Content uploaded by Arlin James Benjamin Jr.
Author content
All content in this area was uploaded by Arlin James Benjamin Jr. on Oct 17, 2017
Content may be subject to copyright.
Content uploaded by Arlin James Benjamin Jr.
Author content
All content in this area was uploaded by Arlin James Benjamin Jr. on Oct 17, 2017
Content may be subject to copyright.
AGGRESSIVE BEHAVIOR
Volume 32, pages 122–136 (2006)
Development and Testing of the Velicer Attitudes Toward
Violence Scale: Evidence for a Four-Factor Model
Craig A. Anderson
1
, Arlin J. Benjamin Jr.
2
, Phillip K. Wood
3
, and Angelica M. Bonacci
1
1
Department of Psychology, Iowa State University, Ames, Iowa
2
Oklahoma Panhandle State University, Oklahoma
3
University of Missouri-Columbia
:::::::::::::::::::::::::::::::::::::::::
The factor structure of the Velicer Attitudes Toward Violence Scale [VATVS; Velicer, Huckel and Hansen, 1989] was examined in
three studies. Study 1 (n5460 undergraduates) found a poor fit for a hierarchical five-factor model earlier advanced by Velicer
et al. [1989], but a good fit for an oblique four-factor model. In Study 2, this alternative model was cross-validated in a
confirmatory factor analysis of an additional 195 undergraduate students. In Study 3, the competing models were compared in
terms of ability to predict self-reported aggression, with 823 undergraduate students. The new four-factor model proved superior.
Other findings included evidence of factorial invariance on the VATVS, and more favorable attitudes toward violence among men
than women. The VATVS appears to measure the same four attitudinal constructs for men and women: violence in war, penal code
violence, corporal punishment of children, and intimate violence. Aggr. Behav. 32:122–136, 2006. r2006 Wiley-Liss, Inc.
:::::::::::::::::::::::::::::::::::::::::
Keywords: aggressive attitudes and behavior; structure of attitudes towards violence; violence attitudes; aggressive and violent
behavior
INTRODUCTION
Attitudes toward violence play an important role
in a wide variety of aggressive behaviors. For
example, negative attitudes toward specific racial
or ethnic groups (i.e., prejudice) are associated with
heightened aggression against those groups [e.g.,
Rogers, 1983]. Similarly, negative attitudes toward
women are associated with heightened sexual and
other forms of physical aggression against women
[e.g., Anderson and Anderson, 1997; Malamuth,
1998]. Closely related constructs, under such labels
as ‘‘hostility’’ or ‘‘cynicism’’ or ‘‘trait anger’’ have
also been strongly implicated in aggression and
violence [e.g., Barefoot, 1992; Buss and Perry, 1992;
Spielberger et al., 1983].
Recently, social psychologists have become cog-
nizant of the similarity of attitudes and other types
of appraisals, such as the formation and modifica-
tion of attributions and prejudice [Mackie and
Smith, 1998]. In terms of the General Aggression
Model [GAM; Anderson and Bushman, 2002;
Anderson and Huesmann, in press; Anderson
et al., 1996], attitudes toward violence can be
thought of as an individual difference variable—
with both affective and cognitive components—that
ultimately affect appraisals of the situation and of
alternative courses of action. As with attitudes in
general, an understanding of an individual’s atti-
tudes toward specific types of violence enables
prediction about specific behaviors [Petty and
Cacioppo, 1981]. In fact, some researchers consider
favorable attitudes toward violence as part of an
aggressive personality syndrome that includes a
well-developed network of aggression-related
knowledge structures (e.g., schemas and scripts),
affect, and reflexive motor responses [e.g., Anderson
and Bushman, 2002; Caprara et al., 1994; Dill et al.,
1997; Dodge and Crick, 1990; Epps and Kendall,
1995].
In addition, because attitudes can be modified by a
variety of situational cues [Petty and Cacioppo,
1986], researchers using a multidimensional attitude
toward violence scale should be able to show a link
between various aggressive cues (e.g., personal
insults, exposure to violent media) and attitudes
Published online in Wiley InterScience (www.interscience.wiley.
com). DOI: 10.1002/ab.20112
Received 1 December 2002; Accepted 5 June 2003
Correspondence to: Craig A. Anderson, Department of Psychology,
Iowa State University, W112 Lagomarcino Hall, Ames, IA 50011-
3180. E-mail: caa@iastate.edu
r2006 Wiley-Liss, Inc.
toward engaging in specific types of violence.
Furthermore, good measures of attitudes toward
violence will enable researchers to evaluate the
effectiveness of persuasive messages and techniques
aimed at changing such attitudes. Dual-process
models of persuasion [e.g., Chaiken, 1980; Petty
and Cacioppo, 1986] are readily applicable in this
regard.
In sum, an individual’s attitudes toward violence
can be thought of as evaluations of various kinds of
violent behavior. For example, one might evaluate
the acceptability of engaging in aggressive or violent
behavior in specific interpersonal settings, such as in
a dispute with one’s spouse. Some efforts have been
made to develop measures of attitudes toward
engaging in specific types of violence, such as rape
or domestic violence [e.g., Burt, 1980]. Until
recently, the only attempt to develop a measure of
more general attitudes toward violence was in the
form of an ostensibly unidimensional scale [Bardis,
1973]. Velicer et al. [1989] developed a multidimen-
sional inventory designed to measure attitudes
toward violence in war, penal code violence,
corporal punishment, extreme interpersonal vio-
lence, and intimate violence.
THE VELICER ATTITUDES TOWARD
VIOLENCE SCALE
The Velicer Attitudes Toward Violence Scale
[VATVS; Velicer et al., 1989] was developed as an
extension of the Violence Scale developed by Bardis
[1973]. The original scale was developed to assess the
favorableness of individuals’ evaluations of violence
in general. The Violence Scale was tested on small
samples of high school students, college students,
and individuals from the community. Bardis [1973]
found evidence that men tended to favor the use of
violence more than did women, and that individuals
with a college education favored violence less than
individuals who had only a high school education.
In addition, Bardis [1973] found that the scale was
internally consistent and stable over time.
After demonstrating that the original Violence
Scale measured four constructs (war, corporal
punishment, penal code violence, and extreme
interpersonal violence), Velicer et al. [1989] devel-
oped the VATVS by combining those items with
items concerned with violence during marriage and
courtship [Owens and Strauss, 1975] as well as some
newly developed items. Velicer et al. [1989] tested
their new scale on 360 undergraduate psychology
students, and found some preliminary support from
exploratory and confirmatory factor analyses for a
hierarchical five-factor model of violent attitudes,
described below.
The model advanced by Velicer et al. [1989]
included five primary and two second-order factors.
As seen in Figure 1, the five primary factors were
Violence in War (e.g., ‘‘War can be just’’), Penal
Code Violence (e.g., ‘‘Capital punishment is often
necessary’’), Corporal Punishment of Children (e.g.,
‘‘Children should be spanked for temper tan-
trums’’), Extreme Interpersonal Violence (e.g.,
‘‘University police should shoot students if they
are demonstrating’’), and Intimate Violence (e.g.,
‘‘The partner is the appropriate one to take out the
frustrations of the day on’’). Each primary factor
included three item parcels that were created by
combining related items (based on prior factor
analytic work). The second-order factor labeled
‘‘Institutional Violence’’ was related to the War,
Penal Code, and Corporal Punishment factors. The
second-order factor labeled ‘‘Interpersonal Vio-
lence’’ was related to Corporal Punishment, Ex-
treme Interpersonal Violence, and Intimate
Violence. Velicer et al. [1989] contended that the
Corporal Punishment factor contained elements of
Institutional Violence and Interpersonal Violence,
thus implying that it should load on both second-
order factors.
YOUTH VIOLENCE
The problem of youth violence in US society has
been highlighted by recent school shootings in
Santee, California (3/5/01), Conyers, Georgia (5/
20/99), and Littleton, Colorado (4/20/99). Two
common myths about youth violence in the US
persist despite evidence to the contrary. First, there
is widespread belief that weapons-related injuries in
schools have increased dramatically in recent years.
In fact, weapons-related injuries in schools have not
changed much in the past 20 years [Surgeon
General, 2001]. Second, many believe that the
epidemic of violence by youth peaked in the early
1990s and has declined significantly since then. In
fact, the prevalence of youth violence has continued
to increase [Surgeon General, 2001].
1
Aggr. Behav. DOI 10.1002/ab
1
Serious violence was defined in this study as: hit an instructor or
supervisor; gotten into serious fight in school or at work; taken part
in a fight where a group of your friends were against another group;
hurt somebody badly enough to need bandages or a doctor (assault
with injury); used a knife or gun or some other thing (like a club) to
get something from a person (robbery with a weapon). Youth
homicide rates have declined in recent years, largely because of a
decrease in use of firearms.
123Attitudes Toward Violence
OVERVIEW
The present studies reexamine the factor structure
of the VATVS [Velicer et al., 1989], and examine
correlations with measures of aggression and vio-
lence. This measure of violent attitudes was chosen
for its high potential utility for research on aggres-
sion in a variety of domains. A careful examination
of the five factors reported by Velicer et al. suggested
to us that a number of items were misclassified. For
example, one of the items classified by Velicer et al.
as a ‘‘corporal punishment’’ item seems more
conceptually related to intimate violence (Partners
should work things out together even if it takes
violence). Furthermore, Velicer et al. did not
examine the validity of the VATVS as a predictor
of aggression or violence, nor did they cross-validate
the five-factor model with another sample.
We conducted three studies designed to build
upon Velicer et al.’s work with a more contemporary
sample. The first two studies compared the original
VATVS five-factor structure with an alternative
four-factor model, using confirmatory factor analy-
sis. We also examined sex differences on the VATVS
to see whether the VATVS subscales measure the
same constructs for men and women and to see
whether men and women differ in terms of their
attitudes toward violence. In Study 3, we compared
the ability of the original five-factor model and the
alternative four-factor model to predict several
standard self-report measures of aggressive and
violent behavior.
STUDY 1
Method
Participants. Participants consisted of 460 in-
troductory psychology students (167 women and
293 men) from the University of Missouri-Colum-
bia, who completed the questionnaire in partial
fulfillment of course requirements. One participant
failed to complete the questionnaire, and was
dropped from the analyses.
Questionnaire. The Velicer et al. [1989] mea-
sure of attitudes toward violence includes 46 items.
Items were scored on a seven-point Likert-type
scale, ranging from ‘‘Strongly Disagree’’ (1) to
‘‘Strongly Agree’’ (7). A higher score indicates a
more favorable attitude toward violence. There are
no reverse-scored items (see individual items in
Table I).
WAR1 WAR2 WAR3 PCV1 PCV2 PCV3 CPC1 CPC2 CPC3 EXT1 EXT2 EXT3 INT1 INT2 INT3
.36
D1
.50
D2
.63
D3
*
D4
.39
D5
-.14
Inst Inter
.93 .87 .54 .63 .921.09
War Penal Corporal Extreme Intimate
.91 .85 .81 .90 .61 .90 .68 .81 .74 .80 .82 .73 .62 .72 .62
.40
E1
.53
E2
.59
E3
.44
E4
.79
E5
.43
E6
.73
E7
.58
E8
.67
E9
.60
E10
.57
E11
.68
E12
.78
E13
.69
E14
.78
E15
Fig. 1. Replication of Velicer et al. (1989) hierarchical five-factor model, Study 1. Notes: ‘‘Inst’’ 5Institutional Violence; ‘‘Inter’’ 5Interpersonal
Violence; CFI 5.82; RMSEA 5.15.
Aggr. Behav. DOI 10.1002/ab
124 Anderson et al.
Results
Confirmatory factor analyses were computed in
order to test the goodness of fit of the models to be
examined. Goodness of fit was based on two indices
[Hu and Bentler, 1999]. These were the comparative
fit index [CFI; Bentler, 1990] and the root mean
square error of approximation [RMSEA; Browne
TABLE I. Composition of Item Parcels for Final Four-Factor Model & Original Five-Factor Model
4’Model Type: Number of Factors-5
Penal Code Violence
PCV1 15) The death penalty should be part of every penal code. PCV2
16) Prisoners should never get out of their sentences for good behavior. PCV1
21) Capital punishment is often necessary. PCV1
PCV2 9) No matter how severe the crime, one should pay with an eye for an eye and a tooth for a tooth. PCV3
11) Violent crimes should be punished violently. PCV2
42) A law enforcement officer should shoot a citizen if they are a murder suspect.
(1) EXT2
PCV3 5) Any prisoner deserves to be mistreated by other prisoners in jail. PCV3
7) Prisoners should have more severe labor sentences than they do. PCV3
34) University police should shoot students if they are demonstrating.
(3) EXT1
43) University police should beat students if they are obscene.
(3) EXT2
Violence in War
War1 3) Any nation should be ready with a strong military at all times. War1
12) Our country has the right to protect its borders forcefully. War3
13) The manufacture of weapons is necessary. War1
22) Our country should be aggressive with its military internationally. War3
War2 1) War is often necessary. War2
2) The government should send armed soldiers to control violent university riots.
(3) War2
8) Killing of civilians should be accepted as an unavoidable part of war. War3
35) Every nation should have a war industry. War1
44) War can be just. War1
War3 6) Violence against the enemy should be part of every nations defense. War3
17) Universities should use armed police against students who destroy university property. War2
18) It is all right for the government to stop violent outbursts in neighboring countries with our armed soldiers.
(1) War2
23) A violent revolution can be perfectly right. War1
32) Spying on our nation should be severely dealt with.
(4) War3
39) War in self-defense is perfectly right. War2
Corporal Punishment of Children
CPC1 10) Punishing a child physically when he/she deserves it will make him/her a responsible and mature adult. CPC3
19) Giving mischievous children a quick slap is the best way to quickly end trouble. CPC3
29) An adult should beat a child with a strap or stick for being expelled. EXT1
CPC2 24) A parent hitting a child when he/she does something bad on purpose teaches the child a good lesson. CPC1
30) Young children who refuse to obey should be whipped. EXT3
38) A teacher hitting a child when he/she does something bad on purpose teaches the child a good lesson. CPC2
CPC3 4) Children should be spanked for temper tantrums. CPC3
25) A child’s habitual disobedience should be punished physically. CPC2
36) An adult should choke a child for breaking the law.
(1) EXT1
Intimate Violence
INT1 14) It is all right for a partner to choke the other if insulted or ridiculed. EXT3
20) It is all right for a partner to slap the other’s face if insulted or ridiculed. INT3
27) Partners should work things out together even if it takes violence. CPC1
37) It is all right for a partner to shoot the other if they flirt with others. EXT2
INT2 26) It is all right for a partner to slap the others face if challenged. INT3
28) The male should not allow the female the same amount of freedom as he has. EXT2
40) The partner is the appropriate one to take out the frustrations of the day on. INT2
46) The dominant partner should keep control by using violence. EXT1
INT3 31) It is all right for a partner to choke the other if they hit a child. EXT3
33) It is all right to coerce ones partner into having sex when they are not willing by forcing them. INT1
41) It is all right for a partner to shoot the other if they are unfaithful. EXT3
45) It is all right to coerce one’s partner into having sex when they are not willing by giving the other alcohol or drugs. INT2
*5Item was dropped from the final four-factor model because of classification discrepancies. Number in parentheses is the number of discrepant
classifications). PCV5Penal Code Violence; War 5Violence in War; CPC 5Corporal Punishment of Children; INT 5Intimate Violence;
EXT 5Extreme Interpersonal Violence (fifth factor in the Velicer et al. model).
Aggr. Behav. DOI 10.1002/ab
125Attitudes Toward Violence
and Cudeck, 1993]. There are no absolute guidelines
for determining the adequacy of various fit indices.
Early research suggested that a CFI of .90 or greater
[Bentler and Bonett, 1980] or a RMSEA of .08 or
less [Browne and Cudeck, 1993] indicated an
acceptable fit. More recently, Hu and Bentler
[1999] have suggested somewhat more stringent
guidelines, specifically, a value of close to .95 for
the CFI and .06 for RMSEA.
The models examined in the present study utilized
item parcels. Although item parcels may be bene-
ficial, there are currently few available published
guidelines for their construction [though for excep-
tions see Kishton and Widaman, 1994; Marsh et al.,
1998]. Marsh et al. [1998] have recently reported
Monte-Carlo findings suggesting that model fit is no
better when parcels are used than when the model is
tested without parceling. However, it is not yet clear
whether Marsh et al.’ [1998] assessment of parceling
holds true for real-world data, or for data in which
items are skewed (as is the case in the present study).
Given those caveats, parcels have been suggested
because: (a) individual items are likely to violate the
assumption of multivariate normality underlying the
maximum likelihood procedure used in many SEM
studies; (b) the use of parcels results in analyses that
are not as likely to be distorted by idiosyncratic
characteristics of individual items; and (c) item
parcels tend to be more reliable [c.f., Byrne, 1988;
Kishton and Widaman, 1994]. Thus, developing an
attitudes toward violence scale and validating it with
measures of aggression and of violence, using a
college-age population, seems both appropriate and
potentially useful.
2
Hierarchical five-factor model. A maximum
likelihood confirmatory factor analysis was com-
puted for the hierarchical five-factor model origin-
ally advanced by Velicer et al. [1989; see Fig. 1].
The model included the 15 item parcels specified
by Velicer et al. [1989] as manifest variables
(see Table I), five first-order factors, and two
correlated higher-order factors. The model was fit
by fixing the following parameters to 1: both higher-
order factors, one item parcel for each of the
five first-order factors, and the error terms related
to each item parcel. All other parameters were
free. As can be seen in Figure 1, the fit indicators
for the model suggest that this model fits the data
quite poorly; CFI 5.82 and RMSEA 5.1472
[.1386, .1559].
Four-factor ‘‘conceptual’’ model. One of our
concerns was the tendency for the factors in the
original Velicer et al. [1989] study to include items
that did not appear related to the construct they
purportedly measured. Therefore, we examined the
items in the measure and reassigned several items to
factors based on their conceptual relatedness to a
particular construct. For example, one item that was
originally included in the Corporal Punishment
factor (Partners should work things out together
even if it takes violence) appeared to be conceptually
related to intimate violence and was therefore
reassigned to an item parcel representing the
Intimate Violence factor in subsequent analyses.
We were ultimately left with four of the original
factors, as all of the items comprising the Extreme
Interpersonal Violence factor were assigned to other
factors. Table I also presents this four-factor model.
Although we categorized items in a way that
appeared intuitively plausible, we wished to double-
check that our categorization scheme would appear
reasonable to individuals who were not familiar with
the VATVS. To do so, we asked 10 psychology
graduate students to sort all 46 VATVS items into
four categories. We then noted any items for which
there were disagreements as well as the number of
people who disagreed for each item. Disagreements
may have occurred for a number of reasons. Some
items may have been classified as war-related by
those who either remember the Vietnam War or who
are knowledgeable about the era (e.g., ‘‘University
police should shoot students if they are demonstrat-
ing’’), but would otherwise be classified as penal
code-related. Similarly, some items may be classified
as war-related to the extent that individuals are
aware of the Cold War era (e.g., ‘‘Spying on our
nation should be severely dealt with’’). Other
discrepancies tended to involve items that could be
conceptualized as either penal code or corporal
punishment related (e.g., ‘‘An adult should choke a
child for breaking the law’’). In general, our intuitive
categorization of VATVS items was in agreement
with the 10 graduate students. However, there were
seven items with at least one discrepancy. These are
depicted in Table I with an asterisk; the number of
discrepant classifications is given by the number in
parentheses following the asterisk. The overall
kappa for agreement among raters was .92.
Like Velicer et al. [1989], we constructed three
parcels of items for each latent variable. In order to
construct these parcels, an exploratory analysis at
the level of individual items was computed. From
Aggr. Behav. DOI 10.1002/ab
2
One reviewer suggested rerunning the models with individual items
to be sure of the results. As expected, the overall model fits were
poorer with individual items than with item parcels. More
importantly, the relative fits of various models remained the same
in each study, with the four-factor models fitting better than the five-
factor models, bolstering our original conclusions.
126 Anderson et al.
that analysis, the factor loadings for each of these
items were examined. The parcels for each factor
were created by assigning the top three items (i.e.,
the three items with the highest loadings) to the
three parcels, followed by the next three items, and
so on until all items had been assigned to item
parcels.
The four latent constructs were assumed to
correlate, and each latent construct was measured
by three item parcels. To fit the model, the following
parameters were fixed to 1: the four factors and the
error terms for each item parcel. All other para-
meters were free. A maximum likelihood confirma-
tory factor analysis of this four-factor model yielded
a poor fit to the data; CFI 5.84 and
RMSEA 5.1545 [.1433, .1660].
The next analysis excluded those items in which
three or more of the 10 raters disagreed about their
classification. A subsequent confirmatory factor
analysis with these items removed yielded a better
fit; CFI 5.90 and RMSEA 5.1225 [.1111, .1341].
Although an improvement, the fit indices still fell
outside the range of what would be considered an
adequate fit.
Finally, all seven discrepant items were removed
from the analysis. Factor loadings and intercorrela-
tions among the four factors are summarized in
Figure 2. The confirmatory factor analysis with the
seven discrepant items removed yielded a good fit;
CFI 5.96 and RMSEA 5.0776 [.0658, .0899]. An
examination of the intercorrelations among the
latent variables suggested that three of the latent
variables (war, corporal punishment, penal code
violence) were strongly related to each other, but
that the latent variable representing intimate
violence was weakly related or unrelated to the
other factors. The three related variables are all
representative of attitudes toward socially sanc-
tioned violence.
The five-factor model without seven discre-
pant items. A final confirmatory factor analysis
was computed on the original five-factor model, but
with the seven discrepant items removed from their
respective item parcels. Even with these items
removed, the hierarchical five-factor model still fits
the data poorly; CFI 5.83 and RMSEA 5.1348
[.1262, .1436].
Sex differences. To examine potential sex
differences, t-tests on the four sub-scales (attitudes
toward war, penal code violence, corporal punish-
ment, and intimate violence) were computed. Men
(vs. women) expressed more favorable attitudes
toward violence in war (M’s 53.73 vs. 2.46,
t(458) 512.16), penal code (M’s 52.85 vs. 2.05,
t(458) 59.34), and corporal punishment domains
(M’s 53.73 vs. 2.51, t(458) 57.90), (all P’so.0001).
WAR1 WAR2 WAR3 PCV1 PCV2 PCV3 CPC1 CPC2 CPC3 INT1 INT2 INT3
.84
War Penal Corporal Intimate
.89 .89 .81 .89 .85 .80 .90 .80 .74 .85 .86 .77
.46
E1
.47
E2
.59
E3
.46
E4
.52
E5
.60
E6
.44
E7
.60
E8
.68
E9
.52
E10
.52
E11
.68
E12
.55 .35
-.02
.63 -.09
Fig. 2. Final four-factor model, Study 1. CFI 5.96; RMSEA 5.08.
Aggr. Behav. DOI 10.1002/ab
127Attitudes Toward Violence
However, attitudes toward the use of intimate
violence showed no significant effect for sex,
t(458) 5.16, ns. Men and women held equally
unfavorable attitudes toward intimate violence
(M’s 51.92 vs. 1.91).
STUDY 2
Study 1 provided evidence that the hierarchical
five-factor model does not fit the attitude structure
of our contemporary college student sample, and
that a four-factor model offers a plausible alter-
native. Study 2 was conducted to cross-validate
these results. In this study, the hierarchical five-
factor model and the new four-factor model were
again tested to determine how well they fit the data.
We expected the new four-factor model to yield a
good fit, and the hierarchical five-factor model to fit
poorly.
Method
Participants. Participants were 195 introduc-
tory psychology students (94 women and 101 men)
from the University of Missouri-Columbia, who
completed the questionnaire in partial fulfillment of
course requirements. About 95% were 23 years old
or younger. Males were slightly older than females
(M’s 519.25 and 18.80, respectively), but not reliably
so (P4.05). There were too few minority participants
(about 8%) for meaningful comparisons.
Materials and procedure. The VATVS was
identical to that used in Study 1, with one exception.
A five-point Likert-type scale was used in the
current study, rather than the seven-point version
used in Study 1.
Results and Discussion
A maximum likelihood confirmatory factor ana-
lysis of the original five-factor model once again
yielded a poor fit to the data; CFI 5.90 and
RMSEA 5.0956 [.0810, .1104]. We also tried run-
ning the original five-factor model without the seven
discrepant items. It yielded a negative Eigen value,
indicating a poor fit. A confirmatory factor analysis
based on the final four-factor model in Study 1
yielded a good fit to the data; CFI 5.96 and
RMSEA 5.0630 [.0403, .0846]. Factor loadings are
reported in Table II, and correlations between the
factors are reported in Table III. The pattern of
correlations between factors differed somewhat
from those found in Study 1. Most notably, the
intimate violence factor, which was essentially
unrelated to the war and penal code factors in
Study 1, showed a weak to moderately positive
relation to the war and penal code factors in Study
2. Why this is the case is not entirely clear.
Nonetheless, the final four-factor model from the
previous study was replicated. In sum, the VATVS
in its present form (i.e., after removing seven items
from the original questionnaire) measures four
latent factors.
As in Study 1, men reported more favorable
attitudes toward violence than women for the war
(M’s 53.14 vs. 2.59, t(193) 56.04), penal code
(M’s 53.31 vs. 2.93, t(193) 53.64), and corporal
punishment subscales (M’s 52.64 vs. 2.14,
t(193) 54.33), all P’so.001. However, unlike
Study 1, men in Study 2 reported more favorable
attitudes toward intimate violence than women
(M’s 51.57 vs. 1.22), t(193) 55.74, Po.0001,
though the means of both men and women were
generally unfavorable.
STUDY 3
Studies 1 and 2 compared the traditional five-
factor model of the VATVS against the new,
TABLE II. Standardized Factor Loadings for Final Four-
Factor Model, Study 2
War Corporal Intimate Penal
WAR1 .85
WAR2 .80
WAR3 .76
CPC1 .84
CPC2 .87
CPC3 .79
INT1 .79
INT2 .82
INT3 .74
PCV1 .71
PCV2 .74
PCV3 .66
TABLE III. Correlations Among Latent Variables for Four-
Factor Model, Study 2
War Corporal Intimate Penal
War 1
Corporal .47 1
Intimate .37 .47 1
Penal .54 .28 .23 1
Aggr. Behav. DOI 10.1002/ab
128 Anderson et al.
conceptually based, four-factor model. Both studies
demonstrated that the four-factor model yielded a
better fit. In Study 3, we again sought to test the
hypothesis that the four-factor model is a better fit
than the five-factor model. We also administered
self-report measures of behavioral aggression—the
physical and verbal aggression subscales of the
Buss–Perry Aggression Questionnaire [Buss and
Perry, 1992] and the violence subscale of the
National Youth Survey [NYS; Elliott et al., 1985].
We expected that the new four-factor model of the
VATVS would predict aggressive behavior at least
as well as the traditional five-factor model, despite
having fewer items and fewer subscales.
Method
Participants. Participants were 823 students
enrolled in introductory psychology courses at Iowa
State University. Participants received extra credit in
exchange for their voluntary participation. There
were 498 females and 325 males. About 95% were 23
years old or younger. Females were slightly older
than males (M’s 519.48 and 18.86, respectively),
F(1,806) 513.38, Po.001. There were too few
minority participants (about 8%) for meaningful
comparisons.
Materials. Participants completed the VATVS
measure using five-point Likert-type scales. They
also completed the physical and verbal aggression
subscales of the Buss–Perry Aggression Question-
naire [Buss and Perry, 1992] and the violence
subscale of the NYS [Elliot et al., 1985]. The
Buss–Perry Aggression Questionnaire is a 29-item
measure assessing physical aggression, verbal ag-
gression, anger, and hostility. Participants indicate
how characteristic each statement is of them (i.e., ‘‘If
someone hits me, I hit back.’’) on a Likert-type scale
ranging from 1 (extremely uncharacteristic of me) to
5 (extremely characteristic of me). Most relevant to
the current studies are the physical and verbal
aggression subscales, which assess aggressive beha-
vior. The Buss–Perry scales have been validated in a
variety of settings, yielding significant correlations
with a range of objective measures of aggression
such as peer-nominated aggression [Buss and Perry,
1992], penalties for aggressive hockey violations
[Bushman and Wells, 1998], and laboratory mea-
sures of physical aggression [Bushman, 1995].
Participants also completed the violence subscale
of the NYS [Anderson and Dill, 2000; Elliot et al.,
1985; Lackey, 2003]. The NYS is a 45-item measure
which assesses how frequently the respondent
engages in aggressive, antisocial, and criminal
behaviors. We chose to use only the violence
subscale of the NYS because it is the scale most
closely related to aggressive behavior. The NYS
violence subscale consists of 10 statements describ-
ing aggressive behavior (i.e., ‘‘yattacked someone
with the idea of seriously hurting or killing him/
her’’). The NYS items, and various versions of them,
are among the most widely used measures of
delinquency in research programs tracking shifts in
antisocial behavior and in assessing intervention
programs [e.g., Esbensen and Osgood, 1999]. In our
version of the scale, participants indicate how often
they have engaged in the behavior within the past
year using an 11-point scale. The scale is anchored at
1 (0 times in the last year) and 11 (more than 27
times in the last year), with the intermediate
responses increasing in intervals of 3 (i.e., 1–3 times
in the last year, 4–6 times in the last year).
3
Procedure. Participants completed the
VATVS, the Buss–Perry physical and verbal aggres-
sion subscales, and the NYS violence scale as part of
a battery of questionnaires administered during
large mass-testing sessions. Participants were as-
sured their responses would be confidential.
RESULTS
Preliminary analyses
Sex differences. Independent groups t-tests were
used to assess gender difference on the VATVS
subscales, Buss–Perry physical and verbal aggres-
sion scales, and the NYS violence scale. Men held
more favorable attitudes toward violence than
woman for each subscale: War M’s 53.03 vs. 2.64,
t(458) 58.78; Penal code violence M’s 52.92 vs.
2.72, t(458) 53.68; Corporal punishment of children
M’s 52.26 vs. 1.81, t(458) 59.08; Intimate violence
M’s 51.56 vs. 1.26, t(458) 57.85, all P’so.001. Men
also reported behaving more aggressively than
woman on the Buss–Perry physical (M’s 53.15 vs.
2.58, t(458) 59.67) and verbal (M’s 53.79 vs. 3.23,
t(458) 57.16) subscales, and for the NYS violence
scale (M’s 5.16 vs. .10, t(458) 54.40), all
P’so.001.
Comparison of the VATVS four- and five-factor
models. We conducted confirmatory factor analyses
to compare the VATVS four- and five-factor
models. As in Studies 1 and 2, a confirmatory factor
analysis based on the four-factor model yielded an
Aggr. Behav. DOI 10.1002/ab
3
The NYS violence subscale score was computed using standardized
scores, because the item variances were so different (Anderson and
Dill, 2000).
129Attitudes Toward Violence
acceptable fit to the data; CFI 5.94 and
RMSEA 5.0927 [.0844, .1011]. Figure 3 presents
this model.
We were unable to fit the original five-factor
model using confirmatory factor analysis, because it
yielded negative Eigen values, thus demonstrating
poor fit to the data. We also tried running the
original five-factor model without the seven dis-
crepant items. It also yielded negative Eigen values.
Thus, the four-factor model again fits the data better
than the original five-factor model.
Assessing factorial invariance of VATVS
One important question is whether the mean sex
differences on the four VATVS sub-scales are due
simply to the general tendency for men to favor the
use of violence to a greater degree than women, or if
men and women interpret items from the VATVS
differently. Systematic differences in participants’
interpretation of a scale as a function of group (e.g.,
sex) can influence interpretability of findings. One
way to detect these differences is to examine the
factorial invariance of the scale, i.e., the similarity in
relationship between items (or item parcels) and
factors across groups. Basically, this tests the
hypothesis that the model fits different populations
equally well [see, e.g., Bentler, 1995; Marsh and
Hocevar, 1985]. Invariance was tested by computing
two confirmatory factor analyses, one in which
factor loadings were allowed to vary between men
and women, the other constraining the factor
loadings to be equal for men and women. Generally,
if there is no significant difference in the fit of
constrained vs. unconstrained models, then the
contention that there are no sex differences in
interpretation of VATVS items is strengthened. If
there is a discrepancy, then further models can be
run to identify the source. For instance, one item
parcel may load more strongly on a latent construct
for males than females.
Studies 1 and 2. To allow for a sufficient sample
size, the data from studies 1 and 2 were combined,
after standardizing each variable within each study.
The resulting correlation matrices (one for males,
one for females) were then examined for factorial
invariance. There were two reasons for this proce-
dure. First, the two data sets used different response
scales (i.e., 1–7 in Study 1, 1–5 in Study 2). Second,
by analyzing the correlation matrices we avoided the
potential problem of finding significant differences
in factor structure due only to differences between
men and women in their variability of responses to
particular items.
WAR1 WAR2 WAR3 PCV1 PCV2 PCV3 CPC1 CPC2 CPC3 INT1 INT2 INT3
.65
War Penal Corporal Intimate
.73 .60 .84 .80 .76 .56 .91 .84 .78 .89 .89 .84
.68
E1
.60
E2
.54
E3
.60
E4
.64
E5
.83
E6
.42
E7
.53
E8
.63
E9
.45
E10
.45
E11
.54
E12
.45 .63
.32
.52 .26
Fig. 3. Replication of four-factor model, Study 3. CFI 5.94; RMSEA 5.09.
Aggr. Behav. DOI 10.1002/ab
130 Anderson et al.
The results are in the top half of Table IV. The
overall fit of all models was in the acceptable range.
The w
2
difference tests showed a significant differ-
ence in fit between the unconstrained model and the
model in which the factor loadings were constrained
w
2
(8) 526, Po.05. Examination of modification
indexes suggested freeing the factor loading between
the War1 parcel and the latent War construct. The
difference in fit between this model and the model
in which all factor loadings were unconstrained
was not statistically significant, w
2
(7) 510. The
War1 parcel loading for males (1.06) was somewhat
larger than for females (.79), but clearly was in the
same direction.
Study 3. The bottom half of Table IV displays the
results of tests of factorial invariance for men and
women from Study 3. The results are essentially the
same as for Studies 1 and 2. Specifically, the factor
loadings were essentially the same for men and
women, except the War1 parcel loaded more
strongly for males (1.13) than for females (.76).
Overall, the four-factor model fits the data for both
men and women pretty well.
VATVS and aggressive behavior
The main analyses tested whether the theoretically
based four-factor model could predict aggressive
behavior at least as well as the five-factor model.
Because the four-factor model is theoretically
cleaner, simpler, and structurally more sound than
the five-factor model, it is not necessary for it to
have higher predictive validity than the five-factor
model in order to be the preferred instrument. It
merely needs to predict as well as the five-factor
model.
Scale reliabilities. The first step was to examine the
coefficient alpha reliabilities of all scales used in this
study (four- and five-factor attitude scales and the
three aggressive behavior measures) to see whether
any subsequent differences in predictive power
might be the result of systematic differences in
measurement reliability. All were acceptably high
for the regression analyses reported in the next
paragraphs. The lowest reliability coefficient was .79
for Buss–Perry verbal aggression. The highest was
.90 for the five-factor subscale of Extreme violence.
Next, a series of stepwise regression analyses using
the maximum R
2
method were run to determine
which VATVS subscales best predicted scores on the
Buss–Perry physical and verbal aggression subscales
and the NYS violence subscale. The same analyses
were also run on a composite aggression variable,
created by standardizing and then averaging scores
on the three aggression measures. Regression
analyses were run for both the four- and five-factor
VATVS models. The results of the regression
analyses, including beta weights and proportion of
variance explained, are presented in Table V. Over-
all, the theoretically based four-factor model pre-
dicted self-reported aggressive behavior better than
the original five-factor model. With respect to the
four-factor model, war, corporal punishment, and
intimate violence each uniquely accounted for a
significant portion of the variance for at least one
aggression measure when all four VATVS subscales
were in the model.
NYS violence. Scores on the NYS violence scale
can be as well predicted by the intimate violence
subscale of the four-factor VATVS as by any
combination of predictors. It accounted for 11%
of the variance in NYS violence scores. The addition
of the other three subscales (war, penal code
violence, and corporal punishment) into the model
did not substantially increase the overall proportion
of variance explained.
Results were similar using the original five-factor
model. Again, intimate violence served as the best
predictor, explaining 10% of the variance. The
addition of the other four subscales (extreme, war,
penal code violence, and corporal punishment) did
not increase the overall proportion of variance
explained.
Buss–Perry physical aggression. Scores on the
Buss–Perry physical aggression subscale can best
be predicted by scores on the war, corporal punish-
ment, and intimate violence subscales of the four-
factor VATVS. By itself, the war subscale accounted
for 15% of the variance. Adding corporal punish-
ment and intimate violence increased the proportion
TABLE IV. Tests of Factorial Invariance on the VATVS for
Men and Women
Factor loading
constraints df w
2
w
2
difference
from All
free (df) cfi
Studies 1 and 2 combined
All free 97 292 .97
All fixed 105 318 26
(8) .96
All fixed except
War1
104 302 10 (7) .97
Study 3
All free 97 486 .95
All fixed 105 522 36
(8) .94
All fixed except
War1
104 494 8 (7) .95
*Po.05.
Aggr. Behav. DOI 10.1002/ab
131Attitudes Toward Violence
of variance explained to 21% and 23%, respectively.
The addition of penal code violence did not increase
the proportion of variance explained beyond 23%.
For the five-factor model, war and corporal
punishment again served as the two best predictors,
explaining 20% of the variance. The third best
predictor was extreme violence, which increased the
proportion of variance explained to 22%. The
addition of penal code violence and intimate
violence did not increase the proportion of variance
explained.
Buss–Perry verbal aggression. The VATVS did not
predict scores on the Buss–Perry verbal aggression
subscale as well as it did on the physical aggression
measures. For the four-factor model, attitudes toward
war served as the best predictor, accounting for 4% of
the variance. The addition of corporal punishment
increased that amount to 6%. The addition of
TABLE V. Regression Analysis of VATVS four- and five-Factor Models as Predictors of Self-Reported Aggression on Buss–Perry
Physical and Verbal Aggression Scales and the NYS Violence Scale
Model/# of
predictors NYS violence Buss–Perry physical Buss–Perry verbal
Aggression
composite
Four-factor/1
variable
R
2
.11 .15 .04 .13
Predictor 1 Intimate (B5.51)
War (B5.70)
War (B5.38)
Intimate (B5.52)
Five-factor/1
variable
R
2
.10 .15 .04 .13
Predictor 1 Intimate (B5.39)
War (B5.70)
War (B5.38)
Extreme (B5.53)
Four-factor/2
variable
R
2
.11 .21 .06 .19
Predictor 1 Intimate (B5.43)
War (B5.50)
War (B5.28)
Intimate (B5.43)
Predictor 2 War (B5.02) Corporal (B5.41)
Corporal (B5.18)
War (B5.30)
Five-factor/2
variable
R
2
.10 .20 .05 .18
Predictor 1 Intimate (B5.22)
War (B5.50)
War (B5.28)
Extreme (B5.43)
Predictor 2 Extreme (B5.22)
Corporal (B5.38)
Corporal (B5.19)
War (B5.31)
Four-factor/3
variable
R
2
.11 .23 .06 .20
Predictor 1 Intimate (B5.43)
War (B5.49)
War (B5.28)
Intimate (B5.34)
Predictor 2 War (B5.06) Corporal (B5.27)
Corporal (B5.18)
War (B5.25)
Predictor 3 Penal (B5.02) Intimate (B5.34
) Intimate (B5.02) Corporal (B5.12)
Five-factor/3
variable
R
2
.10 .22 .05 .19
Predictor 1 Intimate (B5.22)
War (B5.50)
War (B5.26)
Extreme (B5.35)
Predictor 2 Extreme (B5.20)
Corporal (B5.38)
Corporal (B5.17)
War (B5.25)
Predictor 3 War (B5.04) Extreme (B5.36)
Extreme (B5.06) Corporal (B5.12)
Four-factor/4
variable
R
2
.11 .23 .06 .20
Predictor 1 Intimate (B5.43)
War (B5.43)
War (B5.28)
Intimate (B5.34)
Predictor 2 War (B5.06) Corporal (B5.25)
Corporal (B5.18)
War (B5.25)
Predictor 3 Penal (B5.02) Intimate (B5.33)
Intimate (B5.02) Corporal (B5.12)
Predictor 4 Corporal (B5.006) Penal (B5.10)
Penal (B5.01) Penal (B5.01)
Five-factor/4
variable
R
2
.10 .22 .05 .20
Predictor 1 Intimate (B5.22)
War (B5.43)
War (B5.26) Extreme (B5.35)
Predictor 2 Extreme (B5.21)
Corporal (B5.25)
Corporal (B5.18) War (B5.25)
Predictor 3 War (B5.06) Extreme (B5.35)
Extreme (B5.13) Corporal (B5.11)
Predictor 4 Penal (B5.02) Penal (B5.10)
Intimate (B5.09) Intimate (B5.11)
Five-factor/5
variable
R
2
.10 .22 .05 .20
Predictor 1 Intimate (B5.22)
War (B5.43)
War (B5.25) Extreme (B5.25)
Predictor 2 Extreme (B5.22)
Corporal (B5.24)
Corporal (B5.17) War (B5.25)
Predictor 3 War (B5.07) Extreme (B5.31)
Extreme (B5.14) Corporal (B5.11)
Predictor 4 Penal (B5.02) Penal (B5.10)
Intimate (B5.09) Intimate (B5.11)
Predictor 5 Corporal (B5.02) Intimate (B5.05) Penal (B5.02) Penal (B5.01)
*Po.05, N5817.
Aggr. Behav. DOI 10.1002/ab
132 Anderson et al.
intimate and penal code violence did not appreciably
increase the proportion of variance explained.
With respect to the five-factor model, war,
corporal punishment, and extreme violence ac-
counted for 5% of the variance in verbal aggression
scores. The addition of intimate and penal code
violence did not substantially increase the propor-
tion of variance explained.
Aggression composite. The three measures of
aggression correlated modestly but significantly with
each other. Buss–Perry physical aggression corre-
lated .31 with NYS violence and .44 with verbal
aggression, both highly significant (P’so.001). The
verbal and NYS violence measures yielded a small
but significant correlation, r(862) 5.13, Po.0001.
For the four-factor model, the best predictor of
aggression was the intimate violence subscale, which
accounted for 13% of the variance. Adding war and
corporal punishment increased the variance ex-
plained to 19% and 20%, respectively. The penal
code subscale did not significantly increase the
amount of variance explained.
The best predictor of aggression from the five-
factor model was the extreme violence subscale,
accounting for 13% of the variance. Adding the
war and the corporal punishment subscales signifi-
cantly increased the amount of variance explained
to 18% and 19%, respectively. Neither the intimate
nor the penal code violence subscales added
significantly.
Four-factor vs. five-factor model predictions.To
further explore the relative strengths of the four- and
five-factor model subscales, we ran stepwise regres-
sions in which the best 1, 2, 3, 4 and 5 predictor
models were selected based on maximizing R
2
, with
all nine VATVS subscales as potential predictors of
the aggression composite measure. Only three
predictors contributed statistically significant unique
increments. All were from the four-factor model.
Specifically, the intimate violence, war, and corporal
punishment subscales from the four-factor model
accounted for 20% of the variance in aggression
scores, F(3, 813) 567.84, Po.001. No other sub-
scales from either the four- or the five-factor model
contributed significant unique increments, all
F’so1.
Finally, we ran a model which first entered the
three five-factor subscales that had added signifi-
cantly to the prediction of aggression (i.e., the five-
factor/three-variable model in Table V), and then
added the best four-factor model subscale (intimate
violence). This four-factor subscale still added
significantly to the prediction of aggressive behavior,
F(1, 812) 54.33, Po.04.
DISCUSSION
Results from the present analyses suggest that the
VATVS measures four latent constructs. Four
factors (War, Penal Code Violence, Corporal
Punishment, and Intimate Violence) from the
original Velicer et al. [1989] study remained in the
present analyses, though with somewhat different
item compositions. Items from the Extreme Inter-
personal Violence factor from the Velicer et al.
[1989] study were reassigned to the other factors,
and items that did not clearly fit any of the factors
were dropped.
One of our concerns about the original five-factor
model was that items comprising the item parcels
did not seem to be conceptually related to their
respective latent constructs. Thus, one purpose of
our studies was to test a model that was both
conceptually and statistically plausible. Parcels used
in the final model were constructed with the
following question in mind: Could a reasonable
case be made that item X represents construct Y? In
doing so, we discovered that the items forming
parcels for the Extreme Interpersonal Violence scale
in the original Velicer et al. [1989] model could
reasonably be said to belong to one of the other
latent constructs measured by the instrument. The
resulting model (see Fig. 2) was one in which the
relationship of items to latent factors made sense
from a conceptual standpoint and fit the data
reasonably well. Studies two and three confirmed
the superiority of the four-factor model on empirical
grounds.
Another purpose was to examine the validity of
these attitudes towards violence scale by seeing
whether it could predict self-reported aggression and
violence in a large sample cross-sectional study. Our
four-factor model did very well in predicting
physical aggression (Multiple R5.48) and violence
(Multiple R5.33). It also successfully predicted
verbal aggression, though at a more modest level
(Multiple R5.24).
It is interesting to note that only the intimate
violence subscale of our revised four-factor model
was needed to predict self-reported violent behavior.
The other attitudes toward violence subscales (war,
penal code, and corporal punishment of children) do
not have behavioral representatives on the NYS
violence scale; most of the NYS items describe
violence directed at peers, parents, or teachers.
Thus, this specificity makes sense.
Of course, as shown in Table V there was little
specificity in the prediction of physical aggression as
measured by the Buss–Perry subscale. All four
Aggr. Behav. DOI 10.1002/ab
133Attitudes Toward Violence
attitude subscales contributed unique increments.
Why this is so is not obvious from an examination of
the physical aggression items, though the fact that
none of them directly refer to family, peers, or
teachers may be a partial explanation.
One obvious question concerns why the original
five-factor model fared so poorly. We believe
that part of the answer lies in the passage of time
and the resulting shifts in meaning in some of
the items. To our participant population the
Vietnam War, college student anti-war demonstra-
tions, and the Cold War are dim, historical
blips, rather than the vivid events and memories of
Velicer et al.’s participants. In addition, the lack
of a cross-validation study in the Velicer et al.
article may have allowed the original parcel-selec-
tion procedure to capitalize on chance. In any
case, our findings illustrate the importance of
periodic updating and validation studies of
‘‘standard’’ attitude scales, especially when the
attitude domain is one in which major social events
have transpired [e.g., Carnagey and Anderson, in
preparation].
Our results also showed that men hold more
favorable attitudes toward engaging in the violent
behaviors than do women. This sex difference was
reliable for all four constructs measured by the
VATVS. Finally, the VATVS factor structure
appeared invariant across men and women at the
latent construct level, and largely invariant at the
measurement level with the exception that one set of
the War items was more important for men than
women. The mean sex differences on the sub-scales
are apparently not due to differences in the meaning
of the constructs for men and women. All in all,
these findings are perhaps not too surprising given
that men and women differ in their propensity to
engage in violent behavior. Engaging in violence
may be more efficacious for men than for women,
hence, making such behavior appear more attractive
for men.
Future research could be profitably aimed in
several directions. First, the existence of multi-
dimensional attitudes toward violence questionnaire
enables researchers the flexibility to validate the
scale in a variety of ways. Some research has
examined correlations between the original five-
factor model and measures of dispositional aggres-
siveness [e.g., Dill et al., 1997]. Similar research
should be done between the new four-factor model
and various measures of aggressive dispositions,
such as trait irritability [e.g., Caprara et al., 1985].
Our Study 3 moves the field somewhat in this
direction.
Second, the revised VATVS measures provide a
useful tool for attitudes researchers. For example,
additional work is needed on how attitudes and
other individual differences combine to form an
aggressive personality type is needed, as well as
research on how attitudes toward violence fit with
more general personality structures such as the
Big 5.
Third, the validity of the VATVS can be further
tested by examining the role of specific facets of the
VATVS on specific types of behavior. For example,
the GAM [Anderson and Bushman, 2002; Anderson
and Huesmann, in press; Anderson et al., 1996]
predicts that one’s attitude towards the use of
intimate violence should be related to one’s pro-
pensity to engage in domestic violence against a
spouse, child, or significant other. Study 3 provided
one such test, and found that violence directed
against one’s peers, parents, and teachers was highly
correlated with violent attitudes toward intimates.
Additional work along this line is needed, especially
studies that examine the predictive validity of the
VATVS subscales using behavioral measures that
are not self-reports. For example, it would be useful
to examine the relation between attitudes toward
corporal punishment of children and punishment
choices by parents, or the relation between war
attitudes and participation in war protests and
counter-protests.
The face validity of the items and the lack of
reverse scored items suggest some alternative ex-
planations for our findings. The results may, for
example, reflect a tendency of participants to use
stereotyped response sets to the items rather than
actual attitudes toward violence. In addition, the
possibility that participants responded in a socially
desirable manner cannot be directly ruled out.
However, the fact that Velicer et al. [1989] found
no evidence of a link between social desirability and
participants’ scores on the VATVS argues against
that particular alternative explanation. Similarly,
responses to items on several of the subscales may
reflect individual differences in ideology or dogma-
tism rather than endorsements of the use of violence
per se. Future research should test for this possibi-
lity. The fact that the attitude subscales reliably
predicted three very different measures of aggression
(i.e., the Buss–Perry subscales and the NYS violence
subscale) also argues against these alternative
explanations.
The extent to which these attitudes toward
violence represent stable dispositions has yet to be
directly examined. One useful research direction
would be to examine changes in individuals’ scores
Aggr. Behav. DOI 10.1002/ab
134 Anderson et al.
on the VATVS over time, as well as the extent to
which the factor structure of the VATVS is invariant
over time. Another direction would be to test
whether certain types of life events (e.g., repeated
exposure to violent entertainment media, the Sep-
tember 11, 2001 terrorists attacks) or certain types of
specific interventions (e.g., date rape programs
presented to college students) have a significant
and lasting impact on attitudes toward violence. In
addition, experimental research could examine the
role of a number of situational factors that may
result in short-term fluctuations on participants’
attitudes toward violence. The current VATVS
enables researchers interested in attitude change to
examine the role of persuasive messages designed to
increase or decrease the favorability of attitudes
toward various forms of violence, as well as to
examine the role of aggression-related cues (e.g.,
video game violence) on subsequently held attitudes
toward engaging in violence. Such research would
contribute to our understanding of the psychological
underpinnings of aggressive behavior.
ACKNOWLEDGMENTS
The authors thank Doug Bonett and Todd
Abraham for their assistance with the confirmatory
factor analyses.
REFERENCES
Anderson CA, Bushman BJ. 2002. Human aggression. Annu Rev
Psychol 53:27–51.
Anderson CA, Dill KE. 2000. Video games and aggressive thoughts,
feelings, and behavior in the laboratory and in life. J Pers Soc
Psychol 78:772–790.
Anderson CA, Huesmann LR. 2003. Human aggression: A
social–cognitive view. In Hogg MA, Cooper J (eds): ‘‘Handbook
of Social Psychology.’’ London: Sage Publications pp 296–323.
Anderson CA, Anderson KB, Deuser WE. 1996. A general frame-
work for the study of affective aggression: Effects of weapons and
extreme temperatures on accessibility of aggressive thoughts,
affect, and attitudes. Pers Soc Psychol Bull 22:366–376.
Anderson KB, Anderson CA. 1997. The predictive specificity of
hostility toward women for male-on-female vs. male-on-male
aggression. Presented at the Annual Convention of the Mid-
western Psychological Association, Chicago, May.
Bardis PD. 1973. Violence: Theory and quantification. J Polit Mil
Sociol 1:121–146.
Barefoot JC. 1992. Developments in the measurement of hostility. In
Friedman H (ed): ‘‘Hostility, Coping, and Health.’’ Washington,
DC: American Psychological Association, pp 13–31.
Bentler P. 1995. ‘‘EQS: Structural Equations Programming Man-
ual.’’ Encino, CA: Multivariate Software.
Bentler PM. 1990. Comparative fit indices in structural models.
Psychol Bull 107:238–246.
Bentler PM, Bonett DG. 1980. Significance tests and goodness-of-fit
in the analysis of covariance structures. Psychol Bull 88:588–600.
Browne MW, Cudeck R. 1993. Alternative ways of assessing model
fit. In Bollen K, Long J (eds): ‘‘Testing Structural Equation
Models.’’ Newbury Park, CA: Sage, pp 136–162.
Burt M. 1980. Cultural myths and supports for rape. J Pers Soc
Psychol 38:217–230.
Bushman BJ. 1995. Moderating role of trait aggressiveness in the
effects of violent media on aggression. J Pers Soc Psychol
69:950–960.
Bushman BJ, Wells GL. 1998. Trait aggressiveness and hockey
penalties: Predicting hot tempers on the ice. J Appl Psychol
83:969–974.
Buss AH, Perry M. 1992. The aggression questionnaire. J Pers Soc
Psychol 63:452–459.
Byrne BM. 1988. Measuring adolescent self-concept: Factorial
validity and equivalency of the SDQ III across gender. Multi-
variate Behav Res 23:361–375.
Caprara GV, Cinanni V, D’Imperio G, Passerini S, Renzi P,
Travaglia G. 1985. Indicators of impulsive aggression: Present
status of research in irritability and emotional susceptibility
scales. Pers Individ Diff 6:665–674.
Caprara GV, Perugini M, Barbaranelli C. 1994. Studies of individual
differences in aggression. In Portegal M, Knutson JF (eds): ‘‘The
Dynamics of Aggression: Biological and Social Processes in
Dyads and Groups.’’ Hillsdale, NJ: Erlbaum, pp 123–153.
Carnagey NL, Anderson CA. in review. Changes in aggressive
behavior and attitudes towards violence as a result of the
September 11, 2001, terrorist attacks in the US.
Chaiken S. 1980. Heuristic versus systematic information processing
and the use of source versus message cues in persuasion. J Pers
Soc Psychol 39:752–756.
Dill KE, Anderson CA, Anderson KB, Deuser WE. 1997. Effects of
aggressive personality on social expectations and social percep-
tions. J Res Pers 31:272–292.
Dodge KA, Crick NR. 1990. Social information-processing bases of
aggressive behavior in children. Pers Soc Psychol Bull 16:8–22.
Elliot DS, Huizinga D, Ageton SS. 1985. ‘‘Explaining Delinquency
and Drug Use.’’ Beverly Hills, CA: Sage.
Epps J, Kendall PC. 1995. Hostile attribution bias in adults.
Cognitive Ther Res 19:159–178.
Esbensen F, Osgood DW. 1999. Gang Resistance Education and
Training (GREAT): Results from the national evaluation. J Res
Crime Delinquency 36:194–225.
Hu L, Bentler PM. 1999. Cutoff criteria for fit indexes in covariance
structure analysis: Conventional criteria versus new alternatives.
Struct Eq Model 6:1–55.
Kishton JM, Widaman KF. 1994. Unidimensional versus domain
representative parceling of questionnaire items: An empirical
example. Educ Psychol Meas 54:757–765.
Lackey C. 2003. Violent family heritage, the transition to adulthood,
and later partner violence. J Fam Issues 24:74–98.
Mackie DM, Smith ER. 1998. Intergroup relations: Insights from a
theoretically integrative approach. Psychol Rev 105:499–529.
Malamuth NM. 1998. The confluence model as an organizing
framework for research on sexually aggressive men: Risk
moderators, imagined aggression, and pornography consump-
tion. In Geen R, Donnerstein E (eds): ‘‘Human Aggression:
Theories, Research, and Implications for Social Policy.’’ San
Diego, CA: Academic Press, pp 230–247.
Marsh HW, Hocevar D. 1985. Application of confirmatory factor
analysis to the study of self-concept: First and higher order factor
models and their invariance across groups. Psychol Bull
97:562–582.
Aggr. Behav. DOI 10.1002/ab
135Attitudes Toward Violence
Marsh HW, Hau K, Balla JR, Grayson D. 1998. Is more ever too
much? The number of indicators per factor in confirmatory factor
analysis. Multivariate Beh Res 33:181–220.
Owens DJ, Strauss MA. 1975. The social structure of violence in
childhood and approval of violence as an adult. Aggr Behav
1:193–211.
Petty RE, Cacioppo JT. 1981. ‘‘Attitudes and Persuasion: Classic
and Contemporary Approaches.’’ Dubuque, IA: William C.
Brown.
Petty RE, Cacioppo JT. 1986. ‘‘Communication and Persuasion:
Central and Peripheral Routes to Attitude Change.’’ New York:
Springer-Verlag.
Rogers RW. 1983. Race variables in aggression. In Geen R,
Donnerstein E (eds): ‘‘Aggression: Theoretical and Empirical
Reviews,’’ Vol. 2. New York: Academic Press, pp 27–50.
Spielberger CD, Jacobs G, Russell S, Crane RS. 1983. Assessment of
anger: The state-trait scale. In Butcher J, Spielberger C (eds).
‘‘Advances in Personality Assessment,’’ Vol. 2. Hillsdale, NJ:
Erlbaum Associates, pp 161–189.
Surgeon General. 2001. ‘‘Youth Violence: A Report of the Surgeon
General.’’ Washington, DC: US Department of Health and
Human Services.
Velicer WF, Huckel LH, Hansen CE. 1989. A measurement model
for measuring attitudes toward violence. Pers Soc Psychol Bull
15:349–364.
Aggr. Behav. DOI 10.1002/ab
136 Anderson et al.
