Available online at www.sciencedirect.com
Psychiatry Research 168 (2009) 32 – 39
www.elsevier.com/locate/psychres
Factorial composition of the Aggression Questionnaire:
A multi-sample study in Greek adults
Silia Vitoratou a,c , Ioannis Ntzoufras c , Nikolaos Smyrnis a,b , Nicholas C. Stefanis a,b,⁎
a
c
University Mental Health Research Institute, Soranou Efesiou 2, Papagou, 156 01, Athens, Greece
b
Department of Psychiatry, National and Kapodistrian University of Athens Medical School,
Eginition Hospital 74 Vas. Sofias Ave., Athens 115 28, Greece
Department of Statistics, University of Economics and Business, Patision 76, 104 34 Athens, Greece
Received 26 July 2007; received in revised form 7 January 2008; accepted 28 January 2008
Abstract
The primary aim of the current article was the evaluation of the factorial composition of the Aggression Questionnaire (AQ29)
in the Greek population. The translated questionnaire was administered to the following three heterogeneous adult samples: a general
population sample from Athens, a sample of young male conscripts and a sample of individuals facing problems related to substance
use. Factor analysis highlighted a structure similar to the one proposed by Buss and Perry [Buss, A.F., Perry, M., 1992. The
Aggression Questionnaire. Journal of Personality and Social Psychology 63, 452–459]. However, the refined 12-item version of
Bryant and Smith [Bryant, F.B., Smith, B.D., 2001. Refining the architecture of aggression: a measurement model for the Buss-Perry
Aggression Questionnaire. Journal of Research in Personality 35, 138–167] provided a better fit to our data. Therefore, the refined
model was implemented in further analysis. Multiple group confirmatory factor analysis was applied in order to assess the variability
of the 12-item AQ across gender and samples. The percentage of factor loading invariance between males and females and across the
three samples defined above was high (higher than 75%). The reliability (internal consistency) of the scale was satisfactory in all cases.
Content validity of the 12-item AQ was confirmed by comparison with the Symptom Check-List 90 Revised.
© 2008 Elsevier Ireland Ltd. All rights reserved.
Keywords: CFA; AQ; Psychometrics
1. Introduction
Since the seminal conceptualization of aggression by
Moyer (1968), subsequent efforts have strived to further
refine, operationalise and quantify persistent aggressive
behavior. This is not merely an academic endeavor, given
⁎ Corresponding author. Tel.: +30 693 660 8887; fax: +30 210
7242020.
E-mail address: nistefan@med.uoa.gr (N.C. Stefanis).
rising public health concerns reflected in the international
attention directed to prevention programs designed to
address aggressive and violent behavior (Ahmad, 2004).
Aggression is a potential predictive factor of later criminal activity during adolescence (Huesmann et al., 1984;
Huesmann and Efron, 1992; Pulkkinen and Pitkanen,
1993). Furthermore, aggression is a multidimensional
construct that develops within a complex interaction of
biological, psychological, social, and cultural factors.
While aggressive acts are state phenomena, the tendency
0165-1781/$ - see front matter © 2008 Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.psychres.2008.01.016
S. Vitoratou et al. / Psychiatry Research 168 (2009) 32–39
to engage in aggressive behavior over the lifetime is
relatively stable (Olweus, 1979; Huesmann et al., 1984;
Coccaro et al., 1991).
Self-rated aggression measures remain a popular
method to assess trait aggression. Amongst these, the
Aggression Questionnaire (AQ29) was created by Buss
and Perry (1992) and represents an updated and psychometrically improved version of the Buss-Durkee
Hostility Inventory (Buss and Durkee, 1957) that has
quickly become the gold-standard for the measurement
of aggression.
The AQ29 has been cross-validated in several nonUS samples including Dutch (Meesters et al., 1996),
Japanese (Nakano, 2001), Spanish (Garcia-Leon et al.,
2002), Italian (Fossati et al., 2001), Chinese (Maxwell,
2007), Greek (Tsorbatzoudis, 2006) and German
populations (Collani and Werner, 2005). Further,
Williams et al. (1996), Morren and Meesters (2002)
and Diamond et al. (2005) assessed the properties of the
AQ29 in offender samples.
As initially reported by Buss and Perry (1992) and
further elaborated by other authors (Harris, 1995; Bernstein
and Gesn, 1997; Garcia-Leon et al., 2002; Fossati et al.,
2001), the AQ29 is a multidimensional instrument consisting of the following four subscales: Physical aggression,
Verbal aggression, Anger and Hostility1. Two further proposals for the latent structure of the AQ29 are proposed
in the literature, namely: a) a one-factor model, which
assumes that all items load on one first order factor, and b) a
second order factor implicated in the initial four-factor
model.
Further suggestions have been made about the
questionnaire's items. Specifically, Harris (1995) proposed to omit two items from the Hostility subscale (H6
and H8), thereby raising its reliability estimators (model
c). Meesters et al. (1996) reported a closer fit after further
omission of the first indicator of the Verbal aggression
sub-scale, resulting in a modified 26-item AQ (model d).
In addition, Bryant and Smith (2001) proposed a modified
version of the AQ excluded a substantial number of items
(model f). The selection of the excluded items was based
upon i) factor complexity (omitting items that load on
more than one factor) and ii) salience of loadings
(omitting items with loadings lower than 0.40), while
they also omitted the two items with reverse scoring.
The primary aim of the present study was to evaluate
the proposed models in the literature regarding the
Aggression Questionnaire in three Greek adult samples.
1
Hereafter the items are indicated by the item's identification
number as presented in Buss and Perry (1992).
33
Moreover, the score and structure discrepancies among
these samples were evaluated.
2. Methods
2.1. Participants
2.1.1. Sample 1
The translated AQ29 was administered to 307 individuals selected from the general population in the region
of Rafina-Athens (randomly selected using the municipality registry). Among the individuals, 143 (46.6%) were
male (mean age= 49.13, sd= 10.8, range: 22–73 years)
and 164 (53.4%) were female (mean age = 49.14, sd =
11.0, range: 21–67 years). No significant age differences across genders were found (t = − 0.012, df = 301,
P = 0.979).
2.1.2. Sample 2
The AQ29 was also randomly administered to 1228
male conscripts (aged 19–24 years; mean = 20.83, sd =
1.87), who were recruited from the Greek Air Force
during their first 2 weeks of admission in the National
Basic Air Force training centre in the city of Tripoli. This
sample can be considered representative of males in this
age group, since military service is obligatory in Greece.
2.1.3. Sample 3
The third sample consists of 165 volunteers, members of the Greek methadone program of the Organisation Against Drugs (ΟΚΑΝΑ) in Athens. Among the
individuals, 127 (77%) were males (mean age = 36.6,
sd = 5.5, range: 24–55 years) and 33 (20%) were
females (mean age = 36.4 years, sd = 5.2, range: 28–
51 years); for five individuals, gender was not reported.
Age did not differ significantly across gender (t = 0.176,
df = 156, P = 0.861).
2.2. Measurements
The Symptom Check-List 90 Revised (SCL-90-R;
Derogatis, 1993; Donias et al., 1991) was also administered, within the Athens Study of Psychosis Proneness
and Incidence of Schizophrenia (ASPIS; Stefanis et al.,
2004). The SCL-90-R is a 90-item multidimensional selfreport inventory consisting of nine subscales and three
indices of psychological distress. The reliability coefficient
of SCL-90-R was high for the total scale (alpha = 0.97)
while the corresponding coefficients for each subscale were
found to range from 0.66 to 0.87 (Stefanis et al., 2004).
The AQ29 (Buss and Perry, 1992) is a self-administered
inventory that consists of 29 Likert type items (scored 1 to
34
S. Vitoratou et al. / Psychiatry Research 168 (2009) 32–39
5). Twelve of these items constitute the AQ12 (Bryant and
Smith, 2001), which nonetheless retains the four-factor
latent structure.
for significantly variant loadings were excluded. Data
analysis was conducted in AMOS 5 (Arbuckle, 2003).
3. Results
2.3. Translation procedure
The AQ29 adaptation for the Greek population followed the recommendations of Van de Vijver and
Hampleton (1996). Specifically, the questionnaire was
translated into Greek and back-translated into English
by an independent official translator. Comparison of the
original and the first English drafts produced a modified
Greek version that received minor further changes when
administered to a test sample of 15 young employees
of the University Mental Health Research Institute of
Athens (UMHRI).
2.4. Statistical analyses
Three criteria were used to evaluate the items'
intercorrelations: Measures of Sampling Adequacy for
each item (MSA), the Kaiser–Meyer–Olkin measure of
sampling adequacy (KMO) over all items and Bartlett's
test of Sphericity (Dziuban and Shirkey, 1974). For
purposes of comparison with previously reported results, Exploratory Factor Analysis (EFA) was implemented in a random split-half of the second sample
(selected due to its large size). However, the factorial
structure of the Greek version of the questionnaire was
essentially assessed utilizing CFA via a structural equation modelling approach (Bollen, 1989). The measures of fit that are reported are in concordance with
the current AQ29 literature. In this way, comparison
between studies can be easily obtained. Three absolute fit indices are reported, namely the Goodness of
Fit Index (GFI; Jöreskog and Sörbom, 1984), the Root
Mean Square Error of Approximation (RMSEA;
Browne and Cudeck, 1993) and the relative chi-square
(χ2 /df; Hoelter, 1983). Furthermore, the fit of each
model compared to the null (or independence) model
was assessed by using two relative fit indices, namely
the Non-Normed Fit Index (NNFI; Bentler and Bonett,
1980) and the Comparative Fit Index (CFI; Bentler,
1990).
The variation of the factor loadings across samples and the gender effect were assessed via Multiple
Groups CFA. Equality of loadings is necessary to make
comparisons between groups (Anderson and Gerbing,
1988). By examining the critical ratios of the loading
difference across samples for each item (i.e. the ratio
of the difference's estimate divided by the estimate
of difference's standard error), the equality constraints
3.1. Sampling adequacy and exploratory factor
analysis
The assessment of the sampling adequacy diagnostics led to satisfactory MSA values (0.66 to 0.94). Values
were lower than 0.73 for only two items. Furthermore,
Bartlett's test of sphericity (chi-square = 4796.7,
df = 406, P b 0.001) indicated that the intercorrelations
were satisfactory, while the KMO measure was high
(0.89), indicating low partial intercorrelations among
items. The above findings support the existence
of possible latent factors. EFA using principal axis
factoring (Oblimin rotation) was implemented. Seven
eigenvalues above one were found (6.8, 2.6, 1.8, 1.5,
1.2, 1.1 and 1.1). Four factors existed explaining the
43.7% of the total variance, while the next three eigenvalues represented scree2. Further investigation of the
questionnaire's latent structure was implemented by
CFA methodology, which follows.
3.2. Confirmatory factor analysis
CFA was applied in all three Greek samples separately. Seven distinct models that have been proposed in
the literature were evaluated:
a) One-factor model (all items load on one first order
factor),
b) Initial Buss and Perry's (1992) four-factor model
(four factors explain the items' covariation),
c) and d) model b with reduced items as proposed
correspondingly by Harris (1995) and Meesters et al.
(1996),
e) Hierarchical model (one second order factor
underlies the four factors of the initial model b),
f ) Bryant and Smith's (2001) refined model (the
initial model b with the number of items reduced to
12) and
g) Hierarchical refined model (one second order factor underlies the four factors of the refined model f ).
The one-factor model (a) inadequately fit the
data in all samples (Table 1). The indices were improved in all cases after fitting Buss and Perry's four-
2
Full EFA results are available upon request.
35
S. Vitoratou et al. / Psychiatry Research 168 (2009) 32–39
Table 1
Absolute and relative fit indices for proposed models in the literature
of AQ29 (samples 1–3).
Model
a) Buss and
Perry (1992):
1 global factor
b) Buss and
Perry (1992):
4-factor model
e) Buss and Perry
(1992): 1 second
order factor
f ) Bryant and
Smith (2001):
12 items-4 factors
g) Bryant and Smith
(2001): 1 second
order factor
a
b
c
Sample c Absolute fit measures
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3
Relative fit
measures
χ2/df
GFI
RMSEA NNFI CFI
4.830
5.196
2.382
3.688
6.123
1.842
3.676
3.652
1.892
2.686
2.373
1.717
2.377
2.903
1.700
0.620
0.771
0.695
0.719
0.861
0.775
0.718
0.844
0.774
0.963
0.942
0.929
0.940
0.958
0.928
a
0.115
a
0.085
a
0.093
a
0.096
a
0.067
a
0.074
a
0.096
a
0.068
a
0.075
b
0.054
b
0.069
b
0.067
b
0.069
b
0.057
b
0.066
0.397
0.627
0.573
0.577
0.767
0.730
0.579
0.764
0.724
0.922
0.864
0.740
0.864
0.911
0.875
0.440
0.653
0.603
0.613
0.787
0.753
0.613
0.783
0.747
0.943
0.901
0.907
0.897
0.933
0.906
P ≤ 0.001.
P N 0.05.
Sample 1: N=290, Sample 2: N=577 and Sample 3: N=160.
factor model (b). The GFI increased substantially, while
the RMSEA decreased reaching a reasonably close fit in
the second and third samples but not in the first one
(0.096). As can also seen by the relative chi-square, the
fit of the model was adequate only in the third sample
(Table 1). The fit of the hierarchical model (e) was
slightly worse than the former one, in terms of both
absolute and relative fit measures.
In order to evaluate the proposals formulated by
Harris (1995) and Meesters et al. (1996), we compared
models (c) and (d) with the corresponding one for AQ29
(b). cases, the 29-item model proposed by Buss and
Perry (1992) provided the best fit to model in our data,
since the Browne–Cudeck Criterion (Browne and
Cudeck, 1989) was smaller in all samples than the
corresponding value for: the other two models (Table 2).
The 12-item AQ proposed by Bryant and Smith
(2001) was further evaluated. The measures of fit of
model (f) were satisfactory, indicating adequate fit in all
samples. The values of GFI (0.94–0.96) indicate the
high proportion of variance explained. The RMSEA was
not significantly higher than 0.05 with the exception of
the second sample, which nevertheless corresponds to a
reasonably close fit according to Browne and Cudeck
(1993). The relative fit indices (NNFI and CFI) reached
their highest values, with CFI being higher than 0.9 in
all three samples. The relative chi-square value was
lower than the threshold value of 2 only in the first
sample, while its values were 2.4 and 2.7 in the other
two samples. Further, the corresponding hierarchical
refined model (model g) provided similar indices with
the latter one (Table 3).
3.3. Multiple group confirmatory factor analysis
3.3.1. Gender invariance
The gender invariance of the best-fitting model (i.e.
Bryant and Smith model f) was examined in the first and
third samples (Tables 4 and 5). The chi-square statistic
did not indicate a difference in the general population
sample of Athens. Hence the AQ12 satisfies the
requirement of loadings invariance and scores can, be
meaningfully compared across genders.
Even though the chi-square statistic indicated a significant gender effect for the third sample (namely items
V2 and V4, which were higher in the females' group),
the sample size of the females' group (N = 33) was small
(3:1 ratio of observations versus items) and may lead
to unstable results.
3.3.2. Sample invariance
The sample loading invariance was also assessed for
the Bryant and Smith (2001) four-factor model ( f ). The
chi-square statistic indicated significant sample effect;
hence we proceeded with pair-wise comparisons among
samples (Tables 4 and 5). Differences in the loadings
were present only between the second and the third
sample (items A1, H3 and V4; Table 5).
3.4. Descriptive indices and internal consistency
Descriptive indices and Cronbach's (1951) alpha
coefficients for each subscale and total scores for
both AQ29 and AQ12 are presented in Table 1. Analysis of variance was used to compare the mean scores
between the three samples. The highest scores appeared
in the third sample (individuals facing problem related
to substance abuse), followed by the second one (young
male conscripts), while the lowest values appeared in
the first sample (general population sample of Athens).
Table 2
Browne–Cudeck Criterion for three proposed models for the AQ29
(samples 1–3).
Model
Sample 1
Sample 2
Sample 3
(b) 29 items
(c) 27 items
(d) 26 items
861.98
905.58
910.92
1511.18
1603.91
1614.65
1495.98
1594.62
1597.60
36
S. Vitoratou et al. / Psychiatry Research 168 (2009) 32–39
Table 3
Descriptive indices and internal consistency of the AQ29 (and AQ12 within parentheses) subscale and total scores in samples 1, 2 and 3.
Sample 1 (General population from Athens, n = 307)
Sample 2 (Male conscripts, n = 1228)
Sample 3 (Drug dependent individuals, n = 165)
a,b,c
AQ scale
Alpha
Physical
Verbal
Anger
Hostility
Total
Physical
Verbal
Anger
Hostility
Total
Physical
Verbal
Anger
Hostility
Total
0.82 (0.77)
0.55 (0.55)
0.75 (0.57)
0.75 (0.66)
0.85 (0.74)
0.82 (0.70)
0.50 (0.56)
0.79 (0.60)
0.70 (0.61)
0.88 (0.77)
0.84 (0.76)
0.51 (0.48)
0.74 (0.54)
0.70 (0.63)
0.87 (0.73)
Mean
S.D.
Range
19.4 (5.6)
14.2 (7.3)
a
20.1 (8.3)
c
20.3 (7.3)
74.0 (28.6)
25.7 (7.9)
14.7 (7.0)
a
19.7 (8.5)
c
21.5 (7.2)
81.5 (30.5)
27.0 (8.8)
b
15.5 (7.6)
21.0 (9.5)
24.1 (8.9)
88.6 (34.7)
6.6 (2.6)
3.1 (2.3)
5.0 (2.5)
5.0 (2.4)
14.5 (6.6)
7.3 (3.0)
3.0 (2.3)
5.6 (2.7)
5.4 (2.6)
16.3 (7.5)
7.9 (3.4)
3.0 (2.1)
5.2 (2.6)
5.4 (2.6)
16.1 (7.2)
9–43 (2–15)
5–25 (3–15)
7–34 (3–15)
8–38 (3–15)
44–122 (14–51)
9–45 (3–15)
5–25 (2–15)
7–35 (2–15)
7–40 (3–15)
35–135 (12–57)
9–45 (3–15)
8–24 (3–15)
10–35 (3–15)
8–37 (3–15)
53–123 (14–50)
b
The difference between the pairs of means was non-significant: P N 0.05.
Analysis of variance (using the Bonferroni post-hoc
multiple comparisons test) revealed significant differences in all cases (P b 0.05) with the exception of the
“Anger” scale between samples 1 and 2 (adjusted for
age and gender).
Regarding reliability, Cronbach's alpha coefficient
was high (0.85–0.88) for the total AQ29 and moderate
to high for the subscales (0.50–0.84). Two problematic
items were identified (Verbal subscale items 1 and 3)
with low item-total correlations (lower than 0.20) and no
decrease of alpha at item deletion, in all samples. Unlike
the other three statements included in the “Verbal
aggression” subscale of the AQ29, the items V1 (“I tell
my friends openly when I disagree with them”) and V3
(“When people annoy me, I may tell them what I think
of them”) describe socially acceptable behaviours in the
Greek culture. Possibly, they do not reflect aggressive
behaviour; therefore the low consistency with the rest of
the AQ29 items is justified.
With respect to the AQ12, the corresponding Cronbach's alpha coefficients were lower (0.73–0.77 for
the total scale and 0.48–0.76 for the subscales). This
result was expected for the shortened AQ12 since
Cronbach's alpha depends on the number of items
(Pedhazur and Schmelkin, 1991).
3.5. Age and gender effect
The age and gender effects on AQ12 subscale and
total scores were evaluated implementing normal regression. No age effect, adjusted for gender, was present
in samples 1 and 3. On the contrary, in the first sample the AQ12 scores were significantly different for
Table 4
Tests of invariance of loadings among genders and samples (AQ12 —
model f ).
Hypothesis
for loadings
Sample
Gender invariance 1
3
Sample invariance 1, 2 and 3
1 and 2
2 and 3
3 and 1
a
χ2/df
1.762
1.644
2.257
2.480
2.277
1.987
RMSEA a Nested models
comparison b
0.051
0.065
0.035
0.041
0.042
0.047
δχ2 (df )
P-value
15.5 (8)
27.0 (8)
35.7 (16)
15.0 (8)
25.3 (8)
10.2 (8)
0.050
0.001
0.003
0.058
0.001
0.249
P b 0.001.
Differences in χ2 between the unconstrained and the constrained
models.
b
Table 5
Invariance of loadings among gender and samples (AQ12 — model f ).
Hypothesis
for loadings
Sample Items
excluded
%
Nested models
Invariance comparison a
δχ2 (df ) P-value
Gender invariance 1
3
Sample invariance 1 and 2
2 and 3
3 and 1
a
–
100.0
V2, V4
83.3
–
100.0
A1,V4, H3 75.0
–
100.0
–
07.6 (6)
–
08.9 (5)
–
0.159
0.265
–
0.114
–
Differences in chi-square (δχ2) between the unconstrained and the
partially constrained models.
37
S. Vitoratou et al. / Psychiatry Research 168 (2009) 32–39
males and females, with the exception of “Hostility”
(Table 6). In both samples, males scored higher in
“Physical aggression” than females. Males scored
significantly higher than females in “Verbal aggression”
and lower in “Anger” in the general population sample of
Athens while the opposite effects emerged in the third
sample.
In the young males' sample (sample 2) the age effect
was significant. Specifically, there was a negative age
effect on “Physical Aggression” (regression coefficient =
−0.31, se = 0.05, P b 0.001) and “Hostility” (r = −0.23,
se = 0.04, P b 0.001) subscales. Further, a 1-year age
increase corresponded to a decrease of about half a unit in
the total AQ12 score (r = −0.51, se = 0.12, P b 0.001).
Table 7
Pearson correlation coefficients between AQ12 and SCL-90-R scores
(sample 2).
Correlations a
between SCL-90-R
and AQ12 scores
Physical Verbal Hostility Anger Total
AQ12
Anxiety
Depression
Hostility
Inter. Sensitivity
Obsessive–Compulsive
Paranoid ideation
Phobic anxiety
Psychoticism
Somatization
0.14
0.16
0.38 b, c
0.12
0.13
0.14
0.07
0.12
0.11
0.18
0.19
0.28 b
0.15
0.18
0.14
0.16
0.15
0.15
0.34 c
0.42 b, c
0.26
0.37, c
0.36 c
0.32 c
0.33 c
0.32 c
0.27 c
0.23
0.27
0.36 b
0.17
0.23
0.16
0.16
0.19
0.21
0.31
0.36
0.45
0.28
0.31
0.27
0.25
0.28
0.26
a
All coefficients were statistically significant (P b 0.05).
The coefficient was significantly higher than the others in its
column (P b 0.05) according Olkin's z (Olkin,1967).
c
The coefficient was significantly higher than the others in its row
(P b 0.05) according Olkin's z (Olkin,1967).
b
3.6. Correlations with SCL-90-R
All correlations at the subscale level between AQ12
and SCL-90-R were moderate (Table 7). The coefficients
were compared and two interesting outcomes emerged.
Firstly, all SCL-90-R subscales correlated substantially
higher (0.26–0.41) with AQ12 "Hostility" than with the
other three AQ12 subscales. Regarding the opposite
comparison, the correlations of the AQ12 with the SCL90-R “Hostility” (0.26 to 0.37) were significantly higher
compared to the ones with the remaining eight SCL-90-R
subscales (Olkin's z for comparing dependent correlation
coefficients, P b 0.001 in all cases).
Interestingly, a “Hostility-cross” pattern (Table 7)
emerged which does not include the two “Hostility”
subscales. Regarding the AQ12, “Hostility” correlated
higher with “Depression” (r = 0.42, P b 0.001). With respect to SCL-90-R, “Hostility” correlated lower with
Table 6
Differences in means of AQ12 scores between genders in samples 1
and 3.
Mean difference
between males and
females a
Sample 1
Physical
(General population from Athens, Verbal
nmales = 143, nfemales = 164)
Anger
Hostility
Total
Physical
Sample 3
(Drug dependent individuals,
Verbal
nmales = 127, nfemales = 33)
Anger
Hostility
Total
a
Note: in that order.
Diff.
t
P
1.00
0.65
0.59
−0.24
2.00
2.04
−0.27
−1.30
−1.10
−0.62
3.43
2.53
2.08
− 0.86
2.69
3.21
− 0.65
− 2.66
− 2.24
− 0.45
0.001
0.012
0.038
0.390
0.007
0.002
0.516
0.009
0.026
0.655
the homonymous AQ12 subscale than the other three
AQ12 subscales.
4. Discussion
The initial Buss and Perry (1992) AQ29 and the
counterpart hierarchical model provided an inadequate
fit to our data in terms of both absolute and relative fit
CFA measures. On the contrary, the Bryant and Smith
(2001) revised model (model f) provided the best fit as
well as its corresponding hierarchical model (g). The
above results were replicated in three dissimilar samples
of Greek adults. Despite the reduction in the number of
the items, the internal consistency of the AQ12 remained
satisfactory, providing evidence that the revised model
by Bryant and Smith should be implemented in the
Greek population.
Multiple groups CFA did not reveal substantial
discrepancies in the loadings among samples or genders.
On the assumption of metric invariance, comparisons
were carried out between the scores. The AQ12 scores
were substantially higher in the third sample. This fact
indicates the existence of augmented aggressive personality traits in individuals facing problems related to
substance abuse. The association between substance
abuse and aggressiveness is undoubtedly complex, but a
series of investigations with first graders revealed that
boys who were identified by teachers/peers as more
aggressive are more likely to use drugs in the future
(Kellam et al., 1980, 1982, 1983, 1989). Early aggressive behaviour was found predictive of later substance
abuse (Moffitt, 1993) and conduct disorder symptoms
have been observed to begin some years before regular
38
S. Vitoratou et al. / Psychiatry Research 168 (2009) 32–39
drug use (Young et al., 1995). In conclusion, there is
evidence suggesting that aggressive personality traits
may predate addictive behaviour.
Regarding age, even though a negative relationship
was found with aggression (in the conscripts' sample),
its magnitude was low and not replicated in the other
samples, and hence a substantial age effect is not apparent. With respect to the gender effect, males scored
higher than females in “Physical aggression” in
samples 1 and 3. This is consistent with most studies
that have utilized the AQ. In contrast, no clear pattern
emerged when analyzing gender effects on the other
three aggression subscales.
All nine SCL-90-R subscales were preferentially
correlated with the AQ12 “Hostility” subscale rather
than with the other three aggression subscales in accordance with Morren and Meesters (2002). Specifically,
the highest correlations of the AQ12 “Hostility” subscale
emerged with the SCL-90-R “Depression” and “Interpersonal sensitivity” subscales. This is intuitive since
the AQ12 “Hostility” subscale includes items mostly
reflecting depressive traits, suspiciousness and bitterness. On the other hand, the AQ12 subscales are more
correlated with SCL-90-R “Hostility” rather than with
the other eight SCL-90-R subscales. This finding is also
makes intuitive sense, since the SCL-90-R "Hostility"
items reflect mostly anger and physical aggression. This
would explain additionally why the two homonymous
subscales are not highly inter-correlated. Conclusively,
these results provide evidence for content validity for
both the AQ12 and SCL-90-R questionnaires.
With reference to the limitations of the present study,
no test-retest analysis was performed to evaluate the
stability of the AQ. Another limitation concerns the lack
of further evaluation of convergence and discriminant
validity. Finally, even though the Bryant and Smith
(2001) revised model had a reasonably close fit to our
data we acknowledge the need for a model with improved fit. The Greek version of the AQ12 had satisfactory properties and thus can be considered a valuable
instrument in the assessment of aggressive behaviour.
References
Ahmad, K., 2004. Violence prevention receives international attention.
Lancet 363, 220.
Anderson, J.C., Gerbing, D.W., 1988. Structural equation modelling in practice: a review and recommended two step approach.
Psychological Bulletin 103, 411–423.
Arbuckle, J.L., 2003. Amos 5.0 update to the Amos 5.0 Update to the
Amos 4.0 User's Guide. Smallwaters Corporation, Chicago.
Bentler, P.M., 1990. Comparative fit indexes in structural models.
Psychological Bulletin 107, 238–246.
Bentler, P.M., Bonett, D.G., 1980. Significance tests and goodness of
fit in the analysis of covariance structures. Psychological Bulletin
88, 588–606.
Bernstein, I.H., Gesn, P.R., 1997. On the dimensionality of the Buss/
Perry Aggression Questionnaire. Behaviour Research and Therapy
35, 563–568.
Bollen, K.A., 1989. Structural equations with latent variables. Wiley,
New York.
Browne, M.W., Cudeck, R., 1989. Single sample cross-validation indices
for covariance structures. Multivariate Behavioural Research 24,
445–455.
Browne, M.W., Cudeck, R., 1993. Alternative ways of assessing
model fit. In: Bollen, K.A., Long, J.S. (Eds.), Testing Structural
Equation Models. Sage, Newbury Park C.A., pp. 136–162.
Bryant, F.B., Smith, B.D., 2001. Refining the architecture of aggression: a measurement model for the Buss–Perry Aggression
Questionnaire. Journal of Research in Personality 35, 138–167.
Buss, A.F., Durkee, A., 1957. An inventory for assessing different
kinds of hostility. Journal of Consulting Psychology 21, 343–349.
Buss, A.F., Perry, M., 1992. The Aggression Questionnaire. Journal of
Personality and Social Psychology 63, 452–459.
Coccaro, E.F., Harvey, P.D., Kupsaw-Lawrence, E., Herbert, J.L.,
Bernstein, D.P., 1991. Development of neuropharmacologically
based behavioral assessments of impulsive aggressive behavior.
Journal of Neuropsychiatry 3, 44–51.
Collani, G., Werner, R., 2005. Self-related and motivational constructs as
determinants of aggression. An analysis and validation of a German
version of the Buss–Perry Aggression Questionnaire. Personality
and Individual Differences, 38, 1631–1643.
Cronbach, L.J., 1951. Coefficient alpha and the interval structure of
tests. Psychometrika 16, 297–334.
Derogatis, L.R., 1993. Symptom Checklist-90-R, SCL-90-R. Computer Systems Inc., Minneapolis.
Diamond, P.M., Wang, E.W., Buffington-Vollum, J., 2005. Factor
structure of the Buss–Perry Aggression Questionnaire (BPAQ)
with mentally ill male prisoners. Criminal Justice and Behaviour
32, 546–564.
Donias, S., Karastergiou, A., Manos, N., 1991. Standardization of the
Symptom Check-List-90-R rating scale in a Greek population.
Psychiatriki 2, 42–48.
Dziuban, C.D., Shirkey, E.C., 1974. When is correlation matrix
appropriate for factor analysis? Psychological Bulletin 81, 358–361.
Fossati, A., Di Ceglie, A., Acquarini, E., Barratt, E.S., 2001. Psychometric properties of an Italian version of the Barratt Impulsiveness Scale-11 (BIS-11) in non clinical subjects. Journal of Clinical
Psychology 57, 815–828.
Garcia-Leon, A., Reyes, G.A., Vila, J., Perez, N., Robles, H., Ramos,
M., 2002. The Aggression Questionnaire: a validation study in
student samples. The Spanish Journal of Psychology 5, 45–53.
Harris, J.A., 1995. Confirmatory factor analysis of the Aggression
Questionnaire. Behaviour Research and Therapy 33, 991–993.
Hoelter, J.W., 1983. The analysis of covariance structures: goodness of
fit indices. Sociological Methods and Research 11, 325–344.
Huesmann, L.R., Efron, L.D., 1992. Childhood aggression and adult
criminality. In: McCord, J. (Ed.), Facts, Frameworks and
Forecasts: Advances in Criminological Theory. Transaction, New
Brunswick, NJ, pp. 137–156.
Huesmann, L.R., Efron, L.D., Lefkowitz, M.M., Walder, L.O., 1984.
Stability of aggression over time and generations. Developmental
Psychology 20, 1120–1134.
Jöreskog, K.G., Sörbom, D., 1984. LISREL-VI User's Guide, 3rd ed.,
Scientific Software, Mooresville.
S. Vitoratou et al. / Psychiatry Research 168 (2009) 32–39
Kellam, S.G., Ensminger, M.E., Simon, M.B., 1980. Mental health in
first grade and teenage drug, alcohol and cigarette use. Drug and
Alcohol Dependence 5, 273–304.
Kellam, S.G., Stevenson, D.L., Rubin, B.R., 1982. How specific are
the early predictors of teenage drug use? National Institute on Drug
Abuse Research Monograph, Series 43, 329–334.
Kellam, S.G., Brown, C.H., Fleming, J.P., 1983. Relationship of firstgrade social adaptation to teenage drinking, drug-use, and smoking.
Digest of Alcoholism Theory & Application 2, 20–24.
Kellam, S.G., Ialongo, N., Brown, H., Laudolff, J., Mirsky, A., Anthony,
B., Ahearn, M., Anthony, J., Edelsohn, G., Dolan, L., 1989. Attention
problems in first grade and shy and aggressive behaviours as antecedents to late heavy or inhibited substance use. National Institute on
Drug Abuse Research Monograph, Series 95, 368–369.
Maxwell, J.P., 2007. Development and validation of a Chinese version
of the Buss–Perry Aggression Questionnaire. Journal of Personality Assessment, 88, 1–11.
Meesters, C., Muris, P., Bosma, H., Schouten, E., Beuving, S., 1996.
Psychometric evaluation of the Dutch version of the Aggression
Questionnaire. Behaviour Research and Therapy 34, 839–843.
Moffitt, T.E., 1993. Adolescence-limited and life-course-persistent antisocial behavior: a developmental taxonomy. Psychological Review
100, 674–701.
Morren, M., Meesters, C., 2002. Validation of the Dutch version of the
Aggression Questionnaire in adolescent male offenders. Aggressive Behaviour 28, 87–96.
Moyer, K.E., 1968. Kinds of aggression and their physiological basis.
Communications in Behavioral Biology 2, 65–87.
Nakano, K., 2001. Psychometric evaluation on the Japanese adaptation
of the Aggression Questionnaire. Behaviour Research and Therapy
39, 853–858.
39
Olkin, I., 1967. Correlations revisited. In: Stanley, J.C. (Ed.), Improving
Experimental Design and Statistical Analysis. Rand McNally,
Chicago, pp. 102–128.
Olweus, D., 1979. Stability of aggressive reaction patterns in males: a
review. Psychological Bulletin 86, 852–875.
Pedhazur, E.J., Schmelkin, L.P., 1991. Measurement, Design, and
Analysis: An Integrated Approach. Lawrence Erlbaum, Hillsdale,
NJ.
Pulkkinen, L., Pitkanen, T., 1993. Continuities in aggressive behaviour from childhood to adulthood. Aggressive Behaviour 19,
249–263.
Stefanis, N.C., Smyrnis, N., Avrampoulos, D., Evdokimidis, I., Ntzoufras,
I., Stefanis, C.N., 2004. Factorial composition of self rated schizotypal
traits amongst young males undergoing military training. Schizophrenia Bulletin 30, 335–350.
Tsorbatzoudis, H., 2006. Psychometric evaluation of the Greek version
of the Aggression Questionnaire. Perceptual and Motor Skills 102,
703–718.
Van de Vijver, F., Hampleton, R.K., 1996. Translating tests: some
practical guidelines. European Psychologist 1, 89–99.
Williams, T.Y., Boyd, J.C., Cascardi, M.A., Poythress, N., 1996.
Factor structure and convergent validity of the Aggression Questionnaire in an offender population. Psychological Assessment 8,
398–403.
Young, S.E., Mikulich, S.K., Goodwin, M.B., Hardy, J., Martin, C.L.,
Zoccolillo, M.S., Crowley, T.J., 1995. Treated delinquent boys'
substance use: onset, pattern, relationship to conduct and mood
disorders. Drug and Alcohol Dependence 37, 149–162.