Extending a Social Control Framework to Explain the Link between Romantic Relationships and Violent Victimization by Non-Intimate Perpetrators: A Study of Actor and Partner Effects*

Abstract

Victim data reveals that romantic relationships correspond to significant reductions in violent victimizations committed by strangers and acquaintances. This study offers a more detailed exploration of this finding. Specifically, we investigate the effect of relationship quality and structure on victimization risk in combination with mechanisms a social control perspective would suggest as theoretically possible. We also consider the independent contributions of both romantic partners toward this apparent protective effect. We use two waves of adult data from the National Longitudinal Study of Adolescent Health, which include data from both respondent and their partner and adjust for treatment selection a series of models. The protective benefits of relationship quality for the individual respondent appeared to have: (1) functioned irrespective of their partner’s perception of relationship quality, (2) were negated or related to even greater victimization risk if the relationship quality was poor, and (3) were explained by reductions in lifestyle choices associated with risk of violent victimization.

Keywords:

• Victimization
• violence
• dyads
• romantic relationships

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

2 We recognize that one perspective in criminological theory is that theories of crime, like social control theory, are intended to explain crime only, with no clear implications for victimization risk (Schreck et al., 2008). Nevertheless, the assumptions behind control theories—a rational individual pursuing a self-interested course of action—draws explicitly from a general theory of human behavior and not a theory narrowly designed to explain crime only (Gottfredson, 2011). Variations of control theories have been applied successfully to victimization for the past quarter century, and Gottfredson (2021) has recently expressed the view that using the control perspective in this manner is fully consistent with theoretical logic.

3 In other instances, such as the case of a person simply being in the wrong place at the wrong time, victimization would promote concern for the targeted person (Schreck & Berg, 2021).

4 We are aware that relationship effects may also reflect selection, as partners tend to choose those with similar values, personalities, and backgrounds. For our purposes, we would expect that those who are at low risk of victimization would tend to choose partners who are also low risk. We mitigate this by controlling for each partner’s demographic backgrounds and levels of self-control (Schreck, 1999); however, we acknowledge that there are many elements driving the selection of romantic partners (Giordano et al., 2002; Sampson et al., 2006).

5 Add Health provides survey weights to account for sampling and research design effects. We conducted analyses with and without the survey weights (not presented) using the svyset and associated commands in Stata. Although the coefficients were of comparable magnitude, the standard errors were, unsurprisingly, much larger in the analyses where we applied the weights. Following Winship and Radbill (1994), the analyses presented here are based on the unweighted sample and we adjust for variables that were used to create the weights.

6 We exclude the item that asked respondents whether they had seen or witnessed someone shoot or stab someone else because our focus is on direct experiences with victimization. Although our victimization measures do not explicitly exclude victimizations at the hands of a romantic partner, we have valid reasons to assume that non-intimates committed most of these victimizations. Data from the NCVS suggests that strangers and acquaintances perpetrate most violent victimizations (National Institute of Justice, 2002). In addition, our estimate of the protective effect of romantic relationships will be conservative to the extent that respondents were victimized by their romantic partners. If respondents are victimized by a romantic partner, it will increase their victimization risk to a greater extent than respondents not in a romantic relationship, as those not in a relationship do not have the opportunity to be victimized by a romantic partner. It is not clear what direction this ambiguity could bias our estimate of the effect that relationship quality has on victimization.

7 Three of the items ranged from 0 to 4 and one of the items ranged from 0 to 3. We standardized these items before creating the scale using the std option in Stata’s alpha command.

8 We describe these variables as mediating variables based on our theoretical argument. We recognize, however, that a proper mediation analysis requires establishing temporal ordering between the variables. We describe the implications of this decision further below.

9 In other analyses, we controlled for respondents’ level of self-control at Wave 1, measured with seven different items that characterized how respondents interacted with their teachers and peers during school, their ability to pay attention during school, and their impulsivity. We removed the variable at the suggestion of the reviewers, because the scale was a rough approximation of self-control. The analyses produced results that are similar to those presented here. We do not use self-control items asked at Wave 3, as they may reflect changes in self-regulation resulting from respondents’ romantic relationships and not solely the respondent’s own level of self-control. It is, therefore, not clear whether the item would be acting as a mediator, confounder, or both.

10 We conducted a supplemental analysis (not presented) in which we examined the association between relationship quality and victimization among respondents in a romantic relationship. We found that relationship quality was negatively related to victimization, consistent with results presented here.

11 The distribution of the standardized score was negatively skewed (skewness = -2.19; kurtosis = 8.36).

12 When we conducted these correlations, we removed respondents not currently in a relationship or who did not have a partner participate in the survey.

13 Like any other method based on propensity scores, IPTW requires that we assume there are no unmeasured covariates that affect treatment status and treatment outcomes (Austin & Stuart, 2015). The effectiveness of IPTW for resolving bias depends on the extent to which this assumption is valid.

14 In accordance with the literature, we use robust standard errors clustered around a respondent identification variable.

15 The coefficient for the relationship status variable should, at least in theory, be comparable to what we found using the respondent’s perceptions. The status of their relationship was not a product of the partner’s perceptions. We found some variation which may be noise. It may also be the case that the partners who chose to participate in the survey were more committed to/satisfied with the respondent than those who refused to participate.

Additional information

Notes on contributors

Christopher J. Schreck

Christopher J. Schreck is professor and chair of criminal justice at the Rochester Institute of Technology. His scholarship focuses on crime target decision-making, precautionary behavior, and bridging victimology with criminological theory. He is the 2021 recipient of the Academy of Criminal Justice Sciences’ Bruce Smith Sr. award.

Andrew Krajewski

Andrew Krajewski is an assistant professor in the Criminology and Criminal Justice Program at the University of Texas at Dallas. His research interests include aggressive behavior, interpersonal conflict, and the underlying decision-making processes.

Mark T. Berg

Mark T. Berg is a professor and collegiate scholar at the University of Iowa in the Department of Sociology and Criminology.