Exceptional Cases: Examining Intimate Partner Sexual Assault Cases Cleared by Exceptional Means

1. In line with previous research (O’Neal and Spohn Citation2017), this study defines IPSA as a ‘sexual assault incident involving a suspect and victim who are married, cohabitating, dating, legally separated or divorced, or who have children together’ (p. 8).

2. The victim will be referred to in the feminine and the offender in the masculine. Less than 2% of the cases in the original data set used for this research involve male victims. As a result, no male victims were included in analysis in this study.

3. The original data set appeared to feature a ‘reason for clearance’ variable, distinguishes clearances by exceptional means based on victim refusal to cooperate, prosecutor refusal to prosecute and Other. When inspecting and cleaning data, it was determined that this variable did not accurately correspond to case outcomes. For instance, some cases that were open or unfounded had a reason for clearance listed. The variable, therefore, was removed from analysis.

4. Some new variables may have been created by this researcher from formulas and recoding of existing variables. No data related directly to the reporting, response, or outcome of sexual assault cases were entered by this researcher.

5. Since the original data set included an ‘alcohol’ variable, missing entries are considered to be representative of an absence of alcohol consumption. In other words, if a law enforcement officer leaves that entry blank it is interpreted as ‘nothing’, or no alcohol consumption.

6. This original variable was identified as ‘relationship’ but was not clear specifically which person (i.e. suspect or victim) was being referenced. It should not impact this variable, however, as the groupings include both relationship titles of the victim and suspect (e.g. ‘husband’ and ‘wife’; ‘boyfriend’ and ‘girlfriend’.)

7. Because of several cases in which reporting occurred many months after the date of the incident, the natural log was taken of the hours between the time of incident and the time it was reported to police to help normalize the distribution of this variable for multivariate analyses.

8. The jurisdiction being studied did not make publicly available its annual reports for the years 2011 through 2014. In fact, the 2017 annual report was the next the be disseminated publicly. Therefore, that information could not be included with this variable for cases that occurred during those years. This resulted in missing data (missing at random) in approximately 23% of cases involved in the current analysis. This is a substantial number of missing cases; further, the fact that the missing cases where time-based (i.e. missing data as a result of a lack of information for years 2011 to 2014) affected the longitudinal nature of the data set. If those cases were deleted, the data set would essentially be truncated at 2010. Mean substitution was considered but ultimately would also negate the longitudinal nature of the data as it would introduce a fixed number for a dynamic, continuous variable. As such, imputation was completed for missing values using the linear trend at point method, which is designed for use with time series data. This method is essentially a regression, with missing values replaced with their predicted values.

9. For example, according to the 2010 census data, district 4 had the following racial populations: 35,381 Hispanic residents, 9,925 White residents, 76,399 Black residents and 1,870 residents who could be grouped into the Other category. Instead of including several variables with all populations for each district, the district was identified by the race with which a majority of the residents were classified. Thus, district four would be considered a Black majority district for cases in the years 2010, 2011, 2012, 2013, and 2014.

10. The Hosmer and Lemeshow test showed significance, χ² (df = 8) = p =.025. The Hosmer and Lemeshow test was most likely affected by imputation of missing data in the total yearly reported sexual assault by district variable. Further, the Hosmer and Lemeshow goodness of fit test is known to be less effective with an n of greater than 1,000 (see Bertolini et al. Citation2000; Kramer and Zimmerman Citation2007). As such, it was determined that a significant Hosmer and Lemeshow findings did not necessarily mean the predictive model was not useful.

11. Prior to calculating for outliers, the mean reporting time in hours was 733.4, or 30.5 days. There may have been issues with entering original data or they simply may have been skewed by victims who reported cases much after they occurred. To wit, the maximum value was 87,653 hours, or approximately 10 years.