Reliability assessment of the attitude towards computers instrument (ATCI) Teresa M. Shaft * , Mark P. Sharfman, Wilfred W. Wu Michael F. Price College of Business, 307 W. Brooks, University of Oklahoma, Norman, OK 73019-4006, USA Available online 18 November 2003 Abstract Individual’s attitude towards computers is a key component to understanding user’s accep- tance and satisfaction with computer-based information systems. As such, individuals’ attitudes towards computers have been of interest to researchers in a variety of settings for sometime. Therefore, numerous instruments have been developed to assess this construct. We describe these instruments and discuss issues for researchers to consider when selecting an instrument with which to assess attitude towards computers. We find few instruments that are suitable for a general setting or have had their reliability thoroughly assessed. We, therefore, present a reli- ability assessment of the Attitude Towards Computers Instrument (ATCI), which was designed to be applicable in a wide variety of settings. The reliability assessment includes latent structure (confirmatory factor) analysis, internal consistency (Cronbach’s alpha) analysis and stability (test–retest) analysis. We find that the reliability of the ATCI compares favorably with existing instruments, making it a better choice for many research settings. Reliability analysis such as presented in this paper helps move the information systems field forward by providing researchers with a reliable instrument with which to assess attitude towards computers. Ó 2003 Elsevier Ltd. All rights reserved. Keywords: Computer attitudes; Attitude measures; Attitude measurement; Computers 1. Introduction Computer-based information systems have become an integral part of managerial decision making (cf. Galletta & Lederer, 1989). Despite the widespread availability * Corresponding author. Tel.: +1-405-325-2880; fax: +1-405-325-7482. E-mail address: [email protected](T.M. Shaft). 0747-5632/$ - see front matter Ó 2003 Elsevier Ltd. All rights reserved. doi:10.1016/j.chb.2003.10.021 www.elsevier.com/locate/comphumbeh Computers in Human Behavior 20 (2004) 661–689 Computers in Human Behavior
29
Embed
Reliability assessment of the attitude towards computers instrument (ATCI)
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
omputers inuman Behavior
CH
www.elsevier.com/locate/comphumbeh
Computers in Human Behavior 20 (2004) 661–689
Reliability assessment of the attitudetowards computers instrument (ATCI)
Teresa M. Shaft *, Mark P. Sharfman, Wilfred W. Wu
Michael F. Price College of Business, 307 W. Brooks, University of Oklahoma, Norman,
OK 73019-4006, USA
Available online 18 November 2003
Abstract
Individual’s attitude towards computers is a key component to understanding user’s accep-
tance and satisfaction with computer-based information systems. As such, individuals’ attitudes
towards computers have been of interest to researchers in a variety of settings for sometime.
Therefore, numerous instruments have been developed to assess this construct. We describe
these instruments and discuss issues for researchers to consider when selecting an instrument
with which to assess attitude towards computers. We find few instruments that are suitable for a
general setting or have had their reliability thoroughly assessed. We, therefore, present a reli-
ability assessment of the Attitude Towards Computers Instrument (ATCI), which was designed
to be applicable in a wide variety of settings. The reliability assessment includes latent structure
(confirmatory factor) analysis, internal consistency (Cronbach’s alpha) analysis and stability
(test–retest) analysis. We find that the reliability of the ATCI compares favorably with existing
instruments, making it a better choice for many research settings. Reliability analysis such as
presented in this paper helps move the information systems field forward by providing
researchers with a reliable instrument with which to assess attitude towards computers.
662 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
of information systems, many organizations do not gain the full benefit of their
systems because some individuals resist using them. In fact, ‘‘(u)nderstanding why
people accept or reject computers has proven to be one of the most challenging issues
in information systems research’’ (Davis, Bagozzi, & Warshaw, 1989, p. 982). Pre-
vious research suggests that attitudes play a key role in predicting user acceptance
and satisfaction (Bailey & Person, 1983; Coffin & MacIntyre, 1999; Ives, Olson, &Baroudi, 1983). In response to this challenge, a large body of research has been
conducted to determine the effects of attitudes and beliefs on individuals’ use of
1982). As such, attitudes towards computer have been used to predict constructs as
varied as satisfaction with end-user computing (e.g. Rivard & Huff, 1988) and the
effect of implementing IS on organizational power distributions (Burkhardt & Brass,
1990). However, Davis et al. (1989, p. 983) suggested that these research findings
have been mixed and inconclusive. In part this may be due to the wide array ofdifferent attitude, belief and satisfaction measures which have been employed, often
without adequate theoretical or psychometric justification.
This paper takes a step towards developing theoretically and psychometrically
justified instruments by first considering the many attitude towards computers in-
struments that exist and their psychometric properties. We then present a reliability
analysis of the Attitudes Towards Computers Instrument (ATCI). The ATCI differs
from many related instruments. Specifically, it was developed to be applicable to a
broad range of populations, including business professionals, and to be relativelyshort, such that it can be used in studies that require the assessment of numerous
constructs.
We organize the remainder of the paper as follows. First, we describe the theo-
retical relationship between attitude and behavior and distinguish attitude toward
computers from related constructs. Then, we present an overview of computer at-
titude instruments and related reliability studies. We then present evidence of the
ATCI’s reliability through latent structure (confirmatory factor) analysis, internal
consistency (Cronbach’s alpha) analysis and stability (test–retest) studies. We pres-ent the purpose, methods, analytic techniques and results for each analysis separately.
2. The theoretical relationship between attitude and behavior
Studying Management Information Systems (MIS) users’ attitudes was basic to
the creation of ‘‘a theory of MIS development in which MIS success and failure is
explained’’ (Swanson, 1982, p. 157). Attitudes provide people with a framework tointerpret the world and integrate new experiences (Galletta & Lederer, 1989). Ajzen
and Fishbein argue that by understanding an individual’s attitudes towards some-
thing, one can predict that individual’s ‘‘overall pattern of responses to the object’’
(1977: 888) as new experiences occur. Further, when there is a clear linkage between
the target action and any attitudes that are formed, the degree of predictability will
be highest (Ajzen & Fishbein, 1977). Given the pervasiveness of computers in soci-
ety, it is likely that most individuals have developed some attitudes towards these
Fig. 1. The relationships among attitudes, intentions, and behaviors, developed from Fishbein and Ajzen
(1975).
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 663
machines. As such, intentions concerning computer use should also be well devel-oped. Therefore, consistent with Ajzen and Fishbein (1977), if we understand atti-
tudes towards computers, we should be able to predict computer-related behaviors
and choices. To predict these behaviors and decisions with assurance requires a re-
liable attitudes towards computers instrument.
As we show in Fig. 1, an attitude can be exogenous to intention and (indirectly)
exogenous to behavior or endogenous from behavior. In researching computer users’
attitudes, investigators can use attitude as either an independent variable that pre-
dicts computer-related behaviors (see, for example, Burkhardt & Brass, 1990) or as adependent variable by which the researchers examine the antecedents of users’ at-
titudes (see, for example, Burkhardt, 1994; or Igbaria & Parasuraman, 1989). Be-
cause of the variety of ways that attitudes towards computers can be used as a
construct within information systems research, a theoretically and psychometrically
justifiable measure of attitudes towards computers has the potential to be useful in a
variety of research settings.
3. Distinction between attitude towards computers and related concepts
Although we focus on instruments that assess attitudes towards computers, there
are three related concepts commonly found in the IS literature: computer anxiety,
user information satisfaction (UIS), and the technology acceptance model (TAM).
We consider attitudes towards computers distinct from each of the other three and
our reasoning follows. Of the other three constructs, perhaps the most closely related
construct is computer anxiety. In the past, computer anxiety and attitudes towardscomputers were seen as synonymous (i.e. an individual who experiences high levels of
computer anxiety is said to have a negative attitudes towards computers) or as
separate variables with common antecedents. However, evidence suggests that
computer anxiety is an intervening variable between other variables such as
664 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
demographics and attitudes towards computers (Igbaria & Parasuraman, 1989).
Therefore, it appears that computer anxiety and attitude towards computers are
distinct constructs.
Regarding UIS, such instruments measure the success or failure of a particular IS
(Galletta & Lederer, 1989). Although attitudes and satisfaction are related (Galletta
& Lederer, 1989), instruments to assess UIS and attitudes towards computers op-erate at different levels. The focus when measuring attitude towards computers is on
a relatively stable individual characteristic or trait, irrespective of a specific system.
UIS instruments focus on a referent, a particular IS or a specific program and is
considered a state rather than a trait variable (Galletta & Lederer, 1989). Although
an interplay between attitudes towards computers and UIS exists (Galletta & Le-
derer, 1989), the focus on individual’s attitudes towards computers in general and as
a trait rather than a state distinguishes it from UIS.
TAM (Davis, 1989) is an important theory that has been widely investigated andapplied in the IS literature. TAM builds upon the Ajzen and Fishbein (1977) Theory
of Reasoned Action. In TAM, an individual’s perceived usefulness and perceived
ease of use of a particular information system influences their attitude toward using
that system, which affects their intention to use the system and, in turn, their actual
use of the information system. However, similar to user information satisfaction,
TAM focuses on an individual’s acceptance of a particular IS rather than assessing a
general trait. Hence, we regard TAM as distinct from attitude toward computers.
These four concepts (attitude, anxiety, information satisfaction and technologyacceptance) are important concepts in understanding individual’s use of computer
systems. We focus on attitude towards computers because it has been conceptualized
as a trait variable rather than state variable relevant to a specific IS. As such, an
instrument to assess attitude towards computers has application to many research
settings. In addition, despite the importance of attitude towards computers and its
long history of use as a research construct, as we discuss in the next section, psy-
chometric justification of such instruments has been limited. Given the wide-spread
use of this construct, it appears that an instrument whose reliability has been rig-orously confirmed could benefit many researchers.
4. Previous research on attitudes towards computers
Table 1 provides an overview of the many instruments developed to assess atti-
tude towards computers. Within Table 1, we present the instruments in chrono-
logical order based on the publication date of the original instrument. In subsequentdiscussion, we refer to each instrument by number, as indicated in the column la-
beled ‘‘entry.’’ The entry identifier is based on the original instrument. We organized
the table such that follow up studies of an instrument appear directly below the entry
for the original instrument. In developing Table 1, we focused on reports of new
instruments or studies that investigated the psychometric properties of an instru-
ment. Therefore, studies that utilized an existing instrument that did not include
assessment of the instrument’s psychometric properties we excluded from our review.
Re-evaluation Woodrow (1991) 30 Students Student Teachers Likert Exploratory
(Varimax)
r¼ 0.94
(split-half)
n.a.
Re-use &
Re-evaluation
Nash and Moroz
(1997)
40 Educators Certified
Teachers
Likert Confirma-
tory
(Varimax)
a¼ 0.86 n.a.
9 Computer Attitude
Scale
Collis (1984) 24 Students Secondary
Students
Likert Exploratory n.a. n.a.
10 Attitude Towards
MIS (ATMIS)
Kjerulff and
Counte (1984)
20 Students Hospital Staff
Volunteers
Likert n.a. a¼ 0.89 n.a.
Attitude Toward
Computers in
General (ACG)
Kjerulff and
Counte (1984)
20 Students Hospital Staff
Volunteers
Semantic
Differential
n.a. a¼ 0.85 n.a.
11 Computer Attitude
Scale (CATT)
Dambrot,
Watkins-Malek,
Silling, Marshall,
& Garver (1985)
20 Students College
Freshman
Likert n.a. a¼ 0.79–
0.84
n.a.
12 Cognitive &
Affective computer
attitudes
Bannon et al.
(1985)
17 Educators &
Students
Students &
Educators
Likert Exploratory
(Varimax)
a¼ 0.90–
0.93
n.a.
13 Attitude Toward
CAI
Allen (1986) 21 Students
(professional:
nursing, EE,
etc.)
Nursing
Students
Semantic
Differential
Exploratory
(Varimax)
a¼ 0.58–
0.83
n.a.
14 Computer Attitude
Scale (CAS)
Nickell and
Pinto (1986)
20 General
Population
Students Likert n.a. a¼ 0.81 2-weeks
r¼ 0.86
666
T.M
.Shaftet
al./Computers
inHumanBehavio
r20(2004)661–689
15 Computer Attitude
scale
Abdel-Gaid,
Trueblood, and
Shrigley (1986)
23 Teachers Secondary
Teachers
Semantic
Differential
Exploratory a¼ 0.89 n.a.
16 Attitudes Toward
Computer Usage
Scale (ATCUS)
Popovich, Hyde,
and Zakrajsek
(1987)
20 General
Population
Undergraduate
College Students
Likert Exploratory a¼ 0.84–
0.88
2-weeks
r¼ 0.81
to .91
Re-evaluation Belleau and
Summers (1993)
20 General
Population
Undergraduate
students
Likert Exploratory
(Varimax)
a¼ 0.4259 n.a.
17 Bath County
Computer Attidues
Inventory (BCCAS)
Bear et al. (1987) 26 Students 4–12th Graders Likert n.a. a¼ 0.94 n.a.
18 Attitudes Toward
Computers Scale
(ATCS)
Rosen et al.
(1987)
26 General
Population
University
Students (in
multiple studies)
Likert Exploratory a¼ 0.76 n.a.
19 Minnesota
Computer Literacy
& Awareness
Assessment
Instrument
(MCLAA)
Swadener and
Hannafin (1987)
17 Students 6th Graders Likert n.a. a¼ 0.68–
0.74
n.a.
Redesign Comber Colley,
Hargreaves, &
Dorn (1998)
12 Students Secondary
School Students
Likert n.a. a¼ 0.79–
0.88
n.a.
20 Attitudes to
Computers of
Managers in the
Hospitality Industry
Gamble (1988) 17 Managers in
Hospitality
Industry
Hospitality
Workers
n.a. n.a. n.a. n.a.
21 Computer Attitude
Measure (CAM)
Kay (1989) 30 General
Population
Student Teachers Likert Scale
& Semantic
Differential
Exploratory
(Varimax)
a¼ 0.87–
0.94
n.a.
T.M
.Shaftet
al./Computers
inHumanBehavio
r20(2004)661–689
667
Table 1 (continued)
Entry Scale Author # of
Items
Target
population
Sample for
reliability
analysis
Item
format
Factor
analysis
Internal
consistency
Test–
retest
22 Computer Attitudes
& Learning Perfor-
mance
Gattiker and
Hlavka (1992)
17 Students Students Likert Exploratory
(Varimax)
a¼ 0.68 n.a.
23 Attitudes Toward
Computers Ques-
tionnaire (ATCQ)
Jay and Willis
(1992)
32 Elderly Elderly in a
Community
Home
Likert n.a. n.a. n.a.
24 Attitude Toward
Computer Scale
(ATCS)
Francis (1993) 24 Students BYU
Undergrads
Likert Exploratory
(Varimax)
n.a. n.a.
25 Computer Attitude
Survey
Klein, Knupfer,
and Crooks
(1993)
15 Students Students Likert Exploratory
(Varimax)
a¼ 0.69–
0.83
n.a.
26 Computer Attitude
Scale for Secondary
Students (CASS)
Jones and Clarke
(1994)
40 Students Secondary
Students
Likert n.a. a¼ 0.95 2-weeks
r¼ 0.84
27 Teacher Computer
Attitude Scale
Huang et al.
(1995)
23 Educators Teachers Likert n.a. a¼ 0.73–
0.95
n.a.
28 ETSU Computer
Attitude Scale
Dubois et al.
(1995)
23 Educators Student Teachers Semantic
Differential
n.a. n.a. n.a.
29 Attitudes toward
Technology
Pelton and Pel-
ton (1997)
42 Educators College Students Likert Exploratory
(Varimax)
n.a. n.a.
30 Computer Attitudes
of non-computing
Academics
Seyal et al. (1999) 12 Non-comput-
ing Academics
Academics Likert Exploratory
(Principal)
a¼ 0.79 n.a.
31 Computer Self-effi-
cacy Scale
Young (2000) 48 Students Middle & High
School Students
Likert Exploratory
(Principal)
a¼ 0.64–
0.87
n.a.
668
T.M
.Shaftet
al./Computers
inHumanBehavio
r20(2004)661–689
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 669
This diverse set of instruments designed to study attitudes towards computers
dates back to Lee’s (1970) study of social attitudes towards ‘‘electronic brain ma-
chines’’ (p. 55). Since Lee’s pioneering study, many instruments to measure attitudes
towards computers have appeared, particularly after the theoretical link between
attitudes and behavior became well accepted (Ajzen & Fishbein, 1977). Rather than
discuss these instruments in the chronological fashion of the table, we organize ourdiscussion around several issues that we believe are salient to researchers when se-
lecting an instrument with which to measure attitude towards computer. In partic-
ular, we noted when some instruments were particularly long, complex, developed
for a specialized setting, or lack psychometric justification.
4.1. Instrument length
One issue for researchers to consider when assessing attitudes towards computersis its length. The longer an instrument, the more likely participant fatigue could
impact the results (cf. Kerlinger, 1986). Participant fatigue is a threat to internal
validity as it may lead to response bias, such as providing the same response to all
items (Hinkin, 1995). Further, in survey research participant fatigue associated with
long instruments could lower response rates. Participant fatigue is especially salient
in situations in which researchers are measuring multiple constructs in a single study
(e.g. attitudes towards computers and UIS, as well as a variety of individual and
organizational variables) as is frequently necessary in IS research. As such, longerinstruments may be poor choices in such settings. While there is no rule regarding
when an instrument should be considered ‘‘long,’’ we selected 20 items as a cut-off.
Four–six items per construct has been recommended as a goal for construct devel-
opment (Hinkin, 1995). Further, adding items beyond 19 does little to improve in-
ternal consistency (Cortina, 1993). Therefore, with 20 items a scale is more than three
times Hinkin’s (1995) recommendation and beyond what should be required to reach
reasonable levels of internal consistency.
We classified eight instruments as long (entries: 1, 10, 11, 14, 15, 23, 24 and 28).Note that we do not include in this group instruments designed to assess multiple
constructs, which are discussed in the following section. While longer instruments
tend to possess higher levels of internal consistency (Cortina, 1993), many of the
shorter attitude towards computers instruments demonstrate levels of internal con-
sistency similar to that of the longer scales. Therefore, in circumstances where one is
concerned with participant fatigue, one of the more parsimonious instruments is a
better choice.
4.2. Instrument complexity
While we classified an instrument as ‘‘long’’ if it contained 20 or more items, in
some cases longer instruments evolved because researchers were assessing multiple
issues, requiring more complex instruments. We identified 16 (over half of the) in-
struments (entries: 3, 6, 7, 8, 9, 12, 13, 16, 17, 18, 19, 21, 26, 27, 29 and 31) that
intentionally were developed to assess multiple constructs. Other instruments that
670 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
have been found to possess multiple latent constructs (such as through exploratory
factor analysis), but were not designed to assess multiple issues distinctly are not
included in this count.
The Attitudes Towards Computers (3) (Zoltan & Chapanis, 1982) contains 64
items to assess six views of computers (sound working machines, negatively toned,
desirable and useful, slave to man, useful and fun and stimulating, and ease of use).The Computer Use Questionnaire (6) (Griswold, 1983) assesses applications of
computers and social implications of computers with 10 items for each construct.
wards computers with respect to ten sectors of society (society, values, cognition,
education, medicine, counseling, mathematics, banking, politics, and criminal jus-
tice). The Computer Attitude Scale (8) (Loyd & Gressard, 1984) has three sub-
components (computer liking, computer confidence, and computer anxiety). The
three sub-scale structure was confirmed by Bandalos and Benson (1990), but not byWoodrow (1991). Hence, it seems important that any future studies utilizing this
instrument assess the latent structure. An additional consideration is that this in-
strument has also been investigated with a fourth component (attitudes towards
academic endeavors associated with computer training) (Loyd & Loyd, 1985) and
this four factor solution confirmed by Nash and Moroz (1997).
The Computer Attitude Scale [9] (Collis, 1984) examines computer related atti-
tudes and self-confidence. The Attitudes Towards Computer Assisted Instruction
(13) (Allen, 1986) contains components to assess comfort, creativity and function.The Attitudes Towards Computer Usage Scale (16) (Popovich, Hyde, & Zakrajsek,
1987) measures attitudes towards computers as well as attitudes towards other
electronic equipment. The Bath County Computer Attitudes Inventory (17) (Bear,
Richards, & Lancaster, 1987) was developed to predict attitudes towards computers
based on other factors, such as computer usage and experience and educational and
career plans. Hence, it includes items to assess some exogenous variables in addition
to attitudes towards computers. The Attitudes Towards Computers Scale (18)
(Rosen, Sears, & Weil, 1987) measures computer attitudes as well as perceptionsconcerning the impact of computers on future job prospects, job creation and pri-
vacy. The Minnesota Computer Literacy and Awareness Assessment Instrument (19)
(Swadener & Hannafin, 1987) includes items to assess computer self-confidence,
computer utility, sex bias in addition to attitude towards computers. The Teacher
Computer Attitude Scale (27) (Huang et al., 1995) used elements of other scales to
assess four components: sex difference, comfort, value and liking. The Attitudes
Towards Technology instrument (29) (Pelton & Pelton, 1997) was structured to
examine attitudes towards technology, the perceptions of the importance of certaintechnology and confidence to use. The Computer Self-Efficacy Scale (31) (Young,
2000) was developed to consider four dimensions: confidence in using computers,
perception of computers as a male domain, perceived usefulness of computers, and
teachers’ attitudes.
Additionally, three instruments were designed to assess cognitive, affective and
behavioral components of attitude distinctly: The Cognitive and Affective Computer
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 671
Measure (21) (Kay, 1989), and Computer Attitude Scale for Secondary Students
(CASS) (26) (Jones & Clarke, 1994).
Although it may be more effective to assess constructs separately, the broader
orientation of these instruments has benefits in research settings that require the
assessment of all or most of the constructs in a particular instrument. If the addi-
tional constructs are not relevant to a particular research setting, then a shorterinstrument would seem to be a better choice to assess attitudes towards computers.
4.3. Instruments with a specialized focus
Much research on attitudes towards computers occurred in educational settings.
We identify eight instruments that target educators (we do not include re-designs of
these instruments in our counts) (entries: 2, 6, 12, 15, 22, 27, 28 and 29). Most of
these instruments have had their latent structure and internal consistency assessed.Their focus makes them especially appropriate in the educational settings for which
they were developed. Examples of the some of the items that focus on educational
issues include ‘‘Computers can be a useful instructional aid in almost all subject
areas.’’ (2); or ‘‘Computer will relieve teachers of routine duties’’ (6).
In addition to the instruments that focus on educators, an additional set focuses
on students. In all 16 instruments, over half of all the instruments we identified,
24, 25, 26 and 31). Examples of the types of items that focus on students include:‘‘Having computers in the classroom would be fun for me.’’ (4), or ‘‘I would rather
have a computer present my instruction than a teacher.’’ (11), and ‘‘Other students
look to me for help when using the computer.’’ (26).
Some of the instruments are quite focused. For instance, the Computer Attitude
towards computers among the elderly; and the Computer Attitudes of Non-
Computing Academics (30) (Seyal, Rahim & Rahman, 1999).
A final instrument to consider in this context is the Cybernetics Attitude Scale (7)
(Wagman, 1983). This instrument examines attitudes towards computers with
672 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
respect to ten sectors of society (society, values, cognition, education, medicine,
counseling, mathematics, banking, politics, and criminal justice). Hence, it appears
that one could use only those items relevant to the particular segment of society
relevant to a study. While the focus on a specific population (such as educators,
students, hospitality managers, the elderly, or non-computing academics) is a
strength in that particular setting, it renders these instruments less appropriate inother circumstances.
4.4. Psychometric issues
When discussing the psychometric properties of these instruments, we focus on
three fundamental properties that provide evidence of reliability: latent structure,
internal consistency and stability. Assessment of these properties is consistent with
Hinkin’s (1995) recommendations for scale development. Latent structure analysisassesses how well the items in an instrument related to the underlying (latent)
construct(s), in this case ‘‘computer attitude,’’ and is conducted via factor analysis.
Confirmatory factor analysis (CFA) is preferred to exploratory factor analysis
(EFA) when prior theory or hypotheses are available (Nunnally & Bernstein, 1994).
Internal consistency analysis investigates the relationship between the items in an
instrument. Internal consistency is usually assessed via Cronbach’s alpha (1951),
which can be conceptualized as the mean of all split-half reliabilities (Cortina,
1993). It is important to assess latent structure prior to internal consistency as it ispossible to identify a single factor even when inter-item correlations are low or to
achieve high inter-item correlations from items related to multiple latent constructs
(Cortina, 1993). Finally, we are interested in an instrument’s stability over time, i.e.
test–retest analysis. Test–retest analysis is conducted by administering an instru-
ment to the same set of subjects at two different time periods then assess the re-
lationship between the two administrations, either through a correlation assessment
or t-test. Given that attitude towards computers is conceptualized as a trait rather
than a state construct, it is reasonable to desire instruments that are capable ofassessing the construct consistently over time. Further, as some researchers would
like to alter participant’s attitudes towards computers based on a manipulation
(such as exposure to a computer system), it seems important to ensure that an
instrument’s assessment of attitude towards computers is stable in the absence of a
manipulation.
When we examine the psychometric analyses conducted on previous instruments
(see Table 1) with respect to latent structure analysis, we find that 65% (20 of 31)
instruments had their latent structure assessed either in the initial or a follow-upstudy. However, all but one instrument’s latent structure was assessed via explor-
atory factor analysis (EFA). EFA is appropriate for those circumstances where there
is not a priori theory to available. In the context of attitude towards computers, a
fairly well established construct, CFA seems appropriate. However, the infrequent
use of CFA may be due to the fact that the structural modeling tools used to conduct
CFA have become wide-spread only relatively recently. It will be interesting to see if
future studies rely more on CFA to examine latent structure.
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 673
Assessment of internal consistency has been conducted on 71% (22 of 31) of the
instruments. Of these, all but one relied upon alpha (Cronbach, 1951), the remaining
instrument reports a split-half reliability (entry: 2). Of those assessed via alpha
(Cronbach, 1951), only one instrument (entry: 22) does not report an alpha that
exceeds the recommended 0.70 threshhold (Nunnally & Berstein, 1994). However,
internal consistency alone is not adequate to assess unidimensionality (Cortina,1993) and in those circumstances where internal consistency has been assessed, but
not latent structure, it seems important to assess the latent structure of these in-
struments in future studies.
Finally, we found that only four instruments’ stability over time has been
assessed (entries: 5, 14, 16 and 26). We found this surprising. As noted above, as
a trait variable, it is reasonable that researchers would desire that instruments to
assess attitude toward computers be stable over time. More surprising was that
we were able to identify four studies (entries: 10, 18, 23 and 28) that used in-struments to assess changes in attitude after a manipulation yet we found no
evidence that the instruments have been established to be stable over time.
Without first establishing that the instruments were stable over time, it is difficult
to conclude that changes in attitude were due to the manipulation rather than
random variance.
Each of the above instruments has strengths, and for research in an educational
setting there are appear to be many instruments with which to examine attitude
towards computers. However, for researchers in non-educational settings, thechoices are less obvious. We were able to identify only one instrument for a general
population for which the latent structure, internal consistency, and stability has been
assessed (entry: 16). However, this instrument contains 20 items. As such, a shorter
instrument that addresses the theoretical areas of interest and demonstrates good
psychometric properties would be useful for researchers. In the remainder of this
paper, we describe the development and reliability assessment of the ATCI, which we
believe meets the above criteria.
5. Development of the ATCI
At the time of the ATCI’s initial development, no single instrument had beencommonly accepted for measuring basic attitudes towards computers. The ATCI
was developed consistent with the theoretical argument that attitude is composed of
three elements: affective, behavioral, and cognitive components (Triandis, 1971). In
creating the items, the authors intended to represent the various areas where research
had occurred on the effect of ‘‘affective or evaluative reactions towards using com-
puters’’ (p. 36). Consistent with these criteria, the instrument included cognitive (cf.
Huber, 1973) and affective items (Lucas, 1975). The behavioral items were subdi-
vided into two components, ease of use (Lucas, 1975) and productivity enhancement(Fuerst & Cheney, 1982), consistent with the growing body of literature at the time
that suggested that these two categories were equally important. Fig. 2 shows the
Fig. 2. Analysis of the attitude towards computer instrument’s theoretical components showing corre-
sponding items.
674 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
relationship between each area and the items that comprise the instrument (see the
Appendix).
The instrument was developing using the semantic differential format for several
measurement-related reasons. First, the semantic differential form ‘‘may be adapted
through choice of concepts and scale to study numerous phenomena’’ (Miller, 1964:
269). Further, research on this format indicates that different types of subjects use
these scales in essentially similar ways (Osgood, Suci, & Tannenbaum, 1957). Fur-
ther, the semantic differential format is especially suited to assessing basic attitudes(Mehling, 1959) and has yielded consistently high reliability and validity scores
across applications (Miller, 1964).
To limit response bias, the items were randomly distributed throughout the ATCI,
and four items were reverse scaled (noted in the Appendix). Reverse scaling was
accomplished by switching the anchors within an item so that a positive response
became a low score (rather than a high score). Reverse scaling decreases the likeli-
hood that a participant will select a response (usually high or low) and give that
answer for every item. When using the instrument, the researcher recodes the scoreassociated with reverse scaled items such that all positive responses are counted as a
high score. While we designed the instrument to include reverse scaling, the use of
reverse scaling has been questioned (Hinkin, 1995). The semantic differential format
allows the reverse scaling to be removed by switching the anchors on those items per
the preferences of the researcher.
Subsequent to its creation, four experienced researchers examined the instrument,
none made recommendations concerning changing the items. The instrument was
then piloted with student subjects. The ATCI’s first use was in research concerningdecision making within a manufacturing information system (Sharfman & Gleeson,
1984). In this study the instrument demonstrated good internal consistency (Cron-
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 675
bach a¼ 0.76) and some preliminary ability to predict choices in the context of a
manufacturing information system. Since the data reflected good internal consis-
tency and some predictive ability, there was no justification to alter items. Subse-
quently, other researchers used the instrument (Burkhardt, 1994; Burkhardt & Brass,
Martocchio, 1990), thereby providing data and motivation for a more comprehen-sive examination of the instrument’s psychometric properties.
6. Reliability analysis
Consistent with our arguments above, we are interested in assessing the latent
structure, internal consistency and stability of the instrument. The combination of
latent structure and internal consistency analysis provides evidence that the instru-ment is assessing a single construct (attitude toward computers) (Cortina, 1993).
Stability analysis via test–retest analysis provides evidence that the instrument is
assessing a stable trait, rather than a temporary state.
6.1. Latent structure analysis
6.1.1. Purpose of the analysis
A first step in assessing the reliability of an instrument is to establish that all itemsin an instrument are measuring the same latent construct. We address this question
with respect to the Attitudes Towards Computers Instrument (ATCI) by conducting
a confirmatory factor analysis to test the hypothesis of unidimensionality. The more
that the items of a scale converge into a single factor, the more confident that one can
be that they measure the same construct. Fig. 3 presents the univariate model for the
instrument that reflects the following specific hypothesis:
H1. A one factor solution fits the items of the ATCI.
6.1.2. Method and analysis
We conducted a CFA using Analysis of Moment Structures (AMOS) software
(Arbuckle, 1997), a structural equations modeling tool. AMOS is in the same class of
analytic tools as LISREL and EQS but provides easier use and more extensive ex-
amination of the fit between models and data. CFA ‘‘involves the specification of one
or more putative models of factor structure, each of which proposes a set of latent
variables (factors) to account for covariance among a set of observed variables’’(Doll, Xia, & Torkzadeh, 1994, p. 453). A sample of 176 juniors and seniors from a
large southwestern university completed the ATCI. The participants were volunteers
in a larger study of computer-based and inventory-management choices (Sharfman
& Gleeson, 1984). The students were volunteers and were given class credit for
participation.
Fig. 3. A univariate model of attitude towards computers.
676 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
6.2. Results
Table 2 presents the standardized regression weights of the observed items on thecomputer attitude construct (analogous to the factors loadings in an exploratory
factor analysis), the estimated standard error of the estimate and the Critical Ratios
(CR). This CR statistic, distributed as ‘‘t’’, tests the hypothesis that the regression
weight is different from zero. Critical Ratios greater than 1.96 demonstrate signifi-
cance at P < 0:05. The coefficient of ATT8 was set at 1.00 in unstandardized form as
Table 2
Standardized regression weights, standard errors and critical ratios of the observed items
Regression weights Standard error Critical ratio
ATT1 0.56 0.35 4.971
ATT2 0.61 0.26 5.170
ATT3 0.73 0.44 5.574
ATT4 0.72 0.37 5.567
ATT5 0.44 0.24 4.319
ATT6 0.70 0.39 5.494
ATT7 0.57 0.42 4.995
ATT8 0.46 NA NA
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 677
the minimum identification requirement for the model (Arbuckle, 1997). As such, no
CR or standard error is calculated for this variable. Note that all CRs exceed 1.96,
and hence are significant at P < 0:05.We calculated several indices to determine the degree to which the hypothesized
model fits the data. The most commonly used index of fit is the v2 statistic. For thisanalysis the v2 is 59.94 with df¼ 20 and P¼ 0.0000 supporting the hypothesis thatthe univariate model is appropriate. We calculated several additional fit indices as
follows: Goodness of Fit Index¼ 0.92, Adjusted Goodness of Fit Index¼ 0.86, In-
cremental Fit Index¼ 0.90 (Bollen, 1989), Comparative Fit Index¼ 0.90 (Bentler,
1990) and v2/df ratio¼ 2.997. All of these statistics indicate good to substantial fit for
the single factor model (Doll et al., 1994), confirming the hypothesis that the uni-
variate model fits the items of the ATCI, thereby supporting the premise that the
ATCI is measuring the single construct of attitude towards computers.
6.3. Internal consistency analysis
6.3.1. Purpose of the analysis
In addition to latent structure analysis, internal consistency analysis is needed to
provide evidence of unidimensionality (Cortina, 1993). Therefore, in this section we
present data relevant to the internal consistency of the instrument to determine if
participants respond consistently across items in the scale. The most commonly
accepted method of assessing internal consistency is provided by Cronbach’s (1951) a(alpha) statistic, with 0.70 considered as sufficient to demonstrate reliability (Nun-
nally & Bernstein, 1994).
6.4. Method and analysis
We present internal consistency statistics from studies conducted by various re-
searchers that have used the ATCI. In total, these studies provide internal consis-
tency analysis data from 349 subjects in three distinct settings (please see theindividual papers for the specifics on each study).
6.5. Results
As one can see from Table 3, the alpha statistics for the ATCI have consistently
exceeded the 0.70 threshold. In fact, over the three samples presented in Table 3, the
average alpha for the instrument was .80. These data provide evidence that the ATCI
demonstrates a high degree of internal consistency. Note that in the past, longerinstruments have been considered superior to short instruments for achieving high
levels of internal consistency. However, recent work in measurement indicates that
shorter instruments can be more reliable than longer instruments as it can be difficult
to develop a large number of items that address the construct space (Embretson &
Hershberger, 1999). Hence, although the ATCI is relatively short, the instrument
achieves good levels of internal consistency because of the items’ tight focus on the
Table 3
Summary of internal consistency analyses
Reported study Population Sample size Alpha
Sharfman and Gleeson (1989) Business students and
professional production
control personnel
128 0.76
Burkhardt and Brass (1990) Federal employees 75 0.84
Webster et al. (1990) Business students 146 0.82
678 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
latent construct. These levels of internal consistency combined with the CFA, pre-
sented earlier, provide evidence that the ATCI is unidimensional.
7. Stability analysis
7.1. Purpose of the analysis
As the ATCI appears to be unidimensional, the next step in reliability analysis is
to assess whether responses to the instrument are stable over time (Kerlinger, 1986).Only after an instrument has been demonstrated to be stable over time, can re-
searchers conclude that changes detected by the instrument are due to actual changes
in attitude (as influenced by manipulations) rather than measurement error. The
length of the retest period is a function of what one is trying to measure. It is critical
to determine the nature of the construct underlying their instrument so one can select
the proper interval between test–retest administrations. If the interval is too short,
test–retest ‘‘reliability exhibits carryover effects. . . when subjects remember their
responses or become uncooperative. When the interval is too long, the actual atti-tudes or situations may change and the responses are no longer comparable’’
(Galletta & Lederer, 1989, p. 424).
Following the procedures of previous studies of test–retest reliability in the IS
literature, we assessed test–retest reliability over both a short and long interval (cf.
Torkzadeh & Doll, 1991). The short interval (2.5 h) allows us to ‘‘detect effects of
such temporary conditions as fluctuations in attention or a �momentary set� (i.e.transient bias) for particular items’’ (Galletta & Lederer, 1989, p. 424). For the long
interval we selected a 4-week time frame. A 2-week interval is considered necessaryto assess long range stability (Nunnally & Bernstein, 1994) and has been used in
previous IS research to examine user information satisfaction (Torkzadeh & Doll,
1991). However, user satisfaction is considered a state rather than trait variable
(Galletta & Lederer, 1989). Therefore, as attitude towards computers is conceptu-
alized as a state variable, we believed that a longer interval was appropriate for the
ATCI and selected 4 weeks for the longer interval. This interval it is long enough to
assure that participants would not remember their responses to the first adminis-
tration of the instrument and any instability in attitude would likely becomeapparent.
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 679
Two separate studies were performed so that the results from the short interval
would not contaminate those from the long interval. For both intervals, the key to a
test–retest analysis is the rejection of a hypothesis of differences based on when the
instrument is administered. Therefore, for both intervals, the analysis tests the fol-
lowing hypothesis:
H2. There is no difference in responses to the ATCI as a function of when the in-
strument is administered.
7.2. Short interval test–retest: methods and analysis
The subjects for this test–retest were freshmen and sophomores in a quantitative
methods course at a Southwestern university. On the administration day, 25 students
were present. At the beginning of the class, the authors administered the instrument toall students. Participation was voluntary, and no class credit was given, but students
used class time to complete the instrument. Participants were told that the data were
being collected as part of an overall study about computer use. The second adminis-
tration was conducted approximately 2.5 h later. Only 19 students also participated in
the second administration of the instrument, six students left class prior to the second
administration. The use of student participants in this type of research is appropriate
because only the psychometric properties of the instrument are of interest, not the
effect of context or experience on behavior (cf. Gordon, Slade, & Schmitt, 1986).A score for each participant was computed by first reverse scaling the indicated
items (see the Appendix), then summing responses and dividing by the numbers of
items to place scores on the same metric as the original items. The Cronbach alpha
for this sample was a¼ 0.91, n¼ 25, for the first administration and a¼ 0.85, n¼ 19,
for the second administration.
We examine test–retest reliability by examining both the t-test for differences
between the administrations and the correlations between the administrations. In the
case of t-tests for differences between the two administrations, a stable instrumentwill demonstrate no differences between the two sets of results. Hence, one wishes to
retain the null hypothesis, a weak hypothesis test. Therefore, to be conservative, we
set a at 0.1. For the correlation analysis, a stable instrument will demonstrate a high
level of correlation between the two administrations. Hence, one wishes to reject the
null hypothesis of no association, a strong hypothesis test. Therefore, to be con-
servative we set a at 0.05. In the case of the t-test (weak hypothesis test) we also
present power analysis for both levels of a.
7.3. Short interval test–retest: results
Table 4 presents the results of the t-test used to examine the application of Hy-
pothesis 2 in the short interval test–retest analysis. The mean difference between the
administrations was 0.2, and the t-test on the difference was not significant
(t¼)1.17, P¼ 0.26, df¼ 18). These results retain the hypothesis that there is no
difference in scores on the ATCI as a function of when it was administered. The
Table 4
Two and one half-hour test–retest interval statistics
Administration Number of cases Standard mean Standard
deviation
Error
Means and standard deviations for the two administrations of the ATCI
First 19 5.49 1.27 0.29
Second 19 5.69 0.89 0.20
Mean difference Standard
deviation
Degrees of
freedom
t-value Significance of
t-value
Results of analysis comparing the two administrations of the ATCI
0.20 0.76 018 )1.17 0.26
680 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
correlation between the administrations of the instrument was significant (r¼ 0.81,
P < 0:005), indicating that we may reject the null hypotheses of no relationship
between the two responses provided in the two administrations.
As the t-test analysis results in a weak hypothesis test (retaining a null hypothesis),
we conducted a power analysis to establish whether the statistical test possessed
adequate power to detect any differences in attitude. As recommended by Cohen(1988), effect sizes were set as small (0.2 of a standard deviation difference between
population means), medium (0.5 of a standard deviation difference between popu-
lation means), and large [0.8 of a standard deviation difference between population
means) and two levels of Type I errors were examined (a¼ 0.05 and 0.1)]. As we
present in Table 5, given a small effect size and a¼ 0.05, and a¼ 0.10, the power of
the t-test was 0.28 and 0.40, respectively. The power for a medium effect size at
a¼ 0.05 was 0.78, just slightly below the 0.8 recommended by Cohen (1988). All
other power levels were greater than 0.995. These power levels are comparable tothose reported by previous IS researchers (Baroudi & Orlikowski, 1989; Torkzadeh
& Doll, 1991). Further, in the context of this study, a small effect size is the difference
attributed to one person in six changing his or her response by a value of one be-
tween the two administrations of the instrument (based on the standard deviation
observed in the test–retest study). Therefore, the amount of change attributed to a
small effect size is negligible and may not be of interest to many researchers. As such,
it is likely that the somewhat lower power associated with a small effect size will not
concern most researchers. The hypothesis test and follow up power analysis provideevidence that subjects’ responses to the ATCI were stable over the short time
interval.
Table 5
Two and one half-hour test–retest interval-power associated with three effect sizes and two levels of Type I
error
Effect size Level of Type I Error
a¼ 0.05 a¼ 0.10
Small (0.2) 0.28 0.40
Medium (0.5) 0.78 0.86
Large (0.8) >0.99 >0.99
Table 6
Four-week test–retest interval statistics
Administration Number of cases Mean Standard
deviation
Standard error
Means and standard deviations for the two administrations of the ATCI
First 70 5.57 0.798 0.095
Second 70 5.46 0.782 0.093
Mean difference Standard
deviation
Degrees of
freedom
t-value Significance of
t-value
Results of analysis comparing the two administrations of the ATCI
0.1071 0.621 69 1.44 0.153
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 681
7.4. Long interval test–retest: method and analysis
In this study, participants completed the instrument on two occasions approxi-
mately 4 weeks apart. This time interval should eliminate potential ‘‘carryover ef-
fects’’ associated with short intervals (i.e. when subjects might remember their
responses) and also situations associated with overly long intervals that may render
the responses no longer comparable (i.e. when the attitudes or situation may change)
(Galetta & Lederer, 1989). For the long interval test–retest study, 112 junior andsenior business students from two Southwestern universities participated. Partici-
pation was voluntary, no class credit was provided. The first administration of the
instrument was during the first week of class and the second occurred approximately
4 weeks later. Participants were told that the data were being collected as part of an
overall study about computer use. During this study, we tested the same hypothesis
and created subjects’ scores in the manner used in the short interval test–retest.
Cronbach’s alpha for the first administration in the 4-week test–retest was a¼ 0.82,
n¼ 106, and for the second administration it was a¼ 0.80, n¼ 76. Consistent withour approach for the short interval test–retest analysis, we set a at 0.1 for the t-test ofdifferences between the two administrations of the instrument and a¼ 0.05 for the
correlation analysis.
7.5. Long interval test–retest: results
Seventy of the 112 subjects completed the instrument on both occasions. Table 6
presents the results of these analyses. The t-test comparing each individual’s responsesat the two points in time was not significant (t¼ 1.44, P¼ 0.153, df¼ 69), therefore we
retain the null hypothesis of no differences between the two administrations of the
ATCI. 1 Further, the correlation between participants’ scores on the two adminis-
trations was significant (r¼ 0.69, P¼ 0.001), therefore we reject the null hypothesis of
no relationship between the responses to the two administrations of the ATCI.
1 A between-groups t-test did not detect a significant difference between the two groups of students
(P¼ 0.893 comparing the two groups on the first administration, P¼ 0.678 on the second administration).
Because a difference was not detected, the responses were combined to provide a single set of data.
Table 7
Four-week test–retest interval-power associated with three effect sizes and two levels of Type I error
Effect Size Level of Type I Error
a¼ 0.05 a¼ 0.10
Small (0.2) 0.56 0.68
Medium (0.5) >0.99 >0.99
Large (0.8) >0.99 >0.99
682 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
The test–retest results presented in Table 6 indicate that respondents’ attitudes
towards computers were consistent over time. Again, the issue of power arises.
Therefore, we conducted a power analysis using the same effect sizes and levels of
Type I error as for the short test–retest interval. Table 7 presents the results of the
power analysis for the long test–retest interval. Given a small effect size and a¼ 0.05
and a¼ 0.10, the power of the t-test was 0.56 and 0.68, respectively. The power levels
for detecting the other effect sizes were greater than 0.99, which exceeded the .8
recommended by Cohen (1988) and compares favorably with those reported byprevious IS researchers (Baroudi & Orlikowski, 1989; Torkzadeh & Doll, 1991).
The test–retest results demonstrate the instrument’s ability to measure attitudes
towards computers reliably over both a short and a relatively long time interval. The
respondents did not show a significant shift in attitudes towards computers over
either time period. Further, adequate statistical power existed to detect all but trivial
changes in scores.
8. Summary of reliability findings
The conclusion from the latent structure, internal consistency and stability
analyses is that the ATCI can be considered highly reliable. We summarize the re-
sults of these analyses in Table 8. From the latent structure analysis conducted via
CFA, it appears that the instrument is assessing a single latent construct. Further,
based on the internal consistency analysis assessed via Cronbach’s alpha (1951),
participants appear to respond to the instrument in a consistent fashion. Finally,participants’ responses to the instrument appear to be stable over both short (2.5 h)
and long (4-week) time intervals. In addition to the evidence described above,
Table 8
Summary of reliability assessment of the attitude towards computer instrument (ATCI)
# of
Items
Target
population
Sample for
reliability
analysis
Item
format
Latent
structure
analysis
Internal
consistency
Test–retest
2.5-h
interval
4-week
interval
8 General
population
Students Semantic
differential
Confirmatory
factor
v2 ¼ 59.94,
df¼ 20,
P¼ 0.0000
a ¼ 0.76–0.91 t¼)1.17,P¼ 0.26;
r¼ 0.81,
P\0:005
t¼ 1.44,
P¼ 0.15;
r¼ 0.69,
P¼ 0.001
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 683
Burkhardt’s (1994) data demonstrates that scores on the ATCI remained stable for
12 months.
9. Discussion
When we contrast the ATCI with previous attitudes towards computers instru-
ments we find that the ATCI possesses qualities that may make it appealing to
other researchers. The ATCI is relatively short (eight items) which limits problems
associated with participant fatigue. Further, the ATCI was designed as a general
assessment of attitude towards computers, rather than for a specific setting as with
over three-fourths (25 of 31) of the other instruments. Whether or not a specific
focus is a strength or a weakness depends upon the research setting. The use of the
semantic differential provides the ability to include reverse scaling to limit responsebias, but allows other researchers to switch anchors if consistent scaling is
preferred.
When we consider the results of the reliability analysis, the ATCI compares fa-
vorably with other instruments. Although latent construct analysis has been con-
ducted on 65% of the instruments, only two (besides the ATCI) relied on CFA. The
reliance on EFA to assess the structure of earlier instruments may be attributed to
the fact that CFA has become easily accessible only since the somewhat recent ad-
vent of Structural Equations Modeling tools such as LISREL and AMOS. However,it would be beneficial to assess the latent structure of these instruments in future
studies that rely on them.
In contrast to relatively infrequent use of CFA, assessment of internal consistency
via Cronbach’s alpha (1951) is quite wide-spread. The alphas reported for the ATCI,
like most other attitude towards computers instruments, exceeds the recommended
0.70 threshold (Nunnally & Berstein, 1993).
We also provided evidence of stability over time through short and long interval
test–retest analysis. The ATCI is one of only five attitude towards computers in-struments for which stability has been assessed. Of the four other instruments, three
are specific to an educational setting. Given the long history and relatively
wide-spread use of this construct, we were surprised that the stability of so few in-
struments had been assessed. The ATCI’s evidence of stability provides future re-
searchers with confidence that changes detected in a study are due to actual changes
in individual’s attitudes rather than random variation or measurement error. It also
provides further evidence that attitude toward computers is a relatively stable trait
rather than a state variable.
10. Future research
Having demonstrated the psychometric properties of the ATCI, it is reasonable
to discuss its possible role in future IS research. As argued earlier, based on the
theory of reasoned action (Fishbein & Ajzen, 1975), a better understanding of
684 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
computer-related attitudes will increase our understanding of computer-related be-
haviors. Several researchers have demonstrated that the attitudes towards computer
construct affects users’ decisions about system acceptance and their overall satis-
faction with computer systems (e.g. Burkhardt, 1994; Sharfman & Gleeson, 1989;
Swanson, 1988). In the future, as information technology becomes more widely
available to users, the ability to measure easily attitudes towards computers with theATCI will allow researchers and practitioners to understand the reasons why some
users embrace information systems and other users remain passive or even resist
using new systems.
For instance, researchers investigating the implementation of new systems may
wish to make use of the ATCI. Because attitude predicts behavior under many
circumstances (cf. Fishbein & Ajzen, 1975), the ATCI may help researchers develop
better theories about who will choose to be information systems adopters as well
who will choose to be non-adopters (e.g. Burkhardt & Brass, 1990). Further, theATCI may even help us understand which individuals are likely to decide to make
greater or lesser use of a system within an ‘‘adopter’’ group.
Practitioners may find it helpful to administer the ATCI prior to implementing a
new system to identify users who might benefit from extra training. If, prior to
implementation, key users were found to have very negative attitudes towards
computers, the implementation leaders could provide extra attention that might
make the difference between a successful and an unsuccessful implementation.
11. Conclusion
We examined existing instruments that assess attitude towards computers and
found that the majority (over three-fourths) were developed for specific settings,
typically educational settings. Further, although most have had some psychometric
characteristics assessed, we were surprised that only four earlier instruments have
had their stability over time examined. Of these four, only one was applicable to ageneral setting. As such, it appears that a parsimonious, reliable instrument appli-
cable to a general setting would be useful to many research settings. Therefore, we
described the development of and assessed the reliability of the ATCI. The unidi-
mensionality of the ATCI was assessed via latent structure (CFA) and internal
consistency (Cronbach’s alpha) analysis. Instrument stability was assessed via two
test–retest studies (short and long intervals). Results indicated that participants’
responses were consistent and reliable over time. These results provide researchers an
instrument to assess attitudes towards computers that is grounded in theory andjustified psychometrically. 2
When we compare the ATCI to other instruments that measure attitude toward
computers we find that it compares favorably in many respects. It is one of only two
instruments designed for a general setting for which latent structure, internal con-
2 We assessed concurrent and predictive validity in separate studies whose results are available in the
form of a working paper that may be obtained from the first author upon request.
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 685
sistency and stability have been assessed. Further, the ATCI is considerably shorter
than the other instruments (8 v. 20 items), which should limit participant fatigue and
response bias. Developing and validating measures such as the ATCI moves the
information systems field one step closer to the goal of a common set of measures
that ‘‘provide a common frame of reference within which to integrate various re-
search streams’’ (Davis et al., 1989, p. 983).
Appendix. Attitudes towards computers instrument
This questionnaire contains eight pairs of adjectives that are used to describe
computers. Please circle the number that best reflects your opinion. Think of com-
puters in general terms and do not dwell on each specific answer.
686 T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689
References
Abdel-Gaid, S., Trueblood, C. R., & Shrigley, R. L. (1986). A systematic procedures for constructing a
valid microcomputer attitude scale. Journal of Research in Science Teaching, 23, 823–839.
Ajzen, I., & Fishbein, M. (1977). Attitude-behavior relationsa theoretical analysis and review of empirical
research. Psychological Bulletin, 84(5), 888–918.
Allen, L. R. (1986). Measuring attitude toward computer assisted instructionthe development of a
semantic differential tool. Computers in Nursing, 4(4), 144–151.
Arbuckle, J. L. (1997). AMOS users’ guide version 3.6. Chicago, IL: SmallWaters Corporation.
Bailey, J. E., & Pearson, S. W. (1983). Development of a tool for measuring and analyzing computer user
satisfaction. Management Science, 29(5), 530–545.
Bandalos, D., & Benson, J. (1990). Testing the factor structure invariance of a comptuer attitude scale over
two grouping conditions. Education & Psychological Measurement, 49–60.
Bannon, S. H., Marshall, J. C., & Fluegal, S. (1985). Cognitive and affective computer attitude scalesa
validity study. Educational and Psychological Measurement, 45, 679–681.
Baroudi, J. J., & Orlikowski, W. J. (1989). The problem of statistical power in MIS research. MIS
Quarterly, 13(1), 87–106.
Bear, G. G., Richards, H. C., & Lancaster, P. (1987). Attitudes towards computersvalidation of a
computer attitude scale. Journal of Educational Computing Research, 3, 207–218.
Belleau, B. D., & Summers, T. A. (1993). Comparison of selected computer attitude scales. Journal of
Consumer Studies and Home Economics, 17, 275–282.
Bentler, P. (1990). Comparative fit indexes in structural models. Psychological Bulletin, 107, 238–246.
Bollen, K. A. (1989). A new incremental fit index for general structural equation models. Sociological
Methods and Research, 17, 303–316.
Burkhardt, M. E. (1994). Social interaction effects following a technological changea longitudinal
investigation. Academy of Management Journal, 37(4), 869–898.
Burkhardt, M. E., & Brass, D. J. (1990). Changing patterns or patterns of changethe effects of a change in
technology on social network structure and power. Administrative Science Quarterly, 35, 104–127.
Coffin, R., & MacIntyre, P. D. (1999). Motivational influences on computer-related affective states.
Computers in Human Behavior, 15, 549–569.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (Rev. ed.). New York: Academic
Press.
Collis, B. A. (1985). Psychosocial implications of sex differences in attitudes towards computersresults of a
survey. International Journal of Women’s Studies, 8, 207–213.
Comber, C., Colley, A., Hargreaves, D. J., & Dorn, L. (1997). The effects of age, gender, and computer
experience upon computer attitudes. Educational Research, 39(2), 123–133.
Cortina, J. M. (1993). What is coefficient alpha? An examination of theory and applications, 78(1), 98–104.
Cronbach, L. (1951). Coefficient alpha and the internal structure of tests. Psychometrica, 16, 297–334.
Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information
technology. MIS Quarterly, 319–340.
Davis, F. D., Bagozzi, R. P., & Warshaw, P. R. (1989). User acceptance of computer technologya
comparison of two theoretical models. Management Science, 15(8), 982–1003.
Dambrot, F. H., Watkins-Malek, M. A., Silling, S. M., Marshall, R. S., & Garver, J. A. (1985). Correlates
of sex differences in attitudes towards and involvement with computers. Journal of Vocational Behavior,
27, 71–86.
DeSanctis, G. (1983). Expectancy theory as an explanation of voluntary use of a decision support system.
Psychological Reports, 52, 247–260.
Doll, W. J., Xia, W., & Tokkzadeh, G. (1994). A confirmatory factor analysis of the end-user computing
satisfaction instrument. MIS Quarterly, 453–461.
Dubois, D., Linek, W., Gentsch, K., & McEneaney, J. (1995). Field-based student Attitudes and the
Integration of Technology. Electronic Publication.
Ellsworth, R., & Bowman, B. E. (1982). A ‘‘beliefs about computers’’ scale based on Ahl’s questionnaire
items. The Computing Teacher, 10, 32–34.
T.M. Shaft et al. / Computers in Human Behavior 20 (2004) 661–689 687
Embretson, S. E., & Hershberger, S. L. (1999). The new rules of measurement: What every psychologist and
educator should know. Mahway, NJ: Lawrence Erlbaum.
Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior. Reading MA: Addison-Wesley.
Francis, L. J. (1993). Measuring attitude toward computers among undergraduate college studentsthe
affective domain. Computers in Education, 20(3), 251–255.
Fuerst, W. L., & Cheney, P. H. (1982). Factors effecting the perceived utilization of computer based
decision support systems. Decision Sciences, 13, 554–569.
Galletta, D. F., & Lederer, A. L. (1989). Some cautions on the measurement of user information
satisfaction. Decision Sciences, 20(3), 419–438.
Gamble, P. R. (1988). Attitudes to computers of managers in the hospital industry. Behavior and
Information Technology, 7(3), 305–321.
Gattiker, U. E., & Hlavka, A. (1992). Computer attitudes and learning performanceissues for management
education and training. Journal of Organizational Behavior, 13, 89–101.
Gordon, M. E., Slade, A. L., & Schmitt, N. (1986). The ‘‘science of the sophomore’’ revisitedfrom
conjecture to empiricism. Academy of Management Review, 11(1), 191–207.
Gressard, C. P., & Loyd, B. H. (1986). Validation studies of a new computer attitude scale. Association for
Educational Data Systems Journal, 18(4), 295–301.
Griswold, P. A. (1983). Some determinants of computer awareness among education majors. AEDS
Journal, 93–103.
Hinkin, T. R. (1995). A review of scale development practices in the study of organizations. Journal of