Specifying the Strain-Violence Link: The Role of Emotions in Women's Descriptions of Violent Incidents

Abstract

Research on General Strain Theory has not considered fully the range of emotions that might be linked to violent behavior. Moreover, quantitative analyses of strain and emotions have assessed emotional traits rather than emotional states that directly precede violence. We use narrative data from incarcerated women to examine how they describe the strains and resulting emotions that emerged during incidents in which they used violence. Our findings confirm the importance of anger, but also suggest the presence of other related emotions during violent interactions. We argue that the study of General Strain Theory can be advanced by considering situational factors.

Keywords:

• general strain theory
• emotions
• situational violence

Acknowledgments

This research was supported by a grant from the National Consortium on Violence Research (NCOVR) to professors Julie Horney, Sally Simpson, Rosemary Gartner, and Candace Kruttschnitt. The authors would like to thank the Women's Experiences of Violence team for allowing us access to the data. In addition, we would like to thank Bob Bursik for his role in motivating this research and Sally Simpson, Richard Wright, and two anonymous reviewers for their useful comments.

Notes

1. Emotions are generally considered to have two forms—state and trait. State emotions are generally of limited duration and experienced in the face of specific demands or dangers. In contrast, trait emotions reflect the existence of stable, individual differences in the likelihood of responding to stimuli with particular emotions.

2. This is not to suggest that traits are irrelevant to GST. Trait emotions may affect offending by increasing the likelihood that individuals will experience strain (Mazerolle et al., 2000) or by increasing the likelihood of experiencing state anger (Capowich, Mazerolle, & Piquero, 2001).

3. Behaviors, or violent actions, that could be interpreted as suggesting emotion were not coded as emotions because our goal was to clearly differentiate between emotional states as described by the respondents themselves and presumed emotions based on one's behavior. For example, in a narrative in which a woman described using drugs while pregnant, we may think she should feel shame—yet in her description of the incident she describes only anger; therefore, only anger was coded. While we agree this coding decision is conservative and may result in the omission of some emotions, we wanted to avoid imposing our views about what we thought the women should be feeling.

4. The omission of chronic stressors is a limitation of the current study given that Agnew (1992, 2001) argues that chronic strains are especially likely to lead to criminal behavior.

5. Comparisons were made between the emotion and nonemotion narratives using chi-square tests of independence (alpha = .05). We recognize that because some individuals contribute multiple narratives to the dataset, the p values might be inaccurate. Therefore, we also tested for differences across narrative types using a series of logistic regressions with robust standard errors. Here the dependent variable is a dichotomous variable scored 1 if the narrative contained an emotion and 0 if not. The independent variables are the characteristics of the situation recoded as dummy variables. Our substantive conclusions did not change using this alternative method of comparison.

6. For a discussion of centrality measures in SNA see Borgatti and Everett (1997) and Freeman (1979). Concept-specific measures of centrality for this two-mode network are available upon request.

7. To generate this matrix of co-occurring concepts, we multiplied the affiliation matrix by its transpose, resulting in a new matrix whose ij ^th cell indicates the number of narratives in which both strain i and emotion j were present.

8. The algorithm used to generate this visualization is based on multidimensional scaling and the calculation of geodesic distances between concepts.

9. We estimate a core-periphery model to further examine the centralization of the overall data structure (Everett & Borgatti, 2005) because the results presented above indicate that the observed network is sparse and dependent on a few strains and emotions. This core-periphery model assesses the degree to which the overall structure of links between strains and emotions is dependent on a few core concepts weakly connected to a set of peripheral concepts. Guided by prior research (Borgatti & Everett, 1999; Everett & Borgatti, 2005) we estimate the degree of coreness of each node (concept) by fitting a continuous model of core/periphery to the co-occurrence matrix. Coreness refers to the degree of closeness of each node to a core of densely connected nodes in the observable network. We begin with a continuous model because we are uncertain about the two-class partition between core-periphery obtained from a categorical algorithm. Coreness estimates for each strain and emotion are available upon request.

10. To examine whether the nesting of narratives (191) in individual respondents (114) influenced the patterning of relationships identified in our analyses, we repeated our analyses using one randomly selected narrative for individuals with multiple narratives. Results were not substantively different than those presented here, although some of the less frequent emotions and strains tended to be dropped from the sample (e.g., disputes over tasks, predatory incidents, feelings of stress).

11. The most common strain was dispute over sexual jealousy, yet few women explicitly described being jealous. This discrepancy has several possible explanations. First, many of the disputes over sexual jealousy involved situations in which the partner was jealous of the respondent's behavior, which caused the respondent to feel angry or frustrated. Second, in incidents explosive enough to lead to violence, respondents may have felt angry, as opposed to jealous, upon learning of their partner's infidelity.