15
The Factors That Determine Sexual Assault Reporting as They Compare to Non-Sexual Assaults What factors cause a sexual assault to be reported to the police? And further, are these factors unique to sexual assault, or are they similar to other forms of assaultive violence? This research an- swers these questions by employing data collected by the National Crime Victimization Survey through each of the years from 2000 to 2013. By developing a probit regression model, this analysis predicts the police-reporting behavior of female victims of both sexual assaults and assaults of a non-sexual nature. While the results of our analysis agree with previous literature that the presence of a weapon and the incidence of an injury have significant explanatory power in predicting whether a sexual assault is reported, we propose that these factors are not unique. Rather, these factors ex- plain all types of assault, and not sexual assault in exclusivity. The only unique explanatory factors found in our research were whether the victim of the assault was employed at the time of the inci- dent, and the number of years she had been educated. The distinction between factors that influence all types of assault and those that only influence sexual assault is important as it may provide in- sight into the unique procedures necessary to encourage sexual assault victims to report the crimes committed against them to the police. Introduction ore than half of the nation’s vio- lent crimes, or nearly 3.4 million violent victimizations per year, went unreported to the police between 2006 and 2010 (Langton et. al 2012). Of these violent crimes, sexual victimizations were found to be the most gravely underreported, with an estimated 211,200 rapes and sexual assaults not being brought to police attention (Langton et. al 2012). Yet, the fact that these crimes are hidden does not lessen their seri- ousness, and the disparity between the inci- dence of sexual assault and the reporting of it represents a brazen failure of the criminal justice system. Obviously, reporting the incidence of an assault crime is a necessary condition for any police response. Indeed, an offender cannot be arrested if law enforcement is not aware of the occurrence of the crime com- mitted. This holds consequence both at the individual level, as the victim of a sexual assault sees their offender walk free, and at a greater level, as the weakened threat of con- viction erodes the potential to deter further crime. If rapists and would-be rapists per- ceive the likelihood of conviction to be low, dissuasion by threat of punishment dimin- ishes. M

The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

The Factors That Determine Sexual Assault Reporting as They Compare to Non-Sexual Assaults

What factors cause a sexual assault to be reported to the police? And further, are these factors unique to sexual assault, or are they similar to other forms of assaultive violence? This research an-swers these questions by employing data collected by the National Crime Victimization Survey through each of the years from 2000 to 2013. By developing a probit regression model, this analysis predicts the police-reporting behavior of female victims of both sexual assaults and assaults of a non-sexual nature. While the results of our analysis agree with previous literature that the presence of a weapon and the incidence of an injury have significant explanatory power in predicting whether a sexual assault is reported, we propose that these factors are not unique. Rather, these factors ex-plain all types of assault, and not sexual assault in exclusivity. The only unique explanatory factors found in our research were whether the victim of the assault was employed at the time of the inci-dent, and the number of years she had been educated. The distinction between factors that influence all types of assault and those that only influence sexual assault is important as it may provide in-sight into the unique procedures necessary to encourage sexual assault victims to report the crimes committed against them to the police. Introduction

ore than half of the nation’s vio-lent crimes, or nearly 3.4 million violent victimizations per year,

went unreported to the police between 2006 and 2010 (Langton et. al 2012). Of these violent crimes, sexual victimizations were found to be the most gravely underreported, with an estimated 211,200 rapes and sexual assaults not being brought to police attention (Langton et. al 2012). Yet, the fact that these crimes are hidden does not lessen their seri-ousness, and the disparity between the inci-dence of sexual assault and the reporting of

it represents a brazen failure of the criminal justice system. Obviously, reporting the incidence of an assault crime is a necessary condition for any police response. Indeed, an offender cannot be arrested if law enforcement is not aware of the occurrence of the crime com-mitted. This holds consequence both at the individual level, as the victim of a sexual assault sees their offender walk free, and at a greater level, as the weakened threat of con-viction erodes the potential to deter further crime. If rapists and would-be rapists per-ceive the likelihood of conviction to be low, dissuasion by threat of punishment dimin-ishes.

M

Page 2: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

Much of the literature devoted to explain-ing the paucity of sexual assault reporting has focused attention on the motivations be-hind the reporting behavior of victims of sexual assault in isolation (Bachman, 1993; Bachman, 1998; Williams, 1984; Allen 2007). While the results of this analysis may begin to explain the factors that lead women who have been sexually assaulted to report the crimes committed against them, it does not shed light on whether these factors are unique. That is, do factors exist which would increase the probability of a sexual assault being reported, which would not necessarily do the same for a non-sexual assault? The comparison between sexual and non-sexual assaults is necessary in order to pro-vide a measure for the uniqueness of how rapes are reported. This has direct policy implications as it may reinforce the belief that distinct procedures are essential in en-couraging sexual assault victims to report the crimes committed against them to the police. Data Set and Sample Selection The data used for this analysis was taken through the National Crime Victimization Survey (NCVS) for each of the years from 2000-2013, with the exception of 2006. In this year, a number of methodological changes to the survey suggest that the 2006 findings represent a temporary anomaly in the data, and thus we decided to exclude it from our analysis. Datasets were merged within each year, as data from the NCVS is stored in a hierarchically structured format. Data was then appended across the 13 years of interest. The NCVS has been collecting data on personal and household victimization through an ongoing survey of a nationally representative sample of residential address-

es since 1972. The survey, which is the pri-mary source of information on the character-istics of criminal victimization and on the number and types of crimes not reported to law enforcement authorities, is administered by the U.S. Census Bureau on behalf of the Bureau of Justice Statistics. The NCVS sample consists of approxi-mately 50,000 sample-housing units selected with a stratified, multi-stage cluster design. Twice each year, data is obtained from a na-tionally representative sample of roughly 49,000 households comprising about 100,000 persons. Included in the sample are individuals 12 years of age and older living in households and group quarters within the United States and the District of Columbia. Excluded are persons who are crews of ves-sels, in institutions (e.g., prisons and nursing homes) or members of the armed forces liv-ing in military barracks.

Our analysis focuses on female victims of assault crime, allowing us to compare across incidences of both sexual and non-sexual assault, without the potential of bias through gender differences. The exclusion of male respondents is especially apt in this particu-lar analysis, as sexual assault victims are predominately female.

Further, this research concentrates exclu-sively on respondents who had been victims of violent crimes. To determine whether a respondent has been assaulted, the NCVS asks, “How were you attacked?” and re-spondents can answer yes to one or more of various classifications of attack including rape, attempted rape, shot, stabbed, hit, etc. If respondents answered yes to being at-tacked in any of the preceding ways, they were considered to be victims of assault. A sexual assault was considered to be a crime of rape, attempted rape or any other assault of a sexual nature.

Page 3: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

Of the 4000 cases of assault, 12% had been sexually assaulted, while 88% had been assaulted in a non-sexual way. Further, 55% of assault victims had reported the crime committed against them to the police, while 45% had not. Table 1 illustrates the propor-tion of victims of each type of assault that reported and the proportion that did not re-port the incident to police. The results sup-port the notion that there is something unique about sexual assaults that make a vic-tim of the crime less likely to report.

As previously mentioned, the sample for the NCVS is obtained using a stratified, mul-ti-stage cluster design. This is opposed to a simple random sample, which ensures that each observation is independent from all other observations in the dataset. However, due to the method of data collection, we cannot make this assumption of independ-ence in our analysis while conducting tests of significance. Since there may exist corre-lation between the observations within cer-tain groups of the data, the standard errors of the estimates may be underestimated. Thus, our analysis took the clustering of the data into account before analyzing levels of sig-nificance.

Variable Choice

Dependent Variable Reported The Reported variable is a binary variable recording a no/yes response. The respondent is asked whether she reported the crime to the police at any point in time following the incident. Reported is an effective variable to model as it has a significantly strong re-sponse rate, minimizing the number of ob-

servations disregarded. The NCVS deals with a very sensitive matter that has the pro-pensity to sway forthright responses. That being said, the Reported variable has little room for deterring a truthful response. The distribution of “no” and “yes” responses is near equal; this is interesting when distin-guishing between two control groups. Predictor Variables Although we cannot definitively conclude the mechanisms by which the predictor vari-ables influence the respondent’s choice to report, we believe that there is at least a plausible, justifiable reason to include each of the predictors in our regression model. This is based off the fact there are logical explanations (sometimes both positive and negative) for each variable to influence the respondents decision. Income The respondents were asked to place themselves in an annual household income bracket. Income is included as a predictor variable in our model as we determined there to be potential differences in the incentives or opportunity costs associated with incident reporting for individuals of different income levels. For example, a respondent may be less inclined to report if they are from a low-er income bracket as legal pursuit is time intensive and expensive. Dependents Respondents were asked for the number of dependents living in their household, where a “dependent” is classified as any in-dividual under the age of 12. This was rec-

Table 1: Reporting Rates of Type of Assault Type of Assault Did Not Report Reported Sexual 66% 34% Non-Sexual 42% 58%

Page 4: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

orded in a continuous variable and has been included in our regression for reasons simi-lar to those stated for the Income variable. Specifically, respondents responsible for more dependents may be less likely to report their incident on the basis of time and mone-tary restrictions. Population The population of the city/town in which the respondent resides was recorded in a cat-egorical variable, with population brackets represented by the various categories. The Population variable was included in our model in efforts to capture potential geo-graphical factors explaining a victim’s re-porting behavior. More explicitly, we be-lieve there to be potential for different popu-lation sizes to influence a victim’s decision of whether or not to report. Consider a small town where there is an increased likelihood of the victim knowing the offender or ac-quaintances of the offender; a victim may be less likely to report in fear of reprisal from the offender or their acquaintances. Con-versely, it can also be the case that in a small town a victim is more likely to report, as it may be easier to identify and find the of-fender. Age The respondent’s age is recorded in a continuous variable and is included in our regression, as we believe age differentials can greatly influence the decision to report. Older respondents are likely more informed and understand the importance of convicting criminals. Younger individuals tend to be more impressionable, and may not under-stand the severity of the incident, or the pro-cess to go about reporting.

Years Of Education Respondents were asked for their highest level of academic achievement and accord-ingly were placed in a category within the education variable. We then recoded this variable to have continuous responses re-flecting the number of years of schooling the respondent obtained based on national aver-ages for certain levels of education. This was included in our model as we believe that more educated, informed individuals may have a greater appreciation for the im-portance of crime reporting and crime deter-rence, thus increasing the likelihood that they report the incident. Alternatively, indi-viduals of higher education are likely more informed on the nature of the legal system, and the uphill battle often associated with a successful sexual assault conviction. Black The respondents were required to specify their race, which was recorded in a categori-cal variable. Although there were multiple categories listed including Asian, American Indian, etc., the “White” and “Black” cate-gories made up the majority of the sample, as shown in Table 2. Thus, we decided to focus only on respondents who were of these two races. The Black variable was included in the regression to capture the potential dif-ferences in race on the decision to report. It might be the case that black individuals fear differential treatment by authorities, thus are more likely to refrain from reporting. Hispanic Origin Because it is highly likely that respond-ents fall within a race category, as well as specify that they are of Hispanic origin, a separate binary no/yes response variable, Hispanic, is included in our regression. Re-spondents simply answered no or yes to “are

Page 5: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

you of Hispanic origin” and the answer was recorded. Justifications for incorporating this variable in our analysis are parallel to those that explained the inclusion of Race. Weapon The respondent was asked whether or not the offender used a weapon, or a regular ob-ject such as a bottle or wrench as a weapon when committing the crime. The use of a weapon tends to intensify any crime, and can make for a much more traumatic experience for the victim. It is plausible for Weapon to have both positive or negative effect on inci-dent reporting; increased fear of the victim during the attack may incentivize incident reporting, while increased fear that the of-

fender could return with the weapon in re-sponse to the report may sway the offender from reporting. Either way, we aim to affirm or deny the significance of weapon use on incident reporting as proposed by Bachman (1998). Value Property Taken The dollar value of the respondent’s property stolen at the time of the incident was recorded in a continuous variable. It is reasonable to infer that the likelihood of an assault victim to report the crime may be greater if there was also property stolen. If property was taken at the time of the assault, respondents may have a greater incentive to report the crime, either because they feel fur-

Table 2: Race Responses Race Frequency Percentage White Only 3,365 78.24% Black Only 652 15.16% American Indian, Alaska Native Only 67 1.56% Asian Only 66 1.53% Hawaiian/Pacific Islander Only 2 0.05% White-Black 33 .77% White-American Indian 75 1.74% White Asian 12 .28% White-Hawaiian 3 0.07% Black-American Indian 10 .23% Black-Asian 1 .02% Black-Hawaiian/Pacific Islander 0 0% American Indian-Asian 0 0% Asian-Hawaiian/Pacific Islander 0 0% White-Black-American Indian 7 .16% White-Black-Asian 0 0% White-American Indian-Asian 0 0% White-Asian Hawaiian 3 .07% 2 or 3 Races 0 0% 4 or 5 Races 5 .12% Total 4017 100%

Page 6: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

ther victimized, or they are motivated by the possibility of recouping their lost property. Employed At Incident The respondent was asked whether or not they were employed at the time of the inci-dent. Because there is an opportunity cost with reporting, time and cost, this may sway the respondent’s decision to report if they are employed. Furthermore, if a respondent is unemployed at the time, they may have incentive to report and pursue charges in ef-forts to obtain compensation for the crime. Either way, we believe that the employment status of the respondent has potential to af-fect the choice to report. Pay Lost The amount of pay that was forgone by the respondent as a result of the crime was recorded in a continuous variable. Forgone pay could have resulted from injuries requir-ing hospital time, or time off due to trauma. Lost pay has the potential to incentivize a report; if a respondent has lost a large por-tion of pay, she may be more likely to report in efforts to charge the offender, and seek compensation. This variable seems similar to the EmployedAtIncident variable, but there exists a subtle, yet significant, distinction. PayLost addresses the monetary loss already incurred by the victim due to their injuries. EmployedAtIncident addresses the potential monetary loss the victim faces if she chooses to report. Type Of Assault The survey provided a binary variable indicating whether the respondent had an-swered ‘yes’ to any of the 14 types of assault questions asked beforehand. As previously mentioned, in our analysis, we focused only on the respondents who were recorded as

being assaulted and thus only those who an-swered positively to any of these questions. Three of the 14 categories concerned as-sault of sexual nature, namely Rape, Tried to Rape, and Sexual Assault, while the remain-ing 11 categories concerned assaults of non-sexual nature (hit with object, stabbed, shot at, etc.). We created a categorical variable called TypeOfAssault, which recorded whether the respondent was a victim of ei-ther at least one of the 3 types of sexual as-sault crimes or none of the 3 sexual assault crimes but at least one of the 11 types of non-sexual assault crimes. Stranger The respondent was asked a series of no/yes response questions about the nature of their relationship with the offender. We decided to generate a new binary variable called Stranger which recorded whether the respondent answered positively to any of the relationship categories (and thus had a rela-tionship) or responded negatively to all questions (and thus had no relationship). There are a couple implications of the of-fender being a stranger, which lead us to be-lieve it may have an effect on the choice to report. For example, if the offender is a stranger it is more likely that they are uni-dentifiable which may sway the respondent from reporting. At the same time, if there exists a relationship between the offender and the victim, she may be less likely to re-port out of feelings of loyalty, guilt or trust. Married The respondent was required to report their marital status at the time of the inci-dent, which was recorded in a categorical variable with such options as “widowed”, or “used to be married”. We created a new bi-nary variable called Married assigning a

Page 7: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

no/yes distinction to whether the respondent was married at the time of the incident or not. We are uninterested if the respondent used to be married (i.e. widowed, divorced or separated). Our rationale behind inclusion was that a respondent may choose not to re-port in efforts of keeping the incident from their spouse. Alternatively, the presence of a spouse may encourage the victim to report out of a desire for justice. Injured The respondents were asked to respond to a list of different types of injuries suffered from the assault/sexual-assault with a no/yes response. We created a binary no/yes Injured variable which recorded a 0 (no) response if the respondent did not indicate they had been injured in any of the injury categories. A 1 (yes) response was recorded if the re-spondent has answered positively to any of the injury categories. We believe that adding an injury to the incident intensifies the expe-rience for the respondent; the respondent may be more likely to report for this reason. Further, if the respondent experienced an injury resulting in hospital attention, it may have been mandatory for an incident report. Interaction Variables Offender Right

In the NCVS survey, respondents were asked whether their offender had the right to be in the location of the attack. Although we thought this variable would be valuable in our regression, the way the question was worded implied an element of subjectivity, as there may be different interpretations of what having a ‘right’ means. To solve this problem, we created an interaction term be-tween PrivateLocation and Stranger. Since both are binary variables, multiplying the two creates another variable, which assigns a

1 to respondents who reported their attack occurred both in a private location and by someone they did not know. Further, be

cause the two variables used to create the interaction are objective, we avoid the prob-lem of subjectivity.

The PrivateLocation x Stranger interac-tion is important to include in the model, as respondents may be more likely to report assaults when the offender did not have the right to be in the location of the incident. For example, if someone broke into the respond-ent’s house and assaulted the respondent, the offender has now committed two crimes, which likely further incentivizes an incident report.

Summary statistics of each of the variables chosen to include in our regression are given in Table 3: Empirical Specification The data was modeled using probit re-gression in an effort to predict the likelihood of a given respondent reporting the occur-rence of an incident. Through probit analysis we determine the marginal effects of our predictor variables on the choice of reporting the incident; that is the percentage change in the probability of reporting. Our analysis considers 3 distinct but related specifica-tions, each fit to answer the particular ques-tion at hand. Master Specification The first probit regression is referred to as the ‘Master’. This includes all independ-ent variables including the variable describ-ing the type of assault. Further, it includes all observations; that is, those assaulted sex-ually and those assaulted non-sexually. The Master model is specified as follows:

Page 8: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

𝑌!"#$%&∗ = 𝛼 + 𝛽! 𝑖.𝑇𝑦𝑝𝑒𝑂𝑓𝐴𝑠𝑠𝑎𝑢𝑙𝑡 + 𝛽! 𝐼𝑛𝑐𝑜𝑚𝑒

+ 𝛽!𝐷𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡𝑠 + 𝛽! 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛+ 𝛽! 𝐴𝑔𝑒 + 𝛽! 𝑌𝑒𝑎𝑟𝑠𝑂𝑓𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛+ 𝛽! 𝑖.𝑅𝑎𝑐𝑒 + 𝛽!𝐻𝑖𝑠𝑝𝑎𝑛𝑖𝑐𝑂𝑟𝑖𝑔𝑖𝑛+ 𝛽! 𝑂𝑓𝑓𝑒𝑛𝑑𝑒𝑟𝑅𝑖𝑔ℎ𝑡 + 𝛽!"𝑊𝑒𝑎𝑝𝑜𝑛+ 𝛽!! 𝑉𝑎𝑙𝑢𝑒𝑃𝑟𝑜𝑝𝑇𝑎𝑘𝑒𝑛+ 𝛽!" 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑑𝐴𝑡𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑡+ 𝛽!" 𝑃𝑎𝑦𝐿𝑜𝑠𝑡 + 𝛽!" 𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛+ 𝛽!"𝑀𝑎𝑟𝑟𝑖𝑒𝑑 + 𝛽!" 𝐼𝑛𝑗𝑢𝑟𝑒𝑑+ 𝛽!"𝑃𝑟𝑖𝑣𝑎𝑡𝑒𝐿𝑜𝑐𝑎𝑡𝑖𝑜𝑛 + 𝜖

𝑌!!"#$%& =

0 = 𝑁𝑜𝑡 𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑, 𝑖𝑓 𝑌!"#$%&∗ ≤ 01 = 𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑, 𝑖𝑓 𝑌!"#$%&∗ > 0

The purpose of the Master regression is to evaluate the marginal expectation of TypeO-fAssault on the Reported dependent variable, controlling for all independent variables. Depending on the sign of the coefficient, we establish which type of assault victim, sexual or non-sexual, is more likely to report the incident. This interpretation will have merit if it is proven statistically significant.

Non-Sexual Specification The second probit regression is referred to as “NSA”, short for non-sexual assault. As the name would suggest this regression was run for observations that were victims of non-sexual assaults exclusively. Again, the model is regressed with Reported as the de-pendent variable. Naturally this model is net of the TypeOfAssault variable as it is used for control purposes in this case rather than explanatory purposes. The NSA model is specified as follows: 𝑌!"#∗ = 𝛼 + 𝛽! 𝐼𝑛𝑐𝑜𝑚𝑒 + 𝛽!𝐷𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡𝑠

+ 𝛽! 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 + 𝛽! 𝐴𝑔𝑒+ 𝛽! 𝑌𝑒𝑎𝑟𝑠𝑂𝑓𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛+ 𝛽! 𝑖.𝑅𝑎𝑐𝑒 + 𝛽!𝐻𝑖𝑠𝑝𝑎𝑛𝑖𝑐𝑂𝑟𝑖𝑔𝑖𝑛+ 𝛽!𝑂𝑓𝑓𝑒𝑛𝑑𝑒𝑟𝑅𝑖𝑔ℎ𝑡 + 𝛽!𝑊𝑒𝑎𝑝𝑜𝑛+ 𝛽!" 𝑉𝑎𝑙𝑢𝑒𝑃𝑟𝑜𝑝𝑇𝑎𝑘𝑒𝑛+ 𝛽!! 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑑𝐴𝑡𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑡+ 𝛽!" 𝑃𝑎𝑦𝐿𝑜𝑠𝑡 + 𝛽!" 𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛+ 𝛽!"𝑀𝑎𝑟𝑟𝑖𝑒𝑑 + 𝛽!" 𝐼𝑛𝑗𝑢𝑟𝑒𝑑+ 𝛽!!𝑃𝑟𝑖𝑣𝑎𝑡𝑒𝐿𝑜𝑐𝑎𝑡𝑖𝑜𝑛+ 𝜖,𝑊ℎ𝑖𝑙𝑒 𝑇𝑦𝑝𝑒𝑂𝑓𝐴𝑠𝑠𝑎𝑢𝑙𝑡 == 0

Table 3: Variable Summary Statistics Variable Obs Mean Std. Dev. Min Max Income 4017 34,704.07 25,973.67 2500 82,500 Dependents 4017 .7259 1.0286 0 8 Population 4017 324,471 922,549.8 0 5,750,000 Age 4017 31.4996 14.4247 12 90 YearsOfEduc. 4017 12.1003 2.8700 0 22 Black 4017 .1623 .3688 0 1 HispanicOrigin 4017 .1250 .3307 0 1 Weapon 4017 .1598 .3665 0 1 ValuePropTaken 4017 83.5395 1414.421 0 70,000 Reported 4017 .5499 .4976 0 1 Employed 4017 .5327 .4990 0 1 PayLost 4017 19.9686 215.8535 0 7,000 TypeOfAssault 4017 1.1205 .3256 1 2 Stranger 4017 .9696 .1716 0 1 Married 4017 .1982 .3987 0 1 Injured 4017 .6336 .4819 0 1 PrivateLocation 4017 .4820 .4997 0 1 OffenderRight 4017 .4618 .4986 0 1

Page 9: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

𝑌!!"# =

0 = 𝑁𝑜𝑡 𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑, 𝑖𝑓 𝑌!"#∗ ≤ 01 = 𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑, 𝑖𝑓 𝑌!"#∗ > 0

While the Master specification explains why assault victims in general will choose to report or not, it does not give us information on how the factors leading to the report deci-sion change for each type of assault. The Non-Sexual specification allows us to inter-pret the marginal effects of the independent variables with respect to observations from the non-sexual assault category in isolation. Sexual Specification The final probit model, “SA”, deals with the observations from the sexual assault cat-egory exclusively. The dependent variable remains Report, and again the model is net of the TypeOfAssault independent variable as it is being used as a means of control. 𝑌!"∗ = 𝛼 + 𝛽! 𝐼𝑛𝑐𝑜𝑚𝑒 + 𝛽!𝐷𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡𝑠 + 𝛽! 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛

+ 𝛽! 𝐴𝑔𝑒 + 𝛽! 𝑌𝑒𝑎𝑟𝑠𝑂𝑓𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛+ 𝛽! 𝑖.𝑅𝑎𝑐𝑒 + 𝛽!𝐻𝑖𝑠𝑝𝑎𝑛𝑖𝑐𝑂𝑟𝑖𝑔𝑖𝑛+ 𝛽!𝑂𝑓𝑓𝑒𝑛𝑑𝑒𝑟𝑅𝑖𝑔ℎ𝑡 + 𝛽!𝑊𝑒𝑎𝑝𝑜𝑛+ 𝛽!" 𝑉𝑎𝑙𝑢𝑒𝑃𝑟𝑜𝑝𝑇𝑎𝑘𝑒𝑛+ 𝛽!! 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑑𝐴𝑡𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑡+ 𝛽!" 𝑃𝑎𝑦𝐿𝑜𝑠𝑡 + 𝛽!" 𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛+ 𝛽!"𝑀𝑎𝑟𝑟𝑖𝑒𝑑 + 𝛽!" 𝐼𝑛𝑗𝑢𝑟𝑒𝑑+ 𝛽!"𝑃𝑟𝑖𝑣𝑎𝑡𝑒𝐿𝑜𝑐𝑎𝑡𝑖𝑜𝑛 + 𝜖,𝑊ℎ𝑖𝑙𝑒 𝑇𝑦𝑝𝑒𝑂𝑓𝐴𝑠𝑠𝑎𝑢𝑙𝑡 == 1

𝑌!𝑆𝐴 =0 = 𝑁𝑜𝑡 𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑, 𝑖𝑓 𝑌!"∗ ≤ 01 = 𝑅𝑒𝑝𝑜𝑟𝑡𝑒𝑑, 𝑖𝑓 𝑌!"∗ > 0

This model allows us to interpret the marginal effects of the independent variables with respect to observations from the sexual assault category on their own. These effects are then compared to the effects observed in the NSA model.

Results

We now turn to the question of whether there are unique factors that have significant explanatory power over the decision to re-port a violent sexual crime. Table 4 presents the results of three separate yet related re-gressions. The results of the “Master” re-gression are shown in the first column, while the alternative regressions – non-sexual as-sault and sexual assault – are shown in the second and third columns, respectively.

Master Specification: Type of Assault As previously mentioned, the purpose of the Master specification was to evaluate the marginal expectation of the type of assault on the Reported dependent variable, control-ling for all other independent variables. More explicitly, we ascertained which type of assault victim was more likely to report the crime committed against them, and by how much. In the below output, we can see that the predicted probability of reporting for a vic-tim of sexual assault is 26 percent less than that of a non-sexual assault victim, holding all other independent variables at their means. Further, this marginal estimation is significant at the 1 percent level, indicating very probable explanatory power. This result is consistent with the previously published literature on the subject (Bachman, 1993; Bachman, 1998; Williams, 1984; Allen 2007) that suggest there is a distinction be-tween the crimes of sexual assault and other types of assault with respect to the reporting behaviors of their victims. We discuss whether there exist unique factors that ex-plain this inconsistency between the two types of crime in the following sections.

Page 10: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

Table 4: Regression Results

VARIABLES (Master) (Non-Sexual Assault) (Sexual Assault) Predicted prob. Predicted prob. Predicted prob.

TypeOfAssault 2.Sexual Assault

-0.258***

-

-

(0.0304) - - OffenderRight 0.0847 0.0876 -0.260 (0.125) (0.133) (0.347) Income -1.73e-06*** -1.94e-06*** -3.73e-08 (4.34e-07) (4.59e-07) (1.11e-06) Dependents 0.0255** 0.0261** 0.0193 (0.0109) (0.0116) (0.0262) Population -7.21e-09 -1.48e-08 3.81e-08 (1.24e-08) (1.31e-08) (2.45e-08) Age 0.0022** 0.0028*** -0.0023 (0.0009) (0.0009) (0.0023) YearsOfEducation 0.0015 0.0030 -0.0230* (0.0046) (0.0048) (0.0130) Black 0.132*** 0.153*** 0.0090 (0.0322) (0.0348) (0.0750) HispanicOrigin 0.0794** 0.0780** 0.154 (0.0326) (0.0337) (0.101) Weapon 0.118*** 0.104*** 0.266*** (0.0287) (0.0300) (0.0878) ValuePropTaken 2.71e-05 2.56e-05 0.0011 (1.89e-05) (1.75e-05) (0.0010) EmployedAtIncident 0.0120 0.0323 -0.116** (0.0242) (0.0259) (0.0580) PayLost 0.0004** 0.0004* 0.0003 (0.0002) (0.0002) (0.0003) Stranger 0.0655 0.0555 0.405 (0.0982) (0.102) (0.306) Married 0.0264 0.0304 -0.0228 (0.0285) (0.0297) (0.0868) Injured 0.160*** 0.160*** 0.138** (0.0218) (0.0227) (0.0641) PrivateLocation -0.0295 -0.0163 0.175 (0.124) (0.132) (0.341) Observations 4,017 3,533 484 Log pseudolikelihood -2545.1807 -2230.8824 -277.7861 Prob > chi2 0.0000 0.000 .0002

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

NOTES: All predictors at their mean value. All values reported to nearest 4 decimal places.

Page 11: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

Alternative Specifications: Weapon Presence Recall that this analysis is designed to distinguish the factors which cause a sexual assault to be reported to the police, and fur-ther, whether these factors unique to sexual assault rather than shared with other forms of assaultive violence. The two alternative specifications allow us to determine this by testing the explanatory power of each inde-pendent variable on each type of crime in isolation. As we can see above, the marginal effect of the presence of a weapon during a sexual assault is statistically significant at the 1 per-cent level, and thus we can conclude that this particular factor is very likely to have a cas-ual relationship with the dependent variable being tested. More explicitly, it indicates that the probability of reporting increases by 26.6 percent when the offender has a weap-on. While previous literature (Bachman; 1998) has found results consistent with this finding, it has not been investigated further to determine whether this factor is unique to sexual assaults. However, this specification is necessary in order to establish the distinct policy measures needed to encourage sexual assault victims to report. Interestingly, weapon presence is equally significant when tested against the sample of respondents who indicated being victims of non-sexual assaults. While the marginal in-crease in probability is not as strong as in sexual assault cases (a 10.4 percent increase in the likelihood of reporting), the analysis suggests that it is equally probable of having explanatory power. By consequence, our findings suggest that although weapon pres-ence has a significant casual effect on the probability that a victim of sexual assault

reports the crime to police, it would also in-crease the likelihood that a victim of a non-sexual assault reports as well. Thus, it is not unique. Alternative Specifications: Injured When the victim sustained an injury dur-ing her sexual assault, our analysis deter-mined that she was 13.8 percent more likely to report the incident to police compared to another respondent who had not been injured at the hands of her offender, holding all oth-er independent variables at their means. Moreover, we found that this result was sig-nificant at the 5 percent level. Again, this result is consistent with the research undertaken by Bachman (1998) who reported that the factors both presently and previously discussed – the sustainment of physical injuries and the presence of a weapon, respectively – were the only factors that appeared to significantly increase the likelihood of reporting to police. Parallel to the discussion concerning weapon presence, when the analysis was taken further and compared with assaults of a non-sexual nature, it was found that the explanatory significance was not unique. That is, the sustainment of injury was deter-mined to influence the reporting behavior of victims of both sexual and non-sexual as-saults. Furthermore, we found that a victim who incurs an injury in a non-sexual assault is 16 percent more likely to report than a vic-tim of the same type of crime who was not injured. This marginal effect is actually greater than in the case of sexual assault, where injured victims were 13.6 percent more likely to report. Thus, we found that the incidence of injury is not only equally significant across types of assault, but is also more a powerful indicator in the cases of non-sexual assault.

Page 12: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

Alternative Specifications: Employed at Incident One of only two factors in our analysis that showed uniqueness in terms of having significant explanatory power for sexual as-sault exclusively was whether the victim was employed at the time of her sexual assault. Namely, this factor showed significance at the 5 percent level when tested against the sample that experienced a sexual assault, but did not show significance when tested against those who experienced a non-sexual assault. Our regression analysis suggested that a person who is employed is 11.6 per-cent less likely to report than someone who is not employed, holding all other independ-ent variables at their means. Allen (2007) also found significance when he incorporated this variable in his analysis. His justification for its inclusion was as follows: “Employed victims possibly have a greater network of support in the form of co-workers, but alternatively their employment may present a greater oppor-tunity cost of reporting and participating further in the justice system, in light of the time required to do so.” (p. 632) Since the research at hand focuses on the unique factors that determine sexual assault reporting, it is important to briefly comment on why this factor may be significant for one type of assault and not the other. As men-tioned by Allen (2007), there exists an op-portunity cost of reporting. However, this additional cost is present in all types of as-sault cases. What distinguishes the oppor-tunity cost of reporting a sexual assault ver-sus a non-sexual assault is the expected re-turns from reporting. Sexual assault victims may feel that the probability of securing a conviction is low,

as these types of crimes are notoriously dif-ficult to prove occurred, or alternatively, to prove occurred but were not consensual. If the decision to report an assault is motivated by a simple cost-benefit analysis, it may be the case that, in the incidence of a sexual assault, the expected benefits of reporting do not often outweigh the expected costs. Another distinguishing feature of sexual assaults is the feeling of shame that a victim may experience after her attack. Although not empirically measurable, this shame that a victim is undoubtedly grave. We cannot jus-tify any causal claims of shame negatively weighing on the choice to report, but we can hypothesize how it would add to the argu-ment of employment’s role and the oppor-tunity cost of reporting. The pursuit of re-porting not only presents the opportunity cost of forgone work hours, but also the cost of the shame in revealing the nature of the attack to the authorities and eventually the public if a conviction is made. Furthermore, as conviction requires absence from work, questioning from bosses and managers arise. These questions bear answers the victim of a sexual assault may not want to reveal in fear of the shame associated. Alternative Specification: Years of Education The measure of the respondent’s educa-tion level, YearsOfEducation, also proved unique in having explanatory power for sex-ual assault victims exclusively. For each ad-ditional year of education attained by the respondent, the likelihood of reporting fell by 2.3 percent holding all other independent variables at their means. This was proven statistically significant at the 10 percent lev-el.

Page 13: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

Individuals of higher education are likely more informed on the nature of the legal sys-tem, and the uphill battle often associated with a successful sexual assault conviction. We believe this finding to be in accord with our hypothesis regarding the cost benefit analysis of reporting. Since the probability of conviction in sexual assault cases is low relative to other types of assault, the ex-pected benefits of reporting is correspond-ingly low. Further, a low probability of con-viction may increase the expected costs of reporting as well, as victims may fear repris-al by the accused. Thus, if the costs associ-ated with reporting outweigh the low ex-pected benefits, a victim is unlikely to re-port. If we assume that with more years of education come greater awareness of the le-gal justice system, and in particular the low conviction rate for crimes of sexual assault, then this is especially true for well-educated victims. It is interesting to note that while the marginal effect of YearsOfEducation on re-porting is negative for sexual assault, it is positive for non-sexual assault. That is, our model suggests that while education causes a sexual assault victim less likely to report the crimes committed against them, it makes a victim of non-sexual assault more likely to report. This has the effect of making the var-iable YearsOfEducation unique to sexual assault not only in terms of significance, but also in terms of predicted effect. Measures of Fit: Probit Model Prediction Evaluation Using predicted values generated by our models, we determined the percent correctly

classified values to serve as a basic indicator of our models accuracy and fit. As reported in Table 5, in all cases the Master, Non-Sexual Assault, and Sexual Assault models proved to predict the true value of the Report variable more than 60% of time signifying moderate strength of the regression. Alt-hough Sexual Assault’s PCP of over 70% may seem pleasing, the percent correctly predicted method should be interpreted with discretion as it does not take into account uncertainty in the estimated values Yi. This is a result of the framework of PCP; estimat-ed values of Yi equaling .51 and .99 are es-sentially treated the same (correctly predict-ed) although separated by 48% percent of certainty. Discussion and Policy Implications Our analysis of present day assault vic-timization data reinforces the findings of the earlier works by Bachman (1993; 1998) in that the major causal factors of reporting by sexual assault victims were found to be the presence of a weapon and the incurrence of injuries. However, a crucial element was missing in Bachman’s research; determining whether these factors were unique to the crime of sexual assault. When tested against the victims of non-sexual assault crimes, both weapon presence and injuries incurred were found to be significant, and thus we cannot conclude that these factors uniquely contribute to sexual assault reporting. In our research, a unique factor is one that is found to have a significant causal ef-fect on reporting for sexual assault, but not

Table 5: Percent Correctly Predicted Model Percent Correctly Predicted Master 63.41% Non-Sexual Assault 63.46%

Sexual Assault 71.07%

Page 14: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

for non-sexual assaults. Our model suggest-ed that two factors fit this classification of unique – whether the victim was employed at the time of incident, and the number of years of education that the victim had ob-tained. The distinction between factors that influence all types of assault and those that only influence sexual assault is important as it may provide insight into the procedures necessary to encourage sexual assault vic-tims to report the crimes committed against them to the police. Our model showed that a sexual assault victim who is employed is 11.6 percent less likely to report that someone who is not em-ployed. As suggested by Allen (2007), this may be a result of the opportunity cost asso-ciated with taking time off of work to partic-ipate further in the justice system, should the report turn into an arrest and trial. If a victim makes a reporting decision through a basic cost-benefit analysis, it follows that in order for her to report, the benefits must outweigh the costs. Although the personal benefits of report-ing (the satisfaction felt by receiving justice for the crimes committed, the loss of fear of another attack by the same offender) are hard to measure, the costs of reporting are very tangible and have direct policy implica-tions. To encourage sexual assault reporting, government may subsidize sexual assault victims for any compensation forgone in the criminal justice process. It should be noted that although our evidence suggests this so-lution may increase reporting rates, it does not necessarily prove that the proposed poli-cy is feasible or efficient, as it may incentiv-ize individuals to report when there was no incident to begin with. This may, in turn, discredit the claims of legitimate sexual as-sault victims, who already have trouble proving the guilt of their offenders in court.

Our research also suggested the signifi-cance of the number of years of education obtained by the victim on her likelihood of reporting. More specifically, for each addi-tional year of education attained by the re-spondent, the probability of reporting fell by 2.3 percent holding all other independent variables at their means. We suggested that this was a result of educated victims being aware of the uphill battle often associated with a successful sexual assault conviction, thus creating low expected returns and high expected costs of reporting. Unlike the em-ployment factor discussed previously, the policy implications of this variable are more subtle. Although we would be hard pressed to enact policy that directly increases the effi-ciency of the legal system with respect to sexual offense convictions, there still exist positive policy implications to draw from this result. An example of this would be the funding of ad campaigns that raise aware-ness on the benefit of reporting of crimes regardless of the report outcome. Although a victim may feel swayed from reporting in anticipation of no conviction, there are still expected benefits to be gained from a report nonetheless. Sexual assault reports can allow police to pinpoint the areas in which the in-cidents are occurring, and even issue public reports warning the community to stay alert. Thus, public campaigns may encourage vic-tims to have a greater appreciation for the importance of crime reporting in terms of its effect on crime deterrence – increasing bene-fits, and thus the likelihood, of reporting. References Allen, W. David. (2007). The Reporting and Underreporting of Rape. Southern Economic Journal. 73(3). 623-641. DOI:10.2307/20111915.

Page 15: The Factors That Determine Sexual Assault Reporting as They …qed.econ.queensu.ca/faculty/ferrall/Econ452FinalReport.pdf · 2015. 2. 4. · The Factors That Determine Sexual Assault

Bachman, Robert. (1993). Predicting the Reporting of Rape Victimizations: Have Rape Reforms Made a Difference? Criminal Justice and Behavior, 20, 254-270. DOI:10.1177/0093854893020003003 Bachman, Robert (1998). The Factors Related to Rape and Reporting Behavior and Arrest: New Evidence from the National Crime Victimization Survey. Criminal Justice and Behavior, 25(1), 8-29. DOI:10.1177/0093854898025001002 Langton, Lynn; Berzofsky, Marcus; Krebs, Christopher; Smiley-McDonald, Hope. (2012). Victimizations Not Reported to the Police 2006-2010. U.S. Bureau of Justice Statistics (NCJ Publication No. 238536). Washington DC: Retrieved from http://www.bjs.gov/content/pub/pdf/v nrp0610.pdf Williams, Linda S. (1984). The Classic Rape: When Do Victims Report? Social Problems. 31(4). 459-467. DOI:10.2307/800390. Links to Source Data All data was obtained from the ICPSR database. Each year was separated into 5 da-ta subsets. Three of these subsets, House-hold, Person, and Incident, were downloaded and merged for each year. The merged year-ly sets were then appended into one data set containing the 13 years 2000 through 2013, net of year 2006 due to inconsistencies in data collection. 2000 https://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/22921?q=ncvs&amp;searchSource=icpsr-landing

2001 https://www.icpsr.umich.edu/icpsrweb/ICPSR/series/00095/studies/22920?archive=ICPSR&amp;sortBy=7&amp;paging.startRow=26 2002 https://www.icpsr.umich.edu/icpsrweb/ICPSR/series/00095/studies/22902?archive=ICPSR&amp;sortBy=7&amp;paging.startRow=26 2003 https://www.icpsr.umich.edu/icpsrweb/ICPSR/series/00095/studies/22901?archive=ICPSR&amp;sortBy=7&amp;paging.startRow=26 2004 https://www.icpsr.umich.edu/icpsrweb/ICPSR/series/00095/studies/22900?archive=ICPSR&amp;sortBy=7&amp;paging.startRow=26 2005 https://www.icpsr.umich.edu/icpsrweb/ICPSR/series/00095/studies/22746?archive=ICPSR&amp;sortBy=7 2007 https://www.icpsr.umich.edu/icpsrweb/ICPSR/series/00095/studies/25141?archive=ICPSR&amp;sortBy=7 2008 https://www.icpsr.umich.edu/icpsrweb/ICPSR/series/00095/studies/26382?archive=ICPSR&amp;sortBy=7 2009 https://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/28543 2010 https://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/31202 2011 https://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/34061 2012 https://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/34650 2013 https://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/35164