Professional Documents
Culture Documents
Cognition
Optimizing
Efficiency,
Durability, and
Generalizability
Edited by
Alice F. Healy and
Lyle E. Bourne, Jr.
First published 2012
by Psychology Press part of the Taylor and Francis Group
711 Third Avenue, New York, NY 10017
Typeset in Bookman
by EvS Communication Networx, Inc.
Contents
Preface vii
16 Conclusions 326
LYLE E. BOURNE, JR. AND ALICE F. HEALY
Index 337
Preface
References
Healy, A. F., & Bourne, L. E., Jr. (Eds.). (1995). Learning and memory
of knowledge and skills: Durability and specificity. Thousand Oaks,
CA: Sage.
Healy, A. F., & Bourne, L. E., Jr. (Eds.). (1998). Foreign language learn-
ing: Psycholinguistic studies on training and retention. Mahwah, NJ:
Erlbaum.
1 Introduction
Training and Its Cognitive
Underpinnings
Lyle E. Bourne, Jr. and Alice F. Healy
University of Colorado
Important Distinctions
Content of Chapters
The purpose of this volume is to present the current state of
knowledge about how to train people mentally to perform jobs
and tasks in the real world. The plan is to address the process of
training from a number of vantage points, all based on research
in cognitive psychology. The effort begins, in chapter 2, with
a review of principles of training that have strong empirical
validation. Each principle is defined, its evidential support is
described, and possible guidelines for its use are suggested.
Chapters 3 through 7 describe a set of interrelated original
research projects designed to produce new empirical evidence
on training. Specifically, chapter 3 summarizes research aimed
at identifying and empirically supporting training principles
for procedural and declarative memory skills. Experiments are
presented illustrating a range of issues, including (a) generality
across tasks of individual principles, (b) tests of multiple
principles in a single task, (c) tests of principles in complex,
dynamic environments, and (d) development and testing of
new principles. Chapter 4 discusses the role of attention in the
training process. It identifies a number of component aspects of
attention, such as selectivity (i.e., filtering incoming information
from the environment), resource allocation (i.e., how to deploy
attention to meet the specific demands of a target task or job),
and effort investment (i.e., the role of motivation in attention),
and it demonstrates their contribution to the training process.
Chapter 5 summarizes research on response selection and
stimulus-response compatibility as contributing to speed and
accuracy in rudimentary binary choice tasks. The purpose is to
elucidate the most fundamental cognitive processes involved in
skill acquisition, retention, and transfer during training. Chapter
6 addresses issues related to level of automation in the training
process, individual differences in cognitive aptitude, and how
these factors interact to influence the effectiveness of training.
Finally, chapter 7 describes research that has investigated
methods for producing effective training using simulation and
other technologies. It focuses on technology-based training for
both individuals and collectives, especially in military settings.
The next set of chapters (chapters 8 through 12) focuses on
training theory, rather than on new empirical data. Specifically,
chapter 8 provides a taxonomic analysis of tasks, methods,
Introduction 9
performance measures, and principles involved in training. The
range of variables that can affect training and the multiplicity
of tasks that may require training prevent an exhaustive
quantification of training effects for specific tasks and training
scenarios. To render the study of training effects tractable and to
guide research, this chapter develops a taxonomy that includes
separate dimensions for task descriptions, training procedures,
training principles, and assessments of performance. The
taxonomy provides a framework by which training effects can be
assessed and predicted componentially for any task. Chapter 9
describes three computational models of training effects using the
ACT-R framework (Anderson & Lebiere, 1998) and the instance-
based learning (IBL) theory (Gonzalez, Lerch, & Lebiere, 2003).
The models address (a) fatigue effects in data entry, (b) stimulus-
response compatibility and Simon effects, and (c) visual detection
in a dynamic environment. A major conclusion from this work
is the robustness of an instance-based theory of training based
on memory for exemplars rather than on rules. Chapter 10
describes three models of training using a computational tool
called IMPRINT, which has previously been used mainly in large-
scale modeling for making resource allocation and personnel
decisions in the military. The goal of chapter 10 is to show
that this platform can be extended to capture the relationship
between training variables and performance based on smaller-
scale cognitive tasks. Three different tasks are examined: (a)
data entry, (b) visual detection in a dynamic environment, and
(c) information integration in a networked environment. The
goal of the modeling is to predict performance that reflects
underlying cognitive processes as revealed by the experimental
data in these three tasks. Chapter 11 describes the use of
contemporary applied mathematical techniques to evaluate and
compare computational models of the type described in chapters
9 and 10. The work demonstrates that models based on various
modeling platforms are comparable in their ability to account
for human performance in a couple of different tasks. However,
extant models require an excessive amount of computation time,
which often precludes parameter optimization and rapid precise
prediction. To solve this problem, selected models have been
reprogrammed in Matlab, which allows them to be speeded up
substantially. The translation to Matlab also permits extremely
rapid optimization of model parameters. Chapter 12 describes
an overarching quantitative framework designed to account for
the basic cognitive components of training in a single global
equation. The framework incorporates the separate functions
identified with each of the three basic processes of training
10 Lyle E. Bourne, Jr. and Alice F. Healy
(acquisition, retention, transfer). It incorporates an activation
function, similar to that in the ACT-R modeling platform,
describing the strength of a person’s knowledge or skill for a
given task or domain, as a function of past experience with
that and similar tasks. The global equation accommodates
established principles of generalization and transfer.
Earlier, a distinction was made between training and education.
Training focuses on specific tasks or jobs with an emphasis on
skill learning, whereas education seems more general, with a
focus on a particular domain of knowledge and an emphasis
on fact learning. Chapter 13 focuses on retrieval practice as
a training technique to enhance knowledge acquisition in an
educational context. It reviews classroom studies designed to
assess retrieval training when used in combination with spacing
of training sessions and with interleaving of subtasks or of
subparts of the material to be learned. The results illustrate the
close connection or similarity of cognitive processes involved in
training and education.
The overall goal of this volume is to provide enough information
about training to permit some speculation about, if not outright
application to, task and job requirements in the real world. To
make applications possible, however, we must first understand
the “real world.” To that end, chapter 14 describes the reality
of complex operations, particularly those conducted by airline
pilots, and discusses the implications of that reality for research
on training. A critical issue in this discussion is the demand of
concurrent task management in many everyday environments,
such as driving a car, and the ways in which training research
must address such demands.
The focus in this volume has been on the normal training
process. But what if the person being trained is deficient or
defective in some serious way, making the training required
unusual or out-of-the-ordinary? The importance of this question
arises, in part, from the current public and political interest in
acquired brain injuries and the consequent effects on physical
and cognitive health. Chapter 15 reviews the current status
of cognitive retraining programs and evaluates their success.
The review identifies training principles, like those described
in chapter 2, which function effectively in retraining programs.
In addition, it suggests possibilities for the transition of as-yet-
unused training principles into cognitive retraining efforts in
the future.
The final chapter reviews the current status of scientific
knowledge about training. Chapter 16 ties together the major
points of the preceding chapters and arrives at some general
Introduction 11
conclusions and implications of research in cognitive psychology
for application outside the laboratory so as to enhance the
training process.
References
Anderson, J. R. (1982). Acquisition of cognitive skill. Psychological
Review, 89, 369–406.
Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought.
Hillsdale, NJ: Erlbaum.
Bourne, L. E., Jr., Healy, A. F., Parker, J. T., & Rickard, T. C. (1999).
The strategic basis of performance in binary classification tasks:
Strategy choices and strategy transitions. Journal of Memory and
Language, 41, 223–252.
Delaney, P. F., Reder, L. M., Staszewski, J. J., & Ritter, F. E. (1998). The
strategy-specific nature of improvement: The power law applies by
strategy within task. Psychological Science, 9, 1–7.
Ebbinghaus, H. (1913). Memory: A contribution to experimental psy-
chology. New York: Teachers College, Columbia University. (Original
work published 1885)
Gonzalez, C., Lerch, J. F., & Lebiere, C. (2003). Instance-based learn-
ing in dynamic decision making. Cognitive Science, 27, 591–635.
Graf, P., & Masson, M. E. J. (Eds.) (1993). Implicit memory: New direc-
tions in cognition, development, and neuropsychology. Hillsdale, NJ:
Erlbaum.
Healy, A. F. (2007). Transfer: Specificity and generality. In H. L. Roedi-
ger, III, Y. Dudai, & S. M. Fitzpatrick (Eds.), Science of memory: Con-
cepts (pp. 271–275). New York: Oxford University Press.
Healy, A. F., Kole, J. A., Buck-Gengler, C. J., & Bourne, L. E., Jr. (2004).
Effects of prolonged work on data entry speed and accuracy. Journal
of Experimental Psychology: Applied, 10, 188–199.
Ivancic, K., & Hesketh, B. (2000). Learning from errors in a driving
simulation: Effects on driving skill and self-confidence. Ergonomics,
43, 1966–1984.
Kole, J. A., Healy, A. F., & Bourne, L. E., Jr. (2008). Cognitive com-
plications moderate the speed-accuracy tradeoff in data entry: A
cognitive antidote to inhibition. Applied Cognitive Psychology, 22,
917–937.
Kolers, P. A., & Roediger, H. L., III (1984). Procedures of mind. Journal
of Verbal Learning and Verbal Behavior, 23, 425–449.
Lee, Y., & Vakoch, D. A. (1996). Transfer and retention of implicit and
explicit learning. British Journal of Psychology, 87, 637–651.
Leibowitz, N., Baum, B., Enden, G., & Karniel, A. (2010). The expo-
nential learning equation as a function of successful trials results
in sigmoid performance. Journal of Mathematical Psychology, 54,
338–340.
Maxwell, J. P., Masters, R. S. W., Kerr, E., & Weedon, E. (2001). The
implicit benefit of learning without errors. Quarterly Journal of
12 Lyle E. Bourne, Jr. and Alice F. Healy
Experimental Psychology: Human Experimental Psychology, 54A,
1049–1068.
Newell, A., & Rosenbloom, P. S. (1981). Mechanisms of skill acquisition
and the power law of practice. In J. R. Anderson (Ed.), Cognitive
skills and their acquisition (pp. 1–55). Hillsdale, NJ: Erlbaum.
Osgood, C. E. (1949). The similarity paradox in human learning: A
resolution. Psychological Review, 56, 132–143.
Pachella, R. G. (1974). The interpretation of reaction time in informa-
tion-processing research. In B. H. Kantowitz (Ed.), Human informa-
tion processing: Tutorials in performance and cognition (pp. 41–82).
Hillsdale, NJ: Erlbaum.
Postman, L., & Underwood, B. J. (1973). Critical issues in interference
theory. Memory & Cognition, 1, 19–40.
Reber, A. S. (1967). Implicit learning of artificial grammars. Journal of
Verbal Learning and Verbal Behavior, 6, 855–863.
Rickard, T. C. (1997). Bending the power law: A CMPL theory of strategy
shifts and the automatization of cognitive skills. Journal of Experi-
mental Psychology: General, 126, 288–311.
Roediger, H. L., III. (2008). Relativity of remembering: Why the laws of
memory vanished. Annual Review of Psychology, 59, 225–254.
Rubin, D. C., & Wenzel, A. E. (1996). One hundred years of forgetting:
A quantitative description of retention. Psychological Review, 103,
734–760.
Salas, E., Cannon-Bowers, J. A., & Blickensderfer, E. L. (1999).
Enhancing reciprocity between training theory and practice: Princi-
ples, guidelines, and specifications. In G. R. Ferris (Ed.), Research in
personnel and human resources management (Vol. 17, pp. 291–321).
Stamford, CT: JAI Press.
Schmidt, R. A. (1975). A schema theory of discrete motor skill learning.
Psychological Review, 82, 225–260.
Schmidt, R. A., & Bjork, R. A. (1992). New conceptualizations of prac-
tice: Common principles in three paradigms suggest new concepts
for training. Psychological Science, 3, 207–217.
Schneider, V. I., Healy, A. F., & Bourne, L. E., Jr. (2002). What is learned
under difficult conditions is hard to forget: Contextual interference
effects in foreign vocabulary acquisition, retention, and transfer.
Journal of Memory and Language, 46, 419–440.
Shepard R. N. (1987). Toward a universal law of generalization for psy-
chological science. Science, 237, 1317–1323.
Tulving, E. (1985). How many memory systems are there? American
Psychologist, 40, 385–398.
Wickelgren, W.A. (1974). Single-trace fragility theory of memory dynam-
ics. Memory & Cognition, 2, 775–780.
Wixted, J. T., & Carpenter, S. K. (2007). The Wickelgren power law
and the Ebbinghaus savings function. Psychological Science, 18,
133–134.
Wulf, G., & Schmidt, R. A. (1997). Variability of practice and implicit
motor learning. Journal of Experimental Psychology: Learning, Mem-
ory, and Cognition, 23, 987–1006.
2 Empirically Valid Principles
of Training
Alice F. Healy, Vivian I. Schneider,
and Lyle E. Bourne, Jr.
University of Colorado
Deliberate Practice
The old adage that practice makes perfect is true, but not all
practice is equivalent in effectiveness. Deliberate (i.e., effortful,
highly focused, and highly motivated) practice is best in terms of
promoting skill acquisition and expertise (Ericsson, Krampe, &
14 Alice F. Healy, Vivian I. Schneider, and Lyle E. Bourne, Jr.
Tesch-Römer, 1993). Indeed, learners, even those who might be
talented or have a high aptitude for the training domain, will not
acquire their highest level of performance if they do not engage
in deliberate practice over a prolonged period of time with many
repetitions of the skill to be performed.
Guideline: By initial instructions, trainers should try to
engage trainees in deliberate practice throughout the training
process.
Depth of Processing
One aspect of deliberate practice relates to how deeply the mate-
rial to be learned is processed. Activities during training that
promote deep or elaborate processing of materials yield supe-
rior retention (e.g., Craik & Lockhart, 1972; but see Roediger,
2008, for exceptions). The depth of processing principle can be
achieved in various ways. For example, asking individuals to
rate the pleasantness of each word to be memorized enhances its
memory over simply viewing the word. Memory is also enhanced
when the material is presented in a format that requires a trans-
lation process or speech coding. Counter to intuition, when
numerical data must be entered into some system, the num-
bers should be presented in word format (e.g., three-five-two)
rather than numeral format (3-5-2) if the goal is to maximize
memory for the numbers. Word format, but not numeral format,
requires translation from the words to the digits represented
on a keyboard and facilitates speech coding of the digits. This
additional process enhances long-term memory for the material
(Buck-Gengler & Healy, 2001).
Guideline: To enhance the durability of training material, pro-
mote deep processing of the material to be learned either by
explicit instructions or by incidental task demands.
Generation Effect
Another recommendation related to deliberate practice is to
promote active, rather than passive, learning techniques. For
example, if the task is to memorize a set of procedures for trou-
bleshooting a piece of equipment, the trainees should try to gen-
erate the procedures from memory, rather than simply to read
or to reread them. Then the trainees should check the accuracy
of their actively generated responses against the correct list and
make note of any errors. They should actively generate the list
repeatedly until they are able to produce it without error. This
recommendation follows from the generation effect (the finding
Empirically Valid Principles of Training 15
that people show better retention of learned material when it
is self-produced, or generated, than when it is simply copied or
read; e.g., Crutcher & Healy, 1989; McNamara & Healy, 1995,
2000; Slamecka & Graf, 1978). This generation effect is incorpor-
ated into a common study strategy known as the 3R technique
(read-recite-review), which has been shown to improve learning
from educational texts (McDaniel, Howard, & Einstein, 2009).
Guideline: Trainers should use whatever methods are possible
to engage trainees actively in the learning process; this advice
would include periodically requiring them to generate answers
to questions.
Focus of Attention
It is possible for a learner to deploy or focus attention in vari-
ous ways during training. Furthermore, a learner might be
instructed about how to focus attention effectively. Some stud-
ies of motor skill acquisition have compared an external focus
of attention (i.e., attention to the results of a movement) to an
internal focus of attention (i.e., attention to the body movements
themselves). That research has consistently found, at least after
some initial training, that there is an advantage for the external
focus of attention with respect to learning, retention, and trans-
fer of motor skills (Lohse, Sherwood, & Healy, 2010; Lohse, Wulf,
& Lewthwaite, in press; C. H. Shea & Wulf, 1999; Wulf, 2007).
This result is explained by the constrained action hypothesis,
according to which well-developed motor skills are represented
in bodily mechanisms that are impaired by conscious attention
to them (Beilock, Bertenthal, McCoy, & Carr, 2004).
Guideline: Trainers of motor skills should encourage learn-
ers to adopt an external focus of attention on the target of their
movements rather than an internal focus on the bodily move-
ments themselves.
Procedural Reinstatement
The procedural reinstatement principle implies that dupli-
cating at test procedures that were required during learning
facilitates subsequent retention and transfer (Clawson, Healy,
Ericsson, & Bourne, 2001; Healy & Bourne, 1995; Healy, Claw-
son, et al., 1993; Healy, Fendrich, et al., 1992; Healy, Wohld-
mann, & Bourne, 2005). This principle is similar to others that
were derived primarily from studies of list learning, including
the principles of encoding specificity (memory for information is
best when retrieval cues elicit the original encoding operations;
e.g., Tulving & Thomson, 1973), transfer appropriate processing
(memory performance will be best when test procedures evoke
the procedures used during prior learning; e.g., C. D. Morris,
Bransford, & Franks, 1977; Roediger, Weldon, & Challis, 1989),
and context-dependent memory (memory for information is worse
when tested in a new context than when tested in the original
context in which it was learned; e.g., Kole, Healy, Fierman, &
Bourne, 2010; S. M. Smith & Vela, 2001). An important corol-
lary to this procedural reinstatement principle is that specificity
(limited transfer) occurs for tasks based primarily on procedural
information, or skill, whereas generality (robust transfer) occurs
for tasks based primarily on declarative information, or facts
(Healy, 2007; Healy, Wohldmann, et al., in press). Thus, for skill
learning, retention is strong but transfer is limited, whereas for
fact learning, retention is poor but transfer is robust.
Guideline: Trainees should reinstate the conditions of study
as closely as possible when taking a test or performing in the
field. If trainees are able to anticipate the test or field conditions,
then they should modify their study conditions to match them.
Training can be expected to be generalizable when it involves
declarative facts, whereas it can be expected to be durable when
it involves procedural skills.
Part-Task Training
Under certain conditions training only a part of a task before
training the whole task is more effective than training the whole
task from the beginning. For complex tasks that can be divided
into parts, the conditions for part-training superiority appear
to be a function of the organization of the parts. A segmented
task contains parts that are performed sequentially, whereas
a fractionated task contains parts that are performed simulta-
neously. For segmented tasks, part training can either involve
forward chaining (when the initial segment of a task is trained
first) or backward chaining (when the final segment of the task
is trained first). In this case, part-task training is most benefi-
cial when a backward-chaining procedure is used (Wightman
& Lintern, 1985) because there is a strong association between
performance level on the terminal task and knowledge of results
(i.e., the feedback resulting from task completion). In contrast, for
a fractionated task, Adams and Hufford (1962) found that part
training initially disrupted performance on the whole procedure.
In both segmented and fractionated tasks, during the initial
part-training phase, the trainee constructs independent pro-
cedural representations for each part of the whole task. When
transfer to the whole task occurs, there is only a single inter-
ruption between two successive parts in a segmented task. In
contrast, there can be multiple interruptions among the parts
in a fractionated task. Thus, the procedural representations
Empirically Valid Principles of Training 19
can remain intact and independent only in a segmented task;
in a fractionated task a new procedural representation must be
established, which requires integration of the two parts, because
the parts in that case are performed as an interlocking unit
(Marmie & Healy, 1995).
Chapter 4 provides a more detailed account of part-task train-
ing, and chapter 6 discusses the effects of automating parts of
a whole task.
Guideline: Whether or not initial training of a complex task
should involve only parts of that task depends on a number of
task characteristics including forward versus backward chain-
ing of the parts and the segmented versus fractionated nature
of the whole task.
Easy–Difficult Ordering
Complex tasks can be divided into parts based on aspects of the
stimuli involved, such as their difficulty. When a task involv-
ing a stimulus set is trained incrementally, the question arises
as to whether the easier or the more difficult stimuli in the set
should be trained first. Pellegrino, Doane, Fischer, and Alderton
(1991) found that initial training on a difficult subset of stimuli
was beneficial relative to initial training on an easy subset of the
stimuli in a visual discrimination task. According to Pellegrino
et al. (1991; see also Doane, Alderton, Sohn, & Pellegrino, 1996;
Doane, Sohn, & Schreiber, 1999), incremental training should
begin with the part of the stimulus set that yields the most effec-
tive strategic skills. However, it is not always the more difficult
part that yields the optimal strategic skills. For example, Clawson
et al. (2001) found that initial training on easy stimuli in a Morse
code reception task led participants to adopt an effective unitiza-
tion strategy for representing codes, whereas initial training on
difficult stimuli led to a less effective strategy in which individual
elements were separately represented and then integrated.
A related issue that has been explored by Maxwell, Masters,
Kerr, and Weedon (2001) is what they call errorless learning. For
a motor skill, subjects should begin with the easiest task, where
few if any errors are made, and progress to increasingly harder
tasks to minimize the overall number of errors made. It has been
shown that skills that have been learned in an error-prone man-
ner require more attentional resources than do skills learned
in an errorless manner. Because there is less attention needed
to perform the skill learned in errorless training, distractions,
such as a secondary task, cause less disruption. Also related are
the concepts of training wheels and scaffolding, usually used in
20 Alice F. Healy, Vivian I. Schneider, and Lyle E. Bourne, Jr.
the context of fact-learning tasks, in which training begins with
easy items and subsequent training introduces more difficult
items that build on the easier ones (see chapters 4 and 13 for
further discussions of these concepts).
Guideline: Whether or not training should begin with the easi-
est or most difficult components of a task depends on task char-
acteristics. Trainers need to be sensitive to these characteristics
before deciding on the order of subtask training. An important
factor to consider is which parts of the task yield the most effec-
tive strategic skills.
Spacing
When training new skills involves repeated practice trials, per-
formance on the skill improves more rapidly when rest inter-
vals are interpolated between trials (i.e., spaced or distributed
practice) than when the trials are administered without rest
intervals (i.e., massed practice; e.g., Bourne & Archer, 1956). In
contrast, when training involves new knowledge (i.e., fact acqui-
sition), performance improves more rapidly when practice tri-
als are massed rather than spaced (Bjork, 2010; Underwood &
Ekstrand, 1967). For both skill and fact learning, however, per-
formance after a retention interval is better for spaced than for
massed practice. A related spacing effect, based on interleaving
materials (see discussion in chapter 13) involves the separation
of repetitions of a given item within a list of items (e.g., Glenberg,
1976; Hintzman, 1974). Although usually some rest between
repetitions improves performance, the rest interval cannot be
increased indefinitely with benefit to the learner. There is an
optimal rest interval for at least some tasks (Bourne, Guy, Dodd,
& Justesen, 1965), but more research needs to be done to deter-
mine the generality of this effect. With respect to retention of
the learned material, the spacing effect does not always hold
when the retention interval (interval between the last repetition
and the test) is very short. Generally, the advantage of spacing
holds for mixed lists including spacing intervals varying across
Empirically Valid Principles of Training 21
different items as well as for pure lists with a single spacing
interval (Kahana & Howard, 2005). All of this work is based on
single-session training paradigms with short spacing and reten-
tion intervals.
In a different paradigm, Bahrick (1979) used long spacing
intervals to separate learning sessions and long retention inter-
vals between the end of learning and final testing. He found
that the level of performance on the final test session depended
more on the spacing between learning sessions than it did on
the level of performance achieved in the final learning session.
Unlike findings from experiments with short intervals between
practice trials or items (cited above), which generally show an
advantage for spaced practice, performance on the final learn-
ing session of Bahrick’s study was greatest when the interses-
sion intervals were shortest, but performance on the final test
session was highest when the intersession intervals were longest
(so that they resembled the retention interval). Bahrick, thus,
concluded that for optimal knowledge maintenance, practice
should be spaced at intervals approximating the length of the
eventual retention interval.
More recently, Pashler, Rohrer, Cepeda, and Carpenter (2007)
looked at the effects of varying the training intersession inter-
val (ISI). They showed strong effects of ISI over long retention
intervals (RIs). In addition, test performance after a given RI
was found to be optimal when the ISI was intermediate in value.
Making spacing longer than optimal was, however, less harmful
to retention than making it shorter than optimal. These authors
suggest that it is more effective to use an ISI of several months or
years than to use shorter intervals when retention is tested after
a delay of several years (see also Cepeda, Pashler, Vul, Wixted, &
Rohrer, 2006, for a review of the literature).
Guideline: For optimal benefits from training, repeated prac-
tice on particular items or responses should be spaced in time.
The amount of spacing (length of the time interval between rep-
etitions) should be related to the amount of time that is likely to
pass between training and eventual testing. Generally, it seems
desirable to match the time between repetitions during training
to the time between training and test.
Feedback
Two distinct and major questions have been asked about the
effects of feedback: what form it should take and when to pro-
vide it.
22 Alice F. Healy, Vivian I. Schneider, and Lyle E. Bourne, Jr.
What Kind of Feedback Should be Provided? What type of
feedback to provide is a crucial issue for optimizing training and
retention of knowledge and skills (Schmidt & Bjork, 1992). Trial-
by-trial feedback has been shown to facilitate rate of learning
in many tasks, possibly by motivating participants to set
increasingly higher standards of performance or by identifying
errors and how to correct them. But, if participants have a
good sense anyway of how well they responded, then trial-by-
trial feedback might be a distraction that results in inferior
performance on later acquisition trials, on retention tests, or on
tests with tasks requiring slightly different responses. In such
circumstances, periodic summary feedback, given only on some
proportion of training trials, is often a more effective procedure
for promoting long-term retention than is trial-by-trial feedback
(see, e.g., Schmidt, Young, Swinnen, & Shapiro, 1989, for
illustration of this finding in a ballistic timing task). Indeed
there is some suggestion in the literature that the frequency
of feedback given during acquisition can be gradually reduced
or faded without serious or adverse effects on acquisition
performance and at the same time produce beneficial effects
on long-term retention (Schmidt & Bjork, 1992). Other studies
have shown that in some tasks (e.g., speeded aiming) feedback
effects during training might not persist into later testing for
retention (Bourne, Healy, Pauli, Parker, & Birbaumer, 2005).
Thus, differences among tasks will need to be examined in the
future to assess the boundary conditions on effects involving
what kind of feedback to provide.
Rehearsal
Testing
Tests are usually thought of as performance assessment tools,
but there is increasing evidence that people learn from tak-
ing tests often as much as or more than they learn from pure
study (see chapter 13). This phenomenon has been referred to
as a “testing effect” (Carpenter & DeLosh, 2005; Izawa, 1992;
McDaniel & Fisher, 1991). Specifically, the testing effect is the
advantage in retention for material that is tested relative to
material that is presented for additional study. Pashler et al.
(2007) point out that the testing effect has been found for vari-
ous types of tests and materials. Specifically, the effect is evident
for free recall (e.g., Allen, Mahler, & Estes, 1969; Carpenter &
DeLosh, 2006) and cued recall (Carrier & Pashler, 1992) and for
face–name associations (Carpenter & DeLosh, 2005), definitions
(Cull, 2000), and general knowledge facts (McDaniel & Fisher,
1991). The testing effect is found even when the subject’s answer
is incorrect, at least when corrective feedback is provided (Kor-
nell, Hays, & Bjork, 2009), and even when the testing occurs
after an initial correct recall response (Karpicke & Roediger,
2008). Pashler et al. also found that covert retrieval practice, a
form of mental rehearsal, in which subjects are asked to retrieve
without providing an observable response, enhances learning.
Note that the testing effect is similar to the generation effect, but
there are subtle though important differences demonstrated by
Karpicke and Zaromb (2010).
McDaniel, Roediger, and McDermott (2007) summarized
experiments on the testing effect in the context of a university
course. They found that providing short-answer and multiple-
choice tests initially, compared to providing no tests initially,
significantly aided performance on a subsequent test. They also
found that short-answer tests (requiring production or recall)
were more helpful to later test performance than were multiple-
choice tests (requiring only recognition), even when the later
tests involved multiple-choice questions. Finally, they found that
26 Alice F. Healy, Vivian I. Schneider, and Lyle E. Bourne, Jr.
short-answer tests were more effective than focused restudy of
the same material when those tests involved corrective feedback.
Note that the testing effect has been examined primarily in
declarative learning tasks, where it is possible to separate pure
study from test performance. In skill learning tasks, study and
tests are usually integrated into the trial-by-trial acquisition
procedure, with each trial necessarily including a testing com-
ponent. The testing effect is really, thus, not directly applicable
to skill learning although mental practice (or even observation)
might be considered an analogue of studying without testing.
Guideline: A lot of fact learning occurs during test taking.
Therefore tests should be embedded in the training process
whenever possible.
Overlearning
Training usually ends when the trainee reaches some predes-
ignated performance criterion, such as one or more error-free
training trials. Overlearning refers to practice beyond the per-
formance criterion (Pashler et al., 2007). It has been found that
overlearning, relative to less practice, improves later perfor-
mance (Krueger, 1929). Consequently, overlearning has been
proposed as a useful, general strategy when long-term reten-
tion is the goal (Driskell, Willis, & Copper, 1992). However, over-
learning is not always an efficient way to strengthen acquired
knowledge and skill. For example, in a study by Rohrer, Taylor,
Pashler, Wixted, and Cepeda (2005) subjects were taught novel
vocabulary pairs. They saw each word pair either 5 or 10 times.
After 1 week, the subjects who saw the pairs 10 times showed a
substantial benefit over the subjects who saw the pairs 5 times,
but the difference had disappeared after 4 weeks. Rohrer and
Taylor (2006) conducted a similar study using a new math
skill. One group of subjects had three times the number of
practice problems, but no difference was found after either the
1-week or the 4-week retention interval. Thus, Pashler et al.
concluded that for long-term memory, overlearning is ineffi-
cient as a training technique (confi rming results were obtained
by Rawson & Dunlosky, 2011). They pointed out, however, that
in some cases overlearning might be the only alternative when
a skill needs to be performed with absolutely no errors at a
much later time (e.g., performing CPR or landing a space shut-
tle). They also concluded that, even when retrieval accuracy
is at ceiling, overlearning might improve speed of responding
(e.g., Logan & Klapp, 1991), and speedup could be useful when
rapid responding is a prime consideration.
Empirically Valid Principles of Training 27
Guideline: Overlearning is recommended as a training tech-
nique only when training time is not severely limited and when
it is crucial to have the strongest possible representations of
knowledge and skill.
Task Difficulty
Interference is a source of difficulty in training that occurs
when conditions allow incorrect answers to come to the train-
ee’s mind, along with the correct answer, thereby requiring
the trainee to choose the correct answer from among several
alternatives. Increasing interference during training has been
shown to impede training speed but ultimately to enhance the
durability and flexibility of what is learned. For example, mixing
material across categories during training, as opposed to group-
ing the material by category, enhances interference, which may
inhibit initial acquisition, but should yield better retention and
transfer (see chapter 13 for a discussion of interleaving). In fact,
it has been shown that many things that make learning difficult
(not just categorical interference) facilitate transfer to a new task
as well as long-term retention of the original task (see chapter
3 for an example). This recommendation to introduce difficulty
during training follows from both the effects of contextual inter-
ference (interference during learning facilitates later retention
and transfer; Battig, 1972, 1979; Carlson & Yaure, 1990; Lee
& Magill, 1983; Schneider, Healy, & Bourne, 1998; Schneider,
Healy, Ericsson, & Bourne, 1995; J. B. Shea & Morgan, 1979;
but see Wulf & Shea, 2002, for some exceptions) and, more
generally, the training difficulty principle (any condition that
causes difficulty during learning facilitates later retention and
transfer; Schmidt & Bjork, 1992; Schneider, Healy, & Bourne,
2002; but see Young, Healy, Gonzalez, Dutt, & Bourne, 2011, for
some qualifications).
Not all sources of difficulty during training are desirable, how-
ever (see Bjork, 1994). McDaniel and his colleagues (McDaniel &
Butler, 2011; McDaniel & Einstein, 2005) argue that difficulties
introduced during training are facilitative only when they cause
the learner to apply task-relevant cognitive processes (e.g., gen-
eration from answer fragments) to the to-be-learned material
that otherwise would not be engaged.
Guideline: Counter to intuition, trainers should consider
introducing sources of interference into any training routine.
If durable retention and flexible transfer are the goals of train-
ing, then mixing materials during training is advisable for most
learners. Trainers might consider enhancing the difficulty of
28 Alice F. Healy, Vivian I. Schneider, and Lyle E. Bourne, Jr.
training exercises in other ways as well with the caveat that
task-relevant cognitive processes must be engaged.
Stimulus-Response Compatibility
Cognitive skills can be divided into three stages: (a) perception
of the stimulus, (b) decision making and response selection, and
(c) response execution (Proctor & Dutta, 1995). The most ubiq-
uitous phenomenon observed in the second stage of skill acqui-
sition is the effect of stimulus-response compatibility (Fitts &
Deininger, 1954; Fitts & Seeger, 1953; Proctor & Vu, 2006; see
also chapter 5). This effect reflects a difference in performance
attributable to the mapping of individual stimuli to responses,
such that performance is best when the stimulus set and the
response set are configured in a similar way and each stimulus
is mapped to its corresponding response (e.g., left-right stimu-
lus locations are mapped to left-right responses). The detrimen-
tal effects of incompatibility are not easily overcome, even after
extensive practice (e.g., Dutta & Proctor, 1992; but see Miles
& Proctor, 2010, for an exception). Making a task difficult by
introducing stimulus-response incompatibilities is clearly not a
desirable difficulty for learning or retention.
Guideline: It is important to maintain stimulus-response
compatibility during training to avoid the prolonged, detrimen-
tal effects that incompatibility can have on performance.
Serial Position
Better memory has been found for the initial and final items in a
to-be-learned list of items (Nipher, 1878). This bow-shaped serial
position function, with both primacy and recency components,
is found at the start of learning but diminishes as repeated tri-
als on the same material are given (Bonk & Healy, 2010). The
same effect is observed for short lists (as few as 4 items) and
long lists (40 items or more), for tasks that require item learn-
ing or response-sequence learning, and for both immediate and
delayed memory. The relative magnitude of primacy and recency
effects differs depending on many variables, especially the test-
ing procedure. In any event, the items in the middle of a list are
at a disadvantage for both learning and memory (see chapter 3).
Thus, training will require more practice on items in the middle
of a list than on those at either end.
Guideline: For tasks that require training on a sequence
of informational items or responses, the trainer should place
Empirically Valid Principles of Training 29
greater emphasis on items in the middle of the sequence than on
those at the beginning or end.
Variability of Practice
Variable practice conditions (in which individuals train on a
number of different tasks) typically yield better performance at
transfer testing than do constant practice conditions (in which
individuals train on a single task), even when testing is con-
ducted on the same task as trained under constant practice.
For example, practice that varied the distance from a target at
which beanbags were tossed (2 or 4 feet) produced greater accu-
racy at test on an intermediate distance (3 feet) than did prac-
tice limited to the same intermediate target distance (Schmidt
& Bjork, 1992). The benefits of variable practice were first recog-
nized by Schmidt (1975) for discrete motor tasks and explained
by him in terms of a schema theory, according to which vari-
ability promotes effective and general use of rules (schemata)
relating external task requirements to internal movement com-
mands. Wulf and Schmidt (1997) extended these findings to a
continuous, feedback-regulated tracking task, and Schmidt and
Bjork (1992) extended them further to tasks that do not involve
motor learning, such as concept formation and text process-
ing. Goode, Geraci, and Roediger (2008) also found that vari-
able practice yielded superior transfer over repeated practice on
anagram solutions.
Contrary to these findings, in a feedback-regulated non-
tracking perceptual-motor task, Healy, Wohldmann, Sutton,
and Bourne (2006) found that performance was worse for vari-
able practice conditions relative to constant practice conditions
involving the same task used during transfer testing. The tasks
in this study were defined in terms of particular perceptual-
motor reversal conditions, all involving the same targets. It is
possible to introduce a different form of variability by varying
the targets rather than the type of perceptual-motor reversal. In
a subsequent study involving the same perceptual-motor task,
Wohldmann, Healy, and Bourne (2008b) found benefits of vari-
able practice when subjects were given multiple targets under
the same perceptual-motor reversal condition. The conclusion is
that varying task parameters within a single generalized motor
program (schema) enhances transfer, whereas varying the motor
programs themselves has no beneficial transfer effect.
Guideline: Trainers should vary the conditions of practice
to facilitate generalization of the trained skill. There are some
30 Alice F. Healy, Vivian I. Schneider, and Lyle E. Bourne, Jr.
limits, however, which involve how variability is introduced into
the task.
3.0
Training
Test
2.5
Mean Response Time per Click
2.0
1.5
1.0
0.5
0.0
Match Partial Match No Match
Figure 3.1 Response time per click (in s) as a function of phase and
category continuity condition in Kole et al. (2010, Experi-
ment 2). Error bars show standard errors of the mean.
0.6
0.4
0.2
0.0
Match Partial Match No Match
Exemplar Continuity Condition
Message Comprehension
Another principle in need of a generality check is the optimal
modality principle; that is, learning is better when information
is seen than when it is read, and it is best when it is both read
and seen. By this principle, the ordering of modalities to use
for presenting information from worst to best is (a) read, (b) see,
and (c) both read and see. This principle is popular in the field
of classroom education (Dale, 1969), but there is little empiri-
cal support for it in other contexts. Is the ordering of modalities
Basic Research on Training Principles 45
specified by this principle valid generally and, thus, appropriate
for optimizing training?
A message comprehension task, meant to mimic communica-
tion between air traffic controllers and pilots (Barshi & Healy,
2002, 2011), was used to examine this principle in a context out-
side of the classroom. Subjects were given messages about navi-
gating in a space shown on the computer screen, and then they
followed those instructions by making mouse clicks to indicate
movement in the space. Three format conditions were examined
(McCormick, Schneider, & Healy, 2010), one with messages pre-
sented in only one modality one time each (single), one with mes-
sages presented in only one modality two times each (double),
and one with messages presented first in one modality and then
in the other (mixed). For the single and double conditions, mes-
sages that were read (words) were compared with messages that
were seen (symbols).
Figure 3.3 shows proportion of correct responses as a func-
tion of format condition. Performance was worst when the com-
mands were shown only once (single), and performance was no
worse (in fact there was a trend for performance to be better)
when the commands were shown twice in the same modality
(double) than when two modalities (mixed) were used. Perfor-
mance in the word modality improved more consistently across
0.8
Proportion Correct
0.7
0.6
0.5
double mixed single
Condition
0.7
0.6
0.5
1 2 3 4 5 6
Block
Double
0.8 0.8 Equal
Proportion Correct
Proportion Correct
Double None
Equal
0.6 None 0.6
0.4 0.4
0.2 0.2
0.0 0.0
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
Serial Position Serial Position
0.8 0.8
Proportion Correct
Proportion Correct
0.6 0.6
Double
Equal
0.4 None 0.4
Double
Equal
None
0.2 0.2
0.0 0.0
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
Serial Position Serial Position
Figure 3.5 Proportion correct for the fi rst two attempts as a function
of presentation condition, distractor condition, and serial
position in experiment by Bonk and Healy (2010). Error
bars show standard errors of the mean.
Basic Research on Training Principles 49
double conditions. In addition, there was a surprising small but
significant advantage for the no moves condition (with no extra
spatial information) relative to the moves condition. Finally, the
shape of the serial position function depended on attempt as
well as on both presentation and distractor conditions, with the
primacy and recency effects largest when overall performance
was worst.
Although primacy and recency effects were found in all cases,
thus supporting a general serial position principle, the finding
of different serial position patterns for different conditions is not
captured by quantitative models of the serial position function
and therefore poses a challenge to those models (e.g., Brown,
Neath, & Chater, 2007; Murdock, 1960). The results from this
experiment also document the intuitive expectation that dis-
tractors at test disrupt recall. There was, however, no support
for the dual coding principle because the spatial information
provided by the moves condition proved to be distracting instead
of helpful to the recall of verbally encoded items.
The three principles under examination in this experiment
clearly have interactive effects, with the magnitude of primary
and recency effects depending on retrieval distraction and the
type of coding. Therefore these principles need to be considered
in concert in any training routine. The more general theoreti-
cal implication is that single principles should not be applied
indiscriminately or in isolation because they might affect perfor-
mance differently depending on training and testing contexts.
The task used in this experiment resembles tasks that involve
learning a series of steps in a standardized procedure, such as
how to clean and reassemble a device (although the steps in that
case are not necessary independent). Finding best performance
for the initial items in the series and worst performance for the
middle items should lead trainers to concentrate their training
efforts toward the middle items, assuming all items in the series
are equally important. But the extent to which specific middle
items should be emphasized clearly depends on both the learn-
ing and the testing conditions.
Clicker Technique
A popular new educational technology involves clickers. Periodi-
cally, during a classroom lecture, students are asked multiple-
choice questions and use a hand-held clicker device to produce
answers, with the distribution of answers visible to the class as
a whole and to the instructor. The instructor might decide to
review, elaborate, or provide more information about material
if many students in the class give wrong answers to the clicker
50 Alice F. Healy and Lyle E. Bourne, Jr.
questions. The clicker technique provides a possible method of
classroom teaching time conservation or compression, whereby
any material that is already known is eliminated from further
presentation. In addition, like the well-established testing effect,
it uses tests, not just study episodes, as a way to strengthen
classroom learning. There are, thus, two training principles
that are relevant to the clicker technique. First is the principle
of training compression, by which training can be truncated by
eliminating practice on known facts. Second is the principle of
testing, by which tests can be used to strengthen a person’s
knowledge of material as much as, or possibly even more than,
can further study.
How do these two training principles interact? In the first
of two experiments reviewed here, a laboratory analogue of
the clicker technique was compared to a more standard tech-
nique, in which all material was presented whether or not it was
already learned, and a technique based on individualized train-
ing, in which material was dropped out from presentation to a
particular subject if the subject responded correctly to the mate-
rial (Anderson, Healy, Kole, & Bourne, 2010). Students learned
64 facts about plants. The facts were all correct, with the plant
names changed to fictitious ones to eliminate effects of prior
knowledge. Each fact was stated in either specific or general
form. Subjects were tested with four-alternative multiple-choice
questions, in which they had to recognize the fictitious plant
name associated with a fact. There were four study-test training
rounds, in which subjects were shown blocks of up to 8 facts
for study and then immediately tested on those facts. The first
and fourth training rounds involved all 64 facts; the second and
third rounds involved fewer facts in some conditions.
The four training conditions were (a) full, the standard condi-
tion involving presentation of all facts on each round; (b) drop-
out, the individualized condition involving presentation and
testing on Rounds 2 and 3 of only those facts missed on the pre-
ceding round by a given subject; (c) yoked, the control condition
involving presentation and testing on Rounds 2 and 3 of only
those facts missed on the preceding round by a matched subject
in the dropout condition; and (d) clicker, a condition simulat-
ing the clicker technology. The clicker condition was equated
to the dropout condition of a preliminary experiment in terms
of the number of facts presented and tested on Rounds 2 and
3 on average (26 of the 64 facts on Round 2 and 10 on Round
3). The specific facts selected were determined on the basis of a
full condition of the preliminary experiment to be the facts most
often missed by subjects on the previous round. An immediate
Basic Research on Training Principles 51
test followed the training rounds, and a retention test was given
1 week later. Both tests involved all 64 facts tested without
any prior study. At each test, half of the facts were in the same
format (specific or general) as at training and half were in dif-
ferent formats at training and test. The full condition involved
no training compression. The dropout, yoked, and clicker con-
ditions all involved compression of training. The compression
was individualized in the dropout condition and was based on
group performance in the clicker condition. The yoked condition
served as a control because the compression depended on the
performance of only one individual but not of the subject.
Performance was better for the clicker, dropout, and full con-
ditions than for the yoked condition at the immediate test. A
similar pattern was evident at the retention test 1 week later, but
forgetting occurred, and the advantage for the clicker condition
was reduced to some extent because of greater forgetting in that
case (see Figure 3.6). Finding test performance in the clicker
and dropout conditions comparable to that in the full condi-
tion indicates that teaching time can be conserved successfully
without sacrificing the amount learned, in accordance with the
principle of training compression. In addition, this pattern of
Test
1.0
Immediate test
Retention test
0.8
Proportion Correct
0.6
0.4
0.2
0.0
Clicker Dropout Full Yoked
Condition
General
Specific
0.8
Proportion Correct
0.6
0.4
0.2
0.0
Study-Study Study-Test Test-Test
Condition
Figure 3.7 Proportion correct during training in Experiment 2 by
Anderson et al. (2010) as a function of condition and train-
ing fact type. Error bars show standard errors of the mean.
0.9
0.8
Hit Rate
Train-No Fire
0.7 Train-Fire
0.6
0.5
TRAINING Test-No Fire Test-Fire
TEST
Experiment 2
1.0
0.9
0.8
Hit Rate
Train-No Fire
Train-Fire
0.7
0.6
0.5
TRAINING Test-No Fire Test-Fire
TEST
Figure 3.8 Hit rate at training and test in Experiment 1 (top panel)
and in Experiment 2 (bottom panel) of study by Young et
al. (2011, Experiments 2 & 3) as a function of training con-
dition and test condition. Error bars show standard errors
of the mean.
6.2
No Threat
Mean Distance from Location Threat
6.0
5.8
5.6
5.4
5.2
5.0
1 2 3 4 5 6 7
Serial Position
Recall
1.00
Proportion of Errors
Free
0.95 Serial
0.90
0.85
1 2 3 4 5 6 7
Serial Position
0.8 2-Back
8-Back
Proportion Correct
0.6
0.4
0.2
0.0
Default Special
Response Location
Experiment 2
Color Accuracy
1.0
0.8 2-Back
8-Back
Proportion Correct
0.6
0.4
0.2
0.0
Default Special
Response Location
9.5
Mean Total Number Correct
9.0
8.5
8.0
All Positive Blocked Mixed
Acquisition
5.0
4.5
Mean Confidence Rating (1-6)
4.0
3.5
3.0
All Positive Blocked Mixed
Acquisition
Figure 3.11 Mean total number correct (out of 15 possible) (top panel)
and mean confidence rating (1–6) (bottom panel) at test as
a function of acquisition condition in study by Young and
Healy (2011). Error bars show standard errors of the mean.
64 Alice F. Healy and Lyle E. Bourne, Jr.
Conclusions
The purpose of this chapter was to review in some detail a col-
lection of recent basic research experiments designed to extend
in a number of directions the body of research on training prin-
ciples reviewed in chapter 2. The chapter addresses four ques-
tions following up on the implications drawn from that review.
The questions examined in this chapter concern (a) the general-
ity of training principles, (b) the interaction of principles when
combined, (c) the operation of principles in complex, dynamic
environments, and (d) finally the possibility that yet unidentified
principles would emerge.
These results provide an expanded and more general picture
of how training principles affect performance in specific tasks.
Much of what was observed is consistent with expectations
derived from the review of laboratory-based cognitive psycho-
logical studies of these principles. In many cases, however, the
expectations needed to be modified in light of the experimental
results that contradict or qualify some of the training principles
in these different contexts. Moreover, some of the results led to
unexpected findings that require formulation of principles that
could not be anticipated on the basis of previous experiments.
Thus, the experiments presented in this chapter provide a more
complete picture about how cognitive psychological effects can
be incorporated into training principles.
The research reviewed in this chapter also suggests possi-
ble ways of making transitions from the laboratory to real life
training needs. Those transitions will require the development
of training guidelines and, eventually, training specifications for
specific tasks and jobs. That work will require applied research
of the type discussed in chapter 7 and elsewhere in this volume.
But the applications must begin with empirically valid training
principles of the type illustrated in the present chapter. There is
every reason to believe that a transition of this sort is the opti-
mal way of enhancing the effectiveness of real-world training.
References
Anderson, L. S., Healy, A. F., Kole, J. A., & Bourne, L. E., Jr. (2010,
November). The clicker technique: An effective way to compress
teaching time. Paper presented at the 51st Annual Meeting of the
Psychonomic Society, St. Louis, MO.
Baddeley, A. (2007). Working memory, thought, and action. New York:
Oxford University Press.
Basic Research on Training Principles 65
Barshi, I., & Healy, A. F. (2002). The effects of mental representa-
tion on performance in a navigation task. Memory & Cognition, 30,
1189–1203.
Barshi, I., & Healy, A. F. (2011). The effects of spatial representation
on memory for verbal navigation instructions. Memory & Cognition,
39, 47–62.
Bjork, R. A. (1994). Memory and metamemory considerations in the
training of human beings. In J. Metcalfe & A. Shimamura (Eds.),
Metacognition: Knowing about knowing (pp. 185–205). Cambridge,
MA: MIT Press.
Bonk, W. J., & Healy, A. F. (2010). Learning and memory for sequences
of pictures, words, and spatial locations: An exploration of serial
position effects. American Journal of Psychology, 123, 137–168.
Bourne, L. E., Jr., Healy, A. F., Bonk, W. J., & Buck-Gengler, C. J.
(2011). Intention to respond in a special way offers some protection
against forgetting associations. American Journal of Psychology,
124, 23–36.
Brown, G. D. A., Neath, I., & Chater, N. (2007). A temporal ratio model
of memory. Psychological Review, 114, 539–576.
Dale, E. (1969). Audio-visual methods in teaching (3rd ed.). New York:
Holt, Rinehart & Winston.
Gomez, R. L. (1997). Transfer and complexity in artificial grammar
learning. Cognitive Psychology, 33, 154–207.
Healy, A. F., Fendrich, D. W., & Proctor, J. D. (1990). Acquisition and
retention of a letter-detection skill. Journal of Experimental Psychol-
ogy: Learning, Memory, and Cognition, 16, 270–281.
Healy, A. F., Wohldmann, E. L., & Bourne, L. E., Jr. (2005). The pro-
cedural reinstatement principle: Studies on training, retention, and
transfer. In A. F. Healy (Ed.), Experimental cognitive psychology and
its applications (pp. 59–71). Washington, DC: American Psychologi-
cal Association.
Karpicke, J. D., & Roediger, H. L., III. (2008). The critical importance of
retrieval for learning. Science, 319, 966–968.
Ketels, S. L., Healy, A. F., Wickens, C. D., Buck-Gengler, C. J., & Bourne,
L. E., Jr. (2010, April). Spatial list learning and decision making in
the fusion paradigm. Poster presented at the 80th Annual Conven-
tion of the Rocky Mountain Psychological Association, Denver, CO.
Ketels, S. L., Healy, A. F., Wickens, C. D., Buck-Gengler, C. J., & Bourne,
L. E., Jr. (2011, March). A dual-process account of decision making:
Memory and anchoring. Paper presented in the Training Symposium
at the 57th Annual Meeting of the Southeastern Psychological Asso-
ciation, Jacksonville, FL.
Kole, J. A., Healy, A. F., Fierman, D. M., & Bourne, L. E., Jr. (2010).
Contextual memory and skill transfer in category search. Memory &
Cognition, 38, 67–82.
McCormick, B. A., Schneider, V. I., & Healy, A. F. (2010). [Comparing
verbal and symbol modalities in a navigational task]. Unpublished
raw data.
66 Alice F. Healy and Lyle E. Bourne, Jr.
Murdock, B. B., Jr. (1960). The distinctiveness of stimuli. Psychological
Review, 67, 16–31.
Roediger, H. L., III, & Karpicke, J. D. (2006). The power of testing mem-
ory: Basic research and implications for educational practice. Per-
spectives on Psychological Science, 1, 181–210.
Schneider, V. I., Healy, A. F., & Bourne, L. E., Jr. (2002). What is learned
under difficult conditions is hard to forget: Contextual interference
effects in foreign vocabulary acquisition, retention, and transfer.
Journal of Memory and Language, 46, 419–440.
Staal, M. A., Bolton, A. E., Yaroush, R. A., & Bourne, L. E., Jr. (2008).
Cognitive performance and resilience to stress. In B. Lukey & V.
Tepe (Eds.), Biobehavioral resilience to stress (pp. 259–300). Boca
Raton, FL: CRC Press.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty:
Heuristics and biases. Science, 185, 1124–1130.
Young, M. D., & Healy, A. F. (2011, March). Artificial grammar learn-
ing: Implicit and explicit components for retention and transfer. Paper
presented in the Training Symposium at the 57th Annual Meeting of
the Southeastern Psychological Association, Jacksonville, FL.
Young, M. D., Healy, A. F., Gonzalez, C., Dutt, V., & Bourne, L. E., Jr.
(2011). Effects of training with added difficulties on RADAR detec-
tion. Applied Cognitive Psychology, 25, 395–407.
4 Attention and Cognitive
Resource Load in Training
Strategies
Chris Wickens
University of Colorado
John Cumming
Colorado State University
A B C D
“fixed”
resource Germane load
supply
Intrinsic load
resources
Extraneous load
D S D S D S D S
Summary
Collectively then, all of these substrategies make the same point:
Any vehicle that can induce the allocation of more resources into
the task to be mastered can improve learning if those resources
go into germane load (Example B in Figure 4.1). But “the task”
must be carefully trisected into its three load components. A
vehicle that makes the task too difficult, temporarily increasing
the resources demanded by intrinsic load (necessary to sustain
performance) can induce “thrashing” (Example C)—this idea is
sometimes referred to as staying within the “zone of learnabil-
ity” during training (Wolfe et al., 1998). A vehicle that misdirects
attention to irrelevant (albeit sometimes engaging) aspects of the
pedagogy will simply impose added extrinsic load (Example D).
Scanning Training
Another way of dividing up the parts of a multiple task scenario
when the tasks are visual is to simply train the attention allo-
cation component. The logic of such an approach is inherent in
the findings that experts show qualitatively different scan pat-
terns in complex multitask skills, like flying a plane (Bellenkes,
Wickens, & Kramer, 1997; Schriver, Morrow, Wickens, & Talleur,
2008; Wickens et al., 2008), or assisting in the hospital operat-
ing room (Koh, Park, Wickens, Ong, & Chia, 2011). Hence one
should be able to capture experts’ scan patterns, extract their
general features, and train the appropriate allocation behav-
ior separately from the rest of the task. Shapiro and Raymond
(1989) have done so for a complex Space-Fortress video game in
a way that was more productive than whole task training (but
less so than other forms of part task training). However, in con-
trast, Bellenkes (1999) failed to reveal any effectiveness of scan
training on the fundamental aspects of multi-axis control in
Attention and Cognitive Resource Load 81
aircraft flight. That is, training the scan pattern directly failed
to improve the mental model of multi-axis flight dynamics. But
Bellenkes found greater success was obtained with a second
group that was first trained by developing that mental model of
flight dynamics, via a paper-based instructional sheet, showing
how changes on one axis affected changes on others.
Additional Considerations
A set of training principles and associated strategies has been
described above within the context of the resource allocation
metaphor. In considering their collective effects, three general
findings emerge, that transcend individual strategies.
Conclusion
This chapter provides an “attention-centric” view of training,
anchored in cognitive load theory. In so doing, many other
principles and strategies discussed in this volume, based on
memory and learning theory have been ignored. It is strongly
believed that attention and resource allocation are necessary,
but far from sufficient for effective learning and training to
take place, and so these are important considerations in the
design of any training environment. The general trichotomy
distinguishing between intrinsic, germane, and extraneous load
can always be applied to any training environment, even as the
boundaries between the three are sometimes fuzzy (Kalyuga,
2011). The greatest challenge for researchers would appear to
be identifying that boundary between germane and intrinsic
load, and understanding the intriguing paradox that sometimes
allocating more resources into intrinsic task load, to sustain
performance, does not assist skill acquisition, as intrinsic load
competes with resources for germane load. A challenge for
training research will be to continue to disentangle these issues.
Acknowledgments
Portions of this work were supported by the US Army Research
Institute under Contract #: W91WAW-09-C-0081 to Alion Sci-
ence and Technology titled Understanding the Impact of Train-
ing on Performance.
References
Bellenkes, A. (1999). The use of pilot performance models to facilitate
cockpit visual scanning. Unpublished Doctorial Dissertation. Uni-
versity of Illinois at Urbana Champaign.
Bellenkes, A. H., Wickens, C. D., & Kramer, A. F. (1997). Visual scan-
ning and pilot expertise: The role of attentional flexibility and mental
model development. Aviation, Space, and Environmental Medicine,
68, 569–579.
Bjork, R. A. (1999). Assessing our own competence: Heuristics and illu-
sions. In D. Gopher & A. Koriat (Eds.), Attention and performance:
Vol. 17. Cognitive regulation of performance: Interaction of theory and
application (pp. 435–459). Cambridge, MA: MIT Press.
Briggs, G., & Naylor, J. (1962). The relative efficiency of several training
methods as a function of transfer task complexity. Journal of Experi-
mental Psychology, 64, 505–512.
Attention and Cognitive Resource Load 85
Carroll, J. (1990). The Nurnberg Funnel: Designing minimalist instruc-
tion for practical computer skills. Cambridge, MA: MIT Press.
Carroll, J. M., & Carrithers, C. (1984). Blocking learner error states in
a training-wheels system. Human Factors, 26, 377–389.
Colquitt, J. A., LePine, J. A., & Noe, R. A. (2000). Toward an integrative
theory of training motivation: A meta-analytic path analysis of 20
years of research. Journal of Applied Psychology, 85, 678–707.
Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A frame-
work for memory research. Journal of Verbal Learning and Verbal
Behavior 11, 671–684.
Damos, D., & Wickens, C. D. (1980). The identification and transfer of
time-sharing skills. Acta Psychologica, 46, 15–39.
Endsley, M. R. (1995). Towards a theory of situation awareness in
dynamic systems, Human Factors, 37, 32–64.
Endsley, M. R., & Kiris, E. O. (1995). The out-of-the-loop performance
problem and level of control in automation. Human Factors, 37,
381–394.
Fabiani, M., Buckely, J., Gratton, G., Coles, M., Donchin, E., & Logie,
R. (1989). The training of complex task performance. Acta Psycho-
logica, 71, 179–199.
Gopher, D. (2007). Emphasis change as a training protocol for high-
demand tasks. In A. Kramer, D. Wiegmann, & A. Kirlik (Eds.), Atten-
tion: From theory to practice (pp. 209–224). Oxford, England: Oxford
University Press.
Gopher, D., Weil, M., & Barakeit, T. (1994). Transfer of skill from a com-
puter game trainer to fl ight. Human Factors, 36, 387–405.
Gopher, D., Weil, M., & Siegel, D. (1989). Practice under changing pri-
orities: An approach to training of complex skills. Acta Psychologica,
71, 147–197.
Halford, G. S., Wilson, W. H., & Phillips, S. (1998). Processing capac-
ity defined by relational complexity: Implications for comparative,
developmental, and cognitive psychology. Behavioral and Brain Sci-
ences, 21, 803–864.
Holding, D. H. (1976). An approximate transfer surface. Journal of
Motor Behavior, 8, 1–9.
Horrey, W., Lesch, A., & Garabet, A. (2009). Dissociation between driv-
ing performance and driver’s subjective estimates of performance
and workload in dual task conditions. Journal of Safety Research,
40, 7–12.
Kaber, D. B., & Endsley, M. R. (2003). The effects of level of automation
and adaptive automation on human performance, situation aware-
ness and workload in a dynamic control task. Theoretical Issues in
Ergonomic Science, 3, 1–40.
Kahneman, D. (1973). Attention and effort. Englewood Cliffs, NJ: Pren-
tice Hall.
Kalyuga, S. (2011) Cognitive load theory: How many types of load does
it really need? Educational Psychology Review, 23, 1–19.
Kalyuga, S., Chandler, P., & Sweller, J. (1998). Levels of expertise and
instructional design. Human Factors, 40, 1–17.
86 Christopher D. Wickens et al.
Karpicke, J., & Roediger, H. (2008). The critical importance of retrieval
for learning. Science, 319, 966–968.
Koh, R., Park, T., Wickens, C., Ong, L. T., & Chia, S. N. (2011) Differ-
ences in attentional strategies by novice and experienced operating
theatre scrub nurses. Journal of Experimental Psychology: Applied,
17, 233–246.
Kramer, A. F., Larish, J. F., & Strayer, D. L. (1995). Training for atten-
tional control in dual task settings: A comparison of young and old
adults. Journal of Experimental Psychology: Applied, 1, 50–76.
Lintern, G., & Roscoe, S. (1980). Visual cue augmentation in contact
flight and simulation. In S. N. Roscoe (Ed.), Aviation psychology (pp.
227–238). Ames, IA: Iowa State University.
Lintern, G., & Wickens, C. D. (1991). Issues for acquisition in transfer
of timesharing and dual-task skills. In D. Damos (Ed.), Multiple-task
performance (pp. 123–138). London: Taylor & Francis.
Mane, A., Adams, J., & Donchin, E. (1989) Adaptive and part-whole
training in the acquisition of a complex perceptual-motor skill. Acta
Psychologica, 71, 179–196.
Manzey, D., & Parasuraman, R. (2010). Complacency and bias in
human use of automation: An attentional perspective. Human Fac-
tors, 52, 381–410.
Mayer, R. (2007). Research guidelines for multi-media instructions. In
F. Durso (Ed.), Reviews of human factors and ergonomics (Vol. 3, pp.
127–147). Santa Monica, CA: Human Factors.
Mayer, R., Griffith, E., Jurkowitz, T., & Rothman, D. (2008). Increas-
ing interestingness of extraneous details in a multimedia science
presentation leads to decreased learning. Journal of Experimental
Psychology: Applied, 24, 329–363.
Mayer, R. E., & Moreno, R. (2003). Nine ways to reduce cognitive load in
multimedia learning. Educational Psychologist, 38, 45–52.
Montgomery, H., Sharafi, P., & Hedman, L. R. (2004). Engaging in
activities involving information technology: Dimensions, mode and
flow. Human Factors, 46, 334–348.
Naylor, J., & Briggs, G. (1963). Effects of task complexity and task orga-
nization on the relative efficiency of part and whole training meth-
ods. Journal of Experimental Psychology, 65, 217–224.
Neimick , R., Sikorski, C., & Walborg, H. (1996) Learner control effects:
A review of reviews and a meta-analysis. Journal of Educational and
Computing Research, 15, 157–174.
Norman, D. A., & Bobrow, D. G. (1975). On data-limited and resource-
limited processes. Cognitive Psychology 7, 44–64.
Paas, F., Renkl, A., & Sweller, J. (2003). Introduction: Cognitive load
theory and instructional design: Recent developments [Guest Edito-
rial Statement]. Educational Psychologist, 38, 1–4.
Paas, F., Tuovinen, J., van Merriënboer, J. J. G., & Darabi, A. (2005).
A motivational perspective on the relation between mental effort
and performance: Optimizing learners’ involvement in instructional
conditions. Educational Technology, Research and Development, 53,
25–33.
Attention and Cognitive Resource Load 87
Paas, F., & van Gog, T. (2009) Principles for designing effective and effi-
cient training for complex cognitive skills. In F. Durso (Ed.), Reviews
of human factors and ergonomics (Vol. 5, pp. 166–194). Santa Mon-
ica, CA: Human Factors.
Parasuraman, R., Sheridan, T., & Wickens, C. D. (2008). Situation
awareness, mental workload, and trust in automation: Viable,
empirically supported cognitive engineering constructs. Journal of
Cognitive Engineering and Decision Making, 2, 140–160.
Parasuraman, R., & Wickens, C. D. (2008). Humans still vital after
all these years of automation [Special issue]. Human Factors, 3,
511–520.
Pea, R. D. (2004). The social and technological dimensions of scaf-
folding and related theoretical concepts for learning, education, and
human activity. The Journal of the Learning Sciences, 13, 423–451.
Pearce, P. (1981) Route maps: A study of traveller’s perception of a sec-
tion of countryside. Journal of Environmental Psychology, 1, 141–153.
Pollock, E., Chandler, P., & Sweller, J. (2002). Assimilating complex
information. Learning and Instruction, 12, 61–86.
Povenmire, H. K., & Roscoe, S. N. (1973). Incremental transfer effective-
ness of a ground-based general aviation trainer. Human Factors, 15,
534–542.
Renkl, A., Stark, R., Gruber, H., & Mandl, H. (1998). Learning from
worked-out examples: The effects of example variability and elic-
ited self-explanations. Contemporary Educational Psychology, 23,
90–108.
Roediger, H., & Karpicke, J (2006) Test-enhanced learning: Taking
memory tests improves long-term retention. Psychological Science,
17, 249–255.
Sarter, N. B., & Schroeder, B. (2001). Supporting decision making and
action selection under time pressure and uncertainty: The case of
in-fl ight icing. Human Factors, 43, 573–583.
Schmidt, R. A., & Bjork, R. A. (1992). New conceptualizations of prac-
tice: Common principles in three paradigms suggest new concepts
for training. Psychological Science, 3, 207–217.
Schneider, W. (1985). Training high-performance skills: Fallacies and
guidelines. Human Factors, 27, 285–300.
Schneider, W., & Detweiler, M. (1988). The role of practice in dual-task
performance: Toward workload modeling in a connectionist/control
architecture. Human Factors, 30, 539–566.
Schriver, A. T., Morrow, D. G., Wickens, C. D., & Talleur, D. A. (2008).
Expertise differences in attentional strategies related to pilot deci-
sion making. Human Factors, 50, 864–878.
Shapiro, K. L., & Raymond, J. E. (1989). Training of efficient oculo-
motor strategies enhances skill acquisition. Acta Psychologica, 71,
217–242.
Sheridan, T., & Parasuraman, R. (2006). Human–automation interac-
tion. Reviews of Human Factors and Ergonomics, 1, 89–129.
88 Christopher D. Wickens et al.
Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of
a phenomenon. Journal of Experimental Psychology: Human Learn-
ing and Memory, 4, 592–604.
Sweller, J. (1988). Cognitive load during problem solving: Effects on
learning. Cognitive Science, 12, 257–285.
van Merriënboer, J. J. G., Kester, L., & Paas, F. (2006). Teaching com-
plex rather than simple tasks: Balancing intrinsic and germane load
to enhance transfer of learning. Applied Cognitive Psychology 20,
343–352.
Wickens, C. D. (1992). Virtual reality and education. Proceedings of
the IEEE International Conference on Systems, Man, and Cybernetics
(Vol. 1, pp. 842–847) New York: IEEE.
Wickens, C. D. (2008a). Multiple resources and mental workload
[Golden Anniversary Special issue]. Human Factors, 50, 449–455.
Wickens, C. D. (2008b). Situation awareness: Review of Mica Endsley’s
articles on situation awareness [Golden Anniversary Special issue].
Human Factors, 50, 397–403.
Wickens, C. D., & Hollands, J. (2000) Engineering psychology and
human performance (3rd ed.). Upper Saddle River, NJ: Prentice Hall.
Wickens, C. D., Hutchins, S., Carolan T., & Cumming, J. (2011). Investi-
gating the impact of training on transfer: A meta-analysis. Proceed-
ings of the 2011 Conference of the Human Factors and Ergonomics
Society. Santa Monica, CA: Human Factors.
Wickens, C. D., Hutchins, S., Carolan, T., & Cummings, J (2012). Part
task training and increasing difficulty training strategies: A meta-
analysis approach. Human Factors, 54, ppxx.
Wickens C. D., Li, H., Santamaria, A., Sebok, A., & Sarter, N. (2010).
Stages & Levels of automation, An integrated meta-analysis. In
proceedings, 2011 conference of the Human Factors & Ergonomics
Society. Santa Monica, CA: Human Factors.
Wickens, C. D., & McCarley, J. (2008) Applied attention theory. Boca
Raton, FL: CRC Press.
Wickens, C. D., McCarley, J. S., Alexander, A. L., Thomas, L. C.,
Ambinder, M., & Zheng, S. (2008). Attention-situation awareness
(A-SA) model of pilot error. In D. Foyle & B. Hooey (Eds.), Human
performance models in aviation (pp. 213–239). Boca Raton, FL: CRC
Press.
Wightman, D., & Lintern, G. (1985). Part task training for tracking and
manual control. Human Factors, 27, 267–284.
Williams, H. P., Hutchinson, S., & Wickens, C. D. (1996). A comparison
of methods for promoting geographic knowledge in simulated air-
craft navigation. Human Factors, 38, 50–64.
Wolfe, M. B. W., Schreiner, M. E., Rehder, B., Laham, D., Foltz, P. W.,
Kintsch, W., & Landauer, T. K. (1998). Learning from text: Matching
readers and text by latent semantic analysis. Discourse Processes,
25, 309–336.
5 Acquisition and Transfer
of Basic Skill Components
Robert W. Proctor, Motonori
Yamaguchi, and James D. Miles
Purdue University
25 25
a b
20 20
r---
15 15
I 10
I 10
u u
~ ~
w w
c c
0 0
E E
u; u;
L______j
-5 -5
-10 -10
No Practice After Practice 100 300 600
Practice Type Number of Practice Trials
Figure 5.1 Simon effect (a) with no prior practice and after < 100 tri-
als of practice with an incompatible spatial mapping and
(b) as a function of number of practice trials (Proctor,
Yamaguchi, Zhang, & Vu, 2009).
94 Robert W. Proctor, Motonori Yamaguchi, and James Miles
through which spatial information is conveyed (physical location
of a circle, pointing direction of an arrow, the meaning of spatial
words; Proctor et al., 2009). In particular, the transfer effect
was evident after less than 100 trials of practice with physical
locations or with the pointing direction of arrows. Although the
Simon effect tended to be larger for the arrow stimuli than for
the location stimuli, the size of the transfer effect was equivalent
for the two stimulus modes. In contrast, after practice with the
word stimuli for less than 100 trials, there was little indication
of the transfer effect. Nevertheless, when the number of practice
trials was increased to 300, the transfer effect was observed (as
shown in Figure 5.1b), and it was as large as that for the location
and arrow stimuli. Because responses were made by pressing
the left and right keys, set-level compatibility was higher for the
location and arrow stimuli than for the word stimuli (cf. Proctor
& Wang, 1997). Therefore, a similar experiment was conducted
with vocal responses (i.e., saying “left” or “right”; Proctor et al.,
2009), for which set-level compatibility should be higher for the
word stimuli than for the location and arrow stimuli. However,
in this case, the transfer effect still was evident for the location
stimuli after less than 100 practice trials, but it appeared for
the word stimuli only after the number of trials was doubled.
This outcome suggests that the learning rate is not dependent
on set-level compatibility but, regardless of the response mode,
is determined by the stimulus type.
Specificity of Transfer
Another aspect of learning is its limited transfer to other situ-
ations. According to the framework, learning is utilized best in
a context that is similar to the original context in which the
learning has taken place, the principle of transfer specificity (see
chapter 2, this volume). The influence of contextual similarity
of the current trial to past trials is expressed by the exponen-
tial component of Equation 1, where Si is the similarity of the
ith practice trial to the current trial. A well-known nonmetric
theory of similarity judgment is Tversky’s (1977) contrast model
in which an object or event is considered to be a set of unique
features. Then, the similarity between two objects Xi and Xj is
expressed by
S ij = f (X i ∩ X j ) − g (X i / X j ) − h(X j / X i ) , (2)
Study Context (C p )
S(C p , C t )
Cp / C t
Cp ∩ C t
Ct / Cp
Test Context (C t )
Figure 5.2 A feature overlap account of contextual similarity.
S (C p , C t ) = f (C p ∩ C t ) . (3)
Implementation Instructions
Although practiced skills influence performance of a subse-
quent task, there is evidence to suggest that instructions alone
may lead to similar effects. Gollwitzer (1999) has presented con-
siderable evidence that implementation intention instructions
provide an effective and targeted way to improve performance
without the need for practice. Rather than designating a general
task goal, such instructions are presented in an “if … then …”
format that defines the relation between a specified target (“if
a certain situation occurs”) and its associated response (“then
take a particular action”). By prespecifying this association,
responses purportedly become automatic; therefore, they are
immune to interference and lead to faster and more accurate
performance without a trade-off for alternative, nonspecified
actions. For example, Brandstatter, Lengfelder, and Gollwitzer
(2001) had subjects respond to digits in a stream of letters and
digits, and included the implementation instruction to respond
especially fast to the digit 5. Responses to the digit 5 showed
a performance benefit that was not influenced by the level of
Acquisition and Transfer of Basic Skill Components 97
cognitive load imposed by a concurrent task, indicating that
the target of the implemented instruction was not susceptible
to interference.
Cohen, Bayer, Jaudas, and Gollwitzer (2008) used implemen-
tation intention instructions to designate an opposite response
on specific incompatible trials in the Simon task (e.g., “if a red
circle appears on the left side, then I press the right button”)
and found an elimination of the Simon effect. Miles and Proc-
tor (2008) showed that implementation intention instructions
led to faster responses on both incompatible and compatible tri-
als, though, suggesting that the instructions do not replace the
long-terms associations responsible for the Simon effect but set
up new associations that act in addition to them. However, the
instructions were still beneficial to performance of the specified
actions and did not lead to a performance cost for nonspeci-
fied actions. Exactly why implementation intention instructions
produce benefits in performance on a subsequent task is a topic
deserving of further research (see also the related discussion of
mental practice in chapter 2).
Encode display
information
Determine mapping
rule
Make response
Conclusion
The research described in this chapter illustrates that there
are benefits of applying individual principles in the training of
Acquisition and Transfer of Basic Skill Components 107
specific tasks. A small amount of practice responding to a stim-
ulus dimension may have a long-lasting impact on performance
when that dimension is no longer relevant. Many results are
consistent with skill learning being an increasing function of
the number of trials, the rate of learning for a particular stimu-
lus mode, and the amount of allocated attention. The degree
of transfer varies directly as a function of contextual similar-
ity, with the greatest transfer occurring when the stimulus and
response modes are the same in practice and transfer condi-
tions. The amount of transfer is largely uninfluenced by the
passage of time, although intervening tasks with different S-R
mappings tend to reduce the amount of transfer.
The expression of preexisting biases is also influenced by
the current task environment. Factors such as similarity in
stimulus and response modes for two mixed or concurrent
tasks, as well as consistencies of mapping rules across the
tasks, are critical. Relative payoffs and frequencies for the dif-
ferent tasks and mappings in these more complex task envi-
ronments need to be taken into account in characterizing
skilled performance.
References
Allport, A., Styles, E. A., & Hsieh, S. (1994). Shifting intentional set:
Exploring the dynamic control of tasks. In C. Umiltà & M. Mosco-
vitch (Eds.), Attention and performance: Vol. 15. Conscious and non-
conscious information processing (pp. 421–452). Cambridge, MA:
MIT Press.
Alluisi, E. A., & Warm, J. S. (1990). Things that go together: A review of
stimulus-response compatibility and related effects. In R. W. Proctor
& T. G. Reeve (Eds.), Stimulus-response compatibility: An integrated
perspective (pp. 3–30). Amsterdam, the Netherlands: North-Holland.
Bae, G. Y., Cho, Y. S., & Proctor, R. W. (2009). Transfer of orthogonal
stimulus-response mappings to an orthogonal Simon task. Quar-
terly Journal of Experimental Psychology, 62, 746–765.
Brandstatter, V., Lengfelder, A., & Gollwitzer, P. M. (2001). Implementa-
tion intentions and efficient action initiation. Journal of Personality
and Social Psychology, 81, 946–960.
Cohen, A. L., Bayer, U. C., Jaudas, A., & Gollwitzer, P. M. (2008). Self-
regulatory strategy and executive control: Implementation intentions
modulate task switching and Simon task performance. Psychologi-
cal Research, 72, 12–26.
Duncan, J. (1979). Divided attention: The whole is more than the sum
of its parts. Journal of Experimental Psychology: Human Perception
and Performance, 5, 216–228.
Dutt, V., Gonzalez, C., Yamaguchi, M., & Proctor, R. W. (2012). Instance-
based learning models of SRC and Simon effects. Manuscript sub-
mitted for publication.
108 Robert W. Proctor, Motonori Yamaguchi, and James Miles
Dutta, A., & Proctor, R. W. (1992). Persistence of stimulus-response
compatibility effects with extended practice. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 18, 801–809.
Eder, A. B., Rothermund, K., & Proctor, R. W. (2010). The prepared
emotional reflex: Intentional preparation of automatic approach and
avoidance tendencies as a means to regulate emotional responding.
Emotion, 10, 593–598.
Fitts, P. M., & Deininger, R. L. (1954). S-R compatibility: Correspon-
dence among paired elements within stimulus and response codes.
Journal of Experimental Psychology, 48, 483–492.
Fitts, P. M., & Seeger, C. M. (1953). S-R compatibility: Spatial charac-
teristics of stimulus and response codes. Journal of Experimental
Psychology, 46, 199–210.
Gollwitzer, P. M. (1999). Implementation intentions: Strong effects of
simple plans. American Psychologist, 54, 493–503.
Gonzalez, C., Lerch, J. F., & Lebiere, C. (2003). Instance-based learn-
ing in dynamic decision making. Cognitive Science, 27, 591–635.
Greenwald, A. G. (2005). A reminder about procedures needed to reli-
ably produce perfect timesharing: Comment on Lien, McCann,
Ruthruff, and Proctor (2005). Journal of Experimental Psychology:
Human Perception and Performance, 31, 221–225.
Kiesel, A., Steinhauser, M., Wendt, M., Falkenstein, M., Jost, K., Philipp,
A. M., & Koch, I. (2010). Control and interference in task switching—
A review. Psychological Bulletin, 136, 849–874.
Lewandowsky, S., Oberauer, K., & Brown, G. A. (2009). No temporal
decay in verbal short-term memory. Trends in Cognitive Sciences,
13(3), 120–126.
Lien, M., McCann, R. S., Ruthruff, E., & Proctor, R. W. (2005). Con-
firming and disconfirming theories about ideomotor compatibility
in dual-task performance: A reply to Greenwald (2005). Journal of
Experimental Psychology: Human Perception and Performance, 31,
226–229.
Lien, M.-C., & Proctor, R. W. (2002). Stimulus-response compatibility
and psychological refractory period effects: Implications for response
selection. Psychonomic Bulletin & Review, 9, 212–238.
Logan, G. D. (1988). Toward an instance theory of automatization. Psy-
chological Review, 95, 492–527.
Miles, J. D., & Proctor, R. W. (2008). Improving performance through
implementation intentions: Are preexisting response biases
replaced? Psychonomic Bulletin & Review, 15, 1105–1110.
Miles, J. D., & Proctor, R. W. (2010). Attention is required for the
acquisition but not expression of new spatial associations. Journal
of Experimental Psychology: Learning, Memory, and Cognition, 36,
1554–1560.
Mordkoff, J. T., & Hazeltine, E. (2011).Responding to the source of stim-
ulation: J. Richard Simon and the Simon effect [Special issue]. Acta
Psychologica, 136(2).
Müsseler, J., Aschersleben, G., Arning, K., & Proctor, R. W. (2009).
Reversed effects of spatial compatibility in natural scenes. American
Journal of Psychology, 122, 325–336.
Acquisition and Transfer of Basic Skill Components 109
Newell, A., & Rosenbloom, P. S. (1981). Mechanisms of skill acquisition
and the law of practice. In J. R. Anderson (Ed.), Cognitive skills and
their acquisition (pp. 1–55). Hillsdale, NJ: Erlbaum.
Notebaert, W., De Moor, W., Gevers, W., & Hartsuiker, R. J. (2007). New
visuospatial associations by training verbospatial mappings in the
first language. Psychonomic Bulletin & Review, 14, 1183–1188.
Pashler, H. (1984). Processing stages in overlapping tasks: Evidence
for a central bottleneck. Journal of Experimental Psychology: Human
Perception & Performance, 10, 358–377.
Portrat, S., Barrouillet, P., & Camos, V. (2008). Time-related decay or
interference-based forgetting in working memory? Journal of Experi-
mental Psychology: Learning, Memory, and Cognition, 34, 1561–1564.
Proctor, R. W., Koch, I., Vu, K.-P. L., & Yamaguchi, M. (2008). Influ-
ence of display and cue formats on switch costs and the prevalence
effect for horizontal and vertical dimensions. Memory & Cognition,
36, 998–1012.
Proctor, R. W., & Lu, C.-H. (1999). Processing irrelevant location infor-
mation: Practice and transfer effects in choice-reaction tasks. Mem-
ory & Cognition, 27, 63–77.
Proctor, R. W., Marble, J. G., & Vu, K.-P. L. (2000). Mixing incompat-
ibly mapped location-relevant trials with location-irrelevant trials:
Effects of stimulus mode on the reverse Simon effect. Psychological
Research, 64, 11–24.
Proctor, R. W., & Shao, C. (2010). Does the contribution of stimulus-
hand correspondence to the auditory Simon effect increase with
practice? Experimental Brain Research, 204, 131–137.
Proctor, R. W., & Vu, K.-P. L. (2002). Mixing location irrelevant and
relevant trials: Influence of stimulus mode on spatial compatibility
effects. Memory & Cognition, 30, 281–294.
Proctor, R. W., & Vu, K.-P. L. (2006a). Laboratory studies of training,
skill acquisition, and retention of performance. In K. A. Ericsson, N.
Charness, P. J. Feltovich, & R. R. Hoffman (Eds.), Cambridge hand-
book of expertise and expert performance (pp. 265–286). Cambridge,
England: Cambridge University Press.
Proctor, R. W., & Vu, K.-P. L. (2006b). Stimulus-response compatibility
principles: Data, theory, and application. Boca Raton, FL: CRC Press.
Proctor, R. W., & Vu, K.-P. L. (2009a). Determinants of the benefit for
consistent stimulus-response mappings in dual-task performance
of four-choice tasks. Attention, Perception and Psychophysics, 71,
734–756.
Proctor, R. W., & Vu, K.-P. L. (2009b). Determinants of the benefit for
consistent stimulus-response mappings in dual-task performance
of three-choice tasks. Attention, Perception, and Psychophysics, 71,
1771–1781.
Proctor, R. W., & Vu, K.-P. L. (2009c). Task-defined associations are
mode specific for selection of relevant dimension but mode inde-
pendent for selection of mapping. Quarterly Journal of Experimental
Psychology, 62, 371–392.
110 Robert W. Proctor, Motonori Yamaguchi, and James Miles
Proctor, R. W., & Vu, K.-P. L. (2010a). Stimulus-response compatibil-
ity for mixed mappings and tasks with unique responses. Quarterly
Journal of Experimental Psychology, 63, 320–340.
Proctor, R. W., & Vu, K.-P. L. (2010b). Universal and culture-specific
effects of display-control compatibility. American Journal of Psychol-
ogy, 123, 425–435.
Proctor, R. W., Vu, K.-P. L., & Marble, J. G. (2003). Eliminating spa-
tial compatibility effects for location-relevant trials by intermixing
location-irrelevant trials. Visual Cognition, 10, 15–50.
Proctor, R. W., & Wang, H. (1997). Differentiating types of set-level
compatibility. In B. Hommel & W. Prinz (Eds.), Theoretical issues in
stimulus-response compatibility (pp. 11–37). Amsterdam, the Neth-
erlands: North-Holland.
Proctor, R. W., Yamaguchi, M., Dutt, V., & Gonzalez, C. (2012). Disso-
ciation of S-R compatibility and Simon effects with mixed tasks and
mappings. Manuscript submitted for publication.
Proctor, R. W., Yamaguchi, M., & Vu, K.-P. L. (2007). Transfer of non-
corresponding spatial associations to the auditory Simon task. Jour-
nal of Experimental Psychology: Learning, Memory, and Cognition,
33, 245–253.
Proctor, R. W., Yamaguchi, M., Zhang, Y., & Vu, K.-P. L. (2009). Influ-
ence of visual stimulus mode on transfer of acquired spatial asso-
ciations. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 35, 434–445.
Shaffer, L. H. (1965). Choice reaction with variable S-R mapping. Jour-
nal of Experimental Psychology, 70, 284–288.
Shin, Y. K., Cho, Y. S., Lien, M.-C., & Proctor, R. W. (2007). Is the
psychological refractory period effect for ideomotor compatible tasks
eliminated by speed-stress instructions? Psychological Research,
71, 553–567.
Shin, Y. K., & Proctor, R. W. (2008). Are spatial responses to visuo-
spatial stimuli and spoken responses to auditory letters ideomo-
tor-compatible tasks? Examination of set-size effects on dual-task
interference. Acta Psychologica, 129, 352–364.
Simon, J. R. (1990). The effects of an irrelevant directional cue on
human information processing. In R. W. Proctor & T. G. Reeve
(Eds.), Stimulus-response compatibility: An integrated perspective
(pp. 31–86). Amsterdam: North-Holland.
Soetens, E., Maetens, K., & Zeischka, P. (2010). Practice-induced and
sequential modulations of the Simon effect. Attention, Perception, &
Psychophysics, 72, 895–911.
Tagliabue, M., Zorzi, M., Umiltà, C., & Bassignani, F. (2000). The role of
LTM links and STM links in the Simon effect. Journal of Experimen-
tal Psychology: Human Perception and Performance, 26, 648–670.
Telford, C. W. (1931). Refractory phase of voluntary and associative
response. Journal of Experimental Psychology, 14, 1–35.
Tversky, A. (1977). Features of similarity. Psychological Review, 84,
327–352.
Acquisition and Transfer of Basic Skill Components 111
Vu, K.-P. L. (2007). Influences on the Simon effect of prior practice with
spatially incompatible mappings: Transfer within and between hori-
zontal and vertical dimensions. Memory & Cognition, 35, 1463–1471.
Vu, K.-P. L., Ngo, T. K., Minakata, K., & Proctor, R. W. (2010). Shared
spatial representations for physical locations and location words in
bilinguals’ primary language. Memory & Cognition, 38, 713–722.
Vu, K.-P. L, & Proctor, R. W. (2004). Mixing compatible and incompat-
ible mappings: Elimination, reduction, and enhancement of spatial
compatibility effects. Quarterly Journal of Experimental Psychology,
57A, 539–556.
Vu, K.-P. L., & Proctor, R. W. (2006). Emergent perceptual features in
the benefit of consistent stimulus-response mappings on dual-task
performance. Psychological Research, 70, 468–483.
Vu, K.-P. L., & Proctor, R. W. (2008). Age differences in response selec-
tion for pure and mixed stimulus-response mappings and tasks.
Acta Psychologica, 129, 49–60.
Vu, K.-P. L, Proctor, R. W., & Urcuioli, P. (2003). Transfer effects of
incompatible location-relevant mappings on subsequent visual or
auditory Simon tasks. Memory & Cognition, 31, 1146–1152.
Wang, D.-Y. D., Proctor, R. W., & Pick, D. F. (2007). Acquisition and
transfer of attention-allocation strategies in a multiple-task work
environment. Human Factors, 49, 995–1004.
Wang, D.-Y. D., Proctor, R. W., & Pick, D. F. (2009). Allocation of effort
as a function of payoffs for individual tasks in a multitasking envi-
ronment. Behavior Research Methods, 41, 705–716.
Yamaguchi, M., & Proctor, R. W. (2006). Stimulus-response compat-
ibility with pure and mixed mappings in a flight task environment.
Journal of Experimental Psychology: Applied, 12, 207–222.
Yamaguchi, M., & Proctor, R. W. (2009). Transfer of learning in choice
reactions: Contributions of specific and general components of man-
ual responses. Acta Psychologica, 130, 1–10.
Yamaguchi, M., & Proctor, R. W. (2010). Compatibility of motion infor-
mation in two aircraft attitude displays for tracking task. American
Journal of Psychology, 123, 81–92.
Yamaguchi, M., & Proctor, R. W. (2011a). Automaticity without exten-
sive training: The role of memory retrieval in automatic implemen-
tation of task-defined rules. Psychonomic Bulletin & Review, 18,
347–354.
Yamaguchi, M., & Proctor, R. W. (2011b). The Simon task with multi-
components responses: Two loci of response-effect compatibility.
Psychological Research, 75, 214–226.
Yamaguchi, M., & Proctor, R. W. (in preparation). Transfer of an incom-
patible spatial mapping to the Simon task: Influence of an interven-
ing task with a different stimulus mode.
Zorzi, M., & Umiltà, C. (1995). A computational model of the Simon
effect. Psychological Research, 58, 193–205.
6 How Cognitive Ability
and Automation Influence
Training Performance and
Transfer
Eric D. Heggestad
University of North Carolina, Charlotte
Benjamin A. Clegg
Colorado State University
Adrian Goh
University of North Carolina, Charlotte
Robert S. Gutzwiller
Colorado State University
High
Input
Information Perceptual
Processors
Acquisition Speed
Low
Working
High Memory
Information Cognitive
Analysis Processors
Low
Spatial
High
Decision Memory
Automation Processors Reasoning
Low
High
Action Motor Psycho-
Automation Processors motor
Low
References
Ackerman, P. L. (1988). Determinants of individual differences dur-
ing skill acquisition: Cognitive abilities and information processing.
Journal of Experimental Psychology: General, 117, 288–318.
Anderson, J. R. (1982). Acquisition of cognitive skill. Psychological
Review, 89, 369–406.
Bainbridge, L. (1983). Ironies of automation. Automatica, 19, 755–779.
Ballas, J. A., Heitmeyer, C. L., & Pérez, M. A. (1992). Evaluating two
aspects of direct manipulation in advanced cockpits. Proceedings
of ACM CHI 92 Conference on Human Factors in Computing Systems
(pp. 127–134).
Billings, C. E. (1991). Human-centered aircraft automation: A concept
and guidelines (NASA Technical Memorandum 103885). Moffet
Field, CA: NASA Ames Research Center.
130 Eric D. Heggestad et al.
Blitch, J. G., & Clegg, B. A. (2011). The influences of automation and
trainee aptitude on training effectiveness. Proceedings of 55th
Annual Meeting of the Human Factors and Ergonomics Society.
Carroll, J. B. (1992). Cognitive abilities: The state of the art. Psychologi-
cal Science, 3, 266–270.
Carroll, J. B. (1993). Human cognitive abilities: A survey of factor-ana-
lytic studies. New York: Cambridge University Press.
Carroll, J. B. (1997). The three-stratum theory of cognitive abilities.
In Contemporary intellectual assessment: Theories, tests, and issues
(pp. 122–130). New York: Guilford.
Clamann, M. P., Wright, M. C., & Kaber, D. B. (2002). Comparison of
performance effects of adaptive automation applied to various stages
of human-machine system information processing. Proceedings of
the Human Factors and Ergonomics Society 46th Annual Meeting
(pp.342–346). Santa Monica, CA: Human Factors and Ergonomics
Society.
Clegg, B. A., & Heggestad, E. D. (2010). Automation and effective train-
ing (Technical Report for U.S. Army Research Office MURI Grant
W911NF-05-1-0153). Boulder, CO: Colorado State University.
Clegg, B. A., Heggestad, E. D., & Blalock, L. D. (2010). The influences of
automation and trainee aptitude on training effectiveness. Proceed-
ings of 54th Annual Meeting of the Human Factors and Ergonomics
Society.
Cronbach, L., & Snow, R. (1977). Aptitudes and instructional methods:
A handbook for research on interactions. Oxford, England: Irvington.
Cummings, M. L., & Guerlain, S. (2007). Developing operator capacity
estimates for supervisory control of autonomous vehicles. Human
Factors, 49, 1–15.
Dixon, S. R., Wickens, C. D., & Chang, D. (2003). Comparing quantita-
tive model predictions to experimental data in multiple-UAV fl ight
control. Proceedings of the 47th Annual Meeting of the Human Fac-
tors and Ergonomics Society.
Dzindolet, M. T., Beck, H. P., Pierce, L. G., & Dawe, L. A. (2001). A
framework of automation use (Report Number ARL-CR-2412). Aber-
deen Proving Ground, MD: Army Research Laboratory.
Endsley, M. R. (1996). Automation and situation awareness. In R. Para-
suraman & M. Mouloua (Eds.), Automation and human performance:
Theory and applications (pp. 163–181). Mahwah, NJ: Erlbaum.
Endsley, M. R., & Kaber, D. B. (1999). Level of automation effects on
performance, situational awareness, and workload in a dynamic
control task. Ergonomics, 42, 462–492.
Endsley, M. R., & Kiris, E. O. (1995). The out-of-the-loop performance
problem and level of control in automation. Human Factors, 37,
381–394.
Ferreti, R. P., & Butterfield, E. C. (1992). Intelligence-related differ-
ences in the learning, maintenance, and transfer of problem-solving
strategies. Intelligence, 16, 207–223.
Finn, A., & Scheding, S. (2010). Developments and challenges for auton-
omous unmanned vehicles. Berlin, Germany: Springer-Verlag.
Cognitive Ability and Automation 131
Fitts, P. M. (1951). Human engineering for an effective air-navigation
and traffic-control system. Columbus Ohio: Ohio State University
Research Foundation.
Fitts, P. M., & Posner, M. I. (1967). Human performance. Belmont, CA:
Brooks-Cole.
Funke, G., Matthews, G., Warm, J. S., & Emo, A. K. (2007). Vehicle
automation: A remedy for driver stress? Ergonomics, 50, 1302–1323.
Groeger, J. A., & Clegg, B. A. (2007). Systematic changes in the rate of
instruction during driver training. Applied Cognitive Psychology, 21,
1229–1244.
Gustafsson, J. E. (1984). A unifying model for the structure of intel-
lectual abilities. Intelligence, 8, 179–203.
Gustafsson, J. E. (1989). Broad and narrow abilities in research on
learning and instruction. In R. Kanfer, P. L. Ackerman, & R. Cudeck
(Eds.), Abilities, motivation, and methodology: The Minnesota Sym-
posium on Learning and Individual Differences (pp. 203–237). Hills-
dale, NJ: Erlbaum.
Hancock, P. A., & Scallen, S. F. (1998). Allocating functions in human-
machine systems. In R. R. Hoffman, M. F. Sherrick, & J. S. Warm
(Eds.), Viewing psychology as a whole: The integrative science of Wil-
liam M. Dember (pp. 509–539). Washington, DC: APA Press.
Kaber, D. B., & Endsley, M. R. (2004). The effects of level of automa-
tion and adaptive automation on human performance, situational
awareness, and workload in a dynamic control task. Theoretical
Issues in Ergonomics Science, 5, 113–153.
Kanfer, R., & Ackerman, P. L. (1989). Motivation and cognitive abili-
ties: An integrative/aptitude-treatment interaction approach to skill
acquisition. Journal of Applied Psychology, 74, 657–690.
Kieras, D., & Meyer, D. E. (1997). An overview of the EPIC architecture
for cognition and performance with application to human-computer
interaction. Human-Computer Interaction, 12, 391–438.
Kirlik, A. (1993). Modeling strategic behavior in human-automation
interaction: Why an “aid” can (and should) go unused. Human Fac-
tors, 35, 221–242.
Klahr, D., Langley, P., & Neches R. (1987). Production system models of
learning and development. Cambridge, MA: MIT Press.
Lee, J. D. (1992). Trust, self confidence and operators’ adaptation to
automation (Unpublished doctoral dissertation). University of Illinois
at Urbana-Champaign.
Lee, J. D., & Moray, N. (1992). Trust, control strategies and allocation of
function in human-machine systems. Ergonomics, 35, 1243–1270.
Liu , D., Wasson, R., & Vincenzi, D. A. (2009). Effects of system automa-
tion management strategies and multi-mission operator-to-vehicle
ratio on operator performance in UAV systems. Journal of Intelligent
and Robotic Systems, 54, 795–810.
Marmie, W. R., & Healy, A. F. (1995). The long-term retention of a
complex skill. In A. F. Healy & L. E. Bourne, Jr. (Eds.), Learning
and memory of knowledge and skills: Durability and specificity (pp.
30–65). Thousand Oaks, CA: Sage.
132 Eric D. Heggestad et al.
Martin, E., Lyon, D. R., & Schreiber, B. T. (1998). Designing synthetic
tasks for human factors research: An application to uninhabited air
vehicles. Proceedings of the Human Factors and Ergonomic Society.
Meyer, D. E., & Kieras, D. E. (1997). A computational theory of execu-
tive control processes and human multiple-task performance: Part
1. Basic Mechanisms. Psychological Review, 104, 3–65.
Moray, N. (1986). Monitoring behaviour and supervisory control. In K.
R. Boff, L. Kaufman, & J. P. Thomas (Eds.), Handbook of perception
and human performance (pp. 1–51). New York: Wiley.
Moray, N., & Inagaki, T. (1999). Laboratory studies of trust between
humans and machines in automated systems [Special issue on
Man–Machine Systems]. Transactions of the Institute for Measure-
ment and Control, 21, 203–211.
Moray, N., Rodriguez, D., & Clegg, B. A. (2000). Levels of automation
in process control. Proceedings of 44th Annual Meeting of the Human
Factors and Ergonomics Society.
Mourant, R. R., & Rockwell, T. H. (1970). Mapping eye-movement pat-
terns to the visual scene driving: An exploratory study. Human Fac-
tors, 12, 81–87.
Mourant, R. R., & Rockwell, T. H. (1972). Strategies of visual search by
novice and experienced drivers. Human Factors, 14, 325–335.
Muir, B. M., & Moray, N. (1996). Trust in automation. Part II. Experi-
mental studies of trust and human intervention in a process control
simulation. Ergonomics, 39, 429–461.
Norman, D. A. (1991). Cognitive artifacts. In J. M. Carroll (Ed.), Design-
ing interaction: Psychology at the human-computer interface (pp.
17–38). New York: Cambridge University Press.
Parasuraman, R., & Riley, V. (1997). Humans and automation: Use,
misuse, disuse, abuse. Human Factors, 39, 230–253.
Parasuraman, R., Sheridan, T. B., & Wickens, C. D. (2000). A model of
types and levels of human interaction with automation. IEEE Trans-
actions on Systems, Man, and Cybernetics. Part A: Systems and
Humans, 30, 286–297.
Parasuraman, R., & Wickens, C. D. (2008). Humans: Still vital after all
these years of automation. Human Factors, 50, 511–520.
Pashler, H., & Baylis, G. (1991). Procedural learning: Locus of practice
effects in speeded choice tasks. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 17, 20–32.
Robinson, J. A., & Kingsley, M. E. (1977). Memory and intelligence: Age
and ability differences in strategies and organization of recall. Intel-
ligence, 1, 318–330.
Ruff, H. A., Calhoun, G. L., Draper, M. H., Fontejon, J. V., & Guilfoos,
B. J. (2004). Exploring automation issues in supervisory control of
multiple UAVs. In D. A. Vincenzi, M. Mouloua, & P. A. Hancock (Eds.),
Human performance, situation awareness, and automation: Current
research and trends (Vol. 2, pp. 218–222). Mahwah, NJ: Erlbaum.
Ruff, H. A., Narayanan, S., & Draper, M. H. (2002). Human interaction
with levels of automation and decision-aid fidelity in the supervisory
Cognitive Ability and Automation 133
control of multiple simulated unmanned air vehicles. Presence, 11,
335–351.
Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic
human information processing: I. Detection, search and attention.
Psychological Review, 84, 1–66.
Sheridan, T. B. (1995). Human centered automation: Oxymoron or
common sense? IEEE, 823–828.
Sheridan, T. B. (1997). Supervisory control. In G. Salvendy (Ed.), Hand-
book of human factors (2nd ed, pp. 1275–1327). New York: Wiley.
Sheridan, T. B. (2002). Humans and automation. New York: Wiley.
Sheridan, T. B., & Verplank, W. L. (1978). Human and computer control
of undersea teleoperators (Technical report). Cambridge, MA: Man-
Machine Systems Laboratory, Department of Mechanical Engineer-
ing, Massachusetts Institute of Technology.
Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic
human information processing: II. Perceptual learning, automatic
attending and a general theory. Psychological Review, 84, 127–190.
Shinar, D., Meir, M., & Ben-Shoham, I. (1998). How automatic is man-
ual gear shifting? Human Factors, 40, 647–654.
Snow, R. (1989). Cognitive-conative aptitude interactions in learning.
In R. Kanfer, P. L. Ackerman & R. Cudeck (Eds.), Abilities, motiva-
tion, and methodology: The Minnesota Symposium on Learning and
Individual Differences (pp. 435–474). Hillsdale, NJ: Erlbaum.
U.S. Roadmap. (2007). Unmanned systems roadmap 2007–2032. U.S.
Department of Defense. Retrieved from http://purl.access.gpo.gov/
GPO/LPS91893
Wiener, E. L. (1988). Cockpit automation. In E. L. Wiener & D. C. Nagel
(Eds.), Human factors in aviation (pp. 433–461). San Diego, CA:
Academic.
Wiener, E. L., & Curry, R. E. (1980). Flight-deck automation: Promises
and problems. Ergonomics, 23, 995–1011.
7 Conducting Technology-
Based Applied Training
Research
Stephen L. Goldberg and Paula J. Durlach
Orlando Research Unit, U.S. Army Research
Institute for the Behavioral and Social Sciences
The opinions expressed in this chapter are those of the authors and should not
be construed to represent the official position of the Department of the Army.
Conducting Technology-Based Applied Training Reserach 135
Conducting Applied Military Research
In the U.S. military, training research grew out intelligence
testing and selection and classification research during World
War I, and human factors research during World War II that
was conducted in academic and government laboratories. Since
1972 the organization that has been charged with conducting
training research in the Army has been the U.S. Army Research
Institute for the Behavioral and Social Sciences (ARI). ARI’s mis-
sion has been to apply behavioral science research methods to
improve Army training (Uhlaner, 1977). Technology-based train-
ing has been the research area addressed by ARI’s research unit
in Orlando, Florida. The unit was established in the mid-1980s
to provide behavioral science support to the Army’s agency that
procures its training systems and simulators. Technology-based
training research has been multidisciplinary with psycholo-
gists working closely with engineers and computer scientists to
develop training system research findings that will reduce risk
in technologies’ technical design and their features that support
learning and performance. The unit led research in the design
of after action review (AAR) systems which collect simulation
data to portray, summarize, and provide feedback to soldiers
regarding what happened during a collective training event. The
unit has also been conducting research on how virtual reality
technologies can be applied to training dismounted combatants
in virtual environments or computer games. Recently, the unit
has begun research on the design and authoring of adaptive
computer-based training to include intelligent tutor technology.
Applied research projects collect data from either military or
nonmilitary volunteers. Research employing soldiers as partici-
pants usually is conducted on military installations. Less fre-
quently soldiers are brought to research facilities like those at
the ARI laboratory in Orlando where nonmilitary participants
also have been subjects in experiments.
Conducting research with soldiers can be complicated to
arrange. In order to obtain troop support for research, coordina-
tion must occur through either the Army’s training command or
its command that is responsible for preparing operational units
for deployment. These two commands are responsible for the
majority of the soldiers in the United States. Both commands
schedule “umbrella weeks” during which an Army installation
stands down from its normal training activities to support vari-
ous research organizations’ requests for soldiers to participate
in either surveys or experiments. Typically, the major installa-
tions offer 1 week per year, at different times. The number and
types of soldiers and the amount of time they are available are
136 Stephen Goldberg and Paula Durlach
negotiated prior to the umbrella week with the major command
headquarters and the installation. The wars in Iraq and Afghan-
istan have reduced most installations’ capability to participate
in umbrella weeks because soldiers’ time at home is limited, and
there is much to do to reorganize and train for the next rotation
to a war zone. Training research may require more time than
umbrella weeks allow. In that case it has been possible to nego-
tiate for support during non-umbrella week periods.
Regardless of how soldiers are recruited for a research project,
problems sometimes arise because units and individual soldiers
have competing demands for their time. Supporting research
could become a lower priority than an operational requirement
or medical appointment. As a result no matter what support
was negotiated there could be a difference between what the
researcher requested and who participates in the research.
Applied research involves compromise and adaptability. Given
that it could be impossible to run additional participants,
research results and conclusions may have to be tempered by
the vagaries of conducting research in the field.
Adaptive Training
The ultimate challenge in developing technology-based training
for individuals is to approximate as closely as possible one-on-
one human tutoring. This approximation requires the ability to
represent and employ the rich knowledge and behavioral flexibil-
ity of human tutors, the aim of intelligent tutoring systems (ITS).
Development of ITS has progressed since the 1960s (VanLehn,
2006), but, although a handful of companies profess to produc-
ing ITS, expertise in this area is still primarily in the hands
of academia. ITS ideally consist of several component models,
which interact to control the student experience. These com-
ponents correspond to the knowledge used by human tutors:
a student model (knowledge about the student, a pedagogical
model); a set of instructional strategies and behaviors: a domain
model (knowledge about the subject being taught); and an expert
model (knowledge of how to solve problems in the domain). The
term ITS has been somewhat loosely applied, however; applica-
tions labeled as ITS do not necessarily have all (or any!) of these
models. Many of the “true” ITS have been created to supplement
classroom training in academic subjects, such as algebra and
physics (Graesser et al., 2004; Koedinger & Aleven, 2007; Van-
Lehn et al., 2005).
These ITS provide students with an environment in which to
practice solving step-based problems that require a sequence of
steps to arrive at a solution (e.g., finding the value of x in 35 = 5x
+ 5). Guidance and feedback are typically provided for each step,
and analysis of student performance is used to select the next
problem the student should attempt. Relatively less effort has
been devoted to incorporating the techniques of ITS into sce-
nario-based practice environments; however, Sherlock was one
early exception (Lesgold & Nahemow, 2001). Sherlock trained
technicians on device troubleshooting, targeting rare, but criti-
cal breakdowns. Besides simulating the device environment, and
presenting challenging problems to solve, two methods of help
were provided. One was an opportunity to review brief descrip-
tions of fault-causing phenomena that might be relevant to the
situation. The other was the opportunity to review a structured
summary of actions already taken. The structuring was based
upon application of an expert model to the data already gathered
140 Stephen Goldberg and Paula Durlach
by the trainee plus the data in the original problem statement.
The trainee could click on any action and see a list of questions
the ITS were prepared to answer. Sherlock truly accelerated the
acquisition of adaptive expertise, producing the same amount
of improvement in skill after 25 hours of training as occurred,
on average, in 4 years on the job. One of the critical aspects of
Sherlock development, which made it successful, was the inten-
sive upfront analysis of the domain. This analysis allowed the
creation of an expert model which generated expert solutions to
the troubleshooting problems (A.M. Lesgold, personal communi-
cation, March 2010).
a. Communications Checkpoint
b. Coordinating Altitude
c. Minimum Risk Route
d. Restricted Operating Zone
Final Comments
Applied training research investigates the application of psycho-
logical theories, technologies, and innovative training methods to
improve military training. Intelligent tutor systems, dismounted
soldier simulation, and game-based distributed training were
presented to demonstrate the problems and opportunities inher-
ent in applied research. The controls applied to basic research
experiments are often difficult to impose on research conducted
in the field with soldiers. Sample sizes are small for collective
training research because the unit of measure is the small unit
(fire team, squad, or platoon) and not the individual soldier.
152 Stephen Goldberg and Paula Durlach
When methods for training with technologies are being investi-
gated, the technologies in many cases have not been developed
to the point where they are reliable and stable. While innova-
tion can most easily be incorporated in new training systems
at this point, the innovation must change with the changes in
the overall system. Another frequent problem in applied training
research is the lack of adequate measurement methods, par-
ticularly for collective training.
The common problems inherent in conducting applied research
can sometimes limit the conclusions that can be drawn. However,
the value of the findings that have been produced has and will
continue to have an impact on the way the U.S. Army trains. As
mentioned above the Army is moving to a learner-centric train-
ing approach that emphasizes the use of ITS, simulation, and
collaborative learning (TRADOC, 2011). Applied research will
provide the support for adopting new training methods. It will
also help to reduce the risk of government contractors tasked
with developing new training technologies by demonstrating the
efficacy of approaches that do and do not work.
References
Alluisi, E. A. (1991). The development of technology for collective train-
ing: SIMNET, a case history. Human Factors, 33, 343–362.
Branson, R. K. (1975) Interservice procedures for instructional systems
development: Executive summary and model. Tallahassee, FL: Flor-
ida State University Center for Educational Technology for Naval
Education and Training Command, Pensacola, FL.
Campbell, C. H., Knerr, B. W., & Lampton, D. R. (2004). Virtual envi-
ronments for infantry Soldiers (ARI Special Report 59). U.S. Army
Research Institute. Retrieved from http://handle.dtic.mil/100.2/
ADA464022.
Clark, B. R., Lampton, D. R., Martin, G. A., & Bliss, J. P. (2004). User’s
manual for the Dismounted Infantry Virtual After Action Review Sys-
tem (DIVAARS) (ARI Research Product 2004-03). U.S. Army Research
Institute. Retrieved from http://handle.dtic.mil/100.2/ADA425427
Clark, R. E., & Estes, F. (1996). Cognitive task analysis for training.
International Journal of Educational Research, 25(5), 403–417.
Durlach, P. J., & Dargue, B. W. (2010). Adaptive and non-adaptive
training technology for small unmanned aerial system employment.
Paper presented at the U. S. Army Science Conference, Orlando, FL.
Evans, K. L., Knerr, B. W., & Gesselman, A. N. (2009) Training small unit
leaders and teams (ARI Special Report S-68). U.S. Army Research
Institute. Retrieved from http://handle/dtic.mil/100.2/ADA508029.
Gentner, D., & Stevens, A. L. (1983). Mental models. Hillsdale, NJ:
Erlbaum.
Conducting Technology-Based Applied Training Reserach 153
Graesser, A. C., Lu, S., Jackson, G. T., Mitchell, H. H., Ventura, M.,
Olney, A., & Louwerse, M. M. (2004). AutoTutor: A tutor with dia-
logue in natural language. Behavior Research Methods, Instruments,
and Computers, 36, 180–192.
Hatano, G., & Inagaki, K. (1986). Two courses of expertise. In H. Ste-
venson, H. Azuma, & K. Hakuta (Eds.), Child development and edu-
cation in Japan (pp. 262–272). New York: Freeman.
Kirkpatrick, D. L., & Kirkpatrick, J. D. (2006). Evaluating training pro-
grams (3rd ed.). San Francisco, CA: Berrett-Koehler.
Klein, G. A. (1998). Sources of power: How people make decisions. Cam-
bridge, MA: MIT Press.
Knerr, B. W. (2007). Immersive simulation training for the dismounted
Soldier (ARI Study Report 2007-01). U.S. Army Research Institute.
Retrieved from http://handle.dtic.mil/100.2/ADA464022.
Knerr, B. W., & Lampton, D. R. (2005). An assessment of the Virtual-
Integrated MOUT Training System (V-IMTS; ARI Technical Report
1163). U.S. Army Research Institute. Retrieved from http://handle.
dtic.mil/100.2/ADA438315.
Knerr, B. W., Lampton, D. R., Crowell, H. P., Thomas, M. A., Comer, B.
D., Grosse, J. R., … Washburn, D. A. (2002). Virtual environments
for dismounted soldier simulation, training, and mission rehearsal:
Results of the FY 2001 Culminating Event (ARI Technical Report
1129). U.S. Army Research Institute. Retrieved from http://handle.
dtic.mil/100.2/ADA403147 .
Knerr, B. W., Lampton, D. R., Singer, M. J., Witmer, B. G., Goldberg,
S. L., Parsons, K. J., & Parsons, J. (1998). Virtual environments for
dismounted soldier training and performance: Results, recommenda-
tions and issues (ARI Technical Report-1089). U.S. Army Research
Institute. Retrieved from http://handle.dtic.mil/100.2/ADA360109
Knerr, B. W., Lampton, D. R., Thomas, M., Comer, B. D., Grosse, J.
R., Centric, J. H., … Washburn, D. A. (2003). Virtual environments
for dismounted soldier simulation, training, and mission rehearsal:
Results of the Fy2002 Culminating Event. (ARI Technical Report
1138). U.S. Army Research Institute. Retrieved from http://handle.
dtic.mil/10.2/ADA417360.
Koedinger, K. R., & Aleven, V. (2007). Exploring the assistance
dilemma in experiments with cognitive tutors. Educational Psychol-
ogy Review, 19, 239–264
Lampton, D. R., Knerr, B. W., Goldberg, S. L., Bliss, J. P., Moshell, J. M.,
& Blau, B. S. (1995). The Virtual Environment Performance Assess-
ment Battery (VEPAB): Development and evaluation. (ARI Technical
Report 1029). U.S. Army Research Institute. Retrieved from http://
handle.dtic.mil/100.2/ADA297277.
Lampton, D. R., McDonald, D. P., Rodriguez, M. E., Morris, C. S., &
Parsons, J. (2001). Instructional stategies for training teams in vir-
tual environments (ARI Technical Report 1110). U.S. Army Research
Institute. Retrieved from http://handle.dtic.mil/100.2/ADA389674.
Lesgold, A., & Nahemow, M. (2001). Tools to assist learning by doing:
Achieving and assessing efficient technology for learning. In D.
154 Stephen Goldberg and Paula Durlach
Klahr & S. Carver (Eds.), Cognition and instruction: Twenty-five
years of progress (pp. 307–346). Mahwah, NJ: Erlbaum.
Lussier, J. W. & Shadrick, S. B. (2003, October13–15). Adaptive think-
ing training for tactical leaders. Paper presented at NATO Human
Factors and Medicine Symposium on Advanced Technologies for
Military Training, Genoa, Italy.
McGilvray, D. H., Leibrecht, B. C., & Lockaby, K. J. (2008). Measuring
learning and performance in collective training exercises (ARI Con-
tractor Report 2008-02). U.S. Army Research Institute. Retrieved
from http://handle.dtic.mil/100.2/ADA480052.
Schwartz, D. L., Bransford, J. D., & Sears, D. (2005). Efficiency and
innovation in transfer. In J. P. Mestre (Ed.), Transfer of learning from
a modern mutidisciplinary perspective (pp. 1–51). Greenwich, CT:
Information Age.
Singer, M. J., Barnett, J., & Taylor, G. (2011) Evaluation of two distrib-
uted game-based simulations during exercises. Arlington, VA: U.S.
Army Research Institute for the Behavioral and Social Sciences.
Manuscript in preparation.
Singer, M. J., Grant, S. C., Commarford, P. M., Kring, J. P., & Zavod,
M. (2001). Team performance in distributed virtual environments (ARI
Technical Report 1118). U.S. Army Research Institute. Retrieved
from http://handle.dtic.mil/100.2/ADA396489.
Singer, M. J., & Knerr, B. W. (2010). Evaluation of a game-based simula-
tion during distributed exercises (ARI Research Report 1931). Arling-
ton, VA: U.S. Army Research Institute for the Behavioral and Social
Sciences.
Singer, M. J., Kring, J. P., & Hamilton, R. M. (2007). Instructional fea-
tures for training in virtual environments. (ARI Technical Report
1184). Retrieved from http://handle.dtic.mil/100.2/ADA455301.
Sticha, P. J., Weaver, E. A., Ford, L. A., & Campbell, R. C. (2011). Assess-
ing learning and performance in collective training exercises (Draft
HumRRO Report). Alexandria, VA: Human Resources Research
Organization.
Sweller, J., Kirschner, P. A., & Clark, R. E. (2007). Why minimally
guided teaching techniques do not work: A reply to commentaries.
Educational Psychologist, 42, 115–121.
TRADOC. (2011). The U. S. Army Learning Concept for 2015. TRADOC
Pamphlet 525-8-2. Retrieved from http://www-tradoc.army.mil/
tpubs/pams/tp525-8-2.pdf.
Uhlaner, J. E. (1977). The research psychologist in the Army 1917 to
1977 (ARI Research Report 1155). Arlington, VA: U.S. Army Research
Institute for the Behavioral and Social Sciences.
U.S. Department of Defense (2009). Capstone concept for joint opera-
tions (version 3.0). Retrieved from http://www.jfcom.mil/newslink/
storyarchive/2009/CCJO_2009.pdf
VanLehn, K. (2006) The behavior of tutoring systems. International
Journal of Artificial Intelligence in Education, 16, 227–265.
VanLehn, K., Lynch, C., Schulze, K., Shapiro, J. A., Shelby, R., Taylor,
L., … Wintersgill, M. (2005). The Andes physics tutoring systems:
Conducting Technology-Based Applied Training Reserach 155
Lessons learned. International Journal of Artificial Intelligence in
Education, 15, 147–204.
Witmer, B. G., Bailey, J. H., & Knerr, B. W. (1995). Training dismounted
soldiers in virtual environment: Route learning and transfer (ARI
Technical Report 1022). Retrieved from http://www.dtic.mil/dtic/tr/
fulltext/u2/a292900.pdf.
Witmer, B. G., Sadowski, W. J., & Finkelstein, N. M. (2000). Training
dismounted Soldiers in virtual environments: Enhancing configura-
tion learning (ARI Technical Report 1103).Retrieved from http://
handle.dtic.mil/100.2/ADA381715.
8 A New Taxonomy
for Training
William D. Raymond, Alice F. Healy,
and Lyle E. Bourne, Jr.
University of Colorado
Training Method
The training dimension must include variables that capture
the method of instruction and the types of activities performed
during learning. The taxonomy adopted here builds on earlier
research on educational methods in the classroom. Berliner
(1983) recognized the need for more rigorous definitions of edu-
cational treatments, and he provided a taxonomy for classroom
160 William D. Raymond et al.
Table 8.2 Mapping between the New Task Taxa and IMPRINT Task
Taxa
New task taxa IMPRINT task taxa
Modeling (mimicking =
observe and mimic a
model performing the
task)
Discussion/Question & answer default = 1-way; 2-way
Immersion default = no; yes
(embedded in field
context)
Learning location default = local; remote or
“distance learning”
Individualization default = no; yes - e.g.,
human or intelligent
computer tutoring
Group training default = no; group size
Automation default = no; yes
whole; supplemental
Deep processing default = no; yes
Mediation (e.g., use default = no; yes
of prior knowledge)
Attentional focus default = no focus;
internal, external
Attentional breadth default = intermediate;
global, local
Stimulus-response default = yes; no
compatibility
Mapping type default = consistent;
varied
Contralateral default = no; yes
training
Time pressure default = no; yes
Stressor default = no; yes
Performance Context
The performance context component relates the environment
of posttraining performance to the training environment. The
major component of performance context captures the relation-
ship of performance to the items, context, and task encountered
in training. In addition, performance context is concerned with
A New Taxonomy for Training 165
Table 8.5 Decomposition of the Performance Context Dimension of the
New Taxonomy
Transfer parameters New items, item default = same as
order, or item training; different
distribution items, order, or
distribution
New context default = same as
training; different
context
New task default = same as
training; different
task
Performance Assessment
Complex training goals can be evaluated using systems designed
to facilitate assessment of the acquisition of knowledge, such
as in the taxonomy of cognitive learning developed by Bloom,
Englehart, Furst, Hill, and Krathwohl (1956). In their taxon-
omy, cognitive learning goals can be arranged in a hierarchy
of knowledge complexity. Mastering any level of the hierarchy
requires mastery of the behaviors in the taxa below it. The lev-
els proposed by Bloom et al. are shown in Table 8.6, along with
methods of assessment for each level.
Table 8.6 The Bloom et al. (1956) Taxonomic Hierarchy for the
Cognitive Learning Domain
Learning Goal Assessment
Knowledge Recall or recognize information
Comprehension Comprehend or interpret information
Application Use information to complete a task
Analysis Distinguish, classify, and relate knowledge
Synthesis Originate and combine ideas
Evaluation Appraise and assess ideas based on standards
166 William D. Raymond et al.
The Bloom et al. (1956) taxonomy focuses on the acquisition
of verbal, or declarative, knowledge and associated behaviors.
Skill performance can generally be objectively assessed in terms
of speed or accuracy of task completion. Separate measures are
needed, because it has been shown that there are trade-offs
between speed and accuracy in some tasks. In a digit data entry
task, speed and accuracy show different patterns of results;
speed improves with training while accuracy declines (Healy,
Kole, Buck-Gengler, & Bourne, 2004; see chapters 1 and 2).
However, in other scenarios, the opposite pattern might obtain.
Moreover, situations in which training produces improved effi-
ciency of performance (i.e., faster and more accurate responding)
need to be differentiated from those in which it alters only the
speed–accuracy criterion. It is also important to assess perfor-
mance on subcomponents of a task. For example, the response
times for executing the different steps of a digit data entry task
are not always positively correlated, with the typing of digits
slowing down on some digits in order to be faster on subsequent
ones (Healy et al., 2004).
In some tasks, there is also a necessity to develop some
index of changes in the learner’s cognition during training. For
example, in a binary classification task, Bourne, Raymond,
and Healy (2010) have shown that even when both speed and
accuracy measures show continuous improvement, subjects
use different strategies to guide their responses, often changing
strategies during training, with strategy differences impacting
performance. Measures must be developed to assess changes
in cognitive strategies, because the strategy chosen may impact
speed and accuracy, or even retention and transfer.
Researchers have also expanded the scope of learning out-
comes to include affective or attitudinal learning goals as well as
knowledge and skill acquisition. Drawing on all three areas of
research, Kraiger, Ford, and Salas (1993) proposed a more com-
prehensive taxonomy of learning outcomes, shown in Table 8.7.
They define learning as changes in cognitive, skill-based, and
attitudinal states and discuss how learning in each category
can be measured (see Table 8.7). The Kraiger et al. (1993) classi-
fication forms the basis for the present performance assessment
taxonomy. However, speed and accuracy measures of individual
components can be combined with the different levels to form
a more detailed taxonomy of assessment tests. Having quanti-
fied the outcome of a particular training scenario, the effective-
ness of training can be measured by comparing posttraining
performance with performance before or at the beginning of
A New Taxonomy for Training 167
Table 8.7 The Kraiger, Ford, and Salas (1993) Classification of
Learning Outcomes and Associated Measures of Assessment
Learning outcome Assessment
Cognitive Verbal Knowledge Tests of memory
Outcomes Knowledge Organization Probe cognitive
structures
Cognitive Strategies Probe task protocol
Skill-based Compilation Proceduralization Change in
Outcomes Composition performance
Training Principles
Existing taxonomies put an emphasis on domains and teaching
methods and neglect empirically based principles that impact
performance within those domains and methods. A major con-
tribution of the present new taxonomy is its inclusion of an inde-
pendent dimension identifying those principles. This dimension
is orthogonal to the three others that comprise the new tax-
onomy. Table 8.8 provides a hierarchical listing of the principles,
following the outline of chapter 2. The listing divides the princi-
ples into three major categories relating to (a) resource and effort
allocation, (b) context effects, and (c) task parameters. These
principles have a solid basis in empirical research but might be
differentially effective across task types or domains, training
methods, and performance measures. The full four-dimensional
taxonomy, thus, serves to provide a framework for predicting
training outcomes under various combinations of tasks, meth-
ods, and measures.
168 William D. Raymond et al.
Table 8.8 Principles of Training (from Chapter 2)
Principles Relating to Deliberate Practice
Resource and Effort Depth of Processing
Allocation
Generation Effect
Focus of Attention
Strategic Use of Knowledge
Cognitive Antidote to Fatigue, Boredom, or
Task Disengagement
References
Ackerman, P. L. (2007). New developments in understanding skilled
performance. Current Directions in Psychological Science 16,
235–239.
Alliger, G. A., Tannenbaum, S. I., Bennett, W., Traver, H., & Shotland,
A. (1997). A meta-analysis of the relations among training criteria.
Personnel Psychology, 50, 341–359.
Archer, R., Walters, B., Yow, A., Carolan, T., Laughery, K. R., & Gillis, P.
(1999). Training as a performance shaping factor in computer gen-
erated forces. Proceedings of the 1999 Computer Generated Forces
Conferences, Orlando, FL.
Beier, M. E., & Ackerman, P.L. (2005). Age, ability and the role of prior
knowledge on the acquisition of new domain knowledge. Psychology
and Aging, 20, 341–355.
Berliner, D. C. (1983). Developing conceptions of classroom environ-
ments: Some light on the T in classroom studies of ATI. Educational
Psychology, 18, 1–13.
Bjork, R, A. (1994). Memory and metamemory considerations in the
training of human beings. In J. Metcalfe & A. Shimamura (Eds.),
Metacognition: Knowing about knowing (pp. 185–205). Cambridge,
MA: MIT Press.
Bloom, B., Englehart, M., Furst, E., Hill, W., & Krathwohl, D. (1956).
Taxonomy of educational objectives: The classification of educa-
tional goals. Handbook: Vol. 1. Cognitive domain. New York: Long-
mans, Green.
Bourne, L. E., Jr., Raymond, W. D, & Healy, A. F. (2010). Strategy selec-
tion and use during classification skill acquisition. Journal of Exper-
imental Psychology: Language, Memory, and Cognition, 36, 500–514.
Carpenter, S. K., Pashler, H., Wixted, J. T., & Vul, E. (2008). The
effects of tests on learning and forgetting. Memory & Cognition, 36,
438–448.
Clegg, B., & Heggestad, E. (2010, April). Experiments on levels of auto-
mation, individual differences, and team performance. Paper pre-
sented at the 2010 Ellis-Battig Memory Symposium: Optimizing
the Training of Knowledge and Skills: A Review of Accomplishments
from the Multidisciplinary University Research Initiative (MURI) on
Training, 80th Annual Convention of the Rocky Mountain Psycho-
logical Association, Denver, CO.
Companion, M. A., & Corso, G. M. (1982). Task taxonomies: A general
review and evaluation. International Journal of Man-Machine Stud-
ies, 17, 459–472.
174 William D. Raymond et al.
Fleishman, E. A. (1978). Relating individual differences to the dimen-
sions of human tasks. Ergonomics, 21, 1007–1019.
Gawron, V. J., Drury, C. G., Czaja, S. J., & Wilkins, D. M. (1989). A
taxonomy of independent variables affecting human performance.
International Journal of Man–Machine Studies, 31, 643–672.
Gonzalez, C., & Thomas, R. P. (2008). Effects of automatic detection
on dynamic decision making. Journal of Cognitive Engineering and
Decision Making, 2, 328–348.
Healy, A. F., Kole, J. A., Buck-Gengler, C. G., & Bourne, L. E., Jr. (2004).
Effects of prolonged work on data entry speed & accuracy. Journal of
Experimental Psychology: Applied, 10, 188–199.
Healy, A. F., Kole, J. A., Wohldmann, E. L., Buck-Gengler, C. J., &
Bourne, L. E., Jr. (2011). Data entry: A window to principles of train-
ing. In A. S. Benjamin (Ed.), Successful remembering and successful
forgetting: A festschrift in honor of Robert A. Bjork (pp. 277–296). New
York: Psychology Press.
Jonassen, D., & Tessmer, M. (1996/97). An outcomes-based taxonomy
for instructional systems design, evaluation, and research. Training
Research Journal, 2, 11–46.
Kirkpatrick, D. L. (1987). Evaluation of training. In R. L. Craig (Ed.),
Training and development handbook: A guide to human resource
development (3rd ed., pp. 301–319). New York: McGraw-Hill.
Kole, J. A., Healy, A. F., & Bourne, L. E., Jr. (2008). Cognitive com-
plications moderate the speed-accuracy tradeoff in data entry: A
cognitive antidote to inhibition. Applied Cognitive Psychology, 22,
917–937.
Kraiger, K., Ford, J. K., & Salas, E. (1993). Application of cognitive, skill-
based, and affective theories of learning outcomes to new methods
of training evaluation. Journal of Applied Psychology, 78, 311–328.
Krathwohl, D. R., Bloom, B. S., & Masia, B. B. (1964). Taxonomy of
educational objectives: The classification of educational goals. White
Plains, NY: Longman.
Meister, D. (1976). Behavioral foundations of system development. New
York: Wiley.
Miller, R. B. (1953). A method for man-machine task development (Tech-
nical Report WADC-TR-53-137). Wright-Patterson A.F.B., Ohio:
Wright Air Development Center.
O’Neil, H. F. (Ed.). (2003). What works in distance learning: Guidelines.
Charlotte, NC: Information Age.
Proctor, R. W., & Vu, K.-P. L. (2006). Laboratory studies of training,
skill acquisition, and retention of performance. In K. A. Ericsson, N.
Charness, P. J. Feltovich, & R. R. Hoffman (Eds.), Cambridge hand-
book of expertise and expert performance (pp. 265–286). Cambridge,
England: Cambridge University Press.
Roth, J. T. (1992). Reliability and validity assessment of a taxonomy for
predicting relative stressor effects on human task performance (Tech-
nical Report 5060-1 prepared under contract DNA001-90-C-0139).
Boulder, CO: Micro Analysis and Design.
A New Taxonomy for Training 175
Shute, V, J., Lajoie, S. P., & Gluck, K. A. (2000). Individualized and
group approaches to training. In S. Tobias & J. D. Fletcher (Eds.),
Training and retraining: A handbook for business, industry, govern-
ment, and the military (pp. 171–207). New York: Macmillan.
Szpunar, K. K., McDermott, K. B., & Roediger, H. L., III. (2008). Testing
during study insulates against the buildup of proactive interference.
Journal of Experimental Psychology: Learning, Memory, and Cogni-
tion, 34, 1392–1399.
Wickens, C. D., & Holland, J. G. (2000). Memory, learning, and train-
ing. In C. D. Wickens & J. G. Holland (Eds.), Engineering psychol-
ogy and human performance (3rd ed., pp. 241–284). New York:
HarperCollins.
Wohldmann, E. L., Healy, A. F., & Bourne, L. E., Jr. (2008). A mental
practice superiority effect: Less retroactive interference and more
transfer than physical practice. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 34, 823–833.
Young, M. D., Healy, A. F., Gonzalez, C., Dutt, V., & Bourne, L. E., Jr.
(2011). Effects of training with added difficulties on RADAR detec-
tion. Applied Cognitive Psychology, 25, 395–407.
Appendix A: The IMPRINT Planning Matrix
Pedagogy
Instruction
T
Visual navigation
(noise?)
Numerical Analysis radar (visual radar (fire
detection) decision)
Information processing data entry, time time estimation
navigation estimation, (letter counting),
(correct/incorrect reconstruction data entry
; immediate v. of order, radar (articulatory
delayed) (tone counting) suppression, +/-
termination),
radar (fire
decision)
Fine motor - discrete data entry sequence data entry
learning (articulatory
(tones) suppression)
Fine motor - continuous target finding
(no reversals;
periodic v. trial-
by-trial)
Gross motor - light
Gross motor - heavy
Communication
(reading & writing)
Communication (oral) navigation navigation
(abbreviated
responses)
MURI post-training performance context factors
Transfer parameters
New items, item
order, item Refresher training
New context New task (relative to Retention interval schedule (default =
distribution (default (default none;
=
(same as, training) (same as, none; refresher
IMPRINT task taxons = same as training; time since training)
different from different from schedule)
different items,
training) training)
order, or
distribution)
Ai , t
e
Pi ,t =
∑
Aj , t
e (2)
j
Ai = Bi + ∑W j
j S ji − Di + ε (3)
2.6
2.5
Model
2.4 Prediction
2.3
Human
2.2
2.1
2
1 2 3 4 5 6 7 8 9 10
Block
Exp 1 Accuracy
1
0.9
Proportion Correct
0.8
Model
Prediction
0.7
Human
0.6
0.5
1 2 3 4 5 6 7 8 9 10
Block
Figure 9.1 Experiment 1 data from Healy et al. (2004) and ACT-R
model fits to the data from Gonzalez et al. (2011).
188 Cleotilde Gonzalez
Models of Training Difficulty Principle
in the RADAR Task
The training difficulty principle (see chapters 2 and 3) predicts
that conditions that cause difficulty during learning would facil-
itate later retention and transfer. This principle was tested in a
RADAR target detection and decision-making task (Gonzalez &
Thomas, 2008), using laboratory experiments where, in some
cases, the potential targets were nine military vehicles (e.g.,
submarine, helicopter, and jeep) (Young, Healy, Gonzalez, Dutt,
& Bourne, 2011).
The goal in RADAR is to detect and eliminate hostile enemy
targets by visually discriminating moving targets among mov-
ing distractors. RADAR is similar to military target visual detec-
tion devices, in which a moving target needs to be identified as
a potential threat or not and a decision is made on how to best
destroy that target. The task requires the participant to make
both visual and memory searches. The participant must memo-
rize a set of targets and then seek out one or more targets on
a radar grid. A target threat may or may not be present among
a set of moving blips. The blips—in the form of potential tar-
gets or blank masks—begin at the four corners of the grid and
approach the center at a uniform rate. Detection of an enemy
target must occur before the blips collapse in the center.
Models of the training difficulty principle in the RADAR
task were developed under two perspectives, the IBL and SBL
approaches, and compared (Gonzalez, Dutt, Healy, Young, &
Bourne, 2009). The goal of the model comparison effort was to
understand the processes by which behavior is represented, the
constraints that the different approaches impose upon the task
models, and the comparison of the two approaches’ theoretical
assumptions (Lebiere, Gonzalez, & Warwick, 2009).
The IBL model was based upon the IBL theory as presented
above. The SBL model used four concrete strategies that varied
in their effectiveness at performing the target detection task.
One strategy was an optimal strategy, and three strategies were
suboptimal. These strategies represented practically feasible
ways to go about the task. The utility learning mechanism in
ACT-R (Anderson et al., 2004) was used, by which the different
strategies compete using a reinforcement learning algorithm.
This algorithm produces a gradual transition from the subop-
timal to the optimal strategies. When the model executes, there
is a competition set up between the three suboptimal strategies
and the optimal one. Although the suboptimal strategies are
executed more often initially, the optimal strategy later picks up
Cognitive Models and the Instance-Based Learning Tool 189
in usage because of its increased utility through repeated posi-
tive rewards.
The SBL and IBL models were compared along two different
dimensions: (a) fit: how well each model fits human learning
data in the task; and (b) adaptability: how well each model that
has been able to reproduce the way human beings learned in
one task scenario behaves in new scenarios that are similar to
or different from the training condition. The fit criterion is com-
mon in model comparisons, whereas the adaptability criterion
is relatively new (Gluck, Bello, & Busemeyer, 2008). The adapt-
ability criterion used here is similar to the generalization crite-
rion method (Busemeyer & Wang, 2000), which divides observed
data into two sets: a calibration or training set to estimate model
parameters, and a validation or test set to determine predic-
tive performance. However, the models’ adaptability was further
tested by examining their ability to adapt to new test condi-
tions that are either similar to or different from the training
conditions.
Figure 9.2 presents the average times for correct responses
during the training phase, in four conditions that varied in the
difficulty of target detection (Young et al., 2011). The 1+1 condi-
tion indicated the need to memorize one target and the presence
of only one item on the RADAR screen. Thus, this was the easi-
est condition. The 4+4 condition indicated the need to memo-
rize four targets and the presence of four items on the RADAR
screen, making it the most difficult condition. The mappings of
1800
Average Correct Response Time (ms)
SBL model
1600
800
600
400
200
0
1+1 4+4 1+1 4+4
Consistent Varied
Condition
Figure 9.2 Average correct response times (ms) for CM 1+1, VM 1+1,
CM 4+4, and VM 4+4 blocks in human data and SBL and
IBL models during training. The error bars show 90% con-
fidence intervals.
190 Cleotilde Gonzalez
the targets were either consistent (the target was always a target
within a block of trials) or varied (the target was sometimes used
as a distractor on a different trial within the same block).
As shown in Figure 9.2, both the IBL and SBL models fit the
human data quite well, RMSD = 69 ms for IBL and RMSD = 163
ms for SBL. However, the SBL model seems to generate generally
higher time values compared to human data, and it has a higher
RMSD. This difference may be because the four strategies in
the SBL model execute productions in a fi xed time (50 ms per
production). There is also no speedup in the correct response
times due to this fi xed strategy execution time, whereas the IBL
model speeds up on account of activation-retrieval time accel-
eration. The retrieval time decreases if the instances’ activation
increases over blocks (Anderson & Lebiere, 1998). It is also clear
from Figure 9.2 that both models take more time in the more
difficult (4+4) blocks than in the easier blocks (1+1) for both con-
sistent and varied target mappings. This finding demonstrates
the effects of workload well known from behavioral studies of
automaticity (Gonzalez & Thomas, 2008), which result from the
extra time taken to process additional items.
Figure 9.3 demonstrates the effects of added difficulty in the
task (Young et al., 2011). In the “Tone” condition, participants
were required to count deviant tones (low and high frequency)
among standard tones (medium frequency) playing in the back-
ground during the target detection task. As shown by both mod-
els, the tone condition takes slightly more time to process than
silent trials because of an extra auditory production in both
1000
800
600
400
200
0
Silent Tone
Condition
Figure 9.3 Average correct response times (ms) for silent and tone
conditions for human data and SBL and IBL models during
training. The error bars show 90% confidence intervals.
Cognitive Models and the Instance-Based Learning Tool 191
models that processes the tones. Again, the difference between
the time in the SBL model and human data is greater than the
difference between the time in the IBL model and human data.
The SBL model has no activation-retrieval speedup to compen-
sate for time spent in tone counting, whereas there is such a
speedup in the IBL model that reduces the overall time.
To test the adaptability of both models, transfer was com-
pared from difficult to easier conditions (tone-to-silent) and easy
to more difficult conditions (silent-to-tone). Figure 9.4 shows
these results. The SBL model has an RMSD = 160 ms when
it is trained in tone and transferred to silent, whereas the IBL
SBL Data
Average Correct Response Time (ms)
1600
Human data
1400
IBL model
1200
1000
800
600
400
Training Transfer
Tone Silent
Human data
1400
IBL model
1200
1000
800
600
400
Training Test
Silent Tone
Figure 9.4 Left panel: Average correct response times (ms) for human
data and SBL and IBL models for training in the tone and
testing in the silent condition. Right panel: Average cor-
rect response times (ms) for human data and SBL and IBL
models for training in the silent and testing in the tone
condition. The error bars show 90% confidence intervals.
192 Cleotilde Gonzalez
model’s RMSD = 50 ms. The SBL model’s RMSD when trained in
silent and transferred to tone is 248 ms, whereas the RMSD for
the IBL model is 62 ms. Thus, one can conclude that both mod-
els are quite good according to the adaptability criterion, but
the IBL model produces values closer to the human data than
the SBL model. Although the numerical values are important,
the IBL model has other advantages over the SBL model not
shown in measurements: the changes in environmental condi-
tions are captured in the instances stored and retrieved from
memory, whereas the SBL approach is blind to those changes.
The SBL model continues applying the same strategies at test,
which might not be as effective as they were during training
once the task conditions change. Also, the strategies in dynamic
situations are often unknown a priori or difficult to define at all.
Human beings are often unable to explain any rules or strate-
gies used to solve a dynamic problem. Thus, the IBL approach is
more appropriate for modeling dynamic decision making (Gon-
zalez et al., 2003) than the SBL approach.
700
650
RT(ms)
Model
600 Human
550
500
SRC Incompatible SRC Compatible Simon Non
- Simon
Corresponding Corresponding
Figure 9.5 The IBL model’s fits to human data in different mappings
of the SRC and Simon tasks. The error bars show 95%
confidence intervals around the point estimate.
Recognition
PKase
● Retrieval Threshold
● Utility Threshold
Stopping Rule Similarity
HV
●
Goal
Feedback
Update Utility
PKase
Figure 9.6 The IBL tool with five distinct IBL theory phases (Right)
and the task Environment (Left).
196 Cleotilde Gonzalez
The IBL tool takes a modeler step-by-step through the dis-
tinct process of IBL theory, making it easy to understand and
intuitive for new modelers. Most importantly, these steps are a
generic decision-making process that does not depend on the
modeler’s creativity to define (often complicated) decision-mak-
ing strategies. Research has demonstrated that this process is
generic enough to model most forms of decisions from experi-
ence (Gonzalez & Lebiere, 2005; Gonzalez et al., 2003; Lejar-
raga et al., in press). But the most relevant contribution of the
IBL tool is to make the theory more accessible. The IBL tool
makes it shareable, by bringing the theory closer to the end
users; generalizable, by making it possible to use in different
and diverse tasks; understandable, by making it easy to use
in cognitive models; robust, by abstracting the specifics of its
implementation independent of any specific programming lan-
guage; communicable, by making the tool interact more easily
and in a more standard way with tasks; and usable, by making
the theory more transparent.
A step-by-step demonstration of building a cognitive model
in the IBL tool for a particular task (the Iowa Gambling Task)
is explained in Dutt and Gonzalez (2011). Once a modeler has
defined the model’s parameters and instance structure, the tool
can simulate a number of model participants by connecting it to
the task using well-known computer communication standards.
These simulations provide the model’s predictions regarding
human behavior and performance in the task of interest.
1
0.9
Proportion of Choices of
Advantageous Buttons
0.8
0.7
0.6
0.5
Human
0.4
0.3 Model
0.2
0.1
0
1 2 3 4 5 6
Block
Figure 9.7 The fit of the IBL model developed in the IBL tool to human
data for controls in the Iowa Gambling Task. See Dutt and
Gonzalez (2011).
Cognitive Models and the Instance-Based Learning Tool 197
Figure 9.7 shows the results obtained from running the IBL
model of the Iowa Gambling Task in the IBL tool compared to
that of a group of healthy human (control) participants run
in the same task, as reported by Bishara et al. (2009). These
results were obtained using the default values of the parameters
in the tool. Thus, the intention was not to calibrate the model
parameters to produce the best predictions for human data, but
rather to use the Iowa Gambling Task as an example to explain
the IBL tool. Figure 9.7 shows the proportion of choices for the
two advantageous alternatives (out of a total of four options in
the Iowa Gambling Task) over six blocks of 20 trials each. The
proportion of choices has been averaged across all 32 human
and model participants. Although the exact level of the values
are discrepant, the model in the IBL tool set at default param-
eter values provides a reasonable prediction of the trend in the
observed human behavior over the six blocks of the experiment
(MSD = 0.010; r = 0.86). In addition to the performance data,
running a model in the IBL tool produces data on the values and
dynamics of its mechanisms (e.g., activation, base-level learn-
ing, noise, and the values of the instance’s situation-decision-
value slots).
Conclusion
Three projects involving computational representations of
human behavior for three training principles are summarized:
speed–accuracy trade-off attributable to fatigue, training dif-
ficulty, and stimulus-response compatibility. Taken together,
these studies show that the ACT-R architecture and IBL theory
presents an accurate and robust representation of the learn-
ing process in several training paradigms. Because IBL theory
has also demonstrated accurate representations in many other
tasks (see Gonzalez & Dutt, 2011, for a discussion), the theory
is more general than it was initially conceived to be: IBL theory
accounts for decisions from experience at many different levels.
This ability is illustrated by the precision of the models’ predic-
tions in the projects described here. Moreover, the creation of
an explicit computer tool that represents the theory can also
give rise to interesting demonstrations and new questions and
answers. The theory was embodied in the IBL tool, which is
available for research purposes. This tool should allow for more
widespread from the authors use of IBL theory as it helps facili-
tate a cognitive modeler’s work.
198 Cleotilde Gonzalez
References
Anderson, J. R., Bothell, D., Byrne, M. D., Douglass, S., Lebiere, C.,
& Qin, Y. (2004). An integrated theory of the mind. Psychological
Review, 111, 1036–1060.
Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought.
Hillsdale, NJ: Erlbaum.
Anderson, J. R., & Lebiere, C. (2003). The Newell test for a theory of
mind. Behavioral and Brain Sciences, 26, 587–639.
Ballard, J. C. (1996). Computerized assessment of sustained attention:
A review of factors affecting vigilance performance. Journal of Clini-
cal and Experimental Psychology, 18, 843–863.
Bishara, A. J., Pleskac, T. J., Fridberg, D. J., Yechiam, E., Lucas, J.,
Busemeyer, J. R., et al. (2009). Similar processes despite divergent
behavior in two commonly used measures of risky decision making.
Journal of Behavioral Decision Making, 22, 435–454.
Busemeyer, J. R., & Wang, Y. M. (2000). Model comparison and model
selections based on generalization criterion methodology. Journal of
Mathematical Psychology, 44, 171–189.
Dutt, V., & Gonzalez, C. (2011). Making instance-based learning the-
ory usable and understandable: The Instance-Based Learning Tool.
Manuscript under review.
Dutt, V., Yamaguchi, M., Gonzalez, C., & Proctor, R. W. (2011). An
instance-based learning model of stimulus-response compatibility
effects in mixed location-relevant and location-irrelevant tasks. Man-
uscript under review.
Fu, W., Gonzalez, C., Healy, A. F., Kole, J. A., & Bourne, L. E., Jr. (2006).
Building predictive models of skill acquisition in a data entry task.
In Proceedings of the Human Factors and Ergonomics Society Annual
Meeting (HFES 50th Annual Meeting) (pp. 1122–1126). Santa Monica,
CA: Human Factors and Ergonomics Society.
Gluck, K., Bello, P., & Busemeyer, J. (2008). Introduction to the special
issue. Cognitive Science, 32, 1245–1247.
Gonzalez, C., Best, B. J., Healy, A. F., Kole, J. A., & Bourne, L. E., Jr.
(2011). A cognitive modeling account of simultaneous learning and
fatigue effects. Cognitive Systems Research, 12, 19–32.
Gonzalez, C., & Dutt, V. (2011). Instance-based learning: Integrating
decisions from experience in sampling and repeated choice para-
digms. Psychological Review, 118, 523–551.
Gonzalez, C., Dutt, V., Healy, A. F., Young, M. D., & Bourne, L. E.
(2009). Comparison of instance and strategy models in ACT-R. In A.
Howes, D. Peebles, & R. Cooper (Eds.), Proceedings of the 9th Inter-
national Conference on Cognitive Modeling—ICCM2009. Manchester,
UK.
Gonzalez, C., Fu, W., Healy, A. F., Kole, J. A., & Bourne, L. E., Jr.
(2006). ACT-R models of training data entry skills. In Proceedings of
the Conference on Behavior Representation in Modeling and Simula-
tion (BRIMS 2006) (pp. 101–109). Baltimore, MD.
Cognitive Models and the Instance-Based Learning Tool 199
Gonzalez, C., & Lebiere, C. (2005). Instance-based cognitive models of
decision making. In D. Zizzo & A. Courakis (Eds.), Transfer of knowl-
edge in economic decision-making (pp. 148–165). New York: Palgrave
Macmillan.
Gonzalez, C., Lerch, J. F., & Lebiere, C. (2003). Instance-based learn-
ing in dynamic decision making. Cognitive Science, 27, 591–635.
Gonzalez, C., & Thomas, R. P. (2008). Effects of automatic detection
on dynamic decision making. Journal of Cognitive Engineering and
Decision Making, 2, 328–348.
Healy, A. F., Kole, J. A., Buck-Gengler, C. J., & Bourne, L. E., Jr. (2004).
Effects of prolonged work on data entry speed and accuracy. Journal
of Experimental Psychology: Applied, 10, 188–199.
Hoffman, R. R., Sherrick, M. F., & Warm, J. S. (1998). Viewing psychol-
ogy as a whole: The integrative science of William N. Dember. Wash-
ington DC: American Psychological Association.
Jongman, G. M. G. (1998). How to fatigue ACT-R? In Proceedings of the
2nd European Conference on Cognitive Modeling (pp. 52–57). Not-
tingham, England: Nottingham University Press.
Kole, J. A., Healy, A. F., & Bourne, L. E. (2008). Cognitive complications
moderate the speed-accuracy tradeoff in data entry: A cognitive
antidote to inhibition. Applied Cognitive Psychology, 22, 917–937.
Lebiere, C., Gonzalez, C., & Martin, M. (2007). Instance-based decision
making model of repeated binary choice. In R. L. Lewis, T. A. Polk &
J. E. Laird (Eds.), Proceedings of the 8th International Conference on
Cognitive Modeling (pp. 67–72). Ann Arbor, MI.
Lebiere, C., Gonzalez, C., & Warwick, W. (2009). Convergence and con-
straints revealed in a qualitative model comparison. Journal of Cog-
nitive Engineering and Decision Making, 3, 131–155.
Lejarraga, T., Dutt, V., & Gonzalez, C. (in press). Instance-based learn-
ing: A general model of repeated binary choice. Journal of Behavioral
Decision Making.
Logan, G. D. (1988). Toward an instance theory of automatization. Psy-
chological Review, 95, 492–527.
Lovett, M. C., Reder, L. M., & Lebiere, C. (1999). Modeling working
memory in a unified architecture: An ACT-R perspective. In A.
Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of
active maintenance and executive control (pp. 135–182). New York:
Cambridge University Press.
Mackworth, J. F. (1969). Vigilance and habituation. Harmondsworth,
England: Penguin Books.
Martin, M. K., Gonzalez, C., & Lebiere, C. (2004). Learning to make
decisions in dynamic environments: ACT-R plays the beer game. In
M. C. Lovett, C. D. Schunn, C. Lebiere, & P. Munro (Eds.), Proceed-
ings of the Sixth International Conference on Cognitive Modeling (Vol.
420, pp. 178–183). Mahwah, NJ: Erlbaum.
Newell, A. (1990). Unified theories of cognition. Cambridge, MA: Har-
vard University Press.
Parasuraman, R. (1986). Vigilance, monitoring, and search. In K. Boff,
L. Kaufman, & J. Thomas (Eds.), Handbook of perception and human
200 Cleotilde Gonzalez
performance. Vol. 2: Cognitive processes and performance (pp. 1–43).
New York: Wiley.
Proctor, R. W., & Vu, K. L. (2006). Stimulus-response compatibility prin-
ciples: Data, theory, and application. Boca Raton, FL: CRC Press.
Proctor, R. W., Yamaguchi, M., Zhang, Y., & Vu, K. L. (2009). Influence
of visual stimulus mode on transfer of acquired spatial associations.
Journal of Experimental Psychology: Learning, Memory, and Cogni-
tion, 35, 434–445.
Simon, H. A. (1990). Invariants of human behavior. Annual Review of
Psychology, 41, 1–19.
Wickens, C. D. (1984). The multiple resources model of human perfor-
mance: Implication for display design (AGARD/NATO Report). Wil-
liamsburg, VA: AGARD/NATO.
Yamaguchi, M., Dutt, V., Gonzalez, C., & Proctor, R. W. (2011). Cog-
nitive mechanisms of the stimulus-response compatibility effects in
mixed location-relevant and location-irrelevant tasks. Manuscript in
preparation.
Yamaguchi, M., & Proctor, R. W. (2006). Stimulus-response compat-
ibility with pure and mixed mappings in a flight task environment.
Journal of Experimental Psychology: Applied, 12, 207–222.
Young, M. D., Healy, A. F., Gonzalez, C., Dutt, V., & Bourne, L. E.
(2011). Effects of training with added difficulties on RADAR detec-
tion. Applied Cognitive Psychology, 25, 395–407.
10 Modeling Cognitive Tasks
in IMPRINT
Carolyn J. Buck-Gengler,
William D. Raymond, Alice F. Healy,
and Lyle E. Bourne, Jr.
University of Colorado
access
representation plan response Motor planning
1395
Figure 10.1 Cognitive model of data entry.
Model Assessment
The final model was used to fit the data from both experiments,
with only the parameters of hand use, number repetition, and
overall aggregate RT being changed in the model between exper-
iments. Two seeds used for random number generation were
chosen arbitrarily, then two runs were executed for each experi-
ment, one with each of the two seeds. In each run the model
simulated 32 statistical subjects. Each run’s outputs were then
compared to the data from the 32 experimental subjects from
each of the respective experiments by Healy et al. (2004), as well
as to each other. RTs (TRT and RTs for each keystroke) for cor-
rect trials and error output length were used for comparison. To
evaluate model fit, both block mean r2 and Root Mean Square
Error (RMSE) between the set of individual keystroke RTs (50
data points) from the experiment and each of the model outputs
for that experiment were computed (see Table 10.1). RTs over all
10 blocks for each experiment and for both model and experi-
mental data are shown in Figures 10.2 (individual keystrokes)
and 10.3 (first keystroke).
Table 10.1 Values for r2 and RMSEs for Comparison of the Two Final
Model Runs to the Subject Data as well as to Each Other in the Two
Experiments of the Data Entry Task
Experiment Exp vs. Exp vs. Run 1 vs.
run 1 run 2 run 2
r2 Exp 1 0.960 0.956 0.999
Exp 2 0.994 0.993 0.999
RMSE Exp 1 0.087 0.102 0.023
Exp 2 0.031 0.047 0.023
Modeling Cognitive Tasks in IMPRINT 209
1.6
1.4
1.2 E1 data
E2 data
1
E1 model
RT (s)
0.8 E2 model
0.6
0.4
0.2
0
1 3 5 7 9 1 3 5 7 9 1 3 5 7 9 1 3 5 7 9 1 3 5 7 9
1st 2nd 3rd 4th Enter
Figure 10.2 Mean response times (in s) across blocks from one run of
the IMPRINT model of data entry for each experiment and
the experimental data on which it was based, by keystroke.
1.35
E1 Data
1.3 E2 Data
E1 model
1.25
E2 model
RT (s)
1.2
1.15
1.1
1.05
1 2 3 4 5 6 7 8 9 10
Block
Figure 10.3 Mean response times (in s) for the fi rst keystroke, across
blocks, for both experiments and one model run for each
experiment, showing similar patterns of learning and
fatigue for data and model. The dashed lines show best fit-
ting linear functions.
210 Carolyn J. Buck-Gengler et al.
The RADAR Task
Is this in YES!!
Press
right char
space bar
CM set
Move
Prepare eye to Is this the Wait for
VM1
for next next target next
frame square frame
Are there
VM4 more no
Is this in the squares to
memory set look at
NO
yes
Figure 10.4 Cognitive model of RADAR task.
214 Carolyn J. Buck-Gengler et al.
RTs for frames with hits were the sum of eye movement, deci-
sion, and response times. Eye movement and response times
were based on IMPRINT micromodels for eye movement and key
pressing. CM and VM1 decision times were modeled stochasti-
cally. Greater VM4 decision times were multiples of VM1 times
to model search of the memory set. RTs were increased and hit
rates were decreased to simulate the additional load of the sec-
ondary task and the impairment at test from training with tone
counting.
Model Assessment
The final model was run twice with different random seeds. As
an example, Figure 10.5 shows the comparison of the RTs from
the experiment and simulated RTs from the first of these two
model runs. Table 10.2 includes r 2 and RMSE comparing the two
final model runs to the data and to each other, for RT, hit rate,
and false alarm rate.
The fit is not as good for false alarm rate as it is for hit rate
or RT (see Table 10.2), due in large part to the more variable
and complex false alarm rate pattern. Also, twice as many data
216 Carolyn J. Buck-Gengler et al.
1400
Silent-Silent
1300 Silent-Tone
1200 SS
ST
1100
1000
RT (ms)
900
800
700
600
500
CM1 CM4 VM1 VM4 CM1 CM4 VM1 VM4
Training Test
1400
Tone-Silent
1300 Tone-Tone
1200 TS
TT
1100
RT (ms)
1000
900
800
700
600
500
CM1 CM4 VM1 VM4 CM1 CM4 VM1 VM4
Training Test
Session and Shift Type
Figure 10.5 Comparison of the IMPRINT model and RADAR data for
response time for correct responses, for training in silence
(top) and training with tone (bottom) groups. The spelled-
out labels represent subject data, and the labels with two
letters represent IMPRINT output. SS = silent-silent, ST =
silent-tone, TS = tone-silent, TT = tone-tone, CM = consis-
tent mapping, VM = varied mapping, 1 = one target item
and one fi lled square (LP); 4 = four targets and four fi lled
squares (HP). To see the impact of tone, for the training
half compare the two panels, and for the test half within
a panel compare the first and second of each pair of bars.
Also, means in training are averaged for the groups with
the same training condition, because at that point there is
no difference in method between them.
Modeling Cognitive Tasks in IMPRINT 217
Table 10.2 Values for r and RMSE for the Comparisons of the Two
2
Final Model Runs to the Subject Data in the RADAR Task, as well as
to Each Other. The Unit for the Response time RMSE is ms.
r2 RMSE
Response Time
Model Run 1 vs. Data 0.975 54.2
Model Run 2 vs. Data 0.984 49.1
Model Run 1 vs. Model Run 2 0.980 34.3
Hit Rate
Model Run 1 vs. Data 0.969 0.017
Model Run 2 vs. Data 0.940 0.025
Model Run 1 vs. Model Run 2 0.974 0.017
False Alarm Rate
Model Run 1 vs. Data 0.461 0.062
Model Run 2 vs. Data 0.285 0.067
Model Run 1 vs. Model Run 2 0.210 0.081
Model Assessment
The results of the IMPRINT model compared with the experi-
mental data can be seen in Figure 10.6 (Panel 1: Experiment 3
recall, r2 = 0.947, RMSE = 0.034; Panel 2: Experiment 3 fi ring
decision, r2 = 0.973, RMSE = 0.067), and Figure 10.7 (Panel
1: Experiment 2 recall, r2 = 0.978, RMSE = 0.028; Panel 2:
Experiment 2 firing decision, r2 = 0.987, RMSE = 0.080). As can
be seen from the fits and the resulting r2 and RMSE values,
the IMPRINT model did a good job capturing the firing values
based on the recall values, including capturing the depressed
Modeling Cognitive Tasks in IMPRINT 221
0.7
0.6
Proportion of correct response
0.5
0.4
0.3
0.2
Experimental Subjects
0
1 2 3 4 5 6 7
Serial Position
1.4
Mean item distance from firing position
1.3
1.2
1.1
Initial (IMPRINT)
0.9
Final (IMPRINT)
0.8
End 2nd from end 3rd from end
Figure 10.6 Recall (top) and firing response (bottom) results from
Experiment 3, experimental data vs. IMPRINT model. For
the fi ring response, the initial positions (1st, 2nd, 3rd) are
compared to the final positions (7th, 6th, 5th), which have
the same locations across subjects because of the counter-
balancing. Exp = experimental.
222 Carolyn J. Buck-Gengler et al.
0.4
Experimental Subjects
0.3
IMPRINT Statistical Subjects
0.25
0.2
0.15
0.1
0.05
0
1 2 3 4 5 6 7
Serial position
Mean item distance from firing position
1.4
1.3
1.2
1.1
1
Initial (Exp Ss)
Final (Exp Ss)
0.9
Initial (IMPRINT)
Final (IMPRINT)
0.8
End 2nd from end 3rd from end
Item presentation position
Figure 10.7 Recall (top) and firing response (bottom) results from
Experiment 2, experimental data vs. IMPRINT model. For
the fi ring response, the initial positions (1st, 2nd, 3rd) are
compared to the final positions (7th, 6th, 5th), which have
the same locations across subjects because of the coun-
terbalancing. (Note, the IMPRINT values for the firing
response are identical to those for Experiment 3 because
they are based on the same threshold values.) Exp =
experimental.
Modeling Cognitive Tasks in IMPRINT 223
recall values in Experiment 2 compared to Experiment 3 that
one might assume if having to make the firing decision first
did indeed impair memory. Thus, the IMPRINT model supported
the hypothesis that the firing decision was based in large part
on memory for the locations, such that the decisions made by
subjects depended upon how much they recalled about the
locations.
References
Buck-Gengler, C. J., Raymond, W. D., Healy, A. F., & Bourne, L. E.,
Jr. (2007). Modeling data entry in IMPRINT. In Proceedings of the
Sixteenth Conference on Behavior Representation in Modeling and
Simulation (pp. 205–206). Orlando, FL: Simulation Interoperability
Standards Organization.
Buck-Gengler, C. J., Raymond, W. D., Healy, A. F., & Bourne, L. E.,
Jr. (2010). Modeling a visual search task with a secondary task in
224 Carolyn J. Buck-Gengler et al.
IMPRINT. In Proceedings of the Nineteenth Conference on Behavior
Representation in Modeling and Simulation (pp. 63–64). Charleston,
SC: Simulation Interoperability Standards Organization.
Gonzalez, C., & Thomas, R. P. (2008). Effects of automatic detection
on dynamic decision making. Journal of Cognitive Engineering and
Decision Making, 2, 328–348.
Healy, A. F., Kole, J. A., Buck-Gengler, C. J., & Bourne, L. E., Jr. (2004).
Effects of prolonged work on data entry speed and accuracy. Journal
of Experimental Psychology: Applied, 10, 188–199.
Healy, A. F., Kole, J. A., Wohldmann, E. L., Buck-Gengler, C. J., &
Bourne, L. E., Jr. (2011). Data entry: A window to principles of train-
ing. In A. S. Benjamin (Ed.), Successful remembering and successful
forgetting: A festschrift in honor of Robert A. Bjork (pp. 277–296). New
York: Psychology Press.
Henson, R. N. A. (1998). Short-term memory for serial order: The Start-
End Model. Cognitive Psychology, 36, 73–137.
Ketels, S. L., Healy, A. F., Wickens, C. D., Buck-Gengler, C. J., & Bourne,
L. E., Jr. (2010, April). Spatial list learning and decision making in
the fusion paradigm. Poster presented at the 80th Annual Conven-
tion of the Rocky Mountain Psychological Association, Denver, CO.
Kole, J. A., Healy, A. F., & Bourne, L. E., Jr. (2008). Cognitive com-
plications moderate the speed-accuracy tradeoff in data entry: A
cognitive antidote to inhibition. Applied Cognitive Psychology, 22,
917–937.
U.S. Army Research Laboratory. (2011). Improved performance research
integration tool (last updated March 1, 2011). Retrieved from http://
www.arl.army.mil/www/default.cfm?page=445
Young, M. D., Healy, A. F., Gonzalez, C., Dutt, V., & Bourne, L. E., Jr.
(2011). Effects of training with added difficulties on RADAR detec-
tion. Applied Cognitive Psychology, 25, 395–407.
11 Evaluation and Comparison
of Models of Human
Performance During
Training
Bengt Fornberg, William D. Raymond,
Carolyn J. Buck-Gengler, Alice F. Healy,
University of Colorado
Bradley J. Best
Adaptive Cognitive Systems LCC
RADAR Task
The experiments on which the RADAR models are based are
described in Young et al. (2011). The RADAR task, developed by
Gonzalez and Thomas (2008), is a visual search task in which
subjects look for assigned symbol targets in four squares that
move from the four corners to the center of a radar-like display in
a fi xed amount of time. Each search opportunity, as the squares
converge to the center, is called a frame. In each of seven frames
comprising a trial different sets of target and distractor symbols
may be shown in the squares, and the target symbols may differ
from trial to trial (with only one target possible per trial). The
size of the target memory set assigned for a trial includes either
one or four symbols. Targets and distractors may be taken from
different symbol sets (consistent mapping) or the same symbol
set (varied mapping). The squares that contain the symbols may
also be blank. Subjects are to respond only if a target in the cur-
rent memory set appears in one of the squares, and scoring is
based on both accuracy and correct response speed.
The task can be broken down into six taxa from the new tax-
onomy for training (see chapter 8): visual detection (scanning
for symbols), memory/symbolic representation (remembering
targets in the memory set), imagery/visual representation (of
symbols seen in a frame), decision making (target or distrac-
tor decision), motor response planning, and manipulation/fine
motor output (button push on target detection).
ACT-R Platform
ACT-R (Anderson et al., 2004; Anderson & Lebiere, 1998) is a
modeling platform based on a unified theory of cognition devel-
oped through over 30 years of cumulative improvement. It has
accounted for hundreds of phenomena from the cognitive psy-
chology and human factors literature. The version employed
here, ACT-R 6.0, is a modular architecture composed of interact-
ing modules for declarative memory, perceptual systems (such
as vision and audition modules), and motor systems (such as a
manual module), all synchronized through a central production
system.
ACT-R is a hybrid system combining a tractable symbolic
level, implemented as a production system that enables the
specification of complex cognitive functions, with a subsymbolic
level that tunes itself to the statistical structure of the environ-
ment. The combination of these aspects provides both the broad
structure of cognitive processes and the graded characteristics
of cognition, such as adaptivity, robustness, and stochasticity.
The central part of the architecture is the production mod-
ule. A production can match the contents of any combination of
buffers. Buffers include the goal buffer, which holds the current
context and intentions, the retrieval buffer, which holds the most
recent chunk retrieved from declarative memory, visual and audi-
tory buffers, which hold the current sensory information, and the
manual buffer, which holds the current state of the motor module.
During the matching phase, production rules whose conditions
match to the current state of various information buffers (goal,
memory retrieval, perceptual, etc.) qualify to enter a conflict set.
Because ACT-R specifies that only one production can fire at a
time, the rule with the highest expected utility from among those
that match is selected as the one to fire. Utility is graded both by
the expected value of information, driven by activation, and the
quality or exactness of the match itself.
The general structure of the ACT-R model of the data entry
experiments includes two main steps: (a) noticing and encod-
ing of the stimulus from the computer screen, and (b) entry of
the encoded stimulus using the keypad. The first step further
unpacks to include reading of individual numbers, whereas
the second step includes preparing the proper motor program
to press the desired keys. These steps say little about whether
numbers are encoded more than one at a time, and whether
Evaluation and Comparison of Models 231
any key presses occur before all of the numbers are encoded.
As is described below, human participants actually use mul-
tiple strategies to approach even this simple task, and tend to
vary between individuals in a preference either to encode all
four digits before entering any, or to encode a pair of digits at
a time, entering a pair after it is encoded. Thus, the model was
constructed to support both of these strategies. Again, although
the task is quite simple, it still requires maintenance of encoded
stimuli in working memory, potentially decomposing a task into
subgoals (working on entering one pair at a time), and the inter-
action with skilled actions (keyboard entry), which is simulated
through the application of individual ACT-R productions (e.g.,
typing the “9” key on the keypad).
IMPRINT Platform
The versions of IMPRINT used to create the models of digit data
entry and RADAR examined in the present research, IMPRINT
7 and IMPRINT Pro, are primarily used to create simulations of
military personnel and equipment engaged in military tasks.
The simulations can be used to evaluate planning efficiency,
given constraints on time, accuracy, and equipment function-
ality, as well as human skills, abilities, and capacities. Simu-
lations can also take into account variables in the external
environment that may affect personnel or equipment. IMPRINT
was not specifically designed for modeling cognitive tasks; how-
ever, the current modeling effort shows that cognitive models
can be implemented on the IMPRINT platform.
The model of the digit data entry task was based on a cogni-
tive model of the task that involves three subprocesses: (a) read
each number and create an ordered mental representation of the
digits, one digit at a time; (b) access each of the represented dig-
its in sequence to create a motor plan for typing it; and (c) uti-
lize the motor plan to type each digit, followed by the enter key.
The subprocesses were assumed to occur sequentially for each
number. However, accommodation was made in the simulation
for a phenomenon observed in the experimental data, mentioned
in the last section, in which some subjects tended to group, or
chunk, the first two digits of a number and the last two digits
of a number, as evidenced in these subjects by longer response
times for the third keystroke than for the second and fourth.
The chunking phenomenon presumably entails some additional
cognitive processing between the two chunks, which was simu-
lated in the model.
232 Bengt Fornberg et al.
The IMPRINT model consisted of a main network and a goal
network. In the main network, parameters can be set to dupli-
cate the conditions of Experiment 1 of Healy et al. (2004) (all left
hand typing and number repetition in each half) or of Experi-
ment 2 (typing hand crossed with session half and no repeated
numbers). The goal network was called repeatedly until the
stimuli were exhausted. Each run of the model represented the
output from one statistical subject.
A number of human performance parameters in the model
were assigned values stochastically to simulate human variabil-
ity of performance. Values for stochastic variables were taken
from a variety of probability distributions (viz., normal, uniform,
and gamma), which were chosen, together with their param-
eters, to capture distributions observed in the experimental
data. Other model parameters were predetermined through data
inspection and were not left free to vary in the model.
Matlab Platform
Matlab evolved from FORTRAN in the late 1970s, and has since
become a widely used programming environment in science and
engineering. The language is technically an interpreted one, but
its statements are in effect compiled on their first execution,
and then reused in this latter form, providing greater execution
speed. The language is built around matrix/vector operations
and, when used so that this feature is exploited, model execu-
tion speeds come quite close to what is theoretically possible on
the computer hardware on which models are run.
The hardware of modern PCs often allows many computa-
tional threads to execute simultaneously. Not only are com-
puters typically equipped with one or several dual-core (or
multiple-core) processors, each of these cores may furthermore
be hyperthreaded (doubling again the number of independent
simultaneous threads). The resulting opportunities of parallel
processing are automatically utilized in Matlab’s matrix opera-
tions, with no special user attention needed. Matlab’s parallel
computing toolbox can also be utilized for other types of opera-
tions with (in most cases) only a few lines of extra programming.
In the present project, the Matlab model was a direct trans-
lation of the algorithms used in the IMPRINT code. Several
advantages were found to porting the algorithmic parts of
the IMPRINT code to Matlab. In addition to higher execution
speeds, Matlab code is short and easy to write. It can also be
comprehensively viewed as a single program, unlike IMPRINT
code, which is distributed throughout the simulation interface.
Evaluation and Comparison of Models 233
As mentioned, Matlab also has available within it some pow-
erful tools for graphics, optimization, debugging, and perfor-
mance profi ling. Modeling in Matlab also allowed comparison
of two environments specific to modeling human cognition and
performance (ACT-R and IMPRINT) against one with no such
specialization.
It should be stressed that the choice of Matlab (as opposed to
some other scientific or engineering programming environment)
was made for obtaining outside benchmark assessments on the
evaluation speeds of ACT-R and IMPRINT in the most flexible
and convenient way possible, and not because it was expected
that Matlab should be adopted on a large scale for cognitive
modeling. Matlab allowed a focus on obtaining timing and scal-
ability comparisons with little attention diverted to implementa-
tion technicalities.
Accuracy Comparisons
Generally, there were no significant differences among the
platforms in their ability to simulate the empirical data. Cor-
relations were above .97 between all model pairs for both data
entry experiments. The models of digit data entry rely heavily
on random samplings to simulate human behavioral variability,
234 Bengt Fornberg et al.
so that no two runs of any model give identical outputs. Nev-
ertheless, it should be noted that the differences between any
combination of IMPRINT, ACT-R, and Matlab runs used for the
accuracy comparison were no larger than between two different
IMPRINT, ACT-R, or Matlab runs.
Similarly, there were no significant differences in modeling
accuracy between the Matlab and IMPRINT models of RADAR.
The results for the comparison between the Matlab and
IMPRINT models of digit data entry were expected, because the
Matlab code for the digit data entry task was a direct transla-
tion of the IMPRINT algorithms. However, the Matlab code for
the RADAR task was structured differently from the IMPRINT
model of that task. The logical structure of the Matlab model
was a set of nested loops, whereas the IMPRINT structure
consisted of separate, interacting modules. Nevertheless, the
mathematical algorithms and parameter choices were identi-
cal in the two models, and hence, in comparing the Matlab and
IMPRINT models of RADAR there were again no differences in
modeling accuracy.
Optimization
There are several approaches available for parameter optimi-
zation, some of which are yet to reach their full potential in
cognitive modeling. Although finding the global optimum of a
function of one or two variables can usually be handled effi-
ciently, and finding local optima of functions of many variables
is also relatively straightforward (meaning that effective algo-
rithms and software are available in optimization “toolboxes”),
the problem of finding global optima of functions of many vari-
ables is a daunting one. In Gluck, Scheutz, Gunzelmann, Harris,
and Kershner (2007), a calculation based on ACT-R, exploring a
four-variable parameter space by means of 21, 26, 105, and 31
increments in respective parameters, is reported to have con-
sumed 96,000 processor hours on a cluster at Wright-Patter-
son’s High Performance Computing Center. For each additional
variable sampled in a similar manner, times would be expected
to rise by another factor of around 20. Clearly, both faster gen-
eral optimization algorithms and an increase in computational
speed should be pursued as far as possible.
Evaluation and Comparison of Models 237
In a review in 2000 of the 10 most influential algorithms
developed during the 20th century (Cipra, 2000; Dongarra &
Sullivan, 2000), simulated annealing appeared in fi rst place.
Genetic algorithms is a second approach, whose impact is yet
to be fully realized. In many cases, optimization even over
dozens or tens of dozens of variables can be entirely feasible.
Both simulated annealing and genetic algorithms are avail-
able within the Matlab environment, and adding either of these
optimizers to an existing model requires less than 10 extra
lines of code.
A large number of optional toolboxes are available with Mat-
lab, including one that provides both genetic algorithms (GA)
and simulated annealing (SA) search capabilities. As noted
above, these are two very successful strategies for searching
through high-dimensional parameter spaces for locating global
optima more effectively than an exhaustive parameter space
search. Both search methodologies borrow their key ideas from
processes in nature: biological evolution (for GA), and crystal
formation through slow cooling (for SA). Both GA and SA were
applied to the optimization of the IMPRINT/Matlab model of
digit data entry with five free parameters.
For optimization of the data entry model five important model
parameters were chosen that moderated learning and perfor-
mance. Using these five parameters, GA and SA optimizations
were each run 20 times. The runs executed both Experiments
1 and 2. A GA optimization consisted of allowing a population
of size 30 to evolve through 60 generations, for a total of 1800
evaluations of the RMSE objective function. The typical time for
each GA optimization was about 5 min. The SA optimizations
were stopped after about equally many objective function evalu-
ations, thus again taking around 5 min for each full run. The
results for both GA and SA optimizations are shown in Figure
11.1. The 20 GA and 20 SA runs give results comparable to what
an exhaustive search would have provided, but in a fraction of
the approximately 10 days the latter would have required.
For each of the five selected variables, Figure 11.1 shows two
horizontal lines with short vertical lines between them. The
extent of each of the horizontal lines corresponds to a reason-
able range of values for the respective variable, shown at the left
edge of the lines. Along the top line for each variable is shown
the outcome for that variable of 20 separate global GA optimiza-
tions; along the bottom line is shown the same for 20 SA opti-
mizations. The results show that the values (the “hand derived”
values) found by data inspection and a small number of runs
of the original IMPRINT model are consistent with the global
238 Bengt Fornberg et al.
Cognitive l e~ min g GA
[-0.50, 0.00] SA
Moloric learning GA
[-0.20, 0.00] SA
ReJ:.etition Jlriming GA
[0.00, 0.50] SA
Left-hand J:.enalty GA
[1.00, 3.00] SA
Cognitive slowdown GA
[0.00, 0.20] SA
)0' 0.1]
II III IIIII!!)
)0.98' 1) \'" "''
111 1
11' 3)
)1' 2) II
I
I I l l !! II ' I II" 'II )0.98' 1)
)1 .5' 5) Ill 11111 Il l
0.4 u 2 i: i 0.2
0.2 o i i: i U.I
:,.
"i
0 I : 5
c i. 0
-0.1 •0 l i: 1 -C.I
-0.1 -0.2 J •C 2 k-ll :.'fi:i •o: cognitive 0 12
see
''.? 0.2 r I 0 1
-0.05 t. v,
1 a ie
I t *
0 4
0 i
i b
i.
-0 •
•0 2 -ais •i] i
«* •C Ui i: rap learn
0 3
J 0 1
0 ]5
i.
led penalty 0 i
[ o;
J
Vr-
0 •C. i
i: i: •:•
p h y sic a:
i
vuiinli-
o j 0 3
•I :•• n? S
0 1
0 I 0 I
.1 .ii I.
I 1 I.I I
:>•'•
0.2
i £ 0.2
J j
-0.2 j,-: i; i -o.os I" •11 C C5 'j led penalty r
'•-:-' :••'•'
0 c • •:-!-• ' - • > "
. i.ijl 11- V
0.2
9
1.2 1 2 1.3
5 1 % -:... I
'.
'. I
J.i;
-0.2 -0 .-- •i; i •f o^ c n i •i i -, j "i 0.1 n :•• 0 3 '•A J '• •
1(1 :K-I-A--
] • o i 0 I 0.1
R
A er
OJ
5c o:. i o oz
? o-^
i j . :e w 3 0.05
0.™ 0.08
.1 J *
'.<-• <*
:• j o
*e 0
-0.2 -0 .-" •ii l -O.OS c n i r r -:, j o : o 2 n 1.2
l o t ;:••:-! i ;:iy
Figure 11.3 Displays of RMSE errors (the objective function) for the
Matlab data entry model when any pair of two out of the
fi ve parameters that were optimized was left to freely vary
over a reasonable range, while the remaining three param-
eters were held at their assumed best positions. The same
data are displayed in the top right and the bottom left sub-
plots as surface plots and as contour plots, respectively.
The solid dots in the latter fi gures mark the hand-derived
values. The fact that these dots are located at low spots of
the different functions indicates that the manual param-
eter determination was successful.
242 Bengt Fornberg et al.
u-i 0.1 0.2 0.2
o ;• IJ • 0.1 u i
j il- I 0 IJ •
0 IJ 0 J
-ii.i.;. 02 ij. i.O
-0.1 •u i iep lean- IJ -0.2 -i. j o -c i
II
0.4 u j IJ •
10
0-1 0.1
P'"-v= '-'
0.2 o.i
-0.05 Q.12
r 5 0.3
IJ • u •
s a
0 2
^ o5<*
0.2
- 0.1
ij l
# ij l
j
i
I.:.
•:• IJ i
0-i n .;
u j j . •:• 0.2
IJ
'_• -f. 15 .IJ
ccgrithe
i ••: u'.
,
-0.05 u let penalty i.
:
i.
IJ J
O.OB
left pen.*
0.1
1.2 1.2 1.2
«*S *:•»'
u i
O-075 Q.rf* !.:•-•
:
-IJ .• -c 15 • IJ 1
ij.ir:*:
•'. u-: u -01 • J :. 0 i. t l
•«[. Ie,in
12 IJ j o
o i 0.1 0 1 01
c.i-
0.11
0.1
O.OB
«/°-r
*
1.
'•'- : -
j i: 5 :M:-:,
o-oe i: ri-
j .1 j J
IJ :• -c !•: -01 •: U-: -Q1 ij c •:• j i. i: i 12 0.3 i :•
Figure 11.4 The counterpart to Figure 11.3, with the difference that
the original IMPRINT/Matlab model has been replaced by
a RBF data-fitting model. All the trends from the display
in Figure 11.3 can be recognized again here, but a large
amount of the stochastic noise has been eliminated.
Conclusions
The present project is highly complex, both with regard to quan-
tifying cognitive and motoric concepts and with regard to their
modeling. While pursuing both ACT-R and IMPRINT modeling,
it became clear that remarkable opportunities were available by
performing some tasks in a high-speed scientific programming
environment, here represented by Matlab. Speed gains on the
order of 10,000 seem to be typical relative to IMPRINT and ACT-
R, and optimization tools like GA or SA are then also readily
available. Within this environment, equation-based models can
be developed relatively easily, and then optimized with regard
to their parameters. Once the optimization is complete, conver-
sion to IMPRINT will preserve model quality and is an option for
creating codes that interface well with other military tasks. It
244 Bengt Fornberg et al.
seems to be very beneficial to separate clearly the tasks of model
development and model applications (and use the appropriate
tools for each of these). For the application of finished models,
different tasks should each be carried out in the software envi-
ronments that are best suited for them.
For the critical tasks of model development and parameter
tuning, the equation-based approach implemented in a scientific
computing environment uses tools that will scale well also when
extended far beyond the levels they were encountered in the
present research. For example, additional factors in the thou-
sands are quite readily available through distributed processing
(still within the Matlab environment). Another direction for the
future will most likely be improved visualization tools for dis-
playing multivariate data (via the use of radial basis functions).
As much as it might be desirable to find completely new com-
putational paradigms that will totally revolutionize all model-
ing and prediction for cognitive and training processes (maybe
based on the incredible computing resources provided by the
latest generations of peta-flop or exa-flop supercomputer sys-
tems, by highly cost-effective GPU or SPI processors, or by quan-
tum computers), it is still very hard at present to see any more
solid opportunities than what are offered by an equation-based
framework, when this is combined with cutting edge numerical
algorithms.
References
Anderson, J. R., Bothell, D., Byrne, M. D., Douglass, S., Lebiere, C., &
Qin, Y. (2004). An integrated theory of mind. Psychological Review,
111, 1036–1060.
Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought.
Hillsdale, NJ: Erlbaum.
Best, B. J., Gonzalez, C., Young, M. D., Healy, A. F., & Bourne, L. E.,
Jr. (2007). Modeling automaticity and strategy selection in dynamic
visual detection. In Proceedings of the Sixteenth Conference on
Behavior Representation in Modeling and Simulation (pp. 3–11).
Orlando, FL: Simulation Interoperability Standards Organization.
Buck-Gengler, C. J., Raymond, W. D., Healy, A. F., & Bourne, L. E.,
Jr. (2007). Modeling data entry in IMPRINT. In Proceedings of the
Sixteenth Conference on Behavior Representation in Modeling and
Simulation (pp. 205–206). Orlando, FL: Simulation Interoperability
Standards Organization.
Buck-Gengler, C. J., Raymond, W. D., Healy, A. F., & Bourne, L. E.,
Jr. (2010). Modeling a visual search task with a secondary task in
IMPRINT. In Proceedings of the 19th Annual Conference on Behav-
ior Representation in Modeling and Simulation (BRIMS) (pp. 63–64).
BRIMS Society.
Evaluation and Comparison of Models 245
Cipra, B. (2000). The best of the 20th century: Editors name top 10
algorithms. SIAM News, 33, 4.
Dongarra, J., & Sullivan, F. (2000). Top 10 algorithms of the century.
Computing in Science and Engineering, 2, 22–23.
Fasshauer, G.E. (2007). Meshfree approximation methods with Matlab.
Singapore: World Scientific.
Fayyad, U. M., Piatetsky-Shapiro, G., Smyth, P., & Uthurusamy, R.
(Eds.). (1996). Advances in knowledge discovery and data mining.
Menlo Park, CA: American Association for Artificial Intelligence.
Gluck, K. A., & Pew, R. W. (2001). Lessons learned and future direc-
tions for the AMBR model comparison project. In Proceedings of the
10th Annual Conference on Computer-Generated Forces and Behavior
Representation (pp. 113–121). Orlando, FL: Division of Continuing
Education, University of Central Florida.
Gluck, K., Scheutz, M., Gunzelmann, G., Harris, J., & Kershner, J.
(2007). Combinatorics meets processing power: Large-scale com-
putational resources for BRIMS. In Proceedings of the 16th Confer-
ence on Behavior Representation in Modeling and Simulation (BRIMS)
(pp. 73–83). Orlando, FL: Simulation Interoperability Standards
Organization.
Gonzalez, C., Best, B., Healy, A. F., Kole, J. A., & Bourne, L. E., Jr.
(2011). A cognitive modeling account of simultaneous learning and
fatigue effects. Cognitive Systems Research, 12, 19–32.
Gonzalez, C., Dutt, V., Healy, A. F., Young, M. D., & Bourne, L. E., Jr.
(2009). Comparison of instance and strategy models in ACT-R. In A.
Howes, D. Peebles, & R. Cooper (Eds.), 9th International Conference
on Cognitive Modeling—ICCM2009. Manchester, UK.
Gonzalez, C., & Thomas, R. P. (2008). Effects of automatic detection
on dynamic decision making. Journal of Cognitive Engineering and
Decision Making, 2, 328–348.
Han, J., & Kamber, M. (2006). Data mining: Concepts and techniques
(2nd ed.). Waltham, MA: Morgan Kaufman.
Hardy, R. L. (1971), Multiquadric equations of topography and other
irregular surfaces. Journal of Geophysical Research, 76, 1905–1915.
Healy, A. F., Kole, J. A., Buck-Gengler, C. J., & Bourne, L. E., Jr. (2004).
Effects of prolonged work on data entry speed and accuracy. Journal
of Experimental Psychology: Applied, 10, 188–199.
Healy, A. F., Kole, J. A., Wohldmann, E. L., Buck-Gengler, C. J., &
Bourne, L. E., Jr. (2011). Data entry: A window to principles of train-
ing. In A. S. Benjamin (Ed.), Successful remembering and successful
forgetting: A festschrift in honor of Robert A. Bjork (pp. 277–296). New
York: Psychology Press.
Pitt, M. A., & Myung, J. (2002). When a good fit can be bad. Trends in
Cognitive Sciences, 6, 421–425.
Raymond, W. D., Fornberg, B., Buck-Gengler, C. J., Healy, A. F., &
Bourne, L. E., Jr. (2008). Matlab optimization of an IMPRINT model
of human behavior. In Proceedings of the Seventeenth Conference
on Behavior Representation in Modeling and Simulation (pp. 26–34).
Orlando, FL: Simulation Interoperability Standards Organization.
246 Bengt Fornberg et al.
Thuraisingham, B. M. (1998). Data mining: Technologies, techniques,
tools, and trends. Hug, Switzerland: CRC Press/Informa.
Young, M. D., Healy, A. F., Gonzalez, C., Dutt, V., & Bourne, L. E., Jr.
(2011). Effects of training with added difficulties on RADAR detec-
tion. Applied Cognitive Psychology, 25, 395–407.
Young, M. J. (2003). Human performance model validation: One size
does not fit all. SCSC ’03, 732–736.
12 A Compact Mathematical
Model for Predicting
the Effectiveness
of Training
Matt Jones, Lyle E. Bourne, Jr.,
and Alice F. Healy
University of Colorado
a = t– (1)
a= ∑βti– (2)
i
Predicting the Effectiveness of Training 251
where i indexes all past learning experiences, and ti is the time
elapsed since experience i. When there are n equally spaced
experiences, separated by gaps of length , then it can easily be
shown (by replacing the sum with an integral) that Equation 2
is approximately equal to -γ/(1-)∙n1-γ. Therefore the power law
of practice is obtained under appropriate substitution for b and
c, provided ϵ (-1, 0). Figure 12.1 displays the evolution of activa-
tion under equally spaced practice, illustrating how the power
law of practice emerges from the power law of forgetting.
3.5
2.5
Activation
1.5
0.5
0
0 50 100 150 200 250 300
Time
Figure 12.1 Emergence of power law of practice from power law of for-
getting. Solid curve is activation predicted from Equation
1, assuming regularly spaced training events, each trace
decaying according to a power law. Dashed line is fit of
power law of practice.
252 Matt Jones, Lyle E. Bourne, Jr., and Alice F. Healy
Transfer
Transfer is an important consideration in any learning con-
text, because test situations are generally different from those
in which knowledge was acquired. The general question of how
experience in one task or context affects performance in another
is more complex than any existing psychological theory can
address. However, in simpler cases, where issues of task struc-
ture can be ignored and the relationship between training and
transfer situations can be reduced to some notion of psychologi-
cal distance, modeling of how knowledge is generalized from one
situation to another can be straightforward.
The most successful example of this approach is in research
on stimulus generalization, which investigates how knowledge
associated with one stimulus (e.g., an appropriate response) is
extended to another, novel stimulus. Extensive empirical evi-
dence and normative Bayesian modeling support the proposal
that stimulus generalization is an exponentially decaying func-
tion of perceived similarity between stimuli (Shepard, 1987).
This universal law of generalization can be formalized as
a = e−d (3)
Generalization
Dissimilarity
a = t–e−d (4)
a= ∑βiti–e–d i (5)
i
2
Activation 1.5
1
0.5
0
0
20
40
Tas
k
60
80
90 100
100 60 70 80
40 50
10 20 30
0
Time
Deliberate Practice
Highly focused and highly motivated practice is best in terms
of promoting skill acquisition and expertise. This principle can
be incorporated into the model by assuming that acquisition
strength, , is greater for training experiences associated with
more focused or deliberate practice. The most definitive research
on this principle has been conducted by Ericsson and colleagues
(e.g., Ericsson, Krampe, & Tesch-Römer, 1993), who argued that
deliberate practice is a requirement for optimizing the level of
skill or becoming an expert at a given task.
Depth of Processing
Training that promotes deep and elaborate processing enhances
durability of knowledge (Craik & Lockhart, 1972). Deep process-
ing usually involves semantic relationships between the mate-
rial to be learned and the learner’s existing knowledge, whereas
shallow processing draws merely on the surface characteristics
Predicting the Effectiveness of Training 259
of the material, such as phonological and graphemic features.
This principle is incorporated into the model on the assumption
that the major influence of depth of processing is to mitigate for-
getting. Thus, the decay parameter, , is reduced by enhancing
processing depth.
Testing Effect
Testing after initial presentation of material slows forgetting
(Carpenter, Pashler, Wixted, & Vul, 2008). This principle is
predicted as a corollary to the principle of depth of processing,
under the assumption that testing leads people to process infor-
mation more deeply. That is, testing events during the training
phase can be treated as additional training events with smaller
values of the decay parameter, .
Spacing of Practice
Knowledge is retained for longer periods when training sessions
are spaced in time than when they are massed (Cepeda, Pashler,
Vul, Wixted, & Rohrer, 2006; Cepeda, Vul, Rohrer, Wixted, &
Pashler, 2008; see also chapter 13). This principle is predicted
by the model under the assumption that the decay rate of any
training experience depends on the time since the preceding
encounter (Anderson & Schooler, 1991).
{ }
i = max 0, b(ti – ti–1) –0 (8)
Discussion
Development, implementation, and simulation of training sys-
tems can all benefit from quantitative predictions of performance.
The mathematical model proposed in this chapter offers one way
to generate such predictions, in a way that allows comparison
among arbitrary training schedules consisting of varying tasks
262 Matt Jones, Lyle E. Bourne, Jr., and Alice F. Healy
and contexts. It is founded on basic cognitive theory and has
potential utility in a variety of practical applications.
The model is formulated to be concise and general yet also
to capture a wide range of empirical principles related to the
effectiveness of training. The core activation equation is built
on the delineation of three fundamental cognitive processes of
learning: acquisition, retention, and transfer. Different model
parameters govern the characteristics of each process, thus
determining the initial strength of knowledge from each train-
ing experience, how that knowledge decays over time, and how
it can be applied to situations other than the one in which it was
acquired. The collective contribution of all past training events
determines the model’s prediction for performance in any future
testing situation.
Some work is still needed for the model to be implemented.
First, values of the model parameters must be determined.
These values will in general be domain-specific, so this work
would mostly need to be carried out empirically, for each
domain of interest. There are also interesting questions of vari-
ance and covariance of parameters across subjects, which could
be worked out if one desired to model individual differences.
Second, several qualitative parameter dependencies proposed
in the previous section would need to be formalized for model-
ing certain training principles. For example, it is proposed that
decay is slower for procedural than declarative knowledge and
for abstract rules than specific instances, but the magnitudes
of these differences (as well as possible domain dependencies)
are yet to be worked out. Third, the equations of the model need
to be implemented as computer code or integrated into existing
software. This effort should be fairly straightforward, given the
relative simplicity of the model.
Although the proposed model is meant primarily as a tool
for applying cognitive theory, it does raise interesting theoreti-
cal questions. One question concerns the relationship between
retention and generalization. Both of these phenomena concern
how the effect of learning diminishes as training and testing
contexts become more distant—either in time or in task char-
acteristics. The mathematical forms used to model the two pro-
cesses are compatible, but they are not identical, with retention
decaying according to a power function and transfer decaying
exponentially. The debate over power law versus exponential
memory decay has favored the power law (Wixted & Carpenter,
2007). However, we know of no studies in the generalization lit-
erature that test between exponential and power functions, and
we assume exponential generalization in the present exposition
Predicting the Effectiveness of Training 263
only for consistency with prior modeling efforts (Shepard, 1987).
Therefore, an intriguing topic of future research would be to
evaluate a power law for generalization. If a power law were
empirically supported over the exponential, such a result would
suggest a more mathematically elegant—and perhaps theoreti-
cally deep—connection between retention and generalization.
Another question concerns how multiple prior experiences
combine to determine knowledge strength. Although exemplar
models have been highly successful (Hintzman, 1986; Logan,
1988; Nosofsky, 1992), there is good evidence from concept
learning that people learn more about a category than the union
of its elements (Sakamoto, Jones, & Love, 2008; Tenenbaum &
Griffiths, 2001). The spacing effect in memory is a temporal
analog of this phenomenon. Although Equation 8 above offers
one explanation of the spacing effect that maintains additiv-
ity of memory traces, the analogy to category learning (derived
from the connection between retention and generalization at the
core of the present proposal) suggests that richer forms of learn-
ing may be needed in memory modeling as well. One fruitful
approach may be to augment the current instance-based model
with rule-based abstraction of the sort discussed in chapters 9
and 13.
The above considerations suggest ways in which the model
might be refined or extended, and they might limit its predic-
tive power in its current form. Other limitations arise from
the simplifying view of transfer among tasks, specifically that
strength of transfer can be reduced to a notion of similarity.
A more complete model would need to address the internal
structure of tasks and the applicability of knowledge structures
acquired from one task in performing another. Such an exten-
sion would in turn entail consideration of the sort of construc-
tive learning (as opposed to mere accumulation of knowledge)
that presumably occurs as learners piece together experiences
to develop deeper representations and understandings of task
environments. Nevertheless, we suggest that the simplicity of
the present model makes it well suited for implementation in a
wide variety of applications, and that it can offer effective pre-
dictions of how people will perform following different histories
of training.
Note
1. An undesirable mathematical property of decreasing power laws
(i.e., with negative exponents) is that they predict infinite value
at t = 0. A useful convention for solving this problem is to start
264 Matt Jones, Lyle E. Bourne, Jr., and Alice F. Healy
time at 1, or equivalently to replace t with t+1. This approach has
the added benefit that equals the initial memory strength at the
time of encoding.
References
Anderson, J. R., Bothell, D., Byrne, M. D., Douglass, S., Lebiere, C.,
& Qin, Y. (2004). An integrated theory of the mind. Psychological
Review, 111, 1036–1060.
Anderson, J. R., Fincham, J. S., & Douglass, S. (1999). Practice and
retention: A unifying analysis. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 25, 1120–1136.
Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought.
Mahwah, NJ: Erlbaum.
Anderson, J. R., & Schooler, L. (1991). Reflections of the environment in
memory. Psychological Science, 2, 396–408.
Archer, R., Walters, B., Yow, A., Carolan, T., Laughery, K. R., & Gillis, P.
(1999). Training as a performance shaping factor in computer gen-
erated forces. Proceedings of the 1999 Computer Generated Forces
Conferences, Orlando, FL.
Bourne, L. E., Jr., Healy, A. F., Kole, J. A., & Graham, S. M. (2006).
Strategy shifts in classification skill acquisition: Does memory
retrieval dominate rule use? Memory & Cognition, 34, 903–913.
Bourne, L. E., Jr., Healy, A. F., Parker, J. T. & Rickard, T. C. (1999).
The strategic basis of performance in binary classification tasks:
Strategy choices and strategy transitions. Journal of Memory and
Language, 41, 223–252.
Bourne, L. E., Jr., Raymond, W. D., & Healy, A. F. (2010a). Quantify-
ing performance effects of training manipulations: Performance shap-
ing functions based on selected training principles (Technical report:
CRT Publications). Boulder: University of Colorado.
Bourne, L. E., Jr., Raymond, W. D., & Healy, A. F. (2010b). Strategy
selection and use during classification skill acquisition. Journal
of Experimental Psychology: Learning, Memory, and Cognition, 36,
500–514.
Carpenter, S. K., Pashler, H., Wixted, J. T., & Vul, E. (2008). The
effects of tests on learning and forgetting. Memory & Cognition, 36,
438–448.
Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006).
Distributed practice in verbal recall tasks: A review and quantitative
synthesis. Psychological Bulletin, 132, 354–380.
Cepeda, N. J., Vul, E., Rohrer, D., Wixted, J. T., & Pashler, H. (2008).
Spacing effects in learning: A temporal ridgeline of optimal reten-
tion. Psychological Science, 19, 1095–1102.
Chater, N. & Vitányi, P. (2003). The generalized universal law of gener-
alization. Journal of Mathematical Psychology, 47, 346–369.
Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A frame-
work for memory research. Journal of Verbal Learning and Verbal
Behavior, 11, 671–684.
Predicting the Effectiveness of Training 265
Ebbinghaus, H. (1913). Memory: A contribution to experimental psy-
chology. New York: Teachers College, Columbia University. (Original
work published 1885)
Ericsson, K. A., Krampe, R. T., & Tesch-Römer, C. (1993). The role of
deliberate practice in the acquisition of expert performance. Psycho-
logical Review, 100, 363–406.
Healy, A. F. (2007). Transfer: Specificity and generality. In H. L. Roedi-
ger, III, Y. Dudai, & S. M. Fitzpatrick (Eds.), Science of memory: Con-
cepts (pp. 271–275). New York: Oxford University Press.
Healy, A. F., Fendrich, D. W., Crutcher, R. J., Wittman, W. T., Gesi, A.
T., Ericsson, K. A., & Bourne, L. E., Jr. (1992). The long-term reten-
tion of skills. In A. F. Healy, S. M. Kosslyn, & R. M. Shiffrin (Eds.),
From learning processes to cognitive processes: Essays in honor of
William K. Estes (Vol. 2, pp. 87–118). Hillsdale, NJ: Erlbaum.
Healy, A. F., Wohldmann, E. L., & Bourne, L. E., Jr. (2005). The pro-
cedural reinstatement principle: Studies on training, retention, and
transfer. In A. F. Healy (Ed.), Experimental cognitive psychology and
its applications (pp. 59–71). Washington, DC: American Psychologi-
cal Association.
Hintzman, D. L. (1986). “Schema abstraction” in a multiple-trace mem-
ory model. Psychological Review, 93, 328–338.
Jones, M., Maddox, W. T., & Love, B. C. (2005). Stimulus generalization
in category learning. In Proceedings of the 27th Annual Meeting of
the Cognitive Science Society (pp. 1066–1071).
Kole, J. A., Healy, A. F., Fierman, D. M., & Bourne, L. E., Jr. (2010).
Contextual memory and skill transfer in category search. Memory &
Cognition, 38, 67–82.
Kruskal, J. B. (1964). Nonmetric multidimensional scaling: A numeri-
cal method. Psychometrika, 29, 115–129.
Logan, G. D. (1988). Toward an instance theory of automatization. Psy-
chological Review, 95, 492–527.
Luce, R. D. (1963). Detection and recognition. In R. D. Luce, R. R.
Bush, & E. Galanter (Eds.), Handbook of Mathematical Psychology
(pp. 103–189). New York: Wiley.
Markman, A. B., & Gentner, D. (1993). Structural alignment during
similarity comparisons. Cognitive Psychology, 25, 431–467.
Medin, D. L., Goldstone, R. L., & Gentner, D. (1993). Respects for simi-
larity. Psychological Review, 100, 254–278.
Newell, A., & Rosenbloom, P. S. (1981). Mechanisms of skill acquisition
and the power law of practice. In J. R. Anderson (Ed.), Cognitive
skills and their acquisition (pp. 1–55). Hillsdale, NJ: Erlbaum.
Nosofsky, R. M. (1986). Attention, similarity, and the identification-cat-
egorization relationship. Journal of Experimental Psychology: Gen-
eral, 115, 39–57.
Nosofsky, R. M. (1992). Similarity scaling and cognitive process mod-
els. Annual Review of Psychology, 43, 25–53.
Petrov, A. A., & Anderson, J. R. (2005). The dynamics of scaling: A
memory-based anchor model of category rating and absolute identi-
fication. Psychological Review, 112, 383–416.
266 Matt Jones, Lyle E. Bourne, Jr., and Alice F. Healy
Rubin, D. C., & Wenzel, A. E. (1996). One hundred years of forgetting:
A quantitative description of retention. Psychological Review, 103,
734–760.
Sakamoto, Y., Jones, M., & Love, B.C. (2008). Putting the psychology
back into psychological models: Mechanistic vs. rational approaches.
Memory & Cognition, 36, 1057–1065.
Shepard, R. N. (1957). Stimulus and response generalization: Deduc-
tion of the generalization gradient from a trace model. Psychological
Review, 65, 242–256.
Shepard, R. N. (1962). The analysis of proximities: Multidimensional
scaling with an unknown distance function. Psychometrika, 27,
125–140.
Shepard, R. N. (1987). Towards a universal law of generalization for
psychological science. Science, 237, 1317–1323.
Sutherland, N., & Mackintosh, N. (1971). Mechanisms of animal dis-
crimination learning. New York: Academic Press.
Tenenbaum, J. B., & Griffiths, T. L. (2001). Generalization, similar-
ity and Bayesian inference. Behavioral and Brain Sciences, 24,
629–640.
Tversky, A. (1977). Features of similarity. Psychological Review, 84,
327–352.
Tversky, A., & Gati, I. (1982). Similarity, separability, and the triangle
inequality. Psychological Review, 89, 123–154.
Wickelgren, W. A. (1972). Trace resistance and the decay of long-term
memory. Journal of Mathematical Psychology, 9, 418–455.
Wixted, J. T., & Carpenter, S. K. (2007). The Wickelgren power law
and the Ebbinghaus savings function. Psychological Science, 18,
133–134.
13 Put the SPRINT
in Knowledge Training
Training with SPacing,
Retrieval, and INTerleaving
Mark McDaniel
Washington University in St. Louis
Spacing
A fair amount of laboratory research has indicated that reten-
tion of verbal material is improved with spaced review of material
rather than massed review (see Cepeda, Pashler, Vul, Wixted, &
Rohrer, 2006, for a review). However, for several reasons this
body of work may not be compelling to professionals who are
in the business of knowledge training. The materials used in
such laboratory research are simple (nonsense syllables, word
pairs, word lists), and not reflective of the more complicated and
conceptual material that is central to many knowledge-training
contexts. Further, the retention intervals studied are usually
on the order of hours (if that) rather than the time frame of
weeks and months (and years) that is of interest in the authentic
knowledge training situations noted at the outset. Accordingly, a
legitimate question is whether spacing fosters better retention of
more authentic materials (those of interest in knowledge train-
ing) over long intervals spanning months and years.
270 Mark McDaniel
Recent research has begun to address this key question. In
one experiment, retention of knowledge over a 9-month time
interval was examined as a function of the degree of spacing (for
review of the target information). Students in an 8th grade U.S.
history course were given a set of facts to review that they had
learned in the course (Carpenter, Pashler, & Cepeda, 2009). One
group of students reviewed the facts relatively soon after they
had finished their exams in the course (1 week later), whereas
another group reviewed the facts with much greater spacing
after finishing the exams (16 weeks later). On a recall test that
was administered 9 months after each group’s review session,
the group with greater spacing remembered over 50% more of
the reviewed facts than did the group with minimal spacing.
Moreover, facts reviewed with minimal spacing showed mini-
mal retention relative to tested (control) facts that had not been
reviewed. Thus, spaced review of target knowledge enhanced
retention over long periods that are reflective of those of import
in knowledge training contexts.
In another impressively systematic experiment, a greater
range of spacing gaps (the interval between initial learning and
review) was examined, as well as several long retention intervals
(the time interval between the review and the final test) that
ranged from 1 week to 1 year (Cepeda, Vul, Rohrer, Wixted, &
Pashler, 2008). Subjects initially had to learn 32 trivia facts;
across conditions, subjects reviewed these facts immediately
upon initial learning (no spacing) or after a particular gap (e.g.,
1 day, 1 week, 3 weeks, or 15 weeks). Final recall and recogni-
tion tests administered after the prescribed delay showed that
spaced review produced better long-term knowledge retention
than did massed review.
Importantly, the study also established with ecologically valid
retention intervals that the optimal spacing gap for knowledge
retention depends on the ratio between the spacing gap and the
retention interval. The results were roughly consistent with the
rule of thumb that spacing is most effective when the time inter-
val of the gap is approximately 15 to 20% of the retention inter-
val. For instance, for a 10-week retention interval, to provide the
optimal benefit of spacing, the gap between initial learning and
review should be between 1 and 2 weeks. For more precise opti-
mization of spacing it is worth noting, however, that the opti-
mal ratio of spacing gap to retention interval became smaller
with the longest intervals so that for a 1-year retention inter-
val a spacing gap of about a month (8%) was optimal. Also, the
decline in retention when nonoptimal spacing gaps were used
was more substantial when the gaps were too short than when
Put the SPRINT in Knowledge Training 271
the gaps were longer than the optimal gap. These findings rein-
force the observation that knowledge-training contexts that rely
on confining training to compressed time frames do not favor
long-term retention of the knowledge.
Figure 13.1 The trained points for the function learning experiment
contrasting massed and spaced training.
50
Output (Predicted) Value
40
Output
30 Massed Predict
Spaced Predict
20
10
0
60 70 60 50 100 110 120 130
Input value
Figure 13.2 Mean predicted outputs at test for the trained points for
the massed and spaced conditions.
140
120
Output (Predicted) Value
100
60
Output
Massed Predict
Spaced Predict
60
40
20
0
0 20 40 60 80 100 120 140 160
Input value
Figure 13.3 Mean predicted outputs at test for transfer points for the
massed and spaced conditions.
Put the SPRINT in Knowledge Training 275
In sum, recent experimental work with relatively real-world
materials and retention intervals converges on the conclusion
that spacing knowledge training promotes long-term reten-
tion of target factual information and target skills relative to
compressing training into short time periods. Importantly, the
newest results also suggest that spacing is favored for knowl-
edge training that involves induction of concepts. This pattern
is especially informative and critical for training practice, as
learners’ intuitions (and presumably those of people responsible
for designing training) are that massing supports better induc-
tive learning than does spacing (Kornell & Bjork, 2008).
Retrieval Practice
The Holy Grail for skill training is practice, practice, practice. A
key issue for knowledge training is to identify what constitutes
effective practice. One straightforward idea is to repeatedly
study, review, or reread the target information. An alternative
is to practice retrieval of the information. Retrieval practice
essentially involves testing oneself on the target information
or having tests administered as part of the training protocol.
Theoretically, testing is not just an assessment of knowledge;
it also modifies memory (Carpenter & DeLosh, 2006; McDan-
iel & Masson, 1985). Empirically, the basic-laboratory literature
has shown that repeated testing is substantially more effective
in promoting retention than is repeated study (e.g., Karpicke
& Roediger, 2007, 2008). Though suggestive, these basic labo-
ratory paradigms (which also are the foundation for the theo-
retical work) diverge in potentially critical ways from many real
knowledge-training contexts. In the typical laboratory para-
digms, the materials are presented in a single session, learners
have no opportunity to review the material outside of the labo-
ratory session, and learners may have little inherent motivation
to learn the target material. In contrast, in most training situa-
tions, the target material is emphasized in presentations, dem-
onstrations, simulations, labs, and so on, and can be reinforced
with assigned reading and homework. Accordingly, a legitimate
question is whether retrieval practice has substantial benefits
when applied to ecologically valid training contexts.
80 76.1
70
59.7
60
53.7 54.9
50
40
Chapter Exams End of the Semester
Figure 13.4 6th Grade Social Studies (data from Roediger et al., 2011).
Interleaving
Knowledge-training situations require training a body of knowl-
edge and skills that require differentiation among different kinds
of problems and situations within a particular domain and link-
ing appropriate responses to these problems and situations. For
instance, in medicine, training in oncology requires students
to learn about and differentiate different types of tumors and
appropriate treatments. In business, training in accounting
requires students to learn different computational and concep-
tual components; for example revenue versus income. A straight-
forward training procedure that seems intuitively superior to
many individuals is to block training of each particular problem
type (cf. Kornell & Bjork, 2008). Indeed in teaching math, most
mathematics practice assignments consist almost entirely of a
block of problems on the immediately preceding topic (e.g., a
set of 24 ratio problems after a lesson on ratios; Rohrer & Tay-
lor, 2007). Unfortunately, blocking practice may not foster the
ability to discriminate among similar problems and situations,
thereby disrupting the selection of an appropriate and effective
response.
Research from several arenas suggests that interleaving prac-
tice, such that different types of similar problems are intermixed
rather than blocked, produces much more effective training out-
comes than does the standard blocked practice. This claim is par-
ticularly supported in studies of motor skill learning (reviewed
Put the SPRINT in Knowledge Training 281
in Rohrer & Pashler, 2010). Regarding knowledge training, Kor-
nell and Bjork (2008) presented learners with paintings from
various artists either blocked by artist or interleaved by artist
(so that paintings from the same artist were never successively
presented). After the learning phase, participants were required
to classify new paintings (not seen in training) from the same
artists. Across several experiments, countering popular wisdom
and to the surprise of the researchers, interleaving exemplars
during training promoted more accurate categorization of the
new paintings than did blocking the exemplars.
Building on the findings described above, Rohrer and col-
leagues conducted several clever experiments to isolate and
demonstrate the benefits of interleaved practice in math learning
(Rohrer & Taylor, 2007; Taylor & Rohrer, 2010). In one experiment,
college students were taught four different types of geometric
solids and how to find their volume (wedge, spheroid, spherical
cone, and half cone). One group of students was trained using a
typical blocked procedure: for each solid a tutorial was provided,
followed immediately by four practice problems on the solid.
Another group of students was trained using an interleaved pro-
cedure. The tutorials on all four solids were first provided, and
then the practice problems were presented in a random order (but
constrained so that one of each problem was presented in a set
of four). Both the blocked and interleaved groups were given two
training sessions, spaced 1 week apart. A week after the second
training session, students were tested with two novel problems
for each of the four solids. Students who received the blocked
training procedure, despite two spaced sessions, performed
poorly on the test, solving an average of only 20% of the problems
(less than two problems)! By contrast, students given interleaved
training showed over a three-fold advantage over blocked train-
ing in their ability to solve the test problems (63% of the problems
solved). Similar benefits of interleaving were reported in a related
experiment on math training in young children (Taylor & Rohrer,
2010). Importantly, converging with the claim that blocking is
inimical to learning which response is appropriate for a given
problem or context, the high failure rate on the test problems (for
the blocked group) was nearly always a consequence of selecting
an incorrect formula (see also chapter 2 in this volume; Schnei-
der, Healy, & Bourne, 2002).
Further, performance on the training problems provides some
insight into why instructional designers, as well as trainees,
may erroneously believe that blocked training is preferred. With
blocked training, practice performance was particularly good
(89% averaged across the two sessions), and significantly better
282 Mark McDaniel
than with interleaved training (60% on average; Rohrer & Taylor,
2007). Blocked training may well quickly provide learners with
facility in how to solve each problem, thereby giving the illusion
that learning is complete. However, to apply the knowledge (e.g.,
in the job environment) individuals need to be able to deter-
mine which response is most appropriate for the problem that is
encountered. The evidence available at this time indicates that
interleaving but not blocking assists in training this knowledge.
Loukia Loukopoulos
San Jose State University
Training (Theory)
Anyone who has ever peeked into an aircraft cockpit can easily
imagine the substantial volume and variety of activities that a
pilot must perform during a flight. To be managed, this workload
must be organized. Furthermore, it must be standardized, so
activities are accomplished in the same way by a pilot on every
flight, and also by different pilots on different flights. These
Training for Real-World Job Performance 291
activities are, therefore, dutifully described in each airline’s
Flight Operations Manual (FOM). The FOM includes detailed
procedures for every activity pilots are expected to perform,
from the external checks prior to boarding, through preparing
the aircraft for flight, starting the engines, taxiing from the gate
to the runway, and all the way through shutting the engines
down and securing the aircraft at the end of the day at its final
destination.
Procedures must be, first and foremost, consistent with the
technical characteristics and limitations of the aircraft. The orig-
inal FOM is created by the aircraft manufacturer, based on the
design of the aircraft and on many hours of flight testing, giving
it an engineering focus, so that all aircraft systems function as
designed. The manufacturer, however, has no experience in car-
rying passengers in routine daily operations. It is therefore up to
each airline to take the FOM and adapt it to its own operational
needs, by adding different steps, depending on the particular
type of operation it conducts, and on its corporate culture.
Ultimately, FOMs describe two main kinds of procedures:
flows and checklists. Flows are sequences of actions that have
to be performed from memory. These flows are often based on
the spatial layout of the cockpit, so that a pilot may check all the
gauges on the instrument panel starting from the left and moving
to the right. In addition, actions are often functionally related;
for example, setting the flaps requires hydraulic pressure, and
hydraulic pressure, in turn, requires that the engines be started,
and for that to happen the battery has to be turned on.
Pilots have many opportunities to practice flows through
training and through their operational experience, and can thus
perform them from memory. To guard against memory failures,
there is the second kind of procedure in the FOM, the checklist.
This is a list of selected items from the flow that are considered
absolutely critical and must therefore be checked and confirmed
prior to the next phase of flight.
What does all this have to do with training? Flows and check-
lists form the basis for pilots’ training. In training, pilots spend
many hours in cockpit mockups and flight simulators, prac-
ticing these procedures to perfection. Unfortunately, 100% reli-
ability is not a common human trait and checklist procedures
carry particular challenges (e.g., Barshi & Healy, 1993; also see
the serial position principle, chapters 2 and 3); the focus here
is on the implications of having to perform these procedures in
the real world for the design and development of such training.
In the case of setting the flaps for takeoff, the action takes
place when the pilots are ready to taxi the aircraft from the gate
292 Immanuel Barshi and Loukia Loukopoulos
area to the runway. According to the FOM, each of the two pilots
in the cockpit has specific responsibilities during this phase of
flight. Before setting the aircraft into motion, the pilots set the
aircraft flaps to the takeoff position and request taxi clearance
(i.e., navigational “directions” to the runways) from Air Traf-
fic Control (ATC). While taxiing, the two pilots must perform,
from memory, a set of activities prescribed by a flow, which they
then verify by conducting together the corresponding checklist.
The taxi flow and checklist typically contain the “flaps” item,
because of its critical nature. Oftentimes, the FOM prescribes
yet one more, shorter flow and checklist as the aircraft nears
the runway. In any case, by the time all these taxi activities are
completed, the aircraft is ready for takeoff.
Pilots have additional responsibilities besides flows and
checklists while taxiing. They are expected to monitor several
radio frequencies and respond if called; they must monitor their
progress along the taxi route confirming that they are correctly
following the instructions in their taxi clearance; and they must
ascertain that their movements do not conflict with those of
other aircraft. Furthermore, they might be called upon by the
cabin crew, or receive important dispatch information via com-
puters. In short, taxiing to the runway is a very busy time in
the cockpit. Most of these responsibilities are mentioned in the
FOM, but are not organized in specific procedures and are with-
out specific guidance on how to best accomplish them or ensure
their integrity. Many responsibilities are mentioned in passing,
as side-notes to the precise description of procedural activities.
FOM-prescribed taxi-phase activities (and activities for each
of the flight phases), while large in number, are admittedly
straightforward as they are neatly packed into procedures, flows
and checklists, and laid out on paper. Furthermore, their orga-
nization lends itself perfectly to training thanks to three key
features. First, activities are linear, such that a given activity is
always preceded by and followed by specific activities, always in
the same sequence, because one activity is ostensibly a prereq-
uisite for the next. For example, the checklist is called for only
after each pilot has had a chance to accomplish the prerequisite
flow. Second, activities are predictable. Information and events
from the external environment are anticipated, and occur pre-
dictably at certain points in time. And third, activities are in the
pilots’ control. For instance, the pilots decide when to request the
taxi clearance, and they receive it as soon as they do so.
These three features make procedures attractive for train-
ing, both for pilots, and especially for trainers. Linearity facili-
tates learning. Given sufficient practice in flight simulators,
Training for Real-World Job Performance 293
and taking advantage of natural cognitive characteristics, such
as motor memory, a pilot can learn to perform a linear set of
activities with minimal effort and high reliability. Moreover, an
instructor can assess one’s proficiency simply by ensuring that
all activities are accomplished, and in the right order. Predict-
ability implies no surprises. Consequently, these procedures
lend themselves to learning because a pilot can simply and
automatically perform well-practiced actions, without concern
for unforeseen factors. Likewise, the instructor can focus on
watching that habitual activities are accomplished flawlessly,
and can devote more time to serious contingencies, such as
those requiring emergency procedures. Controllability, the full
control of the initiation, timing, and termination of activities by
the pilot, again implies no surprises, specifically the absence
of factors that might alter or otherwise remove altogether that
control. Such organized situations can be both demonstrated by
pilots and assessed by instructors, so that the expected control-
lability of activities, together with their predictability and linear-
ity, all lend themselves very conveniently to the ease of training.
It’s not difficult to appreciate that people organize many of
their daily activities around procedures that, although not
always explicit or formally taught, give a linear, predictable, and
controllable flavor to behavior. Withdrawing cash from an auto-
matic teller machine (ATM), for example, is accomplished fol-
lowing a fi xed, linear sequence of activities (insert card, type
PIN, select option to withdraw, retrieve card, receive money and
receipt), each of which predictably generates an outcome (typ-
ing in the correct PIN produces a menu of possible actions) over
which the user has a large degree of control (type in the PIN
when ready, select the desired option, etc.).
An important outcome of human performance based on lin-
ear, predictable, and controllable procedures is that it becomes
largely automatic (i.e., it is cheap in cognitive resources) and rel-
atively reliable. But it also leads to the generation of, and subse-
quent reliance on, triggers. Repeated performance of activities in
a particular sequence builds strong links among actions, so that
certain, particularly salient events or actions become triggers
that automatically (i.e., with minimal conscious, controlled effort)
elicit the next step in sequence. An expert money-withdrawer
at the ATM can select from the available menu options without
actually reading them on the screen. When the ATM returns the
card it typically generates an aural signal that automatically
triggers the withdrawer’s action to retrieve the card.
Repeated performance of procedures in the cockpit likewise
generates triggers and leads pilots to rely upon them. Interaction
294 Immanuel Barshi and Loukia Loukopoulos
with an ATM is simple, and matters are obviously more complex
when it comes to interacting with an aircraft. Training developers
and instructors may not expect FOM-prescribed activities to
accurately represent everything about the real world, or the
FOM-based training to completely prepare pilots for the real
operational environment. Pilots are expected to learn to deal
with the real world through exposure and experience. But that
doesn’t always work well, so understanding the differences
between the real operational world and the theoretical world of
the FOM is critical for the design of effective training.
Performing (Reality)
One way to truly understand how procedures are performed in
the real world is to actually observe them in real time. A few
years ago, the authors occupied the spare seat (jump seat) in
cockpits, behind the pilots, while passengers (and the authors)
were being whisked to destinations around the country, and
gained a unique perspective on procedures in real life. This
observational study and its findings are discussed in great
detail in their book, The Multitasking Myth (Loukopoulos, Dis-
mukes, & Barshi, 2009). Certain findings, however, have par-
ticular implications for the design of training and for the basic
research upon which such design depends.
One of the findings confirms that expert, professional, exten-
sively trained pilots indeed perform what they have been trained
to do: They follow detailed, exact sequences of complex activi-
ties, working alone, and coming together in a prescribed, care-
fully orchestrated and executed manner. Procedures generally
serve their purpose well.
Another finding concerns the presence of monitoring require-
ments, neither explicitly described by procedures nor specifi-
cally covered in training. Pilots navigate across busy airports,
watching out for other aircraft, ground servicing equipment and
vehicles, maintaining appropriate distance to avoid collisions,
remaining aware of their position along the taxi route, while
listening to communications with ATC so as to immediately pick
out the instructions issued to their aircraft (and infer the situa-
tion around by listening to instructions to other aircraft). Pilots
undertake this “extra” workload seemingly effortlessly. They
simply interleave the monitoring demands with everything else
they are doing and largely depend on skills that enable them to
work well in unison and to back up one another.
Yet another finding concerns the presence of operational fac-
tors requiring on-the-spot solutions that are also not explicitly
Training for Real-World Job Performance 295
(if at all) described by procedures nor introduced in training.
For example, ice and snow on the taxiways may necessitate
(for technical reasons) postponing setting the flaps until right
before takeoff. Radio calls often interrupt the pilots in the mid-
dle of conducting checklists, and the calls must be responded to
before resuming the interrupted checklist. Wind changes cause
controllers to issue changes to the takeoff runway, making
pilots temporarily unavailable to monitor one another or con-
duct checklists together because they are unexpectedly busy
recalculating takeoff data.
Such instances (and the many more that were observed in the
course of this study) lead to the conclusion that activities in the
real operational world are not as straightforward as procedures
portray. The prevailing feature appears to be a flurry of events
that occurs during the course of a flight, at the same time that
pilots are busy carrying out normal procedures per the FOM.
Such events arise from operating within the larger context of
busy airports, changing weather patterns, and traffic-ridden air
space, in association with scores of other professional groups
(e.g., dispatchers, ground services, air traffic controllers) and
under the heavy influence of the operational culture of the air-
line and the economic constraints of the industry. Their pres-
ence “perturbs” the nominal, anticipated sequence of normal
activities prescribed by procedures, and creates a dynamic,
multidimensional (in terms of the many, often-conflicting priori-
ties) environment, in which pilots are called to make on-the-spot
decisions about interleaving, postponing and resuming, and
combining the accomplishment of various activities.
Perturbations do not qualify as “emergencies” because they
do not require pilots to perform radically different steps from
those prescribed by the normal procedures for which they have
been trained. Instead, pilots must execute those procedures
with small variations in the sequence of some activities, a cer-
tain degree of resourcefulness in the timing of other activities,
and, on occasion, simultaneously with other activities. Here,
then is a true picture of the real operational world. Activities
are not linear, but dynamic. The checklist is typically called at
the conclusion of the preceding flow, but if ATC has issued new
instructions, the crew must first deal with them before the cap-
tain can call for the checklist. Activities are semi-predictable;
that is, many perturbations are unpredictable (such as a prob-
lem encountered by another aircraft that delays all movements
on the taxiway), and some can be anticipated though their spe-
cific timing is unpredictable (such as when a takeoff clearance
is issued much sooner than expected). Activities, finally, are
296 Immanuel Barshi and Loukia Loukopoulos
semi-controllable. The pilots may want to call for a clearance, but
the radio frequency may be congested with exchanges between
the controller and other crews, so the pilots have to wait before
they can make their call.
The question then becomes whether it matters that the real
world is drastically different from the theoretical world implied
by the FOM and for which procedures are written? Does it make
any difference for the design of training? More importantly, does
it make any difference for how the research that guides the
design of training is structured?
The short answer is that it does matter. According to the Pro-
cedural Reinstatement Principle (Healy et al., 2005), for train-
ing to be effective, it must reinstate the conditions in which the
skills learned will actually be put to use. The long answer is
that it matters a lot, because these features of the environment
that should be reinstated pose a safety risk given current pro-
cedures. This risk is evidenced in the way pilots’ responses to
operational perturbations interact with cognitive processes.
Multitasking
The issue with the real operational environment is not just one
of workload: perturbations do not simply increase its volume
(number of tasks), they increase its complexity. It’s not that
pilots are unable to cope with this complexity. On the contrary,
they are so used to perturbations (and so capable of incorporat-
ing them into their routine), that they consider them “business
as usual.” Recall, however, the initial discussion of the puzzling
occurrences of expert pilots forgetting to accomplish critical
tasks explicitly prescribed by procedures they had been trained
to perform, and had considerable experience performing in the
real world. These occurrences can be understood by examining
the link between the complexity of the real operational world
and inadvertent omissions (for an extensive discussion, see Lou-
kopoulos et al., 2009).
Perturbations give rise to interruptions and distractions, the
need for tasks to be executed in other than the normal, prac-
ticed sequence, unanticipated new tasks, the need to interleave
multiple tasks, and, of course, any combination of the above! In
doing so, they force pilots to engage in “multitasking”—a situa-
tion of doing several things at the same time. Multitasking with-
out negative effects on performance is not a human strength
because individuals are only able to focus their attention on
one thing at a time (Basak & Verhaeghen, 2011). Multitasking
therefore involves task switching, rather than parallel execution
Training for Real-World Job Performance 297
of tasks. In turn, this means that the degree to which it can
be done successfully depends on many factors, such as the
number, nature, prior experience, and extent of practice of the
particular combination of tasks. When it involves familiar, well-
practiced activities that can be executed with little conscious
attention, multitasking is possible with a low risk for errors, as
can be attested by the pilot who taxis the aircraft and still moni-
tors the radio frequency for calls addressed to him. Throw in
a third, perhaps less familiar, more “involved,” or unexpected
activity such as having to make mental calculations and input
data into the computer, and the risk of errors increases because
task-switching has costs (Rubinstein, Meyer, & Evans, 2001). In
cases such as those encountered in real flight operations, where
multitasking typically involves a mixture of habitual and novel
activities, it proves to be particularly risky because it creates
situations that push the capacity of human memory and atten-
tion to the limits, rendering any pilot vulnerable to errors.
It is now possible to appreciate that behind each reported fail-
ure lie none other than perturbations of the routine flight opera-
tional environment; that is, events that gave rise to the need for
multitasking. A flight crew reported aborting the takeoff after
the configuration warning horn sounded to indicate that flaps
were not in their proper position. The reason? The crew, fol-
lowing instructions from the ground controller, had rushed to
vacate the gate area for another aircraft that had just arrived
and was assigned to the same gate. In rushing to release the
parking brake and get moving, the crew inadvertently forgot
to first set the flaps. The perturbing event (the other aircraft)
and the sense of time pressure it generated effectively masked
the habitual trigger for checking the flaps before releasing the
parking brake. A few moments later, when the aircraft was mov-
ing, the implicit assumption was generated that all preceding
activities had been completed, as usual. Operators everywhere
are vulnerable to attempt to respond to interruptions that arise
from various operational demands, oftentimes without realizing
the potential risks involved.
What about the crew that inadvertently struck a ground truck
after failing to obtain the expected signal from the truck driver?
The pilots were busy discussing the taxi instructions. Becoming
engaged and preoccupied by important and salient operational
issues can sidetrack performance, and allow the brain to jump
forward in the sequence of habitual activities and make assump-
tions that certain activities have already happened (on occasion,
having the strong, though wrong, feeling that they have actu-
ally taken place). False memories are a particularly well-known
298 Immanuel Barshi and Loukia Loukopoulos
and studied phenomenon in the retrieval of declarative informa-
tion (e.g., eyewitness testimonies) and one of the mechanisms
giving rise to such confusion (i.e., source confusion, Johnson,
Hashtroudi, & Lindsay, 1993) may well also be behind pilots’
mistaken assumption that an action has already been accom-
plished. However, false memory for procedural knowledge has
rarely been studied in the laboratory, and remains a challenge
to the research community.
How about the first officer who failed to adequately monitor the
captain, because she was busy reprogramming the computer in
response to just-issued changes, thus not paying enough atten-
tion to prevent the captain from inadvertently taxiing past an
intended turn? A change in runways issued while the aircraft
is already taxiing is a fairly common event. The first officer was
called to multitask, and was set up in a particularly challenging
situation combining an attention-absorbing activity (program-
ming) entailing physical direction of the body/head/eyes away
from the scene outside, with the notoriously nonsalient activity
of monitoring (especially another person’s actions). Monitoring,
even by itself, requires deliberate effort and, when combined
with another task, falls easily out of the scope of attention. Mon-
itoring is at high risk of being “dropped” in the face of perturba-
tions, both because of its nature (it has few salient cues) and
because it is often not considered a task in itself, but rather a
background activity.
The sources of reported and observed perturbations are truly
variable, their effects as well. No task seems immune to the risk
of being omitted, regardless of its criticality. Perturbations are
clearly related to errors and their effects in rendering the real
operational environment nonlinear, only semi-predictable, and
only semi-controllable must be factored into training to make it
effective.
Other Domains
Omissions and other types of human performance errors are
not exclusive to the cockpit. Multitasking and operational
perturbations are features of many professions such as air traffic
control (Loukopoulos, 2010), train dispatching (Department
of Transportation/Federal Rail Administration [USDOT/FRA],
2001), nursing (Kalisch & Aebersold, 2010), even working in
space (Burgess, 2000). A number of studies discuss operator
errors, including omissions, in a wide range of activities including
aircraft maintenance (Hobbs & Williamson, 2003), nuclear power
plants (Alder & Hausmann, 1995; Hirotsu, Suzuki, Kojima,
Training for Real-World Job Performance 299
& Takano, 2001), the pipeline industry (American Petroleum
Institute [API], 2005), administering medication (McBride-Henry
& Foureur, 2006), and identifying the correct site for surgery
(Kwaan, Studdert, Zinner, & Gawande, 2006). Naturally, there
are large differences in the degree to which activities in each
of these work settings are dictated by precise procedures, the
manner with which such procedures are trained and carried
out, and the types and effectiveness of safeguards in place to
guard against unintended deviations from such procedures. The
brief descriptions below illustrate how the dynamic, multitasking
features of these real operational environments, as in aviation,
present a risk to human performance and therefore need to be
studied and reinstated in training.
References
Alder, H. P., & Hausmann, W. (1995). Analysis of human factors in inci-
dents reported by Swiss nuclear power plants to the Inspectorate.
In International Atomic Energy Agency (IAEA), Organizational factors
influencing human performance in nuclear power plants (pp. 115–
Training for Real-World Job Performance 305
119; IAEA-TECDOC-943). Vienna, Austria: International Atomic
Energy Agency.
American Petroleum Institute (API). (2005). Facilities piping and equip-
ment: Focus on items involved and causes of incidents (PPTS Opera-
tor Advisory 2005-4). Retrieved from http://committees.api.org/
pipeline/ppts/docs/advisories/20054advisoryFacilitiessubset.pdf
Barshi, I., & Healy, A. F. (1993). Checklist procedures and the cost of
automaticity. Memory & Cognition, 21, 496–505.
Basak, C., & Verhaeghen, P. (2011) Three layers of working memory:
Focus-switch costs and retrieval dynamics as revealed by the
N-count task. Journal of Cognitive Psychology, 23, 204–219.
Bundesstelle für Flugunfalluntersuchung (BFU). (2004). Accident near
Überlingen/Lake of Constance/Germany, 1 July 2002 (Investigation
Report AX001-1-2/02). Retrieved from http://ocw.mit.edu/courses/
aeronautics-and-astronautics/16-358j-system-safety-spring-2005/
assignments/ueberlingen.pdf
Burgess, P. W. (2000). Real-world multitasking from a cognitive neuro-
science perspective. In S. Monsell & J. Driver (Eds.), Control of cogni-
tive processes (pp. 465–472). Cambridge, MA: MIT Press.
Confidential Human Factors Incident Reporting Program—Mainte-
nance Error Management System (CHIRP-MEMS). (n.d.). Learning
from experience: Report 2—Torch left in nosewheel steering cable
run. Retrieved from http://www.chirp-mems.co.uk/experiance.asp
Comisión de Investigación de Accidentes e Incidentes de Aviación Civil
(CIAIAC). (2008). Preliminary report A-32/2008. Retrieved from
http://www.fomento.es/NR/rdonlyres/C58972BC-B96C-4E14-
B047-71B89DD0173E/76728/PreliminaryReportA_032_2009.pdf
Dismukes, R.K., Berman, B., & Loukopoulos, L.D. (2007). The limits of
expertise: Rethinking pilot error and the causes of airline accidents.
Aldershot, England: Ashgate.
Gamache, P-L., Hudon, C., Teasdale, N., & Simoneau, M. (2010). Alter-
native avenues in the assessment of driving capacities in older driv-
ers and implications for training. Current Directions in Psychological
Science, 19, 370–374.
Healy, A. F., Wohldmann, E. L., & Bourne, L. E., Jr. (2005). The pro-
cedural reinstatement principle: Studies on training, retention, and
transfer. In A. F. Healy (Ed.), Experimental Cognitive Psychology and
Its Applications (pp. 59–71). Washington, DC: American Psychologi-
cal Association.
Hirotsu, Y., Suzuki, K., Kojima, M., & Takano, K. (2001). Multivari-
ate analysis of human error incidents occurring at nuclear power
plants: Several occurrence patterns of observed human errors. Cog-
nition, Technology, and Work, 3, 82–91.
Hobbs, A. N., & Williamson, A. (2003). Associations between errors
and contributing factors in aircraft maintenance. Human Factors,
45, 186–201.
Johnson, M. K., Hashtroudi, S., & Lindsay, D.S. (1993). Source moni-
toring. Psychological Bulletin, 114, 3–28.
306 Immanuel Barshi and Loukia Loukopoulos
Kalisch, B., & Aebersold, R. N. (2010). Interruptions and multitask-
ing in nursing care. The Joint Commission Journal on Quality and
Patient Safety, 36(3), 126–132.
Kwaan, M. R., Studdert, D. M., Zinner, M. J., & Gawande, A. A. (2006).
Incidence, patterns, and prevention of wrong-site surgery. Archives
of Surgery, 141, 353–358.
Loukopoulos, L. D. (2010). Air traffic controllers do it too! Hindsight: A
Eurocontrol Journal, 10, 48–51.
Loukopoulos, L. D., Dismukes, R. K., & Barshi, I. (2009). The multi-
tasking myth: Handling complexity in real-world operations. Alder-
shot, England: Ashgate.
McBride-Henry, K., & Foureur, M. (2006). Medication administration
errors: Understanding the issues. Australian Journal of Advanced
Nursing, 23(3), 33–41.
National Aeronautics and Space Administration, Aviation Safety
Reporting System. (2005). Retrieved November 22, 2010, from http://
asrs.arc.nasa.gov/search/database.html
National Transportation Safety Board (NTSB). (1988). Northwest Air-
lines, Inc., McDonnell Douglas DC-9-82, N312RC, Detroit Met-
ropolitan Wayne County Airport, Romulus, Michigan, August 16,
1987 (Report no. PB88-910406, NTSB/ AAR-88-05). Retrieved from
http://libraryonline.erau.edu/online-full-text/ntsb/aircraft-acci-
dent-reports/AAR88-05.pdf
National Transportation Safety Board (NTSB). (1989). Delta Airlines,
Inc., Boeing 727-232, N473DA, Dallas-Fort Worth International
Airport, Texas, August 31, 1988 (Report no. PB89-910406, NTSB/
AAR-89-04). Retrieved from http://libraryonline.erau.edu/online-
full-text/ntsb/aircraft-accident-reports/AAR89-04.pdf
National Transportation Safety Board (NTSB). (2003). Collision of two
Burlington Northern Santa Fe freight trains near Clarendon, Texas,
May 28, 2002 (Report No. PB2003-916301, NTSB/RAR-03/01).
Retrieved from http://www.ntsb.gov/publictn/2003/RAR0301.pdf
Ring, D. C., Herndon, J. H., & Meyer, G. S. (2010). Case 34-2010—A
65-year-old woman with an incorrect operation on the left hand.
New England Journal of Medicine, 363, 1950–1957. Retrieved from
http://www.nejm.org/doi/full/10.1056/NEJMcpc1007085#t=article
Rubinstein, J. S., Meyer, D. E., & Evans, J. E. (2001). Executive con-
trol of cognitive processes in task switching. Journal of Experimental
Psychology: Human Perception and Performance, 27(4), 763–797.
U.S. Pharmacopeia (USP). (2003, September). Distractions contribute
to medication errors. USP Patient Safety CAPSLink. Retrieved from
http://www.usp.org/pdf/EN/patientSafety/capsLink2003-09-01.
pdf
U.S. Department of Transport, Federal Rail Administration (DOT/FRA/
ORD-01/02). (2001). Understanding how train dispatchers manage
and control trains: Results of a cognitive task analysis. Retrieved
from http://www.fra.dot.gov/downloads/Research/ord0102.pdf
15 Cognitive Retraining
Following Acquired Brain
Injury
Keith R. Lohse and Lyle E. Bourne, Jr.
University of Colorado
References
Adekoya, N., Thurman, D. J., White, D. D., & Webb, K. W. (2002). Sur-
veillance for traumatic brain injury deaths—United States, 1989–
1998. MMWR Surveillance Summaries, 51, 1–14.
Armstrong, B. D., Hu, Z., Abad, C., Yamamoto, M., Rodrigeuz, W. I.,
Cheng, J. T., … Waschek, J. A. (2003). Lymphocyte regulation of
neuropeptide gene expression after neuronal injury. Journal of Neu-
roscience Research, 74, 240–247.
Basso, A., & Caporali, A. (2001). Aphasia therapy, or the importance of
being earnest. Aphasiology, 15, 307–332.
Burke, J. W., McNeill, M. D. J., Charles, D. K., Morrow, P. J., Cros-
bie, J. H., & McDonough, S. M. (2009). Optimising engagement for
stroke rehabilitation using serious games. The Visual Computer, 25,
1085–1099.
Cappa, S. F., Benke, T., Clarke, S., Rossi, B., Stemmer, B., & van
Heug ten, C. M. (2003). EFNS guidelines on cognitive rehabilitation:
Report of an EFNS task force. European Journal of Neurology, 10,
11–23.
322 Keith R. Lohse and Lyle E. Bourne, Jr.
Carr, T. H., & Levy, B. A. (1990). Reading and its development: Compo-
nent skills approaches. Orlando, FL: Academic Press.
Cepeda, N. J., Pashler, H., Vul, E., Wixted J. T., & Rohrer, D. (2006).
Distributed practice in verbal recall tasks: A review and quantitative
synthesis. Psychological Bulletin, 123, 354–380.
Cicerone, K. D., Dahlberg, C., Kalmar, K., Langenbahn, D. M., Malec,
J. F., Bergquist, T. F., … Morse, P. A. (2000). Evidence based cogni-
tive rehabilitation: Recommendations for clinical practice. Archives
of Physical Medicine & Rehabilitation, 81, 1596–1615.
Cicerone, K. D., Dahlberg, C., Malec, J. F., Langenbahn, D. M., Felicetti,
T., Kneipp, S., & Catanese, J. (2005). Evidence based cognitive reha-
bilitation: Updated review of the literature from 1998 through 2002.
Archives of Physical Medicine and Rehabilitation, 86, 1681–1692.
de Kruijk, J. R., Leffers, P., Meerhoff, S., Rutten, J., & Twijnstra, A.
(2002). Effectiveness of bed rest after mild traumatic brain injury: A
randomised trial of no versus six days of bed rest. Journal of Neurol-
ogy, Neurosurgery, and Psychiatry, 73, 167–172.
Denes, G., Perazzolo, C., Piani, A., & Piccione, F. (1996). Intensive ver-
sus regular speech therapy in global aphasia: A controlled study.
Aphasiology, 10, 385–394.
Duncan, J. (1986). Disorganization of behavior after frontal lobe dam-
age. Cognitive Neuropsychology, 3, 271–290.
Ellul, J., Watkins, C., Ferguson, N., Barer, D., & Rowe, J. (1993).
Increasing patient engagement in rehabilitation activities. Clinical
Rehabilitation, 7, 297–302.
Elman, R. J., & Bernstein-Ellis, E. (1999). The efficacy of group com-
munication treatment in adults with chronic aphasia. Journal of
Speech, Language, and Hearing Research, 42, 411–419.
Emery, D. L., Royo, N. C., Fischer, I., Saatman, K. E., McIntosh, T. K.
(2003). Plasticity following injury to the adult central nervous sys-
tem: Is recapitulation of a developmental state worth promoting?
Journal of Neurotrauma, 20, 1271–1292.
Fasoli, S. E., Trombly, C. A., Tickle-Degnen, L., & Verfaellie, M. H.
(2002). Effect of instructions on functional reach in persons with
and without cerebrovascular accident. American Journal of Occupa-
tional Therapy, 56, 380–390.
Fasotti, L., Kovacs, F., Eling, P. A., & Brouwer, W. H. (2000). Time pres-
sure management as a compensatory strategy training after closed
head injury. Neuropsychological Rehabilitation, 10, 47–65.
Fisk, A. D., & Eboch, M. (1989). An automatic/controlled process-
ing theory application to training component map reading skills.
Applied Ergonomics, 20, 2–8.
Glisky, E. L. (1995). Acquisition and transfer of word processing skill by
an amnesic patient. Neuropsychological Rehabilitation, 5, 299–318.
Gordon, W. A. (1990). Cognitive remediation: An approach to the ame-
lioration of behavioral disorders. In R. L. Wood (Ed.), Neurobehav-
ioural sequelae of traumatic brain injury (pp. 175–193). London:
Taylor & Francis.
Cognitive Retraining Following Acquired Brain Injury 323
Hillis, A. E. (1998). Treatment of naming disorders: New issues regard-
ing old therapies. Journal of the International Neuropsychological
Society, 4, 648–660.
Hinckley, J. J., & Carr, T. H. (2005). Comparing the outcomes of inten-
sive and non-intensive context-based aphasia treatment. Aphasiol-
ogy, 19, 965–974.
Hinckley, J. J., & Craig, H. K. (1998). Influence of rate of treatment on
the naming abilities of adults with chronic aphasia. Aphasiology,
12, 989–1006.
Hinckley, J. J., Patterson, J. P., & Carr, T. H. (2001). Differential effects
of context and skill-based treatment approaches: Preliminary find-
ings. Aphasiology, 15, 463–476
Jacobs, H. E. (1987). The Los Angeles head injury survey: Project ratio-
nale and design implications. Journal of Head Trauma Rehabilita-
tion, 2, 37–50.
Joshi, J., Ornstein, E., & Young, W. L. (2001). Cerebral and spinal cord
blood flow. In J. E. Cottrell & D. C. Smith (Eds.), Anesthesia and
neurosurgery (4th ed.). St Louis, MO: Mosby.
Kaschel, R., Della Sala, S., Cantagallo, A., Fahlböck, A., Laaksonen,
R., & Kazen, M. (2002). Imagery mnemonics for the rehabilitation
of memory: A randomised group controlled trial. Neuropsychological
Rehabilitation, 12, 127–153.
Kasten, E., Wüst, S., Behrens-Baumann, W., & Sabel, B. A. (1998).
Computer-based training for treatment of partial blindness. Nature
Medicine, 4, 1083–1087.
Landers, M., Wulf, G., Wallman, H., & Guadagnoli, M. A. (2005). An
external focus of attention attenuates balance impairment in Par-
kinson’s disease. Physiotherapy, 91, 152–185.
Langlois, J. A., Rutland-Brown, W., & Wald, M. M. (2006). The epidemi-
ology and impact of traumatic brain injury: A brief overview. Journal
of Head Trauma Rehabilitation, 21, 375–378.
Laufer, Y., Rotem-Lehrer, N., Ronen, Z., Khayutin, G., & Rozenberg, I.
(2007). Effect of attention focus on acquisition and retention of pos-
tural control following ankle sprain. Archives of Physical Medicine
and Rehabilitation, 88, 105–108.
Levine, B., Robertson, I. H., Clare, L., Carter, G., Hong, J., Wilson, B.
A., … Stuss, D. T. (2000). Rehabilitation of executive functioning:
An experimental-clinical validation of Goal Management Training.
Journal of the International Neuropsychological Society, 6, 299–312.
Livingston, M. G., & Brooks, D. N. (1988). The burden on families of
the brain-injured: A review. Journal of Head Trauma Rehabilitation,
3, 6–15.
Manly, T., Hawkins, K., Evans, J., Woldt, K., & Robertson, I. (2002).
Rehabilitation of executive function: Facilitation of effective goal
management on complex tasks using periodic auditory alerts. Neu-
ropsychologia, 40, 271–281.
McNevin, N. H., Wulf, G., & Carlson, C. (2000). Effects of attentional
focus, self-control, and dyad training on motor learning: Implica-
tions for physical rehabilitation. Physical Therapy, 80, 373–385.
324 Keith R. Lohse and Lyle E. Bourne, Jr.
Medd, J., & Tate, R. L. (2000). Evaluation of an anger management
therapy programme following acquired brain injury: A preliminary
study. Neuropsychological Rehabilitation, 10, 185–201.
Mesulam, M. M. (Ed.). (1985). Principles of behavioral neurology. Phila-
delphia, PA: F. A. Davis.
Miller, L. (1990). Major syndromes of aggressive behaviour following
head injury: An introduction to evaluation and treatment. Cognitive
Rehabilitation, 8, 14–19.
Neimann, H., Ruff, R. M., & Baser, C. A. (1990). Computer-assisted
attention retraining in head-injured individuals: A controlled effi-
cacy study of an outpatient program. Journal of Consulting and Clin-
ical Psychology, 58, 811–817.
Neimeier, J. P. (1998). The Lighthouse Strategy: Use of a visual imagery
technique to treat visual inattention in stroke patients. Brain Injury,
12, 399–406.
Nortje, J., & Menon, D. K. (2004). Traumatic brain injury: Physiol-
ogy, mechanisms, and outcome. Current Opinion in Neurology, 17,
711–718.
Ownsworth, T. L., & MacFarland, K. (1999). Memory remediation in
long-term acquired brain injury: Two approaches in diary training.
Brain Injury, 13, 605–626.
Parikh, S., Koch, M., & Narayan, R. K. (2007). Traumatic brain injury.
International Anesthesiology Clinics, 45, 119–135.
Pulvermuller, F., Neininger, B., Elbert, T., Mohr, B., Rockstroh, B., Koe-
bbel, P., & Taub, E. (2001). Constraint-induced therapy of chronic
aphasia after stroke. Stroke, 32, 1621–1626.
Robertson, I. H. (1996). Goal management training: A clinical manual.
Cambridge, England: PsyConsult.
Robey, R. R. (1994). The efficacy of treatment for aphasic persons: A
meta-analysis. Brain and Language, 47, 585–608.
Robey, R. R. (1998). A meta-analysis of clinical outcomes in the
treatment of aphasia. Journal of Speech, Language, and Hearing
Research, 41, 172–187.
Rutland-Brown, W., Langlois, J. A., Thomas, K. E., & Xi, Y. L. (2006).
Incidence of traumatic brain injury in the United States, 2003. Jour-
nal of Head Trauma Rehabilitation, 21, 544–548.
Ryan, L. M., & Warden, D. L. (2003). Post concussion syndrome. Inter-
national Review of Psychiatry, 15, 310–316.
Schmidt, R. A., & Bjork, R. A. (1992). New conceptualizations of prac-
tice: Common principles in three paradigms suggest new concepts
for training. Psychological Science, 3, 207–217.
Sohlberg, M. M., McLaughlin, K. A., Pavese, A., Heidrich, A., & Pos-
ner, M. I. (2000). Evaluation of attention process training and brain
injury education in persons with acquired brain injury. Journal of
Clinical & Experimental Neuropsychology, 22, 656–76.
Tanielian T., & Jaycox, L. H., (Eds.). (2008). Invisible wounds of war:
Psychological and cognitive injuries, their consequences, and ser-
vices to assist recovery. Santa Monica, CA: RAND.
Cognitive Retraining Following Acquired Brain Injury 325
Teasdale, B., & Jennett, B. (1974). Assessment of coma and impaired
consciousness: A practical scale. Lancet, 2, 81–84.
Warden, M. D. (2006). Military TBI during the Iraq and Afghanistan
Wars. Journal of Head Trauma Rehabilitation, 21, 398–402.
Westerberg, H., Jacobaeus, H., Hirvikoski, T., Clevberger, P., Östens-
son, M. L., Bartfai, A., & Klingberg, T. (2007). Computerized working
memory training after stroke—A pilot study. Brain Injury, 21, 21–29.
Wilson, B. A. (1995). Memory rehabilitation: Compensating for memory
problems. In L. Bäckman & R. Dixon (Eds.), Psychological compen-
sation (pp. 171–190). Hillsdale, NJ: Erlbaum.
Wilson, B. A., Emslie, H. C., Quirk, K., & Evans, J. J. (2001). Reduc-
ing everyday memory and planning problems by means of a paging
system: A randomized control crossover study. Journal of Neurology,
Neurosurgery, & Psychiatry, 70, 477–482.
Wulf, G., Landers, M., Lewthwaite, R., & Töllner, T. (2009). External
focus instructions reduce postural instability in individuals with
Parkinson disease. Physical Therapy, 89, 162–168
Ylvisaker, M., Szekeres, S. F., Henry, K., Sullivan, D. M., & Wheeler, P.
(1987). Topics in cognitive rehabilitation theory. In M. Ylvisaker &
E. M. Gobble (Eds.), Community re-entry for head injured adults (pp.
137–220). Boston, MA: College-Hill Press.
16 Conclusions
Lyle E. Bourne, Jr. and Alice F. Healy
University of Colorado
Important Distinctions
This volume starts by making a set of important distinctions
that apply to training research, theory, and application: (a)
training and education; (b) acquisition, retention, and transfer;
(c) declarative and procedural information; and (d) principles,
guidelines, and specifications. The book focuses on training,
in contrast to education. Training is the attempt to develop
knowledge and skills particular to a well-defined task or job.
Education, in contrast, focuses on the acquisition of general
knowledge and skills pertaining to a relatively broad area or
domain, not necessarily limited to a well-defined task. In training,
the fundamental underlying cognitive processes are acquisition
(the initial learning), retention (the durability over time), and
transfer (the flexibility or generalizability) of knowledge and
skills. The emphasis on training for particular tasks requires
differentiating declarative information (facts, knowing that) and
procedural information (skills, knowing how). Finally, when it
comes to utilizing the body of scientific knowledge on training,
there is a logical progression from basic science to application.
This progression has been described as beginning with training
principles (which have been the primary focus of the basic
Conclusions 327
research), to developing guidelines for the transition of those
principles to real-life situations, and finally to specifications for
actual training practices in a given context. These distinctions
appear in various forms throughout the chapters of the book.
Chapter Synopses
The major points in each chapter after the first, which sets out
the major distinctions reviewed above, are summarized here,
including their take-home messages with respect to training.
Chapter 2 reviews the existing cognitive psychological litera-
ture supporting the conclusion that certain principles of train-
ing have strong empirical validity at least in the context of the
laboratory. The effects of these principles have been shown in
all three cognitive component processes of training (acquisition,
retention, and transfer), although the effects are not necessarily
the same on each component. This review points to some pos-
sible applications of the laboratory findings to a variety of train-
ing situations.
Expanding on the review of the literature, chapter 3 reports a
series of recent experiments, some unpublished, that examined
four types of questions left unanswered by the literature. These
experiments revealed the generality of certain training princi-
ples and the limitation of others; illuminated the way in which
principles interact in combination; assessed the applicability of
laboratory based principles to complex, dynamic environments;
and identified novel principles not previously established by the
literature.
Chapter 4 describes the implications of cognitive load theory
for the effectiveness of different attention-related training strate-
gies by analyzing the limited cognitive resources of the learner
as allocated to the intrinsic load of the task to be performed,
the germane load necessary for skill acquisition, and the extra-
neous load in the training environment that competes with
the other two sources of load. Data from a meta-analysis are
reported to support the effectiveness and influence of moderat-
ing variables on the attention-related strategies of active choice,
increasing difficulty training, error prevention, and part task
training, including the training of visual attention. Implications
for training environments are discussed.
Factors that affect the acquisition and transfer of rudimen-
tary components of skill are the focus of chapter 5. Much of
the research described uses basic choice reaction tasks, which
permit isolation of fundamental cognitive processes and rapid
acquisition of skill. The strategy of this research approach is to
328 Lyle E. Bourne, Jr. and Alice F. Healy
isolate and reveal components of skill acquisition and transfer
by examining influences of task parameters on the preexisting
bias to make spatially corresponding responses.
The use of automation as a training aide is explored in chap-
ter 6. Working from an aptitude-treatment interaction perspec-
tive, the results are reported from three studies examining how
different instantiations of automation as a training aide interact
with the learner’s level of cognitive ability to influence learn-
ing and subsequent task performance. Building on these results
and extant theory and research in the automation, training,
and cognitive abilities literatures, a framework is proposed for
understanding how, when, and for whom automation can have a
positive influence on learning, retention, and transfer.
Technology-based training research, as reviewed in chapter
7, seeks to develop findings that reduce risk in training sys-
tems’ design and incorporate features supporting learning and
performance. Applied research projects conducted by the Army
Research Institute in a wide range of technology-based training
applications are described. Taken together they demonstrate the
strengths and weaknesses of applied training research.
Chapter 8 outlines the development and current form of a
four-dimensional taxonomy with which to categorize training
variables according to (a) task type, (b) training methodology,
(c) performance measures, and (d) training principles. This tax-
onomy should facilitate predictions about performance in given
training situations based on their locations in the taxonomic
space. Taxonomic analysis should help guide future research
into areas not yet fully explored in the taxonomic space and
should aid in the optimization of current training regimens.
Computational models of human behavior for three train-
ing principles are studied in chapter 9: (a) speed-accuracy
tradeoff attributable to fatigue, (b) training difficulty, and (c)
stimulus-response compatibility. These studies show that the
ACT-R architecture and IBL theory present an accurate and
robust representation of the learning process in several train-
ing paradigms. The creation of a computer tool that represents
IBL theory gives rise to interesting new demonstrations of other
learning and training principles.
Chapter 10 reviews three models of basic psychological tasks,
involving data entry, target detection, and information integra-
tion. IMPRINT was the modeling platform used in each case.
Although in the past IMPRINT has been used for larger scale
models of military scenarios, the present study demonstrates
that it is also capable of modeling cognitive tasks at the level of
component processes.
Conclusions 329
Chapter 11 describes a procedure used to compare and evalu-
ate models of cognitive training, which had been created using
both the ACT-R and IMPRINT modeling platforms, finding that
the models were equally accurate and relatively slow. Because
the IMPRINT models are equation based, they could be trans-
lated into equivalent models in Matlab, which had simulation
times that were orders of magnitude shorter than those of the
original models. This translation along with the functional-
ity available in Matlab enabled rapid optimization of multiple
parameter values and visualization of the high-dimensional
parameter space.
Predicting performance can be important for implementation
and design of alternative training scenarios. Chapter 12 pro-
poses a compact mathematical model, founded on existing cog-
nitive theory, that can generate predictions for test performance
as a function of training history on various tasks. The model
incorporates many of the training principles identified in other
chapters of this volume.
Chapter 13 addresses the challenge of training knowledge
so that it is retained and accessed for application to particular
jobs or tasks. Evidence is reviewed from laboratory experiments
and educational research indicating that SPacing, Retrieval
practice, and INTerleaving (SPRINT) can improve the retention
and transfer of trained knowledge. The chapter illustrates how
education-based practices might be modified to improve knowl-
edge retention in knowledge-training programs, and in turn, to
improve the success of these knowledge-training programs.
To make applications to the real world possible, it is neces-
sary first to understand the “real world.” Chapter 14 describes
the reality of complex operations, particularly those conducted
by airline pilots, and discusses the implications of such reality
for research on training. A critical issue in this discussion is
the demands of concurrent task management in many every-
day environments, such as distracted driving, and the ways in
which training research must address such demands.
Acquired brain injury (ABI) is one of the most common causes
of death and disability in the United States and around the
world. Because the nature of ABI and the resulting psychologi-
cal deficits can be so diverse, unique rehabilitation techniques
are being developed to retrain cognitive skills and restore inde-
pendence in patients with ABI. Chapter 15 compares training
methods found in rehabilitation practice with the training prin-
ciples of psychological science detailed in the rest of this volume.
There are many instances of overlap between rehabilitation
practices and empirical principles, but recommendations are
330 Lyle E. Bourne, Jr. and Alice F. Healy
made for future directions to be explored in order to optimize
rehabilitation science.
Chapter Clusters
The chapters in the volume fall into three logical clusters. The
first is a collection of chapters dealing with basic scientific
knowledge about training derived primarily from the cognitive
psychology laboratory. This cluster includes chapter 2 (reviews
the existing experimental literature on training principles);
chapter 3 (presents new experiments identifying new training
principles, refining existing principles, and extending the range
of applicability of those principles); chapter 4 (provides a frame-
work for studying the role of attention in training); chapter 5
(describes experiments on the rudimentary components of skill);
and chapter 6 (examines the possible use of automation in train-
ing and the impact of individual differences on training). These
chapters complement each other to provide an all-encompassing
overview of empirically based principles of the training process.
The second cluster consists of a set of chapters aimed at devel-
oping a theoretical overview of the training process. The chap-
ters include chapter 8 (presents a four-dimensional taxonomic
analysis of tasks, methods, measures, and principles, which
maps out the general domain of training); chapters 9 and 10
(provide specific, detailed models of training in particular tasks,
using ACT-R and IMPRINT as modeling platforms and yielding
precise quantitative predictions of the effects of selected train-
ing principles in those tasks); chapter 11 (compares the models
described in chapters 9 and 10 as to their accuracy and speed
of execution); and chapter 12 (develops a high-level formulation
of training that captures the three fundamental cognitive pro-
cesses and the generalized influence of training principles on
performance).
The third cluster brings together a set of chapters that exam-
ines training in the real world. These chapters show a sample of
the possible applications of basic research and theory. Included
in this cluster are chapters 7 (summarizes the use of training
technology in the Army); chapter 13 (identifies possible transi-
tions from educational research to knowledge training required
by real jobs); chapter 14 (describes the need for training in multi-
tasking in most real-world jobs); chapter 15 (considers possible
uses of training principles developed in normal circumstances for
cognitive rehabilitation after an acquired brain injury). Training
research has applications in many real-world circumstances.
These chapters provide only a sample but illustrate the range of
Conclusions 331
possibilities and justify the attempt to extend the empirical and
theoretical findings from the laboratory to the field.
Empirical Research
The chapters of this book describe the current state of the sci-
ence of training from the perspective of cognitive psychology.
These chapters reveal an impressive set of facts about the vari-
ables that affect training, pointing to possible ways to optimize
training. But these chapters do not close the book on the various
issues discussed. Just about every issue is in need of some fur-
ther scientific examination. A poignant example concerns how
to optimize simultaneously the three component cognitive pro-
cesses of training—acquisition, retention, and transfer. Studies
have shown that variables that improve one process might not
act in the same way for the other two processes. For instance,
making acquisition difficult slows the acquisition process while
at the same time benefits retention and possibly also transfer. If
the idea is to optimize all three processes simultaneously, then
these differential effects or trade-offs have to be evaluated and
accommodated.
Chapter 3 provides several examples of the need for future
research. One case that seems most in need of additional work
concerns ways to overcome biases that influence decision mak-
ing. Many decisions require integration of informational items
that are presented sequentially to the decision maker. Exist-
ing data show that end-of-sequence decisions depend heavily
on the first items in the sequence (primacy) and, to a somewhat
lesser extent, on the items presented at the end of the sequence
(recency). Items presented in the middle are, thus, largely
ignored even though they provide an equal amount of informa-
tion. Optimal decisions usually require equal weighting of all
the items in the sequence. Finding ways to de-bias the over-
weighting of items from the ends of the sequence is an important
issue for future research. Finding ways to make items from the
middle of the sequence more salient, or distinctive, is one pos-
sible approach. Research might, for example, compare instruc-
tions that emphasize middle items to item-neutral instructions
of the sort used in prior research. Another possibility is to use
perceptual manipulations to make the middle items stand out.
Any such manipulations that shift the weighting toward middle
items should serve to de-bias the decisions away from the typi-
cal primacy and recency effects.
332 Lyle E. Bourne, Jr. and Alice F. Healy
The effects of part task training are ambiguous in tasks where
the parts are performed concurrently, requiring time-sharing.
Some studies show an advantage for initial training with only
part of the task, but others show a cost associated with this part
task training. Chapter 4 analyzes the problem and gives some
possible reasons for these ambiguous results. First, the order in
which parts are practiced and the way in which, once practiced,
are joined together makes a major difference in the success of
the training regimen. Second, independent part training does
not provide a sound basis for developing a time-sharing skill
that might be required of the task as a whole. Third, in some
cases the parts, while separable in principle, are integrated in
the whole task, and integration needs to be trained for success-
ful whole task performance. To resolve the ambiguity in these
results, further experimentation is needed to evaluate the rela-
tive contributions of the three possible causes of training failure.
Related to the issue of part task training is the question of
when automation might be used in training to simplify and,
thereby, enhance performance of a complex task. Automation is
usually introduced to eliminate the need to train on parts of a
task, which are performed automatically for the trainee, allow-
ing the trainee to concentrate on the remaining parts. Chap-
ter 6 demonstrates that eventual performance is, if anything,
adversely affected by introducing automation in the training
sequence. That outcome is entirely consistent with the training
principle of task difficulty described in chapter 2, whereby more
difficult training hinders acquisition but enhances retention
and transfer. Automation makes the training task simpler by
eliminating the need for training on certain parts of the overall
task. But this type of training does not eliminate the need to
perform on those parts that had been automated when whole
task requirements are introduced. Future research needs to be
aimed at how much automation to use at the beginning of train-
ing and how best to sequence the reintroduction of automated
parts into the training regimen.
One final empirical example in need of future research relates
to the connection between education and training. Chapter 13
discusses a set of empirical results related to fact learning and
fact memory that demonstrate that fact memory can be strength-
ened by repeated retrieval of those facts over time. Fact memory
is moderated by variables such as spacing of retrieval oppor-
tunities, self-testing, and interleaving of several sets of facts.
The emphasis was on training to enhance the memorability of
the facts. Throughout the book, however, a distinction has been
made between facts and skills (i.e., between declarative and
Conclusions 333
procedural components). Retrieval practice certainly serves to
strengthen the declarative component of performance. But there
is an additional possibility, mentioned in passing, that retrieval
practice might also strengthen the skill of retrieving, that is
the procedural component of performance. Thus, practicing the
retrieval of one set of facts might not work only to the benefit
of those facts but also might enhance the learners’ ability to
perform retrieval of other fact sets that might be encountered
in the future. To date there have been no direct experimen-
tal tests of that possibility although the research required is
straightforward.
Theoretical Research
Some of the work reported in this volume, specifically in chap-
ters 9 and 10, was designed to provide precise quantitative
computational models of well-defined tasks, such as data entry
and target detection. Although those models were successful
in accounting for and predicting performance by human sub-
jects in the base tasks, they are by their nature limited to the
particular tasks for which they were designed. These models
provide, however, a blueprint for model development that would
address on the same level other training tasks of interest. Using
this blueprint, future theoretical research is likely to produce
additional models specific to those tasks with the same compu-
tational platforms (ACT-R and IMPRINT).
These new models can be evaluated and compared using the
same procedures described in chapter 11 for the data entry mod-
els of chapters 9 and 10. That is, the mathematical soundness
and computational feasibility of these models can be assessed
by the procedures that were outlined. More importantly, for mul-
tiparameter models whose computations can be translated into
equations, the procedures developed in chapter 11 can be used
to find efficiently and rapidly parameter values that are opti-
mal for consistency with the data across a wide variety of task
conditions. The new techniques described in chapter 11 can
also be used to enhance visualization of the high-dimensional
parameter space. Knowledge of the best parameter values and
the ability to view the high-dimensional parameter space will
ensure that the predictions provided by future models are most
accurate.
The final, and possibly most important, future theoretical
development involves the compact mathematical model pre-
sented in chapter 12. This model captures all three component
cognitive processes involved in training in a quantitative form
334 Lyle E. Bourne, Jr. and Alice F. Healy
in a single equation with separate parameters for each of the
three processes. It also accounts mathematically for the impact
of most major training principles. In the future, this model can
be used to analyze the data from specific tasks. Parameters esti-
mated from performance on particular tasks can be used to
evaluate the importance of each of the three processes inde-
pendently. A given experiment exploring a particular task will
include manipulations that could differentially influence the
three component processes, and these influences can be rep-
resented by estimated parameter values. Thus, the model can
enable an investigator to localize observed training effects to
the acquisition, retention, and/or transfer processes. Further-
more, the model generates ideas or hypotheses that remain to be
tested empirically. One example concerns predictions of trans-
fer performance based on similarity between the conditions of
a particular task used for training and for testing. The model
expresses this relationship quantitatively whenever the similar-
ity relationship is known and measurable.
Applications
The various chapters of this volume have already identified
applications of training research to real-world problems. For
example, chapter 7 discusses applications to training in the
military using simulators and other high-tech support systems.
Chapter 13 describes applications to knowledge training in the
classroom for professional skill development. Chapter 14 reviews
the need for training in multitasking as a fundamental compo-
nent of complex real-world tasks such as piloting an aircraft.
There are many vast additional areas of potential application
of training principles that could benefit from a deeper scientific
examination. One area of considerable current interest and con-
cern, because of head injuries suffered by military personnel,
sports injuries (e.g., in professional football), and random acci-
dents, is cognitive retraining or rehabilitation. The issue here is
not the training of a novice but the retraining of a person who
once had a set of knowledge and skills but lost them because of
an acquired brain injury. As reviewed in chapter 15, research-
ers and clinicians have barely begun to scratch the surface of
this problem. There is some obvious success in current and past
efforts, but it is also obvious that there is room for improvement.
In particular, cognitive retrainers will need to consider the prin-
ciples for normal training outlined in this volume as possible
candidates for the facilitation of retraining efforts.
Conclusions 335
Another example comes from the possible application of theo-
retical ideas outlined in this volume to military planning. As
noted in chapter 10 and elsewhere, the Army uses a compu-
tational platform called IMPRINT to assist resource allocation
decision making in the field. IMPRINT contains parameters that
relate primarily to different task requirements and personnel
capabilities. Predictions using IMPRINT can be improved by
incorporating into the system the dimension of training based on
well-established training principles. The compact mathematical
model developed in chapter 12 encompasses a set of quantitative
expressions describing the impact of these training principles
on performance after training. These quantitative expressions
lend themselves directly to the computational device known as
performance shaping functions, and these functions are a fun-
damental part of IMPRINT. Thus, there is a direct path from the
theoretical work presented in this volume to an application in
military practice. That path could be followed in a future mili-
tary application.
Applications of training research to date have been limited
to a relatively small number of domains. The principles, how-
ever, should be much more general than those domains would
suggest. As noted above, various chapters have discussed appli-
cations in pilot training, military training, and classroom educa-
tion. But there are other domains in which principles identified
should also be useful. Among the possibilities are sports train-
ing, musical training, medical training, training astronauts for
long-haul missions (e.g., to Mars), and training digital skills. In
some cases, the principles would appear to apply directly. In
other cases, where there are imponderables, such as on a trip to
Mars, or where there is a continually evolving technology, such
as digital devices, the principles themselves might need modifi-
cation or extension. Nonetheless, these domains, in which there
are currently few applications of training research to practice,
are prime targets for future examination.