You are on page 1of 26

Homicide Studies

http://hsx.sagepub.com/

Homicide in the Netherlands: On the Structuring of Homicide Typologies


Catrien Bijleveld and Paul Smit
Homicide Studies 2006 10: 195
DOI: 10.1177/1088767906290413

The online version of this article can be found at:


http://hsx.sagepub.com/content/10/3/195

Published by:

http://www.sagepublications.com

On behalf of:
Homicide Research Working Group

Additional services and information for Homicide Studies can be found at:

Email Alerts: http://hsx.sagepub.com/cgi/alerts

Subscriptions: http://hsx.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations: http://hsx.sagepub.com/content/10/3/195.refs.html

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Homicide Studies
Volume 10 Number 3
August 2006 195-219
© 2006 Sage Publications
10.1177/1088767906290413
Homicide in the Netherlands http://hs.sagepub.com
hosted at
On the Structuring of Homicide http://online.sagepub.com

Typologies
Catrien Bijleveld
NSCR Institute for the Study of Crime and Law Enforcement, Free University
Paul Smit
Research and Documentation Centre of the Dutch Ministry of Justice

The findings of a survey of all homicides in 1998 in the Netherlands are briefly presented.
After describing characteristics of the incident, the offender, and the victim, multivariate
relations between these characteristics are investigated. It appears that homicide cases are
structured in an interpretable way in which a previous classification can be accommo-
dated. The analysis, however, also indicates that homicide types do not constitute distinct
groups but instead rank along a circular continuum.

Keywords: homicide; homicide typologies; cluster analysis; Netherlands

A lthough homicide in the Netherlands has attracted quite a bit of public and media
attention, relatively little scientific research has been devoted to this topic. This
is at least partly due to the fact that, in absolute terms, the Netherlands witnesses few
homicides every year. Since 1980, the number of homicides (excluding attempts) has,
after an initial rise, fluctuated between 200 and 250 each year and has maybe even
receded slightly during the past 3 to 4 years.
Based on 1998 data, the Netherlands, with 1.5 homicide victims per 100,000
inhabitants, occupies a middle position. Countries such as the United States (6.1),
Belgium (2.1), Sweden (2.1), Finland (2.2), Greece (1.7), and Canada (1.8) have
higher rates of homicide per capita; Germany (1.2), England and Wales (1.4), Ireland
(1.4), Denmark (0.9), Norway (0.9), and Japan (1.1) have lower rates (Smit, 2001).
It should be noted that of all these, the United States assumes an extreme position.
In the Netherlands, homicide,1 defined as the nonjustifiable (i.e., not out of self-
defense) intentional killing of another person, is covered in articles 287 to 292 of the
penal code. Not covered in these articles are euthanasia, assistance to suicide, and abor-
tion. In the Netherlands, these are considered crimes only in exceptional circumstances.
The clearance rate for homicide that hovers between 70% and 80% is the highest of
all offences (apart from trivial offenses such as driving while under the influence) in the

Authors’ Note: Please address correspondence to Catrien Bijleveld, P.O. Box 792, 2300 AT Leiden,
Netherlands; e-mail: bijleveld@nscr.nl.

195

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


196 Homicide Studies

Netherlands. The dark number is believed to be very small. Nevertheless, a number of


people reported missing may actually have been murdered. As the number of persons still
missing after 1 year is only 10 to 15, even in the unlikely case that all these are in fact
homicides, they would entail only 5% to 7% of all cases (van den Eshof, Pieterse, Prins,
Schouten, & Tromp, 1996). Some missing active criminals, as well as missing illegal
immigrants, are believed to remain unreported and may also constitute a (small) dark
number. Citing a totally different reason for the existence of a dark number, German
pathologists stated recently that many apparently natural deaths are in fact “perfect mur-
ders,” that is to say homicides that go undetected by medical doctors or pathologists. A
Dutch pathologist stated that there is no reason to assume that the same would not hold
for the Netherlands. Figures as high as an additional 175 to 350 homicides per year have
been mentioned in this respect (Torenbeek, 1999). There is no way to validate this
number or to estimate the number of homicides among these unnatural deaths.
In the Netherlands, studies on (special types of) homicide cannot be conducted from
official statistics such as those published by Statistics Netherlands but have to be tailored
to the question at hand and thus need special data collection efforts. Homicide could not,
until recently, be extracted as such from the available statistical (judicial) databases.
In these databases, information is mostly aggregated (e.g., for all life crimes—a much
wider category). Police databases are maintained for operational purposes, and not all the
stored information is updated or validated; therefore, their accuracy and completeness are
questionable for research purposes. A specialized database, ViCLAS (Violent Crime
Linkage Analysis System), maintained by the National Police Agency (KLPD) contains
fairly extensive information on a selection of homicide cases, especially those that may
have been committed because of a bizarre, sexual, or psychotic motivation.
To extract reliable and valid information on all homicide cases, researchers must
go back to the police and court records for every event. Court records have the
advantage that the available information has been validated; police records have the
advantage that they contain information on unsolved homicides as well. Either way,
this is a fairly painstaking research effort, as neither police records nor court records
are stored centrally but according to regional differentiation.
Previous studies in the Netherlands investigated mainly trends and types of homi-
cides. Berghuis and de Jonge (1993) showed that homicide figures rose between 1989
and 1992, mainly because of the increase in homicides within the criminal milieu. Van
Eck (1997) investigated cases of homicide for family honor amongst Turkish immi-
grants in the Netherlands. Van de Port (2001) conducted a study of contract killings.
Van de Port (2001) and Kleemans, van den Berg, and van de Bunt (1998) conclude—
for the Netherlands—that it is virtually impossible to classify homicides as homoge-
neous and distinct groups. Even within fairly well-delineated categories, a lot of
variability remains; between categories, much uncertainty in classification remains.
All in all, relatively little general information is available about homicide in the
Netherlands, particularly about criminologically relevant aspects such as the relationship
between offender and victim, the setting in which the homicide occurred (as an escalation

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 197

to a quarrel, a contract killing, etc.), to what extent drugs were involved, whether there
are trends in the modus operandi, or the kind of weapons used. For an overview, see
Smit, Bijleveld, and van der Zee (2001) and Leistra and Nieuwbeerta (2003).
Homicides have theoretically relevant properties. These properties can be those of the
victim, the offender, and the event itself. Messner and Rosenfeld (1999) identified four
important correlates of homicide: social class, race, gender, and age. Race is not regis-
tered by the police or by any judicial agency in the Netherlands, but instead, ethnicity can
be inferred. Ethnicity is important not only because it correlates with socioeconomic sta-
tus and economic deprivation (Parker, 1989) but also because of the link with social dis-
organization through immigration (see, for instance, Wilson, 1987). Cultural differences
may be captured as well (Wolfgang & Ferracuti, 1967; for an overview, see Corzine,
Huff-Corzine, & Whitt, 1999). Daily activities (primarily employment) and criminal
involvement are further relevant indicators of social class and social integration. From the
work by Goldstein, Brownstein, and Ryan (1992), Parker (1993), and Parker and Rebhun
(1995), it can be inferred that alcohol and drugs are important not only for understand-
ing dynamics of the homicide incident but also for the classification of the homicide.
Drug and alcohol addiction are additional relevant indicators of social disintegration and,
with socioeconomic status and family problems, are situational precursors to homicide
events. At the incident level, an important aspect of the homicide event is the relation
between offender and victim. Other important aspects are the location where the incident
happened as well as the kind of violence that was exerted to kill the victim.
Based on the motive (if known) of the offender and the relation between offender
and victim, homicide events can be classified. Following Polk (1994), Smith and Zahn
(1999) remark that one of the major questions to ask in homicide research is whether
homicides can be distinguished into meaningful and homogeneous groups. Following
the classic work by Wolfgang (1958), it is generally assumed that homicides constitute
a heterogeneous group. Distinct subtypes of homicide have their own characteristics
and may be linked to different causes and correlates. The main dimensions along
which classifications are generally made are the victim-offender relation and the cir-
cumstances of the event. Other dimensions are instrumental/expressive (Salfati, 2000)
and primary/nonprimary (Smith & Parker, 1980). The classification that is employed
is generally dictated as much by the research questions as by the constraints posed by
data quality and research tradition (for an overview, see Flewelling & Williams, 1999).
One problem with such classifications is, however, that classifications are gener-
ally based on a logical scheme and one cannot be sure that the thus derived classifi-
cation actually manifests itself empirically. Previous studies have attempted to assess
the validity of classifications by studying whether disaggregated rates are differen-
tially influenced by causal factors and predictors (Smith & Parker, 1980; again, for
an overview, see Flewelling & Williams, 1999).
Instead of identifying such correlational variability across homicide types, it is also
possible to investigate the implied homogeneity itself by multivariate analysis of the rela-
tions between pertinent variables. If homicide types are indeed distinct and homogeneous

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


198 Homicide Studies

groups, multivariate analysis should be able to identify clusters of homicide types, to


each of which distinctive combinations of characteristics are particular.
This article investigates the extent to which such homogeneous and distinct
groups or types are present within the 1998 Dutch homicides. Our study does not
work from one single theoretical perspective, nor does it attempt to test one specific
theory on homicide. The framework from which the structuring of homicide typolo-
gies is investigated is exploratory.
In the following, we first briefly summarize the results from a study into all homi-
cides committed in 1998 in the Netherlands, reported on more extensively by Smit,
Bijleveld, and van der Zee (2001). Next, we employ exploratory multivariate analy-
ses to investigate how homicide events are structured and to what extent a previously
chosen classification of homicides is empirically supported.

Research Questions

The main research question in this article is whether homicide events in 1998 in
the Netherlands constitute a heterogeneous group in the sense that particular subgroups
of homicide events with particular properties can be distinguished. To address this
question, we first investigate whether the properties of the homicide events them-
selves, of offenders, and of victims are structured in an interpretable way. Next, we
investigate whether this structure reflects and thus supports the classification of
homicide events according to a logical scheme that was employed in the study by
Smit, Bijleveld, and van der Zee (2001).

Data Collection
We briefly summarize the description of the data collection method given by Smit,
Bijleveld, and van der Zee (2001). Homicide events were included if they had taken
place in 1998 and when a court decision had pronounced an offense to constitute a
homicide. However, as Smit, Bijleveld, and van der Zee reasoned, if only homicides
with a court decision are collected, a lot of cases are lost; some defendants have simply
not been led before a court, or the decision is still pending. For those cases, the offense
qualification as used by the prosecutor’s office was taken. Last, some cases had not (yet)
reached the prosecutor’s office. This may happen when the offender has been killed,
when the case is unsolved so that no offender has been identified, or when the offender
is under 12, which is the age of legal accountability in the Netherlands. For instance,
two very young boys who had killed an even younger child were included as offenders
even though they could not by law be (and were not) prosecuted, so no prosecution and
sentencing information was available. For those cases, police data had been adopted.
All in all, five different sources were used to collect and verify the data on homicide
in the Netherlands in 1998. These included sources from the police (databases and

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 199

interviews with police officers), a database containing all prosecution and court deci-
sions, and newspaper clippings.
Given legal definitions in the Netherlands and given the scope of the study, only
completed homicides were the object of study. The first reason for this was pragmatic:
Completed homicides constitute a clearly delineated group. The second reason was
conceptual: If attempts were included, there would be the risk of investigating a mix-
ture of criminologically distinct phenomena. Next, because of the data collection strat-
egy mentioned above, the term offender in this study is not used in a strictly legal
sense. Whenever the police had identified someone as a suspect or a defendant was
prosecuted but not yet convicted, we labeled the individual as an offender. Defendants
who were proclaimed not guilty of homicide were excluded, as were defendants iden-
tified by the police as homicide suspects but prosecuted for serious assault leading to
death. This means that in our classification of persons as homicide offenders, we adopt
the qualification of the last criminal justice agency dealing with the case. As plea bar-
gaining does not exist in the Netherlands, judges’ verdicts may be considered the most
valid qualification of an event. Thus, the common denominator in the offenses studied
is that there was intent to kill and that the outcome was lethal.

Results

Descriptive Results
The year 1998 witnessed 202 homicide incidents, with a total of 225 victims. For
155 of the incidents and for 176 of the victims, at least one offender had been identi-
fied, which gave a clearance rate of 77% (based on the incidents) or 78% (based on the
victims). In total, there were 230 offenders. Only one of these offenders was involved
in two homicide incidents in 1998. Thus, in the majority (111 of the 155 solved inci-
dents) of homicide incidents, there was one victim and one offender. The remaining
44 solved incidents had more than one victim and/or more than one offender. Of the
47 unsolved incidents, only 3 incidents had more than one victim. Apart from 2 cases
in which a father killed his entire nuclear family, mass murders, serial killings, and
group murders are highly uncommon in the Netherlands.
Inspired by other authors (Flewelling & Williams, 1999; Smith & Zahn, 1999),
in line with classifications previously adopted for the Netherlands, and restricted by
the way police in the Netherlands register homicides, a general and pragmatic way
of classifying homicides had been chosen. Two main principles or dimensions were
used to classify homicides: first, the motive of the offender and second, the relation
between offender and victim. The motive of the offender had four main categories:
criminal background, where the homicide is related to criminal activities in which
offender and victim are involved; sexual, where the homicide was preceded by a sex
offense; robbery, where the basic motivation of the offender is to steal property from

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


200 Homicide Studies

the victim; and dispute, where the homicide followed on or in the course of a dispute
or a fight. Where a homicide fell into more than one category, the first category was
leading.
If a homicide was classified as “criminal background,” a further division in three
categories was made: contract killing, a planned killing between criminals; drug
related, for example when a drugs dealer is killed in a dispute by a customer; and
criminal, other, when neither of the previous applied. In the case of “dispute,” a fur-
ther classification into three categories was made based on the relation between
offender and victim. These three categories were intimate, the so-called family homi-
cides; acquaintances; and strangers. In the Netherlands, family members and acquain-
tances may hire a contract killer to carry out a homicide, so we could in principle have
added this as a further subdivision. Because we feared that this would result in
extremely small subgroups and because, according to the available information, in all
likelihood it does not occur as often as has been reported for other countries (e.g.,
Black, 2000), we did not further subdivide within the dispute category.
Homicides that could not be classified in one of the four main categories were divided
into the following three categories: other, with another motive than the aforementioned;
psychotic, with a bizarre or seemingly psychotic motive; and unknown, when there is
not enough information available to classify the homicide. Table 1 gives the distribution
over the homicide types (taken from Smit, Bijleveld, and van der Zee, 2001).
To summarize the results further, many homicides occurred between intimates
and within a house. In line with findings from other countries (see Messner &
Rosenfeld, 1999), women constitute a small proportion of offenders but a much
larger proportion of victims. This applies even more so to children; contrary to the
evidence from the literature (Daly & Wilson, 1996), the percentage of stepparents
was very small here (it should be noted that absolute numbers were also very small).
Ethnicity was prominent in that nonethnic Dutch offenders were overrepresented rel-
ative to the proportion of nonethnic Dutch in the population (Corzine et al., 1999;
Wolfgang & Ferracuti, 1967). In interpreting this figure, it should be noted that tem-
porary residents are, however, not always in the denominator of the equation. As
attested by other authors (Goldstein et al., 1992; Parker, 1993; Parker & Rebhun,
1995), alcohol played a prominent role, either as a precursor to or during the event
or as a background characteristic of the offender. Victims led more mundane lives
than offenders in that they were more often married, less often had a criminal record,
were less often addicted, and so forth.

Exploration of Typologies
In this section, two questions are addressed. The first pertains to the multivariate
relations between the variables. The second pertains to the empirical validity of
homicide classification employed (see “Research Questions” above).

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 201

Table 1
Distribution of Homicide Types (N = 225 Victims)
Homicide Type n %

Criminal, contract killing 21 9


Criminal, drug related 17 8
Criminal, other 10 4
Sexual 9 4
Robbery 23 10
Dispute, intimates 70 32
Dispute, acquaintances 34 15
Dispute, strangers 8 4
Other 5 2
Psychotic 5 2
Unknown 23 10

Method

The first research question is addressed using multiple correspondence analysis or


homogeneity analysis. Using this technique, the researchers explored the multivariate
relations between the various characteristics of the homicides, offenders, and victims.
In homogeneity analysis, complicated multivariate data structures were reduced to low
dimensional plots that displayed respondents and their characteristics. The software
program HOMALS (Homogeneity analysis by Alternating Least Squares) in the soft-
ware package Statistical Package for the Social Sciences (SPSS) was used for this
purpose. The technique has been described elsewhere (Bijleveld, van der Kamp, et al.,
1998; Gifi, 1990; SPSS, 1990). We describe its technical properties in the appendix.
HOMALS is one of many clustering techniques. The advantage of using
HOMALS over other clustering techniques is that it makes no assumptions on the
distribution or measurement level of the data or on the number of clusters that need
to be identified. In addition, homicide events, as well as their characteristics, are
depicted in the solution. HOMALS is not affected by small numbers of missing
values on variables. As such, it is well suited for analysis of the type of exploratory
questions posed here.
HOMALS produces a solution in which respondents, as well as characteristics, are
positioned. In general, a HOMALS solution consists of a one-, two-, or higher dimen-
sional representation of analysis units (such as respondents) and their characteristics.
In this study, instead of respondents, homicides are used. The characteristics of these
homicides (e.g., the relation between offender and victim or the offender’s criminal
background) are studied. For ease of inspection, usually two dimensions are chosen,
and the same is done here.

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


202 Homicide Studies

In HOMALS, respondents are related to their characteristics according to the prin-


ciple that a respondent is located within the solution as closely as possible to his or her
characteristics and, conversely, that a characteristic is positioned as closely as possible
to the respondents that share that characteristic. In doing so, HOMALS will attempt to
achieve as much discrimination between the categories as possible. As it will not be pos-
sible to accommodate all patterns of characteristics of all respondents perfectly (some
respondents may share one characteristic but diverge on another variable), the technique
arrives at a compromise solution; a fit measure is indicative of how well the technique
has succeeded. No absolute value can be given as to what fit is acceptable, although
fit values under 0.3 or 0.4 are generally considered to be on the brink. Acceptability
depends on technical as well as substantive issues and is ultimately decided by the inter-
pretability of the solution. This type of technique has become an accepted method of
analysis in various branches of the social sciences. It is relatively uncommon in crimi-
nology. Examples of the use of comparable methods in homicide research can be found
in Salfati and Canter (1999) and Salfati (2000); outside of homicide research, Bijleveld,
Bakker, and Hendriks (1998) related personality characteristics and background vari-
ables to offense characteristics using OVERALS (SPSS, 1990).
The multivariate structuring of characteristics and respondents can be interpreted.
If, for instance, three characteristics are typically shared by a number of respondents,
the technique will in all likelihood position these characteristics in the same area of the
solution. The respondents are located close to these characteristics. These respondents
then constitute a so-called homogeneous group, as the group is homogeneous with
respect to these three characteristics.
In HOMALS, the position of a respondent in the solution is referred to as the object
score. The position of a characteristic (i.e., a category of a variable) is called a category
quantification. The category quantification of a characteristic is equal to the average
object score of all respondents who share that characteristic; conversely, a respondent’s
object score equals the average of the category quantifications of his or her characteris-
tics. By looking at the category quantifications, one can interpret the solution: Category
quantifications that are placed closely together tend to occur together. By looking at the
spread of the object scores in the solution space, one can see to what extent distinct
groups emerge: Clusters of object scores indicate that homogeneous subgroups exist for
which patterns or profiles of category quantifications are typical. By combining the
interpretation of the two—category quantifications and (groups of) object scores—one
can label the groups of respondents in terms of the analysis characteristics. In general,
the more peripheral groups are more particular and distinct; respondents with an object
score in the center have either a fairly average pattern or unusual combinations of char-
acteristics. Conversely, category quantifications that are placed in the center are not typ-
ical for certain individuals but are shared by many. Dimensions can be interpreted just
as they are in techniques such as factor analysis.
Of course, it is also possible to look at the average object scores of certain groups
of individuals. For instance, it is possible to compare the average object score of the

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 203

males with the average object score of the females or to compare the average object
scores of various age groups. This is sometimes referred to as the passive analysis of
variables (i.e., the categories of a variable are not taken along in the analysis, in which
case the technique would attempt to place them in the solution with as much dis-
crimination as possible), but differences between categories are computed afterward.
In the analysis to be carried out here, instead of respondents, homicides are analyzed.
Instead of analyzing the relation between respondents’ characteristics, the rela-
tions between characteristics such as victims’ age, cause of death, offenders’ crimi-
nal background, and so forth, are then analyzed. The second issue focuses on the
classification of homicides described above. The implicit assumption in such a clas-
sification is that the homicides do also empirically constitute disjunctive and homo-
geneous groups. The extent to which this is the case can be investigated using the
analysis results. Having carried out the multiple correspondence analysis, one can
inspect the average object scores of the 11 homicide types (see Table 1) and assess
the extent to which they assume different positions in the solution.
For investigating the clustering of offender, victim, and incident characteristics, the
categories from a total of 17 variables were used (see Table 2). Within the available
data, the chosen variables capture relevant aspects of the homicides under investiga-
tion in the sense that they tell us something about the violence that was exerted on the
victim (cause of death), the relationship between offender and victim, and the place
where the homicide incident took place. The offender’s, as well as the victim’s, age and
gender are other pertinent characteristics of the interaction between the two. Ethnicity
was included as well (see above and see Messner & Rosenfeld, 1999; Parker, 1989;
Wilson, 1987) as an indicator of cultural norms and strain induced by immigration.
The daily activities and criminal involvement are both indicative of social integration;
addiction and possible intoxication during the homicide (to alcohol as well as drugs)
as well as the setting in which the homicide took place are telling of lifestyle.
To evaluate the empirical support for the previous classifications of the homicide
incidents, researchers excluded these classifications from the analysis. If the subjective
classifications had been included, these homicide types would then have partly deter-
mined the structuring of the objective homicide characteristics. Then, it would not have
been possible to evaluate the classification against an objective criterion (i.e., against
the structuring of the characteristics). For the same reason, the motive variable was
excluded, as this had been the leading variable in the classification of homicides.
The analysis was conducted at the level of the homicide interaction or victim-
offender relation. For instance, if someone was killed by two offenders, two records
would be entered into the analysis data set. Or, as another example, if one incident
relates to two offenders and three victims, there are six victim-offender relations and
six records in the analysis data set. When no offender was known, still one victim-
offender relation for each victim was entered into the data set excluding all offender
characteristics. Thus, every homicide relation constituted one record (N = 295, of
which 49 were missing offender data), and for each characteristic of the incident, the

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


204 Homicide Studies

Table 2
Offender, Victim, and Event Characteristics in Multivariate Analysis
Homicide event
Location
Victim’s home
Victim and offender’s home
Offender’s home
Unknown house
Dwelling
Vehicle
Open nature
Street
Bar
Other outside

Relation offender/victim
Strangers
Intimate
Acquainted
Unknown

Cause of death/weapon
Shooting
Strangulation
Stabbing
Cuts
Beating
Other external force
Suffocation

Offender and victim characteristics


Age
< 18
18 to 24
25 to 34
35 to 44
45 to 65

Daily activities
Student
Employed
Unemployed
Housewife/houseman
Pensioner
Handicapped scheme
Criminal activities
Drug crimes
Prostitution

(continued)

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 205

Table 2 (continued)
Sex
Male
Female

Ethnicity
Dutch
Surinamese/Antillean
Turkish
Eastern Europe
Middle East
Africa

Addiction
Alcohol
Drugs
None

Intoxication
Drunk
Stoned
Sober

Criminal record
Yes
No

offender and the victim were entered. This choice implies that offenders with more
than one victim, as well as victims with more than one offender, play a proportion-
ally bigger role in the solution, as they appear several times. There is no ideal solu-
tion here, as leaving out relations implies loss of information either on the victim
side or on the offender side.

Multivariate Analysis of Relations


Between Homicide Characteristics
In the first analysis, HOMALS produced a solution in which the unsolved homi-
cides were contrasted with the solved homicides. Such a solution can emerge when
there are a great many missing values on variables, which is the case here with the
unsolved homicides, as all values for the offender are missing there. This has no the-
oretical significance, as such solutions are considered technically sound but other-
wise trivial; they simply capitalize on the missing values. The analysis was therefore
continued with only the victim-offender relations of the solved homicides (N = 246).
An added advantage is that as full information is available (namely, on victims and

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


206 Homicide Studies

on offenders), it is entered into the analysis. The disadvantage is, obviously and as
has been shown, that there may be significant differences between solved and
unsolved homicides (see, e.g., Mouzos & Muller, 2001). Inclusion of the unsolved
homicides would have meant running the analyses without offender information,
which has great disadvantages as well.
After a number of sparsely filled categories had been recoded into larger cate-
gories to prevent outliers, a satisfactory solution was reached.2 It should be noted
that the fit of the solution is fairly low (0.51). In general, the first explanation for
such a low fit is that the data contain quite a bit of noise. Another explanation for
these data may be that the clustering of characteristics is actually higher dimen-
sional. The price paid for a higher dimensional analysis is, however, a much more
difficult interpretation, so the two-dimensional analysis was chosen. To be able to
study the emerging patterns, researchers plotted the category quantifications. Figure
1 depicts this plot of category quantifications.
The solution can be interpreted by inspecting patterns while moving clockwise
through the periphery of the solution. As noted above, the more central categories
are less typical, and for this analysis, not much significance will be attached to them.
The analysis then proceeds as follows. In the bottom of the left-hand corner, a small
number of homicides of criminally active Eastern Europeans committed by Eastern
Europeans have been placed. Very little else is typical for the few homicides here, as the
analysis has not positioned any other variables nearby. This means that, apparently,
homicides among eastern Europeans took place, in 1998, within fairly strict ethnic
boundaries. Moving clockwise in the plot, homicides carried out in bars are encoun-
tered, with Surinamese/Netherlands Antillean (Surinam/NA) victims between 18 and 24
years of age, some of whom are students. Further upward are homicides carried out
on the streets by criminally active Surinamese/NA, where the victim was involved in
criminal activities related to drugs. These homicides would be labeled as crimi-
nal/drugs related. Close by and in between this group and the next group are located
homicides carried out on the street where the victim dies from gunshot wounds.
Victims tend to have a criminal record here, and victim and offender are acquainted.
Victims are often between 25 and 34 years of age. Again, a little further on, a group
of homicide events has been placed where victims as well as offenders are of Turkish
origin or from the Middle East; addiction to alcohol or hard drugs is typical here, as
well as homicides committed while the victim was alcohol or drug intoxicated. Higher
up in the figure, more characteristics of the events, such as the ethnicity of the offender,
the cause of death, and so forth, tend to be unknown. This is often the case when the
offender is a stranger to the victim. Victims tend to be between 35 and 44 years of age
here, and unemployment is typical for victims. These homicides one could character-
ize as the homicides of the drug-addicted socially excluded, not necessarily drug-
related like the previous type. Shootings, beatings, and (to a lesser extent) stabbings
are typical. Victim and offender may be acquainted or strangers. Homicides are gener-
ally carried out in a house: either some unknown house or a dwelling (here the offender

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 207

Figure 1
Category Quantifications

strangers
1 hard addiction victim’s home
alcohol addiction beating handicapped scheme
stoned (hd)
drug addiction drunk <65
unemployed alcohol addiction
Middle East Turkish drunk Dutch vehicle
drug crimes <35
Middle East Dutch
Turkish criminal record female strangulation
dwelling open nature
unemployed other external force
unknown house 35-44 prostitution
criminal record stabbing female housewife/-man
0 >45 other outside
acquainted
pensioner
Surinamese/Antillean
dimension 2

offender’s home
street 25-34 <18
shooting Africa
student sober
suffocation intimate
criminal activities not addicted victim & offender’s home
Surinamese/Antillean
sober no criminal record
<24
bar other
−1 Africa

>65
<12 <4

cuts

−2
eastern Europe

Legend
Helvetica bold face - relation between victim and offender,
cause of death and event location
eastern Europe Helvetica plain face - offender characteristics
Helvetica italic face - victim characteristics

−3
−2 −1 0 1 2 3

dimension 1

is often an acquaintance of the victim) or in the victim’s house (here the offender is
more often a stranger to the victim).
In the top of the picture, aggressive homicides involving beatings with objects as
the typical cause of death are located. These homicides are often committed in the
home of the victim; victims may be handicapped (i.e., in a special scheme for hand-
icapped workers), Dutch pensioners, and between 45 and 65 years of age. Some of
the victims were found killed in a vehicle. The offenders are often Dutch. Typical for
this group as well as the homicides of the drug addicted is that the victim and

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


208 Homicide Studies

offender are often strangers. Also, the offender is often addicted to alcohol and drunk
while committing the homicide. These homicides could perhaps best be typified as
rage or stranger murders that may possibly be a spontaneous “overreaction to a con-
flict situation” precipitated by the consumption of alcohol.
Toward the right-hand side of the solution, more and more female victims are
found. These women often died from strangulation or from causes such as poison-
ing or suffocation. Somewhat less peripheral, but still clearly discernible, are several
homicides of drug-addicted prostitutes. Here, the homicide often took place in the
open, such as in a park or in the countryside; the offender may be (but the position
of this category is fairly central) slightly older and known to the victim.
The entire lower, right-hand side of the figure is taken up by homicides where
victim and offender are intimate relations. At the extreme end are the truly defense-
less victims, such as very young children and people older than 65. Young offenders
(younger than 18) are also positioned here, as are pensioners. Oftentimes, the homi-
cide is committed in the home of the offender and victim, and cuts are typical for
victims. The offender often does not have a criminal record, and the offender, as well
as victim, often cares for children at home as a daily occupation.

Empirical Support for Classification


Moving on to the second question, the average object scores of the homicides in
each of the 11 types of homicides were computed. These are graphically represented
in Figure 2.
Figure 2, in combination with the plot of object scores (not shown here, but later
on in Figure 3), shows that the various categories do not represent distinct and homo-
geneous classes of homicides. The object scores are fairly evenly spread, with little
or no clustering. The same, as shown above, applies to the category quantifications:
They are spread all through the solution and do not cluster into distinct and isolated
patterns. In combination with the fairly low fit of the analysis, the conclusion from
the analysis is, therefore, that the homicides cannot be separated into clearly distinct
groups.
The homicide types “dispute acquaintance” and “unknown” are placed quite cen-
trally, indicating that the technique has trouble identifying unique characteristics for
them. An explanation for this placement of the acquaintance disputes is that they are
quite common and will thus have a fairly average profile. Apart from two types of homi-
cides, one can see how the other homicide types are ranked along a more or less circular
structure: from “contract killing” to “drug related” homicides to “criminal other,” then
from “dispute strangers” on to “other,” “robbery,” and “psychotic,” and, turning to “sex-
ual” and “intimate” homicides, a circular or oval shape is formed.
Combining the placement of the homicide types in this plot with the information
from Figure 1, one can deduce the following. In the lower, left-hand side of the solu-
tion, the contract killings and drug-related homicides are placed; they share a number

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 209

Figure 2
Position of Average Object Scores of Homicide
Classifications in Analysis Structure

1 psychotic
other robbery

dispute strangers
criminal other
dimension 2

dispute acquaintances
0 unknown sexual

intimate
drugs related

contract killings
−1

−2
−2 −1 0 1 2
dimension 1

of characteristics, as they have been placed not far apart. From Figure 1, it can be seen
that the offenders here are not addicted themselves; that there is shooting; that homi-
cides are often carried out on the streets; by Surinamese/NA on Surinamese/NA, and
by Eastern Europeans on Eastern Europeans; and that the victims have a criminal
record. These homicides therefore tend to be carried out by offenders in charge: not
addicted themselves but dealing or involved in the drug trade, using a “distant” means
to kill the victim, and who have a criminal record. These are the criminal settlings of
accounts.
Higher up in the solution, “other criminal disputes” and “disputes between
strangers” are placed closely together. These homicide types share the fact that quite

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


210 Homicide Studies

a few offenders (as well as victims) are originally from the Middle East or Turkey,
victims are addicted to alcohol or drugs, the offender is often intoxicated, and the
offender often has a criminal record. Tentatively, one could say that these offenders
and victims are the socially excluded. They may be acquainted or unacquainted and
are slightly older than those in the previous two types.
Moving on, homicides with a property motive and psychotic murders have been
placed. What these homicides share (combining their positioning with the information
from Figure 1) is that the offender and victim tend to be Dutch and strangers here. The
homicides become more angry and frenzied; the cause of death is often beating, and the
offender is more often drunk during the homicide. Thus, from the bottom of the solution
to the top of the solution, a gradual move has been made from homicides that are fairly
cool and distant settlings to homicides that arise from intoxicated disputes to homicides
that are angry and involved. It is remarkable that the “other” type of homicide is closest
to the robbery type. The property motive is helpful here to interpret the homicides located
a little more to the right, of older and handicapped people, of those in the home of the
victim, and by addicted offenders often under the influence, as it can be seen that these
homicides are probably quite often robberies, with fairly defenseless victims.
Next, the sexual and intimate homicides are encountered about midway on the sec-
ond axis of the solution and well along the first axis. It is not surprising that the sexual
homicides more often have female victims as well as prostitutes. The victim is typically
killed by strangulation or other kinds of external force. Offenders can be much older.
Lower down, the intimate homicides are typically carried out by offenders with no crim-
inal record at all, in the home of the offender and victim, and cuts are typically found on
the sometimes very young victims here. In a sense, these are the emotional/intimate set-
tlements, as opposed to the business settlements in the opposite corner of the structure.
Based on the interpretation of the combined figures, one could, tentatively, interpret
the first dimension as a business-personal dimension. The second dimension could be
interpreted as a personal settlement–impersonal escalation/angry brawl dimension.
Thus, it appears that the various homicide types do not imply dramatically distinct
or unique profiles. Rather, shifts are in all likelihood fairly gradual, as homicide types
are often located fairly closely, in which case homicides in one type will have certain
characteristics but may also share these characteristics with homicide types located
nearby.

Unsolved Homicides
There is one group of homicides that is now completely left unconsidered, and these
are the unsolved homicides. The issue whether unsolved homicides are inherently dif-
ferent from solved homicides is, in itself, interesting. One could, for instance, presume
that unsolved homicides would less often be intimate homicides (these being easier for
the police to solve) and would more often be contract killings (see also Mouzos &
Muller, 2001). If that were the case, exclusion of the unsolved homicides would have

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 211

skewed the data. To investigate whether such a distortion has occurred for the
Netherlands, one would want to check whether the unsolved homicides are distributed
evenly in the solution (in which case, such skewing is less likely) or whether they
resemble certain homicide types more (in which case, skewing is more likely).
As the unsolved homicides did not play a role in the solution, positions for the
unsolved homicides have to be computed afterward. Just as it is possible to find cat-
egory quantifications for the categories of variables that did not play a role in the
solution (i.e., the passive variable homicide type, see above), it is also possible to
find object scores for the homicides that did not play a role in the solution. In that
case, the category values of the variables on which valid measurements are available
for every unsolved homicide (i.e., the victim and events characteristics) are replaced
by the respective category quantifications, and the average for each homicide on
these category quantifications is computed. This average is now the “passive” object
score for this homicide. This procedure is discussed in detail by van de Geer (1993).
Applying this procedure, we obtain the results depicted graphically in Figure 3.
Figure 3 shows that the unsolved homicides are not all concentrated in one partic-
ular area. In fact, they do have a fairly even spread over the two dimensions. On visual
inspection, they do tend to occur more often in the top of the structure, where the
victim and offender are often strangers and homicides are harder to solve; this is, to a
certain extent, in accordance with Mouzos and Muller’s (2001) findings that victims
of unsolved homicides tend to be older and that such homicides tend to occur in the
course of another crime. The left-hand side of the picture where the crime- and drug-
related homicides occur also has perhaps an overrepresentation of unsolved homicides,
although absolute numbers are small. Mouzos and Muller indeed also found that
unsolved homicides tend to be carried out more often with a firearm. Unsolved homi-
cides are, however, also located in the area where intimate homicides are typically
located. These results are also in line with the findings of Petee, Weaver, Corzine,
Huff-Corzine, and Wittekind (2001) and Regoeczi and Riedel (2003). So, all in all, we
do not see evidence that they typically cluster in one particular area of the plot. The
passive positioning of the unsolved homicides shows therefore that, by having had to
exclude these unsolved homicides from the analysis, the data may have been skewed
somewhat but in all likelihood not to a great extent.

Discussion

This study did not attempt to be theoretical but rather to describe and explore the
data. The distribution of several characteristics deemed by previous studies to be rel-
evant was described briefly. In addition, multivariate relations between the variables
were explored, and the empirical support for a pragmatic classification assessed. The
study showed that unsolved homicides that had to be excluded from the multivariate
analysis for technical reasons probably did not seriously distort our conclusions.

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


212 Homicide Studies

Figure 3
Positioning of Object Scores of Solved and Unsolved Homicides

0
dimension 2

−1

−2

−3

−4
−3 −2 −1 0 1 2 3 4
dimension 1

solved homicides unsolved homicides

The multivariate analysis first showed that homicides can be structured meaning-
fully. Homicide events are structured in a two-dimensional solution, for which the
axes can be interpreted as a business-personal dimension and a settlement-escalation/
angry brawl dimension. Along the first axes, we say how homicides ranged from
business-like settlings of organized crime feuds to the settling of family issues,

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 213

where for example a husband killed his wife. Along the second dimension, homi-
cides ranked from more planned and distant executions to angry, involved killings
that often took place in the course of a fight. The second dimension bears resem-
blance to the instrumental-expressive dimension often found in homicide research as
well as to the planned-spontaneous distinction found in many homicide studies.
Further analyses showed that a homicide typology based on pragmatic and theoreti-
cal notions (Smit, Bijleveld, and van der Zee, 2001), although not entirely rejected,
may not be empirically optimal. First, it appears defensible to aggregate some homi-
cide subtypes into larger groups. For instance, the contract killings and drug-related
homicides are placed quite close in the solution. This means that they are empirically
indistinguishable: They share the same characteristics, which is why the technique
places them so close to each other. The typology may thus be theoretically relevant
but empirically not very useful. The question is, of course, particularly given the fact
that the Netherlands does not witness a lot of homicides each year, whether a clus-
ter analysis on the data for subsequent years will yield the same structure and the
same grouping of homicides. Although analyses of slightly different subsets of this
data set yielded more or less the same results, the stability of the empirical classifi-
cations should be tested by combining data from several years.
Second, the study gave an indication that instead of a grouping of homicides into
disjunct categories, they should perhaps be ranked along a circular or circumplex
structure. Such circumplex structures have also been found in the study of personal-
ity characteristics (Wiggins, 1996) and in the study of emotions (Plutchik & Conte,
1997). For the 1998 homicides in the Netherlands, the continuum ranged from con-
tract killings as the most business like and distant, settling-of-dispute type of homi-
cide to disputes between strangers and back to emotional-intimate settling types of
homicides. Results from studies by others in other areas of criminology can be rec-
ognized in the multivariate structure analyzed here: For instance, the instrumental-
expressive dimension that is employed in many studies on violence (for results on
homicide, see, e.g., Salfati, 2000) might also provide appropriate labelling for the
first dimension of the analysis. The advantage of employing dimensions over distinct
classes is that the inevitable “fuzzy” homicides can be easily accommodated. Causes
and correlates are then not tied to distinct types of homicide; instead, their impact
can be investigated relative to the positioning of the homicide on the respective
dimensions. This would perhaps lead to a much more intricate model of lethal
violence.
The fact that the data also contained unsolved cases may be considered a weak
point of this study. And for those offenders who have not yet been found guilty by a
judge, there is no certainty that this offender has actually committed the offense. On
the other hand, excluding these cases entails loss of substantial information, espe-
cially in regard to victims and the circumstances of the event. The advantage of hav-
ing as complete data as possible was therefore deemed to outweigh the
methodological problems.

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


214 Homicide Studies

One question that remains is whether the diffuse empirical classification is not
simply an artifact of the classification that this study employed. Perhaps, one could
argue, if a different classification had been employed, the homicide types might have
emerged as much more distinct and well-clustered groups. This is a matter that will have
to be studied by further in-depth analyses. We believe that chances are slight that this
would lead to radically different structurings. First, the structures of the event and victim
and offender characteristics are independent of the classification chosen. Second, the
low fit itself indicates that there does not exist in this data set a subset of unique differ-
ent homicides, clearly and wholly separate from the rest. Homicides apparently share
many characteristics, which inevitably gives fuzzy classification. One could, of course,
work the other way around and derive typologies not from theoretical notions or prag-
matic considerations but, rather, choose clusters as they emerge empirically. Ultimately,
however, homicide typologies should be employed on empirical and theoretical bases.

Appendix
Homogeneity Analysis by Alternating Least Squares (HOM)

HOMALS is a method for multiple correspondence analysis that uses optimal scaling. Optimal
scaling techniques assign numerical values to the categories of variables with a limited number
of categories. This scaling process is also called quantification. After quantification, variables
can be treated as if they were continuous variables. A host of such optimal scaling techniques
are described in Gifi (1990). Here we somewhat formally describe one type of such a technique,
the HOMALS technique. In doing so, we base ourselves on Bijleveld and Van der Burg (1998).
The ALS algorithm that the optimal scaling techniques use is called such as it consists of
two alternating steps. In one step, the categories of the variables are rescaled. In the other step,
a summary analysis solution is reached. The steps are carried out consecutively, arriving at a
final solution in which the loss function is not decreased substantively anymore. As the quan-
tification of the categories is carried out alternatingly with the solution optimization, this
means that the two are in a sense dependent; that is, an analysis with different variables that
interrelate differently might produce different quantifications.
An asset of optimal scaling techniques is that they do not make any distributional assump-
tions. Second, these techniques can easily handle missing data. The techniques are particularly
useful when the analysis objective is not so much confirmatory data analysis or statistical
model testing but, rather, data exploration and summarization.
Multiple correspondence analysis (Benzécri, 1973) is often used when the objective is to
investigate the interrelations between a large number of categorical variables. The categories of
these variables may be ordered or not; what is essential is that each variable has a limited number
of categories. To investigate the relations between the variables, we could of course also
tabulate them in cross-tables. We would then with M variables have to investigate M(M – 1)
cross-tables. In addition to this being cumbersome, we are not interested in so many bivariate
tabulations but rather in a summary of the most pertinent multivariate relations in the data. We
would like to be able to investigate the multivariate relations between all categorical variables
simultaneously.

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 215

In multiple correspondence analysis, this is achieved as follows. First, the categorical vari-
ables are characterized by their so-called indicator matrices. The N × kj indicator matrix Ij of
each variable j contains dummy variables that show to which kj categories a respondent
belongs (1) and to which categories he or she does not belong (0). The quantifications of the
categories of each variable j are contained in a kj × 1 vector bj. The N × 1 product Ijbj thus
contains the transformed category values for all the respondents. Simultaneously, an N × 1
vector z of so-called object scores is defined, which should summarize similarities and dif-
ferences between respondents as measured by the variables in the analysis.
Multiple correspondence analysis now aims at maximizing the similarity between all
rescaled variables and the object scores z, that is, at maximizing the average squared correlation
between the object scores z and the quantified variables Ijbj, which amounts to minimizing the
following loss function:
M
∑ SSQ (z – Ijbj),
j=1

where SSQ (..) stands for the sum of squares of the elements of the vector between brackets.
As the equation can be trivially minimized by setting all elements of the vector z equal to zero
as well as all elements of each vector bj, it is generally requested that z′ z = N; also, the object
scores are generally defined in deviation from a mean of zero.
The expression defined above gives a one-dimensional solution. More dimensions can be
added by defining a p dimensional matrix Z instead of a vector z and by giving each category
as many quantifications as dimensions, thus arriving at a kj × p dimensional matrix Bj of cat-
egory quantifications. HOMALS is available in the SPSS procedure Categories.
A first measure for interpreting solutions is the badness of fit or loss, the minimum value of
M
∑ SSQ (Z – IjBj).
j =1

The badness of fit is inversely related to the goodness of fit, which is defined as the number
of dimensions minus the loss. Another way to define the goodness of fit is in terms of the so-
called eigenvalues. Each dimension of the solution has an eigenvalue. The eigenvalue of a
dimension corresponds to the mean variance of all the quantified variables accounted for by
that dimension, in short as an indication of explained variance. The eigenvalues in multiple
correspondence analysis are thus comparable to the eigenvalues in principal components
analysis, although it should be kept in mind that the eigenvalues in multiple correspondence
analysis reflect the explained variance of the quantified variables, not that of the original vari-
ables. The sum of the p eigenvalues equals the goodness of fit of the total solution.
Each participant is assigned a point in p dimensional space. The object scores can be rep-
resented in an orthonormalized system. They then form a cloud of points with (N) unit vari-
ance for each dimension. As stated above, the categories of variables on which participants
score simultaneously receive similar quantifications. Conversely, the object scores of partici-
pants whose answering patterns are similar will also be fairly similar, and participants with
different answering patterns will have different object scores.
To illustrate the interpretation of such spaces with object scores and category quantifica-
tions, we give a small illustrative example. Suppose we have the following data for four
respondents and two categorical variables.

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


216 Homicide Studies

Name Hair Color Sex

John Blond Male


Beth Blond Female
Pete Black Male
Mary Red Female

In this data set, two respondents share the characteristic “male” (John and Pete), and two
share the characteristic “female” (Beth and Mary). The technique should position John and
Pete close to the position (i.e., the two-dimensional category quantification) of the character-
istic “male” and should position Beth and Mary close to the position of the characteristic
“female.” However, at the same time, John and Beth share the characteristic “blond,” so that
John and Beth should also be positioned in the vicinity of this characteristic. It is clear that
not all these demands can be met at the same time and that compromises will have to be made.
For the given example, such a compromise could look like this:

John

blond
Beth
male

female
Pete
black
Mary
red

Pete is the only one who has black hair, and Mary is the only one with red hair. It is there-
fore no problem for the technique to position Pete close to the characteristic “black hair” and
Mary close to the characteristic “red hair.” In fact, Pete and “black hair” are positioned in exactly
the same spot, and the same goes for Mary and “red hair.” For the other hair colors and for sex,
compromises have to be found. We see in the figure how the property “male” is positioned
between John and Pete, who share this property, and how Beth and Mary have landed at com-
parable distances from the characteristic “female.” The male John and the female Beth share the
characteristic “blond hair,” and this characteristic is positioned between the two of them.
This correspondence between category quantifications and object scores is also a formal
one. The quantification of a category is equal to the centroid of the object scores of all

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 217

participants who have scored that category. Conversely, the object score of a participant is
equal to a multiple of the average category quantification of all categories that belong to this
participant. Thus, category quantifications and object scores are intrinsically linked. Both are
part of the same solution and should therefore be interpreted in relation to one another.

Notes
1. The Dutch language does not have just one word that is equivalent to homicide in English. Instead,
homicide encompasses two acts: premeditated murder (moord in Dutch) and nonnegligent manslaughter
(doodslag in Dutch).
2. A number of object scores were quite peripheral. On removal of these cases, however, the remain-
ing part of the structure did not change meaningfully, so the peripheral cases were kept in the analysis.

References
Benzécri, J. P. (1973). L’analyse des données [Data analysis]. Paris: Dunod.
Berghuis, A. C., & de Jonge, L. K. (1993). Moord en doodslag in 1989 en 1992—Een secundaire analyse
[Homicide in 1989 and 1992—A secondary analysis]. Tijdschrift voor Criminologie, 35, 55-62.
Bijleveld, C. C. J. H., Bakker, L., & Hendriks, J. (1998). Contact crimes in relation to neuroticism, impul-
siveness, conscience formation and intelligence: An exploratory discriminant analysis. Psychology,
Crime & Law, 4, 341-360.
Bijleveld, C. C. J. H., & Van der Burg, E. (1998). Analysis of longitudinal categorical data using optimal
scaling techniques. In C. C. J. H. Bijleveld & L. J. Th. van der Kamp (Eds.), Longitudinal data analy-
sis (pp. 46-154). London: Sage.
Bijleveld, C. C. J. H., van der Kamp, L., Mooijaart, A., van der Kloot, W. A., van der Leeden, R., & Van
der Burg, E. (Eds.). (1998). Longitudinal data analysis. London: Sage.
Black, J. A. (2000). Murder for hire: An exploratory study of participant relationships. In P. H. Blackman,
V. L. Leggett, B. L. Olson, & J. P. Jarvis (Eds.), The varieties of homicide and its research: Proceedings
of the 1999 Meeting of the Homicide Research Working Group (pp. 236-248). Washington, DC: Federal
Bureau of Investigation.
Corzine, J., Huff-Corzine, L., & Whitt, H. P. (1999). Cultural and subcultural theories of homicide. In
M. D. Smith & M. A. Zahn (Eds.), Homicide: A sourcebook of social research (pp. 42-57). Thousand
Oaks, CA: Sage.
Daly, M., & Wilson, M. (1996). Violence against stepchildren. Current Directions in Psychological
Science, 5, 77-81.
Flewelling, R. L., & Williams, K. R. (1999). Categorizing homicides: The use of disaggregated data in
homicide research. In M. D. Smith & M. A. Zahn (Eds.), Homicide: A sourcebook of social research
(pp. 96-106). Thousand Oaks, CA: Sage.
Gifi, A. (1990). Nonlinear multivariate analysis. Chichester, UK: Wiley.
Goldstein, P. J., Brownstein, H. H., & Ryan, P. J. (1992). Drug-related homicide in New York: 1984 and
1988. Crime & Delinquency, 38, 459-476.
Kleemans, E. R., van den Berg, E. A. I. M., & van de Bunt, H. G. (1998). Georganiseerde criminaliteit
in Nederland [Organized crime in the Netherlands] (O&B Series 173). The Hague, Netherlands:
Wetenschappelijk Onderzoek en Documentatiecentrum Research & Documentation Center, Ministry
of Justice.
Leistra, G., & Nieuwbeerta, P. (2003). Moord en doodslag in Nederland. Amsterdam: Prometheus.

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


218 Homicide Studies

Messner, S. F., & Rosenfeld, R. (1999). Social structure and homicide. In M. D. Smith & M. A. Zahn
(Eds.), Homicide: A sourcebook of social research (pp. 27-41). Thousand Oaks, CA: Sage.
Mouzos, J., & Muller, D. (2001). Solvability factors of homicide in Australia (Trends and Issues in
Criminal Justice Series No. 216). Canberra: Australian Institute of Criminology.
Parker, R. N. (1989). Poverty, subculture of violence and type of homicide. Social Forces, 67, 983-1007.
Parker, R. N. (1993). Alcohol and theories of homicide. In F. Adler & W. Laufer (Eds.), Advances in crim-
inological theory (vol. 4, pp. 113-142). New Brunswick, NJ: Transaction.
Parker, R. N., & Rebhun, L. (1995). Alcohol and homicide: A deadly combination of two American tra-
ditions. Albany: State University of New York Press.
Petee, T. A., Weaver, G. S., Corzine, J., Huff-Corzine, L., & Wittekind, J. (2001). Victim-offender rela-
tionships and the situational context of homicide. In P. H. Blackman, V. L. Legett, & J. P. Jarvis (Eds.),
The diversity of homicide: Proceedings of the 2000 Annual Meeting of the Homicide Research
Working Group (pp. 117-127). Washington, DC: Federal Bureau of Investigation.
Plutchik, R., & Conte, H. R. (1997). Circumplex models of personality and emotions. Washington, DC:
American Psychological Association.
Polk, K. (1994). When men kill: Scenarios of masculine violence. Cambridge, UK: Cambridge University Press.
Regoeczi, W. C., & Riedel, M. (2003). The application of missing data estimation models to the problem
of unknown victim/offender relationships in homicide cases. Journal of Quantitative Criminology, 19,
155-183.
Salfati, C. G. (2000). The nature of expressiveness and instrumentality in homicide. Implications for
offender profiling. Homicide Studies, 4, 265-293.
Salfati, C. G., & Canter, D. V. (1999). Differentiating stranger murders: Profiling offender characteristics
from behavioral styles. Behavioral Sciences and the Law, 17, 391-406.
Smit, P. R. (2001). Nederland in internationaal perspectief [The Netherlands in international perspective].
In F. W. M. Huls, M. M. Schreuders, M. H. Ter Horst-van Breukelen, & F. P. van Tulder (Eds.),
Criminaliteit en Rechtshandhaving 2000 (O & B Series No. 189, pp. 200-219). The Hague,
Netherlands: Ministry of Justice, Wetenschappelijk Onderzoek en Documentatiecentrum Research
and Documentation Centre.
Smit, P. R., Bijleveld, C. C. J. H., & van der Zee, S. (2001). Homicide in the Netherlands: An exploratory
study of the 1998 cases. Homicide Studies, 5, 293-310.
Smith, M. D., & Parker, R. N. (1980). Type of homicide and variation in regional rates. Social Forces, 59,
136-147.
Smith, M. D., & Zahn, M. A. (Eds.). (1999). Homicide: A sourcebook of social research. Thousand Oaks,
CA: Sage.
Statistical Package for the Social Sciences. (1990). SPSS categories. SPSS user’s guide. Chicago: Author.
Torenbeek, R. (1999). Perfecte moorden en zinloze secties [Perfect murders and pointless sections].
Modus, 8, 15-19.
Van de Geer, J. P. (1993). Multivariate analysis of categorical data (vols. 1-2). London: Sage.
Van de Port, M. (2001). Niet geschoten is altijd mis. Antropologische beschouwingen over de liqui-
datiemoord [Nothing ventured, nothing gained. Anthropological reflections on contract killings].
Utrecht, Netherlands: University of Utrecht, Pompe Instituut.
Van den Eshof, P., Pieterse, L., Prins, H. P., Schouten, S. A., & Tromp, R. F. M. (1996). Vermiste personen;
Registratie en opsporing [Missing persons: Registration and tracing]. Zoetermeer, Netherlands: Centrale
Recherche Informatiedienst, Centrale Recherche Informatiedienst Themaboek.
Van Eck, C. (1997). Hoe Zeynep het leven liet [How Zeynep left this world]. Tijdschrift Voor Criminologie,
39, 217-234.
Wiggins, J. S. (1996). An informal history of the interpersonal circumplex tradition. Journal of Personality
Assessment, 66, 217-233.
Wilson, W. J. (1987). The truly disadvantaged: The inner city, the underclass, and public policy. Chicago:
University of Chicago Press.

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014


Bijleveld, Smit / Homicide in the Netherlands 219

Wolfgang, M. E. (1958). Patterns in criminal homicide. Philadelphia: University of Pennsylvania Press.


Wolfgang, M. E., & Ferracuti, F. (1967). The subculture of violence: Toward an integrated theory in crim-
inology. London: Tavistock.

Catrien Bijleveld was trained in psychology and law. She is a senior researcher at Nederlands
Studiecentrum Criminaliteit en Rechtshandhaving and professor of research methods in criminology at the
Free University, Amsterdam. Her recent publications are on sex offenders, criminal careers, and research
methodology.

Paul Smit is a senior researcher at the Research and Documentation Centre of the Dutch Ministry of
Justice. He specializes in cross-national comparisons of crime and sentencing statistics. He is a member
of the project group for the European Sourcebook of Crime and Criminal Justice Statistics and also
various publications on homicide in the Netherlands.

Downloaded from hsx.sagepub.com at TEXAS SOUTHERN UNIVERSITY on December 6, 2014

You might also like