOOKLET

Number of years since commencing forensic handwriting examination training:....

If none, do you have any knowledge of forensic handwriting examination? ...........

Length of handwriting training program (to the nearest year): ................................

Are you authorised by your laboratory, or have you released opinions regarding
forensic handwriting examination? ........................................................................

Would you like to be informed of the results of this project? ..................................

If yes, please provide an address (preferably email) to which they can be sent

...............................................................................................................................

Multiple choice answer instructions

The answer code for each of the questioned handwriting samples refers to your
opinion regarding whether the sample is naturally written or disguised. The
answer code you will write for each questioned sample will be a 1, 2 or 3. The
levels you have to choose from are as follows:

1. There is evidence that the questioned handwriting sample was written

naturally (that is, there is no evidence of a disguise process).

2. There is evidence that the questioned handwriting sample was written

using a disguise process.

3. No opinion can be expressed as to the process by which the questioned

handwriting sample was produced. If you submit an inconclusive (level
3) opinion, you are required to indicate in which direction you
believe the evidence points (forced opinion) by placing the
appropriate code (1 or 2) in the ‘Forced response’ box.

This forced response is an opinion that an examiner would not normally

express, but may suspect is correct, or be the examiner’s best guess. It is
clearly understood by the researchers that a forced response opinion would
not normally be expressed by examiners.

Disguise process: an attempt by the writer to purposefully alter their handwriting

in order to avoid being identified or for them to deny writing the text.

146
If you express an opinion that a handwriting sample was written using a disguise
process, you are required to record the strategy or strategies you believe the
writer used to disguise. Do this by placing a tick in the appropriate disguise
strategy box/es for each handwriting sample.
If you believe a disguise strategy that is not listed has been used, write it in the
"Other" box for that handwriting sample.

If your forced response is that a disguise process was used, recording strategies
is optional.

001 002 003 004 005 006 007 008 009 010
Answer code
Forced response
Disguise strategy
Internal differences in
writing speed
Inconsistencies in letter
construction
Unusual/forced/stylised
letter construction
Unusual/forced/stylised
ascenders/descenders
Unusual/forced/stylised
terminal strokes
Internal differences in
style
Internal differences in
size/proportions
Unusually small letters
Unusually large letters
Unusually thin letters
Unusually wide letters
Internal differences in
spacing
Tremor
Poor line quality
Suspicious variation in
slope
Suspicious variation in
baseline
Unusual punctuation
Unusual diacritics (eg i
dot)
Internal differences in
writing pressure
Internal differences in
connectivity
Other
(please comment)

147
 References
Agresti, A. (1996). An Introduction to Categorical Data Analysis. New York, NY: John Wiley &
Sons, Inc.
Alewijnse, L.C., van den Heuvel, C.E. & Stoel, R.D. (2011). Analysis of signature
complexity. Journal of Forensic Document Examination, 21: 37-49.
Alford, E.F. (1970). Disguised handwriting. Journal of Forensic Sciences, 15: 476-488.
American Standards for Testing Materials International. (2005). ASTM E2388-05 Standard
Guide for Minimum Training Requirements for Forensic Document Examiners.
Beacom, M.S. (1960). A study of handwriting by twins and other persons of multiple births.
Journal of Forensic Sciences, 5(1): 121 – 131.
Bird, C., Found, B. & Rogers, D. (2010). Forensic handwriting examiners' skill in
distinguishing between natural and disguised handwriting behaviours. Journal of Forensic
Sciences, 55: 1291–1295.
Boot, D. (1998). An investigation into the degree of similarity in the handwriting of identical
and fraternal twins in New Zealand. Journal of the American Society of Questioned Document
Examiners, 1: 70 – 81.
Cabanne, R.A. (1974). The Clifford Irving hoax of the Howard Hughes autobiography.
Journal of Forensic Sciences, 20(5): 5-17.
Cain, S. (1998). Disguised writings. Proceedings of the Association of Forensic Document Examiners
Meeting, Milwaukee, WI.
Chapran, J., Fairhurst, M.C., Guest, R.M. & Ujam, C. (2008). Task-related population
characteristics in handwriting analysis. IET Computer Vision, 2: 75-87.
Collaborative Testing Services, Inc (2010). CTS statement on the use of proficiency testing
data for error rate determinations.
Conway, J.V.P. (1959). Evidential Documents. Springfield, IL: Charles C Thomas.
Crane, A. (1999). The frequency of round handwriting in Edmonton, Alberta schools.
Canadian Society of Forensic Science Journal, 32: 169-174.
Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993).
Dewhurst, T., Found, B. & Rogers, D. (2008). Are expert penmen better than lay people at
producing simulations of a model signature? Forensic Science International, 180: 50-53.
Dick, D., Found, B., Lacey, K. & Rogers, D. (2004). Handwriting speed: Experimental
validation trial. Paper presented to the 17th International Symposium on the Forensic Sciences,
Wellington, New Zealand, March 28 –April 2.

148
Dillon, D.J. (1993). Fraud, counterfeiting and forgery in early California – questioned
document testimony and exhibits in a 1854 Mexican land grant case. Proceedings of the
American Association of Forensic Sciences Annual Meeting, Boston, MA, February 15-20.
Dyer, A.G., Found, B. & Rogers, D. (2006). Visual attention and expertise for forensic
signature analysis. Journal of Forensic Sciences, 51: 1397-1404.
Dyer, A.G., Found, B. & Rogers, D. (2008). An insight into forensic document examiner
expertise for discriminating between forged and disguised signatures. Journal of Forensic
Sciences, 53: 1154-1159.
Eiserman, H.W. & Hecker, M.R. (1986). FISH - computers in handwriting examinations.
Presented at the 44th Annual Meeting of the American Society of Questioned Document Examiners,
Savannah, Georgia, USA.
Eldridge, M.A., Nimmo-Smith I., Wing A.M. & Totty, R.N. (1984). The variability of
selected features in cursive handwriting: Categorical measures. Journal of the Forensic
Science Society, 24: 179-219.
Ellen, D. (1989). The Scientific Examination of Documents: Methods and Techniques. Chichester,
UK: Ellis Horwood Limited.
Embretson, S.E. & Reise, S. (2000). Item Response Theory for Psychologists. Mahwah, NJ:
Erlbaum Publishers.
Estabrooks, C.B. (2000). Measuring relative pen pressure to authenticate signatures. Journal
of the American Society of Questioned Document Examiners, 3(2): 56-64.
Evett, I.W. (1998). Towards a uniform framework for reporting opinions in forensic science
casework. Science and Justice, 38: 198-202.
Fitts, P.M. (1064). Perceptual-motor skills learning. In A.W. Melton (Ed.), Categories of Human
Learning (pp. 243-285). New York, NY: Academic Press.
Fitts, P.M. & Posner, M.I. (1967). Human Performance. Belmont, CA: Brooks/Cole.
Found, B. (1997). The forensic analysis of behavioural artefacts: Investigations of theoretical and analytical
approaches to handwriting identification. Unpublished Doctoral Thesis, La Trobe University.
Found, B. & Rogers, D. (1995). Contemporary issues in forensic handwriting examination.
A discussion of key issues in the wake of the Starzecpyzel decision. Journal of Forensic
Document Examination, 8: 1-31.
Found, B. & Rogers, D. (1998). A consideration of the theoretical basis of forensic
handwriting examination. International Journal of Forensic Document Examiners, 4: 109-118.
Found, B. & Rogers, D. (Eds.) (1999). Documentation of forensic handwriting comparison
and identification method: A modular approach. Journal of Forensic Document
Examination. 12: 1-68.

149
Found, B. & Rogers, D. (2003). The initial profiling trial of a program to characterize
forensic handwriting examiners’ skill. Journal of the American Society of Questioned Document
Examiners, 6: 72-81.
Found, B. & Rogers, D. (2005a). Problem types of questioned handwritten text for forensic
document examiners. In A. Marcelli & C. De Stefano (Eds.), Proceedings of the 12th
Conference of the International Graphonomics Society (pp. 8-12). Salerno, Italy, June 26-29.
Civitella, Italy: Editrice Zona.
Found, B. & Rogers, D.K. (2005b). Investigating forensic document examiners’ skill relating
to opinions on photocopied signatures. Science & Justice, 45: 199-206.
Found, B. & Rogers, D. (2008). The probative character of forensic handwriting examiners’
identification and elimination opinions on questioned signatures. Forensic Science
International, 178: 54–60.
Found, B., Rogers, D., Dick, D. & Lacey, K. (2002). Revision and Corrective Action
Package. Handwriting speed: Experimental validation trial March 2002. Melbourne,
Australia: La Trobe University.
Found, B., Rogers, D. & Herkt, A. (2001). The skill of a group of forensic document
examiners in expressing handwriting and signature authorship and production process
opinions. Journal of Forensic Document Examination, 14: 15-30.
Found, B., Rogers, D., Rowe, V. & Dick, D. (1998). Statistical modelling of experts’
perceptions of the ease of signature simulation. Journal of Forensic Document Examination,
11: 73-99.
Found, B., Sita, J. & Rogers, D. (1999). The development of a program for characterizing
forensic handwriting examiners’ expertise: Signature examination pilot study. Journal of
Forensic Document Examination, 12: 69–80.
Franke, K. & Grube, G. (1998). The automatic extraction of pseudodynamic information
from static images of handwriting based on marked grayvalue segmentation. Journal
Forensic Document Examination, 11 (Fall): 17-38.
Franke, K., Schomaker, L., Vuurpijl, L. & Giesler, S. (2003). FISH-New: A common ground
for computer-based forensic writer identification. Forensic Science International. 136: 84.
Galbraith, O., Galbraith, C.S. & Galbraith, N.G. (1995). The principle of the “Drunkard’s
Search” as a proxy for scientific analysis: The misuse of handwriting test data in a law
journal article. International Journal of Forensic Document Examiners, 1: 7-17.
Gamble, D.J. (1979). The handwriting of identical twins. Canadian Society of Forensic Science
Journal, 13: 11-30.
Gelman, A. & Hill, J. (2007). Data Analysis Using Regression and Multilevel/Hierarchical Models.
Cambridge: University Press.
150
Guest, R.M., Fairhurst, M.C., Abreu, M. & Linnell, T.A. (2011). Exploiting inference
mechanisms in the assessment of forensic document examination methodologies for
signatures. Journal of Forensic Document Examination, 21: 5-15.
Halder-Sinn, P. & Funsch, K. (1998). Intra-individual changes in handwriting features
depending on handwriting velocity. Journal of Forensic Document Examination, 11 (Fall):
1-16.
Harrison, W.R. (1958). Suspect Documents: Their Scientific Examination. London, UK: Sweet and
Maxwell Limited.
Hecker, M.R. (1993). Forensic Handwriting Examination. Translated from Forensische
Handschriftenuntersuchung: Eine systematische Darstellung von Forschung, Begutachtung und
Beweiswert by Manfred R. Hecker. Heidelberg, Germany: Kriminalistik Verlag.
Herkt, A. (1986). Signature disguise or signature forgery? Journal of the Forensic Science Society,
26: 257-266.
Hilton, O. (1982). Scientific Examination of Questioned Documents. (Rev. ed.). New York, NY:
Elsevier North Holland, Inc.
Holmes, L., Ostrum, B. & Barton, A. (2011). Online proficiency testing for signature com-
parison by forensic document examiners and non-examiners. Journal of the American
Society of Questioned Document Examiners, 14(1): 19-36.
Horton, R.A. (1996). A study of the occurrence of certain handwriting characteristics in a
random population. International Journal of Forensic Document Examiners, 2: 95-102.
Huber, R.A. & Headrick, A.M. (1999). Handwriting Identification: Facts and Fundamentals. Boca
Raton, FL: CRC Press.
Hull, J.M. (1993). The relationship between disguised handwriting and years of formal
education: Final results. Proceedings of the American Society of Forensic Sciences Annual
Meeting, Boston, MA, February 15-20.
Kam, M. (ca. 2010). Proficiency Testing and Procedure Validation for Forensic Document Examination.
Technical Support Working Group, Requirement Number: 000/IS-001-DREXEL-03-
FP.
Kam, M., Fielding, G. & Conn, R. (1997). Writer identification by professional document
examiners. Journal of Forensic Sciences, 42: 778-786.
Kam, M., Gummadidala, K., Fielding, G. & Conn, R. (2001). Signature authentication by
forensic document examiners. Journal of Forensic Sciences, 46: 884-888.
Kam, M., Wetstein, J. & Conn, R. (1994). Proficiency of professional document examiners
in writer identification. Journal of Forensic Sciences, 39: 5-14.
Kapoor, T.S., Kapoor M. & Sharma, G.P. (1985). Study of the form and extent of natural
variation in genuine writings with age. Journal of the Forensic Science Society, 25: 371-375.
151
Keckler, J.A. (1997). Felonious disguise. International Journal of Forensic Document Examiners, 3:
154-158.
Keele, S.W. & Summers, J.J. (1976). The structure of motor programs. In G.E. Stelmach
(Ed.), Motor Control: Issues and Trends (pp. 109-142). New York, NY: Academic Press.
King, B.H. (2003). Class characteristics of Mexican hand printing. Journal of the American
Society of Questioned Document Examiners, 6: 1-33
Konstantinidis, S. (1987). Disguised handwriting. Journal of the Forensic Science Society, 27: 383-
392.
Kumho Tire Co., Ltd. v. Carmichael, 526 U.S. 137 (1999).
Leung, S.C., Cheng, Y.S., Fung, H.T. & Poon, N.L. (1993). Forgery I - Simulation. Journal of
Forensic Sciences, 38: 402-412.
Leung, S.C., Chung, M.W.L., Tsui, C.K. & Cheung, Y.L. (1988). A comparative approach to
the examination of Chinese handwriting Part 3 - Disguise. Journal of the Forensic Science
Society, 28: 149-165.
Leung, S.C., Chung, M.W.L., Tsui, C.K., Cheung, W.L., Cheung, S.Y.L. & Mok, M.M.C.
(1987). A comparative approach to the examination of Chinese handwriting Part 2 -
Measurable parameters. Journal of the Forensic Science Society, 27: 157-173.
Li, C., Poon, N., Fung, W. & Yang, C. (2005). Individuality of handwritten Arabic numerals
in local population. Journal of Forensic Sciences, 50: 1-7.
Liang, K-Y. & Zeger, S. (1986). Longitudinal data analysis using generalized linear model.
Biometrika, 73: 13-22.
Liwicki, M., Malik, M.I., van den Heuvel, C.E., Chen, X., Berger, C., Stoel, R., Blumenstein,
M. & Found, B. (2011). Signature verification competition for online and offline
skilled forgeries (SigComp2011). In P. Kellenberger (Ed.), Proceedings of International
Conference on Document Analysis and Recognition 2011 (pp.1480-1484). Beijing, China,
September 18-21. Los Alamitos, CA: IEEE Computer Society.
Matley, M.M. (1999). Case citations relating to court ordered exemplars and disguise of same
as contempt of court and obstruction of justice: A discussion and interpretation.
International Journal of Forensic Document Examiners, 5: 146-174.
Mattijssen, E.J.A.T., van den Heuvel, C.E. & Stoel, R.D. (2011). The relation between
signature complexity and the perceived quality of signature simulations. In E. Grassi &
J.L. Contreras-Vidal (Eds.), Proceedings of the 15th Conference of the International Graphonomics
Society (pp.185-188). Cancun, Mexico, June 12 – 15. College Park, MD: University of
Maryland.
McAlexander, T.V. & Maguire, K.B. (1991). Eliminating ill-founded eliminations in
handwriting comparison cases. Journal of the Forensic Science Society, 31: 331-336.
152
Michel, L. (1978). Disguised signatures. Journal of the Forensic Science Society, 18: 25-29.
Michel, L. & Baier, P.E. (1985). The diaries of Adolf Hitler. Implication for document
examination. Journal of the Forensic Science Society, 25: 167-178.
Mohammed, L., Found, B., Caligiuri, M. & Rogers, D. (2009). Pen pressure as a
discriminatory feature between genuine and forged signatures. In A. Vinter & J. Velay
(Eds.), Advances in Graphonomics: Proceedings of IGS 2009 (pp. 26-29). Dijon, France:
Imprimerie Vidonne.
Morris, R.N. (2000). Forensic Handwriting Identification: Fundamental Concepts and Principles.
London, UK: Academic Press.
Muehlberger, R.J. (1990). Identifying simulations: Practical considerations. Journal of Forensic
Sciences, 35: 368-374.
Muehlberger, R., Newman, K.W., Regent, J. & Wichmann, J.G. (1977). A statistical
examination of selected handwriting characteristics. Journal of Forensic Sciences, 22: 206-
215.
National Research Council of the National Academies (2009). Strengthening Forensic Science in
the United States: A Path Forward. Washington, D.C.: The National Academies Press.
Nickell, J. (1997). The "Jack the Ripper diary": History or hoax. International Journal of Forensic
Document Examiners, 3: 59-63.
Osborn, A.S. (1929). Questioned Documents. (2nd Edition). Chicago, IL: Nelson-Hall Co.
Phillips, J.G, Noutsis, S., Hughes, C. & Rogers, D. (2000). Effects of writing speed upon
modes of signature simulations: A kinematic analysis. Journal of Forensic Document
Examination, 13 (Fall): 1-14.
Radley, R. (2005). Practised simulations – If a practised forger forges, does the practised
forgery favour the forger? Or does practice make perfect? Proceedings of the American
Society of Questioned Document Examiners Meeting. Montreal, Canada, August 11-16.
Railbert, M.H. (1977). Motor Control and Learning by the State Space Model. Technical report (AI-
TR-439), Artificial Intelligence Laboratory, MIT.
Ramsey Lines, S. & Franck, F.E. (2003). Triplet and sibling handwriting study to determine
degree of individuality and natural variation. Journal of the American Society of Questioned
Document Examiners, 6: 48-55.
Regent, J. (1977). Changing slant: Is it the only change? Journal of Forensic Sciences, 22: 216-
221.
Risinger, D.M., Denbeaux, M.P. & Saks, M.J. (1989). Exorcism of ignorance as a proxy for
rational knowledge: The lessons of handwriting identification "expertise". University of
Pennsylvania Law Review, 137: 731-792.

153
Risinger, D.M., Denbeaux, M.P. & Saks, M.J. (1998). Brave new "post-Daubert world" - A
reply to Professor Moenssens. Seton Hall Law Review, 29: 405-485.
Robertson, B. & Vignaux, T. (1995). Interpreting Evidence: Evaluating Forensic Science in the
Courtroom. Chichester, UK: John Wiley and Son Ltd.
Saks, M.J. & Koehler, J.J. (2005). The coming paradigm shift in forensic identification
science. Science, 309: 892-895.
Saks, M.J. & Risinger, D.M. (1996). Science and nonscience in the courts: Daubert meets
handwriting identification expertise. Iowa Law Review, 82: 21-74.
Saks, M.J. & VanderHaar, H. (2005). On the "general acceptance" of handwriting
identification principles. Journal of Forensic Sciences, 50: 1-8.
Sciacca, E., Langlois-Peter, M., Gilhodes, J., Margot, P. & Velay, J. (2008). The range of
handwriting variability under different writing conditions. Journal of Forensic Document
Examination, 19: 5-13.
Sciacca, E., Langlois-Peter, M., Gilhodes, J., Margot, P. & Velay, J. (2009). Comparing
handwriting variability under four postural conditions. In: A.Vinter & J. Velay (Eds.),
Advances in Graphonomics: Proceedings of IGS 2009 (pp. 108-111). Dijon, France:
Imprimerie Vidonne.
Sciacca, E., Langlois-Peter, M., Margot, P. & Velay, J. (2011). Effects of different postural
conditions on handwriting variability. Journal of Forensic Document Examination, 21: 51-
60.
Seaman Kelly, J. (2002). Habits observed in naturally written numbers. Journal of the American
Society of Questioned Document Examiners, 5: 58-66.
Simner, M.L. (1998). The origin of class characteristics: An empirical investigation of a
major principle in forensic document examination. Journal of Forensic Document
Examination, 11: 39-50.
Sita, J., Found, B. & Rogers, D.K. (2002). Forensic handwriting examiners' expertise for
signature comparison. Journal of Forensic Sciences, 47: 1117-1124.
Sita, J. & Rogers, D. (1999). Changes in forgers’ handwriting pressure related to the original
writer’s dynamics. Journal of Forensic Document Examination, 12 (Fall): 101-112.
Srihari, S.N., Cha, S.H., Arora, H. & Lee, S. (2002). Individuality of handwriting. Journal of
Forensic Sciences, 47: 856-872.
Srihari, S., Huang, C. & Srinivasan, H. (2008). On the discriminability of the handwriting of
twins. Journal of Forensic Sciences, 53: 430-446.
Srihari, S.N. & Leedham, G. (2003). A survey of computer methods in forensic document
examination. In H.L. Teulings & A.W.A. van Gemmert (Eds.), Proceedings of the 11th

154
 Conference of the International Graphonomics Society (pp. 278-281). Scottsdale, Arizona,
USA, November 2-5. Tempe, AZ: NeuroScript.
Summers, J.J. & Anson, J.G. (2009). Current status of the motor program: Revisited. Human
Movement Science, 28: 566-577.
Tabachnick, B.G. & Fidell, L.S. (2007a). Using Multivariate Statistics (5th ed.) Boston, MA:
Allyn and Bacon.
Tabachnick, B.G. & Fidell, L.S. (2007b). Experimental Designs Using ANOVA. Belmont, CA:
Duxbury.
Teulings, H.L. (1996). Handwriting movement control. In S.W. Keele & H. Heuer (Eds.),
Handbook of Perception and Action. Vol 2: Motor Skills (pp. 561-613). London, UK:
Academic Press.
Teulings, H.L., Thomassen, A.J.W.M. & van Galen, G.P. (1986). Invariants in handwriting:
The information contained in a motor program. In H.S.R. Kao, G.P. van Galen & R.
Hoosain (Eds.), Graphonomics: Contemporary Research in Handwriting (pp. 305-315).
Amsterdam, The Netherlands: North-Holland Publishing Company.
Thomassen, A.J.W.M. & van Galen, G.P. (1997). Temporal features of handwriting:
Challenges for forensic analysis. Journal of Forensic Document Examination, 10 (Fall): 85-
108.
Totty, R.N. (1991). Recent developments in handwriting examination. In A. Maehly & R.L.
Williams (Eds.). Forensic Science Progress 5 (pp. 91-128). Berlin, Germany: Springer-
Verlag.
Tsui, D.C.K. (1997). Disguised Chinese handwriting by Chinese-Canadian writers. Canadian
Society of Forensic Science Journal, 30: 137-150.
Turnbull, S.J., Jones, A.E. & Allen, M. (2010). Identification of the class characteristics in the
handwriting of Polish people writing in English. Journal of Forensic Sciences, 55: 1296-
1303.
Tytell, P.V. (1998). Pen pressure as an identifying characteristic of signatures: Verification
from the computer. Journal of the American Society of Questioned Document Examiners, 1:
21-31.
United States of America v. Roberta and Eileen Starzecpyzel, 880 F.Supp. 1027 (S.D.N.Y. 1995).
van Galen, G.P. (1980). Storage and retrieval of handwriting patterns: a two-stage model of
complex motor behaviour. In G. Stelmach & J. Requin (Eds.), Tutorials in Motor
Behaviour (pp. 567-578). Amsterdam, The Netherlands: North-Holland Publishing
Company.

155
van Galen, G.P. & van Gemmert, A.W.A., (1996). Kinematic and dynamic features of
forging another person's handwriting. Journal of Forensic Document Examination, 9 (Fall):
1-25.
van Gemmert, A.W.A. & van Galen, G.P. (1996). Dynamic features of mimicking another
person's writing and signature. In M.L. Simner, C.G. Leedham & A.J.W.M.
Thomassen (Eds.), Handwriting and Drawing Research: Basic and Applied Issues (pp. 459-
471). Amsterdam, The Netherlands: IOS Press.
van Gemmert, A.W.A., van Galen, G.P., Hardy, H.J.J. & Thomassen, A.J.W.M. (1996).
Dynamical features of disguised handwriting. Paper presented to the 5th European Conference
for Police and Government Handwriting Experts, The Hague, The Netherlands, November
13-15.
Vastrick, T.W. (1998). The uniqueness of handwriting. Journal of the American Society of
Questioned Document Examiners, 1: 4-7.
Webb, F.E. (1977). The question of disguise in handwriting. Journal of Forensic Sciences, 23:
149-154.
Will, E.J. (2011). Validating features for inferring handwriting speed from static trace. In E.
Grassi & J.L. Contreras-Vidal (Eds.), Proceedings of the 15th Conference of the International
Graphonomics Society (pp.185-188). Cancun, Mexico, June 12 – 15. College Park, MD:
University of Maryland.
Wing, A.M. & Nimmo-Smith, I. (1987). The variability of cursive handwriting measure
defined along a continuum: letter specificity. Journal of the Forensic Science Society, 27: 297-
306.
Zlotnick, J. & Lin, J.R. (2001). Handwriting evidence in federal courts – From Frye to Kumho.
Forensic Science Review, 13: 87-99.

156