A Three-Factor Model of Trait Anger: Dimensions of Affect, Behavior

A Three-Factor Model of Trait Anger:
Dimensions of Affect, Behavior,
and Cognition
René Martin
David Watson
University of Iowa
Choi K. Wan
Centers for Disease Control and Prevention
ABSTRACT The structure of trait anger was tested in a study of 24 self-report
scales. Exploratory factor analyses in an undergraduate sample (N = 457)
yielded a two-factor model (comprising cynicism and aggression) and a threefactor model (representing angry emotions, aggressive behaviors, and cynicism). Subsequent evaluations, including confirmatory factor analyses,
indicated that the three-factor model provided the best characterization of the
trait anger domain. The three-factor solution was consistent with an “ABC”
René Martin and David Watson, Department of Psychology, the University of Iowa.
Choi K. Wan, Division of HIV/AIDS Prevention, Centers for Disease Control and
Prevention.
This research is based in part on René Martin’s doctoral dissertation submitted to the
University of Iowa. Portions of this article were presented at the Nags Head Conference
on Personality and Social Behavior, June 18, 1996. This research was supported, in part,
by NIH Grant HL46448. We wish to thank Alan Christensen, Lee Anna Clark, Thomas
Rocklin, Jacob Sines, and Jerry Suls for their comments on a previous version of this
article.
Correspondence should be addressed to René Martin, Department of Psychology, 11
Seashore Hall E, University of Iowa, Iowa City, Iowa 52242; e-mail: rene-martin
@uiowa.edu
Journal of Personality 68:5, October 2000.
Copyright © 2000 by Blackwell Publishers, 350 Main Street, Malden, MA 02148,
USA, and 108 Cowley Road, Oxford, OX4 1JF, UK.
870
Martin et al.
conceptualization of trait anger, consisting of the dimensions of affect, behavior,
and cognition. The three factors showed strikingly different associations with
the Big Five personality traits. Angry Affect was most strongly related to
Neuroticism, whereas Behavioral Aggression was associated with low Agreeableness. Cynical Cognition represented a blend of neurotic and disagreeable
characteristics. Modest mean-level differences were observed between the
genders for each factor.
Trait anger and related constructs have enjoyed considerable attention
from personality and social psychologists in recent years. For example,
65 articles addressing anger, aggression, or cynicism appeared in the
Journal of Personality and the Journal of Personality and Social Psychology in the decade spanning 1988 to 1998. Moreover, anger and aggression scales are included in many popular omnibus personality
inventories, such as the Revised NEO Inventory (Costa & McCrae,
1992), the Personality Research Form (Jackson, 1984), the Schedule for
Nonadaptive and Adaptive Personality (Clark, 1993), and the Multidimensional Personality Questionnaire (Tellegen, in press).
Despite widespread interest in the consequences of trait anger (e.g.,
King, 1997; Martin et al., 1999; Miller, Smith, Turner, Guijarro, & Hallet,
1996; Smith, 1992; Suls, Wan, & Costa, 1995), discussions of anger and
related dimensions typically begin with definitional caveats (e.g., Biaggio & Maiuro, 1985; Smith, 1992; Spielberger, Jacobs, Russell, & Crane,
1983). The complexities in operationalizing anger are so extreme that
Bandura (1973) referred to the domain as a “semantic jungle” (p. 2).
Recent definitional attempts, however, consistently have included three
constructs: anger, hostility, and aggression. For example, Spielberger and
associates (Spielberger, Johnson, Russell, Crane, Jacobs, & Worden,
1985) referred to anger, hostility, and aggression together as the “AHA!
syndrome” (p. 7). Buss and Perry (1992) also provided a definition of
aggression that incorporated both angry emotions and hostile attitudes.
Finally, Barefoot (1992) defined hostility in terms of interrelated elements of cynical beliefs and attributions, angry emotional states, and
aggressive or antagonistic behaviors.
As used by psychologists, the term “anger” usually refers to affective
experience, ranging in intensity from mild annoyance to fury and outrage
(Barefoot, 1992; Buss, 1961; Spielberger et al., 1983). “Hostility” typically characterizes cognitive elements such as attitudes and beliefs; more
specifically, the term “hostility” generally reflects a person’s tendency to
Trait Anger ABCs
871
view the world in a negative, cynical fashion (e.g., Barefoot, 1992; Buss,
1961; Buss & Perry, 1992; Spielberger et al., 1983). Whereas anger is
used to describe the experience of an emotional state that is primarily
intrapersonal, hostility is distinctly “other-directed” and centers on the
relational implications of trait anger. Finally, the term “aggression”
consistently is used to describe behavior (Bandura, 1973; Barefoot, 1992;
Baron, 1977; Berkowitz, 1962; Buss, 1961; Geen, 1990). Aggressive
behavior may be precipitated by anger, but it may be motivated by many
other factors as well.
These definitions suggest that anger, hostility, and aggression describe
related, but distinct domains. Unfortunately, the terms often are used
interchangeably and without clear operationalizations. The subsequent
confusion makes it difficult to identify precisely what is assessed by
various self-report measures of anger and hostility. For example, the same
questionnaires that are identified as measures of anger and hostility by
some researchers (e.g., Biaggio & Maiuro, 1985; Matthews, Jamison, &
Cottington, 1985) also are described as measures of aggression by others
(e.g., Edmunds & Kendrick, 1980; Megargee & Menzies, 1971). The
conceptual ambiguity surrounding these constructs makes it problematic
to draw meaningful conclusions about the social and health consequences
of anger, hostility, and aggression.
We propose that the distinctions among anger, hostility, and aggression
are critical in understanding three unique dimensions assessed by selfreport measures of trait anger. Anger corresponds to affect, aggression to
behavior, and hostility (or cynicism) to cognition. Together the three
constructs form a three-factor “ABC” model of trait anger. The conceptualization of mental activity in terms of affect, behavior, and cognition
has a long history in both philosophy and psychology. For example,
Hilgard (1980) reviewed the role of a taxonomy comprising cognition,
affection, and conation in the history of psychology. McGuire (1969) also
emphasized the distinctions among affect, cognition, and conation,
which he operationalized as overt behavior. We believe that an ABC
perspective can lend clarity to the trait anger domain. Anger-related
emotions, behaviors, and cognitions are likely to be interrelated; however, it is reasonable to expect these dimensions to be separable. For
example, not every angry person (affective dimension) resorts to shouts
or physical violence (behavioral dimension). Similarly, one might hold
cynical interpersonal expectations (cognitive dimension) without feeling
angry. The ABC model is detailed enough to facilitate understanding and
872
Martin et al.
provide predictive specificity in the trait anger domain. The approach is
not so elaborate, however, as to become unwieldy. We now review prior
factor analyses of trait anger measures, which provide some preliminary
support for an ABC model of trait anger.
Prior Factor Analytic Studies of Trait
Anger Measures
Several factor analytic studies of the Buss-Durkee Hostility Inventory
(BDHI) have yielded a two-factor model reflecting the components of
anger experience and anger expression (Bendig, 1962; Bushman, Cooper,
& Lemke, 1991; Buss & Durkee, 1957; Edmunds & Kendrick, 1980;
Sarason, 1961). Costa, McCrae, and Dembroski (1989) also found support for the anger experience and anger expression dimensions in an
item-level factor analysis that included questions extracted from the
Anger Self-Report (ASR; Zelin, Adler, & Myerson, 1972), Multidimensional Anger Inventory (MAI; Siegel, 1986), State-Trait Anger Scale
(Spielberger et al., 1983), and BDHI. Musante, MacDougall, Dembroski,
and Costa (1989) obtained similar results in a scale-level factor analysis
of the ASR, BDHI, MAI, and the Anger Inventory (Novaco, 1975).
Whereas factor analytic studies featuring the BDHI consistently have
yielded components of anger expression and anger experience, the Cook
and Medley Hostility Scale (Ho Scale; Cook & Medley, 1954) seems to
measure a rather different aspect of trait anger. Costa, Zonderman,
McCrae, and Williams (1986) extracted two factors from the Ho Scale.
The first factor, “Cynicism,” reflected a negative outlook on the world
and other people and the second factor, “Paranoid Alienation,” was
marked by items describing feelings of persecution and emotional
isolation.
As previously described, Musante et al. (1989) found support for the
anger experience–anger expression model. Their report, however, included a discussion of a three-factor solution that ultimately was rejected
as incompatible with their theoretical presentation. The third factor of the
rejected solution focused on interpersonal attitudes and feelings of suspicion, guilt, and mistrust. This hostile attitude factor appears to be
conceptually similar to the cynicism component measured by the Ho
Scale, and suggests that other trait anger questionnaires also may measure anger-related cognitions, in addition to anger expression and anger
experience.
Trait Anger ABCs
873
Finally, some factor analyses have included trait anger questionnaires
with inventories measuring potentially related constructs, such as the
Type A behavior pattern and hardiness (Friedman, Tucker, & Reise, 1995;
Miller, Jenkins, Kaplan, & Salonen, 1995). In confirmatory factor analyses that combined the Anger Expression (AX) scale (Spielberger et al.,
1985), a brief version of the MAI, a short form of the Ho Scale, and a
selection of hostility-related items from several Type A behavior scales,
Miller et al. (1995) obtained eight scales (Anger-out, Anger Control,
Anger Frequency, Ease of Anger Provocation, Brooding, Hostile Outlook, Cynicism, and Sulleness). These results are somewhat difficult to
interpret, however. Several scales were substantially interrelated, indicating that they probably reflected a smaller number of underlying higherorder dimensions. In addition, a small group of trait anger measures was
used. These measures, which were selected for their potential to predict
cardiovascular disease, may not have represented the trait anger domain
adequately.
Friedman et al. (1995) included four trait anger measures—Ho Scale,
BDHI, State-Trait Anger Scale, and AX—in exploratory factor analyses
of personality scales selected for their relevance to health. Measures of
Type A behavior, anxiety, depression, optimism, and hardiness also were
used, as well as the Eysenck Personality Questionnaire (Eysenck &
Eysenck, 1975). Five factors were obtained: Neuroticism, Anger-out,
Alienation-Suspicion, Introversion, and Conscientiousness. The Angerout factor was similar to the anger expression component obtained in
factor analyses featuring the BDHI (Bendig, 1962; Bushman et al., 1991;
Buss & Durkee, 1957; Costa et al., 1989; Musante et al., 1989). The
Alienation-Suspicion factor was marked by the Ho Scale; this factor
appeared to approximate the cynicism component observed in previous
studies (Costa et al., 1986; Musante et al., 1989). Anger experience did
not emerge as a discrete factor in Friedman et al.’s study; however, scales
related to the experience of angry emotions (e.g., BDHI Irritability and
Resentment, AX Anger-in) were grouped together as markers of the
broader Neuroticism factor. Thus, Friedman et al. (1995) provided evidence consistent with an ABC conceptualization of trait anger. Results
from Friedman et al. should be interpreted cautiously, however, as it was
not a study of trait anger per se. Variables were selected for their relation
to health outcomes; as such, a small group of trait anger questionnaires
was factor analyzed along with several measures unrelated to the hostility
or anger domain.
874
Martin et al.
Limitations of Prior Studies
These prior factor analytic studies suggest the presence of at least three
distinct dimensions within the domain of trait anger. Moreover, the three
most commonly identified dimensions appear to be broadly consistent
with an ABC model. It is noteworthy, however, that with the exception
of Musante et al. (1989), no investigators actually have reported a
three-factor structure that closely resembles the ABC model. In addition,
previous factor analytic research in this domain is limited in several ways.
First, only small subsets of the available trait anger questionnaires have
been assessed in these studies. Moreover, previous factor analyses have
focused exclusively on isolated scales that were developed with the sole
intent of assessing trait anger. Relevant measures of anger, hostility, and
aggression that are included in broader multidimensional inventories of
personality and emotionality have been neglected (e.g., the NEO Personality Inventory–Revised, NEO PI-R; Costa & McCrae, 1992).
Second, previous investigators in this area have tended to assume that
popular trait anger scales were reliable. Unfortunately, many of these
scales are heterogeneous and lack internal consistency. For example,
investigators typically report Cronbach’s alpha values of .80 or greater
for the Ho Scale (e.g., Cook & Medley, 1954; Smith & Frohm, 1985).
However, a decomposition of these values reveals unacceptably low
mean interitem correlations of .10 or less, indicating that the scale is too
diverse to be represented meaningfully by a single score (see Briggs &
Cheek, 1986, and Clark & Watson, 1995). This perspective is consistent
with Contrada and Jussim’s (1992) contention that the Ho Scale “may
lack the coherent internal structure required for measuring distinctive
psychological traits” (p. 615). Steinberg and Jorgensen (1996) similarly
noted the multidimensionality of the Ho Scale.
A final limitation of previous work in this area is that it has failed to
specify exactly how the components of trait anger fall within broader
structural frameworks of personality, such as the five-factor model. This
issue is critically important because several trait anger questionnaires,
such as the Ho Scale (Blumenthal, Williams, Kong, Schanberg, &
Thompson, 1978; Carmody, Crossen, & Weins, 1989), BDHI (Sarason,
1961; Siegman, Dembroski, & Ringel, 1987), MAI (Siegel, 1986), and
AX (Spielberger et al., 1985), are known to be positively correlated with
Neuroticism. This association is of particular interest to health psychologists, because it has been suggested that neurotic hostility is unassociated
Trait Anger ABCs
875
with disease, whereas other facets of hostility are considered pathogenic
(Costa et al., 1989; Siegman et al., 1987). Consequently, it is important to
know which aspects of trait anger are strongly saturated with Neuroticismrelated variance. To date, the best evidence along these lines was reported
by Costa et al. (1989), who linked the affective, anger experience dimension of trait anger to Neuroticism, and further proposed that the behavioral component was related to Agreeableness versus Antagonism. Few
data, however, on the association between Agreeableness and trait anger
are available in the existing literature. Similarly, little evidence is available regarding how the cynical cognition component relates to the higher
order traits of personality. Measures such as the Ho Scale, which are
thought to tap cynicism (Costa, Zonderman, McCrae, & Williams, 1985;
Johnson, Null, Butcher, & Johnson, 1984), are known to be correlated
with Neuroticism. Negative attitudes toward others, however, including
cynicism and callousness, would seem to be closely linked to low
Agreeableness. At a conceptual level, therefore, it seems likely that the
cognitive component of trait anger should be related to both Neuroticism
and (low) Agreeableness.
The Present Research
As we have seen, although several studies have found support for a
two-factor model of anger experience (i.e., affect and expression), other
evidence suggests the possible presence of an additional dimension of
cynicism (i.e., cognition). Note, moreover, that earlier studies (a) failed
to assess the domain comprehensively and (b) included unreliable, heterogeneous scales. It therefore remains possible that a comprehensive
analysis of the domain actually might yield additional dimensions beyond these three. The primary goal of this research was to identify the
structure of trait anger, as assessed by self-report inventories. In pursuing
this goal, we asked a large sample of undergraduates to complete a
comprehensive battery of trait anger questionnaires, as well as two
inventories assessing the Big Five traits. In selecting trait anger measures,
we included questionnaires commonly used in personality, social, and
health psychology. These inventories, including the Ho Scale (Cook &
Medley, 1954), BDHI (Buss & Durkee, 1957), Aggression Questionnaire
(AQ, Buss & Perry, 1992), ASR (Zelin et al., 1972), MAI (Siegel, 1986),
Trait Anger Scale (TAS; Spielberger et al., 1983), and AX (Spielberger
et al., 1985), were constructed by the authors as discrete measures of trait
876
Martin et al.
anger. In an effort to adequately capture the trait anger domain, we also
administered six trait anger scales drawn from broad affect and personality inventories, including the Angry Hostility, Trust, and Compliance
facet scales from the NEO PI-R (Costa & McCrae, 1992), the Hostility
scale from the Expanded form of the Positive and Negative Affect
Schedule (PANAS-X; Watson & Clark, 1994), and the Aggression and
Mistrust scales from the Schedule for Nonadaptive and Adaptive Personality (SNAP; Clark, 1993). Together, these measures provided a reasonably comprehensive assessment of the domain.
The study proceeded through three major phases. First, we conducted
preliminary psychometric analyses on each trait anger scale. The purpose
of these analyses was to assure that we were working with variables that
were reasonably homogeneous and independent. Second, we subjected
the trait anger scales to exploratory factor analyses. Exploratory analyses
were necessary at this stage because (a) the previous literature does not
clearly establish the structure of trait anger and (b) several of the assessed
scales had not been included in any prior structural analyses of this
domain. These analyses yielded two interpretable models of trait anger.
Finally, confirmatory factor analyses then were conducted to compare
the relative fit of these models. The ultimate purpose of the study was to
clarify a confusing literature by identifying the constructs assessed in
extant trait anger measures.
METHOD
Participants
Undergraduate students (N = 459) received partial course credit for their
participation. Data from two participants were omitted from analyses because
of failure to follow instructions. The final sample was composed of 457
participants; 242 participants (53%) were women. Participants ranged from 18
to 32 years of age, with a mean of 19.02 years (SD = 1.49). Most participants
(n = 408, 89%) were Caucasian; 26 participants (6%) were Asian or Asian
American, 16 (4%) were African American, and seven participants (1%) declined to report their ethnic background.
Trait Anger Measures
Preliminary psychometric analyses. Participants responded to several trait anger questionnaires, which are described in detail below. As previously discussed,
Trait Anger ABCs
877
prior investigations in this area often have overlooked issues of internal reliability. Consequently, extensive preliminary psychometric analyses were conducted
for each measure. The goal of these preliminary analyses was to identify
potentially heterogeneous measures and to rescore them to maximize the
interpretability of the constructs assessed. All measures initially were scored in
the hostile direction according to their authors’ original instructions (except the
ASR; see below). Descriptive and reliability statistics for these measures are
reported in Table 1. Optimal reliability was operationalized as a coefficient alpha
of .70 or greater, with a mean interitem correlation between .20 and .40. Table 1
illustrates that most scales demonstrated satisfactory internal consistency and
these reliable scales were retained in further analyses. Preliminary analyses,
however, revealed significant reliability problems with three instruments—the
Ho Scale, BDHI, and MAI. In each case, the instrument in question was
subjected to exploratory factor analyses, with an evaluation of the full range of
possible solutions. The results then were used to develop scoring formats that
offered improved reliability and interpretability. Because of scoring difficulties
associated with the ASR (see below), this measure also was subjected to
exploratory analyses and scale development. Table 2 presents the descriptive
and reliability statistics and representative items for the newly developed scales.
Additional information regarding the preliminary psychometric analyses is
briefly presented below.1
Cook and Medley Hostility (Ho) Scale. The Ho Scale (Cook & Medley, 1954)
consists of 50 true/false items, typically represented by a composite score. As
shown in Table 1, the coefficient alpha associated with the composite scale was
satisfactory; however, the mean interitem correlation was unacceptably low.
Issues of internal consistency were not resolved using rationally derived scales
developed by Barefoot and associates, also shown in Table 1 (Barefoot, Dodge,
Peterson, Dahlstrom, and Williams, 1989). Exploratory factor analyses of the
Ho Scale items yielded a two-factor solution, which provided the basis for
scoring two new scales (Antagonism and Alienation). As reported in Table 2,
the mean interitem correlations for these newly developed scales remained
somewhat suboptimal, but the reliability of the both scales represented an
improvement over the composite Ho Scale and most of Barefoot et al.’s (1989)
rationally derived scales. Furthermore, the Antagonism and Alienation scales
did not closely replicate the two-factor Ho Scale solution reported by Costa et
al. (1986).
Buss-Durkee Hostility Inventory (BDHI). The BDHI (Buss & Durkee, 1957)
consists of 75 true/false items, 66 of which assess various aspects of hostility.
1. Items, scoring format, and further analytic information regarding the newly developed
scales are available from the first author upon request.
878
Martin et al.
Table 1
Study 1: Descriptive and Reliability Statistics for Trait Anger Measures
Scale
Ho Composite (50)
Ho Cynicism (13)
Ho Hostile Attribution (12)
Ho Hostile Affect (5)
Ho Aggressive Responding (9)
Ho Social Avoidance (4)
BDHI Composite (66)
BDHI Assault (10)
BDHI Indirect Hostility (9)
BDHI Negativity (5)
BDHI Irritability (11)
BDHI Resentment (8)
BDHI Suspicion (10)
AQ Physical Aggression (9)
AQ Verbal Aggression (5)
AQ Anger (7)
AQ Hostility (8)
MAI Composite (25)
MAI Anger-arousal (8)
MAI Range of Situations (7)
MAI Hostile Outlook (4)
MAI Anger-in (5)
MAI Anger-out (2)
TAS Angry Temperament (4)
TAS Angry Reaction (4)
AX Anger-in (8)
AX Anger-out (8)
PANAS-X Hostility (6)
SNAP Mistrust (19)
SNAP Aggression (20)
NEO PI-R Angry Hostility (7)
NEO PI-R Trust (6)
NEO PI-R Compliance (5)
M
SD
α
Mr
21.81
6.83
4.41
2.33
3.76
1.63
30.04
3.50
4.90
2.63
5.59
2.91
3.29
20.11
14.41
16.03
20.14
70.54
17.96
23.42
12.72
13.17
6.32
6.59
9.41
16.90
16.14
12.76
5.92
4.61
12.32
15.91
8.75
7.94
2.81
2.54
1.33
1.95
1.11
10.58
2.70
2.09
1.30
2.45
2.02
2.35
7.57
4.04
5.44
6.02
14.20
6.19
4.77
3.22
3.63
1.85
2.76
2.69
4.39
4.02
4.76
4.32
4.12
4.57
3.31
3.23
.85
.70
.66
.47
.57
.32
.89
.81
.63
.47
.66
.65
.68
.86
.76
.83
.79
.89
.87
.77
.67
.66
.67
.89
.78
.78
.78
.81
.84
.85
.77
.81
.59
.10
.16
.14
.15
.12
.11
.10
.29
.16
.15
.15
.18
.18
.41
.39
.42
.32
.45
.33
.33
.26
.51
.24
.68
.47
.31
.31
.41
.22
.24
.32
.42
.21
Note. The number of items appears in parentheses after each scale. M r = mean interitem
correlation.
Trait Anger ABCs
879
Table 2
Newly Developed Scales: Descriptive and Reliability Statistics and
Sample Items
Scale
M
Ho Scale Antagonism (17)
8.51
“I am often inclined to go out of my
way to win a point with someone
who has opposed me.”
Ho Scale Alienation (16)
6.65
“I am not likely to speak to people
until they speak to me.”
BDHI Physical Aggression (15)
6.61
“If somebody hits me first,
I let him have it.”
BDHI Cynicism-Mistrust (19)
6.54
“I know people tend to talk about
me behind my back.”
BDHI Angry Reactivity (13)
7.66
“When I am angry, I sometimes sulk.”
ASR Anger Expression (17)
–8.12
“I have physically hurt someone
in a fight.”
ASR Neuroticism (16)
–13.84
“I never do anything right.”
ASR Anger Valuation (12)
10.60
“People should never get irritated.”
ASR Cynicism-Mistrust (8)
–3.71
“Often people are friendly when
they want something. . .”
MAI
56.59
“I am surprised at how often I
feel angry.”
SD
α
Mr
3.59
.75
.15
3.41
.74
.15
3.71
.81
.22
4.13
.80
.18
2.68
.65
.13
17.84
.85
.26
13.96
.81
.21
8.98
.68
.14
8.58
.77
.30
13.72
.90
.30
Note. The number of items appears in parentheses after each scale. M r = mean interitem
correlation.
(The nine remaining items measure guilt.) The BHDI often is scored as a
composite index. Furthermore, the hostility items sometimes are grouped into
six rationally constructed scales (e.g., Resentment, Irritability). As reported in
Table 1, the BDHI composite and most of its traditional subscales were not
optimally reliable. In an attempt to improve internal consistency, exploratory
factor analyses were conducted on the 66 BDHI hostility items. A three-factor
880
Martin et al.
solution was adopted, and three new scales, Physical Aggression, CynicismMistrust, and Angry Reactivity, were developed (see Table 2).
Aggression Questionnaire (AQ). The AQ (Buss & Perry, 1992) includes 29
items, rated from 1 (extremely uncharacteristic of me) to 5 (extremely characteristic of me). Four scales were scored: Physical Aggression (nine items), Verbal
Aggression (five items), Anger (seven items), and Hostility (eight items). The
item content of the Physical and Verbal Aggression scales is reasonably clear
from the scale titles. The names of the “Anger” and “Hostility” scales may be
confusing, however, because these two terms are often used interchangeably.
Anger scale items primarily describe the experience of angry emotion (e.g., “I
sometimes feel like a powder keg ready to explode”). In contrast, Hostility scale
items relate to cynical, mistrustful interpersonal attitudes (e.g., “I am suspicious
of overly friendly strangers”).
Anger Self-Report (ASR). The ASR (Zelin et al., 1972) consists of 64 items,
rated on a 6-point scale ranging from –3 (strong disagreement) to +3 (strong
agreement). Unfortunately, scoring instructions provided by the ASR’s authors
presented several ambiguities that made it difficult to score or evaluate the
instrument with any confidence. In addition, the internal structure of the ASR
had never been examined using factor analysis. The ASR therefore was subjected to exploratory factor analyses, which led to the development of four
scales—Anger Expression, Cynicism, Neuroticism, and Anger Valuation. Note
that Neuroticism scale items were scored in the neurotic direction and high
scores on the Anger Valuation scale reflected more positive evaluations of anger.
Multidimensional Anger Inventory (MAI). The MAI (Siegel, 1986) includes 38
items that are rated on a scale ranging from 1 (completely undescriptive) to 5
(completely descriptive). Siegel (1986) summed 25 of these items to create a
total MAI score, and also developed five subscales using factor analysis. As
shown in Table 1, initial psychometric analyses indicated that the alpha reliabilities for three of these subscales were below .70. In addition, the mean interitem
correlations for two scales were above .40, suggesting some redundancy in item
content. Finally, several of the subscale correlations were large (e.g., Anger-in
correlated .66 with Anger-Arousal, and Hostile Outlook correlated .67 with
Range of Situations). Preliminary exploratory factor analyses led us to construct
a single internally reliable scale consisting of 22 items (see Table 2).
Trait Anger Scale (TAS). The TAS (Spielberger et al., 1983) is comprised of 10
items, which are rated from 1 (almost never) to 4 (almost always). Two scales
were scored: Angry Temperament (4 items) and Angry Reaction (4 items). The
Angry Temperament scale is characterized by items such as “I have a fiery
temper,” whereas Angry Reaction items have an interpersonal component (e.g.,
“I get angry when slowed down by others’ mistakes”).
Trait Anger ABCs
881
Anger Expression (AX) Scale. The AX Scale (Spielberger et al., 1985) includes
20 items, rated from 1 (almost never) to 4 (almost always). Two scales were
scored: Anger-in (eight items) and Anger-out (eight items). The AX Anger-in
scale includes items related both to the non-expression of anger (e.g., “I boil
inside, but don’t show it”) as well as to generalized distress (e.g., “I withdraw
from people”; “I pout or sulk”). Previous research (Martin et al., 1999; Martin
& Watson, 1997) has indicated that AX Anger-in is related diffusely to negative
affect and Neuroticism, and is not a specific index of anger per se. In contrast,
most AX Anger-out scale items are directly related to overt acts of aggression.
NEO PI-R Angry Hostility, Trust, and Compliance Facet Scales. Th e Angry
Hostility, Trust, and Compliance facet scales were extracted from the NEO PI-R
(Costa & McCrae, 1992). Each scale is composed of eight items; items are rated
on a 5-point scale, ranging from strongly agree to strongly disagree. The Angry
Hostility scale is a facet of the Neuroticism domain and was constructed to
measure anger experience. The Trust scale assesses attitudes toward others,
contrasting perceived trustworthiness with dishonesty. The Compliance scale
measures aggressive behavioral responses to conflict. The Trust and Compliance
scales are facets of the Agreeableness domain, so that hostile responses receive
lower scores on both scales. It should be noted that one item from the Angry
Hostility scale, two Trust items, and three Compliance items also appear in
another of the personality inventories completed by participants, the NEO Five
Factor Inventory (NEO-FFI; Costa & McCrae, 1992). Because retention of the
duplicated items would spuriously increase correlations between trait anger and
the Neuroticism and Agreeableness domains of the NEO-FFI, these duplicated
items were deleted from the facet scales prior to scoring.
PANAS-X Hostility Scale. The Hostility scale of the PANAS-X (Watson &
Clark, 1994) consists of six terms assessing anger-related affect, such as “angry”
and “irritable.” Participants were instructed to rate how well each term described
their usual feelings on a 1 (very slightly or not at all) to 5 (extremely) scale.
SNAP Aggression and Mistrust Scales. Two scales, Aggression (20 items) and
Mistrust (19 items), were extracted from the SNAP (Clark, 1993). SNAP
Aggression scale items pertain to aggressive acts and angry emotions, whereas
the SNAP Mistrust scale measures negative interpersonal attitudes. Both scales
are composed of true/false items.
Big Five Measures
In addition to the trait anger scales, participants completed two measures of the
five-factor model of personality: the Big Five Inventory (BFI; John, Donahue,
& Kentle, 1991) and the NEO-FFI (Costa & McCrae, 1992). Two Big Five
882
Martin et al.
indices were used to provide a comprehensive assessment of the personality
dimensions. The 44 BFI items are rated from 1 (very uncharacteristic of me) to
7 (very characteristic of me). The BFI Openness scale includes 10 items, the
Agreeableness and Conscientiousness scales each include nine items, and the
Neuroticism and Extraversion scales are comprised of eight items each. The
NEO-FFI is derived from the longer NEO PI-R. It includes 60 items (12 for each
scale) that are rated from 1 (strongly agree) to 5 (strongly disagree). Previous
research has demonstrated that corresponding scales from the BFI and NEO
PI-R are strongly convergent with each other (Watson, Clark, & Harkness,
1994). All BFI and NEO-FFI scales demonstrated good internal reliability.
Procedure
Participants reported to the laboratory in groups of 10 to 20. After informed
consent was obtained, participants completed the individual difference questionnaires. The order of instruments was arranged so that trait anger inventories
were separated by other types of individual difference questionnaires as much
as possible. Given this restriction, the order of questionnaires was counterbalanced across participants. Most participants required approximately 90 min
to complete the questionnaires; all participants took a brief break halfway
through the session. Participants received partial course credit and nominal
monetary compensation for their efforts.
RESULTS
Exploratory Factor Analyses
Principal axis factor analyses were conducted on the 24 trait anger scales.
We inspected the eigenvalues for the first several unrotated factors; the
eigenvalues were 10.86, 3.09, 1.43, 1.03, 0.83, and 0.70 for the first
through sixth factors. The full range of solutions, up to and including five
factors, was evaluated. Both oblique and orthogonal rotations were
examined, and the two approaches generated convergent, virtually identical solutions. We begin by presenting the results using a Varimax
rotation for the two- through five-factor solutions. Oblique rotations are
discussed subsequently. A marker was defined as any scale that loaded
|.30| or greater on a factor and had its highest loading on that factor; the
number of markers (out of 24 scales) for each factor in each solution is
presented in Table 3. The most noteworthy aspect of these data is that all
factors beyond the third were small and poorly defined, suggesting that
a maximum of three factors be retained.
Trait Anger ABCs
883
Table 3
Number of Markers for One- Through Five-Factor Trait Anger Solutions
Number of markers for factor number
Number of factors
in solution
1
2
3
4
5
1
2
3
4
5
24
13
11
11
9
11
8
7
7
5
5
6
1
1
1
All 24 trait anger scales had loadings of greater than |.30| on the
single-factor solution, indicating the presence of a large general factor.
Because our primary goal was to explore conceptual distinctions among
various types of trait anger, however, more complex solutions were
considered. Both the two- and three-factor solutions were interpretable
and are discussed further. The four- and five-factor solutions are not
presented in detail, as they each yielded dimensions that were defined by
a single marker.
The first factor of the two-factor solution was composed primarily of
scales related to cynicism and mistrust (e.g., BDHI Cynicism-Mistrust,
SNAP Mistrust). This factor, however, included two scales (PANAS-X
Hostility and BDHI Angry Reactions) that focused primarily on the
experience of angry feelings rather than cynicism per se. We named this
first factor “Cynicism/Anger.” Scales related to anger expression and
behavioral aggression (e.g., ASR Anger Expression, BDHI Physical
Aggression) made up the second factor of the two-factor solution. This
factor also included three scales (AQ Anger, TAS Angry Temperament,
and NEO PI-R Angry Hostility) relevant to anger experience. Thus, the
second factor was labeled “Aggression/Anger.” The two-factor solution
is presented in Table 4.
When three factors were extracted, the scales related to angry affect
and experience formed their own cluster, distinct from the cynicism and
behavioral aggression scales. As such, the three-factor solution closely
resembled the proposed ABC structure of trait anger. The cognitive
dimension of the ABC model was represented by the first factor (“Cynical Cognition”), which was composed of cynicism-mistrust scales. The
884
Martin et al.
Table 4
Two-Factor Solution for Trait Anger Scales
Trait Anger Factor
Scale
Cynicism/Anger
Aggression/Anger
BDHI Cynicism-Mistrust
SNAP Mistrust
Ho Scale Alienation
AQ Hostility
MAI
ASR Cynicism-Mistrust
AX Anger-in
ASR Neuroticism
PANAS-X Hostility
NEO PI-R Trust
Ho Scale Antagonism
TAS Angry Reactivity
BDHI Angry Reactivity
.86
.82
.81
.77
.72
.69
.69
.63
.55
–.53
.48
.48
.41
.18
.12
.08
.25
.49
.22
.03
.18
.45
–.19
.45
.36
.37
ASR Anger Expression
BDHI Physical Aggression
AQ Physical Aggression
AX Anger-out
NEO PI-R Compliance
SNAP Aggression
AQ Anger
TAS Angry Temperament
AQ Verbal Aggression
NEO PI-R Angry Hostility
ASR Anger Valuation
.14
.10
.18
.23
–.17
.32
.40
.35
.20
.54
.03
.86
.82
.75
.75
–.75
.74
.68
.64
.63
.58
.40
second factor (“Behavioral Aggression”), marked by anger expression
and aggression scales, represented the model’s behavioral component.
Finally, the affective domain of the ABC model was represented by the
third factor (“Angry Affect”), which included scales pertaining to the
experience of angry emotion. The three-factor solution is presented in
Table 5.2
2. Note that in both the two- and three-factor solutions, we opted to apply the label
“cynicism” to the dimensions representing anger-related cognitions. This label may not
Trait Anger ABCs
885
Choosing between the two- and three-factor solutions. Four main issues, detailed below, were considered in evaluating the relative merit of
the two- versus three-factor solutions. First, we compared our two- and
three-factor solutions to the extant literature. Second, we evaluated the
average intermarker correlations both between and within factors of
both solutions. Third, the associations between each model and the Big
Five traits were explored. Finally, confirmatory factor analyses were
conducted to evaluate the relative fit of the two- versus three-factor
solutions.3
Comparing the two- and three-factor models to the trait anger literature.
Our two-factor solution was decidedly different from the two-factor
representations of trait anger reported by previous researchers. As reviewed earlier, a number of previous factor analyses (Bendig, 1962;
Bushman et al., 1991; Buss & Durkee, 1957; Costa et al., 1989; Musante
et al., 1989), based on questionnaires such as the BDHI, ASR, MAI, TAS,
and Novaco Anger Inventory (Novaco, 1975), resulted in two interpretable factors, representing anger expression and anger experience. Our
two-factor solution, however, clearly departs from earlier findings in that
one of the factors is dominated by scales assessing cynicism, rather than
anger experience. The present study apparently represents the first factor
analysis of trait anger measures to yield a two-factor solution reflecting
the basic dimensions of cynicism and behavioral aggression. The unusual
character of the two-factor solution suggested that our three-factor solution actually might be more compatible with the literature.
fully characterize the subtleties of anger-related cognitions. It does, however, capture the
fundamental theme of interpersonal distrust and disparagement that characterizes the
relevant scales (e.g., BDHI Cynicism-Mistrust, SNAP Mistrust). In addition, the cynicism
label also is consistent with common usage among personality, social, and health
psychologists (e.g., Costa et al., 1986).
3. The issue of model replicability also was considered in evaluating the three-factor
model. The data were separated on the basis of subject sex. Exploratory factor analyses
of the male and female data sets yielded virtually identical three-factor solutions. The
convergence between the male and female three-factor solutions then was examined
quantitatively using regression-based scoring weights (Gorsuch, 1983; Harman, 1976).
The results indicated that the three-factor model of trait anger was replicable and that the
structure of trait anger was highly robust across gender.
886
Martin et al.
Table 5
Three-Factor Solution for Trait Anger Scales
Trait Anger Factor
Cynical
Cognition
Behavioral
Aggression
Angry
Affect
BDHI Cynicism-Mistrust
SNAP Mistrust
Ho Scale Alienation
ASR Cynicism-Mistrust
AQ Hostility
AX Anger-in
MAI
NEO PI-R Trust
ASR Neuroticism
Ho Scale Antagonism
TAS Angry Reactivity
.84
.83
.81
.75
.71
.62
.59
–.57
.56
.50
.41
.16
.12
.07
.25
.18
–.04
.32
–.23
.09
.47
.27
.24
.16
.16
.05
.31
.27
.58
–.04
.34
.14
.33
ASR Anger Expression
BDHI Physical Aggression
AQ Physical Aggression
NEO PI-R Compliance
SNAP Aggression
AX Anger-out
AQ Verbal Aggression
ASR Anger Valuation
.16
.14
.21
–.15
.26
.15
.17
.00
.90
.89
.78
–.70
.65
.63
.59
.36
.16
.09
.13
–.27
.40
.44
.25
.16
NEO PI-R Angry Hostility
TAS Angry Temperament
AQ Anger
BDHI Angry Reactivity
PANAS X Hostility
.37
.17
.24
.27
.45
.35
.42
.47
.17
.32
.72
.69
.68
.57
.47
Scale
Average intermarker correlations. All factor analytic techniques attempt to summarize the pattern of covariation among measures using a
much smaller number of latent dimensions (Floyd & Widaman, 1995;
Watson, Clark, & Harkness, 1994). Consequently, another approach to
evaluating the relative merit of the two- versus three-factor solutions is
to examine the average intermarker correlations both within and between
factors. Following the general logic of factor analysis, a clear, well-defined
Trait Anger ABCs
887
solution (i.e., one that summarizes the pattern of covariation particularly
well) should maximize the correlations among the markers defining the
same factor, while at the same time minimizing the correlations among
variables defining different factors.
We therefore computed the average intermarker correlations for both
the two- and three-factor solutions (in these analyses, the correlations
were subjected to an r to z transformation before averaging). For the
two-factor solution, the mean intermarker correlation within factors was
.51. The mean correlation between markers of the cynicism/anger experience and aggression/anger experience factors was notably lower (.34),
demonstrating that at least two well-defined variable clusters could be
identified in these data.
The three-factor solution offered a modest improvement in the mean
within-factor correlations, along with a corresponding small increase in
the between-factor values. Specifically, the mean correlation among
variables defining the same factor was .54 for the three-factor solution.
In contrast, the cross-factor correlations were substantially lower, with a
mean intermarker correlation between factors of .38. Although a comparison of the average intermarker correlations did not clearly indicate
that one solution was superior to the other, the analyses of the three-factor
solution demonstrated that three well-defined variable clusters could be
identified in these data. In other words, the average intermarker correlations revealed that the three-factor solution captured meaningful conceptual distinctions within this array of trait anger scales.
Associations with the Big Five traits. To compare and contrast the twoand three-factor solutions further, scores on each factor were correlated
with the Big Five dimensions of personality. As described in the Method
section, participants completed two Big Five inventories, the NEO-FFI
(Costa & McCrae, 1992) and the BFI (John et al., 1991). Five general
trait scores (Neuroticism, Extraversion, Openness to Experience, Agreeableness, and Conscientiousness) were scored from each of these instruments. A principal axis factor analysis then was conducted on the 10 scale
scores. Five factors, clearly representing the five general domains of
personality, were extracted and rotated using Varimax. Next, regressionbased factor scores were computed to measure each of the five broad
dimensions. These scores, in turn, were correlated with regression-based
scores assessing the trait anger factors that had emerged in both the twoand three-factor solutions. Table 6 presents the correlations between the
888
Martin et al.
Big Five factor scores and factor scores from both the two- and threefactor solutions.
In both the two- and three-factor solutions, aggression was strongly
associated (r = –.61 and r = –.58, respectively) with low levels of
Agreeableness. In the two-factor solution, cynicism was strongly correlated with Neuroticism (r = .53) and more moderately related to Agreeableness (r = –.41). The relations between cynicism, Neuroticism, and
Agreeableness shifted substantially, however, when a third factor was
extracted. Specifically, in the three-factor solution, the correlations between cynicism and Neuroticism (r = .37) and cynicism and Agreeableness (r = –.41) were nearly equal. In this solution, it was the anger
experience factor that was strongly correlated with Neuroticism (r = .58),
and more modestly related to Agreeableness (r = –.27).
The pattern of associations between cynicism, anger experience,
Neuroticism, and Agreeableness should be of considerable interest to
health psychologists. Individuals with high Neuroticism scores are
known to offer frequent somatic complaints, but do not appear to suffer
greater morbidity or mortality than individuals who are low in Neuroticism (Costa & McCrae, 1987; Watson & Pennebaker, 1989). Consequently, it is important for researchers attempting to examine the
contribution of hostility to cardiovascular and other physical disorders to
be able to separate measures tapping Neurotic aspects of trait anger from
those related to low Agreeableness. This line of reasoning suggests that
the two-factor solution obscured potentially important information by
combining content related to Cynicism and Anger Experience, because
only the latter is strongly saturated with Neuroticism-related variance.
Relative fit of the two- versus three-factor models. As a final step in
evaluating the merits of the three-factor model, confirmatory factor
analyses were conducted. The results of the exploratory factor analyses
previously described were used to generate the two- and three-factor
models to be tested. More complex models were not tested because the
exploratory factor analyses yielded four- and five-factor solutions that
included extremely small, uninterpretable factors. Before proceeding, the
data were examined for multivariate normality in order to establish the
appropriateness of using a maximum likelihood solution in the confirmatory factor analyses. The skewness ranged from –0.32 to 1.40 and the
kurtosis ranged from –1.01 to 1.42. The data did not depart from
Trait Anger ABCs
889
Table 6
Zero-Order Correlations Between Two- and Three-Factor Trait Anger
Solutions and Big Five Factor Scores
Two-Factor Solution
Cynicism/Anger
Aggression/Anger
Three-Factor Solution
Cynical Cognition
Behavioral Aggression
Angry Affect
N
E
O
A
C
.53
.07
–.23
.17
–.14
.03
–.41
–.61
–.14
–.12
.37
–.14
.58
–.24
.15
.04
–.11
.06
–.08
–.41
–.58
–.27
–.13
–.11
–.07
Note. Correlations > .09 are statistically significant at the .05 level. N = Neuroticism; E =
Extraversion; O = Openness; A = Agreeableness; C = Conscientiousness. N = 457.
multivariate normality and a maximum likelihood estimation was
deemed appropriate.
The adequacy of fit of each model was evaluated and then the relative
fit of the two- and three-factor models was compared. All analyses were
conducted with LISREL 8. A two-factor model was formulated using the
results obtained from the two-factor solution. A three-factor model was
constructed in a similar fashion. Scales were allowed to load on multiple
factors. The fit indices were stronger for the three-factor model, χ2(239)
= 1056.95, p < .0001, CFI = .90, GFI = .83, than for the two-factor model,
χ2(242) = 1518.83, p < .0001, CFI = .84, GFI = .72.
Nested χ2 tests were conducted to assess the relative fit of the twoversus the three-factor model. When the two-factor model was compared to the three-factor model, the difference was statistically
significant, χ2(3) = 461.88, indicating that the three-factor model’s
representation of the observed correlation matrices was superior to that
associated with the two-factor model. The anger experience factor correlated .47 and .53 respectively with the aggression and cynicism factors.
The correlation between the aggression and cynicism factors was .43.
Additional Issues
Three-factor solution with oblique rotation. To this point, all analyses
were based on orthogonal Varimax rotations. Anger experience, aggression,
890
Martin et al.
and cynicism, however, all reflect a common underlying dimension of
anger and hostility; thus, one would expect these different manifestations
of hostility to be intercorrelated. As such, it is possible that an oblique
rotation might offer greater clarity in representing the structure of trait
anger. To explore this possibility, the three-factor principal axis factor
analysis was conducted once again on the 24 trait anger scales. An
Oblimin rotation was used, with a delta value of zero. Markers were
defined as before. The three-factor oblique solution was highly similar
to its orthogonal counterparts, which is consistent with previous findings
that factor structures are highly robust across different types of rotation,
given a large sample and well-defined factors (Goldberg, 1990; Guadagnoli & Velicer, 1988; Snook & Gorsuch, 1989; Watson, 1988). Cynicism,
aggression, and anger experience factors again were obtained. Twentythree of 24 scales marked the same factor in the oblique and orthogonal
three-factor solutions. The MAI was the only exception. In the threefactor orthogonal rotation, the MAI marked the cynicism factor, but in
the oblique rotation, it marked the anger experience factor. In the threefactor oblique solution, the correlation between the cynicism and aggression factor was .38, whereas the correlation between cynicism and anger
experience was .53. The correlation between the aggression and anger
experience factors was .52.
Mean-level gender differences. As previously described in footnote 3,
the structure of trait anger did not differ between men and women. To
further explore issues of gender and trait anger, t-tests were used to
compare mean factor score differences (computed using the Varimaxrotated factors) between men and women for the three-factor solution.
Men (M = .45, SD = .85) were found to have significantly higher
Behavioral Aggression factor scores than women (M = –.40, SD = .88),
t(455) = 10.47, p < .001. This pattern was reversed for Angry Affect factor
scores. Women’s Angry Affect factor scores (M = .17, SD = .85) were
significantly higher than men’s (M = –.19, SD = .95), t(455) = –4.25, p <
.001. Men’s Cynical Cognition factor scores (M = .01, SD = .97) were
somewhat higher than women’s (M = –.01, SD = .95); however, this
difference was not significant, t(455) < 1.
Trait Anger ABCs
891
DISCUSSION
Our analyses indicated that the three-factor solution provided the optimal
representation of the data. This solution was conceptually clearer than
the two-factor solution, in which anger experience overlapped with both
cynicism and aggression. The three-factor solution additionally clarified
associations among anger experience, cynicism, Neuroticism, and
Agreeableness that were obscured in the two-factor solution. Consistent
with prior research, the Angry Affect factor was most strongly related to
Neuroticism, whereas Behavioral Aggression was strongly related to low
Agreeableness. The Cynical Cognition factor emerged as a Neuroticism/low Agreeableness hybrid. Furthermore, confirmatory factor analyses indicated that the three-factor model fit the data significantly better
than the two-factor model. Thus, it was decided to retain the three-factor
solution. Consistent with the broad dimensions of affect, behavior, and
cognition, we named the first factor “Cynical Cognition,” the second
factor “Behavioral Aggression,” and the third factor “Angry Affect.”
Our results differ from several prior studies which produced two-factor
solutions composed of affect and behavior (e.g., Bendig, 1962; Bushman
et al., 1991; Buss & Durkee, 1957; Costa et al., 1989; Musante et al.,
1989). Our results are compatible, however, with a handful of studies that
suggested that cynicism is an additional factor that should not be overlooked in the conceptualization of trait anger (Costa et al., 1986, Friedman et al., 1995; and Musante et al., 1989). The three-factor solution
obtained in the present research was a consequence of carefully selecting
a comprehensive array of individual difference measures that attempted
to fully capture the affective, behavioral, and cognitive elements of trait
anger.
Although it ultimately was decided that the three-factor solution
provided the best representation of the data, the two-factor solution was
carefully evaluated as well. Importantly, the two-factor solution differed
substantially from the anger experience–anger expression model popular
in the behavioral medicine literature (e.g., Costa et al., 1989; Smith, 1992;
Suls et al., 1995). The two-factor solution was dominated by the dimensions of cynicism and aggression, whereas measures of anger experience
were distributed equally across the two factors. It is likely that cynicism
emerged in the present two-factor solution because several of the assessed trait anger scales were carefully chosen to measure the cynicism
dimension.
892
Martin et al.
Costa et al. (1989) related anger expression to Agreeableness and
anger experience to Neuroticism. The literature, however, offered little
information regarding the relation of Cynical Cognition to these traits.
Furthermore, it was unclear whether components of trait anger were
systematically related to the other Big Five domains of Extraversion,
Openness to Experience, and Conscientiousness.
Consistent with prior research, the results indicated that trait anger is
most closely related to Neuroticism and Agreeableness. Behavioral
Aggression was related strongly to low Agreeableness, and weakly to the
other four general personality traits. In contrast, Angry Affect was
strongly correlated with Neuroticism. Interestingly, Cynical Cognition
was equally associated with Neuroticism and low Agreeableness in both
studies. Thus, the relation between Behavioral Aggression and low
Agreeableness was relatively straightforward. Similarly, Angry Affect
was dominated by its correlation with Neuroticism. The associations
between Cynical Cognition and the Big Five traits were more complex,
however: Cynical Cognition appeared to represent a blend of neurotic
and disagreeable characteristics.
In interpreting our results, it is interesting to note that the Ho Scale has
been linked to cardiovascular disease (Barefoot et al., 1983; Barefoot et
al., 1989; Shekelle et al., 1983; Williams et al., 1980). In the present study,
the Ho Alienation and Antagonism scales both marked the Cynical
Cognition factor. Consequently, future research should carefully evaluate
the association between the cynical cognitive component of trait anger
and health. Of particular interest is the issue of whether it is the neurotic
or disagreeable aspect of Cynical Cognition that is crucial in the development of cardiovascular disease. In this regard, it is well established that
Neuroticism is a poor predictor of objective health outcomes (Costa &
McCrae, 1987; Watson & Pennebaker, 1989). This might be taken to
indicate that it is the disagreeable component of cynicism that is critically
important for health, and that the neurotic element is relatively uninteresting. If this interpretation were true, however, then one would expect
that Behavioral Aggression—which is even more strongly associated
with low Agreeableness—would be an even better predictor of health.
The empirical literature is mixed regarding this point, however (Miller
et al., 1996; Smith, 1992). Another possibility is that it is the combination
of high Neuroticism and low Agreeableness that predicts health-relevant
outcomes. If so, then behavioral medicine researchers may have been
premature both in classifying the Ho Scale as a measure of neurotic
Trait Anger ABCs
893
hostility (Suls et al., 1995) and in discounting the importance of neurotic
hostility for health outcomes (e.g., Costa et al., 1989; Suls et al., 1995).
Careful investigation into the relations among Cynical Cognition, Agreeableness, Neuroticism and health may help explicate the mysterious
ability of the Ho Scale to predict disease.
The application of an ABC model to the trait anger domain is likely to
be most useful in its ability to improve precision in the assessment of trait
anger. The related, but separable, factors of Angry Affect, Behavioral
Aggression, and Cynical Cognition reflect the fact that all forms of anger
do not share similar manifestations and consequences. The person prone
to the experience of angry emotions, for example, is not necessarily at
increased risk of committing assaultive acts. Thus, the three dimensions
of trait anger may be differentially predictive of different kinds of
consequences or outcomes and, in problematic cases, differentially responsive to prevention and clinical intervention. Physiologic processes
often are implicated in discussions of potential mechanisms that link trait
anger to health and behavioral outcomes. The semi-independence of
affect, behavior, and cognition in the anger domain suggests that multiple
physiologic mechanisms may mediate the effects of trait anger.
It is notable that, regardless of rotational strategy, the exploratory
factor analyses yielded solutions in which several scales loaded on more
than one factor. Furthermore, although confirmatory factor analyses
established the superior fit of the three-factor model, the fit indices were
adequate, but not remarkably strong. These observations suggest that
some subscales almost certainly included heterogeneous item content. It
also is remarkable that the analyses failed to yield a single measure
suitable for assessing the affective, behavioral, and cognitive dimensions
of trait anger. Thus, our findings emphasize the need for the development
of an individual difference measure that reliably assesses all three semiindependent dimensions of trait anger (Martin, Watson, & Wan, 1999).
The dimensions of angry affect, behavioral aggression, and cynical
cognition represent basic ABC “building blocks,” which provide a theoretical foundation for the development of a comprehensive trait anger
inventory.
REFERENCES
Bandura, A. (1973). Aggression: A social learning analysis. Englewood Cliffs, NJ:
Prentice-Hall.
894
Martin et al.
Barefoot, J. C. (1992). Developments in the measurement of hostility. In H. S. Friedman
(Ed.), Hostility, coping, and health (pp. 13–31). Washington, DC: American Psychological Association.
Barefoot, J. C., Dodge, K. A., Peterson, B. L., Dahlstrom, W. G., & Williams, R. B. (1989).
The Cook-Medley Hostility Scale: Item content and ability to predict survival.
Psychosomatic Medicine, 51, 46–57.
Baron, R. A. (1977). Human aggression. New York: Plenum.
Bendig, A.W. (1962). Factor analytic scales of covert and overt hostility. Journal of
Consulting and Clinical Psychology, 26, 200.
Berkowitz, L. (1962). Aggression: A social psychological analysis. New York: McGrawHill.
Biaggio, M. K., & Maiuro, R. D. (1985). Recent advances in anger assessment. In C. D.
Spielberger and J. N. Butcher (Eds.), Advances in Personality Assessment (Vol. 5, pp.
71–111). Hillsdale, NJ: Lawrence Erlbaum Associates.
Blumenthal, J. A., Williams, R. B., Jr., Kong, Y., Schanberg, S. M., & Thompson, L. W.
(1978). Type A behavior and angiographically documented coronary disease. Circulation, 58, 634–639.
Briggs, S. R., & Cheek, J. M. (1986). The role of factor analysis in the development and
evaluation of personality scales. Journal of Personality, 54, 106–148.
Bushman, B. J., Cooper, H. M., & Lemke, K. M. (1991). Meta-analysis of factor analyses:
An illustration using the Buss-Durkee Hostility Inventory. Personality and Social
Psychology Bulletin, 17, 344–349.
Buss, A. H. (1961). The psychology of aggression. New York: Wiley.
Buss, A. H., & Durkee, A. (1957). The measurement of hostility in clinical situations.
Journal of Consulting Psychology, 21, 343–349.
Buss, A. H., & Perry, M. (1992). The aggression questionnaire. Journal of Personality
and Social Psychology, 63, 452–459.
Carmody, T. P., Crossen, J. R., & Weins, A. N. (1989). Hostility as a health risk factor:
Relationships with neuroticism, Type A behavior, attentional focus, and interpersonal
style. Journal of Clinical Psychology, 45, 754–762.
Clark, L. A. (1993). Schedule for Nonadaptive and Adaptive Personality (SNAP).
Minneapolis: University of Minnesota Press.
Clark, L. A., & Watson, D. (1995). Constructing validity: Basic issues in objective scale
development. Psychological Assessment, 7, 309–319.
Contrada, R. J., & Jussim, L. (1992). What does the Cook-Medley Hostility Scale
measure? In search of an adequate measurement model. Journal of Applied Social
Psychology, 22, 615–627.
Cook, W. W., & Medley, D. M. (1954). Proposed hostility and pharisaic-virtue scales for
the MMPI. The Journal of Applied Psychology, 38, 414–418.
Costa, P. T., Jr., & McCrae, R. R. (1987). Neuroticism, somatic complaints, and disease:
Is the bark worse than the bite? Journal of Personality, 55, 297–316.
Costa, P. T., Jr., & McCrae, R. R. (1992). Revised NEO Personality Inventory (NEO PI-R)
and NEO Five-Factor Inventory (NEO-FFI): Professional manual. Odessa, FL:
Psychological Assessment Resources.
Costa, P. T., Jr., McCrae, R. R., & Dembroski, T. M. (1989). Agreeableness versus
antagonism: Explication of a potential risk factor for CHD. In A. W. Siegman &
Trait Anger ABCs
895
T. M .Dembroski (Eds.), In search of coronary prone behavior (pp. 41–63). Hillsdale,
NJ: Lawrence Erlbaum Associates.
Costa, P. T., Jr., Zonderman, A. B., McCrae, R. R., & Williams, R. B., Jr. (1985). Content
and comprehensiveness in the MMPI: An item factor analysis in a normal adult
sample. Journal of Personality and Social Psychology, 48, 925–933.
Costa, P. T., Jr., Zonderman, A. B., McCrae, R. R., & Williams, R. B., Jr. (1986). Cynicism
and paranoid alienation in the Cook and Medley HO Scale. Psychosomatic Medicine,
48, 283–285.
Edmunds, G., & Kendrick, D. C. (1980). The measurement of human aggressiveness.
New York: John Wiley & Sons.
Eysenck, H. J., & Eysenck, S. B. G. (1975). Manual for the Eysenck Personality
Questionnaire. San Diego, CA: Educational and Industrial Testing Service.
Floyd, F. J., & Widaman, K. F. (1995). Factor analysis in the development and refinement
of clinical assessment instruments. Psychological Assessment, 7, 286–299.
Friedman, H. S., Tucker, J. S., & Reise, S. P. (1995). Personality dimensions and measures
potentially relevant to health: A focus on hostility. Annals of Behavioral Medicine,
17, 245–253.
Geen, R. G. (1990). Human aggression. Pacific Grove, CA: Brooks/Cole.
Goldberg, L. R. (1990). An alternative “Description of Personality”: The Big-Five factor
structure. Journal of Personality and Social Psychology, 59, 1216–1229.
Gorsuch, R. L. (1983). Factor analysis. Hillsdale, NJ: Erlbaum.
Guadagnoli, E., & Velicer, W. F. (1988). Relation of sample size to the stability of
component patterns. Psychological Bulletin, 103, 265–275.
Harman, H. H. (1976). Modern factor analysis (3rd ed. rev.). Chicago: The University
of Chicago Press.
Hilgard, E. (1980). The trilogy of mind: Cognition, affection, and conation. Journal of
the History of the Behavioral Sciences, 16, 107–117.
Jackson, D. N. (1984). Personality Research Form manual (3rd ed.). Port Huron, MI:
Research Psychologists Press.
John, O. P., Donahue, E. M., & Kentle, R. L. (1991). The Big Five Inventory—Versions
4a and 5 (Tech. Rep.). Berkeley, CA: University of California, Institute of Personality
and Social Research.
Johnson, J. S., Null, C., Butcher, J. N., & Johnson, K. N. (1984). Replicated item level
factor analysis of the full MMPI. Journal of Personality and Social Psychology, 47,
105–114.
King, K. B. (1997). Psychologic and social aspects of cardiovascular disease. Annals of
Behavioral Medicine, 19, 264–270.
Martin, R., Wan, C. K., David, J. P., Wegner, E. L., Olson, B. D., & Watson, D. (1999).
Style of anger expression: Relation to expressivity, personality, and health. Personality
and Social Psychology Bulletin, 25, 1196–1207.
Martin, R., & Watson, D. (1997). Style of anger expression and its relation to daily
experience. Personality and Social Psychology Bulletin, 23, 285–294.
Martin, R., Watson, D., & Wan, C. K. (1999). Development and validation of the Anger
Affect, Behavior, and Cognition Inventory. Manuscript submitted for publication.
Matthews, K. A., Jamison, J. W., & Cottington, E. M. (1985). Appendix: Assessment of
Type A, anger, and hostility: A review of scales through 1982. In A. M. Ostfeld &
896
Martin et al.
E. D. Eaker (Eds.), Measuring psychosocial variables in epidemiologic studies of
cardiovascular disease: Proceedings of a workshop (NIH publication no. 85-2270,
pp. 207–311). Washington, DC: U.S. Department of Health and Human Services.
McGuire, W. J. (1969). The nature of attitudes and attitude change. In G. Lindzey & E.
Aronson (Eds.), The handbook of social psychology (Vol. 3, 2nd ed., pp. 136–314).
Reading, MA: Addison-Wesley.
Megargee, E. I., & Menzies, E. S. (1971). The assessment and dynamics of aggression.
In P. McReynolds (Ed.), Advances in psychological assessment (Vol. 2, pp. 133–156).
Palo Alto, CA: Science and Behavior Books.
Miller, T. Q., Jenkins, C. D., Kaplan, G. A., & Salonen, J. T. (1995). Are all hostility
scales alike? Factor structure and covariation among measures of hostility. Journal of
Applied Social Psychology, 25, 1142–1168.
Miller, T. Q., Smith, T. W., Turner, C. W., Guijarro, M. L., & Hallet, A. J. (1996). A
meta-analytic review of research on hostility and physical health. Psychological
Bulletin, 119, 322–348.
Musante, L., MacDougall, J. M., Dembroski, T. M., & Costa, P. T., Jr. (1989). Potential
for hostility and dimensions of anger. Health Psychology, 8, 343–354.
Novaco, R. W. (1975). Anger control: The development and evaluation of experimental
treatment. Lexington, MA: Lexington Books.
Sarason, I. G. (1961). Intercorrelations among measures of hostility. Journal of Clinical
Psychology, 17, 192–195.
Shekelle, R. B., Gale, M., Ostfeld, A. M., & Paul, O. (1983). Hostility, risk of coronary
heart disease, and mortality. Psychosomatic Medicine, 45, 109–114.
Siegel, J. M. (1986). The Multidimensional Anger Inventory. Journal of Personality and
Social Psychology, 51, 191–200.
Siegman, A. W., Dembroski, T. M., & Ringel, N. (1987). Components of hostility and
severity of coronary artery disease. Psychosomatic Medicine, 49, 127–135.
Smith, T. W. (1992). Hostility and health: Current status of a psychosomatic hypothesis.
Health Psychology, 11, 139–150.
Smith, T. W., & Frohm, K. D. (1985). What’s so unhealthy about hostility? Construct
validity and psychosocial correlates of the Cook and Medley Ho Scale. Health
Psychology, 4, 503–520.
Spielberger, C. D., Jacobs, G., Russell, S., & Crane, R. S. (1983). Assessment of anger:
The State-Trait Scale. In J. N. Butcher & C. D. Spielberger (Eds.), Advances in
personality assessment (Vol. 2, pp. 161–189). Hillsdale, NJ: Lawrence Erlbaum
Associates.
Spielberger, C. D., Johnson, E. H., Russell, S. F., Crane, R. J., Jacobs, G. A., &
Worden, T. J. (1985). The experience and expression of anger: Construction and
validation of an Anger Expression Scale. In M. A. Chesney & R. H. Rosenman (Eds.),
Anger and hostility in cardiovascular and behavioral disorders (pp. 5–30). Washington, DC: Hemisphere Publishing.
Steinberg, L., & Jorgensen, R. S. (1996). Assessing the MMPI-based Cook-Medley
Hostility Scale: The implications of dimensionality. Journal of Personality and Social
Psychology, 70, 1281–1287.
Suls, J., Wan, C. K., & Costa, P. T., Jr. (1995). Relationship of trait anger to resting blood
pressure: A meta-analysis. Health Psychology, 14, 444–456.
Trait Anger ABCs
897
Tellegen, A. (in press). Multidimensional Personality Questionnaire. Minneapolis: University of Minnesota Press.
Watson, D. (1988). The vicissitudes of mood measurement: Effects of varying descriptors, time frames, and response formats on measures of positive and negative affect.
Journal of Personality and Social Psychology, 55, 128–141.
Watson, D., & Clark, L. A. (1994). The PANAS-X: Manual for the Positive and Negative
Affect Schedule—Expanded Form. Iowa City, IA: University of Iowa.
Watson, D., Clark, L. A., & Harkness, A. R. (1994). Structures of personality and their
relevance to psychopathology. Journal of Abnormal Psychology, 103, 18–31.
Watson, D., & Pennebaker, J. W. (1989). Health complaints, stress, and distress: Exploring the central role of negative affectivity. Psychological Review, 96, 234–254.
Williams, R. B., Jr., Haney, T. L., Lee, K. L., Kong, Y., Blumenthal, J., & Whalen, R. E.
(1980). Type A behavior, hostility and coronary atherosclerosis. Psychosomatic
Medicine, 42, 539–549.
Zelin, M. L., Adler, G., & Myerson, P. G. (1972). Anger self-report: An objective
questionnaire for the measurement of aggression. Journal of Consulting and Clinical
Psychology, 39, 340.