Journal of Sports Sciences, 1999, 17, 505± 512

ConWrmatory factor analysis of the Competitive State

Anxiety Inventory-2
1 2 1 2
AN DREW M . LAN E, * DAVID F. SEW ELL, PETER C. TERRY, DAVID BARTRAM
3
and M ARK S. N ESTI
1
Departm ent of Sport Sciences, B r unel U niversity, O ster ley Cam pus, B orough Road, Isleworth, M iddlesex TW 7 5DU ,
2 3
Departm ent of Psych olog y, U niversity of H ull, Cottingh am R oad, H ull H U 6 7RX and School of Leisure and Sport
Studies, Leeds M etropolitan U niversity, B eckett Park, Leeds L S6 3QS, U K

Accepted 17 N ovem ber 1998

The aim of this study was to evaluate the factor structure of the Com petitive State Anxiety Inventory-2 (CSAI-2)
using con W rm atory factor analysis. Volunteer participants (n = 1213) completed the CSAI-2 approxim ately 1 h
before competition and the data were analysed in two samples. The hypothesized model showed poor W t
indices in both samples independently (Robust C omparative Fit Index: sample A = 0.82, sample B = 0.84) and
simultaneously (Com parative Fit Index = 0.83), suggesting that the factor structure proposed by M artens et al. is
X awed. Our W ndings suggest that a limitation of the Cognitive Anxiety scale derives from phrasing items around
the word `concerned’ rather than `worried’ . We suggest that being concerned about an impending performance
does not necessarily m ean that an athlete is experiencing negative thoughts, but that the athlete is acknowledging
the importance and diY culty of the challenge and is attempting to mobilize resources to cope. The present
results question the use of the CSAI-2 as a valid measure of competitive state anxiety.

K eywords : anxiety, C ompetitive State Anxiety Inventory-2, EQS, factorial validity, structural equation
modelling.

Introduction demonstration of the factorial validity of anxiety

Anxiety is one of the m ost frequently investigated
variables in sport psychology (see Jones, 1995; Hardy
et al., 1996). It is usually conceptualized as a m ultidi-
m ensional construct com prising cognitive and som atic
com ponents (M artens et al., 1990b). C ognitive anxiety
is typiW ed by negative self-im ages and self-d oubts, while
som atic anxiety is typiW ed by increased heart rate, tense
m uscles and clam my hands. The Com petitive State
Anxiety Inventory-2 (CSAI-2: M artens et al., 1990a)
has been the m easure of choice for m ost researchers of
com petition anxiety during the past decade. The CSAI-
2 also assesse s self-conW dence, which is characterized
by positive expectations of success. The CSAI-2 has
27 items overall, with nine item s in each of three sub-
scales: C ognitive Anxiety, Som atic Anxiety and Self-
conW dence. G iven the research interest in com petitive
state anxiety and self-conW dence, and the extent to
which tests of theory rely upon valid m easurem ent,
* Author to whom all correspondence should be addressed. e-mail:
andrew.lane@brunel.ac.uk
m easures is im perative.
There are at least three argu m ents to suggest that it
would be prudent to re-evaluate the factor structure of
the C SAI-2. First, the m ethodological rigour app lied
by M artens et al. (1990a) to test factorial validity is
questionable in the light of current knowledge. Vali-
dation of the C SAI-2 involved four exploratory analyses
using principal com ponents analysis with oblique and
varim ax rotations. At each stage, the ratio of participants
to item s was below the m inim um recom m ended (5 : 1)
for trustworthy results (Tabachnick and Fidell, 1996;
T hom pson and D aniel, 1996). Indeed, Tabachnick and
Fidell (1996, p. 640) proposed that, `As a general rule
of thum b, it is com forting to have at least 300 cases
for factor analysis’ . Stage 1 analysed the responses of
162 participants to a 79-item scale (2 : 1 ratio); stage 2
re-analysed data from the sam e participants using a
reduced 36-item scale (4.5 : 1 ratio); stage 3 included
80 participants and a 52-item scale (1.5 : 1 ratio); and
stage 4 used the sam e 80 participants and a 27-item
scale (3 : 1 ratio). M oreover, exploratory factor analy-
sis has been criticized for its inherently atheoretical

0264± 0414/99 Ó 1999 E & FN Spon

506
nature (e.g. T hom pson and D aniel, 1996), which tends
to result in spurious factors, especially when the
participant-to-item ratio is low. F ur ther m ethodological
concerns include the reanalysis of responses from
the sam e data set rather than cross-valid ating to new
sam ples; the collection of anxiety data from som e parti-
cipants based on a hypothetical com petition at stages 1
and 2 (especially surprising given the recom m endation
of M ar tens et al., 1990b, that 1 h before com petition is
the optim um time to assess state anxiety); and the use of
an exclusively undergraduate athletic sam ple, lim iting
the generality of the results to athletes from diV erent
educational backgro unds. Collectively, these m ethodo-
logical lim itations sugge st that cross-va lidation of the
C SAI-2 to new sam ples is desirable.
A second reason for re-evaluating the factor structure
of the C SAI-2 derives from the decision taken at stage 5
of the original validation process to change the word
`worried’ to `concerned’ in the Cognitive Anxiety scale
to reduce the inX uence of social desirab ility. It is
possible that the sem antic diV erence between these
words m ay have threatened the conceptual integrity of
the scale. Also at this stage, M artens et al. argu ed that
low intercorrelations between the three anxiety sub-
com ponents provided suY cient evidence of factorial
validity, although no further factor analysis was con-
ducted on the participant group of 266 athletes (see
M artens et al., 1990a, p. 139).
T he third argu m ent for re-evaluating the C SAI-2
is that recent developm ents in com puter software to
test the factor structures of psych ological question-
naires have prom pted researchers (see H endrick and
H endrick, 1986; Bentler, 1992, 1995; Schutz and
G essaro li, 1993; Thom pson and D aniel, 1996) to
em phasize the beneW ts of structural equation m odel-
ling techniques such as conW rm atory factor analysis.
T his approach has a clear advantage over exploratory
techniques, as data are tested again st a prior m odel and
the W t of the m odel is assessed using m ore stringent cri-
teria. It would app ear incum bent upon contemporary
researchers involved in questionnaire developm ent to
use conW rm atory procedures to establish factorial
validity.
T he aim of the present study was to re-exam ine the
proposed 27-item , three-factor structure of the C SAI-2
using conW rm atory factor analysis techniques.
M ethods
Participants
In total, 1213 volunteer participants (1025 m ales,
188 fem ales) aged 15± 39 years com pleted the CSAI-2.
T he volunteers par ticipated in several sports, including:
Lane et al.
basketball (58 m ales aged 19± 31 years), duathlon (122
m ales and 10 fem ales aged 16± 35 years), hockey
(22 m ales and 11 fem ales aged 18± 24 years), jujitsu
(20 m ales and 18 fem ales aged 18± 24 years), karate (50
m ales and 17 fem ales aged 18± 21 years), rugby league
(193 m ales aged 16± 35 years), rugby union (23 m ales
aged 18± 21 years), soccer (22 m ales and 11 fem ales
aged 18± 21 years), sw im m ing (16 m ales and 15 fem ales
aged 15± 18 years), 10-km running (82 m ales aged
18± 39 years), singles tennis (32 m ales and 61 fem ales
aged 15± 27 years), track and W eld (66 m ales aged 17± 27
years), tae-kwon-do (144 m ales and 34 fem ales aged
18± 27 years) and triathlon (175 m ales and 11 fem ales
aged 17± 39 years).
We suggest that the uneven sex distribution of the
participants is representative of the respective pro-
portions of m ales and fem ales com peting in sport. Al-
though previous research has demonstrated diV erences
between the sexes in the intensity of anxiety responses
(e.g. M artens et al., 1990a), no research has reported
that anxiety is conceptualized diV erently by m ales and
fem ales (Perry and W illiam s, 1998). T he participants in
the present study com peted at diV erent levels of com -
petition, including national (duathlon, singles tennis,
tae-kwon-do and triathlon), club (basketball, jujitsu,
karate, rugby league, 10-km running, track and W eld and
swim m ing) and inter-university com petition (hockey,
rugby union and soccer). F urtherm ore, we suggest
that factorial validity is best demonstrated using large
sam ples that represent the population to which the
W ndings are to be inferred (Tabachnick and Fidell,
1996).
Proposed str ucture of the C SAI-2
It was proposed (M artens et al., 1990a) that the 27
item s of the CSAI-2 describe feelings of cognitive
anxiety, som atic anxiety and self-conWdence. The
nine items in the Cognitive Anxiety scale include eight
that refer to being `concerned’ about a forthcom ing
com petition. Two relate to speciW c outcom es (`I am
concerned about losing’ and `I’ m concerned that others
w ill be disappo inted with my perform ance’ ), W ve to a
self-re ferenced standard (`I am concerned about
perform ing poorly’ , `I’ m concerned I won’ t be able to
concentrate’ , `I’ m concerned about reaching my goal’ ,
`I am concerned that I m ay not do as well as I could’
and `I’ m concerned about choking under pressure’ )
and one assessees general concerns about the com -
petition (`I am concerned about this com petition’ ).
T he rem aining item assesses general doubts (`I have
self-d oubts’ ).
T he nine items in the Som atic Anxiety scale include
two that describe generalized som atic responses (`I feel
nervous’ and `I feel jittery’ ), three that refer to m uscular
M easures of anxiety
tension (`M y body feels tense’ , `M y body feels relaxed’
and `M y body feels tight’ ) and four that describe
som atic responses in speciWc parts of the body (`I feel
tense in the stom ach’ , `M y heart is racing’ , `I feel my
stom ach sinking’ and `M y hands are clam my’ ).
The nine item s in the Self-conW dence scale include
W ve that describe positive exp ectations (`I feel self-
conW dent’ , `I am conW dent I can m eet the challenge’ ,
`I’ m conW dent about perform ing well’ , `I’ m conW dent
because I m entally picture myself reaching my goal’ and
`I’ m conW dent about com ing through under pressure’ )
and four that describe a generalized feeling of calm ness
(`I feel calm ’ , `I feel com fortable’ , `I feel secure’ and `I
feel m entally relaxed’ ).
All item s are rated on a 4-point scale anchored by
1 (`not at all’ ) and 4 (`very m uch so’ ).
Proc edure
T he CSAI-2 was adm inistered to the participants
app roxim ately 1 h before com petition. Before com -
pleting the questionnaire, the `antisocial desirability’
statement of M artens et al. (1990b) was read aloud,
using the response set `H ow are you feeling righ t now?’
Data analyses
ConW rm atory factor analysis using E QS V5 (Bentler
and Wu, 1995) was used to test the three-factor m odel
proposed by M artens et al. (1990a). It has been
suggested that a hypo thesized m odel is exam ined m ore
rigorously by random ly dividing participants into two
sam ples, conducting conW rm atory factor analysis on
one sam ple and then cross-valid ating the results on the
other sam ple (Bynner and Rom ney, 1985). Hence, the
sam ple was split random ly into two sam ples of alm ost
equal size (sam ple A = 606; sam ple B = 607) through
the EQ S V 5 package.
The m odel tested speciW ed that item s were related
to their hyp othesized factor, w ith the varian ce of the
factor W xed at 1, and the three factors were correlated.
As m ultivariate non-norm ality was evident in the data,
the m odel was tested using the Robust M axim um
Likelihood m ethod, w hich has been found to control
for overestim ation of x2 , underestim ation of adjunct W t
indexes and under-identiW cation of errors (see H u and
Bentler, 1995).
Following the recom m endations of H u and Bentler
(1995), several W t indices were used to test the factor
structure. F irst, the x2 statistic was considered. A good
2
W tting m odel tends to produce a non-sign iW cant x
value, although its value is inX ated am ong large
sam ples. Recent research has addressed the issue of how
to interpret a signiW cant x2 am ong large sam ples, with
the ratio of x2 to degrees of freedom being proposed as a
507
superior index. Byrne (1989) suggested that a ratio of 2
or lower indicates an acceptable W t.
Two incremental W t indices were also used. First,
the N on-N orm ed Fit Index (Tucker and L ewis,
1973) assesse s the adequacy of the hypo thesized m odel
in relation to a baseline m odel, taking sam ple size
into account. Secondly, the C om parative F it Index
using the Robust x2 value evaluates the adequacy of
the hypo thesized m odel in relation to the worst (in-
dependent) m odel. If the hypo thesized m odel is not a
signiW cant im provem ent on the independent m odel, the
W t indices w ill be close to zero (Bentler, 1995). Two
absolute indices were also used: the G oodness of F it
Index and Adjusted G oodness of Fit Index, which
indicate the relative am ount of the observed varian ces
and covariances accounted for by the m odel. T he
criterion value asso ciated with an acceptable m odel W t
is 0.90 for all W t indices (Bentler, 1995).
M ultisam ple conW rm atory factor analysis was used
to test the strength of the factor solution across both
sam ples sim ultaneously. In m ultisam ple analysis, it is
assu m ed that data from m ore than one sam ple provide
com parable inform ation about the hypo thesized m odel.
T his assu m ption is tested by analysing data from
diV erent sam ples sim ultaneously to verify whether the
m odel reproduces the data of each sam ple to within
sam pling accuracy (see Bentler, 1992). As with one-
sam ple conW rm atory factor analysis, x2 statistics and
adjunct W t indices represent the extent to which
variance± covariance m atrices from diV erent sam ples
are identical. In m ultisam ple analysis, the Lag range
M ultiplier test assesse s the extent to which the W t of
the m odel would be im proved if equality constraints
were rem oved. Cronbach (1951) alph a coeY cients of
internal consistency were also calculated for each factor.
Tabachnick and Fidell (1996) suggested that C ronbach
alph a coeY cients for an internally consistent scale
should be 0.70 or higher.
Results
T he results of the single-sam ple conW rm atory factor
analyses of the m odel proposed by M artens et al.
(1990a) are shown in Table 1. T he ratio of x2 to degrees
of freedom indicated a questionable W t between the
data and the m odel in both sam ples (sam ple A = 4.07,
sam ple B = 3.88). M ore im portantly, all W t indices were
less than the 0.90 criterion (Robust C om parative F it
Index: sam ple A = 0.82, sam ple B = 0.84) required for
an acceptable Wt (see Bentler, 1995).
The rationale for m ultisam ple conW rm atory factor
analysis in the present study was to test the generality of
the results. As single-sam ple results have dem onstrated
a poor m odel W t, the aim of the m ultisam ple analysis was
508 Lane et al.

Table 1 Con W rmatory factor analysis of the C ompetitive Table 2 M ultisample con W rmatory factor analysis of the
State Anxiety Inventory-2 C ompetitive State Anxiety Inventory-2

Sample A Sample B M ultisample con W rmatory

Fit index (n = 606) ( n = 607) Fit index factor analysis

2
Satorra-Bentler x 1299 1246 x2 2892
Degrees of freedom (d.f.) 321 321 Degrees of freedom (d.f.) 669
Satorra-Bentler x2 /d.f. ratio 4.07 3.88 x2 /d.f. ratio 4.32
NNFI 0.79 0.81 NNFI 0.81
RC FI 0.82 0.84 C FI 0.82
G FI 0.83 0.83 G FI 0.83
AGFI 0.80 0.80 AGFI 0.81

A bbreviations : NFI = Norm ed Fit Index, NNFI = Bentler Bonett Non- A bbreviations : NFI = Norm ed Fit Index, NNFI = Bentler Bonett
norm ed Fit Index, RC FI = Robust Comparative Fit Index, GFI = Non-normed Fit Index, CFI = Comparative Fit Index, G FI = G ood-
G oodness of Fit Index, AGFI = Adjusted G oodness of Fit Index. ness of Fit Index, AGFI = Adjusted G oodness of Fit Index.

Table 3 Standardized factor loadings and error variances of items

Sample A ( n = 606) Sample B (n = 607)

LM T x2
Factor Error Factor Error diV erence
loading variance loading variance test

C ognitive Anxiety
I am concerned about this competition 0.46 0.89 0.46 0.89 0.16
I have self-doubts 0.55 0.83 0.56 0.83 0.95
I am concerned that I may not do as well in this competition as I
could 0.63 0.78 0.64 0.77 0.42
I am concerned about losing 0.54 0.84 0.54 0.84 0.22
I am concerned about choking under pressure 0.52 0.86 0.49 0.87 0.21
I am concerned about performing poorly 0.70 0.72 0.69 0.72 0.16
I’ m concerned about reaching my goal 0.47 0.88 0.47 0.88 0.68
I’ m concerned that others will be disappointed with my perform ance 0.65 0.76 0.65 0.76 0.20
I’ m concerned I won’ t be able to concentrate 0.51 0.87 0.49 0.87 0.04

S om atic A nxiety
I feel nervous 0.58 0.82 0.58 0.81 0.05
I feel jittery 0.62 0.78 0.62 0.79 0.28
M y body feels tense 0.72 0.71 0.68 0.73 0.06
I feel tense in the stomach 0.74 0.68 0.72 0.70 0.01
M y body feels relaxed 0.40 0.92 0.39 0.92 0.18
M y heart is racing 0.70 0.72 0.69 0.73 0.30
I feel my stomach sinking 0.69 0.72 0.70 0.72 0.32
M y hands are clammy 0.55 0.83 0.58 0.82 0.33
M y body feels tight 0.67 0.74 0.68 0.73 0.15

S elf-con W dence
I feel at ease 0.55 0.84 0.52 0.85 0.08
I feel comfortable 0.63 0.78 0.64 0.77 0.30
I feel self-con W dent 0.74 0.68 0.73 0.67 0.28
I feel secure 0.66 0.76 0.63 0.79 0.03
I feel m entally relaxed 0.75 0.66 0.74 0.67 0.01
I am con W dent I can meet the challenge 0.77 0.64 0.75 0.66 0.36
I’ m con W dent about performing well 0.64 0.77 0.64 0.77 0.08
I’ m con W dent because I m entally picture myself reaching my goal 0.63 0.77 0.62 0.78 0.48
I’ m con W dent at coming through under pressure 0.68 0.74 0.68 0.74 0.75

A bbreviation : LM T = Langrange Multiplier test.

M easures of anxiety 509

to exam ine the extent to w hich the parts of the m odel
that were strong and the parts of the m odel that were
weak were consistent across both sam ples. T he m odel
tested the extent to w hich factor loadings were equal in
both sam ples
The results of the m ultisam ple conW rm atory factor
analysis also indicated a poor overall W t (see Table 2).
T he em phasis of m ultisam ple analysis is on the extent
to w hich equality constraints placed on the factor
loadings diV er signiW cantly between sam ples. Standard-
ized correlation coeY cients, error variances and the
results of the x2 test from the m ultisam ple Lag range
M ultiplier test are shown in Table 3. Standardized
factor coeY cients indicated poor relationships between
at least three items and their hypo thesized factor (`I am
concerned about this com petition’ , `I am concerned
about reaching my goal’ and `M y body feels relaxed’ ).
Table 4 C orrelation coe Y cients among the subscales of the
Com petitive State Anxiety Inventory-2
Somatic Anxiety Self-con W dence
Cognitive Anxiety
sample A 0.65* - 0.44*
sample B 0.62* - 0.46*
Somatic Anxiety
sample A - 0.51*
sample B - 0.47*
* P < 0.01.
T hese items demonstrated low factor loadings and high
error variances in one or both sam ples. T he m ultisam ple
Lag range M ultiplier test results indicated that none of
the factor loadings diV ered signiWcantly between the
sam ples. Furtherm ore, the m ultivariate m ultisam ple
Lag range M ultiplier test indicated that diV erences in
item ± factor relationships were not signiW cant. T his casts
substantial doubt upon the inclusion of these three
item s in the C SAI-2, as they do not contribute to their
hyp othesized factor.
Table 4 shows the intercorrelations am ong factors.
T he varian ce shared between C ognitive Anxiety and
Som atic Anxiety scores was 43% for sam ple A and
38% for sam ple B. Self-conW dence and Cognitive
Anxiety shared 19% (sam ple A) and 22% (sam ple B)
com m on varian ce; Self-conW dence and Som atic Anxiety
shared 26% (sam ple A) and 22% (sam ple B) com m on
variance. T he strength and direction of these corre-
lations are consistent with those reported by M artens
et al. (1990a).
The Lag range M ultiplier test results indicated
that the W t of the m odel would be im proved if items
were allowed to load onto m ore than one factor. T he
m ultivariate Lag range M ultiplier test results indi-
cated that the W t of the m odel would be signiW cantly
im proved by adding 17 new param eters in sam ple A
( x2 im provem ent = 478; see Table 5) and by adding
20 new param eters in sam ple B ( x2 im provem ent =
470; see Table 6). M ost notably, the results indi-
cated that three item s (`I have self-d oubts’ : sam ple A,
x2 = 99.0, P < 0.001; sam ple B, x2 = 87.0, P < 0.001;

Table 5 Lagrange Multiplier test scores for adding parameters in sample A: SigniW cant predictors
only ( P < 0.01)

2 2
Item (factor) M ultivariate x Univariate x

I have self-doubts (Self-con W dence) 99 99.0

M y body feels relaxed (Self-con W dence) 154 55.1
I am concerned that I m ay not do as well in this competition as I
could (Self-con W dence) 209 54.5
I’ m concerned I won’ t be able to concentrate (Self-con W dence) 245 36.9
I am concerned about choking under pressure (Somatic Anxiety) 282 36.4
I feel at ease (Somatic Anxiety) 311 28.9
I feel nervous (Cognitive Anxiety) 331 20.4
I’ m concerned about reaching my goal (Self-con W dence) 350 19.4
I feel comfortable (Somatic Anxiety) 368 17.1
I’ m con W dent about perfor ming well (Somatic Anxiety) 386 18.4
I feel secure (Somatic Anxiety) 409 22.6
M y heart is racing (Self-con W dence) 423 14.1
I am concerned about losing (Self-con W dence) 435 12.6
I feel tense in the stomach (Self-con W dence) 447 11.9
I’ m concerned I won’ t be able to concentrate (Somatic Anxiety) 458 11.3
I am concerned about this competition (Self-con W dence) 470 11.1
I have self-doubts (Somatic Anxiety) 478 8.17
510 Lane et al.

Table 6 Lagrange M ultiplier test scores for adding parameters in sample B: SigniW cant predictors only (P < 0.01)

Item (factor) Multivariate x2 Univariate x2

I have self-doubts (Self-con W dence) 87 87.0

M y body feels relaxed (Self-con W dence) 167 79.8
I am concerned that I may not do as well in this competition as I could (Self-con W dence) 222 55.2
I feel nervous (Cognitive Anxiety) 259 36.5
I’ m con W dent because I m entally picture myself reaching my goal (Somatic Anxiety) 283 24.4
I’ m concerned I won’ t be able to concentrate (Self-con W dence) 307 24.3
I feel comfortable (Somatic Anxiety) 327 20.0
I feel at ease (Somatic Anxiety) 351 23.7
I’ m concerned about reaching my goal (Self-con W dence) 366 15.1
I feel tense in the stomach (Self-con W dence) 382 16.3
I feel jittery (Self-con W dence) 395 13.0
I am concerned about performing poorly (Somatic Anxiety) 407 12.0
I feel secure (Somatic Anxiety) 417 10.3
I’ m con W dent about performing well (Somatic Anxiety) 428 10.1
M y heart is racing (Self-con W dence) 436 8.79
I am concerned about choking under pressure (Somatic Anxiety) 445 8.36
I’ m concerned I won’ t be able to concentrate (Somatic Anxiety) 453 8.60
I’ m concerned about this competition (Self-con W dence) 460 7.11
I feel nervous (Self-con W dence) 465 5.03
I’ m concerned that others will be disappointed with my perform ance (Self-con W dence) 470 4.62

`M y body feels relaxed’ : sam ple A, x2 = 55.1, P < 0.001; D iscussio n

sam ple B, x2 = 79.8, P < 0.001; `I am concerned that
I m ay not do as well in this com petition as I could’ :
sam ple A, x2 = 54.7, P < 0.001; sam ple B, x2 = 55.2,
P < 0.001) should cross-lo ad into the Self-conW dence
scale.
T he internal consistency coeY cients for the three
subscales were all above the 0.70 criterion value: C og-
nitive Anxiety, alph a = 0.80; Som atic Anxiety, alpha =
0.85; and Self-conW dence, alpha = 0.88. Although
this provides support for the hypo thesized m odel, the
analysis was re-run with 27 item s, producing an alpha
coeY cient of 0.70. This result could be construed as
evidence to show that including all item s in a single
anxiety dim ension produces an internally consistent
factor. It also reinforces the Lag range M ultiplier test
results, which sugge st that several item s should load
onto m ore than one factor to increase the W t of the
m odel. H owever, Schutz and G essaroli (1993) argu ed
that internal consistency coeY cients are unreliable
estim ates of the inter-reliability of item s, as it is inX u-
enced by the num ber of items included in the analysis.
T herefore, Schutz and G essaro li (1993) suggested that
the hom ogeneity of item s is exam ined better using
factor loadings.
W hen exam ined collectively, the results provide
strong evidence that the m odel proposed by M artens
et al. (1990a) produced an unacceptable W t to exp lain
satisfactorily the obser ved variance within the data.
In this study, we re-evaluated the factorial validity of the
C SAI-2 (M ar tens et al., 1990a). T he rationale for the
investigation was based on the argu m ent that theory
testing and construct m easurem ent are inextricably
linked (Hendrick and Hendick, 1986; T hom pson and
D aniel, 1996). If the validity of a m easurem ent in-
strum ent is in question, then it is not possible to test
the asso ciated theory w ith any accuracy. The results of
the present study bring into question the validity of the
three-factor m odel for the C SAI-2 proposed by M artens
et al. (1990a).
G iven the nature of cognitive anxiety, we hypo th-
esized that an item such as `I have self-d oubts’ should
have shown the strongest relationship with the C ognitive
Anxiety scale rather than the weakest. T herefore, at
a theoretical level, it could be argu ed that the item `I
have self-d oubts’ genuinely assesses cognitive anxiety,
w hereas the other eight item s in the scale that refer to
feeling `concerned’ assess a slightly diV erent construct.
Logically, athletes who are about to take part in an im -
portant com petition are likely to report feeling con-
cerned about perform ance and produce a higher score
for cognitive anxiety, even though they m ay rem ain
conW dent in their ability to m eet the dem ands of the
task. Being concerned about an im pending perform ance
does not necessar ily m ean that an athlete is experiencing
negative thoughts, but that he or she is acknowledging
M easures of anxiety
the im portance and diY culty of the challenge and is
attem pting to m obilize resources to cope.
It has been found that athletes som etim es interpret
cognitive anxiety sym ptom s as facilitative of perform -
ance. Indeed, this has prom pted the developm ent of
a directional scale for the CSAI-2 (Jones et al., 1993),
whereby respondents quantify the extent to which
they feel that anxiety sym ptoms will facilitate or de-
bilitate perform ance. Recent research has suggested
that using the CSAI-2 without a direction scale m ay
provide a m isleading m easure of anxiety (Perry and
W illiam s, 1998). It seems parad oxical that cognitive
anxiety, a construct proposed to be typiWed by negative
expectations, could be perceived as facilitative of
perform ance, or that self-conW dence, typiW ed by
positive expectations, could be seen as debilitative of
perform ance. Interestingly, Jones and co-workers have
abandoned using the directional scale to assess self-
conW dence because of the strong relationship between
intensity and direction of perceptions (see Jones, 1995,
for a review). G iven the proposed nature of cognitive
anxiety, the sam e logic should apply. To reconcile this
contradiction, we suggest that item s in the Cognitive
Anxiety scale should be reworded to reXect the extent to
which individuals are `worried’ about perform ance, as
the notion of worry better captures the negative self-
im ages proposed to be central to the cognitive anxiety
construct. We propose that such a change reX ects m ore
than sem antic nuance and, indeed, lies at the heart of
conceptual integrity.
M artens et al. (1990a) originally used the word
`worried’ in som e item s in the Cognitive Anxiety scale
but replaced it w ith `concern’ in the W nal stage of
the factorial validation process to reduce social de-
sirab ility. It is not unreasonable to assu m e that athletes
would m ore readily acknowledge concern about a
com petition than worry, and perhaps report this as
likely to facilitate good perform ance. H owever, the m ore
`honest’ responses m ay sim ply reX ect the im portance
attached to the event by the individual rather than
negative exp ectations. Therefore, the price of reduced
social desirability bias m ay have been the conceptual
integrity of the cognitive anxiety construct. T his issue
needs fur ther exam ination.
The place of a Self-conW dence scale in an anxiety
inventory needs a strong theoretical rationale. M artens
et al. (1990a) found that the Self-conW dence scale
emerged out of exploratory factor analysis techniques.
T he items in the scale had originally been included
in the item pool to assess cognitive anxiety through
positively worded item s. Recent research has questioned
the reproducibility of the structure of the original self-
conW dence factor. Prapavessis et al. (1996) replicated
the techniques used by M artens et al. (1990a) on a
sam ple of 199 athletes from diV erent spor ts. (These
511
details, w hich were not repor ted by Prapavessis et al.,
1996, were supplied via personal correspondence). T he
results indicated that Self-conW dence divided into two
factors: one com prised W ve item s that describe positive
perform ance expectations (e.g. `I am conW dent I can
m eet the challenge’ and `I’ m conW dent about per-
form ing well’ ) and the other com prised four items
that describe an absence of cognitive anxiety (e.g. `I
feel com fortable’ and `I feel at ease’ ) and therefore
seem to assess w hat could be described as a sense of
calm ness.
The doubts exp ressed about the psychom etric in-
tegrity of the C SAI-2 are founded on analyses that
were not available at the time of the developm ent and
validation of the m easure. C onW rm atory factor analysis
is proposed to be a rigorous test of theory, because data
are tested again st a hypo thesized m odel. T he CSAI-2
was developed using exploratory factor analysis, w hich,
it has been argu ed (Thom pson and D aniel, 1996),
lacks a theoretical basis by virtue of its exploratory
nature. Factors derived from explorator y techniques are
a product of the item s entered into the analysis and m ay
be anom alous to the participants being investigated
rather than general constructs. F ur therm ore, M ulaik
(1987) argu ed that data can inform judgem ents, but the
developm ent of psychological m easurem ents should be
grounded in theory not data.
We conclude that investigators of anxiety responses
to sport com petition cannot have faith in data ob-
tained using the CSAI-2 until further validation studies
have been com pleted and possible reW nements to the
inventory have been m ade.
References
Bentler, P.M . (1992). E QS Str uctural E quation Program
M anual . Los Angeles, C A: BMDP Statistical Software.
Bentler, P.M . (1995). E QS Str uctural E quation Program
M anual. Los Angeles, CA: BMDP Statistical Software.
Bentler, P.M . and Wu, E.J.C. (1995). E Q S/W indow s U ser’ s
G uide . Los Angeles, CA: BM DP Statistical Software.
Bynner, J.M . and Romney, D. (1985). LISREL for beginners.
Canadia n Psych olog y , 26 , 43± 49.
Byrne, B.M . (1989). A Pr im er of Lisrel: B asic Applications
Program m ing for C on W r m ator y Factor Ana lytic M odels . New
York: M acm illan.
Cronbach, L.J. (1951). C oeY cient alpha and internal
structure of tests. Psychom etr ika , 16 , 297± 334.
Hardy, L., Jones, J.G. and Gould, D. (1996). U nderstanding
Psychological Preparation for Sport : Theor y and Practice of
Elite Perform ers . C hichester: W iley.
Hendrick, C. and Hendrick, S. (1986). A theory and method
of love. Jour nal of Personality and Socia l Psychology , 50 ,
392± 402.
Hu, L. and Bentler, P.M . (1995). Evaluating model W t.
In Str uctural Equatio n M odelling : C oncep ts, Issues, and
512
A pplications (edited by R.H. Hoyle), pp. 76± 99. London:
Sage.
Jones, J.G. (1995). M ore than just a game: Research develop-
m ents and issues in competitive anxiety in sport. B ritish
Jour na l of Psychology , 85 , 449± 478.
Jones, J.G., Swain, A.B.J. and Hardy, L. (1993). Intensity
and direction dimensions of competitive state anxiety and
relationships with performance. Jour nal of Sports Sciences ,
11 , 525± 532.
M artens, R., Vealey, R.S., Burton, D., Bump, L. and Smith,
D.E . (1990a). Developm ent and validation of the Com -
petitive Sports Anxiety Inventory 2. In Com petitive Anx iety
in Sport (edited by R. M artens, R.S. Vealey and D. Burton),
pp. 117± 178. Cham paign, IL: Human Kinetics.
M artens, R., Vealey, R.S. and Burton, D. (1990b). Com petitive
A nxiety in Sport . Cham paign, IL: Human Kinetics.
M ulaik, S.A. (1987). A brief history of the philosophical
foundations of exploratory factor analysis. M ultivar iate
B ehavioural Research , 22 , 267± 305.
Perr y, J.D. and Williams, J.M . (1998). Relationship of inten-
Lane et al.
sity and direction of competitive trait anxiety to skill level
and gender in tennis. The Sport Psychologist , 12 , 169± 179.
Prapavessis, H., Cox, H. and Brookes, L. (1996). A test
of M artens, Vealey and Burton’ s theory of competitive
anxiety. Austra lian Jour na l of Science and M edicine in Sport ,
28 , 24± 29.
Schutz, R.W. and Gessaroli, M .E. (1993). Use, misuse, and
disuse of statistics in psychology research. In H andbook
of Resea rch in Sport Psychology (edited by R.N. Singer, M .
M urphy and L.K. Tennant), pp. 901± 921. New York:
M acmillan.
Tabachnick, B.G. and Fidell, L.S. (1996). U sing M ultivar iate
Statistics , 2nd edn. New York: Harper & Row.
Thompson, B. and Daniel, L.G. (1996). Factor analytic
evidence for the construct validity of scores: A historical
overview and some guidelines. E ducatio nal and Psych o-
log ical M easurem ent , 56 , 197± 208.
Tucker, L.R. and Lewis, C. (1973). A reliability coe Y cient
for maximum likelihood factor analysis. Psychom etr ika , 38 ,
1± 10.