Three studies explored the associations among style of anger expression,
emotional expressivity, Big Five personality traits, somatic complaints, and
self-reported health behaviors among undergraduate and community-residing
participants. Unlike measures of emotional expressivity, which tend to be most
strongly related to Extraversion, anger-in and anger-out primarily were associated
with Neuroticism and Agreeableness, respectively. Anger-in was positively
related to somatic complaints but failed to predict symptoms after controlling for
Neuroticism. Anger-out was positively associated with both somatic complaints
and self-reported health behaviors, even after controlling for Neuroticism and
Agreeableness. Measures of emotional expressivity provided further information
regarding style of anger expression. Anger-in was associated with a general
tendency to be emotionally inexpressive, whereas anger-out was more specifically
related to the expression of angry emotions .
Health researchers have suggested that style of anger expression may have
health-related consequences. Specifically, researchers have contrasted two
distinct expressive styles: "anger-out" and "anger-in" (Gentry, Chesney, Gary,
Hall, & Harburg, 1982; Harburg et al., 1973; Spielberger et al., 1985). Persons
prone to behavioral expressions of anger characterize the anger-out expressive
style, whereas anger-in refers to angry emotions that are experienced but not
expressed. Investigations of the health consequences of style of anger expression
have produced a complex empirical picture that is relevant to the distinction
between anger-out and anger-in.
Some studies have suggested that the Type A behavior pattern and hostile
behaviors that reflect the outward expression of anger are associated with the
development of coronary heart disease (e.g., Barefoot, Dahlstrom, &
Williams,1983; Barefoot, Dodge, Peterson, Dahlstrom, & Williams, 1989;
Matthews, Glass, Rosenman, & Bortner, 1977; Shekelle, Gale, Ostfeld, &
Paul,1983) . A variety of mechanisms have been suggested to explain this
association (T. W. Smith & Anderson,1986; Suls & Rittenhouse, 1990). For
example, the hyperreactivity hypothesis (Krantz & Manuck, 1984) proposes that
persons who tend to express their anger overtly might show especially extreme
changes in blood pressure and heart rate when provoked by other persons.
Alternatively, the exposure hypothesis (T. W. Smith & Anderson,1986) suggests
that the anger-expressive person might not show physiologic hyperreactivity per
se but instead might frequently encounter, or even create, conflict situations that
elicit cardiovascular responses. Although these hypotheses differ in their specifics,
they are not mutually exclusive and both suggest that the association between
anger expression and health outcomes is mediated by arousal of the sympathetic
nervous system. Over time, this chronic arousal is thought to produce physiologic
wear and tear, thus increasing the risk of disease.
In contrast to the literature linking anger-out and health outcomes, other
investigators have reported that it is the nonexpression of anger that engenders
health risk. In particular, some studies have indicated that people who do not
express their anger are vulnerable to hypertension (Gentry et al.,1982; Harburg et
al.,1973) . Although these findings seem at odds with the anger-- expression
literature, sympathetic arousal again is the mechanism thought to mediate the
association between anger-in and health outcomes. This line of reasoning is
founded in a psychoanalytic perspective that proposes that unexpressed anger is
directed inward toward the self, producing physiologic consequences, such as
increased heart rate and blood pressure, that might ultimately encourage the
development of cardiovascular disease (Alexander, 1939; Funkenstein, King, &
Drolette,1954). Gentry et al.,1982, and Harburg et al.,1973, found that people
who tended to be inexpressive when angry were at increased risk of hypertension.
A metaanalysis by Suls, Wan, and Costa (1995) found that anger-in
(conceptualized as part of an anger experience dimension) was related to elevated
blood pressure; however, the effect size was modest and the association highly
variable.
The Assessment of Style of Anger Expression
Interest in the potential health consequences of anger expression and inexpression
prompted researchers to develop trait measures assessing individual differences in
style of anger expression. The initial attempt was undertaken by Harburg et al.
(1973) , who developed the first questionnaire designed to classify people as
either anger-in or anger-out. Spielberger et al. (1985) refined and extended
Harburg's efforts with the construction of the Anger Expression (AX) scale. The
AX includes two subscales: Anger-in (AX/in) and Anger-out (AX/out). The
AX/in scale was constructed to identify people who experience, but tend not to
express, angry emotion. In contrast, the AX/out scale was intended to tap overt
verbal and physical expressions of anger. Note that the AX scales do not measure
the repression of unconscious anger, which would not be accessible to self-report.
Instead, the AX pertains to the conscious inhibition or expression of anger
(Davidson,1996) . Previous researchers have found the two scales to be only
weakly correlated (Deffenbacher et al., 1996; Martin & Watson, 1997;
Spielberger et al., 1985). The independence of AX/in and AX/out initially may
seem counterintuitive because it is tempting to assume that people who hold angry
emotions in do not also express angry emotions . However, the lack of
association between AX/in and AX/out underscores a basic distinction between
affective experience and behavior. For example, angry emotions may be
experienced on an ongoing basis. Whether any particular anger-provoking
episode culminates in expressive behavior will be determined by a variety of
factors, including personality, appraisal, and situational constraints. Thus, a person
who frequently expresses anger might often experience some angry emotions that
are not expressed, yielding high scores on both the AX/in and AX/out dimensions.
Consistent with previous findings that anger-in was associated with hypertension
(Gentry et al., 1982; Harburg et al., 1973), Johnson, Schork, and Spielberger (
1987a, 1987b) found that adolescents with high AX/in scores had significantly
higher blood pressure readings than those with low AX/in scores. These data
suggest that style of anger expression, as measured by the AX scale, might be
related to important health indices. Relatively little is known, however, regarding
the construct validity of the AX scale. Thus, it would be premature to draw any
conclusions about the association between style of anger expression and health.
Current Studies and Hypotheses
The present studies were designed to explicate the construct validity of the AX
scales so as to better understand their implications for health. We addressed three
gaps in the AX literature. First, although the AX scales are conceptualized as a
measure of stable individual differences in the expression of anger, previous
research has not considered the scale's association to other measures of general
emotional expressivity. If AX/in and AX/out actually measure the anger-related
aspect of emotional expression, logic suggests that the scales should be negatively
and positively related to measures of general emotional expressivity, respectively.
In other words, individual differences in style of anger expression should be
captured to some extent by more general measures of emotional expressivity,
which typically include items pertinent to anger and other negative emotions .
Furthermore, as measures of style of anger expression, the AX/in and AX/out
scales should demonstrate some specificity with regard to the expression (or
nonexpression) of angry emotion. That is, the AX/in and AX/out scales should
relate more strongly to the expression of angry emotions than to the expression
of positive or intimate emotions . These hypotheses were tested in Study 1 in a
large sample of undergraduates.
Second, questions remain regarding style of anger expression and more general
personality constructs. Specifically, explorations of the associations between the
AX scales and the five-factor model of personality (including the dimensions of
Neuroticism, Extraversion, Openness to Experience, Agreeableness, and
Conscientiousness) (Digman, 1990; John, 1990) have produced mixed results.
Friedman, Tucker, and Reise (1995) reported that AX/in was positively related to
Neuroticism, whereas AX/out was correlated with both Neuroticism and low
Agreeableness. Marshall, Wortman, Vickers, Kusulas, and Hervig (1994)
replicated the association between anger-in and Neuroticism but found that
anger-out primarily was related to low Agreeableness. Martin and Watson (1997)
recently reported that AX/in scores were correlated with daily self-reports of
anger experience. However, they also observed that AX/in scores were
significantly associated with every other type of state negative affect (e.g., anxiety,
depression) as well as with a trait measure of Neuroticism. Martin and Watson
(1997) found that AX/out scores were essentially unrelated to Neuroticism.
Martin and Watson did not measure the associations between AX scores and the
other dimensions of the Big Five, however.
In the current studies, we hoped to clarify the associations between AX/in,
AX/out, and the five-factor model of personality Placing the specific constructs of
AX/in and AX/out within the context of the broad Big Five traits would be an
important step in identifying both potentially redundant and unique characteristics
of style of anger expression. Based on previous findings (Marshall et al., 1994;
Martin & Watson,1997), we predicted that AX/in would be dominated by its
association with Neuroticism, whereas AX/out was expected to be most strongly
related to low Agreeableness. Finally, we were interested in observing the
associations between AX/in, AX/out, and Extraversion. General measures of
emotional expressivity tend to be positively related to Extraversion (King &
Emmons, 1990; Kring, Smith, & Neal, 1994). Given that the AX scales are
conceptualized as measures of the expressivity of angry emotion, AX/in and
AX/out therefore should be negatively and positively associated with
Extraversion, respectively. However, others have found only AX/in to be
systematically related to Extraversion (Friedman et al.,1995; Marshall et al.,
1994), raising questions regarding the construct validity of the AX scales. These
issues were evaluated in Study 1 and our findings were replicated in Studies 2 and
3 among both undergraduate and community-residing adult samples.
A third aim of the present program of research was to explore the associations
between style of anger expression and indices of health. As previously
described,Johnson et al. (1987a, 1987b) related AX/in scores to blood pressure
readings in high school students. Past studies have not considered how the AX
might relate to other markers of health, such as somatic complaints and use of
health care services. The five-factor model of personality provides a useful
framework within which to explore individual differences in the health domain
(Costa & McCrae, 1987; Marshall et al., 1994; T. W. Smith & Williams,1992) .
Thus, previous data (Friedman et al.,1995; Marshall et al.,1994; Martin &
Watson,1997) are potentially important in suggesting how style of anger
expression might relate to such health indices. It is well documented that
Neuroticism is a stronger predictor of retrospective somatic complaints than of
more objective health variables (Cohen et al., 1995; Costa & McCrae, 1987;
Watson, 1988; Watson & Pennebaker, 1989; cf. Brown & Moskowitz,1997).
The overlap between AX/in and Neuroticism naturally suggests that AX/in might
display a similar pattern of correlations, such as showing a stronger association
with subjective complaints than with health behaviors. In contrast, AX/out
emerges as a particularly interesting construct. Researchers have linked hostile
anger expression to the long-term development of cardiovascular disease. This
raises the question of whether AX/out might therefore be related to more proximal
health variables such as somatic complaints and health behaviors. For example,
one might speculate that the expression of anger might be associated with
increased vulnerability to minor illnesses, such as upper respiratory infections, as a
consequence of hyperreactivity and immune system changes. This would
prompt persons with high AX/out scores to report more minor physical
symptoms, colds, physician visits, and use of over-the-counter medications.
Alternatively, anger expression might compromise social support, which has been
linked to health outcomes (Cohen & Syme,1985). In any case, it would be
unlikely that any association between AX/out and health-related variables would
reflect the effects of Neuroticism because Marshall et al. (1994) observed a weak
correlation between AX/out and Neuroticism. We assessed somatic complaints in
Study 2 among a sample of healthy, community-residing adults. In Study 3, we
went on to evaluate self-report health-related behaviors (e.g., physician visits,
absenteeism, use of over-the-counter medications) in addition to somatic
complaints among a large undergraduate sample. Given the considerations
outlined above, we predicted that both AX/in and AX/out would be positively
related to somatic complaints but that only AX/out would predict self-report
health behaviors.
STUDY 1
The goal of Study 1 was to explore the construct validity of AX/in and AX/out by
relating each scale to measures of emotional expressiveness and the full five-factor
model of personality. Data for Study 1 were collected in the context of a set of
large factor-analytic studies of trait anger measures (Martin, 1996; Martin &
Watson, 1998). Here we focus on those aspects of the data that are relevant to
the construct validity of the AX scales and present emotional expressiveness data
that have not been previously reported.
Method
PARTICIPANTS
A large sample of college students (N= 457) participated; 53% (n = 242) were
women. Participants ranged from 18 to 32 years (M= 19.02 years, SD= 1.49).
Most participants (n = 408, 89%) were of European descent; the remaining
participants were Asian American and African American.
MEASURES
Style of anger expression was assessed with the AX/in and AX/out scales
(Spielberger et al., 1985). Participants completed two general measures of
emotional expressivity: the Emotional Expressiveness Questionnaire (EE) (King &
Emmons, 1990) and the Emotional Expressivity Scale (EES) (Kring et al., 1994).
The Big Five Inventory (BFI) (John, Donahue, & Kentle, 1991) and the NEO
Five Factor Inventory (NEO-FFI) (Costa & McCrae, 1992) were used to
measure the five broad personality traits.
AX. The AX/in and AX/out scales (Spielberger et al., 1985) each include eight
items that are rated from 1 (almost never) to 4 (almost always). As previously
described, the AX/in scale was intended to measure the tendency to suppress
anger but actually appears to assess broader individual differences in the
experience of anger and other emotions (Martin & Watson,1997). The AX/in
scale includes items such as "I keep things in" and "I boil inside but I don't show
it." The AX/out scale was designed to quantify individual differences in the
expressive, behavioral component of anger. The AX/out scale includes items such
as "I do things like slam doors" and "I strike out at whatever infuriates me." Both
scales are scored in the hostile direction. The mean AX/in score was 16.90 (SD=
4.39); the AX/out mean was 16.14 (SD = 4.02). The AX/in and AX/out scales
were internally consistent; both yielded a Cronbach's alpha of .78 in this sample.
Consistent with previous findings (Deffenbacher et al., 1996; Martin & Watson,
1997; Spielberger et al., 1985), the correlation between anger-in and anger-out
was modest (r = .20).
EEQ The EEQ (King & Emmons, 1990) is a 16-item questionnaire measuring
individual differences in expressing both positive and negative emotions . Items
are rated from 1 (strongly disagree) to 7 (strongly agree). Self-report EEQ scores
are moderately associated with peer and family ratings of emotional
expressiveness (King & Emmons, 1990). Three EEQ subscales were constructed
based on factor analyses by King and Emmons (1990): Expression of Positive
Emotions (EEQ/positive, 7 items; M = 5.17, SD = 0.99), Expression of Intimacy
(EEQ/intimacy, 5 items; M = 5.01, SD = 1.09), and Expression of Negative
Emotions (EEQ/negative, 4 items; M = 3.94, SD = 1.03) . Cronbach's alpha
was satisfactory (.77) for the EEQ/positive subscale. Cronbach's alpha was
somewhat low for the EEQ/intimacy and the EEQ/negative subscales (.65 and
.58, respectively). However, the mean interitem correlations were satisfactory for
both subscales (.28 and .25, respectively), suggesting that the low alphas were a
consequence of the small number of items comprising each scale. (See Briggs and
Cheek, 1986, and Clark and Watson, 1995, for a discussion of the mean
interitem correlation as an index of internal consistency.)
EES. The EES (Kring et al., 1994) provided a second measure of emotional
expressiveness. The EES includes 17 items rated on a 6-point scale and scored in
the expressive direction. In contrast to the EEQ (King & Emmons, 1990), EES
items refer to general emotional expressivity rather than to the expression of
specific positive or negative emotions . EES scores are moderately related to
both peer- and laboratory-rated expressiveness (Kring et al., 1994). The mean
EES score was 64.12 (SD= 15.46) and Cronbach's alpha was strong (.93).
BFI. The BFI (John et al., 1991 ) assesses the five major domains of personality.
The inventory includes 44 items rated from 1 (very uncharacteristic of me) to 7
(very characteristic of me) . Scores were obtained for Neuroticism (M= 3.91, SD
= 1.02), Extraversion (M = 4.67, SD = 0.98), Openness (M = 4.83, SD =
0.82), Agreeableness (M = 5.19, SD = 0.81), and Conscientiousness (M= 4.78,
SD = 0.81). All BFI scales were internally consistent, with Cronbach's alphas
ranging from .78 to .84.
NEO-EFI. The 60-item NEO-FFI (Costa & McCrae, 1992) provided a second
measure of the Big Five traits. Items are rated on a 5-point scale ranging from
strongly agree to strongly disagree. Five domains were scored: Neuroticism (M=
23.59, SD = 8.16), Extraversion (M= 31.32, SD = 6.38), Openness (M= 28.69,
SD = 6.51), Agreeableness (M= 31.15, SD= 6.02), and Conscientiousness (M=
30.67, SD = 6.80) . All NEO-FFI scales were internally consistent, with
Cronbach's alphas ranging from .75 to .86.
PROCEDURE
Participants were recruited from an introductory psychology class at a large,
Midwestern university. The individual difference measures were completed in
groups of 10 to 20 in the context of a larger investigation of trait anger. The order
of instruments was counterbalanced. Participants received partial course credit
and $10 in compensation for their efforts.
Results and Discussion
Consistent with the goals of Study 1, we evaluated the associations between style
of anger expression and two measures of emotional expressivity. The relations
between style of anger expression and the five-factor model of personality also
were explored.
STYLE OF ANGER EXPRESSION AND EMOTIONAL
EXPRESSIVENESS
Zero-order correlations. Table 1 reports the zero-order correlations among the
style of anger expression scales and emotional expressivity measures. AX/in was
negatively related to general emotional expressiveness, as measured by the EES.
More modest negative correlations also were observed between AX/in and the
EEQ subscales. AX/out was positively related to the expression of negative
emotions .
Two of the four EEQ/negative subscale items refer to anger expression explicitly.
To provide further detail regarding the relation of the AX scales to the expression
of angry emotions , zero-order correlations were computed between AX/in,
AX/out, and each of the EEQ/negative subscale items. Overall, the correlations
between the AX/in scale and the EEQ/negative items were modest in magnitude.
However, the pattern of correlations was consistent with the interpretation that
although AX/in relates to emotional inexpressiveness, it is not a specific measure
of anger expressivity per se. AX/in was unrelated to the EEQ/negative item "If
someone makes me angry in a public place, I will cause a scene" (r = .05) and
very modestly associated with the item "When I am angry people around me
usually know" (r = -.12). AX/in was more strongly and negatively related to the
item "People can tell from my facial expressions how I am feeling" (r= -.21 ) . A
small positive correlation was observed between AX/in and the item "I always
express disappointment when things don't go as I'd like them to" (r= .17).
AX/out was moderately related to both EEQ/negative anger items (r= .40 with
"When I am angry people around me usually know" and r= .43 with "If someone
makes me angry in a public place, I will cause a scene") . In contrast, the
correlations between AX/out and the two nonanger items were more modest (r=
.28 with "I always express disappointment when things don't go as I'd like them
to" and r = .01 with "People can tell from my facial expressions how I am
feeling"). The associations between AX/out and the EEQ/negative anger
expression items support the construct validity of the AX/out scale because it
appears to discriminate between the tendency to express anger and the
expression of other emotions . However, these findings should be interpreted
cautiously because they are based on correlations between AX/out and single
items.
Structural equation modeling. To evaluate the interpretations suggested by the
preceding pattern of correlations, we tested the structural relations among style of
anger expression and emotional expressivity via structural equation modeling. The
analyses were conducted with LISREL 8 (Joreskog & Sorbom, 1993) . The data
initially were examined for multivariate normality to determine the appropriateness
of using a maximum likelihood solution. The skewness ranged from -1.50 to 1.39
and the kurtosis ranged from -1.01 to 1.42. (Mardia's coefficient of multivariate
kurtosis for the data was 100.88) . Because the data did not depart from
multivariate normality, maximum likelihood estimation was used for all subsequent
analyses. The model addressed the associations among the AX/in, AX/out,
EE/positive, EEQ/intimacy, EEQ/negative, and EES measures, with the individual
items of each scale used as the indicators of each construct. As discussed
previously, we expected that AX/in would be negatively related to the EES,
whereas AX/out would be positively related to the same scale. We further
expected both AX scales to be more strongly associated with the expression of
negative emotions , which include anger, than with the expression of other
emotional states. Table 2 presents the standardized and unstandardized path
coefficients for the model. The chi-square test failed to achieve nonsignificance
(chi^sup 2^ = 394.92, df= 118, p < .01) but the Goodness of Fit Index (GFI)
(.90) and the Comparative Fit Index (CFI) (.95) indicated avery satisfactory fit
for the model.
Table 2 illustrates that AX/in was negatively related to the EES, indicating that
participants with high AX/in scores tended to describe themselves as emotionally
inexpressive. The path coefficients between AX/in and the EEQ subscales
provided additional important information. Consistent with the intended purpose
of the AX/in scale, AX/in was negatively related to the EEQ/negative subscale. In
other words, participants with high AX/in scores systematically characterized
themselves as unlikely to express negative emotions such as anger and
disappointment. However, AX/in also was significantly and negatively related to
both the EEQ/positive and EEQ/intimacy subscales. This suggests that the AX/in
scale is not specific to the inexpression of angry emotions . Rather, the AX/in
scale identifies individuals who tend to be generally inexpressive.
Table 2 also reports a significant, positive association between AX/out and the
EES, reflecting the tendency for participants with high AX/out scores to be
generally expressive of their emotions . Moreover, the AX/out scale appeared to
be specifically related to the expression of angry emotions . AX/out was most
strongly related to the EEQ/negative subscale but unrelated to the expression of
either positive or intimacy-related emotions . Thus, our data indicated that
participants with high AX/out scores tended to describe themselves as emotionally
expressive in general, with an emphasis on the expression of negative emotions .
STYLE OF ANGER EXPRESSION AND THE BIG FIVE TRAITS
Zero-order correlations. As described previously, participants completed two
measures of the Big Five: the BFI (John et al., 1991) and the NEO-FFI (Costa &
McCrae, 1992). Five trait scores (Neuroticism, Extraversion, Openness,
Agreeableness, and Conscientiousness) were scored from each instrument. Then,
each trait score was standardized and the two scales representing each dimension
were added together to create a single unit-- weighted index of each of the Big
Five traits. The zero-- order correlations among the Big Five dimensions, the AX
scales, and the measures of emotional expressiveness are presented in Table 3.
Table 3 illustrates that AX/in was most strongly related to Neuroticism and more
modestly associated with low Agreeableness and low Extraversion. A very
different pattern of findings emerged for AX/out, which was strongly and
negatively related to Agreeableness and only weakly related to the other Big Five
traits. The conceptual differences between AX/in, AX/out, and emotional
expressivity are particularly evident when considering how the EES and the EEQ
subscales relate to the Big Five traits. As reported in Table 2, the emotional
expressivity measures tended to be associated more strongly with Extraversion
than with any of the other Big Five traits. In addition, each measure of emotional
expressivity was more modestly related to Agreeableness. A small correlation
was observed between the EEQ/negative subscale and Neuroticism; however, the
other expressivity measures were not strongly related to Neuroticism. The
associations between the expressivity measures and the Big Five scores contrast
sharply with those previously described for AX/in and AX/out. Recall that AX/in
and AX/out were dominated by their associations with Neuroticism and low
Agreeableness, respectively. Although both AX scales were related to
Extraversion, these correlations were more modest in magnitude. Thus, the
zero-order correlations suggested that measures of emotional expressivity, AX/in,
and AX/out related differentially to the Big Five traits.
Confirmatory factor analysis. The above observations are based on the
interpretation of the zero-order correlations, without consideration of the
measurement errors. To this end, confirmatory factor analysis was used to
examine the associations among AX/in, AX/out, Neuroticism, and Agreeableness.
Based on the literature previously reviewed and the pattern of zero-order
correlations, we expected AX/in to be positively related to Neuroticism, whereas
a negative association was predicted between AX/out and Agreeableness. Once
again, the individual AX items were used as indicators of the AX/in and AX/out
constructs. Neuroticism and Agreeableness each had two indicators, which were
derived from the BFI and NEO-FFI scales. The fit indices for the hypothesized
factor structure among these constructs reflected satisfactory goodness of fit (GFI
= .92, CFI = 0.95), X2 ( 164) = 746.34, p < .01. The structural coefficients
between AX/in and Neuroticism and Agreeableness were .61 and -.40,
respectively, ps < .05. In contrast, the structural coefficient was -.64, p < .05,
between AX/out and Agreeableness and .30, p < .05, between AX/out and
Neuroticism.
Summary. To summarize the major findings of Study 1, AX/in was negatively
related to emotional expressiveness. However, this association reflected the
tendency for participants with high AX/in scores to report difficulty expressing
positive and intimacy-related emotions as well as angry emotions . AX/in was
most strongly related to Neuroticism and more weakly related to low
Agreeableness. Our findings were consistent with Martin and Watson (1997),
who found that AX/in related to general negative affect in addition to angry
emotions . The finding
|