Social Cognitive Theory and Individual Reactions To Computing Technology: A Longitudinal Study
Social Cognitive Theory and Individual Reactions To Computing Technology: A Longitudinal Study
Social Cognitive Theory and Individual Reactions To Computing Technology: A Longitudinal Study
Qarterjy
RE s i A i u n
Nort:
Deborah Compeau
Faculty of Management
University of Calgary
Calgary, Alberta T2N 1N4
CANADA
compeau@ucalgary.ca
Christopher A. Higgins
Ivey School of Business
The University of Western Ontario
London, Ontario N6A 3K7
CANADA
chiggins@ivey.uwo.ca
Sid Huff
tvey School of Business
The University of Western Ontario
London, Ontario N6A 3K7
CANADA
shuff@ivey.uwo.ca
Abstract
A model, based on Bandura's Social Cognitive
Theory, was developed to test the influence of
computer self-efficacy, outcome expectations,
affect, and anxiety on computer usage. The
model was tested using longitudinal data gathered from 394 end users over a one-year interval.
Significant relationships were found betv^^een
computer self-efficacy and outcome expectations, and between self-efficacy and affect and
anxiety and use. Performance outcomes were
found to influence affect and use, while affect
was significantly related to use. Overall, the findings provide strong confirmation that both selfefficacy and outcome expectations impact on an
individual's affective and behavioral reactions to
information technology.
Keywords: IS usage, self-efficacy, causal models,
longitudinal
ISRL Categories: AP, GB02, GB03
Introduction
The study of individual reactions to computing
technology has been an important topic in recent
information systems research. Many authors have
studied different aspects of the phenomenon,
from a variety of theoretical perspectives, including Diffusion of Innovations (DOI) (e.g.,
Compeau and Meister 1997; Moore and
Benbasat 1991), the Technology Acceptance
Model (TAM) (e.g., Davis et al. 1989; Venkatesh
and Davis 1996), the Theory of Planned Behavior
(TPB) (e.g., Mathieson 1991; Taylor and Todd,
1995), and Social Cognitive Theory (SCT) (e.g.,
Compeau and Higgins 1995a, 1995b; Hill et al.
1986, 1987). This research has produced useful
insights into the cognitive, affective, and behavioral reactions of individuals to technology, and
into the factors which influence these reactions.
146
makes drawing causal conclusions more difficult. In any research, without longitudinal separation of hypothesized causes from effects, it is
difficult to draw conclusions about the causal
implications of the relationships observed
(Vitalari 1991). Given the reciprocal relationships posed by Social Cognitive Theory, this
problem is magnified.
Thus, the current study tests a model of individual reactions to computing technology in a longitudinal context. This allows us to make stronger
causal arguments regarding the observed relationships, even given the complex theoretical
context from which the model is derived.
Moreover, studying the effects of self-efficacy
and outcome expectations over time allows us to
understand whether their influences are relatively short in duration or whether they are more
enduring. This evidence will help in building
programs (training, support, etc.) and managing
implementation based on these factors.
Computer
Self-Efficacy
Outcome
Expectations
(Performance)
Outcome
Expectations
relate to expectations of change in image or status or to expectations of rewards, such as promotions, raises, or praise. Affect and anxiety represent the affective responses of individuals
toward using computers. Affect represents the
positive sidethe enjoyment a person derives
from using computerswhile anxiety represents
the negative sidethe feelings of apprehension
or anxiety that one experiences when using computers. Use represents the degree of use of computers at work and at home.
The hypotheses tested are those originally proposed by Compeau and Higgins (1995b). These
are outlined in Table 1. According to the model,
self-efficacy influences both personal and performance-related outcome expectations (HI and
H2), since it is often difficult for individuals to
separate the anticipated consequences of the
behavior from their expectations of performance
attainments (Bandura 1986). That is, if I believe I
will be able to use a computer with great skill, I
am more likely to expect positive outcomes from
H2.
H3.
H4.
H5.
H6.
H7.
H8.
H9.
H10.
H11.
148
The higher the individual's computer selfefficacy, the higher his/her performancerelated outcome expectations.
The higher the individual's computer selfefficacy, the higher his/her personal
outcome expectations.
The higher the individual's computer selfefficacy, the higher his/her affect
(or liking) of computer use.
The higher the individual's computer selfefficacy, the lower his/her computer
anxiety.
The higher the individual's computer selfefficacy, the higher his/her use cf
computers.
The higher the individual's performancerelated outcome expectations, the higher
his/her affect (or liking) for the behavior.
The higher the individual's personal
cutccme expectations, the higher his/her
affect (or liking) for the behavior.
The higher the individual's performancerelated outcome expectations, the higher
his/her use of computers.
The higher the individual's personal
outcome expectations, the higher
his/her use of computers.
The higher the individual's affect fcr
computer use, the higher his/her use
of computers.
The higher the individual's computer
anxiety, the lower his/her use of
computers.
Supporting References
Bandura et al.
Compeau and
1987
Bandura et al.
Ccmpeau and
1987
Bandura et al.
Compeau and
1987
Bandura 1986
Methodology
Procedures
Pretest and pilot studies of the survey instrument
were conducted prior to the initial data collection phase and are reported elsewhere (Compeau
and Higgins 1995b). The survey design and data
collection procedures were those recommended
by Dillman {1978).
Data were collected at two points in time. The
first survey was senl to 2,000 randomly selected
subscribers to a Canadian business periodical.
The response rate was 53.4%. One year later, the
same survey was sent to those who responded to
Measures
Time 1
Computer self-efficacy was measured by the 10Item instrument developed by Compeau and
Higgins (1995b). Outcome expectations were
measured by 11 items developed by Compeau
and Higgins (1995b). Six items relate to performance outcomes, and five items relate to personal outcomes (Table 3 shows the measures for
each of the constructs).
^Analysis of Ihe matching was also conducted by comparing age, gender, and educational background
across the two surveys. Any inconsistent matches were
removed from the final sample.
'Note that this analysis did not compare reported educational level at time 1 with educational level at time
2 (where a difference could reflect maturation in the
sample). The comparison was made on the time 1
demographic data for respondents who completed
only the time 1 survey and those who completed both
the time 1 and time 2 surveys.
Sample Composition
Age
Gender
Men
Women
86%
General Management
Marketing
Accounting/Finance
Information Systems
Engineering
Production
Human Resources
0th er^
28%
Executive
Middle Management
Professional
First Line Management
Technical/Clerical
Other
40%
Graduate Degree
Some Graduate Work
University or College Degree
Some University or College
Secondary School or Less
43%
Business
Science
Arts
Social Science
Other
64%
16%
6%
Educational Background
14%
18%
17%
5%
5%
4%
4%
18%
30%
15%
9%
3%
3%
5%
38%
11%
5%
6%
8%
Assessment of the research model was conducted using Partial Least Squares (PLS) Version
Data Analysis
Factor
Loading
Measure
SE8
SE9
SE 10
Pert.
Pert.
Pert,
Pert.
Perf,
Perf.
IF 1 USE A COMPUTER . . .
, , . 1 will be better organized
. . . 1 will increase my effectiveness on the job
. , . 1 will spend less time on routine job tasks
. . . 1 will increase the quality of output of my job
. . . 1 will increase the quantity of output for the same amount of effort
. . . 1 will be less reliant on clerical support staff
0.565
0.830
0,663
0.835
0.721
0,523
0.734
0.580
0.830
0.698
0.821
0,869
SE 1
SE2
SE3
SE4
SE5
SE6
SE7
Out. 1
Out. 2
Out, 3
Out. 4
Out. 5
Out. 6
Pers. Out.
Pers. Out,
Pers. Out.
Pers. Out.
Pers. Out.
Affect 1
Affect 2
Affect 3
Affect 4
Affect 5
Anxiety 1
Anxiety 2
1
2
3
4
5
0,807
0.791
0.822
0.814
0.821
0.799
0.791
0.711
0.740
0.805
0,816
0.646
0,721
0.693
0.873
Anxiety 3
Anxiety 4
Use
Use
Use
Use
0,776
0,731
0.710
0.657
1
2
3
4
0,792
0,892
0.909
n = 394
2.91.02.08 (Chin and Fry 1995), a regressionbased technique that can analyze structural models with multiple-item constructs and direct and
ICR
1.
2.
1.
2.
3.
4.
5.
6.
0.94
0.85
0.86
0,87
0.92
0.81
0.79
0.31
0.21
0.48
-0.54
0.43
0.70
0.53
0.43
-0.30
0.40
Self-efficacy
Performance. Out- Exp.
Personal Out. Exp,
Affect
Anxiety
Use
3.
0.74
0,27
-0.11
0.15
4.
0.75
-0.64
0.50
5.
6.
0.87
-0,44
0.72
152
path tests, the explained variance in the dependent constructs is assessed as an indication of the
overall predictive strength of the model.
Results
Measurement Model
Individual item loadings (Table 3) for the computer self-efficacy and anxiety constructs were all
above 0.70. While each of the other constructs
showed some weak (< 0.70) loadings, the internal consistency reliabilities were all greater than
0.7 (see Table 4) so no items were dropped. This
allowed consistency with the measures used in
the previous study (Compeau and Higgins
1995b). Further examination of Table 4 shows
that all constructs were more strongly correlated
with their own measures than they were with any
of the other constructs; thus, discriminant validity was observed.
Structural Model
The path coefficients from the PLS analysis are
shown in Figure 2. Consistent with recommended procedures (Barclay et al. 1995), jackknifing
was used to generate standard errors and t-statistics. A jackknife size of 10, yielding 39 sub-samples, was used.
Hypotheses 1 through 5 were supported. Self-efficacy was shown to exert a significant positive
influence on both perform a nee-related (HI) and
personal (H2) outcome expectations, a significant
Affect
= 32.2%
Computer
Self-Efficacy
Outcome
Expectations
(Performance)
= 9.4%
Anxiety
= 28.7%
Outcome
Expectations
(Personal)
= 4,6%
* p < .05
** p < .01
Usage
= 34.3%
*** p < .001 (based on t,3R,, two-tailed test)
Figure 2. Results
positive influence on affect (H3), a significant
negative influence on anxiety (H4), and a significant positive influence on use (H5).
Hypotheses 6 and 7 were also supported; performance outcome expectations exerted a significant positive influence on both affect (H6) and
use (H7). Hypothesis 8, which posited a significant relationship between personal outcome
expectations and affect, was not supported. With
respect to hypothesis 9, a significant relationship
betvi'een personal outcome expectations and use
was observed, but this relationship was negative,
contrary to the hypothesized relation.
Affect for computer use was found to exert a significant positive influence on usage (H10).
Hypothesis 11 was not supported. The path from
anxiety to use was not significant.
Figure 2 also shows the explained variance for
each of the constructs in the model.
Approximately 34% of the variance in usage is
Discussion
The results of this study confirm many of the
results of the earlier cross-sectional study
(Compeau and Higgins 1995b), and strengthen
the findings by showing the continuing predictive capability of self-efficacy and performancerelated outcome expectations, even when measured one year prior to affective and behavioral
responses. Self-efficacy is a strong and significant
predictor of affect, anxiety, and use one year
later. When both the direct and indirect effects
are taken into account, self-efficacy explains a
154
supervisory skill (Latham and Saari 19791, attendance behavior IFrayne and Latham 1987|,
mathematics skill ISchunk 1981], and academic
productivity (Taylor et al. 19891), it becomes evident that self-efficacy with respect to information
technology use will continue to be a factor in our
choices about what technologies to adopt, how
much to use them {if we have that choice), and
how much to persist in the face of obstacles to
successful use of such technologies.
For researchers, the findings of the longitudinal
extension of Compeau and Higgins (1995b) provide evidence of the robustness of tbe Social
Cognitive Theory model of individual reactions
to computing technology, at least in part. Civen
tbe similarities between the Social Cognitive
Theory model and other models of technology
adoption and use discussed earlier, it is reasonable to extend this conclusion, albeit with some
caution, to these other models. Outcome expectations, measured in this study, are similar to the
concepts of perceived usefulness (Davis 1989),
relative advantage and image (Compeau and
Meister 1997; Moore and Benbasat 1991) and
behavioral beliefs (Mathieson 1991; Taylor and
Todd 1995). Thus, the findings that performancerelated outcome expectations at one point in
time predict affect and use one year later can reasonably be extended to these closely related constructs. It would appear that cognitively based
models of Eecbnology use evidence predictive
validity, even over time separations of one year.
A ckno wiedgements
This research was supported by a grant from the
Social Science and Humanities Research Council
of Canada (ref # 410-92-1526).
References
Bandura, A. Social Foundations of Thought and
Action. Prentice-Hall, Englewood Cliffs, NJ,
1986.
Bandura, A., Adams, N. E., and Beyer, |.
"Cognitive Processes Mediating Behavioral
Change," Journal of Personality and Social
Psychology (35:3), 1977, pp. 125-139.
Barclay, D., Higgins, C, and Thompson, R. "The
Partial Least Squares Approach to Causal
Modeling; Personal Computer Adoption and
156
158