COMPUTER SELF-EFFICACY - Human-Computer Interaction and Management Information Systems: Foundations - page 210

The importance of task-focus relates, in part, to the distinction between computer self-efficacy
and related constructs. Component skills, as reflected by the Torkzadeh and Koufteros items, are
more difficult to separate from self-reports of cognitive user competence (Marcolin et al., 2000b).
Thus, they might be seen as antecedent to the ability to use a computer for task-oriented behavior,
but they lack the generative sense of being able to organize and execute courses of action. Sein et al.
(1999) present a useful conceptualization of software knowledge levels that may be helpful in decid-
ing on the types of questions to include in computer self-efficacy measures. They present a six-level
hierarchy, with the first three levels representing various aspects of procedural knowledge. 1 The
lowest is command-based knowledge (e.g., how to click on a button), followed by tool procedural
knowledge (e.g., how to create a document or transaction), and business procedural knowledge (e.g.,
how to query a database for needed transactions). We maintain that a focus on either business pro-
cedural or tool procedural knowledge is preferable for measuring computer self-efficacy; such foci
are also more closely aligned with Bandura's definition of self-efficacy, since the knowledge types
focus on courses of action rather than on the component subskills required to complete the action.
A second issue with respect to the definition of computer self-efficacy relates to the dimensionality
of the measure. Some recent studies suggest evidence of unreliability or multi-dimensionality in the
Compeau and Higgins (1995b) measure of general computer self-efficacy. Gundlach and Thatcher
(2000) argued that this was a multi-dimensional measure, reflecting “human assisted” and “individ-
ual” computer self-efficacy. Data presented by Thompson et al. (under review) supports an alternative
view. It appears as though the items that reflect easier situations in which to use a system have become
subject to a floor effect, and thus have different variance. In their study, the variances for the items on
one factor were all smaller than those for the other factor (Table 11.1) and the distributions were
markedly skewed for this factor as well. Thus, the existence of a second factor may reflect measure-
ment issues rather than a conceptually different factor.
Interestingly, a comparison of the results from two studies, conducted in 1990 (Compeau and
Higgins, 1995b) and 1999 (Marcolin et al., 2000a) do not show an increase in the means for the
self-efficacy items, nor do they show large decreases in the variance (Table 11.2). The work of
Gundlach and Thatcher and of Thompson et al. was conducted using student subjects, so perhaps
the differences reflect those of different subgroups. Thompson et al. measured the construct on a
seven-point scale, rather than using the two-part response recommended by Bandura, but this would
not explain the variance difference. Moreover, data collected by Gravill et al. (2001) using managers
and professionals used a seven-point scale but did not observe multi-dimensionality in the results.
Thus, it seems important to ensure adequate variability in the response items, especially with student
populations. Such variability may not be evident in all of the original questions developed by Compeau
and Higgins.
Generality of Self-Efficacy
Bandura (1986, 1997) argues that self-efficacy (SE) judgments must be specific to the task in order
to consistently predict behavior. He contrasts this domain-specific notion of SE with an overall gen-
eral SE reflecting one's confidence in general (Bandura, 1997). Bandura argues that overall general
SE is too broad to be of high predictive value, and that variations in domain specific confidence must
be taken into account. This makes intuitive sense and has been discussed by several authors in infor-
mation systems research. Yet the definition of “specific” remains relatively unexplored.
Marakas et al. (1998) provide an initial conceptualization of the domain for IS specific self-
efficacy judgments. They suggest that GCSE is the weighted average of a collection of SCSE
(defined at the software package level or the application environment level). They further argue