Psychological Test and Assessment Modeling, Volume 56, 2014 (2), 178-194 [610330]

Psychological Test and Assessment Modeling, Volume 56, 2014 (2), 178-194
Validation of the STAXI-2: A study with
prison inmates
Sonja L. Etzler1, Sonja Rohrmann2 & Holger Brandt3
Abstract
This study investigates the reliability and validity of the State-Trait-Anger-Expression-Inventory-2
(STAXI-2; in the German version by Rohrmann et al., 2013) in a sample of n=57 male inmates
(aged 25-67 years). Internal consistencies reveal a high reliability of the STAXI-2 and subscale intercorrelations are close to those of the standa rdization sample. A confirmatory factor analysis
confirms the four-dispositional-factor structure of the STAXI-2. Construct validity is supported by
correlations with personality traits of inmates that play a key role in inmate misconduct (e.g., ag-
gressiveness & social adjustment). Anger-Expre ssion-Out and Anger-Expression-In are significant
predictors of the length of sentence and Trait-Anger is associated with the age of first offense, both
indicating criterion-related validity. These results support the use of the STAXI-2 for inmates and
suggest that it can serve as an diagnostic tool in correctional institutions.
Keywords: STAXI-2, Anger, inmates, validity

1 Correspondence concerning this ar ticle should be addressed to: Sonja Etzler, PhD, Goethe University
Frankfurt, Department of Psyc hology, Grüneburgplatz 1, 60323 Frankfurt, Germany;email: etzler@psych.
uni-frankfurt.de
2 Department of Psychology, Goethe University Frankfurt, Frankfurt, Germany
3 Center for Educational Science and Psychology, Eberhard Karls University, Tübingen, Germany

Validation of the STAXI-2: A study with prison inmates 179
Anger is an elementary emotion that plays an important role in daily life. Anger is de-
fined as an emotional reaction when an organism is blocked in the attainment of a goal or fulfillment of a need, or as reaction to a provocation (Izard, 1991; Novaco, 1975). Anger is assumed to be a multidimensional construct (Scherer, 1990) which consists of an emo-tional, a physiological, an expressive, a behavioral, and a cognitive component (Hodapp & Bongard, 2009). Spielberger (1988) suggested discriminating two components of Anger, State-Anger and Trait-Anger: State- Anger represents the emotional component
that is present in a specific situation and may change within a subject across different situations. In contrast, Trait-Anger is the general disposition to experience anger fre-
quently and intensively. It is assumed not to vary within a subject across situations but to be a stable disposition. Furthermore, Spie lberger (1988) demarcated Trait-Anger itself
from the processing and handling of angry emotions. He suggested differentiating three tendencies in the handling of angry feelings: First, subjects may tend to introvert anger (Anger Expression-In), second, they may frankly express or show their feelings of anger
(Anger Expression-Out), and third, subjects ma y tend to control angry feelings and sup-
press them (Anger-Control).
The State-Trait-Anger-Expression-Inventor y (STAXI, STAXI-2; Spielberger, 1988,
1999) is based on this definition and allows for a differentiated measurement of these
five dimensions: based on self-report, the STAXI assesses how a person feels at a given moment ( State-Anger ), how frequently, easily, and intensely the person feels angry
(Trait-Anger ) and what the person does when feeling angry ( Anger Expression-In, Anger
Expression-Out, Anger-Control ). The STAXI exhibits good reliability, with alpha coeffi-
cients ranging from .81 to .93 (Spielberger, 1988). The structure of the anger expression subscales (Spielberger, Johnson, Russel, Cran e, Jacobs, & Worden, 1985) has been repli-
cated in an offender sample (Kroner & Reddon, 1992). There are several findings that support the validity of STAXI and STAXI-2. Th ere are, for example, associations be-
tween Trait-Anger and Hostility, Anger Expression-In and Anxiety as well as blood pressures, which indicate that the STAXI measures anger in a highly valid manner (Spielberger, 1988; Schwenkmezger, Hodapp, & Spielberger, 1992). The STAXI has been validated on a variety of normal and clinical populations and has good psychomet-ric properties (Müller, Bongard, Heiligtag & Hodapp, 2001). The revised and extended version, STAXI-2 (Spielberger, 1999), cont ains new items for the subscales Anger-
Control and State-Anger to increase psychometric properties. The German version of the STAXI-2 (Rohrmann, Hodapp, Schnell, Tibubos , Schwenkmezger, & Spielberger, 2013)
shows good reliability, with Cronbach-alpha coefficients ranging from
.82α= to
.90α= . It displays factorial validity and correlations with anger reactions, aggressive-
ness, negative affectivity as well as physical discomfort (Rohrmann et al., 2013).
In this article, we report additional support fo r the validity of the German version of the
STAXI-2 for a specific subgroup of inmates. Since increased anger may be associated with physical aggression or general hostility, application of the STAXI-2 in inmate sam-ples may be of special interest for treatment recommendations or other diagnostic situa-tions. In the following section, we therefore review the main findings concerning the relationship between anger and criminal behavior.

S. L. Etzler, S. Rohrmann & H. Brandt 180
Anger and criminal behavior
Agnew's General Strain Theory (Agnew, 1992, 2001, 2007) is one of the most prominent
theories that can be applied to formulate the relationship between anger and criminal behavior. Agnew (1992) assumed that a subject's exposition to specific types of strain causes increased feelings of anger. Agnew th erefore defined three types of strain: the
failure to achieve positively valued goals, th e removal of positively valued stimuli from a
person, and the presentation of negative stimu li to an individual. Being confronted with
such stimuli, feelings of anger are accompanie d by an urge to take corrective steps. One
way of taking corrective steps is to resort to criminal or aggressive behavior (Agnew,
1992). An outwardly aggressive and criminal be havior is most likely, if there are few
legal alternatives from the person's perspective and if the strain is assumed to be caused by another person. In situations where the st rain cannot be removed directly, aggressive
behavior can serve as revenge. For instance, Deffenbacher et al. (1996) showed for a
sample of college students that those persons who scored at or above the 75th percentile on the STAXI Trait-Anger reported significantly more aggressive behavior when angry
than students who scored below the 25th percentile. Further research also documented that in community samples anger is associated with different forms of antisocial behavior (Hazaleus & Deffenbacher, 1986; Maiuro et al., 1988; Novaco, 1994). These results support Agnew's theory that anger is involved in criminal behavior in non-incarcerated
samples.
Spielberger and Sydeman (1994) postulated that particularly the outward expression of
anger is associated with violent behavior, whereas an inward expression is associated with anxiety. Besides violence and criminal behavior, which are extreme forms of an outward expression of anger, weaker forms of antisocial behavior such as conflicts and
interpersonal problems, are very likely to occur (Baumeister, Stillwell, & Wotman, 1990;
Schmitt & Altstötter-Gleich, 2010). Although researchers emphasize the importance of anger and anger expression in criminological theories, empirical evidence does not al-ways support a prevalence of increased anger in inmates in general. In their study on STAXI scores for male sexual offenders, Dalton, Blain, and Bezier (1998) revealed that
scores of inmates were very similar to thos e of non-incarcerated men, except for a slight-
ly higher State-Anger. Additionally, research has not always indicated a direct relation-ship between anger and criminal behavior or criminal recidivism (Loza & Loza-Fanous,
1999a, 1999b; Mills & Kroner, 2003).
These findings are in accordance with the theory of Anger, Aggression and Hostility
("AHA"-Syndrome; Spielberger, 1988), which postulates that anger is the basic and necessary but not sufficient condition to develop hostile cognitions or manifestations of aggressive behavior (Spielberger & Rehe iser, 2010). Mills & Kroner (2003) concluded
that, although some offenders exhibited serious anger problems which were involved in committing their crimes, there might be several criminal behaviors for which anger only plays a subordinate role. And finally, there is the possibility that anger plays a different role in criminal behavior in inmate samples than in student samples (e.g., Deffenbacher et al., 1996). These findings suggest that there is no linear relationship, but an interaction
between anger and a third moderator variable which is not yet known. This moderator

Validation of the STAXI-2: A study with prison inmates 181
variable might be categorical, such as differe nt types of offenses, or continuous, such as
the degree of the ability to apply stress-reducing coping strategies.
Findings concerning the identification of s ubgroups with increased anger scores are
mainly based on the type of offense committed. These results are inconclusive: Zamble and Quinsey (1997) revealed that anger diffe rentiated assaulters from other offenders.
Cherek, Scott, Dougherty, Moeller, and White (2000) found that violent female offenders
showed higher Trait-Anger, Anger Expre ssion-Out, and lower Anger-Control than non-
violent female offenders. Kalichman (1991) revealed higher Trait-Anger in sexual of-
fenders who committed their crimes on prepube scent children than in offenders with
adult victims. Shealy, Kalichman, Henderson, Szymanowski, and McKee (1991) found that STAXI scales helped to identify se veral subgroups of sexual offenders. Up until
now, there is no evidence that a specific type of offense, for example, sexual or violent offense, can be identified that is associated with an unambiguously increased Trait-Anger
or an increased outward anger expression. But, one may at least conclude that anger is
involved in criminal behavior if not to the same extent in each offense. Mills and Kroner
(2003) discuss an interaction effect between anger and the continuous moderator coping strategies. Increased anger may abet criminal behavior in delicate situations where sub-
jects do not dispose of coping skills to handl e the situation legally. Consequently, higher
Trait-Anger leads to criminal behavior if third variables (e.g., coping skills) are not available.
The behavior of perpetrators who are sentenced to imprisonment continues to be influ-
enced by anger, its expression, and its contro l. Thus, theories concerning anger and devi-
ant behavior may also be applied to imprisone d persons. For example, Cornell, Peterson,
and Richards (1999) present evidence that anger may be related to conduct problems in incarcerated adolescents. As an indication of construct validity of the STAXI-2 for incar-
cerated men, we expect a relationship betw een Trait-Anger, Anger Expression-Out and
self-reported traits concerning interpersonal behavior, i.e. aggressiveness and social
adjustment. Furthermore, we hypothesize th at Anger Expression-In shows strong asso-
ciations with traits that repr esent internalized dispositions, i.e., a positive association with
emotional instability and a negative association with optimism. Finally, we assume that in correctional institutions high Anger-Control is accompanied by high social adjustment and weak Experience of incompetence.
Based on the implications of these criminological theories and on empirical findings, we do
not expect STAXI scores to be generally increas ed in inmates. But, in accord with Agnew's
theory, we expect that anger is associated with indices of severity of criminal behavior. The
severity of criminal behavior implies the severity of damage that is produced by one par-ticular offence (e. g., very brutal crimes) as well as the frequency of several offences. Both
alternatives typically lead to longer sentences when subjects are convicted, so we first oper-ationalized the severity of criminal behavior as length of sentence.
Furthermore, Deffenbacher et al. (1996) dem onstrated that individuals who scored high
in Trait-Anger on the STAXI tended to expe rience more negative consequences as a
result of their anger than did individuals scor ing lower on this scale. For persons with an
increased Trait-Anger, the probability of fee ling angry in intense situations is higher;

S. L. Etzler, S. Rohrmann & H. Brandt 182
therefore, they are also more likely to react with criminal behavior to the perceived
strain, especially when the situation involve s criminal stimuli. Thus, increased Trait-
Anger scores should lead to a higher frequenc y of criminal behavior, which shows a high
association with the Age of the first offe nse (Tolman, 1987), our second operationaliza-
tion of the severity of criminal behavior.
Aim of the study
In order to examine psychometric properties in the population of prison inmates, this study examines reliability coefficients. First, we expect the STAXI-2 to show reliability
coefficients of an acceptable magnitude ( .70
α≥ ; in line with suggestions by Schermel-
leh-Engel & Werner, 2011). Second, we expect subscale intercorrelations of the STAXI-
2 in the current sample of inmates to approach the intercorrelations of subscales in the standardization sample assuming that the STAXI-2 measures anger in an equivalent manner across these two populations. In order to validate the STAXI-2 for the population of prison inmates, the factorial (i.e. construc t) validity of the STAXI-2 will be analyzed.
Factorial validity will be proven for the STAXI-2 if the postulated factor model fits the
data. To examine construct validity, we w ill analyze correlations between anger dimen-
sions and personality traits that play a role in inmate misconduct. We hypothesize that outward anger will show the strongest associ ations with self-repor ted personality disposi-
tions of inmate misconduct and expect a pattern of correlation which fits the theory. For
the assessment of criterion validity, we will analyze the relationship between the STAXI-2 and indices of general criminality. Outward anger expression, which relates to violent behavior, should be associated with some meas ures of criminal behavior, such as age of
first offense and the length of sentence.
Method
Participants
The sample consisted of 57 incarcerated male offenders from a state correctional institu-
tion in Frankfurt/Main, Germany. Participants took part in the study on a voluntary basis,
they were allowed to fill out the questionnaire in their spare time, and received a small allowance (e.g., chocolate) afterwards. 98% of these inmates were finally convicted;
their prison sentences ranged from 2.6 months to 4.25 years with a mean of 21 months. Subjects were mainly convicted for offenses like violation of the narcotics law, robbery, fraud, or theft. Detailed demographic statistic s for the sample are presented in Table 1.
Participants' ages ranged from 25 to 67 years (M ean = 37.11; SD = 9.19). 59.6% of the n
= 57 participants had never been married, 22. 8% lived in a relationship, 10.5% were
divorced, and 5.3% were married at the time of the investigation. Most of the subjects
had left school after the 9
th or 10th grade (40.4%, respectively), 8.8% were classified as

Validation of the STAXI-2: A study with prison inmates 183
Table 1:
Participants' Demographics
Mean SD
Age 37.11 9.19
Absolute
number Percentage
Marital status
Single 34 59.6
Married 3 5.3 Relationship 13 22.8 Divorced 6 10.5 Widowed 0 0.0
Educational level
Early school leavers (without graduation) 5 8.8 Grade 9 (Hauptschule) 23 40.4 Grade 10 (Realschulabschluss) 23 40.4 Grade 12 or more (Fachhochschulreife) 5 8.8 Qualification for university entrance (Abitur) 1 1.8
Treatment during arrest 26 45.6

early school leavers (without school-leaving cer tificate), 8.8% had left school after the
12th grade, and 1.8% had obtained the qualifi cation for university entrance. In total,
45.6% of the subjects had received at least one treatment during their arrest (including
psychotherapy, apprenticeships, or other skill trainings).
Measures
The State-Trait-Anger-Expression-Inventory-2 (STAXI-2; Spielberger, 1999) is a
self-report questionnaire that contains 51 items measuring five domains of anger: State-
Anger, Trait-Anger, Anger Expression-In, Anger Expression-Out, and Anger-Control.
Responses to the items are indicated on a 4-point graded scale, ranging from 1 (com-pletely disagree) to 4 (completely agree). In the current study, the subscale State-Anger was omitted in all analyses, because information on the subjects' emotional state during participation was only of limited interest.
Persönlichkeitsfragebogen für Inhaftierte (P ersonality Questionnaire for inmates;
PFI; Seitz & Rautenberg, 2010). The PFI is a German-language questionnaire that
assesses personality traits of inmates which are relevant to prison life by presenting items which pertain to daily life in prison. The PFI measures ten facets of personality. For
economic reasons, we administered only seven of ten personality facets (i.e. Experience

S. L. Etzler, S. Rohrmann & H. Brandt 184
of Incompetence, Emotional Instability, Optimistic Recklessness, Social Adjustment,
Need for Attention and Encouragement from Inmates, Perceived Autonomy and Domi-nance against Inmates, and Aggressiveness) with those items with the highest discrimi-nability index (3 to 5 items per facet). The reliability coefficients for these short facets
ranged from .48 (acceptable reliability) to .80 (good reliability) considering the rather
small number of items per scale. Results concerning the validity of the PFI are based on the original version of the questionnaire. C oncerning internal validity, substantial rela-
tionships were found between corresponding s ubscales such as aggressiveness and de-
pression from the Freiburger Persönlichkeitsinve ntar (FPI; Fahrenberg, Hampel, & Selg,
2001) and the PFI (Seitz & Rautenberg, 2010). Mean differences between relevant sub-
samples (socio-demography, intelligence, attit ude towards criminal, and legal behavior,
etc.) confirm the external validity of the PFI (see Seitz & Rautenberg, 2010).

Statistical analysis
T-Scores were calculated for the four scales of the STAXI-2 in order to examine the
differences between our prison sample and the standardization sample of the STAXI-2 (Spielberger, 1999).
Missing data were observed for 15 subjects for at least one item of the STAXI-2 (total
missingness was 6.02%). Missi ngness was assumed to be Missing At Random (MAR)
and was accounted for in the analysis of validity by imputing scale scores using the EM algorithm (Dempster, Laird, & Rubin, 1977; L ittle & Rubin, 2002). For the analysis of
factorial validity, the missing data in the item parcels was accounted for appropriately (under the MAR assumption), that is by applying the full information maximum likeli-hood estimator in M
plus (Muthén & Muthén, 1998-2012), or the Gibbs sampler for the
Bayesian analysis (Gelman, Carlin, Stern, & Rubin, 2004).
Results
Descriptives and reliability
Means, standard deviations, Cronbach's α coefficients, and correlations between the four
subscales of the STAXI-2 (Trait-Anger, A nger Expression-In, Anger Expression-Out,
and Anger-Control) are presented in Table 2. The subscales of the STAXI-2 showed high
reliabilities for the sample of inmates, with Cronbach's α coefficients ranging from .79 to
.88. Intercorrelations among the subscales were comparable to correlations of the stand-
ardization sample. As can be seen, almost a ll correlations of the inmate sample were
close to those of the standardization sample. The only exception was the correlation between Trait-Anger and Anger Expression-Ou t, where the inmate sample exhibited a
weaker association between the two subscales (r = .48) than subjects in the standardiza-tion sample (r = .70). We applied a multivariate test to the two correlation matrices in order to examine if all correlation coefficien ts were the same (Bollen, 1989). Therefore,

Validation of the STAXI-2: A study with prison inmates 185
we specified two models: A restricted model that assumed the same population correla-
tion matrices for both samples (R inmate=R stand), and an unrestricted comparison model that
assumed different population matrices (R inmate≠Rstand). We used the χ2 values provided by
Mplus (Muthén & Muthén, 1998-2012) to carry out a χ2 difference test that tested the
invariance of the correlation matrices (based on the sample correlation matrices given in
Table 2). The unrestricted model fitted the data significantly better ( χ2
diff = 19.54, df = 6,
p <.05) which indicated a significant difference between the two groups when all correla-tions were assumed to be equal. An additional model that assumed equal correlations across the two samples except the one between Trait Anger and Expression-Out showed
no significant difference to the unrestricted model (
χ2
diff = 8.89, df = 5, p = .08). This
implied that the correlational pattern of the scales was the same for inmates and subjects from the standardization sample except for the correlation between Trait Anger and Expression-Out.

Table 2:
Means, Standard Deviations, α Coefficients, and Subscale Intercorrelations
(with Confidence Intervals) of the STAXI-2
M (SD) α Correlations
Anger
Expression-
In Anger
Expression-
Out Anger-Control
Trait-Anger 19.55 (5.54) .88 .32* (.06; .53)
.21 * a .48* (.25; .66)
.70* a -.24 (-.48; .02)
-.44* a
Anger Expression-In 18.96 (5.53) .83 .30* (.04; .52)
.05 a .10 (-.16; .36)
.25* a
Anger Expression-Out 12.90 (3.96) .79 -.46* (-.65; -.23)
-.53* a
Anger-Control 29.46 (6.51) .87
Note: n = 56, *p < .05. a Correlations of the standardization sample

Factorial validity: Confirmatory factor analysis
The STAXI-2 assumes five anger dimensions , from which four dispositional constructs
were included into the analysis4: Trait-Anger, Anger-Control , Anger Expression-In, and
Anger Expression-Out. In the present study, we investigated whether the four-

4 Due to the lack of systematic va riation of the situation an appropria te differentiation between states and
traits (e.g., with the Latent-State-Trait-Model; Steyer, 1987) was not possible. Hence the state items were
not included into the analysis.

S. L. Etzler, S. Rohrmann & H. Brandt 186
dimensional structure could be confirmed in the prison sample by analyzing the data with
a confirmatory factor analysis (CFA; see e. g., Bollen, 1989). Because of the small sam-
ple size, complexity of the item information was reduced by forming item parcels (Ban-dalos, 2002). For each latent dimension, three item parcels were formed by aggregating two to four items. We applied two different estimators to the data: A standard Maximum Likelihood estimator (Mplus; Muthén & Muthén, 1998-2012) and a Bayesian estimator (MCMC estimator implemented in OpenBugs ; Lunn, Spiegelhalter, Thomas, & Best,
2009). For small sample sizes the Bayesian estimator has been shown to give reliable results (Dunson, 2000; Lee & Song, 2004, 2012) . Though, information on the prior dis-
tribution is crucial.
The ML analysis of the factor model indi cated a good model fit with a non-significant
χ2-value of 49.60 (df = 60, p = .83). Descri ptive fit indices also suggested a good or
acceptable fit of the data to the model (RMS EA = .00, SRMR =.08, CFI = 1.00, NFI =
1.05, BIC = 2,532.89). No additional residual covariances needed to be specified to
enhance model fit, which would have complicat ed the interpretation of latent constructs
(Bollen, 1989). Additionally, a one-factor model was analyzed that modeled a general Anger construct in order to test whether a more parsimonious one-dimensional structure
suffices to explain the construct. Fit indices indicated a clearly worse fit than the four
factor model (BIC = 2,646.62) and the same was true with regard to the
χ2 test ( χ2 =
187.37, df = 66, p < .05).
For the Bayesian analysis, we used informa tive conjugate priors (e.g., Gelman et al.,
2004) with information for the priors stemming from a factor analysis of the standardiza-tion sample (Rohrmann et al., 2013). Convergence criteria (Gelman plot and density plots, see Gelman, 1996) indicated convergence after about 1,500 iterations, which were
discarded as burn-in iterations. The f our-factor model showed better fit (DIC
5 = 2,187)
than a one-factor model (DIC = 2,537).
Results for parameter estimation are presented in Table 3. The results for both analyses
were fairly similar. All factor loadings were significant and had medium to large effect sizes, with standardized factor loadings ranging from .58 to .93 (for the ML estimator)
and from .56 to .99 (for the Bayesian estimator). The correlations between the latent factors approached the correlations between scales reported in Table 2 above. In all,
results of the CFA, i.e., good model fit, lo w to medium-sized correlations, and high fac-
tor loadings support the hypothesized four-dimensional structure for the STAXI-2 in the prison sample.

5 Deviance information criterion (Spiegelhalter, Be st, Carlin, & van der Linde, 2002; Celeux, Forbes,
Robert, & Titterington, 2006).

Validation of the STAXI-2: A study with prison inmates 187
Table 3:
Parameter Estimates for the Factor Loadings of the Four-Factor Model Based on Item Parcels
for a Standard ML Estimator and a Bayesian Estimator
ML factor analysis B ayesian factor analysis
Factor
loading S.E. Standardized
factor loadingFactor
loading S.E. Standardized
factor loading
Trait-Anger
Item parcel 1 1.00 .58 1.00 .56
Item parcel 2 1.75 .41 .91 1.99 .36 .99
Item parcel 3 1.29 .31 .89 1.47 .29 .97
Anger-Control
Item parcel 1 1.00 .80 1.00 .77
Item parcel 2 .83 .15 .78 .93 .18 .84 Item parcel 3 .81 .14 .80 .90 .18 .85
Anger Expression-In
Item parcel 1 1.00 .66 1.00 .61
Item parcel 2 1.42 .29 .83 1.67 .34 .89 Item parcel 3 1.53 .30 .87 1.79 .34 .94
Anger Expression-Out
Item parcel 1 1.00 .75 1.00 .81
Item parcel 2 .77 .18 .67 .74 .16 .70 Item parcel 3 1.05 .21 .76 1.08 .20 .84

Construct validity
In order to investigate the construct valid ity, we examined relationships between the
STAXI-2 and seven subscales of the PFI. Corre lations are reported that are significant on
the 5% Type I error level. As Table 4 show s, Trait-Anger and Social Adjustment were
significantly and negatively correlated (r = −.29), which implies a tendency that higher
Trait-Anger is related to a lower Social Ad justment. Anger Expression-In was correlated
with Emotional Instability and Optimistic Recklessness with r = .25 and r = -.26, respec-tively, which tended towards significance. While the increase of Anger Expression-In was accompanied by increased Emotional Instability, high scores in Anger Expression-In were associated with lower Optimistic Recklessness. The associations between Anger Expression-Out and the PFI scales showed th ree significant results. Besides a high nega-
tive association of Anger Expression-Out and Social Adjustment (r = -.31), two Anger Expression-Out correlated positively with Perceived Autonomy and Dominance against Inmates and Aggressiveness (r = .30 and r = .32, respectively).

S. L. Etzler, S. Rohrmann & H. Brandt 188
Table 4:
Correlations of STAXI-2 and PFI Scales
Trait-
Anger Anger
Expression
-In Anger
Expression
-Out Anger-
Control
Experience of
Incompetence .02 .05 .15 -.26
Emotional Instability -.04 .25 .12 -.04
Optimistic Recklessness -.09 -.26 -.12 .26 Social Adjustment -.29* .16 -.31* .25 Need for Attention and
Encouragement from Inmates .21 .05 .17 -.18
Perceived Autonomy and
Dominance against Inmates .19 .15 .30* .07
Aggressiveness .20 .18 .32* .00
Note: * p < .05.

Criterion validity
In order to test whether the sample of incarcerated subjects showed higher T-Scores in
the STAXI-2 scales, we calculated means, st andard errors and confidence intervals for
four subscales (see Table 5). In the current study, the subscale State-Anger was omitted,
because information concerning the subjects' emotional state during participation was only of limited interest. Subjects showed a significantly increased average T-score of
54.61 (SD = 1.56, p < .05) for the Anger E xpression-In scale with a 95% confidence
interval ranging from 51.48 to 57.74. Average scor es for the other three scales were close
to 50 and did not differ significantly from the mean of the standardization sample.

Table 5:
Descriptive Statistics for T-scores of the STAXI Scales and their Confidence Intervals (CI)
for STAXI-2
Means SD CI (95%)
Trait-Anger 49.51 1.32 46.87 – 52.17
Anger Expression-Out 49.91 1.18 47.54 – 52.28
Anger Expression-In 54.61 1.56 51.48 – 57.74 Anger-Control 51.87 1.46 48.94 – 54.80
Note: n = 56.

Validation of the STAXI-2: A study with prison inmates 189
To predict the length of sentence with the f our constructs Trait-Anger, Anger Expres-
sion-In, Anger Expression-Out, and Anger-Control we conducted a multiple linear re-gression analysis. The variance explained by a ll four predictors together was 29.0% (F =
4.29, df
h = 4, df e = 42, p < .05), but only Anger Expr ession-In and Anger Expression-Out
contributed significantly to the prediction of length of sentence (with standardized coef-
ficients of β = .339, p < .05, and β = .295, p <.05, respectively). A regression with these
two predictors reduced the explained variance to 26.2% (F = 7.81; df h = 2; df e = 44, p <
.05). The decrement was not significant (F = .83, df h = 2, df e = 42, p = .45).
Next, we examined the correlation between Tr ait-Anger and the Age of first offense,
because we did not assume a direction of association between these two variables. Re-sults indicated that there is a negative association between Trait-Anger and Age of first offense with r = −.28 (p < .05,
n = 55). The higher a subject scored in Trait-Anger, the
younger he was at the time of his first offens e. The other scales showed no significant
correlations with the Age of first offense.
Discussion
In this study, we examined the reliability, cr iterion validity, and construct-related validity
of the STAXI-2 in a sample of incarcerated men. High reliability coefficients for the subscales indicated that a precise measurement of anger could be achieved with the STAXI-2 for the inmate sample. Also, correla tions between the STAXI-2 subscales were
mainly comparable to those found for the standardization sample of the STAXI-2. For the investigation of factorial validity, two c onfirmatory factor analyses were conducted
which both assumed a four-dimensional measurement of anger. Fit indices showed a good model fit, which confirmed the postulated four-factor structure of the STAXI-2 and
replicated findings of the standardization sample by Rohrmann et al. (2013). The fairly similar results in both the ML analysis and the Bayesian analysis suggest that the found four-dimensional is not purely artifactual due to the small sample size (cf. estimation properties of frequentist approaches in e.g., Dunson, 2000; Lee & Song, 2004, 2012).
These results confirm that the STAXI-2 is a reliable and adequate instrument to be ad-ministered to inmates.
To examine the construct validity of the STAXI-2, we analyzed associations between
STAXI-2 and a questionnaire of personality tra its relevant to prison life (PFI; Seitz &
Rautenberg, 2010). Correlations between the STAXI-2 and the PFI showed the expected pattern, which supports the assumptions c oncerning construct validity. A strong and
negative correlation between Trait-Anger and Social Adjustment reflects the theory conform involvement of Trait-Anger in inte rpersonal problems and conflicts. Persons
with high Trait-Anger scores are often reported to show a lack of skills of behaving in a
socially adjusted manner. Furthermore, we found correlations between Anger Expres-
sion-Out and self-reported traits concerning in terpersonal behavior, i.e., Aggressiveness,
Social Adjustment, and Perceived Autonomy and Dominance against Inmates. This supports hypotheses concerning Anger and Aggression which were based on the "AHA"-Syndrome (Spielberger, 1988). Correlations betw een Anger Expression-In and Indices of

S. L. Etzler, S. Rohrmann & H. Brandt 190
internalized dispositions (e.g., Emotional Instability) were obtained which tended to-
wards significance. This pattern of relations hips between STAXI-2 scales and inmates'
personality traits thus confirms the construct validity of the STAXI-2.
To investigate criterion-related validity, we compared the means of T-Scores of the in-
mate sample to those of the standardization sample, expecting no differences between both samples. Contrary to this expectation, Anger Expression-In was significantly higher
in inmates than in non-incarcerated subjects. Based on the findings by Spielberger and Snyder (1992), who claimed a relationship between Anger Expression-In and Anxiety, it can be argued that persons in correctional in stitutions are in a situation that actually
evokes increased feelings of anger. However, the only way to express this anger without
risk of negative consequences is an inward expression of anger. This might result in an increased Anger Expression-In for inmates, which may be a consequence rather than the cause of a conviction. As we hypothesized, all other subscales showed no differences
between the sample of inmates and the standardization sample.
We examined the prediction of length of sentence with four relevant STAXI-2 scales.
The scales Trait-Anger and Anger-Control did not add substantially to the prediction of
length of sentence. As expected, subjects with higher Anger Expression-Out reported a
longer sentence, which can be attributed to a higher severity of criminal behavior. The significant prediction by Anger Expression-In, on the other hand, could also be interpret-ed as a consequence of a longer sentence (see above). Thus, the results can serve as an
indicator for criterion validity; one may concl ude that the STAXI-2 measures anger as it
is theoretically defined. Inspection of the relationship between Trait-Anger and Age at
the first offense indicated that the younger an inmate was at the time of his first offense, the higher the Trait-Anger he reported in the STAXI-2. Those with higher Trait-Anger
showed a younger age of onset of criminal activity and research (Tolan, 1987) demon-strates younger of age of onset of criminal ac tivity is associated with greater involvement
in criminal activity. This could be indirectly interpreted as a support of the assumption
that persons with a higher Trait-Anger are more likely to be involved in situations that
include criminal actions and thus may comm it their first offense at an earlier age.
Limitations of the study
Although our results support the postulated reliability and validity of the STAXI-2
among inmates, some limitations of this study need to be discussed. Due to the small sample size, our results should be interpre ted with caution. Future studies should be
based on larger samples and they should include women, adolescents, and also offenders who committed more severe offenses, e.g., mu rder. But despite the small sample size of
the current study and its potentially low power for inferential tests, the results can con-tribute substantially to the validation of th e STAXI-2, because effect sizes, for example,
for correlations between the scales of the ST AXI-2, were very similar to those found in
the standardization sample. This implies that the observed associations between these variables are comparable in both samples.

Validation of the STAXI-2: A study with prison inmates 191
In this study, we examined the factorial validity of the STAXI-2 in a sample of incarcer-
ated men. Further research should investigate whether measurement invariance can be assumed for incarcerated men and the standardization sample. But for an item-based analysis (e.g., by conducting a Multi Sample Analysis; Bollen, 1989) a larger sample size would be required.
In the current study, the data was based on se lf-report information. Therefore, measures
might be biased by different kinds of respons e styles, for instance, exaggeration, minimi-
zation, malingering, and especially social desirability. Rohrmann et al. (2013) reported
correlations of the STAXI-2 and social desira bility which may also pertain to self-reports
of offenders, who are probably motivated to underreport anger problems. Future research
needs to examine the magnitude of effects of response styles on STAXI-2 scores in in-mate samples by administering validation scales (e.g. for measuring social desirability) and by an additional consideration of behavi oral variables (e.g., verbal anger expres-
sions).
Another topic which could not be addressed in this study concerns potential interaction
effects of anger-related variables and coping skills for the prediction of criminal behav-ior. Although theories postulate associations between anger and criminal behavior in
particular, empirical research could not provide a precise definition of this relationship.
Conclusion
The current study indicates that the STAXI-2 has a good reliability, factorial validity,
criterion validity, and construct validity within a sample of inmates. The results from this
initial study suggest that further research of the STAXI-2 with inmates is in order. Clear-
ly, the instrument needs to be tested in a much larger, representative sample of criminal offenders in prisons before recommending its widespread use. Diagnosticians should
keep in mind that anger does not function as an empirically supported predictor of crimi-nal recidivism. Nevertheless, measuring anger in correctional settings could be helpful to identify reasons for problems of conduct, or to develop treatment recommendations.
References
Agnew, R. S. (1992). Foundation for a general strain theory of crime and delinquency. Crim-
inology , 30, 47-87.
Agnew, R. S. (2001). Building on the foundation of general strain theory: Specifying the
types of strain most likely to lead to crime and delinquency. Journal of Research in
Crime and Delinquency, 36, 123-155.
Agnew, R. S. (2007). Strain theory and violent be havior. In D. J. Flannery, A. T. Vazsonyi, &
I. D. Waldman (Eds.), The Cambridge handbook of vi olent behavior and aggression (pp.
519-529). New York: Cambridge University Press.

S. L. Etzler, S. Rohrmann & H. Brandt 192
Bandalos, D. L. (2002). The effects of ite m parceling on goodness-of-fit and parameter esti-
mate bias in structural equation modeling. Structural Equation Modeling, 9 , 78-102.
Baumeister, R. F., Stillwell A., & Wotman, S. R. (1990). Victim and perpetrator accounts of
interpersonal conflict: Autobiogra phical narratives about anger. Journal of Personality
and Social Psychology, 59 , 994-1005.
Bollen, K. A. (1989). Structural equations with latent variables . New York: Wiley.
Celeux, G., Forbes, F., Robert , C. P., & Titterington, D. M. (2006). Deviance information
criteria for missing data models. Bayesian Analysis, 1 , 651-674.
Cherek, D. R., Scott, D. L., Dougherty, D. M ., Moeller, F. G. & White, S. (2000). Laboratory
and questionnaire measures of aggression am ong female parolees with violent or nonvio-
lent histories. Aggressive Behavior, 26 , 291-307.
Cornell, D. G., Peterson, C. S., & Richards, H. (1999). Anger as a predictor of aggression
among incarcerated adolescents. Journal of Consulting and Clinical Psychology, 67 , 108-
115.
Dalton, J. E., Blain G. H., & Bezier, B. (1998). State-Trait Anger Expression Inventory
Scores of Male Sexual Offenders. International Journal of Offender Therapy and Com-
parative Criminology, 42 , 141-148.
Deffenbacher, J. L., Oetting, E. R., Thwaites, G. A., Lynch, R. S., Baker, D. A., Stark, R. S.,
Thacker, S., & Eiswerth-Cox, L. (1996). State- trait anger theory and the utility of the
Trait Anger Scale. Journal of Counseling Psychology , 43, 131-148.
Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977). Maximum likelihood from incomplete
data via the EM algorithm. Journal of the Royal Statistical Society, Ser. B, 39 , 1-38.
Dunson, D. B. (2000). Bayesian latent vari able models for clus tered mixed outcomes. Journal
of the Royal Statistical Society: Ser. B, 62 , 355-366.
Fahrenberg, J., Hampel, R., & Selg, H. (2001). Das Freiburger Persönlichkeitsinventar,
revidierte Fassung (FPI-R). Handanweisung (7. Aufl.). Göttingen: Hogrefe.
Gelman, A. (1996). Inference and monitoring conve rgence. In W. R. Gilks, S. Richardson, &
D. J. Spiegelhalter (Eds.), Markov Chain Monte Carlo in practice (pp. 131-143). Boca
Raton, FL: Chapman & Hall/CRC.
Gelman, A., Carlin, J. B., Stern, H. S., & Rubin, D. B. (2004). Bayesian Data Analysis . Boca
Raton, FL: Chapman & Hall/CRC.
Hazaleus S. L., & Deffenbacher, J. L. (1986). Relaxation and cognitive treatments of anger.
Journal of Consulting and Clinical Psychology, 54 , 222-226.
Hodapp, V., & Bongard, S. (2009). Ärger. In V. Brandstätter & J. H. Otto (Eds.), Handbuch
der Allgemeinen Psychologie – Motivation und Emotion (pp. 612-622). Göttingen:
Hogrefe.
Izard, C. E. (1991). The psychology of emotions . New York: Plenum Press.
Kalichman, S. C. (1991). Psychopathology and pe rsonality characteristics of criminal sexual
offenders as a function of victim age. Archives of Sexual Behavior, 20, 187-197.

Validation of the STAXI-2: A study with prison inmates 193
Kroner, D. G., & Reddon, J. R. (1992). The A nger Expression Scale and State-Trait Anger
Scale; Stability, Reliability and Factor Structure in an Inmate Sample. Criminal Justice
and Behavior, 19 , 397-408.
Lee, S. Y., & Song, X. Y. (2012). Basic and advanced Bayesian structural equation model-
ing: With applications in the medical and behavioral sciences . New York: Wiley.
Lee, S. Y., & Song, X. Y. (2004). Evaluati on of the Bayesian and Maximum likelihood ap-
proach in analyzing structural equa tion models with small sample sizes. Multivariate Be-
havioral Research, 39, 653-686.
Little, R. J. A., & Rubin, D. B. (2002). Statistical analysis with missing data . New York:
Wiley.
Loza, W., & Loza-Fanous, A. (1999a). Anger a nd prediction of violent and nonviolent of-
fenders' recidivism. Journal of Interpersonal Violence, 14, 1014-1029.
Loza, W., & Loza-Fanous, A. (1999b). The fallacy of reducing rape and violent recidivism by
treating anger. International Journal of Offender Therapy and Comparative Criminology,
43, 492-502.
Lunn, D., Spiegelhalter, D., Thomas, A., & Best, N. (2009). The BUGS project: Evolution,
critique, and future directions. Statistics in Medicine, 28 , 3049-3067.
Maiuro R. D., Cahn, T. S., Vitaliano, P. P., Wagner, B. C., & Zegree, J. B. (1988). Anger,
hostility, and depression in domestically violen t versus generally assaultive men and non-
violent control subjects. Journal of Consulting and Clinical Psychology, 56 , 17-23.
Mills, J. F., & Kroner, D. G. (2003) Anger as a Predictor of Institutional Misconduct and
Recidivism in a Sample of Violent Offenders. Journal of Interpersonal Violence, 18, 282-
294.
Müller, C., Bongard, S., Heiligtag, U., & Hodapp, V. (2001). Das State-Trait-
Ärgerausdrucks-Inventar (STAXI) in der klinischen Anwendung: Reliabilität und faktori-
elle Validität. Zeitschrift für Klinische Psyc hologie und Psychotherapie, 30 (3), 172-181.
Muthén, L. K., & Muthén, B. O. (1998-2012). M plus User's Guide. Version 7. Los Angeles,
CA: Muthén & Muthén.
Novaco, R. W. (1975). Anger Control: The development and evaluation of an experimental
treatment . Lexington, MA: Lexington Books.
Novaco, R. W. (1994). Anger as a risk factor for violence among the mentally disordered. In
J. Monahan & H. Steadman (Eds.), Violence and mental disorder: Developments in risk
assessment (pp. 21-59). Chicago: University of Chicago Press.
Rohrmann, S., Hodapp, V., Schnell, K., Tibubos, A. N., Schwenkmezger, P., & Spielberger,
C. D. (2013). Das State-Trait-Ärgerausdrucks-Inventa r-2 (STAXI-2). Deutschsprachige
Adaptation des State-Trait A nger Expression Inventory-2 (STAXI-2) von Charles D.
Spielberger. Bern: Verlag Hans Huber.
Seitz, W., & Rautenberg, M. (2010). Persönlichkeitsfragebogen für Inhaftierte (PFI). Göt-
tingen: Hogrefe.
Scherer, K. R. (1990). Theorien und aktuelle Probleme der Emotionspsychologie. In K. R.
Scherer (Ed.), Psychologie der Emotion (pp. 1-38). Göttingen: Hogrefe.

S. L. Etzler, S. Rohrmann & H. Brandt 194
Schermelleh-Engel, K., & Werner, C. (2011). Methoden der Reliabilitätsbestimmung. In H.
Moosbrugger & A. Kelava (Eds.), Testtheorie und Fragebogenkonstruktion . Heidelberg:
Springer.
Schmitt, M., & Altstötter-Gleich, C. (2010). Differentielle und Persönlichkeitspsychologie –
kompakt. Weinheim: Beltz.
Schwenkmezger, P., H odapp, V., & Spielberger, C. D. (1992). Das State-Trait-Ärgeraus-
drucks-Inventar (STAXI) . Bern: Huber.
Shealy, L., Kalichman, S. C., Henderson, M. C., Szymanowski, D., & McKee, G. (1991).
MMPI profile subtypes of incarcerated sex offenders against children. Violence and Vic-
tims, 6, 201-212.
Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & van der Linde, A. (2002). Bayesian
measures of model complexity and fit (with discussion). Journal of the Royal Statistical
Society, Series B, 64 , 583-616.
Spielberger, C. D. (1988). State-Trait-Anger-Expression-Inventory . Palo Alto, CA: Consult-
ing Psychologist Press.
Spielberger, C. D. (1999). State-Trait Anger Expression Inven tory-2 (STAXI-2). Professional
Manual . Tampa, FL: Psychological Assessment Resources.
Spielberger, C. D., Johnson, E. H ., Russel, 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. Ches ney & R. H. Rosenman (Eds.), Anger and hostility in car-
diovascular and behavioral disorders (pp. 5-30). New York: Hemisphere/McGraw-Hill.
Spielberger, C. D. & Reheiser, E. C. (2010). The nature and measurement of anger. In M.
Potegal, G. Stemmler, & C. D. Spielberger (Eds.), International handbook of anger (pp.
403-412). New York: Springer Sc ience + Business Media.
Spielberger, C. D., & Sydeman, S. J. (1994). State-Trait Anxiety Inventory and State-Trait
Anger Expression Inventory. In M. E. Maurish (Ed.), The use of psychological tests for
treatment planning and outcome assessment (pp. 292-321). Hillsdale, NJ: Erlbaum.
Steyer, R. (1987). Konsistenz und Spezifität: Defi nition zweier zentraler Begriffe der Diffe-
rentiellen Psychologie und ein einfaches Modell zu ihrer Identifikation [Consistency and
specificity: Definition of two pivotal concep ts of differential psychology and a simple
model for their identification] Zeitschrift für Differentie lle und Diagnostische Psycholo-
gie, 8, 245-258.
Tolan, P. H. (1987). Implications of Age of Onset for Delinquency Risk. Journal of Abnormal
Child Psychology, 15 (1), 47-65.
Zamble, E., & Quinsey, V. L. (1997). The criminal recidivism in process . Cambridge, UK:
Cambridge University Press.