Research Article Review and Data Analysis - Domestic Assault

In "The specific deterrent effects of arrest for domestic assault", Sherman and Berk (1984) identified an inconsistency in sociological theories of deviance on the effects of punishment on domestic assault offenders. Under the specific deterrence doctrine, it would be expected that individuals previously punished for domestic assault would avoid re-offending, in order to avoid the punishment they have experienced once before. However, under labelling theory, individuals may actually be more likely to re-offend after their first offense due to "altered interactional structures, foreclosed legal opportunities, and secondary deviance (Sherman & Berk, 1984, p. 261)." However, empirical support for both of these theories was weak and inconclusive. The researchers used a field experiment that varied police response during domestic assault calls, using three different approaches to the offense, including informal mediation, barring the offender from the premises for a short period of time (eight hours), and immediate arrest. The researchers then followed suspects through the criminal justice system for some period of time, in order to determine whether each offender had re-offended. The research found that individuals that were arrested had lower rates of recidivism than those that were barred from the property, which the authors used as proof for a hypothesis supporting specific deterrence and rejecting labelling theory. However, rather than simply accepting this research uncritically, there are a number of questions that should be asked in order to either satisfy ourselves of the validity of its conclusions or to identify weaknesses in the design that may weaken these conclusions. Discussion of sample identification and selection, data collection and analysis techniques, internal and external validity, and potential for competing explanations to emerge from the results of the study can provide a better level of insight into the soundness of the findings of the study, as well as the conclusions that the researchers have drawn from these conclusions.

Hypothesis

The authors of the study did not state an explicit hypothesis in formal null and alternative hypothesis form. However, the operational assumption contained within the paper was that a lower rate of re-arrest of suspects on domestic assault cases that were initially arrested would support the specific deterrence doctrine. Conversely, increase re-arrest among suspects initially arrested would be supportive of labelling theory. However, as the researchers themselves noted, labelling theory is not a clearly defined or discrete theoretical base, and so lack of support within the analysis would not necessarily disprove labelling theory. The lack of a formal specification of hypothesis did make it difficult to determine what the initial assumptions of the paper were, and given that the researchers used a hypothesis-proving research approach it would (Babbie, 2009).

Sample and Population

The population chosen for this research was not specified explicitly. However, the researcher's discussion of the conditions of law enforcement in domestic assault cases makes it clear that the population is suspects involved in domestic assault cases. The sample of the population included 314 cases of individuals suspected of simple (misdemeanour) domestic assault during the time period of the study. Analysis was at the individual level.

Research Design

The research design used a simple randomized experiment method in order to determine differential effects on the outcomes. There were three groups in the study, including one group where the suspect was arrested, one where informal mediation or advice was offered, and one in which the suspect was required to leave the premises for eight hours. This division was made possible by a change in Minnesota law that allowed police officers some freedom in deciding what path to take under the conditions the case study examined. The use of a control group was not specified, and would actually have been unethical considering the conditions under which the experiment was being conducted (the police simply doing nothing when responding to a domestic assault call would place victims in harm's way unnecessarily, violating the principle of non-malificence). However, the researchers did specify that cases where arrest was clearly required were excluded from the study and officers took the course of action appropriate at that time. This indicates an appropriate focus on ethical research, even though the lack of a control group may weaken the analysis somewhat.

The research design called for 34 police officers to act as field experimenters in this study. The officers carried randomly arranged colour-coded report forms, in which each colour represented a different action - unless there were clear indications for exclusion, the officers chose the action identified by the colour t the top of the pad. This was accompanied by a brief post-incident follow-up form filled out by the police officer, as well as a post-incident interview with the victim. (The researchers were not highly successful in gaining permission for these interviews and only 205 of 314 incidents were accompanied by a victim interview). The researchers then tracked the suspects through the criminal justice system for six months in order to identify re-offenders. Overall, this seems like a robust research design that would provide insight into the areas of interest. However, what was not clear was whether the researchers tracked only evidence of recidivism in domestic crimes, or whether this included all evidence of continued involvement in the criminal justice system.

Data Collection and Analysis

The researchers offered a critique of their own research design and the data collection process within the study. They identified areas of change from the research design, including the exclusion of further situations from the study following review. The authors indicated that this may have caused differential attrition from the experiment. The authors also identified that there was a high rate of attrition among police officers involved in the experiment, which required recruitment of additional officers in order to complete the research on time. The main concept that was used was recidivism, or re-offense by a previous criminal offender. This was operationalized simply by using court records to identify repeat offenders. However, for some respondents (N=205), the researchers also used follow-up interview data with the victim of the assault, which identified whether the behaviour that triggered the initial arrest had been repeated. This provided an aggregate concept of re-offense, which included both unreported offenses (included only for victims that had agreed to be interviewed) and reported offenses that triggered a written citation, warning, or re-arrest. The theoretical constructs that the research was based in directly addressed the concept as operationalized by the researcher. The instruments used in the study were simple and the majority of the information (other than interviews and incident reports) were generated by the court system, reducing the difficulty involved in identifying construct validity of a given concept in the paper. The use of official court and police records lends validity to the research because these records can be expected to be more reliable than self-reports of repeat offenses by victims or offenders themselves. However, the inclusion of self-reports by victims was important to increase the effectiveness of analysis for a very sad reason - other, more recent research suggests that while 25% of women experience domestic violence in their lifetime, only 2.5% to 15.5% report that they are in a domestic violence situation (Garcia, 2004). Thus, relying on only formal court reporting would have significantly reduced the reliability of the study due to this level of underreporting. Of course, the use of follow-up victim interviews is also not perfect, as victims may not be forthcoming with researchers, and many refused to be interviewed or could not be contacted, but it is an improvement over uncritical reliance on formal statistics.

Internal Validity and Construct Validity

There was the potential for internal validity within the research given the reliance on a small number of officers being involved in a large number of the cases. The authors indicated that three of the authors produced around 28% of the cases involved in the study. Although they indicated that this did not pose an internal validity problem due to the randomization of the study forms on the pads, this is not necessarily the case. The officers had some discretion in inclusion, and there were certain conditions under which a case must be excluded from the study (including severe physical assault as well as a few other conditions). Although the police officers could not affect the study outcomes if they chose to include a given assault case in the study, it is possible that they could influence internal validity through not choosing to include specific cases that may have been marginal (or even falsifying conditions to include or not include specific cases). Thus, the assumption of no internal validity problems caused is required for the integrity of the study, but it may not hold up to more in-depth analysis. The authors did acknowledge that it was possible for officers to influence internal validity in other ways, such as switching pads or manipulating calls based on prior information. The researchers do also acknowledge the possibility (or likelihood) that the experimenters failed to follow the experimental protocol at times, which would have reduced the internal validity of the experiment due to experimenter error that was not detected. (This was a possibility because officers were largely unsupervised during the experiment). Thus, although the researchers provided ample evidence of potential for internal validity issues, there is no evidence that researcher bias or unintentional error by experimenters did take place. However, neither was this possibility accounted for in the analysis of the research outcomes. This indicates a potential problem accepting the outcomes of the research.

The main problem with construct validity is the reliance on only a small number of officers for a large number of arrests. If one or more of the three officers that performed 28% of the case collection had the incorrect idea regarding the underlying constructs and their relationship to the operationalized study model, that could have affected the outcome of the study. While the researchers do not give any indication that this was the case, the lack of supervision of the officers overall does leave open the possibility of issues in this area.

Analysis and Findings

The analysis first used an analysis of variance (ANOVA) technique to determine if there was a difference between means of the three random study groups. The initial analysis did indicate some suggestion of deterrent effects of arrest, but as the authors noted, there was some "upgrading" of more serious incidents when the officers felt arrest was more appropriate than the suggested action, which could artificially exaggerate these results. The authors then adjusted variables to take these issues into account and used linear and logit regression models to determine relationships. They then used a proportional hazard analysis to adjust the outcomes for right-hand censoring (that is, inappropriately optimistic models due to the six-month observational limit). In one analysis the researchers did include the self-report data derived from interviews with victims, and indicated that this changed the outcomes of the analysis and rankings.

The overall results indicated that 28.9% of the suspects in the original survey had a second report, citation, or arrest following a new assault during the six-month period during which they were followed. However, the group that was arrested initially showed lower rates of recidivism than interception and separation or advice showed. The researchers attested this outcome to the proving of the specific deterrence doctrine, rather than to other potential influences that could have changed the outcome of the responses. They also held that it falsified the hypothesis of labelling theory being responsible (which would have resulted in an increase in re-arrests, rather than a decrease).

The researchers did consider some of the responses that may have influenced the outcomes. For example, they considered that arrest may reduce the rate of re-offense through taking offenders out of the situation. However, when researchers examined this, they found that only 14% of the suspects were held for longer than a week, indicating that this response was probably insignificant. However, this does not indicate an uncritical acceptance of the specific deterrence doctrine as a causative factor in recidivism. For example, it is possible that an arrest may actually change the victim's view of the situation, allowing her the opportunity to escape the relationship (thus eliminating the chance for re-offense, at least assuming the offender stayed single for six months). It may open up family and couples therapy options that were not available, or provide individual therapy or treatment to the offender, or trigger a substance abuse treatment program recommendation, all of which could affect the outcomes of recidivism. However, these potential changes were not explored in the research, and so there is little evidence actually within the structure that could apply to them.

External Validity

This research was relatively small scale and isolated in geographic coverage (including only a few neighbourhoods of Minneapolis). Additionally, it was conducted 25 years ago, and policing practices and the social meaning of incarceration has changed. Thus, the results of this study have not stayed as applicable over time as they were initially. However, the generally robust design of the study and the analysis do indicate that externally, this research would have been valid across the population it was intended to serve for some period of time.

Conclusion

This study is not perfect and does have some flaws, particularly in the area of unobserved experimenters and their potential effects on internal validity of the study. However, overall the research design was robust and the construct validity was strong. The findings indicate a relationship between arrest and lower recidivism rates, and although the author's theoretical explanation does not immediately follow on form this discussion, it is suggestive of support for this theory. The results of the study overall do indicate that it is generally reliable and valid research.