You are on page 1of 7

This article was downloaded by: [Moskow State Univ Bibliote]

On: 10 February 2014, At: 14:40


Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Journal of Personality Assessment


Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/hjpa20

The Factor Structure of the State-Trait Anxiety


Inventory: An Alternative View
a b
François Vigneau & Stéphanie Cormier
a
College of Psychology , Université de Moncton , Canada
b
Department of Psychology , Université de Montréal , Canada
Published online: 17 Apr 2008.

To cite this article: François Vigneau & Stéphanie Cormier (2008) The Factor Structure of the State-Trait Anxiety Inventory:
An Alternative View, Journal of Personality Assessment, 90:3, 280-285

To link to this article: http://dx.doi.org/10.1080/00223890701885027

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Journal of Personality Assessment, 90(3), 280–285, 2008
Copyright C Taylor & Francis Group, LLC
ISSN: 0022-3891 print / 1532-7752 online
DOI: 10.1080/00223890701885027

The Factor Structure of the State-Trait Anxiety Inventory:


An Alternative View
FRANÇOIS VIGNEAU1 AND STÉPHANIE CORMIER2

1
College of Psychology, Université de Moncton, Canada
2
Department of Psychology, Université de Montréal, Canada

Past studies of the factor structure of the State-Trait Anxiety Inventory (STAI; Spielberger, Gorsuch, Lushene, Vagg, & Jacobs, 1983) have arrived
at various solutions. However, a relatively strong consensus supports a four-factor (State Anxiety Present, State Anxiety Absent, Trait Anxiety
Present, and Trait Anxiety Absent) structure of the scale resulting from the combination of item polarity dimensions and the original two factors
(State and Trait Anxiety). In this article, we assessed the adequacy of an alternative factor model of STAI data, the two-construct, two-method
Downloaded by [Moskow State Univ Bibliote] at 14:40 10 February 2014

model, in three large samples using confirmatory factor analysis. The results revealed a statistical advantage of the two-construct, two-method
model over the one-factor model, the two-construct factor model, and the four-factor model. We discuss possible interpretational advantages of the
two-construct, two-method model of the STAI.

The State-Trait Anxiety Inventory (STAI; Spielberger, Gor- these four factors and, more generally, of their interpretation.
such, Lushene, Vagg, & Jacobs, 1983) is a widely used self- For example, reliability indexes, which are routinely reported
report rating scale designed to measure two dimensions of for the state and trait subscales, are typically not reported for the
anxiety, namely, state anxiety and trait anxiety. State anxiety subscales defined by the four-factor model. Given the limited
has been defined as a transitory feeling of tension and appre- recognition of any heuristic value of the models currently pro-
hension; it may fluctuate over time and can vary in intensity. posed in the literature, we suggest that a new model of the STAI’s
In contrast, trait anxiety denotes relatively stable individual dif- factor structure should be explored. Such a model would need to
ferences in anxiety proneness and refers to a general tendency explicitly account for the complexity of the STAI’s factor struc-
to respond with anxiety to perceived threats in the environment ture while at the same time offering a coherent interpretation of
(Spielberger et al., 1983). STAI scores.
Studies of the dimensionality of the original version of the There has been a general recognition in the measurement
STAI (Form X) have initially reported two factors: a state fac- literature that endorsing an anxiety-present item is not equiva-
tor and a trait factor, in general agreement with the theoreti- lent to not endorsing an anxiety-absent item. In Bernstein and
cal item classification in the two subscales and also consistent Eveland’s (1982) words, “asserting the existence of anxiety’s
with a view of state and trait anxiety as unidimensional, bipolar symptoms is not the same as denying favorable mood” (pp.
constructs (Gaudry, Vagg, & Spielberger, 1975). However, this 370–371). However, such distinctions seem to have come as
simple state-trait factor structure has been questioned, even by an afterthought to the developers of the STAI. In fact, the items
Spielberger et al., 1983, authors of the revised scale, Form Y), from the STAI reflect the very common practice among scale
who reported results indicating that each of the two basic dimen- developers of using a mix of opposite polarity items to create
sions could be further divided on the basis of whether the items what is called a balanced scale. Item polarity is usually con-
were keyed in the direction of the presence or absence of anxi- ceptualized as the “direction” of a rating-scale item relative to
ety (Spielberger, Vagg, Barker, Donham, & Westberry, 1980; see the construct being measured. According to this description, an
also Ramanaiah, Franzen, & Schill, 1983). The resulting four- item is said to be positive when it is keyed in the same direction
factor model (State Anxiety Present, State Anxiety Absent, Trait as the scale’s total score. Conversely, a negative item is one
Anxiety Present, Trait Anxiety Absent; see Figure 1), consis- that is keyed in the opposite direction and should have its score
tent with a view of state and trait anxiety as bidimensional, reflected (e.g., with a 4-point rating scale ranging from 1 to 4,
unipolar constructs, has received empirical support from a num- converting 1 to 4, 2 to 3, 3 to 2, and 4 to 1) before being included
ber of studies (Bernstein & Eveland, 1982; Vagg, Spielberger, in the scale’s total score.
& O’Hearn, 1980) and has been reported in studies using trans- Although the use of items of opposite polarity is often re-
lated versions of the STAI (Gauthier & Bouchard, 1993; Shek, garded as a standard test construction feature designed as a
1988; Suzuki, Tsukamoto, & Abe, 2000). guard against “acquiescence,” a form of response bias (Nun-
Although the four-factor model probably has been the most nally, 1978), it has been criticized as being both unnecessary
widespread account of the factor structure of the STAI at present, and undermining the validity of the scale (see, e.g., Benson
few confirmations have been offered in terms of the validity of & Hocevar, 1985; Pilotte & Gable, 1990; Schriesheim & Hill,
1981). Also, this strategy has been suspected of having an impact
on the factor structure of various scales. Benson and Hocevar
Received November 8, 2006; Revised May 17, 2007. (1985), using an attitude scale in a sample of elementary school
Address correspondence to François Vigneau, École de psychologie, Uni-
children and a randomized design, showed that the factor anal-
versité de Moncton, Moncton, New Brunswick, Canada E1A 3E9; Email:
f.vigneau@umoncton.ca
ysis of a balanced scale of opposite polarity items designed to
280
FACTOR STRUCTURE OF THE STAI 281

1 21 fined by the contributing item’s polarities. The method factors


2 23 are kept orthogonal to one another and to the construct factor.1
One of the main advantages of the one-construct, two-method
5 26
model for theoretically unidimensional constructs is that by con-
8 27
ceptually distinguishing construct dimensions from method di-
10
SN TN
30 mensions, it can offer an adequate statistical account of a scale’s
11 33 items’ covariances, including polarity effects, without necessar-
ily jeopardizing the traditional unidimensional interpretation of
15 34
the construct factor. Pragmatically speaking, this means that as
16 36
long as the items’ loadings on the construct factor are sufficiently
19 39 high, nothing opposes to the calculation of a (unidimensional)
20 22 construct score. This is so because the variance accounted for
24
in the model by the method factors is kept orthogonal relative
3
to the construct. By the same token, the presence of individual
25
4 differences factors associated with item polarity can be esti-
6 28 mated (instead of being blurred with construct variance in some
7 29 unknown proportions or simply ignored).
Downloaded by [Moskow State Univ Bibliote] at 14:40 10 February 2014

31
The frequently suggested four-factor model of the STAI can
9
SP TP
be interpreted as the result of the combined effect of a content-
12 32
based factor structure (state-trait distinction) superimposed on a
13 35 polarity-based factor structure (Suzuki et al., 2000). If this is the
14 37 case, a construct-method model may prove to be an appropriate
17 38
description of the factor structure of the STAI. However, to
be applied to the entire STAI scale, the construct-method model
18 40
has to be expanded to accommodate the two definitional content
constructs at the origin of the measure, state and trait anxiety.
FIGURE 1.—Four-factor model of the State-Trait Anxiety Inventory with stan- The model then becomes a two-construct, two-method model
dardized parameter estimates (Sample 1). SP = State Anxiety, Positive Items
(see Figure 2), and the construct factors are kept independent
factor; SN = State Anxiety, Negative Items factor; TP = Trait Anxiety, Positive
Items factor; TN = Trait Anxiety, Negative Items factor.
from the method factors, which are in turn kept orthogonal to
one another.
In this study, we assessed the factor structure of the STAI
assess one dimension indicated factors based on item polarity, using confirmatory factor analysis. We tested the proposed two-
not content. Matschinger and Angermeyer (1992) and Pilotte construct, two-method model against three other factor models
and Gable (1990) have provided further evidence of the impact of the STAI: the one-factor model, the two-factor (state, trait)
of item polarity on factor structure of scales used with samples model, and the four-factor model. We conducted the analyses
of adults and high school students. Such findings suggest that a separately on three large samples of published and new data:
polarity-based, two-factor model often fit the data significantly 500 English-speaking university students, 888 French-speaking
better than a one-factor model. adults in France, and 625 French-Canadian university students.
We suggest that a construct-method model—that is, a model METHOD
in which construct and method factors are distinguished in the
tradition of the multitrait-multimethod model (Campbell & Fisk, Samples
1959; see Eid, Lischetzke, Nussbeck, & Trierweiler, 2003)— Sample 1. The first sample consisted of 510 undergraduate
may offer an adequate account of the factor structure of STAI students (345 women and 159 men) from an introductory psy-
data. In particular, a one-construct, two-method model is cur- chology course at the University of Toronto at Scarborough.2
rently emerging as a promising model for a variety of scales Participants with incomplete data were excluded, resulting in
thought to be unidimensional but comprised of a mix of positive a final sample of 500 (338 women, 156 men; m age = 20.07
and negative polarity items. This construct-method model has years, SD = 3.23; 6 participants did not report biographical
already demonstrated its superiority in terms of fit, relative to information).
more simple models, when applied to a variety of scales (Bors,
Vigneau, & Lalande, 2006; French, France, Vigneau, French, Sample 2. The data of the second sample were collected by
& Evans, 2007; Gruman & Bors, 2005; Vigneau, Bors, & Dau- Vautier and Jmel (2003) and also used by Vautier (2004).3 As
vier, 2006; Vigneau & Lalande, 2006). The one-construct, two
method model can be conceived as the combination, within a 1 Such orthogonality assumptions are made for three reasons: They simplify
single model, of a one-factor model and a polarity-based, two-
variance decomposition, they facilitate model identification in confirmatory
factor model. Like a one-factor model, the one-construct, two-
analyses, and they ensure that factor interpretation in terms of construct and
method model assumes that all the scale’s items load on one method dimensions is indeed possible (see Marsh, 1989).
latent construct factor that represents the substantial factor pur- 2 We are grateful to D. A. Bors who graciously provided us with the until
ported to be measured by the scale. Similar to a polarity-based, now unpublished Sample 1 data.
two-factor model, it assumes that each of the scale’s items also 3 We thank S. Vautier for making the Sample 2 data available for reanalysis

load on one of two method factors, with these factors being de- through the web (http://www.univ-tlse2.fr/cerpp/annuaire/vautier/).
282 VIGNEAU AND CORMIER

1 21 STAI (STAI–Y; Spielberger et al., 1983) was administered


2 23
to Sample 1, whereas a French version developed in France
(Bruchon-Schweitzer & Paulhan, 1993) was administered to
5 26
Sample 2; and the French-Canadian translation (Gauthier &
8 27 Bouchard, 1993) was used in Sample 3. The STAI was admin-
10 30 istered in class (Samples 1 and 3) and individually (Sample 2)
11 N 33 using the standard instructions.
15 34

16 36

19 39
RESULTS
20 22 We conducted all descriptive statistics and confirmatory fac-
STATE TRAIT
3 24 tor analyses reported in this article using SAS (Version 9.1;
4 25 SAS Institute, 2004). After reflecting scores to negative items,
6 28 we obtained descriptive statistics for the State and the Trait Anx-
7 29 iety scales for each sample. These descriptive statistics, along
9 31
with the Cronbach’s alpha reliabilities for each subscale, are
P
presented in Table 1.
12 32
We analyzed four measurement models with confirmatory
Downloaded by [Moskow State Univ Bibliote] at 14:40 10 February 2014

13 35 factor analysis in each of the three samples: a one-factor model,


14 37
a two-factor model (State Anxiety and Trait Anxiety, corre-
lated), a four-factor model (State Anxiety Present, State Anxiety
17 38
Absent, Trait Anxiety Present, and Trait Anxiety Absent, cor-
18 40 related; see Figure 1), and a two-construct, two-method model
(see Figure 2). We estimated all factor models using the max-
imum likelihood method. Covariance matrices were used as
input data. Five goodness-of-fit indexes were provided for each
model. In addition to the χ 2 and its associated degrees of free-
FIGURE 2.—Two-construct, two-method model of the State-Trait Anxiety Inven- dom, the goodness-of-fit index (Jöreskog & Sörbom, 1996), the
tory with standardized parameter estimates (Sample 1). State = State Content comparative fit index (Bentler & Wu, 1995), the Akaike infor-
factor; Trait = Trait Content factor; P = Positive Polarity method factor; N = mation criterion (Akaike, 1987), and the root mean square error
Negative Polarity method factor.
of approximation (Browne & Cudeck, 1993) are reported.
The goodness-of-fit indexes of the confirmatory factor anal-
reported by Vautier and Jmel (2003), a total of 455 women with a yses are presented in Table 2 for each of the four models ap-
mean age of 44.93 (SD = 8.93) and 451 men with a mean age of plied to each of the three samples. These results indicate that
43.86 (SD = 9.57) were recruited by undergraduate psychology in each of the three samples, the correlated four-factor model
students. Out of these 906 participants, 18 had missing data, provided a better fit to the data than the one-factor model or
which resulted in a final sample of 888 participants (446 women, the two-factor model based on the state versus trait distinction.
442 men). Particularly interesting was the pattern of correlations among
these four factors (see Table 3), the polarity factors within a
Sample 3. For the third sample, 756 undergraduate students construct systematically exhibiting (with one minor exception)
(457 women and 276 men) were recruited from various intro- the highest correlations, followed by correlations between con-
ductory classes at the Université de Moncton. Due to incomplete structs measured using items of the same polarity, and finally by
STAI data, 131 participants were removed, resulting in a total correlations between distinct constructs measured with opposite
sample of 625 (377 women, 229 men; mean age = 20.2, SD = polarity items.
3.2; 19 participants did not report biographical information). The two-construct, two-method model also offered a rela-
tively good fit to the data (see Table 2). Despite the fact that this
Measures model was more constrained than the four-factor model, its fit
The STAI is a self-report inventory comprised of two sub- to the data was marginally superior in each of the three samples.
scales of 20 items assessing state and trait anxiety. Each item in
the State Anxiety subscale is rated on a 4-point intensity scale TABLE 1.—Descriptive statistics and alpha reliabilities of the State and Trait
ranging from 1 (Not at all) to 4 (Very much so). Ten of the state Anxiety subscales.
items are negatively keyed relative to the scale total score (Items
1, 2, 5, 8, 10, 11, 15, 16, 19, and 20); the remaining 10 items are Subscale Sample N M SD Range Skewness Kurtosis α
positively keyed. The Trait Anxiety subscale items are rated on
State Anxiety
a 4-point frequency scale ranging from 1 (Almost never) to 4 1 500 48.23 11.98 20−80 0.01 −0.59 0.937
(Almost always). Nine items are negatively keyed (Items 21, 23, 2 888 36.31 11.57 20–78 0.95 0.61 0.933
26, 27, 30, 33, 34, 36, and 39), whereas the remaining 11 items 3 625 38.84 9.22 20–74 0.48 0.24 0.883
are positively keyed. Participants are instructed to complete the Trait Anxiety
state form, immediately followed by the trait form. 1 500 43.05 10.02 20–76 0.33 −0.17 0.912
2 888 41.59 9.51 20–75 0.44 0.05 0.898
A different version of the STAI was used in each of the 3 625 40.79 9.13 20–77 0.51 0.33 0.904
samples studied. The original English version of the revised
FACTOR STRUCTURE OF THE STAI 283

TABLE 2.—Goodness-of-fit measures for the four measurement models tested in the three samples.

Model χ 2 (df) GFI CFI AIC RMSEA (90% CI)

Sample 1a
1-factor 5008 (740) .484 .588 3528 .108 (.105–.110)
2-factor (correlated; r = .57) 3383 (739) .623 .745 1905 .085 (.082–.088)
4-factor (correlated .36 to .71) 2020 (734) .808 .876 552 .059 (.056–.062)
construct method (r = .50) 1897 (699) .820 .884 499 .059 (.055–.062)
Sample 2b
1-factor 5761 (740) .639 .687 4281 .088 (.085–.090)
2-factor (correlated; r = .70) 4011 (739) .761 .796 2533 .071 (.069–.073)
4-factor (correlated .52 to .89) 3268 (734) .809 .842 1800 .062 (.060–.065)
construct method (r = .65) 2762 (699) .849 .871 1364 .058 (.055–.060)
Sample 3c
1-factor 3938 (740) .681 .670 2458 .083 (.087–.086)
2-factor (correlated; r = .76) 3322 (739) .725 .733 1844 .075 (.072–.077)
4-factor (correlated .55 to .76) 2424 (734) .815 .826 956 .061 (.058–.063)
construct method (r = .75) 2233 (699) .826 .842 835 .059 (.057–.062)

Note. GFI = goodness-of-fit index; CFI = comparative fit index; AIC = Akaike information criterion; RMSEA = root mean square error of approximation; 90% CI = 90% confidence
Downloaded by [Moskow State Univ Bibliote] at 14:40 10 February 2014

interval for RMSEA.


a
n = 500. b n = 888. c n = 625.

We note that the estimated correlation parameters between the against three other factor models of the STAI: the one-factor
state and the trait factors in these models (0.50, 0.65, and 0.75 model, the two construct (state, trait) factor model, and the four-
in Sample 1, Sample 2, and Sample 3, respectively) were in factor model. The analyses indicate that neither a one-factor
close agreement with the values of the correlations between the model nor a content-based two-factor model accounted for the
observed state and trait subscale scores (0.54, 0.63, and 0.70, data adequately. The correlated version of the four-factor model
respectively). repeatedly demonstrated a better fit to the data than those two
In terms of the amount of variance accounted for by the mod- models. One could also argue that the four-factor (correlated)
els, the one-factor model accounted for 28% (Sample 3) to 31% model represented a nearly adequate fit to the data. The proposed
(Samples 1 and 2) of the variance—average item variance— two-construct, two-method model also repeatedly demonstrated
in each of the three samples. The distinction between the two a nearly adequate fit to the data. In fact, the two-construct, two-
constructs, state and trait, as operationalized in the two-factor method model emerged, within each of the samples used, as the
(correlated) model, explained an additional 3% (Sample 3) to best fitting of the four models tested.
8% (Sample 1) of item variance. Finally, adding to the cor- The fit of the commonly suggested four-factor model was,
related two-factor structure two orthogonal method (polarity) in some cases, only marginally inferior to that of the construct-
factors (i.e., the two-construct, two-method model) increased method model. However, fit is not the only element to take
the amount of item variance explained by another 5 to 9 per- into account when trying to determine the factor structure of an
centage points relative to the two-factor (correlated) model. The instrument. In the case of the STAI, we argue that the interpre-
total amount of item variance explained by the two-construct, tation of the four-factor model is not transparent. In particular,
two-method model varied from 39% (in Sample 2) to 47% (in in light of the frequently high interfactor correlations (reaching
Sample 1). The amount of item variance explained by the four- in one instance .89), the empirical value of some of these fac-
factor model was similar, varying from 37% (in Sample 3) to tors is questionable. In any case, under such conditions of high
47% (in Sample 1). interfactor correlation, conclusive external validation evidence
would be required to support the uniqueness of each of the four
DISCUSSION factors. Because they would be based on fewer items, the relia-
The purpose of this study was to evaluate the factor struc- bilities of scores based on the four-factor model would also be
ture of the STAI by testing a two-construct, two-method model expected to be lower than the reliabilities of the traditional state
and trait subscales.
Although only a slightly more adequate representation of
TABLE 3.—Factor intercorrelations in the four-factor model for the three
the STAI factor structure statistically, the two-construct, two-
samples.
method model may offer some interpretational advantages. In
Correlation Sample 1 Sample 2 Sample 3 particular, the two-construct, two-method model remains com-
patible with a unidimensional, bipolar view of state and trait
SP–SN .71 .89 .70 anxiety as well as with a unidimensional use of each of the
TP–TN .67 .75 .74 two corresponding STAI subscales. In fact, this model also
SN–TN .57 .69 .76
SP–TP .56 .63 .71
provides an opportunity to validate the results by comparing
SN–TP .41 .62 .57 the parameter estimate for the state-trait factor correlation with
SP–TN .36 .52 .55 the empirical correlation between the state and trait subscale
scores. A second advantage is that contrary to the four-factor
Note. SP = State Anxiety, Positive items factor; SN = State Anxiety, Negative items
factor; TP = Trait Anxiety, Positive items factor; TN = Trait Anxiety, Negative items
model, the construct-method model provides a means of esti-
factor. mating the size of the polarity effect because it treats construct
284 VIGNEAU AND CORMIER

and method variance separately. In the samples we examined, Bors, D. A., Vigneau, F., & Lalande, F. (2006). Measuring the need for cognition:
we estimated that two independent polarity-based method fac- Item polarity, dimensionality, and the relation with ability. Personality and
tors accounted for up to 9% of item variance. This amount of Individual Differences, 40, 819–828.
systematic variance is relatively substantial given that taking Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit.
In K. J. Bollen & J. S. Long (Eds.), Testing structural equation models (pp.
into account construct distinctions (state, trait) in the two-factor
136–162). London: Sage.
(correlated) model only added 3% to 8% relative to an undiffer- Bruchon-Schweitzer, M. L., & Paulhan, I. (1993). Le manuel du STAI-Y de C. D.
entiated one-factor model. Spielberger, adaptation française [Manual of C. D. Spielberger’s STAI-Y,
The results obtained in this research show remarkable stabil- French adaptation]. Paris: Éditions du Centre de Psychologie Appliquée.
ity across the three samples of participants used. In fact, the rank Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation
ordering of the four models tested in terms of their fit to the data by multitrait-multimethod matrix. Psychological Bulletin, 56, 81–105.
was virtually identical from one sample to the next, this despite Cronbach, L. J. (1946). Response sets and test validity. Educational and Psy-
the fact that the participants came from different cultures and chological Measurement, 6, 475–494.
were administered distinct versions of the instrument. This abil- Edwards, A. L. (1957). The social desirability variable in personality research.
ity to replicate findings across samples and contexts constitutes New York: Dryden.
a strong argument in support of the construct-method model. Eid, M., Lischetzke, T., Nussbeck, F. W., & Trierweiler, L. I. (2003). Separat-
ing trait effects from trait-specific method effects in multitrait-multimethod
However, because two out of three samples were homogeneous
models: A multiple-indicator CT-C(M-1) model. Psychological Methods, 8,
regarding age and student role, more research is needed to assess 38–60.
Downloaded by [Moskow State Univ Bibliote] at 14:40 10 February 2014

the generalizability of our findings. In particular, the validity of French, D. J., France, C. R., Vigneau, F., French, J. A., & Evans, R. T. (2007).
the two-construct, two-method model should be evaluated with Fear of movement/(re)injury in chronic pain: A psychometric assessment of
other samples from the general population. It would also be the original English version of the Tampa scale for kinesiophobia (TSK).
of particular interest to examine the model in various clinical Pain, 127, 42–51.
populations. Furnham, A. (1986). Response bias, social desirability and dissimulation. Per-
The construct-method model explicitly acknowledges the ex- sonality and Individual Differences, 7, 385–400.
istence of systematic individual differences associated with item Gaudry, E. G., Vagg, P., & Spielberger, C. D. (1975). Validation of the state-
trait distinction in anxiety research. Multivariate Behavioral Research, 10,
polarity. It does not, however, offer an explanation of such po-
331–341.
larity effects. Several psychological variables may constitute Gauthier, J., & Bouchard, S. (1993). Adaptation canadienne-française de la
promising targets for research aimed at explaining psycholog- forme révisée du State-Trait Anxiety Inventory de Spielberger [A French-
ically what have been called in this article “method factors.” Canadian adaptation of the revised version of Spielberger’s State-Trait Anx-
Response style variables in general (see Cronbach, 1946), ac- iety Inventory]. Canadian Journal of Behavioural Science, 25, 559–578.
quiescence (Bentler, Jackson, & Messick, 1971) and social de- Gruman, J. A., & Bors, D. A. (2005, July). Two tests of tolerance for ambi-
sirability (Edwards, 1957; Furnham, 1986) in particular, would guity: A psychometric study. Paper presented at the 15th annual meeting of
appear as good starting candidates to explore the nature of the the Canadian Society for Brain, Behaviour and Cognitive Science (BBCS),
polarity factors. The systematic investigation of the individual Université de Montréal, Montréal, Quebec, Canada.
differences factors associated with item polarity appears neces- Jöreskog, K. G., & Sörbom, D. (1996). LISREL 8: User’s reference guide.
Chicago: Scientific Software International.
sary. This task is beyond the scope of this research. However,
Marsh, H. W. (1989). Confirmatory factor analyses of multitrait-multimethod
the proposed two-construct, two-method model of the factor data: Many problems and a few solutions. Applied Psychological Measure-
structure of the STAI, by explicitly recognizing such effects, is ment, 13, 335–361.
a first step in that direction. Matschinger, H., & Angermeyer, M. C. (1992). Effekte der Itempolung auf
das Antwortverhalten [The effect of item polarity on response behav-
ior]. Zeitschrift für Differentielle und Diagnostische Psychologie, 13, 97–
ACKNOWLEDGMENTS 110.
This research was supported by grants from the Natural Sci- Nunnally, J. C. (1978). Psychometric theory (2nd ed.). New York: McGraw-Hill.
ences and Engineering Research Council of Canada (NSERC) Pilotte, W., & Gable, R. K. (1990). The impact of positive and negative item
and from the Social Sciences and Humanities Research Council stems on the validity of a computer anxiety scale. Educational and Psycho-
logical Measurement, 50, 603–610.
of Canada (SSHRC). We thank Nadine LeBlanc for her contri-
Ramanaiah, N. V., Franzen, M., & Schill, T. (1983). A psychometric study
bution to the data collection at the Université de Moncton and of the State-Trait Anxiety Inventory. Journal of Personality Assessment, 47,
for her assistance in the preparation of this article. 531–535.
SAS Institute. (2004). SAS/STAT user’s guide (Version 9.1) [Computer software
manual]. Cary, NC: Author.
REFERENCES Schriesheim, C. A., & Hill, H. D. (1981). Controlling acquiescence response
Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52, 317–332. bias by item reversals: The effect on questionnaire validity. Educational and
Benson, J., & Hocevar, D. (1985). The impact of item phrasing on the validity Psychological Measurement, 41, 1101–1114.
of attitude scales for elementary school children. Journal of Educational Shek, D. T. (1988). Reliability and factorial structure of the Chinese version of
Measurement, 22, 231–240. the State-Trait Anxiety Inventory. Journal of Psychopathology and Behav-
Bentler, P. M., Jackson, D. N., & Messick, S. (1971). Identification of content ioral Assessment, 10, 303–317.
and style: A two-dimensional interpretation of acquiescence. Psychological Spielberger, C. D., Gorsuch, R. L., Lushene, R., Vagg, P. R., & Jacobs, G. A.
Bulletin, 76, 186–204. (1983). Manual for the State-Trait Anxiety Inventory (Form Y). Palo Alto,
Bentler, P. M., & Wu, E. J. C. (1995). EQS for Windows: User’s guide. Encino, CA: Consulting Psychologists Press.
CA: Mulitvariate Software, Inc. Spielberger, C. D., Vagg, P. R., Barker, L. R., Donham, G. W., & Westberry, L.
Bernstein, I. H., & Eveland, D. C. (1982). State vs. trait anxiety: A case study G. (1980). The factor structure of the State-Trait Anxiety Inventory. In I. G.
in confirmatory factor analysis. Personality and Individual Differences, 3, Sarason & C. D. Spielberger (Eds.), Stress and anxiety (Vol. 7; pp. 95–109).
361–372. Washington, DC: Hemisphere.
FACTOR STRUCTURE OF THE STAI 285

Suzuki, T., Tsukamoto, K., & Abe, K. (2000). Characteristic factor structures Vigneau, F., Bors, D. A., & Dauvier, B. (2006, September). Besoin de cogni-
of the Japanese version of the State-Trait Anxiety Inventory: Coexistence tion et tolérance à l’ambiguı̈té : généralisation d’un modèle trait-méthode
of positive-negative and state-trait factor structures. Journal of Personality à deux questionnaires de personnalité [Need for cognition and tolerance of
Assessment, 74, 447–458. ambiguity: Generalizing one trait-method model of two personality question-
Vagg, P. R., Spielberger, C. D., & O’Hearn, T. P., Jr. (1980). Is the State-Trait naires]. Paper presented at the 17th Journées internationales de psychologie
Anxiety Inventory multidimensional? Personality and Individual Differences, différentielle, Université Paris X Nanterre, Paris, France.
1, 207–214. Vigneau, F., & Lalande, F. (2006). Multidimensionnalité de l’Échelle de be-
Vautier, S. (2004). A longitudinal SEM approach to STAI data: Two compre- soin de cognition—Note sur l’utilisation d’items négatifs dans un ques-
hensive multitrait-multistate models. Journal of Personality Assessment, 83, tionnaire de personnalité [Multidimensionality of the Need for Cognition
167–179. scale: A note on using negative items in a personality questionnaire]. In
Vautier, S., & Jmel, S. (2003). Transient error or specificity? An alternative C. Houssemand, R. Martin, & P. Dickes (Eds.), Perspectives de psycholo-
to the staggered equivalent split-half procedure. Psychological Methods, 8, gie différentielle (pp. 267–272). Rennes, France: Presses Universitaires de
225–238. Rennes.
Downloaded by [Moskow State Univ Bibliote] at 14:40 10 February 2014

You might also like