-
PERSONNEL PSYCHOLOGY1993.46
INDIVIDUAL AND SITUATIONAL INFLUENCES ON THEDEVELOPMENT OF
SELF-EFFICACY: IMPLICATIONSFOR TRAINING EFFECTIVENESS
JOHN E. MATHIEU, JENNIFER W. MARTINEAUPennsylvania State
University
SCOTT I. TANNENBAUMState University of New York at Albany
We proposed a model that included individual and situational
an-tecedents of self-efficacy development during training. Initial
perfor-mance and self-efficacy levels, achievement motivation, and
choicewere examined as individual variables. Constraints,
operationalizedat both the individual and aggregate levels of
analysis, were exam-ined as situational influences. Mid-course
efficacy was hypothesizedto have positive linear relationships with
training reactions and subse-quent performance, and an interactive
relationship with performancewhen training reactions were
considered as a moderator. Survey datawere gathered at two points
in time from 215 students enrolled in 15eight-week long university
bowling classes. All of the hypothesized an-tecedents of mid-course
self-efficacy were significant except aggregateand individual
situational constraints, although both constraints re-lated
negatively to training reactions. Time 2 self-efficacy exhibited
sig-nificant positive influences on training reactions and
subsequent per-formance, but the hypothesized moderated
relationship was not sup-ported.
Recent theorizitig and research concerning influences on
training ef-fectiveness have moved beyond a focus on the training
program andits attributes, and adopted a more global or systems
perspective (Tkn-nenbaum & Yukl, 1992). For example, Campbell
(1988) argued thatthe effects of individual and situational
variables on training effective-ness should be considered. He noted
that individual variables such astrainees' goals; their levels of
self-efficacy before, during, and after train-ing; and the
self-regulatory behaviors of trainees could all impact theultimate
effectiveness of a program. Furthermore, Campbell suggestedthat
situational influences such as reinforcement and punishment
contin-gencies, socialization, and group processes that influence
trainees' goals.
We thank three anonymous reviewers for their helpful comments on
an earlier versionof this article.
Requests for reprints and other correspondence concerning this
article should be sentto John Mathieu, 437 Moore Building,
Department of Psychology, Pennsylvania StateUnivereity, University
Park, PA 16802 or via BITNET to NI7 @ PSUVM.
COPYRIGHT © 1993 PERSONNEL PSYCHOLOGiY. INC.
125
-
126 PERSONNEL PSYCHOLOGY
self-efficacy, and instrumentality judgments should be
incorporated intotraining effectiveness studies. In short, training
research is expanding in-quiry beyond the method or learning
techniques used by any particularprogram, and is beginning to
consider the larger context within whichtraining programs reside.
Accordingly, the purpose of the present studyis to examine more
thoroughly the role of individual and situational an-tecedents of
training effectiveness in a skills development program.
Previous work concerning the effects of individual variables on
train-ing effectiveness has yielded many significant insights. In
particular,studies have found significant correlations between
various operational-izations of training related motivation and
subsequent affective reac-tions (e.g., Mathieu, Tannenbaum &
Salas, 1992; Tannenbaum, Math-ieu, Salas & Cannon-Bowers,
1991), learning (e.g., Baldwin, Magjuka &Loher, 1991; Clark,
1990; Mathieu et al., 1992), and performance (e.g..Hicks &
Klimoski, 1987; Rails & Klein, 1991). One related constructthat
has been receiving well-deserved attention in the training
researchhterature is self-efficacy, or a trainee's perceived
capability to performa specific task (Gist, 1987). In the current
study, we present a model ofantecedents of self-efficacy
development during training, and the subse-quent influence of
self-efficacy on trainees' reactions and performanceimprovement.
Below we review the existing research concerning self-efRcacy and
its influence on various training outcomes in order to de-velop our
hypothesized model.
Self-Efficacy and Training Effectiveness
Self-efficacy is ". . . defined as people's judgments of their
capabilitiesto organize and execute courses of action required to
attain designatedtypes of performances. It is concerned not with
the skills one has butwith judgments of what one can do with
whatever skills one possesses"(Bandura, 1986, p. 391).
Self-efficacy has been shown to predict perfor-mance in computer
software training (Gist, Schwoerer, & Rosen,
1989),interpersonal skills training (Gist, Stevens & Bavetta,
1991), and mili-tary training programs (Eden & Ravid, 1982;
Tannenbaum et al., 1991).Furthermore, self-efficacy levels at the
conclusion of training have ex-hibited significant correlations
with post-training transfer and job per-formance measures. Ford,
Quinones, Sego and Sorra (1992) found thatself-efficacy levels
among Air Force mechanic trainees were related totask opportunities
in their initial job assignment. Frayne and Latham(1987; Latham
& Frayne, 1989) found that self-efficacy predicted
jobattendance both during training and 9 months after the
completion oftraining. In short, previous research has
substantiated the important
-
JOHN E. MATHIEU ET AL. 127
Self Efricacy
044)
Parformanco
IndividualSit. Conatrainti
Aggragat*Sil. Conatrainti
N-21G a p
-
128 PERSONNEL PSYCHOLOGY
Model of the Role of Self-Efficacy on Training Effectiveness
Bandura (1986) argued that self-efficacy develops gradually
throughrepeated task-related experiences. Elsewhere, he (Bandura,
1989,1991)and Gist and Mitchell (1992) have outlined several
antecedents of effi-cacy development. Gist and Mitchell classify
these antecedents as ei-ther intemal (i.e., individual) or extemal
(i.e., situational). Mathieu etal. (1992) advanced a similar
framework when discussing antecedentsof a
valence-instrumentality-expectancy training motivation compositeas
either individual or situational. Included in Gist and Mitchell's
in-ternal category are individual attributes such as knowledge,
skills andabilities, performance-related strategies, personality
factors, and moodstates. For example, Bandura (1991) argued that
individuals monitortheir experiences and base subsequent efficacy
judgments, in part, onthe extent to which they perceive their
performances were attributableto their abilities and effort.
Further, he argued that personality needs,such as achievement
motivation, exert indirect influence on performanceby impacting
efficacy perceptions (Bandura, 1989). In the current study,we
examine previous performance, the extent to which individuals
choseto participate in the program, and achievement motivation as
individualinfluences on efficacy development during training.
Gist and Mitchell's (1992) extemal category includes influences
suchas task attributes (e.g., difficulty, complexity),
distractions, and norma-tive information (e.g., models,
persuasion). For example, Bandura andWood (1989) found that
individuals' efficacy levels decreased when theywere placed in a
constrained situation where they were led to believe thatthey had
little control. In the current study, we consider different typesof
situational constraints as extemal influences on efficacy
developmentduring training.
One factor that cannot be neatly classified as either intemal or
exter-nal is enactive mastery; which is defined as individuals'
previous experi-ences in the same or similar situations (Bandura,
1986). While primarilyan individual variable, the relative
difficulty, complexity, and other at-tributes of previous task
experiences will also influence individuals' per-ceptions of
enactive mastery. Thus, while enactive mastery varies
perindividual, it is partly determined by extemal characteristics
such as, thetask assignments, resource availability, or the
experimental conditionsto which subjects have been previously
assigned. For sake of discussion,however, we consider enactive
mastery experiences as an individual vari-able.
-
JOHN E. MATHIEU ET AL. 129
Hypothesized Antecedents of Self-Efficacy Development
Initial self-efficacy. Participants' initial levels of
self-efficacy (i.e., self-efficacy-1), as assessed at the beginning
of training, were hypothesizedto relate positively to later levels
of self-efficacy (i.e., self-efficacy-2).Gist et al. (1991), Latham
and Frayne (1989), and Tiinnenbaum et al.(1991) have provided
evidence that individuals' efficacy levels exhibitsome consistency
over time. That is, self-efficacy levels are iikely tobe somewhat
stable yet are also somewhat malleable over time (Gist
&Mitchell, 1992). The path from self-efficacy-1 to
seIf-efficacy-2 assessesthe degree of efficacy stability over time.
Therefore, the other pathsleading to self-efficacy-2 predict the
residual variance in efficacy, or thechanges that occur from one
time to another.
Initial performance. Initial performance is represented by
partici-pants' scores on the first of approximately 13 games they
bowled through-out the course. In effect, this represents a
baseline measure that playstwo critical roles in the current model.
First, participants' scores at thebeginning of training may act as
a surrogate measure of enactive masterythat reflects not only their
first experience in this class, but also experi-ences that they had
prior to the course. Thus, we anticipated that individ-uals who
initially experienced greater success in the program would
ev-idence greater self-efficacy development than would those who
initiallyperformed poorly. Initial performance in the program was
also includedin the equation predicting performance during the
latter portion of thecourse to control for initial individual
differences, and thereby, to isolatevariance in performance
improvement during training.
Achievement motivation. Achievement motivation, defined as a
de-sire "... to overcome obstacles, to exercise power, to strive to
do some-thing difficult as well and as quickly as possible"
(Murray, 1938, pp. 80-81), was also hypothesized to exhibit a
positive influence on the de-velopment of self-efficacy.
Achievement motivation is a relatively sta-ble individual attribute
that predisposes individuals to approach situa-tions in a
particular manner. High achievement motivated individualsgenerally
prefer challenging tasks and perceive a stronger
relationshipbetween their abilities and efforts and their
performance, than do lessachievement motivated individuals (see
Heckhausen, Schmalt & Schnei-der, 1985; Weiner, 1984). This
pattern of relations led Meyer (1987) tohypothesize that
differences in achievement motivation would lead todifferences in
task specific ability perceptions, and by extension In thepresent
context, to differences in self-efficacy. He reviewed previous
re-search and found support for this contention in 8 of 13 studies
(Meyer,1987). Therefore, we hypothesized that achievement
motivation wouldrelate positively to the development of
self-efficacy during training.
-
130 PERSONNEL PSYCHOLOGY
Choice. Hicks and Klimoski (1987) conducted a field experiment
inwhich they manipulated trainees' choices conceming whether to
attenda training program. Their results showed that participants
who weregiven a choice reported greater satisfaction with the
program, highermotivation to leam, more positive reactions, and
performed better onan achievement test, as compared to individuals
who were not givena choice of whether to attend the program.
Similar results have beenobtained by Ryman and Biersner (1975),
Mathieu et al. (1992), andBaldwin et al. (1991). In the present
setting, the university required allundergraduates to eam three
credits in P.E. to graduate, although theydid not need to enroll in
a bowling class. Consequently, many studentsenroll in P.E. courses
not because they wish to learn a particular skill, butsimply to
fulfill a requirement. In short, some students want to be in
thecourse, whereas others are simply "getting their tickets
punched."
We hypothesized that students who chose, or wanted, to be in
thecourse would report more positive training reactions.
Furthermore, weassumed that individuals who intentionally enrolled
in a bowling classwould do so, in part, because they believed that
they could perform well.Indeed, Bandura (1982,1989,1991) argued
that individuals' self-efficacywould influence their choices
concerning the situations in which they arewilling to participate.
Extending this reasoning to the present setting,we hypothesized
that individuals who wanted to participate in bowlingwould do so,
in part, because they felt that they could develop the neces-sary
skills and perform well in the class. Further, to the extent that
par-ticipants did not wish to be in the course, we expected that
they wouldlikely experience reactance, psychologically disengage
from the activity,and would be less likely to develop self-efficacy
than would those whowanted to be there. Bandura (1989, 1991) has
summarized previous re-search which has suggested that when
individuals do not care about theirperformance in a given
situation, they do not engage in self-regulatoryprocesses. Thus, we
anticipated a positive influence of choice on thedevelopment of
efficacy.
Situational constraints. The final two antecedents depicted in
Fig-ure 1 are individual and aggregate situational constraints.
Situationalconstraints can be defined as characteristics of the
environment that in-terfere with or restrict employees' performance
(Peters & O'Connor,1980; Peters, O'Connor & Eulberg, 1985).
Although typically conceivedof as aspects of the situation,
previous research has yet to differentiatethe level(s) of analysis
at which constraints operate. In the current study,we distinguish
constraints that exist at different levels of analysis.
Specif-ically, situational constraints can covary at individual or
aggregate levels.At the individual level of analysis, situational
constraints refer to otherobligations or pressures placed on
individuals that may differ from one
-
JOHN E. MATHIEU ET AL. 131
person to another. In the current setting, these may include the
num-ber of other classes different students are taking, their
extracurricularactivities, part-time jobs, and so forth. In work
settings, individual-levelconstraints could refer to any inhibitor
that differs from one person toanother (e.g., specific job tasks,
time pressures, nonwork or family diver-sions, lack of feedback,
etc.).
Aggregate situational constraints refer to environmental
featuresthat are common to a set or group of employees and hinder
or limitperformance. Examples of such constraints in the training
environmentwould include, but not be limited to, differences in
training sessions,classes, or group influences that interfere with
leaming processes. Inthe current setting, 15 different bowling
classes were sampled. The pri-mary factors that differed from one
class to another were the instruc-tors and their teaching styles,
but other features such as time and equip-ment availability also
differed somewhat. In more traditional trainingsettings, aggregate
constraints in the training environment could includethe trainer,
equipment, facilities, training methods, and so forth; namely,any
inhibitor that may differ from one training group to another.
We hypothesized that both individual and aggregate
constraintswould relate negatively both to the development of
self-efficacy andto training reactions. Phillips and Freedman
(1984) found support fora negative relationship between
individuals' perceptions of situationalconstraints and their work
motivation. Mathieu et al. (1992) founda negative relationship
between perceived situational constraints andan index of
training-related motivation. Further, Peters et al. (1985)reviewed
several studies that have obtained negative relationships be-tween
individuals' perceptions of constraints and their affective
reac-tions. Gist and Mitchell (1992) argued that when individuals
are focusedupon formidable aspects of a task they report lower
levels of self-efficacy.
What is not clear from the research summarized above, however,
isthe extent to which the detrimental effects of situational
constraints areattributable to individual or aggregate processes.
The effects emanat-ing from individual situationai constraints in
the current study repre-sent individual-level processes, whereas
the effects from aggregate sit-uational constraints represent
cross-level processes (see Mathieu, 1991,and Rousseau, 1985 for
more discussion of these designs). Because pre-vious theory and
research has yet to disentangle these two constmcts, wehypothesized
negative paths from both types of constraints to the devel-opment
of efficacy and reactions to the program.
-
132 PERSONNEL PSYCHOLOGY
Effects of Self-Efficacy on Training Outcomes
Gist (1987; Gist & Mitchell, 1992) argued that trainees'
self-efficacyrepresents a critical mediating variable in the
effectiveness of trainingprograms. For example, Frayne and Latham
(1987) found that trainees'self-efficacy levels mediated the
influence of a self-management trainingprogram on work attendance
levels during training. Given the demon-strated pervasive effects
of self-efficacy on performance in a variety ofsettings (cf.
Bandura, 1989, 1991; Gist, 1987), we hypothesized that ef-ficacy
assessed at the mid-point of the course would relate
significantlyto performance during the latter half of the program,
after controllingfor initial performance levels. Further, we
hypothesized that mid-courseself-efficacy would relate positively
with participants' training reactions.Bandura and Schunk (1981)
argued that when individuals experience asense of self-efficacy in
a situation, they are more likely to develop aninterest in the
activity than are those who fail to develop such efficacy.Thus, we
hypothesized that individuals who develop greater levels
ofself-efficacy by mid-way through the course would report greater
interestin bowling and react more positively to the course.
Besides the effects discussed above, reactions and self-efficacy
werehypothesized to interact as related to performance during the
latter por-tion of the course. Alliger and Janak (1989) conducted a
meta-analysisof studies that included two or more measures of
Kirkpatrick's (1976)4-fold typology of training outcomes: (1)
reactions, (2) learning, (3) be-havior change, and (4) results.
TTiey concluded that although significantpositive relationships are
generally found between learning, behavior,and results, training
reactions tend to be unrelated to other outcomes.Mathieu et al.
(1992) argued that Alliger and Janak's (1989) results donot
preclude other forms of relationships between training reactions
andother outcome measures. Mathieu et al. suggested and found
support fora moderated relationship between training motivation and
training re-actions, as related to a measure of learning due to
training. The form ofthe interaction was such that training
motivation was more strongly re-lated (i.e., exhibited a steeper
positive slope) to learning if participantsreacted positively to
the program. Training motivation was still relatedpositively to
learning among individuals who reacted negatively to theprogram,
but to a lesser degree.
Applied to the present context, the moderated relationship
obtainedby Mathieu et al. (1992) would suggest that trainees'
mid-course self-efficacy might interact with their training
reactions, so that efficacy wouldexhibit a more positive
relationship with subsequent performance amongparticipants who
reacted positively to the program. Finally, although a
-
JOHN E. MATHIEU ETAL. 133
path from training reactions to performance is included in the
hypothe-sized model, based on Alliger and Janak's (1989)
meta-analysis findings,we did not expect it to be significant. The
path is included in the modelsimply to permit the test of the
hypothesized moderated relationships.
Method
Participants/Program
The training program was an 8-week long introductory bowling
courseat Pennsylvania State University in which students were
instmcted ontechniques and were able to practice almost every class
during the semes-ter. Five hundred six students were enrolled in 15
classes. Classes metfor either 50 minutes, three times a week, or
75 minutes, twice a week.
Obviously a University bowling class is markedly different from
most"real" training settings. Nevertheless, this setting provides a
valuableopportunity for examining the processes in question. First,
because stu-dents had to take at least three credits of P.E.
classes in order to graduate(but not necessarily a bowling class),
we expected a wide range of atti-tudes concerning participation.
Second, because the classes were taughtby several different
instmctors, we anticipated that some between-classeffects could
emerge. Third, a significant limitation of conducting re-search in
most corporate training environments is the lack of
meaningful,objective, quantifiable criterion measures (Saari,
Johnson, McLaugh-lin & Zimmerle, 1988). The current setting
does not suffer this short-coming. Bowling scores are directly
quantifiable and objective, and canbe compared from one time to
another and from one course to an-other. When gathered
longitudinally, they provide a fairly uncontami-nated measure of
skill acquisition. TTius, although the extemal validity,or
generalizability to "real world" settings is naturally a concem
here,the current setting offers many protections against threats to
intemaland statistical conclusion validities and offers a "strong
test" of the hy-pothesized relationships (Cook & Campbell,
1979).
Surveys were distributed during the first and fourth weeks of
classes,and contained the measures described below. Students were
asked tocomplete the surveys outside of class time and to retum
them within aweek. Although 280 students completed the first
survey, only 215 alsoreturned usable second surveys. This sample
was 58% male and had anaverage age of 20.33 years. Statistical
contrasts were performed betweenthose who did and did not retum the
second survey, using demographicindices and other measures
available from the first survey (see below).No significant (p
>.O5) differences were obtained. Furthermore, therewere no
significant differences between the performance levels of class
-
134 PERSONNEL PSYCHOLOGY
members who participated in the study and those who did not.
Thus, dif-ferential mortality is not a threat to intemal validity
(Cook & Campbell,1979).
Survey Measures
The first survey, administered on the first day of class,
assessed par-ticipants' demographics, initial self-efficacy,
achievement motivation,choice, and other measures not pertinent
here (e.g., course expecta-tions). The second survey included
measures of individual and aggregatesituational constraints,
self-efficacy, and reactions to training, and wasadministered
during the middle week of the course. Except for the de-mographic
items, all responses were made on 7-point Likert-type scales,with
higher values representing greater amounts of each variable.
Scalescores were computed for each variable by averaging the items
to whicha participant responded.
To assess self-efficacy, students were asked to rate the extent
to whichthey believed that they could score at least each of eight
bowling scores(i.e., 141,131,121, etc.) that corresponded to the
grade levels they couldeam (i.e.. A, A-, B+, etc.). They responded
to each level using a 7-pointscale that ranged from (1) "Almost no
possibility" to (4) "About a 50/50chance" to (7) "Complete
certainty." This scale exhibited high intemalconsistencies both at
time 1 (a = .85) and at time 2 (a = .87).
Further,participants'self-efficacy improved significantly (i(214)=
8.17,p < .001)between the first and second surveys (M = 4.23, SD
= 1.22 and M = 4.86,SD = 1.15, respectively).
Achievement motivation was assessed using a 10-item scale (a =
.74)adapted from Mehrabian and Banks (1978). Six items on this
scale areworded negatively, and were reverse coded prior to
analyses. This scalehas demonstrated acceptable reliability and
predictive validity in previ-ous research (e.g., Mathieu, 1988).
The scale measures the extent towhich participants (a) prefer
challenging situations, (b) are comfortablewith making decisions or
being in high-pressure situations, and (c) wouldwork hard rather
than take it easy.
The university requires students to participate in three credits
of P.E.to graduate. Two items were used to assess participants'
choice of, orpreference for, their enrollment: "If the University
did not require me totake P.E. courses, I would not take any during
my undergraduate career"(reverse coded) and "I would take as many
P.E. courses as I could fit intomy schedule during a given
semester." These two items exhibited a =.62.
Individual and a^egate situational constraints were each
assessed us-ing items included in the second survey. This timing
was adopted so that
-
JOHN H. MATHIEU ETAL. 135
participants would have sufficient time and experience to
evaluate theextent to which their class provided certain
opportunities, and the otherdemands placed upon them during the
semester. The 3-item individualconstraints scale assessed the
extent to which extracurricular activitiesand other classes'
workloads interfered with the amount of time partici-pants had to
practice their bowling skills (a = .63).
The aggregate situational constraints, or those experienced by
all par-ticipants in a given class, were assessed using a 3-item
scale that mea-sured the extent to which adequate equipment, time,
and encourage-ment were given to students. It is important to
emphasize that althoughsurvey responses were collected from
individual students, these itemsreferred to an aggregate class
attribute. Thus, as discussed by Sirotnik(1980), the psychometric
properties of this scale should be examined atthe aggregate level
of analysis using a between groups matrix that con-tains item
averages computed within groups. In order to justify aggre-gating
students' responses within classes, however, it is first necessary
todemonstrate that individuals within each class exhibit reasonably
highlevels of agreement. James, Demaree and Wolf (1984) advanced an
in-terrater reliability index (IRR) for such purposes. James et
al.'s multi-item IRR formula contrasts the average observed item
variance acrossresfjondents within a group, against that which
might be expected from arandom response pattern. Low IRR values
indicate a lack of agreementbetween individuals in a group, whereas
high IRRs suggest that individ-uals within a group agree on some
target variable (Kozlowski & Hattrup,1992). The 15 class IRRs
ranged from .65 to .89 with a median value of.82. Thus, students
within each class evidenced high agreement and theirresponses/wr
item were averaged within each class. Finally, to calculatea
reliability for the aggregate situational constraints, Cronbach's
alphawas computed based on the item averages per class and was .67.
Meanclass scores were then assigned to all students in the class
(see Mathieu,1991 and Rousseau, 1985 for more discussion on this
procedure).
Outcome Measures
Training reactions. Participants' affective reactions to the
course weremeasured using a 4-item scale included in the second
survey that assessedthe extent to which students enjoyed the class
(a = .88). Two exampleitems are "I have had a good time in bowling
so far this semester" and"I am happy that I am taking bowling this
semester."
Performance. Students' course grades were determined, in large
part,by how well they performed on a series of games bowled during
the 8-week course. The exact number of games bowled per student
varied
-
136 PERSONNEL PSYCHOLOGY
somewhat across the 15 classes because of class schedules and
instruc-tion styles. On average, students bowled 13.5 games during
the course.So as to maintain the temporal order of processes
depicted in Figure 1,each participant's performance score was
calculated as the average of thegames he or she bowled after
retuming the second survey. Applying theSpearman-Brown prophecy
formula using the average number of gamesplayed after survey 2 was
retumed (i.e., 8.8), and the average weightedgame-to-game
correlation (i.e., r = .34, p < .05), yields a stability
coef-ficient of .82 for the performance measure. Finally, to
control for initialindividual performance differences, we used
participants' scores on theflrst games they bowled (which occurred
after survey 1 and before survey2 was returned) as an initial
performance, or baseline measure. Individ-uals improved
significantly (t(214) = 5.90, p
-
JOHN E. MATHIEU ET AL. 137
and therefore should be interpreted cautiously. It is useful,
however, fortesting the relative fit of nested models. We employed
a p value of .05for determining the significance of competing
models, and the individualmodel paths. GFI represents an index of
the relative amount of the co-variances among the latent variables
that are collectively accounted forby the hypothesized model.
Generally, GFIs above .90 are consideredindicative of a good fit.
RMSR is a measure of the average of the fittedresiduals, and when
working in covariance metric, must be interpretedrelative to the
magnitude of variables' variances and covariances.
The T-L index compares the relative fit and degrees of
freedomfor a given structural model against a baseline model. One
T-L indexwas calculated comparing the hypothesized model against a
model thatspecified zero correlations among the latent variables
(i.e., a null latentmodel). Values > .90 for this version T-L
index are generally consideredas indicative of a good fit. Sobel
and Bohrnstedt (1985) have argued thata null latent baseline model
is too restrictive and recommended the useof "informed baseline
models" for comparative purposes. Thus, we useda "null structural
model," which hypothesized that none of the paths de-picted in
Figure 1 would be significant, as a second baseline model.
Results
Descriptive statistics and observed correlations among all study
vari-ables, and correlations among the latent variables, are
presented in Tk-ble 1. However, the sample covariance matrix was
used for all modeltests. The results of the hypothesized model,
with and without the mod-erator term, are presented in Figure 1 and
a summary of the model fitindices is presented in T^ble 2. The
hypothesized model represented asignificant improvement over the
null structural model that predicts nosignificant paths
[x^-difference (14) = 228.56, p < .01], and exhibitedhigh fit
indices [(x^(22) = 52.40 p < .01; GFI =.956; RMSR = .062;T-L vs.
Null Latent = .988, T-L vs. Null Structural = .797]. However,some
hypothesized relationships were not statistically significant.
Specif-ically, aggregate situational constraints did not have a
significant impacton self-efficacy-2, and the influence of
individual situationai constraintswas only marginal {0 = - .141, p
< .10). Further, choice did not have asignificant effect on
reactions to training. Finally, as hypothesized, thelinear
relationship between training reactions and perfonnance was
notsignificant.
In order to assess the hypothesized moderated relationship
betweenself-efiicacy-2 and training reactions, as related to
performance, we per-formed a x^ difference test between models
including and excluding the
-
138 PERSONNEL PSYCHOLOGY
•Si
I Ip p ^ ri
r r I I
0 > - H . - H 0 ^ O " ^ ^ ^
• r • • f "^ •
I I I ! I
o
l' I I
CTNOO>noo o c o i nf i r s i n O ' - ' --I f ^ < nO " — '
r — M r - i r ^ i n i ^
I I ^
nr •»•
5o I p —
S o' - OrI I r r00 Q rH
r
r - o o f j
•I 5
-
JOHN E. MATHIEU ET AL. 139
TABLE 2Summary of liirious Structural Model Fit Indices
Model fit indicesT-L vs.
null nullModels df X GFI RMSR latent structural
1. Hypothesizedwith interaction" 22 52.40* .956 .062 .988
.797
2. Hypothesizedwithout interaction" 23 52.79* .956 .063 .989
.810
3. Null structural 36 280.98* .797 .175 NA NA4. Null latent 55
6526.23* .196 .196 NA NA
Notes: dS= degrees of freedom; T-L = TUcker-Lewis incremental
fit index; GF1=goodness of fit index; RMSR= root mean square
residual.
W-215; •p
-
140 PERSONNEL PSYCHOLOGY
Antecedents of Self-Efficacy Development
Individual variables. Several individual-level variables
influenced thedevelopment of self-efficacy, including initial
performance, achievementmotivation, and trainee choice. Trainees'
initial self-efficacy levels alsoexhibited a strong positive
relationship with mid-course self-efficacy.That is, as in previous
research, efficacy levels were somewhat consis-tent over time (Gist
et al., 1991; Latham & Frayne, 1989; Tannenbaumet al., 1991).
Alone, initial self-efficacy accounted for 39.06% of the vari-ance
of mid-course efficacy. However, after controlling for time 1
self-efficacy, the other hypothesized antecedents accounted for a
significantadditional 12.47% of time 2 self-efficacy variance.
Achievement motivation was positively related to the
developmentof self-efficacy. Tt"ainees who entered the course with
a predispositionfor challenging situations and hard work were more
likely to exhibitincreased self-efficacy during training. However,
achievement motiva-tion only appears to influence training
effectiveness as mediated by self-efficacy, as an examination of
the bivariate correlations revealed no di-rect relationships
between it and either performance or training reac-tions. Thus, the
relatively stable personality variable shaped trainees'cognitions
about more specific instances (i.e., self-efficacy), which in
turnhad an influence on training effectiveness. These findings are
consistentwith Bandura's (1989) conception of the linkage between
personal needsand self-efficacy development.
As hypothesized, trainees who chose to take this course were
morelikely to develop increased self-efficacy during training.
Perhaps traineeswho wanted to take the course entered more
motivated to learn (e.g.,Baldwin et al., 1991), exerted more
effort, and thus, correctly believedthat they would learn to bowl
better than before. However, traineechoice was unrelated to
training reactions. It is somewhat surprising thattrainees who
wanted to attend did not report greater satisfaction thantrainees
who had not wanted to be there. We can only speculate whythis
occurred. Perhaps those trainees who wanted to attend this
courseentered it with higher expectations than those who merely
wanted their"ticket punched." There is some evidence that trainee
desires and expec-tations are correlated and that a failure to meet
trainee expectations canlead to reduced satisfaction (Tannenbaum et
al., 1991). Perhaps sometrainees who wanted to take this course had
false expectations, washingout any positive effect of choice on
reactions.
Situational variables. We hypothesized that situational
constraintswould relate negatively to the development of
self-efficacy during train-ing. Previous research has shown the
potentially debilitating effects of
-
JOHN E. MATHIEU ET AL. 141
situational constraints on employee performance (e.g., Peters
& O'Con-nor, 1980). However, this study was the first to test
whether constraintsoperate at different levels as related to
training effectiveness. We exam-ined individual- or trainee-level
constraints and aggregate- or course-level constraints. In
addition, the correlation between the two types ofconstraints was
non-significant (r = .085, n.s.) suggesting that they weretapping
different phenomena. Furthermore, the two levels of
constraintsexhibited different relationships with other
variables.
Individual-level constraints exhibited a marginally significant
(p <.10) negative impact on seIf-efficacy-2 in the hypothesized
model. Thissame relationship became significant (p < .05),
however, when the non-significant path from aggregate situational
constraints to self-efficacy-2was trimmed from the model. Trainees
who felt they had more individ-ual constraints (e.g., competing
demands for their time) were less likelyto develop a belief that
they could master the skills being trained. Thisis an important
finding as it shows that events outside of training
(i.e.,non-training demands) can have a debilitating influence on
training ef-fectiveness. It suggests that managers must give
careful attention to theobligations and pressures that their
employees need to balance while at-tending training. Training does
not occur in isolation from other job andpersonal obligations, and
merely providing release time to attend train-ing may not be
sufficient to maximize training effectiveness.
Trainees who reported the presence of constraints, whether
individ-ual or course related, responded less favorably to the
training. That is,whatever the source of the constraints, trainees
translated any limita-tions in their environment into negative
reactions to the training. Thus,it appears that constraints can
have a negative influence on trainees' re-actions as well as on
their self-efficacy. Many organizations rely on reac-tion forms to
evaluate their training programs, yet most reaction formsdo not ask
trainees anything about non-training issues (e.g., time
con-straints). Therefore, organizations that use standard reaction
measuresto revise their training may make fruitless "improvements"
to a train-ing program if the actual source of discontent lies
outside the trainingcontext.
In contrast to individual constraints, training environment
constraintswere not related to the development of self-efficacy.
Because the aggre-gate-level measure was significantly related to
training reactions, it is un-likely that the non-significant
finding for self-efficacy can be attributablestrictly to
statistical power or to measurement issues. However, it is
pos-sible that in this training context the constraints were not
severe enoughto interfere with actual performance and thus, did not
interfere with thedevelopment of self-efficacy.
-
142 PERSONNEL PSYCHOLOGY
Overall, these findings highlight the importance of identifying
andminimizing constraints within the entire training system—not
just in thetraining program. In addition, they suggest the need to
treat and di-agnose different types of constraints separately. In
the current context,individual-level and training-level constraints
both had a negative influ-ence on trainee reactions, but more
importantly, individual constraintshad a debilitating effect on
trainee self-efficacy. Researchers may wantto examine how
altemative methods of "freeing" employees to attendtraining can
influence trainee self-efficacy, motivation to learn, and
sub-sequent training outcomes. For example, in one study that
reportedhigh training effectiveness, managers performed their
subordinates' jobswhile the employees were attending training (Lee,
1991). While that ap-proach may not always be feasible, researchers
should begin to examinedifferent ways to balance training needs
with other job obligations andstresses.
We should also note that the aggregate-level constraints
examinedin this study dealt specifically with the training program.
Yet, aggregatesituational constraints are also likely to be present
in the work environ-ment and could include different work
technologies; rules, procedures,or policies; supervisor behaviors;
or any other features that are commonto more than one employee.
Work environment constraints are likelyto be most detrimental to
transfer processes. For example, Rouiller andGoldstein (1991) found
that aggregate aspects of the work environment(called climate for
transfer) related significantly to subsequent trans-fer behavior
even after controlling for individual difTerences in
learning.Supportive climates enhanced transfer processes, whereas
nonsupport-ive climates limited transfer. Thus, researchers and
practitioners shouldtry to identify and to minimize aspects of
individuals' jobs, the training,and the work environment, that
constrain the learning process and ul-timately limit performance
improvements. In a similar vein, attentionshould be directed at
identifying features of the work and training envi-ronments that
facilitate such processes (i.e., act as enhancers).
Self-Efficacy —* Performance Improvement
Linear effect. As hypothesized, self-efficacy assessed midway
throughthe course contributed to subsequent performance improvement
andwas positively related to training reactions. This is consistent
with pre-vious research and further confirms the central role that
self-efficacyplays in understanding and enhancing training
effectiveness. Initial self-efficacy was also related to subsequent
self-efficacy, and initial perfor-mance levels were associated with
subsequent performance levels. Other
-
JOHN E. MATHIEU ET AL. 143
studies have shown how successful performance enhances the
subse-quent development of self-efficacy (cf. Bandura, 1982, 1991).
This sug-gests that there may be a form of positive, reinforcing
feedback cycle thatoccurs between self-efficacy and perfomiance
(Gist & Mitchell, 1992).Perhaps initial self-efficacy enhances
skill acquisition and performancewhich in turn fosters subsequent
self-efficacy. It is also conceivable thatthis phenomenon may carry
over to related courses (cf. Gist, Bavetta,& Stevens, 1990),
and over time could contribute to the establishmentof a "continuous
leaming environment" within an organization. Indeed,this process
may account for the pervasive effects of enactive mastery
ex-periences on efficacy levels that has been observed in previous
research.Future research should examine the longitudinal efFects of
self-efficacyand performance over a sequence of training
experiences.
Researchers and practitioners should also consider the use of
main-tenance interventions, such as self-management or goal-setting
applica-tions designed to maintain efficacy levels (cf. Gist et
al., 1990; Gist etal., 1991). For example. Gist et al. (1991) found
that post-training self-management interventions were effective
maintenance strategies for in-dividuals who left training with low
or moderate efficacy levels. In con-trast, goal-setting
interventions proved more successful for individualswho left
training with high efficacy levels. Clearly the interplay
betweentrainees' efficacy levels, both during and after training,
and factors out-side of the training environment need to be
understood better to maxi-mize the effectiveness of training
programs.
Training reactions as moderators. We found no support for a
moder-ated relationship between reactions and mid-course
self-efficacy, as re-lated to performance improvement. This is in
contrast with the resultsfrom the Mathieu et al. (1992) study that
reported a moderated effect be-tween motivation and reactions.
However, there are several differencesbetween the two tests of
moderation that may explain the seemingly con-flicting results.
First, the Mathieu et al. study used a VIE measure oftrainee
motivation rather than a self-efficacy construct. Although
self-efficacy and motivation are similar constructs they are not
synonymous(see Gist, 1987; Gist & Mitchell, 1992). Efficacy is
similar to the effort -»perfomiance expectancy component of VIE
theory, although the formeris a more encompassing concept that
includes individuals' considerationsof task and situational
attributes, and their ability to mobilize resourcesto accomplish
the task. The instrumentality and valence componentsof VIE theory
deal with linkages between performance and differentlyvalued
outcomes. Thus, it may be that the moderating influence of
train-ing reactions found in Mathieu et al. may be more
attributable to theircombination with instrumentalities and
valences than with effort —* per-fomiance expectancies. This
represents a question for future research.
-
144 PERSONNEL PSYCHOLOGY
A second difference between the Mathieu et al. (1992) study and
thisone is that the former used a leaming measure as the criterion
for the testof interaction, whereas the current study used a
performance measure.As noted earlier, a traditional leaming measure
was not as applicable inthe current setting that focused on
physical skill acquisition. Finally, thenature of the two samples
differed markedly, which could account forthe different findings.
In summary, although we failed to find a non-linear relationship
among training criteria in the current context, webelieve that
future researchers should continue to examine
moderatedrelationships among training outcomes. It would be most
informative tocollect both efficacy and VIE measures from a sample
of trainees so asto test some of these altemative explanations
directly.
Study Limitations
Although significant effects stemmed from individuals' choice,
indi-vidual-level constraints, and aggregate constraints, these
variables weremeasured with only a few items each and exhibited
relatively low relia-bilities. We recommend that future researchers
develop more compre-hensive scales for measuring these variables.
Second, we believe that acomment on sampling decisions, as related
to various validity inferences,is in order (Cook & Campbell,
1979). Traditionally, laboratory studiesare thought to yield
greater intemal validity whereas field studies pro-vide more
extemal validity. We chose our sample as a "middle ground"in order
to balance the two competing demands. We wanted a researchsite that
would enable us to control or to test for various threats to
inter-nal validity (e.g., instrumentation, differential mortality),
yet still sam-ple from an actual leaming environment with real
consequences (albeit,simply a course grade). We also desired a
setting that would minimizethreats to statistical conclusion
validity (e.g., minimize range restriction,provide sufficient
power). As part of this decision, we recognize thatsome compromises
are inevitable. In particular, we acknowledge that auniversity
bowling class is inherently different from a company trainingcourse
on computer programming, engine repair, interpersonal skills,and so
forth. Nevertheless, effective training depends, in part, on
skillacquisition processes, whatever those skills may be.
In the current context we were able to study processes
associated withthe acquisition of a physical skill (i.e., average
improvement in bowlingscore was approximately 10%) in an
environment that allowed for thecollection of objective perfonnance
indicators and enhanced internaland statistical conclusion
validity. Task attributes such as the nature, dif-ficulty, and
complexity of activities, are likely to influence the magnitude
-
JOHN E. MATHIEU ET AL. 145
of effects that are observed (Gist & Mitchell, 1992) and the
generalizabil-ity of findings. Therefore, these findings are
probably most generalizableto jobs that require repetitive use of
physical skills such as machine oper-ators, cashiers, pressors, or
assembly line work. The nature of the train-ing program itself may
also operate as an important boundary condition.For example, the
bowling classes were conducted in a group setting andentailed
primarily verbal and written instruction with some demonstra-tions.
Other training methods, such as the use of video-taped
feedback,extensive modeling efforts, and individually designed and
paced pro-grams might influence the nature of the processes
underlying effectiveprograms. As part of a programmatic effort to
better understand theinfluences on training effectiveness, we
encourage the extension of thisresearch into more traditional
organizational settings where issues suchas the role of work
environment constraints on trainees' efficacy and mo-tivation, and
also transfer processes may be more thoroughly examined.
REFERENCES
Alliger GM. Janak EA. (1989). Kirkpatrick's levels of training
criteria: TTiirty years later.PERSONNELPSYCHOLOGY, 42, 331-341.
Anderson JC, Gerbing DW. (1988). Structural equation modeling in
practice: A reviewand recommended two-step approach. Psychological
Bulletin, 103, 411-423.
Baldwin TT. Magjuka RJ, Loher BT. (1991). The perils of
participation: Effects of choiceof training on trainee motivation
and leaming. PERSONNEL PSYCHOLOGY, 44,51-66.
Bandura A. (1982). Self-efficacy mechanism in human agency.
American Psychologist, 37,122-147.
Bandura A. (1986). Social foundations of thought and action.
Englewood Cliffs, NJ:Prentice-Hall.
Bandura A. (1989). Self-regulation of motivation and action
through intemal standardsand goal systems, in Pervin LA (Ed.), Goal
concepts in personality and social psy-chology (pp. 19-85).
Hillsdalc, NJ: Erlbaum.
Bandura A. (1991). Social cognitive theory of self-regulation.
Organizational Behavior andHuman Decision Processes, 50,
248-287.
Bandura A, Schunk DH. (1981). Cultivating competence,
self-efficacy, and intrinsic inter-est through proximal
self-motivation. Joumai of Personality and Social Psychology,41,
586-598.
Bandura A, Wood R. (1989). Effect of perceived controllability
and performance stan-dards on self-regulation of complex decision
making. Joumai of Personality andSocial Psychology, 56,
805-814.
Bohmstedt G. Marwell G. (1978). The reliability of products of
two random variables.In Schuessler KF (Ed.), Sociological
methodology (pp. 254-273). San Francisco:Jossey-Bass.
Busemeyer J, Jones L. (1983). Analysis of multiplicative
combination rules when the causalvariables are measured with error.
Psychological Bulletin, 93, 549-562.
Campbell JP. (1988). TVaining design for performance
improvement. In Campbell JP,Campbell RJ and Associates (Eds.),
Productivity in organizations (pp. 177-216). SanFrancisco:
Jossey-Bass.
-
146 PERSONNEL PSYCHOLOGY
Clark CS. (1990). Social processes in work groups: A model of
the effect of involvement,credibility, and goal linkage on training
success. Unpublished doctoral dissertation.The University of
Tfetinessee, Knoxville.
Cook TD, Campbell DT. (1979). Quasi-experimentation: Design and
analysis issues for fieldsettings. Boston: Houghton MifHin.
Eden D. Ravid G. (1982). Pygmalion versus self-expectancy:
Effects of instructor- andself-expectancy oti trainee performance.
Organizational Behavior and Human Per-formance, 30, 35\-364.
Ford JK. Quiiiones MA, Sego D, Sorra JP. (1992). Factors
afTecting the opportunity toperform traitied tasks on the job.
PERSONNEL PSYCHOLOGY, 45,511-528.
Frayne CA, Latham GP. (1987). The application of social leaming
theory to employeeself-management of attendance. Joumal of Applied
Psychology, 72, 387-392.
Gist ME. (1987). Self-efRcacy: Implications for organizational
behavior and human re-source management. Academy of Management
Review, 12, 472-485.
Gist ME, Bavetta AG, Stevens CK. (1990). IVansfer training
method: Its influence on skillgeneralization, skill repetition, and
performance level, PERSONNEL PSYCHOLOGY, 43,501-523.
Gist ME, Mitchell TR. (1992). Self-efficacy: A theoretical
analysis of its determinants andmalleability. Academy of Management
Review, 17,183-211.
Gist ME, Schwoerer C, Rosen B. (1989). Effects of altemative
training methods on self-efficacy and peifomiance in computer
software training. Journal of Applied Psychol-ogy, 74, 884-891.
Gist ME, Stevens CK, Bavetta AG. (1991). Effects of
self-efficacy and post-training inter-vention on the acquisition
and maintenance of complex interpersonal skills, PER-SONNEL
PSYCHOLOGY, 44, 837-861.
Heckhausen H, Schmalt H-D, Schneider L. (1985). Achievement
motivation in perspective.Orlando, FL: Academic Press.
Hicks WD, Klimoski RJ. (1987). Entry into training programs and
its effects on trainingoutcomes: A field experiment. Academy of
Management Joumal, 30, 542-552.
James LR, Demaree RG, Wolf G. (1984). Estimating within-group
interrater reliabilitywith and without response bias. Joumal of
Applied Psychology, 69, 85-98.
Joreskog KG, Sorbom D. (1989). LISREL VII User's Reference
Guide. Mooresville. IN:Scientific Software.
KJrkpatrick DL. (1976). Evaluation of training. In Craig RL
(Ed.), Training and develop-ment handbook (2nd ed., pp. 18-27). New
York: McGraw-Hill.
Kozlowski SWJ, Hattnip K. (1992). A disagreement about
within-group agreement: Dis-entangling issues of consistency versus
consensus. Joumal of Applied Psychology, 77,161-167.
Latham GP, Frayne CA. (1989). Self-management training for
increasing job attendance:A follow-up and a replication. Joumal of
Applied Psychology, 74, 411-416.
Lee CW. (1991). Effects of supervisor behavior on work group
effectiveness: A field experiment.Unpublished doctoral
dissertation, The State University of New York at Albany.
Mathieu JE. (1988). A causal model of organizational commitment
in a military trainingenvironment. Joumal of Vocational Behavior,
32, 321-335.
Mathieu JE. (1991). A cross-level nonrecursive model of the
antecedents of organizationalcommitment and satisfaction. Joumal of
Applied Psychology, 76, 607-618.
Mathieu JE, Tknnenbaum SI, Salas E, (1992). Influences of
individual and situationa! char-acteristics on measures of training
effectiveness. Academy of Management Joumal,35, 828-847.
Mehrabian A. Banks L. (1978). A questionnaire measure of
individual differences inachieving tendency. Educational and
Psycholo^cal Measurement, 38,475-478.
-
JOHN E. MATHIEU ET AL. 147
Meyer W-U. (1987). Perceived ability and achievement related
behavior. In HalischF, Kuhl J (Eds.), Motivation intention and
volition (pp. 73-86). Berlin: Springer-Verlag.
Murray HJ. (1938). Explorations in personality. New York: Oxford
University Press.Peters LH, O'Connor EJ. (1980). Situational
constraints and work outcomes: The influ-
ences of a frequently overlooked constnict. Academy of
Management Review, 5,391-398.
Peters LH, O'Connor El, Eulberg JR. (1985). Situational
constraints: Sources, conse-quences, and future considerations.
Research in Personnel and Human ResourcesManagement, 3, 79-114.
Phillips JS, Freedman SM. (1984). Situational performance
amstraints and task charac-teristics: Their relationship to
motivation and satisfaction. Joumal of Management,/0,321--331.
Rails RS, Klein KJ. (1991, April). Thainee cogniHve ability and
motivation: Effects oncomputer training performance. A paper
presented at the Sixth Annual Conferenceof the Society of
Industrial and Organizational Psychology, St.Louis.
Rouiller JZ, Goldstein IL. (1991, April). Determinants of the
climate for transfer of tramirig.A paper presented at the Sixth
Annual Conference of the Society of Industrial andOrganizational
Psychology, St. Louis.
Rousseau DM. (1985). Issues of level in organizational research:
Multi-level and crc^s-level perspectives. Research in
Organizational Behavior, 77,1-37.
Ryman DH, Biersner RJ. (1975). Attitudes predictive of diving
success, PERSONNEL PSY-CHOLOGY, 2S , 181-188.
Saari LM, Johnson TR, McLaughlin SD, Zimmerle DM. (1988). A
sutvey of managementtraining and education practices in U.S.
companies. PERSONNEL PSYCHOLOGY, 41,731-743.
Sirotnik KA. (1980). Psychometric implications of the unit of
analysis problem (withexamples from the measurement of
organizational climate). Joumal of EducationalMeasurement, 17,
245-282.
Sobel M, Bohmstedt GW. (1985). Use of null models in evaluating
the fit of covariancestructure models. In TUma NB (Ed.).
Sociological methodology (pp. 290-312). SanFrancisco:
Jossey-Bass,
l^nnenbaum S, Mathieu J, Salas E, Cannon-Bowers J. (1991).
Meeting trainees' expec-tations: The influence of training
fulfillment on the development of commitment,self-efficacy, and
motivation. Joumal of Applied Psychology, 76, 759-769.
Iknnenbaum S, YukI G. (1992). Training and development in work
organizations. AnnualReview of Psychology, 43, 399-441.
Tbcker L, Lewis C. (1973). A reliability coefficient for maximum
likelihood factor analysis.Psychometrika,38, 1-10.
Weiner B. (1984). Achievement motivation and attribution theory.
Morristown, NJ: GeneralLeaming Press.