See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/220079846

Predicting User Intentions: Comparing the Technology Acceptance Model with the Theory of
Planned Behavior
Article in Information Systems Research · September 1991
DOI: 10.1287/isre.2.3.173 · Source: DBLP

CITATIONS READS

2,549 2,890

1 author:

Kieran Mathieson
Oakland University
36 PUBLICATIONS 3,661 CITATIONS

SEE PROFILE
All content following this page was uploaded by Kieran Mathieson on 17 July 2016.

The user has requested enhancement of the downloaded file.

Predicting User Intentions: Comparing the Technology Acceptance Model with

the Theory of Planned Behavior

Abstract

Information systems (IS) cannot be effective unless they are used. However, people sometimes do not
use systems that could potentially increase their performance. This study compares two models that
predict an individual's intention to use an IS: the technology acceptance model (TAM) and the theory
of planned behavior (TPB). The comparison was designed to be as fair as possible, not favoring one
model over the other. Both TAM and TPB predicted intention to use an IS quite well, with TAM
having a slight empirical advantage. TAM is easier to apply, but only supplies very general
information on users' opinions about a system. TPB provides more specific information that can
better guide development.
 1

1. Introduction
Information systems (IS) have the potential to improve organizational performance, but only
if they are actually used. Although firms require that certain systems be used for some tasks, in
other cases system use is voluntary. The more these systems are used, the greater the impact
they can have (Trice & Treacy, 1988). People are sometimes unwilling to use systems, however,
even if the IS could increase their job performance (Nickerson, 1981).

Since systems that are not used cannot be effective, no matter what their technical merits, it is
important to understand how people decide whether they will use a particular IS. The issues that
influence this decision are likely to vary with the system, the individual, and the context. If these
issues can be identified, developers can take them into account during system design.

Note that the emphasis is on IS use, not the IS itself. An individual might feel that a system
could help improve decision making, but not use it because of lack of convenient access to a
terminal, lack of time, etc. Research in social psychology shows that behavior is best predicted
by an individual's attitude towards the behavior itself (such as using an IS), rather than his or her
attitude towards objects involved in the behavior (such as the IS) (Fishbein & Ajzen, 1975). A
positive evaluation of an IS may be a necessary but not always sufficient condition for system
use.

Developers employ a number of techniques to ensure that users will accept the systems they
build. User participation in design is seen as a key to achieving acceptance. Although the
empirical evidence is mixed (Barki & Hartwick, 1989), many believe that systems developed
with user participation will better match user requirements and capabilities than systems
designed solely by IS professionals. In some system development methodologies (e.g., Naumann
& Jenkins, 1982; Gane 1989), a small group of users participate heavily in systems specification
and logical design. However, this approach may only be effective if the users participating in the
design are representative of the final user base. When a system is designed to serve a large
number of people, perhaps the development team should ensure that an IS acceptable to the team
is also acceptable to the broad spectrum of users.

During development, it is difficult to estimate eventual system use, since the system does not
yet exist. An individual's intention to use the system can be measured, however. There is
considerable evidence that intention to perform a behavior predicts actual behavior (Sheppard,
 2

Hartwick, and Warshaw, 1988).

Measuring user acceptance is not important only during design or immediately after
implementation. Over time, there will be changes in the system, the users, and the environment
in which both operate (Swanson, 1988). The business environment might change, affecting
users' information requirements. Changes made to satisfy one group of users may make the
system less suitable for other purposes. Users' expectations might change as they become more
familiar with IS technology, and what was once acceptable may no longer be adequate (Doll &
Ahmed, 1983).

This study compares two models that predict an individual's intention to use an IS. The first
is the technology acceptance model (TAM), specifically designed by Davis (1986) to predict use
of an IS. The second is the theory of planned behavior (TPB), discussed by Ajzen (1985; 1989).
TPB was designed to predict behavior across many settings, and can be applied to IS use.

The models are compared on three criteria. First, how well do they predict the user's
intention to use an IS? If one model predicts intention much better than another, it can provide a
more accurate picture of the issues that developers should consider in addressing system
acceptability. Answering this question requires a fair empirical comparison, that is, a
comparison that is not biased in favor of one model or the other. Second, how valuable is the
information provided by the models? If the models do not supply information that can guide
development, they will not be useful to systems analysts, no matter how well they predict
intention. Third, how difficult are the models to apply? Ideally, the models would provide
valuable information at a low cost. Answers to these questions will help decide, first, whether
the models are useful at all, and, second, the conditions under which one might be more useful
than the other.

The paper proceeds as follows. First, TAM and TPB are described, and the differences
between them examined. The conditions necessary for a fair comparison, identified by Cooper
and Richardson (1986), are reviewed. Second, an empirical study, designed to compare the
extent to which the models predict intention to use a system, is described. Its results are then
presented. Finally, the limitations of the study and the implications of its findings are discussed.
 3

2. The Models

2.1. The Technology Acceptance Model

TAM:

... is specifically meant to explain computer usage behavior.... (p. 983) The goal of TAM
is to [be] ... capable of explaining user behavior across a broad range of end-user
computing technologies and user populations, while at the same time being both
parsimonious and theoretically justified. (p. 985)

(Davis, Bagozzi, & Warshaw, 1989)

Figure 1 shows the model. Ease of use (EOU) is "the degree to which the ... user expects the
target system to be free of effort" (Davis et al., 1989, p. 985). Usefulness (U) is the user's
"subjective probability that using a specific application system will increase his or her job
performance within an organizational context" (p. 985). U is influenced by EOU. Both EOU
and U predict attitude (A), defined as the user's evaluation of the desirability of his or her using
the system. A and U influence the individual's intention to use the system (I). Actual use of the
system is predicted by I.

Figure 1 about here

TAM is fairly new, and has not been extensively tested. The empirical tests that have been
conducted suggest it predicts intention fairly well. Davis et al. (1989) found that TAM
successfully predicted use of a word processing package. Davis (1989) reports that EOU and U
were significantly correlated with use of an office automation package, a text editor, and two
graphics packages.

2.2. The Theory of Planned Behavior

TPB is outlined in Figure 2. Behavior is determined by intention (I) to perform the behavior.
Intention is predicted by three factors: attitude toward the behavior (A), subjective norms (SN),
and perceived behavioral control (PBC). Both A and I are defined as for TAM. SN is the
individual's perception of social pressure to perform the behavior. PBC is the individual's
perception of his or her control over performance of the behavior.

Figure 2 about here

 4

Beliefs are antecedent to attitude, subjective norms, and perceived behavioral control.
Attitude is a function of the products of behavioral beliefs and outcome evaluations. A
behavioral belief is the subjective probability that the behavior will lead to a particular outcome.
The outcomes are fairly specific, utilitarian outcomes, such as "Using the system will save time
compared to current methods." An outcome evaluation is a rating of the desirability of the
outcome. The following equation reflects this process:

nb
A = ∑ bbi oei i=1

where

bb = behavioral belief i i

oe = outcome evaluation of belief i i

nb = number of salient outcomes

As an example, suppose a sales representative is considering using a laptop PC to access a
central database on product availability. A potential outcome from using the system is improved
customer service. The relevant behavioral belief is the extent to which she believes using the
system will improve customer service. The associated outcome evaluation would be the
importance of improving customer service. Because behavioral beliefs and outcome evaluations
are multiplied, they would have the greatest impact on attitude if both (1) the sales representative
felt that the system would improve customer service, and (2) improving customer service is
important.

Subjective norms reflect the perceived opinions of referent others. A "referent other" is a
person or group whose beliefs may be important to the individual. A normative belief is the
individual's perception of a referent other's opinion about the individual's performance of the
behavior. Motivation to comply is the extent to which the person wants to comply with the
wishes of the referent other. In equation form:

no
SN = ∑ nbi mci
i=1
 5

where

nb = normative belief about referent other i i

mc = motivation to comply with referent other i i

no = number of salient others

In the example above, the sales representative might feel that the other sales representatives
would approve of her using the system. This would be a normative belief. The relevant
motivation to comply is the importance she attaches to the opinions of other sales
representatives. Again, the two are multiplied, so even if she felt that other representatives
would approve of her using the laptop, this would not impact her intention to use the system if
she did not care about their opinions.

Perceived behavioral control (PBC) refers to the individual's perceptions of "... the presence
or absence of requisite resources and opportunities" (Ajzen & Madden, 1986, p. 457) necessary
to perform the behavior. PBC depends on control beliefs and perceived facilitation. A control
belief is a perception of the availability of skills, resources, and opportunities. Perceived
facilitation is the individual's assessment of the importance of those resources to the achievement
of

outcomes. The appropriate equation is:

nc
PBC = ∑ cbi pfi
i=1

where

cb = control belief about availability of skill, resource, or opportunity i i

pf = perceived facilitation of skill, resource, or opportunity i i

nc = number of salient skills, resources, or opportunities

Control beliefs can be situational (e.g., having access to a terminal) as well as personal (e.g.,
being able to use a system). PBC goes beyond TAM's ease of use (EOU) construct to embrace
other barriers to system use. Suppose that the laptop PC in the example above requires access to
 6

a telephone to contact a central mainframe. The sales representative often visits building sites,
where there is no telephone available. Her control belief about the availability of a telephone
(one of resources required to perform the behavior) would be low. However, she might rate the
perceived facilitation of telephone availability as high. That is, telephone access is important, but
it is often not available.

Overall, the sales representative might not be inclined to use the system. Although the
behavior (system use) might achieve a valuable outcome (improved customer service), there
would be little social benefit, and she does not have easy access to all of the necessary resources.
The weights in the equations (oe , mc , and pf ) can be measured in two ways. First, the
i i i

individual can be asked to specify them using, for example, a Likert scale (direct assessment).
Second, the weights can be estimated as coefficients in regression equations (indirect
assessment). Direct assessment is useful when subjects disagree about the sign of a weight
(Fishbein & Ajzen, 1975, p. 238). However:
When the evaluative polarity of an outcome is fairly homogeneous across
subjects, the corresponding belief tends to be monotonically related to
attitudes, and statistically estimated weights tend to accurately capture the
actual usage of information cues ... and generally predict dependent variables
as well as subjective weights....
Davis et al. (1989), p. 988

Both approaches are compared below for the IS-use case.

Since the theory of planned behavior (TPB) is fairly new, there have been relatively few
empirical tests of its effectiveness. Schifter & Ajzen (1985) successfully applied TPB to the
prediction of weight loss behavior. Ajzen & Madden (1986) used TPB to predict students'
decisions about attending class and earning a good grade.

There have been more tests of the theory of reasoned action (TRA), on which TPB is based.
The main difference between the models is that TRA does not consider perceived behavioral
control. It predicts behavior solely from attitudes and subjective norms, and is predictive in
those situations were there are no significant barriers to behavioral performance (Fishbein &
Ajzen, 1975). Sheppard et al. (1988) report a meta-analysis of 87 studies from which they
 7

concluded that there is "strong support for the overall predictive utility of the Fishbein and Ajzen
[TRA] model" (p. 336). In the computing domain, Yeaman (1988) found that TRA predicted
intention to learn to use a microcomputer, although subjective norms did not contribute to the
prediction. Davis et al. (1989) reported that TRA predicted intention to use a word processing
program, although, again, subjective norms did not contribute to the explained variance.

2.3. Differences Between the Models

There are three main differences between TAM and TPB. The first is their varying degree of
generality. The second is that TAM does not explicitly include any social variables. The third is
that the models treat behavioral control differently. Each of these points is discussed below.

2.3.1. Degree Of Generality

TAM assumes that beliefs about usefulness and ease of use are always the primary
determinants of use decisions. This was a conscious choice on the part of Davis et al. (1989),
since they wanted to use "a belief set that ... readily generalizes to different computer systems
and user populations" (p. 988). TPB uses beliefs that are specific to each situation. The model
does not assume that beliefs that apply in one context also apply in other contexts. Although
some beliefs may generalize across contexts, others may not.

This difference between the models raises three concerns. First, in some situations there
could be variables besides ease of use and usefulness that predict intention. For example,
accessibility might be an important factor for users who are not always near a terminal.
Identifying such beliefs is part of the standard methodology for using TPB. While such
exploration is not excluded from TAM, it is not an essential part of the model.

Second, TPB is more difficult to apply across diverse user contexts than TAM. TAM's
constructs are measured in the same way in every situation. TPB requires a pilot study to
identify relevant outcomes, referent groups, and control variables in every context in which it is
used. This can be complex if different user groups focus on different outcomes from use of the
same system. For example, students using a computer-aided learning system might be interested
in maximizing exam scores, while instructors are interested in saving class time. Ideally, TPB's
instruments would be tailored to each group.

Third, some TPB items require an explicit behavioral alternative if they are to be as specific
as possible. For example, in asking someone whether using a spreadsheet for sales forecasting
 8

will save time (a behavioral belief), it is best to explicitly identify an alternative behavior so that
the basis for comparison is clear. Potential users might be asked to respond to the following
item: "Using a spreadsheet instead of a calculator will save me time in developing sales forecasts.

(Agree/Disagree)." TAM does not require the identification of a specific comparison behavior.
The advantage of TPB's approach is that all respondents are making the same comparison. The
comparison target is not specified in TAM's instruments, and may vary across subjects (Ryan &
Bock, 1990). The disadvantage of TPB's approach is that this reference point may not apply to
all individuals. For example, some people may be generating sales forecasts using a specialized
DSS instead of a calculator, so the question may not provide a useful comparison to current
practices.

2.3.2. Social Variables

The second major difference between TAM and TPB is that TAM does not explicitly include
any social variables. These are important if they capture variance that is not already explained by
other variables in the model. Davis et al. (1989) point out that social norms are not independent
of outcomes. For example, an individual might perceive pressure from his or her supervisor to
use a system, with an implied outcome of non-use being a poor performance evaluation. That is,
social norms will already have been taken into account to some extent in the evaluation of
outcomes.

However, the social variables in TPB may still capture unique variance in intention. There
could be social effects that are not directly linked to job-related outcomes such as usefulness.
For example, some individuals might use a system because they think they will be perceived by
their coworkers as technologically sophisticated. This motivation is more likely to be captured
by TPB than by TAM.

2.3.3. Behavioral Control

The third major difference between TAM and TPB is their treatment of behavioral control,
referring to the skills, opportunities, and resources needed to use the system. The only such
variable included in TAM is ease of use (EOU). Examining the EOU items used by Davis (1989,

p. 340), it is apparent that EOU refers to the match between the respondent's capabilities and the
skills required by the system. The items include "Learning to operate [the system] would be easy
for me," and "My interaction with [the system] would be clear and understandable."
 9

Although possession of requisite skills is important, sometimes other control issues will arise.
Ajzen (1985) differentiates between internal control factors that are characteristics of the
individual, and external factors that depend on the situation. Internal factors include skill and

will power. External control factors include time, opportunity, and the cooperation of others.
For instance, where connect time and CPU usage are charged to user departments, some people
might not have the resources necessary to use a system, even if they feel they could benefit from
doing so and have the necessary skills. In other words, they are denied the opportunity to use the
system by external factors.

EOU corresponds to the internal factor of skill. However, external control issues are not
considered in TAM in any obvious way. Although it could be argued that the EOU item "I
would find [the system] easy to use" (from Davis, 1989) implies that respondents consider
external control issues, this is not explicit.

Some control factors will be stable across situations, while others will vary from context to
context (Ajzen, 1985). An individual takes the same skills from situation to situation, and to the
extent that similar skills are required for different IS-related tasks, ability should be a fairly stable
control factor. In fact, Hill, Smith, and Mann (1987) found that a general efficacy measure
predicted intentions to use a wide range of technologically advanced products. However, some
control issues will be idiosyncratic to particular circumstances. For example, while the
availability of a telephone line was important to the sales representative, it will not be important
to other people in other situations.

TPB taps the important control variables for each situation independently, and is more likely
to capture such situation-specific factors. TAM is less likely to identify idiosyncratic barriers to
use. This is in keeping with the stated objective of Davis et al. (1989) to develop a model that is
applicable across many situations, but will cause the model to miss control issues that are
important in particular contexts.