Abstract
Despite widespread distrust of police among Black Americans, research indicates that Black crime victims are more likely to call the police than other demographic groups. Using National Crime Victimization Survey data from 1994–2018, I employ logistic regression and compare the average marginal effects of victim race and gender on probabilities of reporting to police. Of four race-sex categories (non-Black and Black male and female victims), I find that Black women are most likely to report and Black men are least likely. But when the offender is a romantic partner, family member, or other household member, Black victims of either gender are more likely to report. Evidence points to the importance of victim gender and victim-offender relationship in understanding racial differences in crime reporting.
Disclosure statement
No potential conflict of interest was reported by the author.
Notes
1 The percentage is certainly higher, as cases deemed by police officers to merit investigation as potential child abuse were not included in the dataset and analysis.
2 It is not clear how to interpret this response. Does reporting to another official rather than the police signify better access to or knowledge of other officials and services, or an active avoidance of police? Did the victim fail to report to police because the other official contacted the police on their behalf? Furthermore, the concept of the “most important reason” for not reporting to police is difficult to interpret. The other response options in the survey (private or personal matter, minor crime, police couldn’t do anything, police wouldn’t think it was important enough, police would be inefficient or ineffective, police would be biased, victim did not want to get offender in trouble with the law, fear of reprisal, too inconvenient, or other reason) could also be applicable. For example, if someone reported to another official because they were fearful of and wanted to avoid the police, would they be more likely to select “reported to another official,” “police would be biased,” or “other reason”?
3 Nearly all published papers using the NCVS conduct complete case analyses—see Baumer, Citation2002; Slocum, Citation2018; Zaykowski et al., Citation2019. I use Stata’s subpopulation option, rather than deleting cases.
4 The NCVS asks about Hispanic ethnicity as a separate question. 137 victims who reported being both Black and Hispanic were coded as Black, in order to provide as expansive a definition of Black as possible. An alternative coding scheme, in which these victims were coded as Hispanic and all victims coded as Black were non-Hispanic, was also conducted. Findings were unchanged.
5 Until very recently, the NCVS only collected data on victims’ self-reported sex, so I use “sex” rather than “gender” in my analyses and when presenting results.
6 Based on responses by survey participants to a series of screening questions, NCVS survey administrators classify incidents with a “type of crime code.” I used these codes to sort the crime types into the six categories. “Rape” includes completed and attempted rape. “Sexual assault” includes sexual attack with serious assault, sexual attack with minor assault, sexual assault without injury, and unwanted sexual contact without force. “Robbery” includes completed robbery with injury from serious or minor assault, completed robbery without injury from minor assault, attempted robbery with injury from serious or minor assault, and attempted robbery without injury. “Aggravated assault” includes completed aggravated assault with injury, attempted aggravated assault with weapon, and threatened assault with weapon. “Simple assault” includes simple assault completed with injury and assault without weapon without injury. “Verbal threat of violent injury” includes verbal threat of rape, verbal threat of sexual assault, and verbal threat of assault.
7 Roommates were grouped with family members based on the notion that non-related household members sometimes function like family. Sensitivity analyses were conducted in which roommates were instead grouped with friends and causal acquaintances; the results were substantively unchanged.
8 I use the command “margins, dydx(*)” to compute the average effect on the conditional mean of y of a hypothetical change in race and/or sex for the observed cases in the dataset (Cameron & Trivedi, Citation2010). Put another way, the AME shows the average difference in the adjusted predicted probabilities for two race-sex groups (Williams, Citation2012). In the presented results, non-Black male victims serve as the baseline category and the average marginal effects of a victim being a Black male, non-Black female, and Black female are calculated.
9 I use victimization weights adjusted for series crimes. “Series victimizations are similar in type but occur with such frequency that a victim is unable to recall each individual event or describe each event in detail (for example, intimate partner violence or domestic violence may be a series victimization). Survey procedures allow NCVS interviewers to identify and classify these similar victimizations/incidents as series and to collect detailed information on only the most recent incident/victimization in the series. In 2018, about 1.5% of all victimizations were series incidents. Weighting series incidents/victimizations as the number of incidents up to a maximum of 10 produces more reliable estimates of crime levels, while the cap at 10 minimizes the effect of extreme outliers on the rates.” (NCVS Codebook)
10 The pseudo-stratum (V2117) variable approximates the stratification in the NCVS, and the half-sample (V2118) code approximates the primary sampling units (PSUs). “The pseudo-stratum and half-sample codes specify the NCVS design within a single year. However, NCVS samples are not independent across years, so an additional stratification variable is needed to correctly specify the overlap in the NCVS yearly samples” (Shook-Sa et al., Citation2015). To account for this sample overlap, I created a “year group” stratification variable based on when new samples were phased in. The year groups are 1994-1996, 1997-2005, 2006-2015, and 2016-2018.
11 I conducted sensitivity analyses in which the sample was restricted to only Black and white victims and offenders, and the coefficients for offender race remained non-significant.
Additional information
Notes on contributors
Rachel D. Butler
Rachel D. Butler holds a PhD in sociology from Johns Hopkins University and an MSEd from the University of Pennsylvania. She studies race, poverty, family life, and criminal justice system contact.