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

net/publication/278047438

The Brief Aggression Questionnaire: Structure, Validity, Reliability, and

Generalizability

Article  in  Journal of Personality Assessment · June 2015

DOI: 10.1080/00223891.2015.1044093 · Source: PubMed

CITATIONS READS
42 10,964

16 authors, including:

Gregory Webster Richard S Pond, Jr.

University of Florida University of North Carolina at Wilmington
97 PUBLICATIONS   6,358 CITATIONS    46 PUBLICATIONS   2,021 CITATIONS   

SEE PROFILE SEE PROFILE

Timothy Deckman Peter K Jonason

13 PUBLICATIONS   763 CITATIONS   
University of Padova
271 PUBLICATIONS   12,419 CITATIONS   
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Mating research View project

Studying daily social interaction View project

All content following this page was uploaded by Austin Lee Nichols on 21 September 2015.

The user has requested enhancement of the downloaded file.

 Journal of Personality Assessment, 0(0), 1–12, 2015
Copyright Ó Taylor & Francis Group, LLC
ISSN: 0022-3891 print / 1532-7752 online
DOI: 10.1080/00223891.2015.1044093

The Brief Aggression Questionnaire: Structure, Validity, Reliability,

and Generalizability
GREGORY D. WEBSTER,1 C. NATHAN DEWALL,2 RICHARD S. POND JR.,3 TIMOTHY DECKMAN,2 PETER K. JONASON,4
BONNIE M. LE,5 AUSTIN LEE NICHOLS,6 TATIANA OROZCO SCHEMBER,1 LAURA C. CRYSEL,7 BENJAMIN S. CROSIER,8
C. VERONICA SMITH,9 E. LAYNE PADDOCK,10 JOHN B. NEZLEK,11,12 LEE A. KIRKPATRICK,11 ANGELA D. BRYAN,13

J. BATOR14
AND RENEE

1
Department of Psychology, University of Florida
2
Department of Psychology, University of Kentucky
3
Department of Psychology, University of North Carolina at Wilmington
4
Department of Psychology, University of Western Sydney, Australia
5
Department of Psychology, University of Toronto, Canada
6
Department of Business, University of Navarra, Pamplona, Navarra, Spain
Downloaded by [University of Florida] at 15:28 09 June 2015

7
Department of Psychology, Stetson University
8
Center for Technology and Behavioral Health, Dartmouth College
9
Department of Psychology, University of Mississippi
10
Chair of Work and Organizational Psychology, ETH-Z€ urich, Switzerland
11
Department of Psychology, College of William and Mary
12
University of Social Sciences and Humanities, Pozna
n, Poland
13
Department of Psychology and Neuroscience, University of Colorado, Boulder
14
Department of Psychology, State University of New York at Plattsburgh

In contexts that increasingly demand brief self-report measures (e.g., experience sampling, longitudinal and field studies), researchers seek
succinct surveys that maintain reliability and validity. One such measure is the 12-item Brief Aggression Questionnaire (BAQ; Webster et al.,
2014), which uses 4 3-item subscales: Physical Aggression, Verbal Aggression, Anger, and Hostility. Although prior work suggests the BAQ’s
scores are reliable and valid, we addressed some lingering concerns. Across 3 studies (N D 1,279), we found that the BAQ had a 4-factor
structure, possessed long-term test–retest reliability across 12 weeks, predicted differences in behavioral aggression over time in a laboratory
experiment, generalized to a diverse nonstudent sample, and showed convergent validity with a displaced aggression measure. In addition, the
BAQ’s 3-item Anger subscale showed convergent validity with a trait anger measure. We discuss the BAQ’s potential reliability, validity,
limitations, and uses as an efficient measure of aggressive traits.

The reliability and validity of new measures must be tested situations where time permits, we also believe that researchers
rigorously and repeatedly if they are to be adopted by face an increasing demand for efficient measures such as the
researchers. The case for brief self-report measures of aggres- BAQ in specific settings that require them, including experi-
sion is no different. Webster et al. (2014) developed the 12- ence sampling studies, daily diary studies, prescreening or
item Brief Aggression Questionnaire (BAQ) as a more effi- mass-testing studies, longitudinal studies, field studies, and
cient alternative to the 29-item Aggression Questionnaire studies with special populations (see Widaman, Little,
(BPAQ; Buss & Perry, 1992). The BAQ uses the three highest Preacher, & Sawalani, 2011). In addition, brief measures can
loading items from each of the BPAQ’s four subscales: Physi- help reduce respondent fatigue and inattentiveness. Thus,
cal Aggression, Verbal Aggression, Anger, and Hostility. In when used in conjunction with several other long-format ques-
five studies (N
4,000), the BAQ was found to have (a) theo- tionnaires, the full 29-item BPAQ might add unnecessary
retically consistent patterns of convergent and discriminant items to a burgeoning item count that can become overly bur-
validity with other self-report measures, (b) a four-factor densome to respondents.
structure using multiple factor analyses, (c) adequate informa- Although there is a clear trade-off between reliability and
tion recovery using item response theory, (d) stable test–retest efficiency regarding the number of items per construct when
reliability across 3 weeks, and (e) convergent validity with creating brief measures, the BAQ uses three items per con-
behavioral measures of aggression (Webster et al., 2014). struct for three reasons. First, confirmatory factor analyses
Although we recommend using the 29-item BPAQ in (CFAs) and item response theory (IRT) analyses found that
the 12-item BAQ can efficiently recover test information
about four latent aggressive traits with only three items per
Received August 13, 2013; Revised February 21, 2015. construct (Webster et al., 2014; see also Bryant & Smith,
Address correspondence to Gregory D. Webster, Department of Psychol- 2001). Second, because the BAQ sought to preserve the
ogy, University of Florida, P.O. Box 112250, Gainesville, FL 32611-2250; BPAQ’s four-factor structure, including four or five items per
Email: gdwebs@ufl.edu construct would have needlessly ballooned the total number of

1
 2 WEBSTER ET AL.

items by a factor of four, thus defeating the purpose of creat- Pashler & Wagenmakers’s [2012] overview), we believe that
ing an efficient measure (i.e., 12 vs. 16 vs. 20 items out of replicating the BAQ’s reliability and factor structure while
29). Third, three items per construct are often a necessary min- addressing some of its lingering limitations is both necessary
imum for model identification and convergence when testing and important.
structural equation models (SEMs; Kline, 2011).
Despite these advances, the BAQ has a least four key limi-
tations. First, prior assessments of the BAQ’s structure have STUDY 1: FACTOR STRUCTURE AND TEST–RETEST
relied solely on principal axis factoring (PAF) and confirma- RELIABILITY
tory factor analysis (CFA; Webster et al., 2014) without first The goals of Study 1 were twofold. First, we aimed to repli-
presenting an exploratory factor analysis (EFA), which is cate and extend prior results regarding the BAQ’s four-factor
often an initial step in scale construction to determine factor structure (Webster et al., 2014). Whereas prior studies have
structure and assess item–factor pairings (Fabrigar, Wegner, relied on PAF and CFA, Study 1 focuses on EFA as a neces-
MacCallum, & Strahan, 1999). Consequently, we present the sary step in assessing structure in scale construction (Fabrigar
first EFA of the BAQ’s structure (Study 1). Second, because et al., 1999). We also used multiple criteria to establish the
the BAQ has shown acceptable test–retest reliability for only plausibility of a four-factor BAQ model. Second, we sought to
a short time interval (3 weeks; Webster et al., 2014), we extend the BAQ’s test–retest reliability. Establishing accept-
sought to address this concern by assessing the BAQ’s test– able test–retest reliability is essential to developing new or
retest reliability for a longer time interval (12 weeks; Study brief scales because trait-level individual differences should
1). Third, although the BAQ’s Physical Aggression subscale be relatively stable over time. We measured the BAQ at two
relates positively to behavioral aggression (noise blasts in an
Downloaded by [University of Florida] at 15:28 09 June 2015

time points 12 weeks apart, which allowed us to test longer

ostensibly competitive two-person game; Webster et al., term test–retest reliability. Although prior research established
2014), it remains unknown whether the BAQ relates to the the BAQ’s test–retest reliability across 3 weeks (Webster
time course of aggressive responding (noise blasts across 25 et al., 2014), initially promising results could be due to mem-
trials in the same game; Study 2). Specifically, we expect a ory biases or carryover effects characteristic of short time
Person (trait) £ Situation (aggressive retaliation over time) periods.
interaction. At Trial 1, the BAQ should positively predict
behavioral aggression (noise blasts) because the trait influence Method
should be strongest when situation is weak (retaliation from Measures. To test factor structure in Study 1, we aggre-
the participant’s ostensible partner has not yet occurred). By gated BAQ data from two independent samples to achieve a
Trial 25, the BAQ should less reliably predict behavioral sufficient sample size (Samples 1 and 2 described later). Spe-
aggression because the trait influence should become compar- cifically, we sought a > 20:1 cases-to-items ratio, which is
atively weaker over time as the situation (aggressive retalia- important for achieving stable estimates (e.g., Kline, 2013;
tion across trials) grows stronger. Fourth, although the BAQ but also see MacCallum, Widman, Zhang, & Hong, 1999). In
has shown acceptable psychometric properties in samples of both samples, participants responded to the 12 BAQ items
U.S. undergraduates (Webster et al., 2014), its generalizability using a 7-point scale ranging from 1 (extremely uncharacteris-
to more diverse, nonstudent samples remains unknown. In tic of me) to 7 (extremely characteristic of me).
Study 3, we address this limitation by surveying a large and
diverse international sample with a broader age range. In addi- Sample 1. Participants were 125 undergraduates (56 men,
tion, we strove to expand the nomological network of the 58 women, 11 did not report gender) enrolled in introductory
BAQ by examining its convergent and discriminant validity psychology courses at a public university in Virginia who
with trait anger and displaced aggression (Study 3). received course credit for their participation in an online ques-
Thus, whereas prior research established and justified item tionnaire (ages: 18–22 years, M D 19.10, SD D 1.22). Regard-
selection for the 12-item BAQ along with gender differences ing race and ethnicity, the sample was 77% White (non-
(Webster et al., 2014), this research focuses on addressing the Hispanic), 7% Asian American or Pacific Islander, 7% Black
limitations already listed and expanding the BAQ’s validity or African American, 3% Hispanic, and 4% other races or eth-
and generalizability. In addition, given psychological scien- nicities. BAQ descriptive statistics and correlations appear in
ce’s renewed emphasis on replication and reproducibility (see Table 1.

TABLE 1.—Brief Aggression Questionnaire (BAQ) descriptive statistics and zero-order correlations of observed scores for Study 1, Sample 1 (below diagonal)
and Study 2 (above diagonal).

Study 1 (N D 125) Zero-Order Correlations Study 2 (N D 307)

BAQ Measure M SD a 1 2 3 4 5 M SD a

1. Physical aggression 3.03 1.76 .84 — .43 .28 .27 .78 2.75 1.65 .83
2. Verbal aggression 3.84 1.31 .66 .54 — .31 .19 .69 3.56 1.24 .62
3. Anger 2.71 1.39 .81 .40 .51 — .36 .66 2.31 1.16 .67
4. Hostility 3.07 1.35 .74 .37 .35 .48 — .63 2.36 1.18 .65
5. BAQ mean 3.16 1.11 .86 .79 .78 .77 .70 — 2.74 0.91 .79

Note. All correlations significant at p < .01.

 THE BRIEF AGGRESSION QUESTIONNAIRE 3

Sample 2. Participants were a convenience sample of 140

undergraduates enrolled in psychology classes at a public uni-
versity in Florida. Each participant was asked to complete a
paper version of the 12-item BAQ in class twice—12 weeks
apart. We chose a 12-week interval for convenience because it
corresponded to the second and penultimate weeks in a semes-
ter, and because it was long enough to avoid possible carry-
over effects associated with prior testing. Of the 140 partici-
pants, 130 (93%) and 123 (88%) completed questionnaires
during Weeks 1 and 12, respectively; 113 (81%) participants
completed both sessions. We used the sample from Week 1
for the EFA because it was larger than the sample from Week
12. Of the 113 participants recruited for the test–retest reliabil-
ity analysis, 88 were women and 25 were men; ages ranged
from 18 to 29 years (M D 20.31, SD D 1.54). Race or ethnicity
information was not collected for this sample.

Results and Discussion FIGURE 1.—Scree plots of eigenvalues by number of factors: Observed and
two threshold criteria.
Factor structure. To assess factor structure while exceed-
Downloaded by [University of Florida] at 15:28 09 June 2015

ing a 20:1 cases-to-items ratio, we aggregated Sample 1

(n D 125) and Sample 2, Week 1 (n D 130) for the EFA Finally, we also used regression-based iterative outlier anal-
(N D 255). Using Mplus 6.1 (Muth
en & Muth
en, 2010), we yses to identify eigenvalues that departed significantly from
specified an EFA with up to four factors using the default obli- linearity in the scree plot. This involved a series of simple
que geomin rotation and full maximum likelihood estimation. regressions in which we regressed eigenvalues onto a number
The EFA procedure estimated models with one to four factors; of factors while using established methods to identify the larg-
fit indexes appear in Table 2. As expected, model fit improved est outlier (via Studentized deleted residual [SDR] and Cook’s
significantly with each additional factor (via Dx2), and only D; see Judd, McClelland, & Ryan, 2009, pp. 301–305),
the four-factor model yielded “good” fit indexes (i.e., compar- remove it, and then rerun the model without the largest outlier.
ative fit index [CFI] and Tucker–Lewis Index [TLI]  .95; The stopping rule was the absence of outliers (both SDR and
root mean square error of approximation [RMSEA] and stan- Cook’s D). After five iterations, outliers became absent,
dardized root mean square residual [SRMR]  .05). thereby suggesting a purely linear relationship between eigen-
We also assessed the appropriateness of the BAQ’s values and number of factors, and thus supporting a four-fac-
expected four-factor structure using multiple methods because tor model (Table 3; Figure 1). To summarize, a parallel
each one has strengths and weaknesses. First, as stated earlier, analysis supported a three-factor model, and, showing some
going from three to four factors produced fit indexes widely consensus, three methods—fit index thresholds, eigenvalues
considered to be in the acceptable range. Second, using a scree  1.0, and iterative outlier analyses—supported a four-factor
plot, the eigenvalues > 1.0 criterion (i.e., Kaiser–Guttman cri- model.
terion [Guttman, 1954; Kaiser, 1960]) also suggested a four- Table 4 shows the factor structure matrix from the four-fac-
factor solution (Figure 1). In contrast, parallel analysis (Horn, tor EFA. All items loaded at .50 or greater on their expected
1965) suggested a three-factor solution. Parallel analysis factors with one exception: “My friends say that I’m some-
assumes eigenvalues from a random matrix with the same what argumentative” loaded more strongly on Anger (.54)
number of items and sample size as the observed eigenvalues than its predicted factor, Verbal Aggression (.44). Thus, with
(see Hayton, Allen, & Scarpello, 2004). one exception, the BAQ items loaded on the four factors
related to their respective constructs: Physical Aggression,
TABLE 2.—Study 1: Exploratory factor analysis results for the Brief Aggres- Verbal Aggression, Anger, and Hostility. Nevertheless, our
sion Questionnaire. moderate sample size (N
250) might have contributed to
this discrepancy. We revisit and address this concern in Study
90% CI 3, where we use CFAs with larger samples (Ns > 500; see
Models or Differences x2
df CFI TLI RMSEA LL UL SRMR

1 factor 439.54 54 .656 .580 .167 .153 .182 .105 TABLE 3.—Study 1: Iterative outlier analyses of eigenvalues regressed on
2 factors 210.46 43 .851 .771 .124 .107 .141 .075 number of factors.
2 vs. 1 difference 229.08 11
3 Factors 121.42 33 .921 .842 .103 .083 .122 .052 Factor Studentized Deleted Residual (t) Cook’s D
3 vs. 2 difference 89.04 10
*
4 Factors 39.29 24 .986 .963 .050 ns
.018 .077 .021 1 14.83 2.01a
4 vs. 3 difference 82.13 9 2 2.13* 0.76
3 4.23* 1.52a
Note. N D 255. Fit indexes and suggested acceptable fit thresholds: Comparative fit 4 5.35* 1.76a
index (CFI) and Tucker–Lewis Index (TLI): > 90. Root mean square error of approxima- 5 0.19 0.02
tion (RMSEA) and standardized root mean square residual (SRMR): < .08. LL and UL
D lower and upper limits for RMSEA. All x2 and RMSEA statistics were significant at p a
Cook’s Ds  1.0 are considered outliers.
< .05 except ns. *p < .05, one-tailed.
 4 WEBSTER ET AL.

TABLE 4.—Study 1: Results of exploratory factor analysis (EFA) for four factors.

Factor
Brief Aggression Questionnaire Subscales and Items 1 2 3 4

Physical aggression
2. Given enough provocation, I may hit another person. .96 .42 .32 .35
5. If I have to resort to violence to protect my rights, I will. .76 .35 .30 .37
6. There are people who pushed me so far that we came to blows. .60 .40 .21 .46
Anger
4. I am an even-tempered person.a .11 .52 –.04 .02
6. Sometimes I fly off the handle for no good reason. .39 .86 .14 .47
7. I have trouble controlling my temper. .48 .90 .16 .45
Verbal aggression
1. I tell my friends openly when I disagree with them. .24 .06 .82 .05
3. When people annoy me, I may tell them what I think of them. .36 .37 .54 .24
5. My friends say that I’m somewhat argumentative. .42 .54 .44 .28
Hostility
3. Other people always seem to get the breaks. .25 .33 .04 .79
7. I sometimes feel that people are laughing at me behind my back. .28 .49 –.09 .54
8. When people are especially nice, I wonder what they want. .30 .33 .12 .60

Note. N D 255. Factor loadings > .50 are shown in bold. Oblique rotation was used.
Downloaded by [University of Florida] at 15:28 09 June 2015

a
Reverse-scored item.

also Webster et al., 2014) while also exploring the BAQ’s residuals for the same items at different time points, and
generalizability. examining the correlation associated with regressing Time 2’s
latent variable onto Time 1’s latent variable. We tested these
Test–retest reliability. Descriptive statistics and test– models as a series of CFAs in Mplus 6.1 (Muth
en & Muth
en,
retest correlations for the BAQ at both time points appear in 2010) using full maximum likelihood estimation. These CFAs
Table 5. As expected, test–retest reliability correlations were showed that the latent BAQ and its latent subscales had strong
strong and significant, ranging from .68 to .80 among the four and significant test–retest reliability correlations ranging from
subscales, and as high as .81 for the BAQ mean. Overall, these .83 to .90 (Table 6). In addition, the models fit the data well
findings show that the BAQ has good test–retest reliability (Table 6), except for the 12-item latent BAQ, which would
even over a longer time interval of 12 weeks, suggesting that typically be modeled with a four-factor approach (vs. a unidi-
it measures stable aggressive traits. mensional one) in most CFA or SEM contexts (see Study 3).
Because traditional test–retest correlations can confound Thus, modeling measurement error improved the temporal
temporal reliability with measurement error (Watson, 2004), reliability of the BAQ and its subscales.
we also examined latent test–retest correlations using an autor-
egressive latent-variable test–retest reliability model, which
separated measurement error from temporal reliability (see STUDY 2: PREDICTIVE VALIDITY WITH BEHAVIORAL
Khoo, West, Wu, & Kwok, 2006, pp. 305–307; see also Nich- AGGRESSION OVER TIME
ols & Webster, 2015). This involved modeling latent variables
for the 12-item BAQ for each of its four three-item subscales Having found some additional support for the BAQ’s four-
at Times 1 and 2, allowing for equal loadings and correlated factor structure and evidence of good test–retest reliability
(Study 1), we next addressed questions about the BAQ’s pre-
dictive validity regarding behavioral aggression over time. In
TABLE 5.—Study 1, Sample 2: Descriptive statistics and zero-order correla- Study 2, we reanalyzed data from Webster et al.’s (2014)
tions of observed scores for the Brief Aggression Questionnaire.

Descriptive Statistics Time 1 Time 2 TABLE 6.—Study 1: Latent test–retest reliabilities (rxx) and fit indexes based
M SD a 1 2 3 4 5 6 7 8 9 on confirmatory factor analyses for the Brief Aggression Questionnaire
(BAQ).
Time 1
1. Physical 2.14 1.17 .72 90% CI
2. Verbal 3.84 1.20 .66 .41
3. Anger 2.19 0.95 .74 .29 .19 Models rxx x 2
df CFI TLI RMSEA LL UL SRMR
4. Hostility 2.71 1.05 .57 .44 .04ns .27
Physical Aggression .88 5.44 7 1.0 1.0 .000 .000 .089 .034
5. Mean 2.72 0.74 .76 .81 .65 .61 .63
Verbal Aggression .90 5.47 7 1.0 1.0 .000 .000 .089 .038
Time 2
6. Physical 2.51 1.30 .76 .80 .51 .25 .25 .69 Anger .89 6.45 7 1.0 1.0 .000 .000 .099 .050
7. Verbal 3.76 1.19 .73 .33 .76 .19 –.06ns .48 .46 Hostility .83 11.17 7 .979 .954 .065 .000 .133 .054
8. Anger 2.32 0.95 .72 .37 .34 .74 .20 .59 .43 .35 BAQ (12 items) .87 591.38* 250 .757 .732 .099* .089 .109 .131
9. Hostility 3.04 1.10 .64 .20 –.05ns .28 .68 .39 .07ns –.03ns .30
Note. N D 140. Fit indexes and suggested acceptable fit thresholds. Comparative fit
10. Mean 2.90 0.76 .77 .66 .60 .51 .39 .81 .77 .68 .74 .47 index (CFI) and Tucker–Lewis Index (TLI): > 90. Root mean square error of approxima-
tion (RMSEA) and standardized root mean square residual (SRMR): < .08. LL and UL
Note. All correlations were significant at p < .05 except ns. N D 113. Twelve-week D lower and upper limits for RMSEA.
test–retest reliability correlations are shown in bold. *p < .05.
 THE BRIEF AGGRESSION QUESTIONNAIRE 5

Study 5, in which participants completed the 29-item BPAQ Multilevel modeling. We used the multilevel modeling
before taking part in a 25-trial competitive reaction-time com- program Hierarchical Linear Modeling 6.0 (HLM; Rauden-
puter game, in which the winners of each trial (randomly bush, Bryk, Cheong, & Congdon, 2006) because multiple tri-
assigned to win 13 of 25 trials) got to blast the losers (ostensi- als (25) were nested within each participant. Using HLM’s
bly in an adjacent room) with bursts of white noise. Although default restricted maximum likelihood estimation and robust
Webster et al.’s Study 5 showed that the BAQ’s Physical standard errors, multilevel modeling allows for the simulta-
Aggression subscale performed as well as or slightly better neous modeling of within- and between-person effects
than that of the BPAQ, change over time—or across trials— (Nezlek, 2008, 2011; Raudenbush & Bryk, 2002). Specifi-
was not examined. In Study 2, we predicted a Time £ BAQ cally, within-person (or between-trial) variance in noise blast
interaction for noise blast duration and intensity: Trait BAQ duration or intensity was modeled at Level 1, and between-
scores should positively predict aggressive behavior more person variance in noise blast duration or intensity was mod-
strongly at Trial 1 than at Trial 25. In other words, we pre- eled at Level 2 as a function of individual differences in the
dicted a classic Person £ Situation interaction (Funder, 2008; BAQ. Specifically, Level 1 of our multilevel was
Krueger, 2009; Swann & Seyle, 2005; Webster, 2009): Trait
effects will be strongest when the situation is weak (Trial 1), Duration or Intensityti D p0i C p1i ðTrial ¡ 13Þi C eti ;
but weaken as the situation grows stronger (i.e., following per-
ceived iterative aggressive retaliation; Trial 25). Decomposing
this same interaction from another angle (see Aiken & West, where Duration or Intensityti represents the noise blast dura-
1991), we also predicted that people with high trait aggression tion or intensity (separate models) given out at time t by indi-
(1 SD above the BAQ mean) would continue to respond
Downloaded by [University of Florida] at 15:28 09 June 2015

vidual i. Each person’s duration or intensity scores are

aggressively regardless of retaliation over trials (i.e., stronger modeled by p0i, which represents the mean or intercept for
trait ! less situational influence), whereas people with low each person at the midpoint of the 25 trials (Trial 13), and
trait aggression (1 SD below the BAQ mean) would increase p1i(Trial – 13)i, which represents the change-over-time (or
their aggressive responding because of retaliation over trials
trial) slope for each person. The error term, eti, represents the
(i.e., weaker trait ! more situational influence). We remained
agnostic, however, as to whether this interaction would be (a) residual Level-1 variance.
stronger for noise blast duration or intensity, or (b) driven by We modeled the Level-1 random-effects intercepts and
any particular BAQ subscale. slopes for each person simultaneously at Level 2 as a function
of their respective BAQ score (grand-mean-centered):

Method
p0i D b00 C b01 ðBAQ ¡ meanÞ C r0i ;
Participants. Participants were 307 undergraduates
enrolled in introductory psychology courses at a public univer- p1i D b10 C b11 ðBAQ ¡ meanÞ C r1i :
sity in Florida who received course credit in exchange for their
participation (91 men, 216 women; ages: 18–41 years, M D
19.34, SD D 2.27). Here, p0i again represents the mean or intercept for each per-
son. The b00 coefficient represents the grand mean—the
between-person average of each person’s average duration or
Measures. Participants completed the 29-item BPAQ intensity score (at Trial 13 and at the mean BAQ score). The
using a 7-point response scale ranging from 1 (extremely b01(BAQ – mean) coefficient represents the moderating effect
uncharacteristic of me) to 7 (extremely characteristic of me).
of individual differences in the BAQ on people’s overall mean
duration or intensity score. In contrast, p1i represents each per-
Procedure. The procedure was a modified Taylor (1967) son’s duration- or intensity-over-time slope, and the b10 coef-
aggression paradigm, where participants were ostensibly ficient represents the average of these respective slopes (at the
paired in adjacent rooms and competed against each other to mean BAQ score). The focal effect, which is the b11(BAQ –
be the quickest responder on each reaction-time trial; the win- mean) coefficient, represents the moderating effect of the
ner of each trial could deliver a white noise blast to his or her
BAQ on people’s change-over-time slopes in duration or
partner. In reality, participants completed their reaction-time
trials against a computer program set to mimic another per- intensity. The error terms, r0i and r1i, represent the residual
son’s actions. Participants controlled the duration (0.0–5.0 Level-2 variances in people’s intercepts and slopes,
sec) and intensity (0–105 dB) of the noise (about the volume respectively.
of a smoke alarm) corresponding to response scales ranging
from 0 (low) to 10 (high), which was consistent with similar Results and Discussion
research using this paradigm with 10- or 11-point response See Table 1 for BAQ mean and subscale correlations and
scales (e.g., DeWall, Bushman, Giancola, & Webster, 2010; descriptive statistics. Average noise-blast duration was related
see Webster et al., 2014). Participants’ noise-blast duration to the BAQ, b01 D 0.46, t(305) D 3.81, p < .001, rp D .21 (see
(M D 4.92, SD D 2.20) and intensity (M D 5.18, SD D 2.26) also Webster et al., 2014). Regarding the focal analyses,
scores were positively correlated (r D .86, p < .001). This lab- noise-blast duration increased over time for the average partic-
oratory procedure provides a valid and established measure of ipant, b10 D 0.041, t(305) D 5.19, p < .001, rp D .28; how-
behavioral aggression (e.g., Anderson & Bushman, 1997; ever, the change-over-time slopes were unrelated to BAQ
Giancola & Chermack, 1998). scores, b11D –0.0016, t(305) D –0.16, p D .87, rp D –.01.
 6 WEBSTER ET AL.

BAQ scores were more susceptible to the demands of the

situation, starting with lower intensity noise blasts, but
then increasing them over 25 trials. Whereas prior research
has examined the BAQ’s association with mean levels of
noise blast duration and intensity (averaged across 25 tri-
als; see Webster et al., 2014), this research extends these
findings to show that the BAQ—but not the BPAQ—pre-
dicted initial differences in behavioral aggression (noise
blast intensity) and its increase over time across 25 trials
in a controlled laboratory experiment.

STUDY 3: EXPANDED GENERALIZABILITY

AND CONVERGENT VALIDITY
Although the BAQ’s reliability and validity have received
some support here (Studies 1 & 2) and elsewhere (Webster
et al., 2014), it has yet to be determined the extent to which its
psychometric properties generalize to more diverse, nonstu-
dent samples. Specifically, nearly all participants to this point
Downloaded by [University of Florida] at 15:28 09 June 2015

FIGURE 2.—Study 2: Noise-blast intensity as a function of trial (1–25) and have been U.S. undergraduates, or more generally, “W.E.I.R.
mean score on the 12-item Brief Aggression Questionnaire (BAQ; Low D –1 D.” people (i.e., Western, educated, industrialized, rich, and
SD, High D C1 SD). democratic; Henrich, Heine, & Norenzayan, 2010). To address
this and other concerns, we conducted Study 3 with three goals
in mind.
Average noise-blast intensity was related to the BAQ mean, First, we tested a series of CFAs to compare the fit of the
b01 D 0.53, t(305) D 4.35, p < .001, rp D .24 (see also Web- BAQ’s predicted four-factor model to a hierarchical model
ster et al., 2014). Regarding the focal analyses, noise-blast (four latent BAQ subscales load onto a second-order latent
intensity increased over time for the average participant, aggression factor) and to a unidimensional model (12 items
b10 D 0.031, t(305) D 3.88, p < .001, rp D .22; and the load directly onto one latent aggression factor). Because the
change-over-time slopes were negatively related to BAQ BAQ was developed to optimize subscale items (vs. mean
scores, b11D –0.018, t(305) D –1.96, p D .050, rp D –.11 score), we expected the unidimensional model to fit the data
(Figure 2). In addition, controlling for gender reduced this worse than other models.
effect only slightly, b11D –0.017, t(304) D –1.86, p D .063, Second, because we believe that the four-factor model will
rp D –.11; and replacing the 12-item BAQ with the 29-item be widely adopted, we tested the extent to which gender (men
BPAQ showed that the full measure did not significantly mod- vs. women), first-language (English vs. other), and sample
erate the change-over-time slopes, b11D –0.015, t(305) D (student vs. non-student) differences moderated the item fac-
–1.64, p D .101, rp D –.09. We also ran a follow-up model tor loadings (metric invariance) and intercepts (scalar invari-
that replaced the BAQ mean with its four subscales; none was ance). We did this because (a) men tend to report and enact
a significant predictor of noise blast intensity change (ps > more aggression—particularly unprovoked physical aggres-
.08, |rp|s < .10). Simple effects tests (see Aiken & West, sion—than women (Archer, 2004; Bettencourt & Miller,
1991) on the cross-level interaction in Figure 2 showed the 1996; Eagly & Steffen, 1986), and (b) prior research using the
predicted effect: At Trial 1, noise-blast intensity was posi- BAQ had U.S. undergraduate participants whose first lan-
tively and significantly related to the BAQ mean, b01 D 0.74, guage was English. Nevertheless, we expected the BAQ to
t(305) D 4.62, p < .001, rp D .26; but by Trial 25, this was no show partial metric and scalar invariance across these three
longer the case, b01 D 0.32, t(305) D 1.94, p D .053, rp D .11. group comparisons.
On an exploratory basis, we also examined the other two sim- Third, we also sought to expand the nomological network of
ple effects: Noise-blast intensity increased significantly over the BAQ and its subscales by examining its correlations with
time for participants scoring 1 SD below the mean, b10 D trait anger and displaced aggression. Trait anger is important
0.047, t(305) D 4.08, p < .001, rp D .23, but the same increase to understanding the BAQ’s Anger subscale, especially
was not significant for participants scoring 1 SD above the because prior research (Webster et al., 2014) stressed validat-
mean, b10 D 0.015, t(305) D 1.34, p D .18, rp D .08. ing the BAQ’s Physical Aggression subscale. Specifically, we
Thus, supporting our prediction and showing a classic expected the Trait Anger Scale (described later) would corre-
Person £ Situation interaction, the BAQ mean was sensi- late more highly with the BAQ’s Anger subscale than its other
tive to trait differences in aggression in predicting initial three subscales. We also measured a brief version of the three-
noise-blast intensity (but not duration), and this relation- factor Displaced Aggression Questionnaire (DAQ; described
ship waned across time (repeated trials) as people became later). Testing relationships among the BAQ and DAQ (and
immersed in a strong, competitive situation. From another their subscales) is important because doing so would show the
perspective, people with high BAQ scores maintained their first evidence of convergent validity between the two multifac-
aggressive behavior (via high-intensity noise blasts) eted trait aggression measures. Although we expected positive
unabated throughout the 25 trials, whereas people with low correlations among the BAQ and DAQ and their subscales,
 THE BRIEF AGGRESSION QUESTIONNAIRE 7

we remained agnostic regarding the precise pattern of correla- Displaced Aggression Questionnaire: Participants also
tions among subscales. completed an abbreviated, nine-item measure of displaced
aggression based on items drawn from the DAQ (Denson,
Method Pedersen, & Miller, 2006). Just as Webster et al. (2014) chose
Participants and procedure. Participants were 611 peo- the highest loading items from the BPAQ’s subscales to make
ple recruited using Amazon’s Mechanical Turk (MTurk; see the BAQ, we chose the three highest loading items (reported
Buhrmester, Kwang, & Gosling, 2011; Goodman, Cryder, & in Denson et al., 2006) from each of the DAQ’s three sub-
Cheema, 2013; Hauser & Schwarz, in press). MTurk is a mar- scales—Angry Rumination, Revenge Planning, and Displaced
ketplace for work that brings together researchers and willing Aggression—to make an abbreviated measure. The DAQ’s
participants who wish to be paid for research (e.g., completing three factor-based subscales measure theoretically informative
surveys, computerized experiments). MTurk has been widely dimensions of displaced aggression: affective (Angry Rumina-
adopted by psychology departments and business schools tion), cognitive (Revenge Planning), and behavioral (Dis-
throughout the world. For example, by 2010, 16 of the top 30 placed Aggression). We chose to measure displaced
of business schools in the United States (53%) were using aggression because it should be positively correlated with trait
MTurk to collect research data (Goodman et al., 2013). Many aggression as reflected in the BAQ; however, we remained
experiments pay participants between 5 cents and 50 cents. In agnostic regarding relationships among specific subscales.
Study 3, we paid participants 25 cents for completing an
online survey. To see our survey, we required MTurk partici- Results and Discussion
pants to have human intelligence task (HIT) approval rates 
Confirmatory factor analyses. We first ran a series of
Downloaded by [University of Florida] at 15:28 09 June 2015

95% based on their prior performance on other tasks or studies

(i.e., in previous tasks or studies, they completed the assigned CFAs in Mplus 6.1 (Muth
en & Muth
en, 2010) using full maxi-
task or study to the satisfaction of the researcher or requester); mum likelihood estimation to test the factor structure of the
this assured some quality control. Indeed, MTurk participants BAQ using our nonstudent sample. The fit statistics for all
with such HIT rates often outperform traditional participant three measurement models (CFAs) appear in Table 7. The
pools in survey attentiveness (Hauser & Schwarz, in press). four-factor model fit the data well and better than each of the
For additional information on MTurk and the demographics of other two measurement models. Thus, comparatively simpler
its users, see Buhrmester et al. (2011).
Using three items designed to detect inattentive participants TABLE 7.—Study 3: Confirmatory factor analysis results for the Brief Aggres-
(e.g., “Please select option 2 for this item”) interspersed sion Questionnaire.
throughout the online survey, we identified and excluded 90% CI
seven participants with inattentive response patterns. This left
a sample of 604 people (336 men, 260 women, 4 transgender, Models or Differences x2 df CFI TLI RMSEA LL UL SRMR
4 did not report gender; ages: 18–79 years, M D 29.53, SD D Measurement models
10.03), representing more than 40 countries and 40 languages. 1. Four-factor 131.11 48 .959 .943 .054ns .043 .065 .050
The five most frequent countries were India (48.7%), the 2. Hierarchical 146.93 50 .952 .936 .057ns .046 .067 .052
United States (31.3%), Canada (4.0%), Pakistan (2.6%), and 2 vs. 1 difference 15.82 2
3. Unidimensional 617.23 54 .720 .657 .132 .122 .141 .086
the Philippines (1.3%); all other countries represented less 3 vs. 2 difference 470.30 4
than 1.0% of the sample. Although all participants could read Four-factor: Gender
a
English, 48% of them reported English as their first language. 1. Configural invariance 202.34 96 .945 .924 .061ns .049 .073 .056
Ethnically, the sample was 94.0% non-Hispanic; racially, the 2. Full metric invarianceb 210.49 104 .945 .930 .059ns .047 .070 .059
2 vs. 1 difference 8.15ns 8
sample was 48.8% Indian, 36.6% White, 4.3% unspecified 3. Full scalar invariancec 219.39 112 .945 .935 .057ns .046 .068 .061
Asian, 2.3% Hispanic, 2.2% Black, 2.2% Filipino, 1.7% 3 vs. 2 difference 8.90ns 8
unspecified, 1.0% Chinese, 0.5% American Indian, and 0.5% Four-factor: First language
Japanese. Regarding education, 0.2% completed only primary 1. Configural invariancea 205.33 96 .945 .924 .062 .050 .073 .058
or grammar school, 7.4% completed secondary or high school, 2. Full metric invarianceb 216.35 104 .943 .928 .060ns .049 .071 .062
2 vs. 1 difference 11.02ns 8
23.6% had some college or university; and 5.7%, 41.8%, 3. Full scalar invariancec 248.28 112 .931 .919 .064 .053 .075 .065
17.4%, and 3.8% had earned an associate’s, bachelor’s, mas- 3 vs. 2 difference 31.39 8
ter’s, or doctoral or professional degree, respectively. Partici- 4. Partial scalar invariance 227.57 110 .940 .928 .060ns .049 .071 .065
ns
pants completed the online survey and demographic questions 4 vs. 2 difference 11.22 6
Four-factor: Sampled
in a timely fashion, averaging 8.5 min (
10 sec per item). 1. Configural invariancea 214.55 96 .934 .909 .069 .057 .081 .060
2. Full metric invarianceb 222.96 104 .934 .916 .066 .054 .078 .062
Measures. 2 vs. 1 difference 8.41ns 8
3. Full scalar invariancec 261.82 112 .917 .902 .072 .060 .083 .066
Brief Aggression Questionnaire: Participants completed 3 vs. 2 difference 38.86 8
the 12-item BAQ using a 7-point response scale ranging from 4. Partial scalar invariance 233.99 109 .931 .916 .066 .055 .078 .062
ns
1 (extremely uncharacteristic of me) to 7 (extremely charac- 4 vs. 2 difference 11.03 5
teristic of me). Note. Fit indexes and suggested acceptable fit thresholds: Comparative fit index (CFI)
Trait Anger Scale: Participants also provided data on the and Tucker–Lewis Index (TLI): > 90. Root mean square error of approximation
15-item Trait Anger Scale (Spielberger, Jacobs, Russell, & (RMSEA) and standardized root mean square residual (SRMR): < .08. LL and UL D
lower and upper limits for RMSEA. All x2 and RMSEA statistics were significant at
Crane, 1983; e.g., “I have a fiery temper”) using a 4-point p < .05 except ns.
scale ranging from 1 (almost never) to 4 (almost always). a
Equal form. bEqual factor loadings. cEqual intercepts. N D 603 except dN D 520.
 8 WEBSTER ET AL.

models significantly worsened the fit. In broad terms of abso- the prior analyses for first language, results indicated that the
lute goodness of fit, the four-factor and hierarchical models BAQ items showed full metric invariance, but not full scalar
each showed good fit, whereas the unidimensional model invariance (equal factor loadings, but not equal intercepts;
showed poor fit. In terms of comparative fit, however, and rep- Table 7). To this end, we again examined modification
licating Webster et al.’s (2014) results, the four-factor model indexes iteratively, allowing the item with the highest inter-
fit the data better than the hierarchical model (i.e., Dx2), which cept difference across groups to vary freely. After freeing the
fit better than the one-factor model. intercepts for three items, partial scalar invariance was
Using multiple-group CFAs (Brown, 2006; Kline, 2011), achieved (equal intercepts for nine items). Specifically, non-
we next tested for metric (item loadings) and scalar (item students (vs. students) scored higher on the items “There are
intercepts) invariance in the BAQ’s four-factor model for people who pushed me so far that we came to blows”; “When
three grouping variables: gender (men vs. women), first lan- people annoy me, I may tell them what I think of them”; and
guage (English vs. non-English), and sample (student vs. non- “I am an even-tempered person” (prior to reverse-scoring).
student; Table 7). We first tested a fully unconstrained Collectively, these CFAs suggest that the BAQ’s four-factor
(configural invariance or equal form) model that freed all structure holds not only for U.S. undergraduates (see Webster
parameters to differ by grouping variable, and then tested et al., 2014), but also for a more diverse sample of nonstu-
models that constrained the item loadings to be the same for dents. The four-factor model again produced the best fitting
both groups (metric invariance or equal factor loadings), fol- solution. Showing full metric invariance (equal factor load-
lowed by also constraining the item intercepts to be the same ings), factor loadings as a set did not vary significantly
for both groups (scalar invariance or equal intercepts); we between men and women, between English- and non-English-
fixed the factor variances at 1.0 for the reference groups in the speaking people, or between North American students and
Downloaded by [University of Florida] at 15:28 09 June 2015

metric and scalar invariance models (see Brown, 2006, pp. nonstudents. Although we showed full scalar invariance (equal
236–304). For gender, tests comparing these three models intercepts) for gender, we only showed partial scalar invari-
showed no significant differences (Table 7), suggesting both ance for first language and sample after freeing a few item
full metric and scalar invariance (equal factor loadings and intercepts. Thus, the BAQ showed evidence of measurement
intercepts). consistency across three broad categorical attributes: gender,
For first language, we grouped the sample into respondents English as a first language, and student versus nonstudent.
who answered “Yes” to the question, “Is English your first Although more diverse than most U.S. student samples,
language?” (48%) and those who answered “No.” Results MTurk participants still represent a select group of English-
indicated that the BAQ items showed full metric invariance speaking people with Internet access. Thus, our findings are a
for first language, but not full scalar invariance (equal factor preliminary—and necessary—step toward broadening the
loadings, but not equal intercepts; Table 7). To address this BAQ’s generalizability.
concern, we examined modification indexes iteratively, allow-
ing the item with the highest intercept difference across groups Convergent validity. Having established some degree of
to vary freely. After freeing the intercepts for two items, par- measurement invariance in our nonstudent sample, we next
tial scalar invariance was achieved (equal intercepts for 10 created mean scores by averaging across items for each scale
items). Specifically, people whose first language was English or subscale. Descriptive statistics and correlations among
scored higher on “Other people seem to get the breaks,” but measures are shown in Table 8. One purpose of Study 3 was
people whose first language was not English scored higher on to test the convergent validity of the BAQ’s Anger subscale
“I have trouble controlling my temper.” with that of an established anger measure—the Trait Anger
To compare student with nonstudent samples, we created a Scale. Indeed, this pair of anger measures had the highest cor-
new data set with the 213 U.S. and Canadian MTurk partici- relation (r D .58), which was significantly higher than the cor-
pants and the 307 students from Study 2 (N D 520). Similar to relations between the Trait Anger Scale and each of the other

TABLE 8.—Study 3: Descriptive statistics and zero-order correlations of observed scores for the Brief Aggression Questionnaire (BAQ), the Abbreviated
Displaced Aggression Questionnaire (DAQ), and the Trait Anger Scale.

Descriptive Statistics Zero-Order Correlations

Variable M SD a 1 2 3 4 5 6 7 8 9

BAQ
1. Anger 3.11 1.23 .67
2. Physical Aggression 2.75 1.40 .82 .50
3. Hostility 3.67 1.25 .65 .35 .35
4. Verbal Aggression 4.15 1.24 .66 .25 .38 .19
5. BAQ Mean 3.42 0.91 .81 .73 .80 .66 .63
DAQ
6. Angry Rumination 3.96 1.54 .88 .35 .29 .49 .18 .46
7. Revenge Planning 3.21 1.54 .87 .56 .53 .47 .32 .67 .53
8. Displaced Aggression 2.84 1.42 .86 .49 .32 .35 .17 .47 .32 .39
9. DAQ Mean 3.34 1.17 .87 .60 .49 .56 .29 .68 .80 .83 .71
Trait Anger Scale 2.05 0.47 .86 .58 .51 .45 .33 .66 .46 .55 .48 .64

Note. Ns D 602–604. All ps < .001.

 THE BRIEF AGGRESSION QUESTIONNAIRE 9

three BAQ subscales (zs  2.16, ps  .03; see Lee & Preacher, Studies 2 and 3 also expanded the BAQ’s nomological net-
2013). After correcting for attenuation (unreliability) in both work by showing some theoretically consistent patterns of
measures—by dividing the correlation by the square root of convergent validity with behavioral aggression over time (tri-
the product of each measure’s reliability coefficient (a; Spear- als), trait anger, and displaced aggression. Specifically, these
man, 1904)—this correlation increased (r’ D .76). Thus, the findings improve on prior BAQ research (Webster et al.,
BAQ’s three-item Anger subscale explained 34% (or 58% 2014), which reported evidence for convergent validity among
after correcting for attenuation) of the variance in the 15-item the BAQ, its precursors, and behavioral aggression averaged
Trait Anger Scale (or vice versa). across time (trials). In conjunction with prior work (Webster
As an exploratory exercise, we also measured brief, three- et al., 2014; see also Jonason & Webster, 2010; Webster,
item versions of the DAQ’s three subscales. As expected, 2006, 2007; Webster & Bryan, 2007; Webster & Crysel,
and showing some convergent validity, each subscale was 2012; Webster, Kirkpatrick, Nezlek, Smith, & Paddock,
positively and significantly correlated with each of the four 2007), these studies add to a growing literature of evidence
BAQ subscales; however, there was some substantial vari- supporting the BAQ as a psychometrically robust measure of
ability in the strength of these correlations. Specifically, large individual differences in trait aggression.
correlations (rs
.50) linked (a) Anger with Revenge Plan-
ning and Displaced Aggression, (b) Physical Aggression with
Revenge Planning, and (c) Hostility with Angry Rumination Limitations and Directions for Future Research
and Revenge Planning. Moderate correlations (rs
.30) Our results provide new and converging evidence support-
linked (a) Anger with Angry Rumination, (b) Physical ing the structure, validity, reliability, and generalizability of
Aggression with Angry Rumination and Displaced Aggres- BAQ scores that was absent from Webster et al.’s (2014)
Downloaded by [University of Florida] at 15:28 09 June 2015

sion, (c) Hostility with Displaced Aggression, and (d) Verbal development of the BAQ. Despite the consistency of these
Aggression with Revenge Planning. Small-to-moderate corre- findings, at least seven limitations remain that might serve to
lations (rs
.20) linked Verbal Aggression with Angry inspire future research.
Rumination and Displaced Aggression. Of the four BAQ sub- First, regarding item selection, although the BAQ maxi-
scales, Anger was most closely related to the three DAQ sub- mized within-factor loadings (Webster et al., 2014), it largely
scales (rs D .35–.56), suggesting that this affective ignored cross-factor loadings, some of which were nontrivial
component of aggression might be more closely linked to dis- (Table 4). These larger-than-desired cross-loadings pose a
placed aggression than either the cognitive (Hostility) or concern to both the convergent and discriminant validity of
behavioral (Physical Aggression and Verbal Aggression) the BAQ’s subscales. For instance, in latent-variable models
forms of aggression. This analysis also suggests that, similar (e.g., CFAs, SEMs) where the cross-loadings are set to zero,
to the BAQ, the DAQ can be measured using three-item the BAQ’s interfactor correlations could be unexpectedly
scales, although further testing will be necessary. high, leading to possible reductions in construct validity.
Thus, researchers should be aware of this fact, especially
when using the BAQ’s subscales in latent-variable contexts.
GENERAL DISCUSSION Second, because we sought to create and validate three-item
Self-report measures of aggression are necessary for assess- measures, construct underrepresentation is a concern. As
ing individual differences in aggressive traits. Short-form noted earlier, creating brief measures requires a trade-off
measures are increasingly in demand (Widaman et al., 2011), between efficiency (number of items) and breadth (represent-
and aggression researchers need brief self-report scales to mea- ing a wide swath of the construct). Because we chose to
sure aggressive traits in contexts that place premiums on time emphasize brevity, and hence efficiency, the BAQ’s subscales
or space (e.g., daily diary studies, field studies, special popula- likely lack the breadth of their parent versions in the BPAQ.
tions, mass testing or prescreening questionnaire packets). The Nevertheless, because the BAQ continues to measure all four
12-item BAQ fulfills this need by providing an efficient mea- of the BPAQ’s original factor-based subscales, the breadth of
sure of anger, hostility, and verbal and physical aggression. the mean BAQ score adequately represents that of its parent
Three studies, including 1,279 participants, offered some measure, the BPAQ (rs D .96; Webster et al. 2014).
converging evidence supporting the structure, validity, reli- Third, although Study 1 supported BAQ’s four-factor struc-
ability, and generalizability of the BAQ’s scores. Study 1 ture across three methods—fit index thresholds, eigenvalues
showed some consistency in the BAQ’s structure as a four-  1.0, and iterative outlier analyses—Study 1 also supported a
factor aggression measure; however, one method—parallel three-factor solution using parallel analysis. Thus, although
analysis—supported a three-factor solution. Study 1 also there was some consensus, some room for debate continues on
showed that the BAQ had acceptable 12-week test–retest reli- the BAQ’s factor structure, owing to its multifaceted nature as
ability, using both traditional and latent-variable methods. either a global aggression measure or one with four subscales.
Using a behavioral aggression measure (noise blasts), Study 2 Study 3’s results also supported the BAQ’s four-factor struc-
showed that the BAQ interacted with time (trial) in a retalia- ture, and although it fit the data better than a hierarchical fac-
tory aggression experiment, such that BAQ mean scores posi- tor model (via Dx2), the fit of both models was acceptable (via
tively predicted initial aggression (noise-blast intensity) more other fit indexes). Again, the use of the BAQ as either a global
strongly at Trial 1 than at Trial 25. Study 3 confirmed the or four-subscale measure appears justifiable; however, further
BAQ’s four-factor structure and highlighted its partial metric research might be needed to resolve some of the BAQ’s struc-
and scalar invariance (equal factor loadings and intercepts) tural ambiguity.
across gender, English as a first language, and student versus Fourth, because Cronbach’s alpha (a) is positively corre-
nonstudent groups using a large, diverse sample. lated with number of scale items (holding mean interitem
 10 WEBSTER ET AL.

correlation [MIC] constant; see Cortina, 1993, p. 101), we constraints are placed on participants, or when space con-
expected the BAQ’s subscales to have acceptable—but not straints are placed on researchers (e.g., number of items
excellent—internal consistency. This research produced 16 allowed), abbreviated measures can and should be used.
BAQ subscale as, and all were acceptable (> .50; see Schmitt, Although the number of scale items contributes to internal
1996) for three-item measures, and these corresponded with consistency reliability (i.e., Cronbach’s a), the main contribu-
respectable MICs (> .30; see Cortina, 1993). Thus, research- tor is the average interitem correlation. If the average interi-
ers should be aware of the internal consistency trade-off that tem correlation is reasonable, and if the items used are not
accompanies the increased efficiency of the BAQ’s three-item redundant, then brief unidimensional scales are possible.
subscales. Indeed, researchers might wish to consider account- Although longer, original scales are generally preferable when
ing for measurement error in the BAQ’s three-item subscales time and space are not concerns, abbreviated versions of those
by disattenuating correlations (as we did for the Trait Anger scales can be used to measure the constructs of interest. Spe-
Scale) or by using latent-variable models (e.g., SEM). cifically, we recommend that aggression researchers use the
Fifth, although we sought to support the BAQ’s validity and original 29-item BPAQ when time and space allows; however,
reliability in our studies, we focused on test–retest reliability given constrained time and space, we recommend they use the
(Study 1), and on predictive and convergent validity (Studies 12-item BAQ. Overall, our findings add to a growing literature
2 & 3). Although the BAQ’s test–retest reliability is good that emphasizes the development of concise, efficient, self-
(Study 1; Webster et al., 2014), its internal consistency reli- report measures of psychological constructs, with the goal of
ability (a) is acceptable at best (as noted earlier). Although the meeting researchers’ contemporary needs (e.g., Ames, Rose,
BAQ showed evidence of predictive validity with noise-blast & Anderson, 2005; Bryant & Smith, 2001; Donnellan,
intensity (Study 2; and duration in Webster et al., 2014, Study Oswald, Baird, & Lucas, 2009; Fraley, Waller, & Brennan,
Downloaded by [University of Florida] at 15:28 09 June 2015

5), effect sizes were small, and generalizability to other behav- 2000; Gosling, Rentfrow, & Swann, 2003; Jonason & Web-
ioral aggression measures remains untested. In addition, ster, 2010; Nichols & Webster, 2013, 2014, 2015; Rammstedt
although we expanded the BAQ’s nomological network, thor- & John, 2007; Robins, Hendin, & Trzesniewski, 2001; Web-
ough convergent (and discriminant) validity tests for its Hos- ster & Crysel, 2012; Webster et al., 2014; Webster & Jonason,
tility and Verbal Aggression subscales are lacking. Future 2013; Widaman et al., 2011; Wood, Nye, & Saucier, 2010).
research could also examine other forms of validity not tested
here (e.g., diagnostic, ecological). For example, peer assess-
ments of trait aggression using the BAQ might be especially ACKNOWLEDGMENTS
informative (in conjunction with self-reports).
Sixth, although we assessed the BAQ in both student (Stud- Different analyses of some of the data included in Study 2
ies 1 & 2) and nonstudent (Study 3) samples, we did not sam- of this article have been published (Webster et al., 2014,
ple any special populations (e.g., children, at-risk youth, Study 5). Aside from parts of the Study 2 Method section, and
prisoners, clinical samples) in which the BAQ might be used some zero-order correlations and descriptive statistics (Ms,
in the future. Thus, although the BAQ showed some conver- SDs, as), these results do not reproduce previously published
gence between North American student and nonstudent sam- findings.
ples, we do not know the extent to which the BAQ’s
measurement properties also extend to various special popula-
tions. In addition, although Study 3 expanded the BAQ’s gen- REFERENCES
eralizability to a more diverse, nonstudent sample, the extent Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpret-
to which it generalizes to other nations, cultures, and age ing interactions. Thousand Oaks, CA: Sage.
groups remains unclear because we did not have the broad Ames, D. R., Rose, P., & Anderson, C. P. (2005). The NPI–16 as a short mea-
sample sizes necessary to adequately test these possibilities. sure of narcissism. Journal of Research in Personality, 40, 440–450.
Thus, further research on a global scale might be necessary for doi:10.1016/j.jrp.2005.03.002
establishing the BAQ’s international, cross-cultural, and gen- Anderson, C. A., & Bushman, B. J. (1997). External validity of “trivial”
erational generalizability. experiments: The case of laboratory aggression. Review of General Psy-
chology, 1, 19–41. doi:10.1037/1089-2680.1.1.19
Seventh, whether the BAQ—and particularly the Physical Archer, J. (2004). Sex differences in aggression in real-world settings: A meta-
Aggression subscale—can assess proclivities toward extreme analytic review. Review of General Psychology, 8, 291–322. doi:10.1037/
violent behavior remains an open question. Prior research 1089-2680.8.4.291
found that scores on the BAQ’s Physical Aggression subscale Bettencourt, B. A., & Miller, N. (1996). Gender differences in aggression as a
correlated positively with a validated laboratory aggression function of provocation: A meta-analysis. Psychological Bulletin, 119,
measure—blasting a stranger with aversive white noise (Web- 422–447. doi:10.1037/0033-2909.119.3.422
ster et al., 2014). Although noise blasts show aggressive Brown, T. A. (2006). Confirmatory factor analysis for applied research. New
behavior, they are clearly not violent aggression. Testing York, NY: Guilford Press.
whether the BAQ can predict extreme violence might be a Bryant, F. B., & Smith, B. D. (2001). Refining the architecture of aggression:
goal of future research. A measurement model for the Buss–Perry Aggression Questionnaire. Jour-
nal of Research in Personality, 35, 138–167. doi:10.1006/jrpe.2000.2302
Buhrmester, M., Kwang, T., & Gosling, S. D. (2011). Amazon’s Mechanical
Implications and Recommendations Turk: A new source of inexpensive, yet high-quality data? Perspectives on
Psychological Science, 6, 3–5. doi:10.1177/1745691610393980
A broad theoretical and methodological implication of our Buss, A. H., & Perry, M. (1992). The Aggression Questionnaire. Journal of
findings is that psychological constructs can be measured Personality and Social Psychology, 63, 452–459. doi:10.1037/0022-
effectively by using brief, efficient scales. When time 3514.63.3.452
 THE BRIEF AGGRESSION QUESTIONNAIRE 11

Cortina, J. M. (1993). What is coefficient alpha? An examination of theory and Lee, I. A., & Preacher, K. J. (2013, September). Calculation for the test of the
applications. Journal of Applied Psychology, 78, 98–104. doi:10.1037/ difference between two dependent correlations with one variable in com-
0021-9010.78.1.98 mon [Computer software]. Retrieved from http://quantpsy.org
Denson, T. F., Pedersen, W. C., & Miller, N. (2006). The Displaced Aggres- MacCallum, R. C., Widman, K. F., Zhang, S., & Hong, S. (1999). Sample size
sion Questionnaire. Journal of Personality and Social Psychology, 90, in factor analysis. Psychological Methods, 4, 84–99.
1032–1051. doi:10.1037/0022-3514.90.6.1032 Muth
en, L. K., & Muth
en, B. O. (2010). Mplus user’s guide (6th ed.). Los
DeWall, C. N., Bushman, B. J., Giancola, P. R., & Webster, G. D. (2010). The Angeles, CA: Author.
big, the bad, and the boozed-up: Weight moderates the effect of alcohol on Nezlek, J. B. (2008). An introduction to multilevel modeling for social and
aggression. Journal of Experimental Social Psychology, 46, 619–623. personality psychology. Social and Personality Psychology Compass, 2,
doi:10.1016/j.jesp.2010.02.008 842–860. doi:10.1111/j.1751-9004.2007.00059.x
Donnellan, M. B., Oswald, F. L., Baird, B. M., & Lucas, R. E. (2009). The Nezlek, J. B. (2011). Multilevel modeling for social and personality psychol-
mini-IPIP scales: Tiny-yet-effective measures of the Big Five factors of per- ogy. In J. B. Nezlek (Ed.), Sage library in social and personality psychology
sonality. Psychological Assessment, 18, 192–203. doi:10.1037/1040- methods. London, England: Sage.
3590.18.2.192 Nichols, A. L., & Webster, G. D. (2013). The single-item Need to Belong
Eagly, A. H., & Steffen, V. J. (1986). Gender and aggressive behavior: A Scale. Personality and Individual Differences, 55, 189–192. doi:10.1016/j.
meta-analytic review of the social psychological literature. Psychological paid.2013.02.018
Bulletin, 100, 309–330. doi:10.1037/0033-2909.100.3.309 Nichols, A. L., & Webster, G. D. (2014). The single-item Need for Consis-
Fabrigar, L. R., Wegner, D. T., MacCallum, R. C., & Strahan, E. J. (1999). tency Scale. Individual Differences Research, 12, 50–58.
Evaluating the use of exploratory factor analysis in psychological research. Nichols, A. L., & Webster, G. D. (2015). Designing a brief measure of social
Psychological Methods, 4, 272–299. doi:10.1037/1082-989X.4.3.272 anxiety: Psychometric support for a three-item version of the Interaction
Fraley, R. C., Waller, N. G., & Brennan, K. A. (2000). An item response the- Anxiousness Scale (IAS-3). Personality and Individual Differences, 79,
Downloaded by [University of Florida] at 15:28 09 June 2015

ory analysis of self-report measures of adult attachment. Journal of Person- 110–115. doi:10.1016/j.paid.2015.01.043
ality and Social Psychology, 78, 350–365. doi:10.1037/0022-3514.78.2.350 Pashler, H., & Wagenmakers, E.-J. (2012). Editors’ introduction to the special
Funder, D. C. (2008). Persons, situations, and person–situation interactions. In section on replicability in psychological science: A crisis of confidence?
O. P. John, R. Robins, & L. Pervin (Eds.), Handbook of personality (3rd Psychological Science, 7, 528–530. doi:10.1177/1745691612465253
ed., pp. 568–582). New York, NY: Guilford Press. Rammstedt, B., & John, O. P. (2007). Measuring personality in one minute or
Giancola, P. R., & Chermack, S. T. (1998). Construct validity of laboratory less: A 10-item short version of the Big Five Inventory in English and Ger-
aggression paradigms: A response to Tedeschi and Quigley (1996). Aggres- man. Journal of Research in Personality, 41, 203–212. doi:10.1016/j.
sion and Violent Behavior, 4, 237–253. doi:10.1016/S1359-1789(97)00004-9 jrp.2006.02.001
Goodman, J. K., Cryder, C. E., & Cheema, A. (2013). Data collection in a flat Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Appli-
world: The strengths and weaknesses of Mechanical Turk samples. Journal cations and data analysis methods. Thousand Oaks, CA: Sage.
of Behavioral Decision Making, 26, 213–224. doi:10.1002/bdm.1753 Raudenbush, S. W., Bryk, A. S., Cheong, Y. F., & Congdon, R. T., Jr. (2006).
Gosling, S. D., Rentfrow, P. J., & Swann, W. B., Jr. (2003). A very brief mea- HLM 6: Hierarchical linear and nonlinear modeling. Lincolnwood, IL: Sci-
sure of the Big-Five personality domains. Journal of Research in Personal- entific Software International.
ity 37, 504–528. doi:10.1016/S0092-6566(03)00046-1 Robins, R. W., Hendin, H. M., & Trzesniewski, K. H. (2001). Measuring
Guttman, L. (1954). Some necessary and sufficient conditions for common global self-esteem: Construct validation of a single-item measure and the
factor analysis. Psychometrika, 19, 149–161. doi:10.1007/BF02289162 Rosenberg Self-Esteem Scale. Personality and Social Psychology Bulletin,
Hauser, D. J., & Schwarz, N. (in press). Attentive Turkers: MTurk participants 27, 151–161. doi:10.1177/0146167201272002
perform better on online attention checks than do subject pool participants. Schmitt, N. (1996). Uses and abuses of coefficient alpha. Psychological
Behavior Research Methods. doi:10.3758/s13428-015-0578-z Assessment, 8, 350–353. doi:10.1037/1040-3590.8.4.350
Hayton, J. C., Allen, D. G., & Scarpello, V. (2004). Factor retention decisions Spearman, C. (1904). The proof and measurement of association between two
in exploratory factor analysis: A tutorial on parallel analysis. Organiza- things. American Journal of Psychology, 15, 72–101. doi:10.2307/1422689
tional Research Methods, 7, 191–205. doi:10.1177/1094428104263675 Spielberger, C. D., Jacobs, G., Russell, S., & Crane, R. S. (1983). Assessment of anger:
Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the The State–Trait Anger Scale. In J. N. Butcher & C. D. Spielberger (Eds.), Advances
world? Behavioral and Brain Sciences, 33, 61–83. in personality assessment (Vol. 2, pp. 161–201). Hillsdale, NJ: Erlbaum.
Horn, J. L. (1965). A rationale and test for the number of factors in factor anal- Swann, W. B., & Seyle, C. (2005). Personality psychology’s comeback and its
ysis. Psychometrika, 32, 179–185. doi:10.1007/BF02289447 emerging symbiosis with social psychology. Personality and Social Psy-
Jonason, P. K., & Webster, G. D. (2010). The Dirty Dozen: A concise measure chology Bulletin, 31, 155–165. doi:10.1177/0146167204271591
of the Dark Triad. Psychological Assessment, 22, 420–432. doi:10.1037/ Taylor, S. P. (1967). Aggressive behavior and physiological arousal as a func-
a0019265 tion of provocation and the tendency to inhibit aggression. Journal of Per-
Judd, C. M., McClelland, G. H., & Ryan, C. S. (2009). Data analysis: A model sonality, 35, 297–310. doi:10.1111/j.1467-6494.1967.tb01430.x
comparison approach (2nd ed.). New York, NY: Routledge. Watson, D. (2004). Stability versus change, dependability versus error: Issues
Kaiser, H. F. (1960). The application of electronic computers to factor analy- in the assessment of personality over time. Journal of Research in Personal-
sis. Educational and Psychological Measurement, 20, 141–151. ity, 38, 319–350. doi:10.1016/j.jrp/2004.03.001
doi:10.1177/001316446002000116 Webster, G. D. (2006). Low self-esteem is related to aggression, but especially
Khoo, S.-T., West, S. G., Wu, W., & Kwok, O.-M. (2006). Longitudinal meth- when controlling for gender: A replication and extension of Donnellan et al.
ods. In M. Eid & E. Diener (Eds.), Handbook of multimethod measurement (2005). Representative Research in Social Psychology, 29, 12–18.
in psychology (pp. 301–317). Washington, DC: American Psychological Webster, G. D. (2007). Is the relationship between self-esteem and physical
Association. aggression necessarily U-shaped? Journal of Research in Personality, 41,
Kline, R. B. (2011). Principles and practice of structural equation modeling 977–982. doi: 10.1016/j.jrp.2007.01.001
(3rd ed.). New York, NY: Guilford Press. Webster, G. D. (2009). The person–situation interaction is increasingly out-
Kline, R. B. (2013). Exploratory and confirmatory factor analysis. In Y. pacing the person–situation debate in the scientific literature: A 30-year
Petscher & C. Schatsschneider (Eds.), Applied quantitative analysis in the analysis of publication trends, 1978–2007. Journal of Research in Personal-
social sciences (pp. 171–207). New York, NY: Routledge. ity, 43, 278–279. doi:10.1016/j.jrp.2008.12.030
Krueger, J. I. (2009). A componential model of situation effects, person Webster, G. D., & Bryan, A. D. (2007). Sociosexual attitudes and behaviors:
effects, and situation-by-person interaction effects on social behavior. Jour- Why two factors are better than one. Journal of Research in Personality,
nal of Research on Personality, 43, 127–136. doi:10.1016/j.jrp.2008.12.042 41, 917–922. doi:10.1016/j.jrp.2006.08.007
 12 WEBSTER ET AL.

Webster, G. D., & Crysel, L. C. (2012). “Hit me, maybe, one more time”: Webster, G. D., Kirkpatrick, L. A., Nezlek, J. B., Smith, C. V., & Paddock, E.
Brief measures of impulsivity and sensation seeking and their prediction of L. (2007). Different slopes for different folks: Self-esteem instability and
blackjack bets and sexual promiscuity. Journal of Research in Personality, gender as moderators of the relationship between self-esteem and attitudinal
46, 591–598. doi:10.1016/j.jrp.2012.07.001 aggression. Self and Identity, 6, 74–94. doi:10.1080/15298860600920488
Webster, G. D., DeWall, C. N., Pond, R. S., Jr., Deckman, T., Jonason, P. K., Widaman, K. F., Little, T. D., Preacher, K. J., & Sawalani, G. M. (2011). On
Le, B. M., Bator, R. J. (2014). The Brief Aggression Questionnaire: Psycho- creating and using short forms of scales in secondary research. In K. H.
metric and behavioral evidence for an efficient measure of trait aggression. Trzesniewski, M. B. Donnellan, & R. E. Lucas (Eds.), Secondary data anal-
Aggressive Behavior, 40, 120–139. doi:10.1002/ab.21507 ysis: An introduction for psychologists (pp. 39–62). Washington, DC:
Webster, G. D., & Jonason, P. K. (2013). Putting the “IRT” in “dirty”: American Psychological Association.
Item response theory analyses of the Dark Triad Dirty Dozen—an effi- Wood, D., Nye, C. D., & Saucier, G. (2010). Identification and measurement
cient measure of narcissism, psychopathy, and Machiavellianism. Per- of a more comprehensive set of person-descriptive markers from the
sonality and Individual Differences, 54, 302–306. doi:10.1016/j. English lexicon. Journal of Research in Personality, 44, 258–272.
paid.2012.08.027 doi:10.1016/j.jrp.2010.02.003
Downloaded by [University of Florida] at 15:28 09 June 2015

View publication stats