Criminal Justice Outcomes in Intimate and Non-intimate Partner Homicide Cases

• Previous Page
• Table of Contents
• Next Page

4. Results (cont'd)

4.2 Multivariate analysis: Isolating the effects of intimacy in criminal law

Because the objective here is to compare the treatment of those who killed intimate partners to the treatment received by those who killed victims with whom they shared more distant relationships, a reduced sample is used in this part of the analysis for two reasons. First, intimate partner homicides rarely involve more than one accused and one victim (Silverman & Kennedy, 1987, 1993). For example, in the Toronto data, 98 percent of those accused of killing intimate partners acted alone whereas 82 percent of the 907 accused of killing other types of victims acted alone. Therefore, while the majority of homicides involve lone offenders, this is almost exclusively the situation in intimate partner homicides. With respect to the number of victims, the proportion of intimate partner and non-intimate partner homicides that involved the killing of more than one victim is small, but similar (4% and 5% respectively). Thus, in order to compare similar types of killings, the reduced sample is restricted to cases that involved one accused and one victim. Second, by virtue of its definition, the category ‘intimate partner’ is comprised primarily of adults. In the Toronto sample, the youngest accused person who killed an intimate partner was 19 years of age while the youngest victim killed by an intimate partner was 17 years of age. Again, to ensure that similar types of cases are compared, the reduced sample is restricted to those victims who were greater than or equal to these ages. The final reduced sample used in the multivariate analysis, then, was 914 cases.

The control variables used in the multivariate analyses are similar to those examined in the bivariate analysis. Of particular interest, however, are the variables that significantly distinguished between intimate partner and non-intimate partner homicide at the bivariate level. For example, there were significant differences between the two types of homicide based on the accused person’s prior record, gender, age, employment status, marital status and psychiatric history. In addition, the victim’s gender, age, prior criminal record, employment and marital status as well as location and method of killing also distinguished between the two types of homicide. It is important, then, to control for these factors in the multivariate analysis to allow for an examination of the independent effects of intimacy on court outcomes.^[30]

Results are presented in Table 4.8 for a series of multivariate models. Models 1 through 7 show the results of the logistic regression analysis that was used to examine the odds of being in the focal group compared to the reference group based on particular characteristics (only the odds are shown here^[31]; see Appendix D for the full results for each model). The final column – Model 8 – shows the multiple regression coefficients for length of sentence. Findings for each stage of the process for the key independent variables – victim-accused relationship, the year the case entered the court, and gender – are discussed first.^[32]

With respect to the entry point of the criminal process, Model 1 demonstrates that those accused of killing intimate partners were less likely to be charged with first-degree murder than those accused of killing victims with whom they shared more distant relationships. Model 2 shows that accused persons in intimate partner homicides were less likely to have their cases resolved at trial than those who killed other types of victims. In other words, cases of intimate partner homicide were more likely to be disposed of by guilty plea than cases of non-intimate partner homicide. However, among those cases that were resolved at trial, Model 3 demonstrates that accused persons who killed intimate partners were significantly more likely to be convicted at this stage than those who killed non-intimate partners (see Box 4). Among those acquitted (Model 4), however, there were no differences in the type of acquittal – not guilty versus not criminally responsible by reason of mental disorder.^[33]

---

• ^[30] To retain the maximum sample size and control for any biases associated with missing values on any of the independent variables, each model includes missing dummy variables for those variables missing information in more than 10 percent of the cases. The missing dummy variables are coded 0 if data were present and 1 if data were missing. None of the regression coefficients for the missing value dummy variables were significant, indicating that the missing information should not bias the results. However, information on two of the independent variables - victim's psychiatric history and substance use by the accused - was missing in approximately 50 percent of the cases and, therefore, these variables were not included in the multivariate analysis. Appendix C lists all the independent and control variables with the number and percent of cases that are missing information on each variable.
• ^[31] If the odds were greater than 1 and significant (as indicated by at least one asterisk), an accused was more likely to fall into the focal category (e.g. more likely to be charged with first-degree murder than a less serious charge). If the odds were less than one and significant, an accused was less likely to fall into the focal category (e.g. less likely to be charged with first-degree murder than a less serious charge).
• ^[32] Because earlier decisions are expected to affect later outcomes, unmeasured variables that affect one stage of the process may be correlated with unmeasured variables that affect a later outcome, producing a correlation between error terms on the two dependent variables. As a result, without including information on how cases are screened into each stage of the process, estimates of the effect of variables included in the next stage of the process may be biased (Berk 1983; Berk & Ray 1982; Klepper et al., 1983). To correct for this, a two-equation estimation procedure was followed to control for incidental sample selection bias in the coefficients of the variables affecting various decision points (Berk, 1983; Berk & Ray, 1982; Heckman, 1976, 1979; Klepper et al., 1983). Because results from the corrected models did not differ significantly from the uncorrected models, the latter findings are used because interpretation and presentation of results is simpler.
• ^[33] This underscores the importance of multivariate analyses that allow one to consider simultaneously other relevant factors when looking at criminal justice decision-making because the bivariate analysis showed that those who killed intimate partners were significantly more likely to be found 'not criminally responsible by reason of mental disorder' when the effects of other factors were not taken into account. For a more detailed analysis of determinants of the 'not criminally responsible by reason of mental disorder' verdict, see Dawson (2003b).

• Previous Page
• Table of Contents
• Next Page