INTEGRATING ERROR S INTO THE TRAINING
PROCESS: THE FUNCTION
OF
ERROR
MANAGEMENT INSTRUCTION S AN D THE ROLE
OF GOAL ORIENTATION
DOERTE HEIMBECK, MICHAEL FRESE, SABINE SONNENTAG
Department
of
Psychology
University
of
Amsterdam, Netherlands
NINA KEITH
Department
of
Psychology
University
of
Giessen, Germany
Error management training explicitly allows participants
to
make
er-
rors.
We
examined the effects of error management instructions
("rules
of thumb" designed
to
reduce
the
negative emotional effects
of er-
rors),
goal orientation (learning goal, prove goal,
and
avoidance goal
orientations)
and
attribute
x
treatment interactions
on
performance.
A randomized experiment with 87 participants consisting
of
3
training
procedures
for
learning
to
work with
a
computer program was
con-
ducted:
(a)
error training with error management instructions,
(b) er-
ror training without error management instructions;
and (c) a
group
that was prevented from making errors. Results showed that short-
and medium-term performance (near
and
far
transfer) was superior
for participants
of
the error training that included error management
instructions, compared with
the
two other training conditions. Thus,
error management instructions were crucial
for
the high performance
effects
of
error training. Prove
and
avoidance goal orientation inter-
acted with training conditions.
Errors appear in nearly
every
learning process because of insufficient
knowledge and
skills.
Moreover, errors appear not just
as a
result
of
too
little knowledge
but
also
as a
consequence
of
inappropriate goals
and
plans,
interruptions during
the
action process,
and
inaccurate interpre-
tation
of
system feedback
(Zapf,
Brodbeck, Frese, Peters,
&
Prumper,
1992).
Errors
are
often frustrating
and can
lead
to
anger
and
despair
(Brodbeck,
Zapf,
Prumper, & Frese, 1993).
This study is based on
the
dissertation
of
the first author in partial fulfillment
for the
doctoral degree
at
the
University
of
Amsterdam. Doerte Heimbeck has since moved
to
Florida, U.S., Michael Frese has moved
to
University
of
Giessen, Germany,
and
Sabine
Sonnentag has moved
to
University
of
Braunschweig, Germany. We are indebted
to
Rick
DeShon
for
a
critical reading
of
a
prior version
of
this
article.
Correspondence
and
requests
for
reprints should
be
addressed
to
Michael Frese,
University
of
Giessen, Germany Otto-Behaghel-Str.
lOF,
35394 Giessen, Germany;
Michael.Frese@Psychol.Uni-Giessen.de
333
First publ. in: Personnel Psychology 56 (2003), 2, pp. 333-361
The definitive version is available at www.blackwell-synergy.com
Konstanzer Online-Publikations-System (KOPS)
URN: http://nbn-resolving.de/urn:nbn:de:bsz:352-opus-83874
URL: http://kops.ub.uni-konstanz.de/volltexte/2009/8387/
334
Because errors appear frequently in the learning process, one would
imagine that there is a strong literature on errors and the function of
errors in training. However, this is not the case. Only very recently has
the function of errors in training been explicitly discussed (e.g.. Gully,
Payne, Koles, & Whiteman, 2002; Ivancic & Hesketh, 2000). Tradition-
ally, there was little explicit discussion of errors in the training literature
and the little discussion there was, was primarily negative. For example.
Skinner (1953) equated errors with punishment and argued that punish-
ment leads to emotional arousal and to temporary suppression of be-
havior but does not tell the learner what needs to be done. Therefore,
Skinner (1953) argued for an error prevention approach that minimizes
errors;
eventually his approach led to the development of the famous
programmed learning machines that were supposed to maximize positive
reinforcement and shaping by positive reinforcement (Skinner, 1968).
Critiques of exploratory learning (e.g., Ausubel, 1968) also warned of
the dangers of errors in the learning process; they suggested (among
others) that allowing people to make errors lets them practice incorrect
and inefficient approaches; this was one reason that exploratory learn-
ing was deemed to be inefficient in comparison to teaching appropriate
cognitive models (e.g., Ausubel, 1968; similarly also Debowski, Wood,
& Bandura, 2001). Discussions with software trainers and informal ob-
servations of their training processes have also convinced us that most
software trainers attempt to prevent errors. Software trainers thought
that errors would lead to negative effects and they studiously avoided
them whenever possible. Thus, errors are often seen to be "nuisance"
factors that need to be prevented in training and, at the same time, there
are relatively few explicit studies of errors in training.
The approach taken in this article is to counter the error prevention
approach. We conceptualize errors as as source of negative feedback
and argue that errors can have a positive and informative function in
training. Thus, errors should not be avoided but explicitly incorporated
into the training process. The training concept suggested here is one
of error training—a training method suggested by Frese and coauthors,
which increases the exposure to errors but which also produces a safe
environment for errors to appear (Frese, 1995; Frese & Altmann, 1989;
Frese et al., 1991). Error training proved to lead to higher performance
than a training that reduced the chances to make errors, the so-called
error-avoidant training (e.g., Frese et al., 1991). Recent research has
been a bit more skeptical about error training (Debowski et al., 2001;
Gully et al., 2002). Debowski et al. (2001) showed that enactive explo-
ration (an approach similar to error training) produced weaker perfor-
mance effects than what they called guided exploration. However, with
more extensive training time, the performance results of both training
335
procedures may become similar
(Wood,
Kakebeeke, Debowski,
&
Frese,
2000).
Debowski et al. (2001) also discussed self-regulatory and affec-
tive processes and suggested that such processes should be studied more
frequently in training research. Gully et
al.
(2002) did not find a general
performance superiority of error training but rather attribute-treatment
interactions; positive performance effects of error training
(as
compared
to an error avoidance training) occurred for high ability and high open-
ness to experience people. Gully et al. (2002), therefore, argue to in-
clude interindividual difference variables into training design research.
The current study takes the error training approach
as a
starting point
and advances our knowledge in four
areas:
First, it replicates the studies
done so far; this replication is of special interest because some other
studies have revealed a somewhat more ambiguous picture about the
effectiveness of error training (Debowski et
al.,
2001).
Second, the study
explicitly examines the function of the error management instructions
that are supposed to reduce the negative emotional effects of errors.
Third, the study examines the effects of error training on two transfer
outcomes, including
a
far transfer outcome obtained one week following
the conclusion of the training. Fourth, we examine goal orientation as a
specific attribute in an attribute-treatment interaction.
Error
Training:
Conceptual and Empirical Issues
The error training concept starts from the assumption that errors—
defined as deviations from goals that are potentially avoidable—provide
feedback and that negative feedback is a necessary prerequisite for
learning (Frese &
Zapf,
1994). Feedback as a learning device is also
the basis of exploratory learning (Bruner, 1966; Carroll, Mack, Lewis,
Grischkowsky, & Robertson, 1985; Greif & Janikowski, 1987; Greif &
Keller, 1990). Exploratory learning
is
similar to error training, although
error training tends to present more difficult tasks earlier
in
training. Al-
though most error training programs also increase difficulty level during
the course of the training, both the starting and the endpoint tasks are
usually more difficult in error training than within the exploratory learn-
ing paradigm. A first evaluation of error training of a software system
was
provided
by
Frese et
al.
(1991); in
this
study,
the error training group
had ample opportunities for making errors by having to solve tasks that
were too difficult for the participants. Error training provides trainees
with only the most basic information and minimal instructions about the
program taught during the training. The error training group in Frese
et al.'s study received only a list of commands that could be used to deal
with the tasks. Therefore, this group had to try out several approaches
and to explore the system until they found a
way
to solve the
tasks.
Dur-
336
ing this process, participants made a number of errors. We call them
errors because participants could not reach the goal of solving the tasks
and knew that there was, in principle, a solution available (cf. Frese &
Zapf,
1994, for a concept of errors). Examples of errors are that partic-
ipants attempted to use the wrong commands (e.g., CTRL T instead of
CTRL D to delete a letter) or that they used a keystroke combination
that led the participants into unknown parts of the software (e.g., getting
into the area of commands for margin setting instead of changing letters
in a text). In contrast, the error avoidant group was given detailed in-
structions of how to solve the tasks. Because these detailed instructions
were correct, they did not make any errors; if an error occurred in spite of
these detailed prescriptions of actions (e.g., because of a typing error or
wrong reading of the instructions), the trainer intervened and corrected
the situation immediately—making sure that the participants had min-
imal exposure to an error situation. After the training, a performance
test was done that examined whether participants could solve everyday
tasks of moderate to high difficulty. For example, in the word processing
training, participants were tested to set margins or design a text (Frese
et al., 1991). The error training group showed higher performance in
tasks of medium and high difficulty than the group that received error-
avoidant training (Dormann & Frese, 1994; Frese et al., 1991; Greif &
Janikowski, 1987; Ivancic & Hesketh, 2000). Performance in easy tasks
did not differ between error and error avoidant trainings (Frese, 1995).
Easy tasks require a low degree of skill and, therefore, do not normally
lead to errors; thus, they are less affected by error training.
Some researchers have argued that error training would be particu-
larly useful to increase transfer of training (Frese, 1995; Hesketh, 1997).
Because successful transfer represents the most important training out-
come (Baldwin & Ford, 1988; Goldstein, 1992), an effective training
method should produce high transfer. Transfer implies that there is a
gap between learning tasks and transfer tasks (Gick & Holyoak, 1987;
Kraiger, 1995). Gick and McGarry (1992) found a positive relationship
between failure during the source problem solving and transfer in an
analogous problem-solving experiment. Most training research assumes
that learning and transfer occur in different environments: The learning
context is supposed to be safe and well structured. Tasks are carefully
sequenced and described, and feedback is provided in such a way that
skill acquisition is facilitated. In contrast, the transfer context is open,
disruptive, and ambiguous (Kraiger, 1995). Thus, the transfer situation
introduces the chance to make errors. Trainees have to deal with errors,
explore new aspects of the program, and solve new and more difficult
tasks in the transfer environment. In this environment, help is typically
not provided, nor is there enough time to look through an extensive man-
337
ual (Brodbeck et al., 1993). Error training reduces the distance be-
tween training and transfer environments as it allows and encourages
errors to occur in the training process, teaching skills to deal with errors
in the training context. Thus, error training should facilitate transfer
performance.
Error Management Instructions
In some ways, error training is frustrating to its participants. Frese
et al. (1991) presented in the 5th and 6th training hour tasks that had
been pretested not to be achievable by any of the participants (for ex-
ample, to write the Danish currency "0re" with the system Wordstar of
the 1980s). Participants usually attempted various approaches but none
of them achieved the goal of writing "0re." Participants were aware
that, in principle, there was a solution. To reduce the frustrating role
of this situation, error management instructions were used
by
explaining
to the participants that errors were necessary to leam and presenting
the following four error management instructions: "I have made an er-
ror. Great!"; "There is always
a way
to leave the error situation"; "Look
at the screen"; "I watch what is on the screen and what is changing."
These instructions were developed to counter ineffective emotions that
appeared in pilot studies in which people's reactions to errors were ob-
served. For
example,
people were sometimes deeply frustrated by errors
and they had the feehng that they could not get out of certain "error
traps."
People also tended not to look at the screen and not to observe
what had changed on the screen and thereby minimized the feedback
they received from errors (Frese
&
Altmann, 1989). The error manage-
ment instructions were displayed prominently and were reiterated
by
the
trainers throughout the training.
Why should error training with error management instructions lead
to high performance? Our argument is twofold. First, we suggest that
using errors in training helps to develop a good operative mental model.
However, this positive effect can only appear if training participants are
not hindered by the negative emotional effects of
errors.
Second, these
negative effects can be countered by error management instructions.
Thus,
only the combination of error training and the error management
instructions should lead to high performance.
First, the development of a good mental model is enhanced by error
training. Errors provide negative feedback and point out areas of mis-
understandings. Therefore, errors can lead people to develop a better
mental model of the system (it
is
an
operative
mental model as the model
is supposed to regulate behavior adequately; Frese &
Zapf,
1994). Er-
rors interrupt the behavior stream and the interruption of the behavior