Self-Regulation in Error Management Training: Emotion Control and Metacognition as Mediators of Performance Effects.

Abstract

In error management training, participants are explicitly encouraged to make errors and learn from them. Error management training has frequently been shown to lead to better performance than conventional trainings that adopt an error avoidant approach. The present study investigated self-regulatory processes mediating this effect. Fifty-five volunteer students learned a computer program under 1 of 3 conditions: error avoidant training, error management training, or error management training supplemented with a metacognitive module. As predicted, both forms of error management training led to better transfer performance than did error avoidant training (d = 0.75). Mediation hypotheses were fully supported: Emotion control and metacognitive activity (from verbal protocols) mediated performance differences. These findings highlight the potential of promoting self-regulatory processing during training.

Full text

Self-Regulation in Error Management Training: Emotion Control and
Metacognition as Mediators of Performance Effects
Nina Keith and Michael Frese
Justus-Liebig-University of Giessen
In error management training, participants are explicitly encouraged to make errors and learn from them.
Error management training has frequently been shown to lead to better performance than conventional
trainings that adopt an error avoidant approach. The present study investigated self-regulatory processes
mediating this effect. Fifty-five volunteer students learned a computer program under 1 of 3 conditions:
error avoidant training, error management training, or error management training supplemented with a
metacognitive module. As predicted, both forms of error management training led to better transfer
performance than did error avoidant training (d ⫽ 0.75). Mediation hypotheses were fully supported:
Emotion control and metacognitive activity (from verbal protocols) mediated performance differences.
These findings highlight the potential of promoting self-regulatory processing during training.
Keywords: self-regulation, adaptive transfer, error management, active learning
“Errors are great because you learn so much from them!” Such
a statement stressing the positive function of errors may sound
quite ironic to many, given the rather negative view of errors
generally held in society. Early in school one learns that errors are
punished by poor grades, and workplace errors can have severe
consequences for individuals and organizations and may even lead
to catastrophes such as the Chernobyl meltdown. From this point
of view, there is nothing good about errors. From a psychological
perspective, however, errors make it possible to learn (Fisher &
Lipson, 1986). This is the perspective taken by proponents of error
management training: Errors provide informative feedback and
should, therefore, be explicitly incorporated into the training pro-
cess (Heimbeck, Frese, Sonnentag, & Keith, 2003). Consequently,
training participants are exposed to many errors during the training
process and are encouraged to use these errors as a learning device
by means of positive error statements such as the one in the
opening sentence.
Error management training studies have shown that error man-
agement training leads to better performance by participants than
error avoidant training, which is designed to prevent participants
from making errors, when learning new computer programs (Chil-
larege, Nordstrom, & Williams, 2003; Dormann & Frese, 1994;
Frese, 1995; Frese et al., 1991; Heimbeck et al., 2003; Nordstrom,
Wendland, & Williams, 1998). Others have added to these find-
ings, for example, by applying error management training to
driving simulation training (Ivancic & Hesketh, 2000), by com-
paring error management training with behavior modeling (De-
bowski, Wood, & Bandura, 2001; Wood, Kakebeeke, Debowski,
& Frese, 2000), and by testing aptitude–treatment interactions of
training condition and person characteristics (Gully, Payne, Koles,
& Whiteman, 2002; Heimbeck et al., 2003). Despite the growing
body of research dealing with error management training, evidence
illuminating the psychological mechanisms underlying its effec-
tiveness remains scarce. Only a few studies explicitly looked at
potentially mediating processes (Debowski et al., 2001; Wood et
al., 2000), and none of these arrived at conclusive results. The aim
of our study was to fill this gap by using both questionnaire and
verbal protocol data to identify mediating processes in error man-
agement training.
The present study contributes to the existing research in the
following ways: First, it replicates existing error management
training studies. Second, the major focus of our study is on
processes mediating the effects of error management training on
task performance. More specifically, we argue that error manage-
ment training— but not error avoidant training—stimulates self-
regulation of emotions (i.e., emotion control) and self-regulation of
cognitions (i.e., metacognition) during skill acquisition. We further
propose that the quality of these self-regulatory processes deter-
mines later task performance. This argument is consistent with
educational theory stressing the importance of metacognition in
self-regulated learning (e.g., Schunk & Zimmerman, 1994) and
with Kanfer and colleagues’ resource allocation perspective in
skill acquisition (e.g., Kanfer & Ackerman, 1989). Third, we
present and test a new variant of error management training
specifically designed to enhance metacognitive activity. To our
knowledge, no previous study has explicitly tried to systematically
change elements of error management training in order to improve
Nina Keith and Michael Frese, Work and Organizational Psychology,
Justus-Liebig-University of Giessen, Giessen, Germany.
Parts of this research were presented at the 18th Annual Conference of
the Society for Industrial and Organizational Psychology in Orlando,
Florida, April 2003. This study was conducted as part of Nina Keith’s
doctoral dissertation at the Justus-Liebig-University of Giessen.
We are indebted to Sabine Sonnentag and Steve Kozlowski for critical
comments on a previous version of this article and to Winfried Hacker for
helpful discussions in earlier stages of our research. We thank Christiane
Haupt and Leonore Schulze for their help in data collection and analysis.
Correspondence concerning this article should be addressed to Nina
Keith, Work and Organizational Psychology, Justus-Liebig-University of
Giessen, Otto-Behaghel-St. 10 F, 35394, Giessen, Germany. E-mail:
nina.keith@psychol.uni-giessen.de
Journal of Applied Psychology Copyright 2005 by the American Psychological Association
2005, Vol. 90, No. 4, 677–691 0021-9010/05/$12.00 DOI: 10.1037/0021-9010.90.4.677
677

its effectiveness. In the following, we briefly describe the basic
concepts underlying error management training. Then, we discuss
processes that potentially mediate the effectiveness of error man-
agement training.
The Concept of Error Management Training
The basic principle of error management training is that partic-
ipants are given opportunities to make errors during training.
Participants are provided with only minimal information (e.g.,
information about the functions of the computer program to be
learned) and are then given the opportunity to individually explore
the system. It can be generally stated that errors occur during
goal-oriented behavior, that they imply that a goal has not been
reached, and that they could have been potentially avoidable (Frese
& Zapf, 1994; Reason, 1990; Zapf, Brodbeck, Frese, Peters, &
Pru¨mper, 1992). In error management training, for example, if the
goal of a participant is to enlarge an object that is visible on the
computer screen, and if he or she instead moves this object, this
would be an error. Errors can, in principle, be distinguished from
inefficient actions, because inefficient actions still lead to the goal.
However, inefficient actions can be conceived as erroneous when
it is assumed that most people hold a standard of efficiency. In
error management training, inefficient actions can occur as well.
For example, if the task was to insert three additional columns into
an existing table, and if the participant first deleted the whole table
and then inserted a new table with the desired number of columns,
this would be inefficient and considered an error even though the
goal was reached.
Error management training is similar to exploratory learning
(Bruner, 1966), which emphasizes the importance of allowing the
learner to actively explore ideas and to test them (e.g., Greif &
Keller, 1990). There are two characteristics of error management
training, however, that show its greater emphasis on making errors
and using them as a learning device, compared with classical
approaches to exploratory learning. First, in contrast to exploratory
training, error management training tasks are quite difficult right
from the start, thereby exposing participants to many error situa-
tions (Heimbeck et al., 2003; Hesketh & Ivancic, 2002). Because
explicit training tasks are given, participants have clear external
objectives during training, whereas pure discovery methods often
lack this kind of structure (Mayer, 2004). The second characteristic
of error management training is that participants are explicitly
informed about the positive function of errors during training and
are presented with error management instructions to reduce poten-
tial frustration in the face of errors (Dormann & Frese, 1994;
Frese, 1995). Error management instructions are brief statements
such as “Errors are a natural part of the learning process!” or “The
more errors you make, the more you learn!”—statements that are
designed to frame errors positively (Frese et al., 1991). Error
avoidant training, on the other hand, mimics many conventional
tutorials adopting a negative attitude toward errors: Step-by-step
instructions are provided to prevent errors from occurring, and
participants are not informed about the positive functions of errors
(Frese, 1995).
In several training experiments, error management training that
included error management instructions proved superior to error
avoidant training across diverse participant samples (students as
well as employees), training content (e.g., computer training, driv-
ing simulator training), and training lengths (1-hr training sessions
to 3-day training sessions). These training experiments consisted
of one or more training phases and subsequent test phases that
assessed performance in terms of number of correct task solutions
(Chillarege et al., 2003; Debowski et al., 2001; Nordstrom et al.,
1998; Wood et al., 2000); ratings of correctness, efficiency, and
speed of solutions in difficult tasks (Dormann & Frese, 1994; Frese
et al., 1991); or number of errors in transfer tasks (Ivancic &
Hesketh, 2000). A recent study by Heimbeck et al. (2003) high-
lighted the crucial role of error management instructions in error
management training: Error management training was superior not
only to error avoidant training but also to pure exploratory training
without error management instructions. Thus, according to this
study, only the combination of providing participants with (a)
ample opportunities to make errors and (b) explicit encouragement
to learn from their errors by means of error management instruc-
tions improved task performance.
Error management training is not expected to affect all types of
learning outcomes at any time. First, error management training
aims at improving performance after (as opposed to during) train-
ing. That is, most error management training studies differentiate
one or more training phases from later test phases. During training,
participants are encouraged to make errors. During the test phase,
however, participants are aware that their performance is being
assessed (e.g., Wood et al., 2000). This distinction is crucial, given
that manipulations positively affecting training performance may
negatively affect performance in the long run and vice versa
(Goodman, 1998; Hesketh, 1997; R. A. Schmidt & Bjork, 1992).
In other words, error management training aims to improve trans-
fer performance, not training performance. In fact, training perfor-
mance may be worse in error management training in terms of
error rate, efficiency, or training time because participants are not
directly guided to correct solutions; rather they experiment, ex-
plore, make errors, and sometimes arrive at wrong solutions.
Second, error management training should affect different types
of transfer tasks differentially. Transfer implies that “knowledge,
skills and attitudes” are “transferred from one task or job to
another” (Hesketh, 1997, p. 318). Two types of transfer can be
distinguished (Ivancic & Hesketh, 2000): (a) Analogical transfer
refers to problem solutions that are familiar or analogous, and (b)
adaptive transfer entails “using one’s existing knowledge base to
change a learned procedure, or to generate a solution to a com-
pletely new problem” (Ivancic & Hesketh, 2000, p. 1968). From a
practical perspective, adaptive transfer is most relevant, because
not all potential work-related problems and solutions can be taught
during training (Hesketh, 1997; Kozlowski et al., 2001). For ex-
ample, not all functions of a new word processing program can be
explained during a 1-day training session. Back on the job, how-
ever, training participants may encounter unexpected problems
while working with the word processing program and, in contrast
to the protected training situation, might not have any assistance at
all. In this respect, error management training resembles the trans-
fer situation more than error avoidant training—an issue that is
captured in the principle of transfer appropriate processing, which
postulates that those processes required on transfer tasks should be
practiced in training (Morris, Bransford, & Franks, 1977).
We expected error management training to be particularly ef-
fective in promoting adaptive transfer, because participants learn
to deal with unexpected problems during training. For analogical
678
KEITH AND FRESE

transfer, the prediction was less clear. As outlined by Ivancic and
Hesketh (2000), errors made during training may facilitate the
retrieval of similar problems and their solutions, thereby promot-
ing analogical transfer. On the other hand, error avoidant training
might be as successful for analogical transfer as error management
training: In order to solve analogical problems, participants who
have undergone error avoidant training need only apply the correct
strategies they learned during training to the new (but analogous)
problem. Therefore, we expected analogical transfer to be the same
in both error management training and error avoidant training. This
prediction is consistent with the results of prior error management
training studies. For example, Heimbeck et al. (2003) predicted
and found group differences only for difficult tasks, not for easy
tasks (cf. Dormann & Frese, 1994; Frese, 1995). They argued that
performance in easy tasks should not benefit from error manage-
ment training, because easy tasks require only a low degree of skill
and do not lead to many errors. In sum, we expected to replicate
the group difference in adaptive transfer that has been found in
earlier studies.
Hypothesis 1: Error management training leads to better
adaptive transfer than does error avoidant training.
Processes in Error Management Training
Several mechanisms for the effectiveness of error management
training have been proposed in the literature, although only a few
studies have attempted to directly test these potential mechanisms.
Two groups of mechanisms have been proposed. First, cognition-
based approaches have highlighted the function of exploration and
associated deeper level processing during training (Dormann &
Frese, 1994; Heimbeck et al., 2003); in addition, other authors
have suggested that metacognition is important (Ivancic & Hes-
keth, 2000). Second, emotion/motivation-based approaches have
investigated the emotional or motivational processes potentially
having a facilitating or a debilitating effect on learning during
training, such as intrinsic motivation (Debowski et al., 2001; Wood
et al., 2000) or frustration (Chillarege et al., 2003; Nordstrom et
al., 1998).
We do not reject these proposed mechanisms but suggest that
they can be integrated in a self-regulatory perspective that ac-
knowledges the significance of both cognitive and emotional pro-
cesses in error management training. Self-regulation refers to
processes “that enable an individual to guide his or her goal-
directed activities over time,” comprising “modulation of thought,
affect, behavior, or attention” (Karoly, 1993, p. 25). In error
management training, self-regulatory processes are particularly
important because of the low degree of structure and the lack of
external guidance entailed in such training (A. M. Schmidt & Ford,
2003). We argue that participants in error management training
learn to use self-regulatory skills that prove valuable when partic-
ipants are confronted with new problems not practiced in train-
ing—problems that require adaptive transfer (Ivancic & Hesketh,
2000). In the following, we refer to emotion control and metacog-
nition as two self-regulatory skills mediating error management
training effectiveness.
Emotion Control in Error Management Training
Emotion control is a skill involving “the use of self-regulatory
processes to keep performance anxiety and other negative emo-
tional reactions (e.g., worry) at bay during task engagement”
(Kanfer, Ackerman, & Heggestad, 1996, p. 186). Emotion control
is expected to and has been shown to be particularly important for
learning in early phases of skill acquisition, in which errors and
setbacks are most likely to occur. Failures in emotion control result
in impaired learning and performance, because negative emotions
divert attentional resources to the self and away from the task at
hand (Kanfer & Ackerman, 1989; Kanfer et al., 1996; Kluger &
DeNisi, 1996). Not all types of emotion control processes, how-
ever, can be expected to be equally beneficial. For example, mere
suppression of negative emotions drains resources (Muraven &
Baumeister, 2000) and can result in cognitive deficits, whereas
reappraisal of the emotional event, modifying emotions before
they unfold, does not (Richards & Gross, 2000). We propose that
error management training helps participants to develop and prac-
tice beneficial skills of emotion control early on in training be-
cause error management instructions frame errors positively and
thereby encourage participants to adopt a positive perspective on
errors. In error avoidant training, however, participants are pre-
vented from making errors, and this does not prepare them to
handle their negative emotional reactions to errors. As a result,
when they are confronted with new tasks in the test phase without
guidance, they are more likely to encounter negative emotions that
have debilitating effects on their performance. In sum, we ex-
pected a mediating effect of emotion control.
Hypothesis 2: Emotion control mediates the effect of training
conditions on adaptive transfer in that (a) error management
training leads to higher emotion control than does error
avoidant training and (b) emotion control positively affects
adaptive transfer.
Metacognition in Error Management Training
Notwithstanding the critical role of emotional control processes
during skill acquisition, cognitive control processes should also be
considered because the mere absence of negative emotions does
not quite ensure learning. Rather, the free attentional resources at
one’s disposal have to be devoted to task-related activities that
maximize task learning (Kanfer & Ackerman, 1989). Following
theorizing by Ivancic and Hesketh (2000; Hesketh & Ivancic,
2002), we propose that metacognition is powerful in promoting
transfer and that error management training fosters metacognitive
activity.
Metacognition implies that an individual exerts self-regulatory
“control over his or her cognitions” (Ford, Smith, Weissbein,
Gully, & Salas, 1998, p. 220), and involves skills of planning and
monitoring as well as evaluation of one’s progress during task
completion (Brown, Bransford, Ferrara, & Campione, 1983;
Schraw & Moshman, 1995). Metacognition has been shown to be
related to academic achievement (e.g., Pintrich & De Groot, 1990;
Schunk & Zimmerman, 1994) and to problem-solving perfor-
mance (Berardi-Coletta, Buyer, Dominowski, & Rellinger, 1995;
Davidson & Sternberg, 1998) and is assumed to be particularly
useful in learning environments that provide little external struc-
679
SELF-REGULATION IN ERROR MANAGEMENT TRAINING

ture or guidance (A. M. Schmidt & Ford, 2003). In error manage-
ment training, metacognitive activities are encouraged because
“errors prompt learners to stop and think about the causes of the
error” (Ivancic & Hesketh, 2000, p. 1968). Participants then need
to come up with solutions to the impasse, implement them, and
monitor their effectiveness (Ivancic & Hesketh, 2000). These
metacognitive activities can be conceived of as higher order strat-
egies (Ford et al., 1998) that help participants to master new tasks
on their own. Error avoidant training, however, does not necessar-
ily offer the opportunity to engage in metacognitive activities
because participants are provided with the correct task solutions
and do not need to explore the system on their own. In sum, we
expected a mediating effect of metacognitive activity.
Hypothesis 3. Metacognitive activity mediates the effect of
training conditions on adaptive transfer in that (a) error man-
agement training leads to higher metacognitive activity dur-
ing training than does error avoidant training and (b) meta-
cognitive activity positively affects adaptive transfer.
In addition, our study aimed to explore whether the effect of
error management training could be improved by supplementing
error management instructions with additional instructions specif-
ically designed to enhance metacognitive activity. Although we
assume that error management promotes both emotion control and
metacognitive activity and that these two processes enhance per-
formance, supplementary instructions could be even more power-
ful: If not all participants spontaneously engage in emotion control
or metacognitive activity, additional instructions specifically de-
signed to improve one of these processes may be more effective
than error management instructions alone. For emotion control, we
would not expect a strong effect of supplemental instructions
aimed at improving emotion control because standard error man-
agement instructions already have a component of emotional relief
(e.g., “There is always a way to get out of an error situation!”), and
because prior studies suggest that error management training alone
can have an effect on emotional outcomes (Frese et al., 1991;
Nordstrom et al., 1998). Thus, an additional effect on the regula-
tion of emotions may be less likely. For instructions specifically
designed to enhance metacognitive activity, however, an addi-
tional effect seems more likely: When left without further guid-
ance in error management training, some participants may rely on
less effective strategies such as unsystematic trial and error (van
der Linden, Sonnentag, Frese, & van Dyck, 2001). Similarly,
Mayer (2004) argued that exploratory training methods can be
improved by providing help in guiding participants’ cognitive
activity in productive directions (cf. Bell & Kozlowski, 2002).
Consequently, additional instructions highlighting the benefits of
and explaining how to make use of metacognitive activity may
direct the participants’ attention to more effective strategies and,
therefore, be more successful than error management instructions
alone. On the other hand, it may be argued that because error
management training already is powerful in promoting metacog-
nitive activity, there might not be any more room for an add-on
effect of any supplementary instructions. Because of these con-
flicting expectations concerning the role of metacognitive instruc-
tions, we put forth an open research question:
Does error management training supplemented by metacogni-
tive instructions lead to better adaptive transfer than error man-
agement training alone?
Method
Participants
Participants were 55 volunteer university students majoring in education
(i.e., primary and secondary education). As an incentive, participants took
part in a lottery that was conducted after completion of the study in which
they could win one of three monetary prizes (equivalent to about $50, $30,
and $20). The sample was composed of 53 women (94%). Mean age was
23.1 years (SD ⫽ 5.2). Most participants reported having had work
experience (86%), with 27% having worked regularly before they started
attending the university and 70% working on regular basis while studying
(M ⫽ 11.6 hr per week, SD ⫽ 9.2). Participants’ experience with comput-
ers differed broadly, but none of them had ever worked with the specific
software used in this study. This was a prerequisite for participation.
Accordingly, when making the appointment for the training session and
again directly before the training started, we asked participants whether
they had used the program before. Participants were randomly assigned to
training conditions.
Experimental Design and Procedure
Participants were trained to create overhead slides with a presentation
program (PowerPoint 2000 for Windows) in one of three training condi-
tions. Sessions were run individually for each participant and lasted 2
1
⁄
2
hr
(including a 10-min break). Sessions comprised (a) an introductory phase
(identical for all participants); (b) the actual training phase, during which
the experimental manipulation took place; and (c) a test phase (identical for
all participants).
Introductory phase. In the beginning, all participants received a 2-page
manual containing general information about the program. This manual
briefly explained the menu and toolbars, how specific functions can be
activated to create objects (e.g., a rectangle), and how existing objects can
be modified (e.g., enlarging a rectangle). Also, participants were informed
about the undo function of the program. All participants received the same
manual so that task information was held constant across training condi-
tions. Reading time was approximately 5 min. Participants were allowed to
refer to their manuals during the entire training session (but not during the
test phase).
After reading the manual and before the actual training started, partici-
pants first worked on a simple slide. In this way they could get accustomed
to handling the mouse for creating objects and to thinking aloud while
working (“warm-up” exercise for verbalization; Taylor & Dionne, 2000, p.
415). This introductory task included creating and modifying a circle, a
rectangle, a text box, and an arrow while following written instructions.
The experimenter demonstrated the first few steps. She read the written
instructions aloud (e.g., “Click on the icon ‘rectangle’ in the drawing
toolbar”) and then carried out the described actions while verbalizing them.
Participants were asked to complete the task following the written instruc-
tions while thinking aloud (for instructions on thinking aloud, see below).
No time limit was given for the introductory task. Mean time for task
completion was 16.80 min (SD ⫽ 5.04) and did not differ between
experimental groups, F(2, 52) ⫽ 0.21, p ⫽ .81.
Training phase. After the introductory phase, the actual training began,
during which the training condition was experimentally manipulated. Par-
ticipants were consecutively given copies of two slides printed on paper.
The task was to reproduce these slides as closely as possible. The first slide
required creating, moving, and modifying (e.g., coloring) diverse objects
such as rectangles, triangles, text boxes, and stars. The second slide
involved creating and modifying a table by simple formatting such as
680
KEITH AND FRESE

coloring cells and centering cell entries. To complete each slide, partici-
pants were given 15 min, resulting in a training time of 30 min. Those
participants who finished the two slides before the training time was up
received a third slide to work on during the remaining time. This third slide
looked different than the former ones but required program functions
similar to those already used. The number of participants who worked on
the third slide did not differ between experimental groups, F(2, 52) ⫽ 1.14,
p ⫽ .33.
Participants completed the training tasks in one of three training condi-
tions: error avoidant training, error management training, or error manage-
ment training supplemented with a metacognitive module. In the error
avoidant training (n ⫽ 18), participants received detailed written instruc-
tions (similar to those in the introductory practice phase) explaining task
solution in a step-by-step manner. This training condition resembled com-
mercially available software tutorials. Participants were asked to follow the
instructions closely. They were told that these instructions would enable
them to learn the most important program functions in the shortest time and
that by following the instructions participants would become familiar with
the correct functions from the very beginning.
In the error management training condition (n ⫽ 17), participants were
not provided with any information on task solution. They received instruc-
tions emphasizing the positive function of errors during training and were
encouraged to make errors and learn from them. In addition, the following
error management instructions derived from earlier error management
training studies were presented (cf. Debowski et al., 2001; Dormann &
Frese, 1994; Heimbeck et al., 2003; Wood et al., 2000): “Errors are a
natural part of the learning process!” “There is always a way to leave the
error situation!” “Errors inform you about what you still can learn!” “The
more errors you make, the more you learn!” During training, the error
management instructions were prominently displayed on a poster and
verbally repeated by the experimenter.
Participants in the error management training plus metacognition con-
dition (n ⫽ 20) initially received exactly the same treatment as participants
in the error management training condition. That is, before they worked on
the first training slide, they were not provided with any information about
the task solution but were given instructions emphasizing the positive
function of errors and error management instructions. When the partici-
pants worked on the second training slide, however, treatment differed in
that participants received additional instructions designed to enhance meta-
cognitive activity. These metacognitive instructions were derived from a
study conducted by King (1991; cf. also McInerney, McInerney, & Marsh,
1997; A. M. Schmidt & Ford, 2003) in which pairs of children were trained
in strategic questioning while solving problems. In King’s study, the
children were provided with an index card listing questions concerning
metacognitive planning, monitoring, and evaluation (e.g., “Are we getting
closer to our goal?” “What worked? What did not work?”). In the present
study, these questions were adopted and slightly modified. Participants
were first given brief written instructions explaining the benefits of stra-
tegic questioning while working on the training tasks. They were then told
to pose and answer these questions to themselves whenever appropriate
while working on the task. For example, when they did not know what to
do next, participants were told to analyze the problem and develop a
strategy by asking and answering questions such as “What is my problem?
What am I trying to achieve?” or “What do I know about the program so
far that can be useful now?” Finally, the list of questions to be posed was
prominently displayed on a poster during training and verbally repeated by
the experimenter. To ensure that participants followed instructions on
strategic questioning during training, they were told that the questions they
posed and answered would be tape-recorded and counted later on (which
was actually the case, because the method of thinking aloud was used
during training; see below).
Participants in all three groups were informed about the undo function of
the program and the delete key before the training started. This was done
to keep knowledge about these error correction options constant. No further
help was provided during training. Only in the few cases in which partic-
ipants could not continue with the task did the experimenter intervene (e.g.,
a participant accidentally closed the working file; another participant
accidentally “lost” a toolbar that was essential for task completion). The
number of interventions by the experimenter did not differ between train-
ing conditions, F(2, 52) ⫽ 0.86, p ⫽ .43.
During the entire training, the method of thinking aloud was used.
Instructions for thinking aloud were carefully constructed following rec-
ommendations by Ericsson and Simon (1993). Instructions were as fol-
lows: “While you are working on the slide, please verbalize all your
thoughts. Just speak out whatever comes into your mind, no matter what it
is.” When participants stopped verbalizing for more than 10 s, they were
prompted to continue (“Please keep on talking”). Empirical evidence
suggests that this type of verbalization instruction is least obtrusive to
participants’ cognitive processing (Ericsson & Simon, 1993; Taylor &
Dionne, 2000). The number of prompts for continuing verbalization did not
differ between training conditions, F(2, 52) ⫽ 1.97, p ⫽ .15.
Test phase. Tasks and instructions in the test phase were identical for
all participants. Participants were handed printed copies of three slides. As
in the training phase, the task was to reproduce these slides as closely as
possible. The test slides, however, were more difficult than the training
slides (cf. Dormann & Frese, 1994; Heimbeck et al., 2003). The first test
slide comprised bullet points with text items and a figure consisting of
several framed and colored text boxes and arrows. The main task of the
second test slide was to produce and to format a table. For the third slide,
a vertical bar chart had to be created and edited with the diagram function
of the program. In addition, all three test phase slides involved picking a
specific design template and predefined layouts of the program. Because
pilot testing had indicated that these were extremely difficult tasks, all
participants were informed about the menu options where they would find
the required functions. Participants were given 12 min to complete each
slide, resulting in an overall testing time of 36 min. Before testing started,
participants were told that this was the test phase in which they were to
demonstrate what they had learned during the training session (cf. Wood et
al., 2000).
Measures
Performance. Performance ratings were conducted on the basis of the
slides the participants had created during the training and the test phase.
Each task was divided into meaningful observable subtasks. For example,
the task to create a figure consisting of several text boxes and arrows was
divided into seven subtasks: “at least one text box present,” “all text boxes
present,” “position of text within text box correct,” “at least one arrow
present,” “all arrows present,” “format of arrows correct,” and “relative
positions of text boxes and arrows correct.” The subtasks served as coding
units and were rated as either correctly completed or not (dichotomous
rating; cf. Heimbeck et al., 2003). A second rater coded a randomly chosen
subset of training and test slides. The two raters were Nina Keith and a
graduate student who was trained to use the coding system. Both raters
were blind to the experimental condition. For the training phase slides,
Cohen’s kappa was .87 (based on a subset of 270 coding units). To arrive
at a measure for overall training performance, the number of completed
subtasks was computed for each participant.
For the ratings of the test phase slides, Cohen’s kappa was .89 (based on
a subset of 768 coding units). We further divided the subtasks of the test
phase into tasks of low and high distinctiveness from training slides. A
subtask was rated as low in distinctiveness if it required mere repetitions of
program functions used in training (e.g., creating a text box, changing the
color of a rectangle) or if a program function used in training had to be
applied in a similar although not exactly identical manner as in training
(e.g., inserting a 4 ⫻ 4 table when the training task was to insert a 3 ⫻ 3
table). A subtask was rated as high in distinctiveness if a completely new
function had to be applied for task completion (e.g., complex formatting of
a table, inserting and editing a diagram). Interrater agreement on this
681
SELF-REGULATION IN ERROR MANAGEMENT TRAINING

distinction of high–low distinctiveness for the 64 subtasks of the test slides
was high (
␬
⫽ .84). Cases in which the ratings of both raters differed were
resolved by discussion. Low-distinctiveness subtasks solved were summed
to represent analogical transfer; high-distinctiveness subtasks solved were
summed to represent adaptive transfer (cf. Ivancic & Hesketh, 2000). All
analyses are based on these sum scores. As outlined in the introduction, we
expected performance differences between experimental groups only for
adaptive transfer, not for analogical transfer.
Verbal protocol data. Participants’ thinking aloud during training was
transcribed verbatim and segmented with each phrase (either complete or
incomplete) constituting a segment (cf. Sonnentag, 1998). Mere expressive
utterances (e.g., “Hum,” “Okay,” “Yup”) were coded as such and excluded
from further analyses. Because of technical problems (microphone dys-
functions and broken videotapes), audio data of 6 participants were lost,
resulting in a sample size of 49 for all analyses comprising verbal protocol
data (error avoidant group, n ⫽ 14; error management training group, n ⫽
16; error management training plus metacognition group, n ⫽ 19).
The second half of the training phase, in which participants created and
formatted a table, was critical for the present research question. Only in this
phase did all three training conditions differ (the first training phase was
identical for the error management training and the error management
training plus metacognition group). Also, the task in this phase was rather
difficult and, therefore, required deliberate and conscious processing,
which is a prerequisite for verbal protocols to “generate rich and valid
data” (Taylor & Dionne, 2000, p. 415). We therefore based our analyses on
the verbal data of the second training phase. Another potential threat to the
validity of verbal protocol data is posed by general verbalization tendencies
of participants (as a person characteristic) influencing critical verbaliza-
tions. Thus, we counted the verbalizations of participants during the
introductory phase in which no experimental manipulation had occurred.
The general verbalization tendencies of participants (as indicated by the
number of phrases per minute) did not differ between the experimental
groups, F(2, 43) ⫽ 0.20, p ⫽ .82. Protocols of the critical training phase
comprised an average number of 166.9 segments (SD ⫽ 57.3) and did not
differ in length between experimental conditions, F(2, 46) ⫽ 1.78, p ⫽ .18.
Each segment was classified into one of two major categories and into
a more specific subcategory within the major category (cf. Berardi-Coletta
et al., 1995; Sonnentag, 1998). The first major category, which was the
focal category in our study, was metacognitive statements; statements
reflecting metacognitive control of planning, monitoring, and evaluation
were categorized here. The second major category, which we called task-
focused statements, subsumed statements that indicated task orientation but
lacked the cognitive control characteristic for metacognitive processing.
Only a very few segments did not fit into either category (M ⫽ 2.31, SD ⫽
2.05) and were deleted from further analyses. The number of nonclassifi-
able segments did not differ between experimental conditions, F(2, 46) ⫽
0.40, p ⫽ .67. The two major categories of metacognitive versus task-
focused statements map the distinction made by Berardi-Coletta et al.
between processing level (i.e., metacognition) and problem level as two
general levels of cognitive-attentional focus during problem solving. The
most frequent subcategories of metacognitive statements and task-focused
statements, along with sample statements, are listed in Table 1. The two
most frequent categories in task-focused statements refer to mere descrip-
tions by participants on what action step they were just performing (see
Table 1, Category 2a) or what action step they were about to perform
(Table 1, Category 2b). These categories may appear similar to the second
metacognitive category listed, “Monitoring—observing changes” (Table 1,
Category 1b). The difference is that statements coded in the latter category
reflected more detailed and attentive observations by participants that did
not refer to the performed action itself (as in Categories 2a and 2b, Table
1) but referred to the visible changes on the computer screen that were the
result of an action performed.
The statements were classified by Nina Keith and a graduate student
who was trained to use the coding system. Cohen’s kappa was .80 for the
distinction between the two major categories (based on a subsample of
2,000 segments). Although interrater agreement remained acceptable on
the level of subcategories (
␬
⫽ .69), we based our main analyses on the
broader level because our hypotheses referred not to specific metacognitive
subprocesses but to overall metacognitive activity of participants. If a
statement was categorized as metacognitive and the same statement was
then merely repeated by the participant, these repetitions were counted as
such and excluded from further analyses. We then calculated the percent-
age of metacognitive statements relative to the total number of statements
to represent metacognitive activity during training.
Emotion control. Emotion control during task engagement was as-
sessed shortly after the test phase using a self-developed eight-item scale.
Items were subjected to a pilot test involving an independent sample (N ⫽
79) while closely following definitions of the construct as outlined by
Kanfer and colleagues (e.g., Kanfer & Ackerman, 1989; Kanfer et al.,
1996). We used this self-developed scale in our study because existing
measures of emotion control or related constructs did not seem to fit our
purposes. Although Kanfer and colleagues used a measure for emotion
control in a study dealing with job search activities (Wanberg, Kanfer, &
Rotundo, 1999), there are two reasons why their items did not seem
suitable for our study. First, their items are mostly specific to their research
question (e.g., “I get anxious even thinking about a job interview”).
Second, their items appear to measure emotion control only indirectly by
measuring negative emotions (i.e., anxiety in the sample item) as an
indicator of lack of emotion control. The items we developed were de-
signed to capture strategies for regulation of negative emotions that par-
Table 1
Two Major Categories and Most Frequent Subcategories in Verbal Protocol Analysis
Category Subcategory
1. Metacognitive statements a. Planning—generation of hypotheses (e.g., “It must be possible to select these cells separately,” “If I mark the whole
thing right here, I should be able to do the frame”)
b. Monitoring—observing changes (e.g., “Now I have these dotted lines again,” “And if I pull the mouse across them,
these turn blue”)
c. Evaluation—derivation of general rules (e.g., “I first have to click on this thing here, then I get these dots and I can
move it,” “I cannot do this until I have inserted the line”)
d. Evaluation—explicit explanation (e.g., “That’s because I have clicked on this pen here,” “No, because I have to
activate it first”)
2. Task-focused statements a. Description of present step (e.g., “I click on textbox,” “Now I pull this,” “And I center this one, too”)
b. Description of next step (e.g., “Now I will enter the text,” “Now I will center it again,” “I will make this more
evenly spread”)
c. Negative evaluation without explanation (e.g., “No, that’s wrong,” “No, I don’t like that,” “I didn’t want that”)
d. Spelling out while typing (text or numbers to be entered into the table)
e. Reading out or repeating instructions (error avoidant group only)
682
KEITH AND FRESE

ticipants actively engage in rather than negative emotions per se. In this
respect, our scale resembled coping questionnaires (e.g., Carver, Scheier,
& Weintraub, 1989; Folkman & Lazarus, 1985) or more recently published
scales on emotion regulation at work in service employees (e.g., Grandey,
Dickter, & Sin, 2004; Totterdell & Holman, 2003), where items directly
refer to regulatory strategies one might use when experiencing a stressful
encounter. However, in line with Kanfer’s conceptualization of emotion
control, on which our research was based, our scale’s emphasis was on
controlling emotions and sustaining attention during completion of a spe-
cific task. We used a modified version of Wanberg et al.’s item instruction:
Participants were asked to rate their reaction to problems they faced during
task completion. All items began with the root “When difficulties arose”
with various stems following, including “I purposely continued to focus
myself on the task” and “I calmly considered how I could continue the
task.” Items were answered on a 5-point Likert-type scale ranging from 0
(does not apply)to4(applies). Cronbach’s alphas were .82 in the pilot
sample and .80 in the present sample.
Error orientation. As a manipulation check, error orientation during
task completion was assessed using two subscales of the Error Orientation
Questionnaire (EOQ; Rybowiak, Garst, Frese, & Batinic, 1999). The
original questionnaire is designed to measure “attitudes to and . . . coping
with errors at work” (Rybowiak et al., 1999, p. 527) of individuals or
groups. For the present study, we chose two of the eight EOQ subscales
covering important individual error orientations that we expected to be
affected by error management instructions (EOQ subscale Error Strain) and
by metacognitive instructions (EOQ subscale Learning From Errors). In
order to fit the present research question, we slightly modified the instruc-
tions and items to capture error orientations during task completion (rather
than general orientations at work). The Error Strain subscale consisted of
five items involving negative emotional reactions to errors and being afraid
of making errors (e.g., EOQ item “I feel embarrassed when I make an
error” was changed to “I felt embarrassed when I made an error”). Cron-
bach’s alpha was .81 for this scale. The Learning From Errors subscale
comprised four items covering the extent to which people used errors to
learn (e.g., EOQ item “Errors help me to improve my work” was changed
to “Errors helped me to improve my work”). Cronbach’s alpha was .82 for
this scale.
Computer experience. Before the onset of the study, participants were
asked how many years they had been using a computer and which com-
puter applications they used (e.g., word processing programs, spreadsheet
programs). We used years of computer usage and number of applications
as two indicators of computer experience and included these variables as
covariates in all analyses. There were no preexperimental differences
between training conditions in years of computer usage, F(2, 52) ⫽ 0.25,
p ⫽ .78, or in number of applications, F(2, 52) ⫽ 0.51, p ⫽ .67.
All participants had worked with text processing programs before (such
as Word for Windows). We asked participants which functions of text
processing programs they used, because the presentation program taught in
this study shares many features with common text processing programs.
For example, participants were asked whether they regularly formatted
text, used bullets, or created and formatted tables. We used the number of
functions regularly used by participants as a third indicator of computer
experience and included this variable as another covariate in all analyses.
There was a preexperimental difference between training conditions in
number of functions: Before the study began, participants who underwent
error avoidant training knew more computer functions relevant to the
program taught, F(1, 52) ⫽ 4.61, p ⬍ .05,
␩
2
⫽ .08 (for the descriptive
statistics, see Table 1).
Results
Intercorrelations of Study Variables
Descriptive statistics and intercorrelations between the study
variables are displayed in Table 2. As expected, intercorrelations
between computer experience and performance variables were in
the middle range with all but one coefficient being significant. We
included all three computer experience variables as statistical
controls in all further analyses.
Manipulation Checks
To ensure that participants had interpreted the error manage-
ment instructions and the metacognitive instructions in the in-
tended way, error orientation of participants during task comple-
tion was compared. We expected error strain to be lower in both
error management training groups than in the error avoidant group
because error management instructions frame errors positively,
and errors should, therefore, be perceived as less threatening.
Planned contrasts revealed that this was the case: Error strain was
significantly higher in the error avoidant group compared with the
error management training groups, F(1, 49) ⫽ 5.81, p ⬍ .05,
␩
2
⫽
.11, but did not differ between error management training groups,
F(1, 49) ⫽ 0.03, p ⫽ .86. We further expected learning from errors
to be particularly high in the error management training plus
metacognition condition because the metacognitive instruction
given to this group stressed the usefulness of metacognitive plan-
ning, monitoring, and evaluation for learning over and above the
rather general positive framing of errors in the error management
instructions. Again, this was the case. Learning from errors was
significantly higher in both error management training groups
compared to the error avoidant group, F(1, 49) ⫽ 8.11, p ⬍ .01,
␩
2
⫽ .14, and a direct comparison of the two error management
groups revealed that it was highest in the error management
condition with the metacognitive instructions, F(1, 49) ⫽ 4.17,
p ⬍ .05,
␩
2
⫽ .08. Taken together, these results suggest that both
the error management instructions and the metacognitive instruc-
tions worked in the intended way.
To assess whether participants in the error management training
plus metacognition group followed instructions to pose and answer
questions related to metacognitive planning, monitoring, and eval-
uation during training, we counted how often participants in all
training groups posed questions similar to those listed in the
metacognitive instructions (
␬
⫽ .71, based on 2,000 segments). As
expected, the number of questions was largest in the verbal pro-
tocols of the error management training plus metacognition con-
dition, F(1, 43) ⫽ 23.30, p ⬍ .01,
␩
2
⫽ .35, indicating that
participants had followed metacognitive instructions.
We also used the verbal protocol data to further illuminate
whether participants in error management training had in fact
made more errors during training than participants in error
avoidant training who received detailed instructions on the task
solution (note that despite these instructions, participants in this
condition could still make errors— e.g., because they did not read
the instructions correctly). Although not an exact count, the cate-
gory “Negative evaluation without explanation” can serve as an
indicator of errors in training, because statements subsumed under
this category imply that participants’ preceding action did not lead
to the desired outcome (Category 2c in Table 1). As expected, the
statements in this category were much more frequent in the error
management training groups (M ⫽ 20.91, SD ⫽ 9.23) compared
with the error avoidant group (M ⫽ 8.79, SD ⫽ 4.68), F(1, 43) ⫽
15.68, p ⬍ .01,
␩
2
⫽ .27.
Finally, to get a better picture of what exactly happened in the
683
SELF-REGULATION IN ERROR MANAGEMENT TRAINING

training conditions, we inspected the frequency of task-focused
statements in the training groups. In the error avoidant training
group, about one third of the statements comprised reading or
repeating the written instructions on task solution (Category 2e in
Table 1; M ⫽ 31.31, SD ⫽ 11.71; numbers refer to percentage
relative to all statements). Participants in this condition also fre-
quently described what they were currently doing (Category 2a;
M ⫽ 19.37, SD ⫽ 6.65) or what they were about to do (Category
2b; M ⫽ 11.56, SD ⫽ 5.39). In error management training, too,
participants frequently described the present step (Category 2a;
M ⫽ 24.92, SD ⫽ 10.30) or the next step (Category 2b; M ⫽
20.79, SD ⫽ 5.46), but the third most frequent category was the
error category (Category 2c; M ⫽ 12.36, SD ⫽ 4.48). Taken
together, these analyses suggest that the experimental manipula-
tion was successful: Participants in error avoidant training worked
along the lines of the written instructions during training, and
participants in error management training frequently made errors
while working on their own.
Main Effects of Training Condition on Transfer
Performance
Hypothesis 1 predicted adaptive transfer performance to be
superior in both error management trainings than in error avoidant
training. In an open research question, we further explored whether
participants in the error management training plus metacognition
condition (i.e., with additional metacognitive instructions) would
perform better than those in classical error management training
without additional metacognitive instructions. For performance in
analogical transfer, we did not expect any differences between
training groups. We first tested this with a repeated measures
analysis of covariance (ANCOVA) with training condition varied
between subjects, transfer type (i.e., analogical and adaptive trans-
fer) varied within subjects, and computer experience variables as
covariates. As expected, a significant interaction between training
condition and transfer type emerged, F(2, 49) ⫽ 4.20, p ⬍ .05,
␩
2
⫽ .15. In line with predictions, analogical transfer did not differ
between groups, F(2, 49) ⫽ 0.39, p ⫽ .68, but adaptive transfer
did, F(2, 49) ⫽ 4.34, p ⬍ .05,
␩
2
⫽ .15.
We found clear support for Hypothesis 1 (see Table 3): Adap-
tive transfer was superior in error management training conditions
compared with error avoidant training with a medium to large
effect size (
␩
2
⫽ .12; note that the correlation reported in Table 2
between the variable contrasting error avoidant training with error
management trainings is not significant, whereas the ANCOVA
contrast from Table 3 is because the correlation does not take the
control variable computer experience into account). The difference
Table 2
Means, Standard Deviations, and Intercorrelations of the Study Variables
Variable 1 2 345678 9 101112
Training conditions
a
1. Contrast 1 —
2. Contrast 2 .05 —
Computer experience
3. Years of computer usage .08 ⫺.05 —
4. Number of applications ⫺.16 ⫺.02 .27* —
5. Number of functions ⫺.28* .11 .34* .44** —
Performance variables
6. Training performance ⫺.15 ⫺.15 .30* .25 .38** —
7. Analogical transfer .00 ⫺.01 .42** .41** .46** .70** —
8. Adaptive transfer .18 .15 .49** .41** .49** .63** .77** —
Mediators
9. Emotion control .42** .15 .07 ⫺.25 ⫺.02 .12 .20 .30* (.80)
10. Metacognitive activity .63** .31* .14 ⫺.16 ⫺.13 ⫺.04 .10 .32* .38** —
Manipulation checks
11. Error strain ⫺.31* ⫺.02 ⫺.15 .17 ⫺.08 ⫺.13 ⫺.24 ⫺.24 ⫺.68** ⫺.18 (.81)
12. Learning from errors .34* .30* .07 ⫺.07 .12 ⫺.07 .09 .19 .48** .44** ⫺.28* (.82)
Total sample
M 0.35 0.05 5.10 1.95 4.36 25.95 24.73 10.87 2.97 4.88 0.85 2.60
SD 0.95 0.83 3.02 1.22 2.03 9.00 4.89 5.35 0.61 3.52 0.72 0.70
Error avoidant group
M ⫺2.00 0.00 4.75 2.22 5.17 27.89 24.72 9.50 2.60 1.40 1.17 2.26
SD 0.00 0.00 2.59 1.22 1.89 7.84 5.11 4.59 0.66 1.81 0.75 0.68
Error management group
M 1.00 ⫺1.00 5.49 1.82 3.65 26.65 24.82 10.53 3.04 5.01 0.69 2.50
SD 0.00 0.00 3.19 1.63 1.87 10.31 5.15 6.07 0.59 1.74 0.69 0.71
Plus metacognition group
M 1.00 1.00 5.07 1.80 4.25 23.60 24.65 12.40 3.24 7.35 0.69 2.99
SD 0.00 0.00 3.35 0.77 2.12 8.71 4.70 5.20 0.41 3.49 0.65 0.52
Note. N ⫽ 55 (error avoidant group, n ⫽ 18; error management training group, n ⫽ 17; error management training plus metacognition group, n ⫽ 20).
For all analyses involving metacognitive activity, n ⫽ 49. Alpha coefficients are shown in parentheses on the diagonal when applicable. Plus metacognition
group ⫽ error management training plus metacognition group.
a
Contrast 1 is a comparison of error avoidant training with both error management training groups (error avoidant training ⫽⫺2, both error management
training groups ⫽⫹1), Contrast 2 is a comparison of error management training groups (error avoidant training ⫽ 0, error management training ⫽⫺1,
error management training plus metacognition ⫽⫹1).
* p ⬍ .05. ** p ⬍ .01.
684
KEITH AND FRESE

between the two error management training conditions (open re-
search question) was not significant ( p ⫽ .26).
Emotion Control and Metacognitive Activity as Mediators
of Adaptive Transfer Performance
In Hypotheses 2 and 3 we predicted that emotion control and
metacognitive activity during training would mediate the effect of
training condition on adaptive transfer performance. We first
tested these hypotheses separately and then simultaneously for
emotion control and metacognitive activity using the procedure
recommended by James and Brett (1984). According to this pro-
cedure, variable b mediates the effect of variable a on variable c if
the following conditions are met: First, a has an effect on b;
second, b has an effect on c; and, third, the effect of a on c
disappears when b is held constant. The first and second conditions
were met: Table 2 reveals that the training condition (i.e., Contrast
Variable 1) was significantly related to both mediators and that
both mediators were significantly related to adaptive transfer per-
formance. The third condition was tested in hierarchical regression
analyses in which training condition was entered as a predictor
after controlling for mediating variables. Results are displayed in
Table 4. When entered after emotion control or metacognitive
activity in separate analyses, the effect of training condition van-
ished (after emotion control,
␤
⫽ .17, ns; after metacognitive
activity,
␤
⫽ .01, ns). Further, when entered after both mediators
(emotion control and metacognitive activity), the effect of training
condition disappeared (
␤
⫽⫺.08, ns), and the effects of both
mediators remained significant (
␤
⫽ .24 for emotion control, and
␤
⫽ .27 for metacognitive activity, both ps ⬍ .05). Thus, Hypoth-
eses 2 and 3 were supported: Emotion control and metacognitive
activity fully and independently mediated the effect of training
condition on performance.
To supplement the ordinary least squares regression analyses,
structural equation modeling (SEM) was conducted using the
maximum-likelihood (ML) procedure in LISREL (Jo¨reskog &
So¨rbom, 1996). These analyses were conducted because SEM
offers the advantages (a) that parameters can be estimated simul-
taneously, (b) that an overall model fit can easily be obtained, and
(c) that additional paths can be introduced into the model and
tested for statistical significance. Although LISREL and similar
approaches are commonly used as large sample size procedures,
recent evidence suggests that SEM-ML can also yield appropriate
estimates in mediation models with small samples (Hoyle &
Kenny, 1999). In our models, to keep the subject-to-parameter
ratio at an acceptable level and to keep the model simple, we did
not include the control variables as exogenous variables but used
residuals instead. That is, we regressed the four study variables
(predictor training condition, the mediators emotion control and
metacognitive activity, and criterion adaptive transfer) on the
computer experience variables and used the covariance matrix of
the residual variables as input for the LISREL analyses. The model
had an excellent fit,
␹
2
(2, N ⫽ 49) ⫽ 1.09, p ⫽ .58, root-mean-
square error of approximation ⫽ .00, adjusted goodness-of-fit
index ⫽ .94, normed fit index ⫽ .98, comparative fit index ⫽ 1.00.
Standardized parameter estimates of the model are depicted in
Figure 1. All hypothesized paths were significant. We further
tested the indirect effects of training condition on adaptive transfer
for significance using Sobel’s first-order solution for standard
errors of indirect effects (MacKinnon, Lockwood, Hoffman, West,
& Sheets, 2002). Both the paths, via emotion control and via
metacognitive activity, were significant (both ps ⬍ .05).
In a second LISREL model we introduced an additional direct
effect of training condition on adaptive transfer (additional to the
paths depicted in Figure 1). This path was estimated to be zero
(⫺.09 in the standardized solution, ns). Also, the model fit did not
improve, ⌬
␹
2
(1, N ⫽ 49) ⫽ 0.34, p ⬎ .75. Thus, replicating results
of the regression analyses (cf. Table 4), the effect of training
condition on adaptive transfer was fully und independently ex-
plained by the mediators emotion control and metacognitive ac-
tivity in LISREL analyses.
We further explored the relationship between emotion control
and metacognition with LISREL. As can be seen in Table 2, the
manifest zero-order correlation between these two variables was
significant (r ⫽ .38, p ⬍ .01). In a third LISREL model we
introduced a correlation between emotion control and metacogni-
tive activity (in addition to the paths depicted in Figure 1). In this
model, the correlation was estimated to be zero (.09 in the stan-
dardized solution, ns), and model fit did not improve, ⌬
␹
2
(1, N ⫽
49) ⫽ 0.76, p ⬎ .50. Thus, the training condition served as an
explanatory variable in the mediation model: Emotion control and
metacognitive activity covaried only to the extent to which both
processes were evoked by the training condition.
Table 3
Effects of Training Condition on Adaptive Transfer (Analysis of Covariance Contrasts
Controlling for Computer Experience)
Contrast F(1, 49)
Effect size
␩
2
Cohen’s d
Error avoidant vs. error management training groups (Group 1 vs. Groups 2 ⫹ 3) 6.75* .12 0.75
Error management training vs. error management training plus metacognition
(Group 2 vs. Group 3) 1.28
Note. N ⫽ 55. For this analysis, the appropriate effect size estimate is
␩
2
, representing the explained variance.
For ease of interpretability, Cohen’s d, based on residuals after controlling for computer experience, was also
calculated.
* p ⬍ .05.
685
SELF-REGULATION IN ERROR MANAGEMENT TRAINING

Discussion
The main goal of our study was to identify processes mediating
error management training effectiveness. In line with resource
allocation theories assuming a limited amount of attentional re-
sources (e.g., Kanfer & Ackerman, 1989; Kluger & DeNisi, 1996),
we argued that error management training helps to exert self-
regulative control that in turn leads to better learning and perfor-
mance. More specifically, we proposed that error management
training enhances both emotional self-regulation (i.e., emotion
control) and cognitive self-regulation (i.e., metacognitive activity).
A second goal of our study was to explore whether there is still
room for an add-on effect of additional metacognitive instructions
over and above the effect of classical error management training
without such instructions.
Our study replicated the main effect on performance that has
frequently been found in error management training studies (e.g.,
Chillarege et al., 2003; Dormann & Frese, 1994; Frese et al., 1991;
Heimbeck et al., 2003; Nordstrom et al., 1998; Wood et al., 2000):
Error management training participants outperformed those in
error avoidant training on an adaptive transfer test. This effect was
appreciable (an effect size equivalent to a Cohen’s d of 0.75),
given that the error avoidant group was not a nontraining control
group. A direct comparison of the two error management training
conditions (i.e., error management training vs. error management
training supplemented by metacognitive instructions) did not re-
veal any performance difference. It is possible that the metacog-
nitive instructions were too weak in this study because participants
were to engage in metacognitive activities individually whereas
Table 4
Summary of Hierarchical Regression Analysis for Emotion Control and Metacognitive Activity
as Mediators of Training Effects on Adaptive Transfer
Step and variable BSEB
␤
R
2
⌬R
2
Direct effect of training condition
Step 1: Computer experience variables (controls) .38**
Years of computer usage 0.60 0.21 .34**
No. of applications 0.85 0.54 .19
No. of functions 0.76 0.33 .29*
Step 2: Training condition .46** .08**
Error avoidant vs. error management 1.69 0.62 .30**
Mediation by emotion control
Step 2: Mediator .51** .12**
Emotion control 3.20 0.91 .37**
Step 3: Training condition .53** .02
Error avoidant vs. error management 0.98 0.65 .17
Mediation by metacognitive activity
Step 2: Mediator .53** .11**
Metacognitive activity 0.54 0.17 .35**
Step 3: Training condition .53** .00
Error avoidant vs. error management 0.03 0.86 .01
Mediation by both emotion control and metacognitive activity
Step 2: Mediators .57** .16**
Emotion control 2.16 1.01 .24*
Metacognitive activity 0.42 0.17 .27*
Step 3: Training condition .58** .00
Error avoidant vs. error management ⫺0.46 0.85 ⫺.08
Note. N ⫽ 55. For all analyses involving metacognitive activity, n ⫽ 49.
* p ⬍ .05. ** p ⬍ .01.
Figure 1. Emotion control and metacognitive activity mediating effects
of training condition on adaptive transfer (standardized parameter esti-
mates from LISREL analysis).
686
KEITH AND FRESE

other studies had participants work in dyads (King, 1991) or in
cooperative groups (McInerney et al., 1997). In a recent study by
A. M. Schmidt and Ford (2003), the effect of a metacognitive
intervention for individual learners depended on their dispositional
goal orientation. Also, given that metacognitive activities require
effortful and time-consuming processing, the practice phase in the
present study might have been too short for the benefits of meta-
cognitive activities to fully develop. Future research should inves-
tigate whether metacognitive instructions can improve error man-
agement training under conditions involving longer time periods or
multiple sessions.
The most intriguing finding of this study is the strong support
for our mediational hypotheses: Group differences between error
avoidant training and error management training in adaptive trans-
fer performance were fully and independently explained by emo-
tion control and metacognitive activity during training. Supple-
mental analyses with LISREL further revealed that the empirical
zero-order correlation between the two mediators was fully ac-
counted for by training condition in the mediation model. In other
words, error management training induced both emotion control
and metacognitive activity during training, and these processes
enhanced performance in tasks that required finding new solutions.
From a self-regulatory perspective, emotional self-regulation
(emotion control) and cognitive self-regulation (metacognition)
were equally important for adaptive transfer to occur.
Our results concerning emotion control are consistent with
theory and research by Kanfer and colleagues (Kanfer & Acker-
man, 1989; Kanfer et al., 1996). They argued that emotion control
is a skill useful in early phases of skill acquisition because it helps
to direct attention away from the self and to the problem at hand
in the face of errors and setbacks. From this point of view, error
management training may be regarded as a form of emotion
control training because participants are confronted with errors
early on in training and learn to exert emotion control in order to
deal with them. Similarly, within the framework of transactional
stress theory (Lazarus & Folkman, 1984), error management train-
ing may be thought of as a form of cognitive reappraisal training,
because error management instructions reframe errors positively.
As a consequence, participants in error management training can
conceive of errors as positive and useful events, rather than threat-
ening ones, which in turn reduces the experience of negative
emotions in the face of errors.
Our results concerning metacognitive activity conform to edu-
cational theory and research highlighting the benefits of metacog-
nition during learning (e.g., Schunk & Zimmerman, 1994; cf. also
A. M. Schmidt & Ford, 2003) and to theorizing by Ivancic and
Hesketh (2000; Hesketh & Ivancic, 2002). Ivancic and Hesketh
delineated that error management training instigates metacognitive
planning, monitoring, and evaluation because errors encourage a
systematic analysis of the error’s cause as well as an implemen-
tation and testing of potential solutions. Error avoidant training
that prevents participants from making errors, in contrast, does not
provide the opportunity to practice emotion control and metacog-
nition because participants simply follow correct instructions and
do not need to work out solutions on their own. This line of
argument is also consistent with cognitive theories of action reg-
ulation. For example, action theory (Frese & Zapf, 1994; Hacker,
1998) posits that errors disrupt premature automatization of ac-
tions because they make learners rethink their strategy. Similarly,
control theory (e.g., Lord & Levy, 1994) proposes that discrepan-
cies between standards and feedback (i.e., errors) initiate an in-
creased allocation of attention to the task and that learning occurs
when these discrepancies are resolved.
Strengths and Limitations
Although our data show emotion control to be an effective
mediator of adaptive transfer performance, one may doubt whether
our measure of emotion control was unbiased because participants
filled out the emotion control items soon after they had completed
the performance test. More precisely, our results might be distorted
because of what is known as self-handicapping in test anxiety
research (e.g., Laux & Glanzmann, 1987): Poor-performing par-
ticipants might have indicated their emotion control to be low
simply because they were aware of their poor performance. This is
an issue that applies not only to the present or other error man-
agement training studies using questionnaire data to measure pro-
cesses (e.g., Debowski et al., 2001; Wood et al., 2000) but to
nearly all studies in which participants are asked for self-ratings of
psychological variables after performance assessment. Although
we are confident that not all interindividual variance in emotion
control was solely due to participants’ self-serving bias, this alter-
native explanation cannot be ruled out based on the self-ratings we
collected after the test phase. A better strategy would be to collect
emotion control data during the training session or right before the
performance phase or, alternatively, to use a method other than
self-reports that is less subject to self-serving bias.
Our measurement of metacognitive activity is unaffected by
participants’ potential self-serving bias. First, we derived this
measure not from participants’ self-ratings but from verbal proto-
col ratings that were blind to experimental condition and perfor-
mance scores. Second, for both methodological and conceptual
reasons, we were careful to make a time-lagged prediction: We
predicted later adaptive transfer performance from the metacog-
nitve activity measure that had been assessed earlier in training.
From a methodological perspective, this time lag has the advantage
that mediator and outcome variable are less likely to be con-
founded. From a conceptual perspective, we were interested in
processes not just concurrent with but predictive of adaptive trans-
fer performance.
Although we feel that our approach of using verbal protocol data
for process analyses was successful, one may raise objections
concerning the validity of thinking aloud protocols in general.
There has been an intensive debate as to whether thinking aloud
protocols reflect cognitive processes of participants or whether the
processes are critically altered (e.g., Schooler, Ohlsson, & Brooks,
1993). For two reasons, we are confident that the conclusions we
drew in our study from verbal protocol analysis are valid. First, we
carefully followed recommendations to avoid obtrusive instruc-
tions for thinking aloud (Ericsson & Simon, 1993; Taylor &
Dionne, 2000). Second, our results concerning the superiority of
error management training compared to error avoidant training
replicated results of other studies with similar effect sizes (Chill-
arege et al., 2003; Dormann & Frese, 1994; Frese et al., 1991;
Heimbeck et al., 2003; Ivancic & Hesketh, 2000; Nordstrom et al.,
1998; Wood et al., 2000). If our verbal protocol data were invalid,
this would imply that our study produced the same effects as other
687
SELF-REGULATION IN ERROR MANAGEMENT TRAINING

studies but that these effects were due to different processes—an
assumption that is of low plausibility.
An obvious drawback to our study is the composition and small
size of our sample. It should be noted, however, that other error
management training studies (e.g., Debowski et al., 2001; Wood et
al., 2000) as well as other studies using verbal protocol analysis
(e.g., Ball, Langholtz, Auble, & Sopchak, 1998) have relied on
small and sometimes even smaller samples. More important, we
found group differences in performance as well as process vari-
ables despite the relatively low statistical power due to the small
sample—a finding that corresponds to the considerable effect sizes
for these differences (Cohen, 1994; Kramer & Rosenthal, 1999;
Sonnentag, 1998). Also, as mentioned above, the superiority of
error management training compared to error avoidant training has
been found in other studies using larger samples. For reasons of
research economy, the use of thinking aloud analysis restricted our
sample to a limited number of volunteer students (despite the
rather small number of participants, about 18,000 coding units
were available from the verbal protocols).
Implications for Future Research
The present study focused on processes in training and did not
look at individual differences potentially affecting performance.
Some studies, however, suggest that participants may differentially
benefit from error management training or error avoidant training
depending on person characteristics such as cognitive ability,
openness to experience, or goal orientation (Gully et al., 2002;
Heimbeck et al., 2003). Future studies should look at differential
processes induced by such interactions of training condition and
person characteristics. Furthermore, given the strong predictive
power of emotion control and metacognitive activity, it would be
interesting to identify person characteristics that promote exertion
of emotion control and metacognitive activity during training. For
example, participants high in learning goal orientation may be
more likely to engage in effortful metacognitive activity during
training. Likewise, avoidance goal orientation, directed at the
avoidance of potential failure in the task and of negative judgment
by others, may be negatively related to emotion control in the face
of setbacks during training. Apart from the influence of goal
orientations as stable person characteristics, goal orientations as
temporal states may be affected by error management training as
well. Error management instructions emphasizing the positive role
of errors during learning may encourage a learning or mastery
orientation. In fact, there are studies that, among other instructions,
have used positive error statements similar to error management
instructions when manipulating learning goal orientation (e.g.,
Kozlowski et al., 2001).
Starting from the notion that errors instigate metacognitive
activity, our study demonstrated the power of overall metacogni-
tive activity in error management training. A more microanalytical
approach that examines actions and cognitions following errors
could further illuminate the processes of how errors instigate
metacognitive activity. For example, do errors trigger metacogni-
tive activity immediately? Or does it take several errors to finally
engage in effortful metacognitive activity? Another possibility is
that there is no simple one-to-one relation between errors and
metacognitive activity but that the low level of structure and the
frequent errors in error management training together induce a
general metacognitive processing mode during training. Future
research could use methods such as behavior observation or anal-
yses of concurrent video and verbal protocol data to gain insight
into the dynamics of errors and metacognitive activity in error
management training.
Such a microanalytical approach could also provide an insight
into component processes of metacognitive activity (i.e., planning,
monitoring, or evaluation) that are specifically important for the
effects of error management training. In post hoc exploratory
analyses, we identified three of the metacognitive subcategories
(cf. Table 1) as significant individual predictors of adaptive trans-
fer (partial correlations controlling for computer experience vari-
ables ⬎ .37, all ps ⬍ .05). These subcategories were “Planning—
generation of hypotheses” (Category 1a in Table 1),
“Monitoring— observing changes” (Category 1b), and “Evalua-
tion— explicit explanation” (Category 1d). Although the results of
these exploratory analyses should be interpreted with caution, they
might provide some initial directions for future research dealing
with metacognitive subprocesses.
Related to this issue, future research could use an error taxon-
omy to identify what types of errors lead to learning because not
all errors can be expected to be equally useful and informative or
to automatically lead to metacognitive activity and subsequent
enhanced performance. Within an action theory framework, an
error taxonomy differentiates errors as to the level of action
regulation involved in the error (Zapf et al., 1992; see also Ras-
mussen, 1982, who distinguishes between knowledge-based, rule-
based, and skill-based regulation). The levels of regulation run
from conscious regulation to automatic regulation. For example,
Zapf et al. validated a taxonomy that distinguished errors in
computer work. First, on the intellectual level, complex problem
analyses are regulated, which may lead to errors (e.g., planning
errors occur when the user selects the wrong course of action for
a task). Second, on the level of flexible action patterns, actions are
regulated by schemas (e.g., habit errors occur when a well-known
action is performed in the wrong situation). Third, on the senso-
rimotor level, stereotyped and automatic movements are organized
(e.g., typing errors or wrong movements of the computer mouse
occur here). We would expect, for example, that most learning in
error management training occurs from errors on higher levels of
regulation rather than from sensorimotor errors such as typos that
can be detected and corrected immediately. Future research could
use this taxonomy and identify types of errors that lead to learning,
both in the present and in other kinds of tasks.
Another related issue deals with the question of how overall
errors and errors of different types relate to adaptive transfer. The
concept of error management training suggests that the number of
errors should positively relate to subsequent transfer performance.
Empirically, however, we would expect not a positive but rather a
negative relationship because most errors are a result of lack of
knowledge, which is usually associated with poor performance.
Another possibility is a nonlinear relationship of errors and per-
formance that corresponds to the concept of an optimum number
of errors for transfer to occur. Ivancic and Hesketh (2000) found a
negative relationship between errors in training and performance
on a transfer task. At the same time, however, participants tended
not to repeat the same errors they made during training. These
results possibly indicate that errors and subsequent transfer per-
formance may be negatively related between persons but may be
688
KEITH AND FRESE

unrelated or even positively related within persons. Future research
could address these questions using a design that involves multiple
tasks and trials.
The present study, like other studies dealing with error manage-
ment training, compared error management training to error
avoidant training, the latter of which hindered participants from
making errors by means of step-by-step instructions. Another,
probably better known, training approach is behavior modeling,
which is based on Bandura’s (1986) social– cognitive theory.
Within Bandura’s framework, building self-efficacy by mastery
experiences is crucial for learning and performance. A training
program using behavior modeling usually involves a live or vid-
eotaped model demonstrating the correct strategies for task solu-
tion followed by the trainees’ imitation of the model’s behavior in
practice (e.g., Gist, Schwoerer, & Rosen, 1989). Thus, behavior
modeling is more structured than error management training. In
complex tasks with ambiguous feedback and in tasks that require
one single best strategy for task solution, behavior modeling prob-
ably results in better performance than error management training
(Debowski et al., 2001). These two training techniques, however,
do not necessarily have to be conceived of as mutually exclusive
alternatives. Also, just like behavior modeling, error management
aims at building self-efficacy. More specifically, error manage-
ment is directed at building self-efficacy in the face of problems
and errors that occur when working on new tasks. Future research
could look at self-efficacy expectations as outcomes of error
management training.
Another interesting issue for future research could be to exam-
ine the exact relation of emotion control and metacognitive activ-
ity. In our study, emotion control and metacognitive activity were
conceptualized as and shown to be independent mediators of
performance effects. The interrelation between the two variables
disappeared when training condition was taken into account. In
other words, error management training enhanced both emotional
and cognitive self-regulation, which in turn led to improved per-
formance. This result raises the question of how both processes are
intertwined. For example, does metacognitive activity positively
affect emotion control because participants engaged in metacog-
nitive activity “forget” to get angry about an error? Or does
emotion control serve as a prerequisite for metacognitive activity,
because only if participants’ negative emotions are controlled can
metacognitive activity be initiated? Theoretically, these kinds of
questions go beyond the academic convention to describe emo-
tional and cognitive processes as distinct phenomena using differ-
ent theoretical models. We believe that the self-regulation perspec-
tive adopted in this article provides a framework for integrating
emotions and cognitions into a common model.
Implications for Theory and Practice
Our study corroborates the notion that emotional and cognitive
self-regulation mediates the effectiveness of error management
training. This finding has important implications for both theory
and practice. From a theoretical perspective, training researchers
have always been interested not only in the question of whether
training works but also in why it works (e.g., Goldstein & Ford,
2002). Only a very few error management training studies to date
have looked explicitly at the processes underlying error manage-
ment training effectiveness, and none of these have provided
conclusive results (Debowski et al., 2001; Wood et al., 2000). Our
study contributes to a better theoretical understanding of why error
management training leads to better performance than does error
avoidant training.
From a practical perspective, identifying effective mediators in
training is particularly important because this information is useful
for modifying error management training or for adapting its prin-
ciples to another area. Our results suggest that error management
training is effective because it provides the opportunity to practice
the metacognitive activities of planning, monitoring, and evalua-
tion—skills that prove useful when it comes to tasks that require a
new solution. Practitioners may consider explicitly integrating
modules of error management training into the training process by
giving participants the opportunity to make errors by working on
difficult training tasks on their own and at the same time encour-
aging them to use their errors as a learning device.
Most important, our research highlights the critical role of the
kind of information processing participants engage in during train-
ing (cf. Hesketh & Ivancic, 2002). Stated differently, in our study
the crucial question for adaptive transfer was not what material
was learned during the training session (the material was identical
in all training conditions) but rather how it was learned. When
planning a training intervention, practitioners may increase the
effectiveness of the training by focusing their attention not only on
the training material to be covered but also on the kind of infor-
mation processing that is most promising for transfer to occur.
We are confident that the principles of error management train-
ing can be incorporated into areas other than computer training,
although research concerning other areas is rare (Gully et al., 2002;
Ivancic & Hesketh, 2000). We suggest that error management
training is useful whenever the material to be learned cannot be
covered completely during the training session and, consequently,
participants will need to “learn to learn” when confronted with
new tasks. This is related to the principle of transfer appropriate
processing, which, as mentioned above, postulates that those pro-
cesses required on transfer tasks should be practiced during train-
ing (Morris et al., 1977). The present transfer task required
discovery-type activities involving learning from errors because
solutions to problems distinct from those worked on during the
training session had to be found. Consequently, error management
training, which required the same type of activities during training,
resulted in superior performance relative to error avoidant training,
which taught the correct solutions during training. In trainings
covering a relatively small amount of material that is highly
structured, however, it is probably more economical to teach the
correct strategies directly because exploring and learning from
errors may be too time-consuming. Related to this issue, in tasks
that require one single best strategy for task solution, behavior
modeling probably results in better performance than does error
management training (Debowski et al., 2001). It should be kept in
mind that although error management training may be successful
in promoting transfer performance, training performance itself
may not be better or may even be worse than in error avoidant
training. Not only will participants in error management training
make more errors during training—after all, they are told to do
so— but it will also take them longer to solve the tasks on their
own or, if time is limited, they will solve fewer tasks during the
same training period than their counterparts in error avoidant
training.
689
SELF-REGULATION IN ERROR MANAGEMENT TRAINING

Also, when tasks are very complex, error management training
should be combined with elements of guided training (Bell &
Kozlowski, 2002) because, given the low level of structure and
guidance in error management training, participants may run the
risk of developing incorrect conceptualizations of the training
content (Frese, 1995; Mayer, 2004). For example, a guided ap-
proach comprising assistance and external feedback by the trainer
could be used to develop basic competencies that subsequent error
management training could build on (Debowski et al., 2001).
Finally, high-fidelity task feedback is probably a prerequisite for
error management training because errors can serve as informative
feedback only in systems that allow self-regulated error detection
and correction. Many of the studies that have successfully applied
error management training have used computer tasks that usually
provide clear task feedback. For example, if a participant takes
action to insert a table into a document, he or she will immediately
see whether the action leads to the desired goal or not. Other tasks
that lack the kind of clarity of task-inherent feedback may not be
well suited for error management training. In a social skills train-
ing, for example, error management training may not be helpful if
a participant is not able to interpret others’ reactions to his or her
actions or speech correctly, so that augmented feedback by a
trainer or by fellow participants may be required. On the other
hand, once basic interpretation skills are developed, error manage-
ment training may be effective in promoting transfer because in
real-life interactions augmented feedback is not provided. It is our
hope that this work encourages researchers and practitioners to
take up error management training principles and apply them to
other areas of skill acquisition.
References
Ball, C. T., Langholtz, H. J., Auble, J., & Sopchak, B. (1998). Resource-
allocation strategies: A verbal protocol analysis. Organizational Behav-
ior and Human Decision Processes, 76, 70 – 88.
Bandura, A. (1986). Social foundations of thought and action: A social
cognitive theory. Englewood Cliffs, NJ: Prentice Hall.
Bell, B. S., & Kozlowski, S. W. J. (2002). Adaptive guidance: Enhancing
self-regulation, knowledge, and performance in technology-based train-
ing. Personnel Psychology, 55, 267–306.
Berardi-Coletta, B., Buyer, L. S., Dominowski, R. L., & Rellinger, E. R.
(1995). Metacognition and problem solving: A process-oriented ap-
proach. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 21, 205–223.
Brown, A. L., Bransford, J. D., Ferrara, R. A., & Campione, J. C. (1983).
Learning, remembering, and understanding. In J. H. Flavell & E. M.
Markman (Eds.), Handbook of child psychology (Vol. 3, pp. 77–166).
New York: Wiley.
Bruner, J. S. (1966). Toward a theory of instruction. Cambridge, MA:
Harvard University Press.
Carver, C. S., Scheier, M. F., & Weintraub, J. K. (1989). Assessing coping
strategies: A theoretically based approach. Journal of Personality and
Social Psychology, 56, 267–283.
Chillarege, K. A., Nordstrom, C. R., & Williams, K. B. (2003). Learning
from our mistakes: Error management training for mature learners.
Journal of Business and Psychology, 17, 369 –385.
Cohen, J. (1994). The earth is round ( p ⬍ .05). American Psychologist, 12,
997–1003.
Davidson, J. E., & Sternberg, R. J. (1998). Smart problem-solving: How
metacognition helps. In D. J. Hacker, J. Dunlosky, & A. C. Graesser
(Eds.), Metacognition in educational theory and practice (pp. 47– 68).
Mahwah, NJ: Erlbaum.
Debowski, S., Wood, R. E., & Bandura, A. (2001). Impact of guided
exploration on self-regulatory mechanisms and information acquisition
through electronic search. Journal of Applied Psychology, 86, 1129 –
1141.
Dormann, T., & Frese, M. (1994). Error management training: Replication
and the function of exploratory behavior. International Journal of
Human–Computer Interaction, 6, 365–372.
Ericsson, K. A., & Simon, H. A. (1993). Protocol analysis: Verbal reports
as data. Cambridge, MA: MIT Press.
Fisher, K. M., & Lipson, J. I. (1986). Twenty questions about student
errors. Journal of Research in Science Teaching, 23, 783– 803.
Folkman, S., & Lazarus, R. S. (1985). If it changes it must be a process:
Study of emotion and coping during three stages of a college examina-
tion. Journal of Personality and Social Psychology, 48, 150 –170.
Ford, J. K., Smith, E. M., Weissbein, D. A., Gully, S. M., & Salas, E.
(1998). Relationships of goal orientation, metacognitive activity, and
practice strategies with learning outcomes and transfer. Journal of Ap-
plied Psychology, 83, 218 –233.
Frese, M. (1995). Error management in training: Conceptual and empirical
results. In C. Zucchermaglio, S. Bagnara, & S. U. Stucky (Eds.), Orga-
nizational learning and technological change, Series F: Computer and
systems sciences (Vol. 141, pp. 112–124). Berlin, Germany: Springer.
Frese, M., Brodbeck, F. C., Heinbokel, T., Mooser, C., Schleiffenbaum, E.,
& Thiemann, P. (1991). Errors in training computer skills: On the
positive function of errors. Human–Computer Interaction, 6, 77–93.
Frese, M., & Zapf, D. (1994). Action as the core of work psychology: A
German approach. In H. C. Triandis, M. D. Dunette, & L. M. Hough
(Eds.), Handbook of industrial and organizational psychology (Vol. 4,
pp. 271–340). Palo Alto, CA: Consulting Psychologists Press.
Gist, M. E., Schwoerer, C., & Rosen, B. (1989). Effects of alternative
training methods on self-efficacy and performance in computer software
training. Journal of Applied Psychology, 74, 884 – 891.
Goldstein, I. L., & Ford, J. K. (2002). Training in organizations: Needs
assessment, development, and evaluation. Belmont, CA: Wadsworth/
Thomson Learning.
Goodman, J. S. (1998). The interactive effects of task and external feed-
back on practice performance and learning. Organizational Behavior
and Human Decision Processes, 76, 223–252.
Grandey, A. A., Dickter, D. N., & Sin, H.-P. (2004). The customer is not
always right: Customer aggression and emotion regulation of service
employees. Journal of Organizational Behavior, 25, 397– 418.
Greif, S., & Keller, H. (1990). Innovation and the design of work and
learning environments: The concept of exploration in human–computer
interaction. In M. A. West & J. A. Farr (Eds.), Innovation and creativity
at work: Psychological and organizational strategies (pp. 231–249).
Oxford, England: Wiley.
Gully, S. M., Payne, S. C., Koles, K. L. K., & Whiteman, J. A. K. (2002).
The impact of error management training and individual differences on
training outcomes: An attribute–treatment interaction perspective. Jour-
nal of Applied Psychology, 87, 143–155.
Hacker, W. (1998). Allgemeine Arbeitspsychologie: Psychische Regulation
von Arbeitsta¨tigkeiten [General industrial psychology: Mental regulation
of working activities]. Bern, Switzerland: Huber.
Heimbeck, D., Frese, M., Sonnentag, S., & Keith, N. (2003). Integrating
errors into the training process: The function of error management
instructions and the role of goal orientation. Personnel Psychology, 56,
333–361.
Hesketh, B. (1997). Dilemmas in training for transfer and retention. Ap-
plied Psychology: An International Review, 46, 317–339.
Hesketh, B., & Ivancic, K. (2002). Enhancing performance through train-
ing. In S. Sonnentag (Ed.), Psychological management of individual
performance (pp. 249 –265). New York: Wiley.
Hoyle, R. H., & Kenny, D. A. (1999). Sample size, reliability, and tests of
690
KEITH AND FRESE

statistical mediation. In R. H. Hoyle (Ed.), Statistical strategies for small
sample research (pp. 195–222). Thousand Oaks, CA: Sage.
Ivancic, B., & Hesketh, K. (2000). Learning from error in a driving
simulation: Effects on driving skill and self-confidence. Ergonomics, 43,
1966 –1984.
James, L. R., & Brett, J. M. (1984). Mediators, moderators, and tests for
mediation. Journal of Applied Psychology, 69, 307–321.
Jo¨reskog, K. G., & So¨rbom, D. (1996). LISREL 8: User’s reference guide.
Chicago: Scientific Software International.
Kanfer, R., & Ackerman, P. L. (1989). Motivation and cognitive abilities:
An integrative/aptitude-treatment interaction approach to skill acquisi-
tion. Journal of Applied Psychology, 74, 657– 690.
Kanfer, R., Ackerman, P. L., & Heggestad, E. D. (1996). Motivational
skills and self-regulation for learning: A trait perspective. Learning and
Individual Differences, 8, 185–209.
Karoly, P. (1993). Mechanisms of self-regulation: A systems view. Annual
Review of Psychology, 44, 23–52.
King, A. (1991). Effects of training in strategic questioning on children’s
problem-solving performance. Journal of Educational Psychology, 83,
307–317.
Kluger, A. N., & DeNisi, A. (1996). Effects of feedback intervention on
performance: A historical review, a meta-analysis, and a preliminary
feedback intervention theory. Psychological Bulletin, 119, 254 –284.
Kozlowski, S. W. J., Gully, S. M., Brown, K. G., Salas, E., Smith, E. M.,
& Nason, E. R. (2001). Effect of training goals and goal orientation traits
on multidimensional training outcomes and performance adaptability.
Organizational Behavior and Human Decision Processes, 85, 1–31.
Kramer, S. H., & Rosenthal, R. (1999). Effect sizes and significance levels
in small-sample research. In R. H. Hoyle (Ed.), Statistical strategies for
small sample research (pp. 59 –79). Thousand Oaks, CA: Sage.
Laux, L., & Glanzmann, P. G. (1987). A self-presentational view of test
anxiety. In R. Schwarzer & H. M. van der Ploeg (Eds.), Advances in test
anxiety research (Vol. 5, pp. 31–37). Berwyn, PA: Swets North Amer-
ica.
Lazarus, R. S., & Folkman, S. (1984). Stress, appraisal, and coping. New
York: Guilford Press.
Lord, R. G., & Levy, P. E. (1994). Moving from cognition to action: A
control theory perspective. Applied Psychology: An International Re-
view, 43, 335–398.
MacKinnon, D. P., Lockwood, C. M., Hoffman, J. M., West, S. G., &
Sheets, V. (2002). A comparison of methods to test mediation and other
intervening variable effects. Psychological Methods, 7, 83–104.
Mayer, R. E. (2004). Should there be a three-strike rule against pure
discovery learning? American Psychologist, 59, 14 –19.
McInerney, V., McInerney, D. M., & Marsh, H. W. (1997). Effects of
metacognitive strategy training within a cooperative group learning
context on computer achievement and anxiety: An aptitude–treatment
interaction study. Journal of Educational Psychology, 89, 686 – 695.
Morris, C. D., Bransford, J. D., & Franks, J. J. (1977). Levels of processing
versus transfer appropriate processing. Journal of Verbal Learning and
Verbal Behavior, 16, 519 –533.
Muraven, M., & Baumeister, R. F. (2000). Self-regulation and depletion of
limited resources: Does self-control resemble a muscle? Psychological
Bulletin, 126, 247–259.
Nordstrom, C. R., Wendland, D., & Williams, K. B. (1998). “To err is
human”: An examination of the effectiveness of error management
training. Journal of Business and Psychology, 12, 269 –282.
Pintrich, P. R., & De Groot, E. V. (1990). Motivational and self-regulated
learning components of classroom academic achievement. Journal of
Educational Psychology, 82, 33– 40.
Rasmussen, J. (1982). Human errors: A taxonomy for describing human
malfunction in industrial installations. Journal of Occupational Acci-
dents, 4, 311–335.
Reason, J. (1990). Human error. Cambridge, Eng1and: Cambridge Uni-
versity Press.
Richards, J. M., & Gross, J. J. (2000). Emotion regulation and memory:
The cognitive costs of keeping one’s cool. Journal of Personality and
Social Psychology, 79, 410 – 424.
Rybowiak, V., Garst, H., Frese, M., & Batinic, B. (1999). Error Orientation
Questionnaire (EOQ): Reliability, validity, and different language equiv-
alence. Journal of Organizational Behavior, 20, 527–547.
Schmidt, A. M., & Ford, J. K. (2003). Learning within a learner control
environment: The interactive effects of goal orientation and metacogni-
tive instruction on learning outcomes. Personnel Psychology, 56, 405–
429.
Schmidt, R. A., & Bjork, R. A. (1992). New conceptualizations of practice:
Common principles in three paradigms suggest new concepts for train-
ing. Psychological Science, 3, 207–217.
Schooler, J. W., Ohlsson, S., & Brooks, K. (1993). Thoughts beyond
words: When language overshadows insight. Journal of Experimental
Psychology: General, 122, 166 –183.
Schraw, G., & Moshman, D. (1995). Metacognitive theories. Educational
Psychology Review, 7, 351–371.
Schunk, D. H., & Zimmerman, B. J. (Eds.). (1994). Self-regulation of
learning and performance: Issues and educational applications. Hills-
dale, NJ: Erlbaum.
Sonnentag, S. (1998). Expertise in professional software design: A process
study. Journal of Applied Psychology, 83, 703–715.
Taylor, K. L., & Dionne, J. P. (2000). Assessing problem-solving strategy
knowledge: The complementary use of verbal protocols and retrospec-
tive debriefing. Journal of Educational Psychology, 92, 413– 425.
Totterdell, P., & Holman, D. (2003). Emotion regulation in customer
service roles: Testing a model of emotional labor. Journal of Occupa-
tional Health Psychology, 8, 55–73.
van der Linden, D., Sonnentag, S., Frese, M., & van Dyck, C. (2001).
Exploration strategies, performance, and error consequences when learn-
ing a complex computer task. Behavior and Information Technology, 20,
189 –198.
Wanberg, C. R., Kanfer, R., & Rotundo, M. (1999). Unemployed individ-
uals: Motives, job-search competences, and job search constraints as
predictors of job seeking and reemployment. Journal of Applied Psy-
chology, 84, 897–910.
Wood, R. E., Kakebeeke, B. M., Debowski, S., & Frese, M. (2000). The
impact of enactive exploration on intrinsic motivation, strategy, and
performance in electronic search. Applied Psychology: An International
Review, 49, 263–283.
Zapf, D., Brodbeck, F. C., Frese, M., Peters, H., & Pru¨mper, J. (1992).
Errors in working with office computers: A first validation of a taxon-
omy for observed errors in a field setting. International Journal of
Human–Computer Interaction, 4, 311–339.
Received March 4, 2004
Revision received August 12, 2004
Accepted August 23, 2004 䡲
691
SELF-REGULATION IN ERROR MANAGEMENT TRAINING