Non-PRO Homework / test English to Malay Other Psychology JUVENILE DELINQUENCY

Attributions Toward Violence of Male Juvenile Delinquents: A Concurrent Mixed-Methodological Analysis
Christine E Daley, Anthony J Onwuegbuzie. The Journal of Social Psychology. Washington: Dec 2004.Vol.144, Iss. 6; pg. 549, 22 pgs
» Jump to full text Full text
»
Translate document into:
» More Like This - Find similar documents
Subjects: Social psychology, Violence, Males, Juvenile offenders, Behavior Social psychology, Violence, Males, Juvenile offenders, Behavior
Author(s): Christine E Daley, Anthony J Onwuegbuzie
Document types: Feature
Publication title: The Journal of Social Psychology. Washington: Dec 2004. Vol. 144, Iss. 6; pg. 549, 22 pgs
Source type: Periodical
ISSN/ISBN: 00224545
ProQuest document ID: 737725421
Text Word Count 7913
Document URL: http://proquest.umi.com/pqdweb?did=737725421&sid=1&Fmt=4&cli...

Abstract (Document Summary)

The authors investigated male juvenile delinquents' causal attributions about others' behavior and the salient pieces of information that the male juveniles used in arriving at their attributions. Participants were 82 male juvenile offenders whom the authors drew randomly from the population of juveniles who were incarcerated at a correctional facility. A concurrent mixed-methodological analysis revealed that the juvenile offenders committed attributional errors about violence nearly 53% of the time. Race and number of prior arrests predicted the number of attributional errors about violence. A phenomenological analysis revealed the following 7 themes stemming from juveniles' reasons for causal attributions: self-control, violation of rights, provocation, irresponsibility, poor judgment, fate, and conflict resolution. Some of these themes were related to age, ethnicity, and number of prior arrests. The authors discussed implications of all findings. [PUBLICATION ABSTRACT]

Key words: attribution errors, juvenile delinquents, mixed methods, violence
Full Text (7913 words)
Copyright HELDREF PUBLICATIONS Dec 2004
[Headnote]
ABSRACT.
The authors investigated male juvenile delinquents' causal attributions about others' behavior and the salient pieces of information that the male juveniles used in arriving at their attributions. Participants were 82 male juvenile offenders whom the authors drew randomly from the population of juveniles who were incarcerated at a correctional facility. A concurrent mixed-methodological analysis revealed that the juvenile offenders committed attributional errors about violence nearly 53% of the time. Race and number of prior arrests predicted the number of attributional errors about violence. A phenomenological analysis revealed the following 7 themes stemming from juveniles' reasons for causal attributions: self-control, violation of rights, provocation, irresponsibility, poor judgment, fate, and conflict resolution. Some of these themes were related to age, ethnicity, and number of prior arrests. The authors discussed implications of all findings.
Key words: attribution errors, juvenile delinquents, mixed methods, violence

VIOLENCE AMONG YOUTHS is epidemic. In the United States every 5 min, police arrest a child for a violent crime. Every 2 hr, a child is murdered. In a single day, individuals wielding guns kill 13 children and wound 30 more; 5,703 teenagers are victims of violent crime; 2,350 juveniles are in adult jails; and 1,200,000 latchkey children return to homes in which there is a gun (Children's Defense Fund, 1993). Violent crime has increased 43% since 1980 and 6-fold since 1960 (American Legislative Exchange Council, 1994). However, despite a plethora of research in the area of violence and aggression, investigators still do not understand why our youth continue to become involved in and be victimized by violent acts. So, in the present mixed-methodological study, we investigated male juvenile delinquents' causal attributions about others' behavior and the salient pieces of information that they use in arriving at their attributions. Moreover, this investigation sought to demonstrate the utility of mixed-methodological analyses (cf. Onwuegbuz.ie & Teddlie, 2003; Tashakkori & Teddlie, 1998) in studying social psychological phenomena.

Many researchers have focused on concomitants of violent behavior. They have identified a number of genetic, biological, and familial antecedents, including intellectual functioning, economic deprivation, chronic parental unemployment, family criminality, poor childrearing, child-abuse history, hyperactivity-impulsivity-attention deficit, early onset of aggression, and antisocial behavior (e.g., Busch, Zagar, Hughes, & Arbit, 1990; Calicchia, Moncata, & Santostefano, 1993). More specifically, Cornell (1987) found that the best single predictor of violence is past violent behavior. Nevertheless, most of these factors are relatively immutable, and thus, at best, identify youth who are at risk for violent behavior, having only minimal implications for intervention.

The area of social cognition, including the theory of attributions, appears to offer investigators a viable avenue for research on the antecedents of violent behavior (Andrews & Brewin, 1990; Breterton, Collins, & Ferretti, 1993). Indeed, Guthrie and Betancourt (1991) found that attribution processes play a role in children's reactions to violence.

Vala, Monteiro, and Leyens (1988) described aggression as a special kind of interaction whose features are socially defined; that is, it is the interpretation of the act, rather than its overt manifestations, that is important. Researchers also recognize that this social meaning may vary according to the different participants-actor, observer, or victim (e.g., Jones & Nisbett, 1972)-and according to the causes that the participants attribute to the act (e.g., Kelley, 1973).

Attribution theory (Kelley, 1973) is concerned with the cognitive processes underlying an individual's causal inferences about events occurring within his or her physical and social environment. According to Kelley, attribution theory examines the information that people use in making inferences about causes and what they do with this information to answer causal questions. Within Kelley's model, the question of interest is whether an event should be attributed to the provocation of a target (i.e., stimulus), to exacerbating conditions (i.e., circumstance), or to the actor's disposition (i.e., person; Zebrowitz, 1990). (For a more recent version of attribution theory, see, for example, Malle, 1999.)

A few investigators have examined the causal attributions made by children identified as aggressive or hostile (e.g., Dodge & Coie, 1987). Such research has suggested that aggressive children process information differently than do nonaggressive children and that the differences are directly associated with aggressive interpersonal behavior (Dodge, 1980).

One group of findings suggests that aggressive children are simply deficient in attributional skills. Dodge and Tomlin ( 1983) found that aggressive youngsters were less likely than their nonaggressive peers to rely on social cues and more likely to rely on past experience when making causal inferences about another's behavior. Additionally, Dodge and Newman (1981) reported that aggressive boys made causal inferences more quickly and paid less attention to available social cues than did nonaggressive boys. Dodge, Murphy, and Buchsbaum (1984) reported that when shown video-recorded stimuli of benign provocations, aggressive children demonstrated relative deficits in accurately interpreting others' intentions. Specifically, aggressive children are more likely than their nonaggressive peers to externalize blame in provocative social interactions (Dodge & Coie, 1987). One implication of these findings is that the aggressive, anti-social behavior that some youth demonstrate may be due to their inaccurate or biased appraisals of interpersonal exchanges.

In summary, despite considerable research on youthful aggression, few investigators have examined the role of social cognitive factors-in particular, attributions-in placing children at risk for involvement in acts of violence. Additionally, from a methodological standpoint, investigators typically have made no attempt to approximate experimental conditions by manipulation of an independent variable. This methodological flaw may have culminated in the difficulties that investigators have experienced in predicting violent acts (Capaldi & Patterson, 1993; Cornell, 1987; Schlesinger, 1983).

Thus, the major purpose of the present study was to assess sensitivity to violence by an attributional measure of incarcerated juvenile delinquents. Specifically, we sought to examine juvenile offenders with respect to the proportions of inaccurate causal attributions (i.e., attributional errors about violence) that they make about others' behaviors and the salient pieces of information that they use in arriving at their attributions (i.e., reasons for violence attributions). For the purposes of the present study, violence attributional errors were defined as errors that occur when an offender does not blame the perpetrator of a violent act (e.g., rape) but instead blames either the victim or the circumstance. (Implicit in this definition is the assumption that the authorities who established the events that we portrayed in the vignette made no errors.)

Other goals of the present study were to develop a typology of reasons for violence attributions and to determine whether these reasons predict juvenile delinquents' attributional errors about violence. In the present inquiry, we also attempted to ascertain the antecedent correlates of juvenile offenders' causal attributions. Finally, of particular interest was whether profiles of juvenile delinquents could be developed on the basis of their violence attribution reasons.

We believed that an understanding of the attributional styles of aggressive, undersocialized youth would have implications for confronting the rising concerns about violence among today's youth. Also, it was hoped that in the present study, we would contribute to the knowledge base relating to juvenile delinquents by determining factors that place them at risk, thus helping to identify effective treatment programs and ultimately reducing the overall rate of incarceration.

Method

Participants

Participants were 82 male juvenile offenders whom we drew randomly from the population of juveniles incarcerated at a correctional facility in a large southeastern U.S. state. The sample size of 82 was selected via an a priori power analysis because it provided acceptable statistical power (i.e., .80) for detecting a moderate correlation, r = .30, at the (two-tailed) .05 level of significance (Erdfelder, Faul, & Buchner, 1996). The 82 participants represented 15% of the offenders who were incarcerated at that facility. The participants ranged in age from 12 years to 18 years (M = 15.46 years, SD = 1.28 years), with an average of 3.17 prior arrests (SD = 2.72). In all, 23.2% of the participants were Caucasian American boys, and 76.8% were African American boys.

The juvenile offenders were wards of the state; thus, informed consent to participate in the study was inherent in the institutional permission to conduct the research. The Department of Juvenile Justice at the state where the study took place granted formal consent for both researchers to collect data.

Instruments and Procedure

We administered to participants the Violence Attribution Survey (VAS), which we had developed specifically for the present investigation. The VAS is a 12-item questionnaire designed to assess attributions made by the juveniles about the behavior of others involved in a variety of violent acts. Each item consists of a vignette, followed by three possible attributions (i.e., person, stimulus, and circumstance) presented in multiple-choice format and an open-ended question asking the juveniles for their reason for choosing the response that they did. We had constructed the vignettes so as to allow for the perceived plausibility of any one of the three possible attributions. For the present investigation, the VAS generated scores that had a classical theory Cronbach's alpha reliability coefficient of .71 (95% confidence interval [CI] = .61, .79).

Secondary school teachers reviewed the VAS, and we analyzed it using Grammatik 5 (Reference Software International, 1992) for readability. The scale was found to be suitable for readers at a 5th-grade level. Although we provided participants with the actual instrument in hardcopy form, each question was read aloud by one of us to control for variation in reading skills. We required participants to write their responses on the survey forms. (All participants were able to write their own responses.) After completing the VAS, participants were given refreshments (e.g., soda, potato chips), which they considered according to staff at the facility to represent a special treat.

Analysis

We undertook a concurrent mixed-methodological analysis (CMMA), as described by Onwuegbuzie and Teddlie (2003), to analyze the data. The analysis involved the use of qualitative and quantitative data analytic techniques in a complimentary manner. In accordance with the framework of Greene, Caracelli, and Graham (1989), the purpose of the mixed-methodological analysis was complementarity, that is, using quantitative and qualitative techniques to measure overlapping but also different aspects of the underlying phenomenon, attributional errors about violence. The VAS generated both quantitative information (i.e., multiple-choice responses) and qualitative responses (i.e., reasons for choosing responses). These two measures assessed similar but distinct aspects of attributional errors.

The CMMA involved six stages. The first stage (i.e., an exploratory stage) consisted of the recoding of the multiple-choice responses (i.e., person, stimulus, and circumstance). Because stimulus and circumstance responses represent attributional errors on the VAS, we then combined these two responses and contrasted them with person attributions. That is, responses representing external attributions (i.e., stimulus and circumstance) were compared to responses signifying dispositional attributions (i.e., person), so that external attributions were given a score of 1, and dispositional attributions were given a score of O. We summed responses to the 12 items of the VAS to produce an index of attributional errors about violence (range = 0-12), with high scores being indicative of persons who committed a high proportion of attributional errors. These scores were then used to determine the juvenile delinquents' overall violence attributional error rate. This error rate served as what Onwuegbuzie (in press) termed as a manifest effect size (i.e., an effect size pertaining to observable content).

The second stage (i.e., an exploratory stage) consisted of a phenomenological mode of inquiry to examine participants' reasons for their attributions (i.e., person, stimulus, and circumstance; Goetz & Lecompte, 1984). Specifically, a modification of Colaizzi's (1978) phcnomenological analytic methodology was used. The procedural steps that we used were as follows:

1. We read all the juveniles' reasons to obtain an overall picture of them.

2. We then unitized (i.e., categorized into units; Lincoln & Guba, 1985) these adolescents' responses.

3. We used these units of information as the basis for extracting a list of nonrepetitive, nonoverlapping significant statements (i.e., horizonalization of data), treating each statement as having equal status; we eliminated units that contained the same or nearly the same statements, so that each unit corresponded to a unique violence attribution reason.

4. We formulated meanings by specifying the meaning of each significant statement (i.e., unit).

5. We organized clusters of themes from the aggregate formulated meanings, with each cluster containing units that appeared similar in content, so that each cluster signified a distinct emergent theme (i.e., method of constant comparison; Lincoln & Cuba, 1985); we compared and contrasted these clusters of themes with the original descriptions to validate them (i.e., to assess trustworthiness of categorization).

We used this five-step method of analysis to reveal a number of themes relating to the offenders' reasons for their attributions.

The third stage (i.e., the exploratory stage) of the mixed-methodological analysis involved using descriptive statistics to analyze the hierarchical structure of the emergent themes (Onwuegbuzie & Teddlie, 2003). In particular, each theme was quantitized (quantitizing is converting qualitative data into numerical codes that can be statistically analyzed; Tashakkori & Teddlie, 1998). Specifically, for each participant, a score of 1 was given for a theme if it represented at least one of the reasons cited by the participant for the 12 attributions made on the VAS; otherwise, a score of 0 was given for that theme. That is, for each sample member, each theme was quantitized to a score of either 1 or 0, depending on whether it was represented by that individual. This dichotomization led to the formation of an interrespondent matrix (i.e., Participant ~{!A~} Theme matrix; Onwuegbuzie, in press). This matrix contained a combination of "0"s and "1"s. The quantitizing of themes allowed the computation of an additional manifest effect size. Specifically, we obtained a measure of frequency effect size (Onwuegbuzie) by calculating the frequency of each theme from the interrespondent matrix and then converting these frequencies to percentages. These percentages represented the prevalence rate of each theme. The interrespondent matrix was used to determine the relationship between responses to each theme (i.e., 0 vs. 1) and the violence attributional error rate. We examined the associations between responses to each theme and the demographic variables (i.e., age, ethnicity, and the number of prior arrests) also.

The fourth stage of the mixed-methodological analysis involved the use of the interrespondent matrix to conduct an exploratory factor analysis to ascertain the underlying structure of these themes (i.e., an exploratory stage). This factor analysis determined the number of factors underlying the themes. These factors, or latent constructs, represented metathemes (Onwuegbuzie, 2003) each of which contained one or more of the emergent themes. The trace, or proportion of variance explained by each factor after rotation, served as a latent effect size for each metatheme (Onwuegbuzie). As described by Onwuegbuzie, we computed an additional latent effect size via the use of odds ratios. Specifically, we determined and used the odds ratios among the metathemes to compare prevalence rates among them (Onwuegbuzie). Also, we computed a manifest effect size for each metatheme by determining the combined frequency effect size for the themes within it (Onwuegbuzie).

The fifth stage (i.e., the confirmatory stage) of the mixed-methodological analysis involved the determination of antecedent correlates of the emergent themes that were extracted in Stage 1 and quantitized in Stage 2. This phase used the interrespondent matrix to undertake (a) a series of correlational analyses and Fisher's Exact tests, depending on whether the demographic variables were represented by the interval scale (i.e., age and number of prior arrests) or the nominal scale (i.e., ethnicity), to test the hypothesis that the selected background variables would be related to each of the themes; and (b) a canonical correlation analysis to examine simultaneously the relationship between the themes and the demographic variables.

The sixth and final stage (i.e., an exploratory stage) of the mixed-methodological analysis involved narrative profile formation. Specifically, we determined the number of average profiles (Tashakkori & Teddlie, 1998) using an ipsative approach in which juveniles' responses to each theme were interpreted relative to their responses to the other themes (Block, 1957) in the following manner: (a) for each adolescent, we ranked the emergent theme scores (i.e., 0 or 1) so that each scale took on a value from one through the number of emergent themes; and (b) the measure of similarity was based on the theme scores ranked from lowest to highest within each profile. We then formed an intra-individual correlation matrix by correlating each pair of profiles, yielding (n)(n - 1 )/2 Spearman Rho values (where n was the number of respondents). This correlation matrix was cluster-analyzed so that we could characterize individual patterns for each member of the offender sample. The formation of average profiles represented the qualitizing (i.e., transforming the quantitative data into data that can be analyzed qualitatively) of previously quantitized themes (Tashakkori & Teddlie, 1998). We compared the eigenvalues for each cluster-solution to determine the number of interpretable profiles. Each profile was compared and contrasted by determining whether, within each theme, the confidence intervals (i.e., standard error bars) overlapped, as well as by computing within-theme manifest effect sizes (Onwuegbuzie & Teddlie, 2003). Finally, we compared each profile group with respect to the selected demographic variables.

Results

Stage 1 Analysen

Scores on the VAS ranged from 0 to 12, with a mean number of attributional errors of 6.30 (SD = 2.82). The 95% confidence interval (CI) associated with this mean number of attributional errors was 5.69 to 6.91. In other words, on average, the juvenile offenders were committing attributional errors 52.99% of the time (SD = 23.44%; 95% CI = 47.92%, 58.06%). According to Onwuegbuzie and Daniel (2002), normality is suggested if both the standardized skewness coefficient (skewness divided by its standard error) and the standardized kurtosis coefficient (kurtosis divided by its standard error) are within the ~{!@~}2 range. Encouragingly, an examination of the standardized skewness (-0.30) and kurtosis (-1.30) coefficients pertaining to the number of attributional errors about violence made by each juvenile sample member, as well as the corresponding normal probability plot, suggested no marked departure from normality. This evidence of normality, with the randomness of the sample, indicates that, for any particular item on the VAS, investigators can expect with 95% confidence that between 47.92% and 58.06% of juvenile delinquents would make attributional errors. This situation suggests a moderate-to-large effect size.

Interestingly, Black juvenile delinquents (M = 7.05, SD = 3.84) were statistically significantly more likely, t(80) = 5.10, p < .0001, to commit attributional errors about violence than were their White counterparts (M = 3.84, SD = 2.36). The effect size that was associated with this difference, as measured by Cohen's (1988) d, was .91, which was extremely large. Further, we found a positive relationship between the number of prior arrests and the number of attributional errors about violence, r(80) = .28, p < .0001. This association represented a moderate effect size. However, no relationship between age and the number of attributional errors about violence emerged, r(80) = -.17, p > .05.

Stage 2 Analyses

The juvenile participants listed a total of 441 unique reasons across the 12 VAS items (M = 36.75, SD = 8.36). Table 1 presents the themes that emerged from the students' violence attribution reasons, alongside their attribution categories, and examples of statements representing each theme. We extracted the following seven themes from the responses: self-control, violation of rights, provocation, irresponsibility, poor judgment, fate, and conflict resolution. The first two themes were associated with the actor's disposition (i.e., person), the middle three themes pertained to the provocation of a target (i.e., stimulus), and the last two themes represented the exacerbating conditions (i.e., circumstance). The overall interrater reliability between the two coders pertaining to the categorization of the units into the six themes was .95.

Stage 3 Analyses

Table 1 presents the prevalence rates of each theme (i.e., [manifest] frequency effect sizes; Onwuegbuzie, in press; Onwuegbuzie & Teddlie, 2003) also. Interestingly, the three stimulus themes-provocation, irresponsibility, and poor judgment-were the most endorsed themes, with more than three fourths of the sample citing one or more reasons that fell into these categories. The two person themes-self-control and violation of rights-were the next most endorsed themes, with 58.5% and 42.7% of the offenders providing violence attribution reasons that pertained to these classifications, respectively. Finally, the two circumstance themes-fate and conflict resolution-were the least endorsed themes.
Table
Enlarge 200%
Enlarge 400%
TABLE 1. Open-Ended Response Categories With Selected Examples of Significant Statements of Attributions and Endorsement Rates

Of the intercorrelations among the seven themes (not presented), after the Bonferroni adjustment (Onwuegbuzie & Daniel, 2002), only the correlation between (a) responses categorized as belonging to the self-control theme and (b) responses belonging to both the violation of rights theme (i.e., r[80] = .33, p < .003) and the conflict resolution theme (i.e., r[80] = .39, p < .0002) were statistically significant. Using Cohen's (1988) criteria, we found that these relationships were moderate. Specifically, juvenile delinquents who tended to cite lack of self-control on the part of the actor as the reason for their violence attributions also tended to provide violation of rights and conflict resolution as explanations for their attributions.

We used a series of independent-samples t tests to compare juveniles who endorsed each of the seven themes to those who did not endorse these themes with respect to the violence attributional error rate. Table 2 shows these results. After the Bonferroni adjustment, it can be seen (a) that juveniles who endorsed the self-control theme tended to make less attributional errors about violence than did their counterparts; (b) that juveniles who endorsed the violation of rights theme tended to make less attributional errors about violence than did their counterparts; (c) that juveniles who endorsed the provocation theme tended to make more attributional errors about violence than did their counterparts; and (d) that juveniles who endorsed the poor judgment theme tended to make more attributional errors about violence than did their counterparts. The Cohen's d effect sizes pertaining to these differences were extremely large, ranging from 0.90 to 1.15.
Table
Enlarge 200%
Enlarge 400%
TABLE 2. Means, Standard Deviations, t Values, and Effect Sizes Pertaining to Attributional Error Rate Differences for Each Theme

Stage 4 Analyses

We used an exploratory factor analysis to determine the number of factors underlying the six themes. Specifically, a maximum likelihood factor analysis was used. This technique, which gives better estimates than does principal factor analysis (Bickel & Doksum, 1977), is perhaps the most commonly used method of common factor analysis (Lawley & Maxwell, 1971). As recommended by Kieffer (1999) and Onwuegbuzie and Daniel (2003), we used the correlation matrix to undertake the factor analysis. An orthogonal (i.e., varimax) rotation was used because of the low degree of correlations among the themes. We used this analysis to extract the latent constructs. As conceptualized by Onwuegbuzie (in press), these factors represented metathemes.

We implemented the eigenvalue-greater-than-one rule, also known as K1 (Kaiser, 1958), to ascertain an appropriate number of factors to retain. This technique resulted in four factors (i.e., metathemes). The scree test (Cattell, 1966; Zwick & Velicer, 1986) also suggested that four factors be retained. Table 3 presents this four-factor solution. After a cutoff correlation of 0.5, recommended by Hair, Anderson, Tatham, and Black (1995), as an acceptable minimum loading value, this table shows (a) that the following themes loaded significantly on the first factor: conflict resolution, self-control, and violation of rights; (b) that the following themes loaded on the second factor: poor judgment and irresponsibility; (c) that the following theme loaded on the third factor: fate; and (d) that the following theme loaded on the fourth factor: provocation.

Clearly, the first metatheme (i.e., Factor 1) can be labeled disposition of actor and interaction with stimulus. The second metatheme can be labeled cognitively based stimulus. The third metatheme can be labeled circumstance. Finally, the fourth metatheme can be labeled emotionally based stimulus. Figure 1 presents the thematic structure. The figure illustrates the relationships among the themes and metathemes arising from offenders' reasons for their violence attributions.

An examination of the trace (i.e., the proportion of variance explained, or eigenvalue, after rotation; Hetzel, 1996) revealed that the metatheme of the disposition of actor and interaction with stimulus (i.e., Factor 1) explained 25.57% of the total variance, the metatheme of the cognitively based stimulus (i.e., Factor 2) explained a further 21.29% of the variance, the metatheme of circumstance (i.e., Factor 3) explained an additional 15.14% of the variance, and the metatheme of the emotionally based stimulus (i.e., Factor 4) explained a further 14.29% of the variance. These four metathemes together explained 76.29% of the total variance. This total proportion of variance represents a latent effect size, which can be considered very large. The manifest effect sizes associated with the four metathemes (i.e., the prevalence rates of the metathemes that were based on the juveniles' violence attribution reasons) follow: for disposition of actor and interaction with stimulus, 71.9%; for cognitively based stimulus, 92.7%; for circumstance, 40.2%; and for emotionally based stimulus, 76.8%.
Table
Enlarge 200%
Enlarge 400%
TABLE 3. Summary of Themes and Factor Loadings From Maximum Likelihood Varimax Factor Analysis: Four-Factor Solution

Compulation of odds ratios revealed that the metatheme of cognitively based stimulus was 6.00 (95% CI = 1.02, 35.40) times more likely to be endorsed than was the metatheme of the disposition of actor and interaction with stimulus, 1.53 (95% CI = 0.29, 8.13) limes more likely to be endorsed than was the metatheme of circumstance, and 3.75 (95% CI = 0.69, 20.38) times more likely to be endorsed than was the metatheme of emotionally based stimulus. Also, the metatheme of emotionally based stimulus was 1.25 (95% CI = 0.41, 3.81) times more likely to be endorsed than was the metatheme of the disposition of actor and interaction with stimulus and 1.21 (95% CI = 0.42, 3.47) times more likely to be endorsed than was the metatheme of circumstance. Finally, the metatheme of the disposition of actor and interaction with stimulus was 1.38 (95% CI = 0.51, 3.75) more likely to be endorsed than was the metatheme of circumstance. In summary, the odds ratios ranged from 1.21 to 6.00, with the difference in endorsement rates between the metatheme of cognitively based stimulus and the metatheme of the disposition of actor and interaction with stimulus being by far the most practically significant one.
Chart
Enlarge 200%
Enlarge 400%
FIGURE 1. Thematic structure pertaining to juvenile delinquents' reasons for their violence attributions.

Stage 5 Analyses

A series of correlational analyses, with the Bonferroni adjustment controlling for family-wise error (p < .05), indicated that older juveniles were more likely to endorse the self-control theme than were younger offenders, r(80) = .30, p < .01. Age was not statistically significantly related to any other theme. Also, delinquents with the greatest number of prior arrests were least likely to endorse the theme of violation of rights, r(80) = -.28, p < .01. The number of prior arrests was not statistically significantly related to any of the other themes.

A series of Fisher's Exact tests, with the Bonferroni adjustment controlling for family-wise error, indicated that the White juvenile delinquents were statistically significantly more likely than were the Black offenders to provide a reason pertaining to self-control (84.21% vs. 50.79%; Cramer's V = 0.29; odds ratio = 5.15; 95% CI = 1.48, 18.18), more likely than were the Black offenders to provide a reason pertaining to violation of rights (78.95% vs. 31.75%; Cramer's V = 0.40; odds ratio = 8.06; 95% CI = 2.37, 27.78), and less likely than were the Black offenders to cite a reason relating to provocation (47.37% vs. 85.71%; Cramer's V = 0.38; odds ratio = 6.67; 95% CI = 2.12, 20.93).

We undertook a canonical correlation analysis to examine simultaneously the relationship between the seven themes and the three demographic variables (i.e., age, ethnicity, and number of prior arrests). The seven themes were treated as the dependent set of variables, whereas the demographic variables were used as the independent multivariate profile. The canonical analysis revealed that the first canonical correlation (R^sub c1^ = .68) appeared to be large, contributing 46.7% (i.e., R^sub c1^^sup 2^) to the shared variance. The remaining two canonical roots were not statistically significant. Consequently, we interpreted only the first canonical correlation.

Table 4 presents data pertaining to the first canonical root. The table provides both standardized function coefficients and structure coefficients (Onwuegbuzie & Daniel, 2003). With a cutoff correlation of 0.3 (Lambert & Durand, 1975), the standardized canonical function coefficients revealed that self-control, violation of rights, provocation, and conflict resolution made important contributions to the set of themes-with violation of rights being the major contributor. With respect to the demographic set, age and ethnicity made noteworthy contributions.

The structure coefficients revealed that self-control, violation of rights, provocation, and poor judgment made important contributions (i.e., were practically significant) to the first canonical variate. The square of the structure coefficient indicated that these variables explained 31.8%, 44.1%, 27.2%, and 16.3% of the variance, respectively. With regard to the demographic cluster, ethnicity made the strongest contribution, with age making a moderate contribution. The square of the structure coefficient indicated that ethnicity and age explained 67.5% and 34.8% of the variance, respectively.

In the canonical function, conflict resolution appeared to serve as a suppressor variable because the standardized coefficients associated with this variable were moderate, whereas the corresponding structure coefficient for age was small. It is likely that conflict resolution was a suppressor variable because of its relationship with one or more of the other themes. In particular, as noted above, conflict resolution was statistically significantly related to self-control. Thus, conflict resolution improved the predictive power of the themes by suppressing variance that was irrelevant to this prediction as a result of its relationship with self-control.
Table
Enlarge 200%
Enlarge 400%
TABLE 4. Canonical Solution for First Function: Relationship Between Seven Themes and Selected Demographic Variables

Stage 6 Analyses

Finally, narrative profile formation qualitized the quantitized dichotomous variables that formed the seven themes (Tashakkori & Teddlie, 1998). Specifically, the number of average profiles (Tashakkori & Teddlie) was determined using an ipsative approach in which participants' responses to each theme were interpreted relative to their responses to the other themes (Block, 1957). We then formed an intra-individual correlation matrix by correlating each pair of profiles, yielding 3,321 (i.e., 82 ~{!A~} 81/2) Spearman Rho values. This correlation matrix was cluster-analyzed by the VARCLUS procedure of the Statistical Analysis System (SAS Institute, 1990) so that individualistic patterns could be characterized for each juvenile delinquent. We expected offenders having similar profiles to cluster together. The criterion of percentage variation explained by each cluster determined the most meaningful cluster solution.

To obtain the minimum cluster solution that explained the maximum variation, we applied the criterion of terminating the splitting of clusters when each cluster has only one eigenvalue greater than one. Also, cluster solutions that added less than 5% to the explained variation were eliminated from consideration (Onwuegbuzie, 2003). Thus, we selected a three-cluster solution, which explained 58.9% of the variation, as the most meaningful and parsimonious.

Figure 2 displays pictorially the profiles for the resulting three clusters. The seven themes are presented on the horizontal axis, whereas the proportion of students who provided an attribution reason belonging to each theme is presented on the vertical axis. Thus, each of the three emergent profiles represented an average set of responses across each theme. As can be seen, members of Cluster 1 (n = 35) were extremely unlikely to endorse the theme of self-control, p = .20, or the theme of conflict resolution, p = .20. These juveniles were moderately likely to endorse the theme of violation of rights, p = .43, and the theme of fate, p = .40. However, they were very likely to endorse the theme of provocation, p = .80, the theme of irresponsibility, p = .80, and the theme of poor judgment, p = .86.
Graph
Enlarge 200%
Enlarge 400%
FIGURE 2. Average profiles relating to juvenile delinquents' reasons for violence attributions.

Individuals in Cluster 2 (n = 23) highly rated the themes of self-control, p = .83, violation of rights, p = .70, provocation, p = .83, irresponsibility, p = .83, and poor judgment, p = .74. Additionally, they were moderately likely to endorse the conflict resolution theme, p = .57. However, they were unlikely to provide a reason associated with fate, p = .17.

Members of Cluster 3, like those of Cluster 2, highly rated the themes of self-control, p = 1.00, provocation, p = .70, irresponsibility, p = .90, and poor judgment, p = .95. Also, they were moderately likely to endorse the theme of fate, p = .63. However, this group was highly unlikely to endorse the theme of violation of rights, p = .10, or the theme of conflict resolution, p = .15.

Interestingly, although we found no difference between members of the three clusters with respect to number of prior arrests, F(2, 79) = 1.02, p > .05, or ethnicity, Fisher's Exact p = .52, ~{&V~}^sup 2^(2) = 1.45, p > .05, a difference among the three groups emerged with regard to age, F(2, 79) = 8.01, p < .001. The effect size associated with this latter difference was moderate (~{&X~}^sup 2^ = . 18, ~{&E~}^sup 2^ = .21). Scheff~{(&~}'s post hoc comparisons revealed that members of Cluster 1 (M = 14.86 years, SD = 1.35 years) tended to be younger than members of Cluster 2 (M = 15.82 years, SD = 1.11 years) and Cluster 3 (M = 16.05 years, SD = 0.94 years).

Discussion

In the present study, we investigated the causal attributions that male juvenile delinquents make for others' violent behavior and the salient pieces of information that they use in arriving at their attributions. We used a six-stage concurrent mixed-methodological analysis. The first stage revealed that the juvenile offenders committed attributional errors about violence nearly 53% of the time. This finding is consistent with the finding of Dodge and Coie (1987) that youth are more likely to externalize blame in what they perceive as provocative interpersonal interactions. Because it is likely that the aggressive, antisocial behavior of some juveniles may be due to their inaccurate or biased appraisals of interpersonal exchanges (Dodge & Coie), the present finding regarding the rate of errors of violence attribution is particularly informative, albeit disturbing. Moreover, the juvenile delinquents' tendency to commit such errors might explain, at least in part, their prison status. Future researchers should investigate further this possible link.

The greater likelihood of Black juvenile offenders' committing attributional errors about violence than that of their White counterparts may be the result of the different experiences of the two groups. Indeed, as noted by Dodge and Tomlin (1983), aggressive youngsters are more likely to rely on past experiences than their nonaggressive peers when making causal inferences about another's behavior. Interestingly, an analysis of the home location of the juveniles revealed that a greater proportion of the Black offenders than of the White offenders lived in the poorest neighborhoods. Additionally, in the current inquiry, the Black youth were significantly more likely to cite a reason for their violence attributions that related to provocation than were the White sample members. Thus, it is possible that level of racism (probably experienced by a large proportion of Black youth in the present sample to some degree) may generally lead Black and other minority adolescents to be more sensitive about being provoked and less confident about the importance of violations of rights than are their White peers (as seemed to be the case in the present study).

The positive relationship that we have noted in the present study between the number of prior arrests and the number of attributional errors about violence suggests that generally a cycle of attributional biases occurs among juvenile delinquents, whereby social cues culminate in attributions that drive feelings, which in turn serve as precursors to behavior, which in turn affect the onset and consequences of future social cues. According to this cue-attribution-emotion-behavior-attribution cycle, following negative social encounters juvenile delinquents tend to make attributional errors, leading to negative emotions and then to at-risk behaviors, the consequences of which affect future attributions. This model predicts that the more a juvenile is arrested, the more likely he is to believe that he is a victim of society, and that any antisocial behaviors in general and violent behaviors in particular that he performs are the result of provocation. Indeed, in the present investigation, delinquents with the greater number of prior arrests were significantly less likely than their counterparts to attribute a reason for their violence that related to the theme of violation of rights.

An additional purpose of the present study was to develop a typology of reasons for violence attributions and to determine whether these reasons predict juvenile delinquents' attributional errors about violence. The phenomenological analysis (Stage 2) revealed the following seven themes that arose from juveniles' reasons for their causal attributions: self-control, violation of rights, provocation, irresponsibility, poor judgment, fate, and conflict resolution. The first two themes were associated with the actor's disposition (i.e., person), the middle three themes pertained to the provocation by a target (i.e., stimulus), and the last two themes represented the exacerbating conditions (i.e., circumstance). This result suggests that offenders' reasons for their violence attributions represent a multidimensional construct. Moreover, the finding that the three stimulus themes-provocation, irresponsibility, and poor judgment-were the most endorsed themes, with more than three-fourths of the sample citing one or more reasons that fell into these categories, indicates that causal attributions of stimulus are most responsible for attributional errors about violence. Simply put, juvenile delinquents appear to blame the victim more often than they blame the perpetrator. Of the three stimulus reasons cited, the adolescents' perception that the victim should be blamed for being violated because of a poor judgment that he or she made (e.g., walking into a bad neighborhood) or because the victim had provoked the actor (e.g., laughing at the actor) appears to be the most pervasive. Indeed, the juveniles who endorsed the poor judgment and provocation themes tended to significantly make more attributional errors about violence than did the rest. At the same time, juveniles who tended to cite attribution reasons that related to both of the person themes (i.e., self-control and provocation) tended to make less attributional errors about violence.

The exploratory factor analysis revealed that the seven themes fell into the following four metathemes: the disposition of actor and interaction with stimulus (comprising self-control, violation of rights, and conflict resolution), the cognitively based stimulus (comprising irresponsibility and poor judgment), the emotionally based stimulus (comprising provocation), and the circumstance (comprising fate). Interestingly, the cognitively based stimulus was the most prevalent metatheme, providing a further explanation for the high incidence of attributional errors about violence.

The result that older juveniles were more likely to endorse the self-control theme-making a person attribution-than were the younger offenders suggests that violence attributions have a developmental context. In fact, it is likely that younger adolescents, who still use concrete thought as opposed to abstract thought, are unable to use moral principles in making social decisions (Kohlberg, 1969). Thus, it is possible that attributional errors about violence can be tied to developmentally based frameworks such as Kohlberg's stages of moral development (Kohlberg). Therefore, future researchers should explore further this possible link between attributional errors about violence and level of moral reasoning. To the extent that moral reasoning is a determinant of violence attributions, increasing adolescents' beliefs in rules and the law as early as possible may be beneficial. This could have implications for school curricula.

A final goal of the current study was to identify profiles of juvenile delinquents on the basis of their violence attribution reasons. Through ipsative or cluster analyses, three profiles emerged (cf. Figure 2). These profiles provide compelling evidence that although juveniles are relatively homogenous with respect to the number of attributional errors about violence made, they differ with respect to the reasons they provide for their causal attributions.

The current sample, although random, represented juvenile delinquents from a geographically restricted region. Thus, the extent to which these findings generalize to juvenile offenders from other geographic regions is not clear, suggesting a need for replication in more diverse samples. Nevertheless, the present findings make an important contribution to the juvenile delinquency literature by simultaneously quantifying and qualifying violence attributions. Use of a concurrent mixed-methodological data analysis technique not only allowed an estimation of the prevalence of attributional errors about violence but also facilitated a typology of the salient pieces of information that the juvenile delinquents use in arriving at their attributions. Moreover, this pragmatist paradigmatic framework, as posited by Onwuegbuzie (2003), led to the determination of the structural relationships among the reason categories. Figure 1 shows these relationships. The figure not only identifies which reason categories (i.e., themes) fall under the same umbrella (i.e., a metatheme) but also reveals which metathemes are associated with the tendency to make attributional errors about violence and which metathemes are associated with the tendency to not make them.

The mixed-methodological data analysis also facilitated (a) the computation of effect sizes associated with attributional errors about violence and reasons and (b) the identification of antecedent correlates of the juveniles' responses. Therefore, future researchers in this area should continue using this pragmatist paradigmatic approach (Onwuegbuzie & Teddlie, 2003). Such an approach will help to improve researchers' and practitioners' understanding of factors that place adolescents at risk for violent behavior, hopefully promoting the identification of effective treatment programs for this vulnerable population.