Fair Treatment or Preferred Outcome? The Impact of Police Behavior on Victim Reports of Domestic Violence Incidents

Abstract

This research is an exploratory test of two hypotheses emerging from debates about how police behavior may influence domestic violence victim reporting. From a procedural justice perspective, victims should be more apt to report victimization when previous encounters with police are viewed as procedurally fair. From a distributive justice perspective, denying victims their preferred outcome may discourage future police utilization. We find that satisfaction with police is related to both distributive and procedural justice but that re-utilization of police is conditioned by preferred outcome. Specifically, if the offender was arrested in accordance with victim preference, the victim is significantly more apt to utilize police in the future.

Introduction

Only relatively recently has official victim reporting captured the attention of researchers (Greenberg & Ruback 1992; Langan & Innes 1986). This is partially due to a long-standing perception of police as gatekeepers of the criminal justice system (e.g., Frisch & Caruso 1996; Buzawa et al. 1992). Contrary to this traditional view, Hindelang and Gottfredson (1976) argue that from the perspective of the criminal justice system, suspects, and criminological inquiry, it is the victim who serves as gatekeeper. By deciding whether to officially report crimes, victims control mobilization of the criminal justice system, of which police are a part. Since victims serve as the first major decision maker in the criminal justice process, understanding the factors that influence victim reporting behavior should be a prominent criminological endeavor (Gottfredson & Gottfredson 1980; Hindelang and Gottfredson 1976).

In the case of victims of domestic violence, official reporting has been a topic of much discussion but very little research. Over the past 30 years, reformers have brought about many changes, perhaps most dramatic of which is the criminalization of domestic violence (Fagan 1996). A necessary condition for application of these reforms is that incidents come to the attention of the criminal justice system. Victims of domestic violence, like victims of other crimes, are the most likely parties to report their victimization to police (Bachman & Coker 1995; Waller 1990). As such, victims are the primary source of what Black (1980:45) calls legal intelligence, or information about violation of the law received by the authorities.

Still unclear is the extent to which domestic violence victims are willing to provide this legal intelligence. According to the National Crime Victimization Survey (NCVS), between 1992 and 1996, about half of the female respondents disclosing a domestic violence incident stated that they had reported the incident to police (Greenfeld et al. 1998). This is roughly equivalent to the rate of violent crime reporting overall (Laub 1997). Estimates derived from the National Family Violence Survey are substantially lower than the NCVS estimates. In this survey, about 7% overall and 14% of those suffering serious violence reported to police (Kantor & Straus 1990). This range is more consistent with other research that generally finds that less than 20% of domestic violence incidents come to the attention of the police (Dutton 1995; Greenberg & Ruback 1992). Whatever the actual rate of official reporting, it is clear that large numbers of victims continue to refrain from seeking the assistance of the police when they are assaulted by a present or former sexual intimate.

In this research, we explore the issue of victim reporting in the context of domestic violence victimization. Of particular interest is whether a relationship exists between the victim's assessment of police behavior during a previous domestic violence encounter and the likelihood that she will report subsequent victimizations. We begin with a review of the domestic violence victim reporting literature.

Domestic Violence Victim Reporting Behavior

Only a handful of studies have focused specifically on identifying factors associated with domestic violence victim reporting, and results are equivocal. There appears to be no important difference in reporting by victim age (Hutchison 1999; Bachman & Coker 1995; Johnson 1990). Marital status has produced inconsistent results. For example, Felson and colleagues found that victims were more likely to report assaults by ex-spouses than spouses, but no difference was identified in reports of violence by other types of partners (Felson, Messner, & Hoskin 1999). Berk and colleagues found that married victims were less likely to report their partners' violence than cohabitating victims (Berk et al. 1984). Other studies, however, have found marital status to be unimportant (Hutchison 1999; Bachman & Coker 1995; Johnson 1990). Additional inconsistencies emerge around the role of race. One study utilizing a shelter sample (Berk et al. 1984) and two utilizing national survey data (Felson, Messner, & Hoskin 1999; Kantor & Straus 1990) found that victim race was unimportant. In data from the pre-redesigned NCVS and from victims identified through police contacts, race was associated with victim reporting. Black victims were more likely to report than white victims (Hutchison 1999; Bachman & Coker 1995).

A common hypothesis about reporting behavior is that middle- and upper-class women are less likely to report domestic violence victimization than other women because they are more financially dependent upon the offender. Reporting victimization to police may jeopardize their partner's source of income and thus the victim's means of support (Hart 1993). In addition, Black (1980) argues that financial dependence is often accompanied by a patriarchal relationship structure. Through their greater financial dependence, middle-class women are under greater authoritarian control than are lower-class women (Black 1980:125). The available research has not supported this contention. Of the three indicators of socioeconomic status employed in the domestic violence reporting literature, few significant differences were noted. One study using a shelter population found offender employment and victim (or offender) education level to be negatively associated with reporting (Berk et al. 1984), but these results have not been replicated (Johnson 1990). Other indicators of social class employed across studies (e.g., victim or household income) were unrelated to reporting (Hutchison 1999; Bachman & Coker 1995; Johnson 1990; Berk et al. 1984). Thus, there is not much reason to suspect that socioeconomic status has a direct impact on victim reporting behavior. However, none of the available studies included a direct measure of financial dependence. Consequently, the relationship between financial dependence and victim reporting has not been adequately addressed.

There are other inconsistencies around the impact of incident seriousness on reporting. Shelter samples show it to be unimportant (Johnson 1990; Berk et al. 1984), while studies employing national samples have found some measure of victim injury or incident seriousness to influence reporting (Felson, Messner, & Hoskin 1999; Bachman & Coker 1995; Kantor & Straus 1990). This could be due to difference between the samples. Shelter samples are generally composed of women with a more extensive history of victimization, both in terms of frequency and severity, than are national samples (Hutchison 1999). It may be that the seriousness of individual victimizations matters for reporting among women with little to no domestic violence history, but that other factors are more important for women with more exposure to violence. Whether and how domestic violence history matters to reporting behavior remains an open question.

We could uncover no research directly examining the impact of police behavior on future victim reporting or the relationship between victim satisfaction with police performance and reporting of subsequent domestic violence. There is some indication that victims may generally feel satisfied with police, particularly when they receive greater amounts of police attention (Jolin & Clavadetscher 1995). Still unclear are what aspects of police behavior affect victim satisfaction and if the level of satisfaction has an impact on future behavior, including calls to police.

Hypotheses

Given the frequency with which this relationship is discussed by researchers, policy makers, and reformers, the present void in the literature is surprising. A common assertion is that domestic violence victims (and victims in general) who have negative experiences with police will refrain from calling them for assistance in the future. It is argued that bad experiences with police lead to negative expectations about future interaction and thus decisions not to seek police involvement. Under these conditions, Lerman (1992) refers to police action as deterring the victim, rather than the offender. By the same token, positive attitudes toward police (primarily derived from previous direct experience) are expected to increase victim reporting of domestic violence and other crimes as well (e.g., Davis & Taylor 1997; Hart 1993; Bowman 1992; Friedman & Shulman 1990; Kelly 1990; Kidd & Chayet 1984; Skogan 1984).

There are two common hypotheses about how police behavior may impact victim attitude or expectations, which in turn impact reporting. One hypothesis is that police demeanor toward victims, defined by Stephens and Sinden as “perceptions of the attitudes of police as expressed in their behavior” (2000:537), increases or decreases future reporting behavior. For example, victims who perceive that police are hostile or blaming are less likely to call them for assistance than victims who perceive that police are caring, supportive, and concerned (Davis & Taylor 1997; Hart 1993; Friedman & Shulman 1990; Kelly 1990; Kidd & Chayet 1984; Skogan 1984). Hereafter, we refer to this as the process hypothesis. The second hypothesis concerns attitudes or expectations formed as a result of the outcome of police involvement, e.g., whether offenders are arrested or ordered away from victims, and how this police action fits with victims' preference of outcomes. According to this hypothesis, victims who get the outcome they want from police are more likely to utilize them in the future than victims whose preference did not match the outcome. This assertion (we refer to it as the outcome hypothesis) commonly appears in arguments against the adoption of mandatory arrest laws (Gelles 1993; Bowman 1992; Buzawa et al. 1992; Sherman 1992; Ford 1991; Miller 1989).

These two hypotheses are repeatedly mentioned in the victim reporting and domestic violence literature. Yet they are not well developed, discussed relative to one another, or combined into a single hypothesis about possible interactive effects of both the process and outcome of police involvement on future reporting behavior. Rather than serving as the focus of empirical investigation, these hypotheses are often proposed to account for unexpected research results. For example, Davis and Taylor (1997) conducted an evaluation of a domestic violence program in New York City teaming police with social workers. According to victim interviews, there were no significant differences in the level of repeat violence between the program and control conditions. In official data, however, the program was associated with a greater number of violent incidents reported to police than was the control condition. The Portland, Oregon Domestic Violence Experiment also revealed a similar discrepancy between official and interview-disclosed re-offenses (Jolin et al. 1998). Both the New York and Portland researchers explain their findings by forwarding the process hypothesis. They state that the greater number of official re-assault reports observed in the program condition (relative to the control) may have been produced by the positive experiences of program group victims with police. Thus, neither program was considered a failure because interview data suggested that it is victim behavior, rather than offender behavior, that produced the higher level of official reports among those receiving program services.

As an introduction to this line of empirical investigation, we present an exploratory analysis of the possible impact of police behavior on the subsequent reporting of female domestic violence victims. This study tests the two predominant hypotheses, i.e., the process and outcome hypotheses. The initial challenge is determining how to operationalize the two hypotheses in a manner consistent with their discursive definitions in the available literature. Fortunately, the meaning of the outcome hypothesis is straightforward, making a measurement scheme clear. This hypothesis, articulated in opposition to mandatory arrest laws, states that a call to police will be more likely when the outcome of a previous call was consistent with a victim's preference.

In theoretical terms, the outcome hypothesis can be understood as reflective of a social learning process (Akers 1973). Social learning theory predicts that behavior will be elicited and strengthened if it has been positively or negatively reinforced. In this case, victims will be more apt to call the police if, in previous encounters, police actions were consistent with their preferences. Importantly, social learning theory would not give greater weight to one outcome over another (i.e., arrest, mediation, or separation). Indeed, as Akers describes, social learning

is a theory of the contingent relationship between behavior and stimuli. The same stimulus can be reinforcing for one person but not for another, it can be reinforcing at one time and place and not at another, and it can strengthen a fairly wide range of behavior in the same person. But what is likely or not likely to be reinforcing or punishing for what and for whom is not part of operant learning principles. (1973:51, emphasis in original)

Therefore, victims who received a preferred outcome, regardless of what it is, should be more apt to call the police again under conditions of revictimization.

Another way to theoretically contextualize the outcome hypothesis is to situate it within an instrumental view of procedural justice (Thibaut and Walker 1975). According to this viewpoint, disputants define procedural justice primarily in control terms, i.e., procedures will be viewed as fair to the extent that disputants maintain significant control over important life decisions. Fair procedures, according to this theory, are those in which disputant preferences are exercised. Thus, when police act according to the preferences of victims, the victims are more satisfied with the procedure (i.e., the police) and more apt to think it fair. This condition increases the likelihood that victims will utilize the police again in response to future victimization (Tyler and Lind 2001). Tyler (1990:175) suggests that this approach to procedural justice is instrumental in two respects. First, it focuses on the interaction between the disputants (offender and victim) and an authority (police) in which the procedure represents what the authority does to resolve the dispute. Second, the disputants seek control over the interaction in order to get a favorable response from the authority. Their reactions to the experience will be shaped by their perceived control over the outcome (either directly or indirectly) (Tyler 1990:139).

With regard to the process hypothesis, a measurement strategy is not as self-evident because it has been articulated in number of different ways. For example, Bower refers to the depressant effect on victim reporting of police demeanor that is “victim aversive” and interaction with victims “in a demeaning fashion” (1982:490). Hart describes “victim-blaming attitudes” of police as discouraging of subsequent reporting (1993:626). Referring to crime victims in general, Skogan states that victims who feel “mistreated by the police are less willing to report their experiences” (1984:155). Similarly, Kidd and Chayet assert that the level of victim reporting is influenced by “the threat of further victimization from authorities” (1984:41). Kury and colleagues refer to the suppressant effect of “less positive relationships between citizens and police” (Kury, Teske, & Wurger 1999:126) on the level of victim reporting. Some phrases used to describe the positive side of the process hypothesis are that reporting is increased among victims who are more “trusting of the police” (Schneider, Burcart, & Wilson 1976) or receive proper “treatment by police” (Greenberg & Ruback 1992). Overall, it seems that the commonality in the numerous descriptions is whether police treat victims fairly during their interaction. Consequently, testing the process hypothesis requires operationalization of the term fair treatment.

Theories of procedural justice provide explicit criteria to assess whether individuals perceive fair treatment from authorities. We used Tyler and Lind's version of procedural justice to test the process hypothesis. A general statement of their theory is that perceptions of fairness are dependent upon the process by which outcomes are distributed rather than an assessment of the fairness or favorableness of the outcomes themselves. Tyler and Lind's group process model (Tyler 1990; Lind & Tyler 1988; Tyler & Lind 1992) is the most commonly applied in the field of criminology. Tyler and Lind (1992) decompose procedural justice into three factors that represent independent influences on individual perception of fair treatment: standing, neutrality, and trust. The standing component of procedural justice refers to the extent to which individuals are treated as valued group members—with politeness, dignity, and respect. The neutrality component refers to perceptions about authorities' bias, honesty, and use of accurate and full information to make decisions. Trust refers to perceptions of the intentions of the authorities to act fairly, show concern for individual needs, and behave in an ethical manner.

Parts of Tyler and Lind's models have been used to examine whether offender perceptions of procedural justice in misdemeanor domestic violence cases decreased recidivism, independent of the favorability of outcome (arrest or nonarrest). Using arrestee interviews from the Milwaukee Domestic Violence Experiment, Paternoster and associates found “at least moderate support for the predictions that spouse assaulters' perceptions of procedural justice and fair treatment by the police are important determinants of the propensity for future conduct” (Paternoster et al. 1997:192). Like Paternoster and his colleagues, our research interests fall outside the parameters of most procedural justice research—investigations tend to examine whether people who fail to receive distributive justice outcomes are nevertheless satisfied with the outcome if they receive procedural justice. Consistent with the policy debates, we focus on whether receiving procedural or distributive justice from the police affects the likelihood that domestic violence victims will utilize the police again following subsequent victimization.

Methodology

Although the five National Institute of Justice-funded Spouse Assault Replication Program (SARP) Experiments were potential sources of data for this analysis, we selected the Metro-Dade County (Florida) SARP Experiment (Pate, Hamilton, & Annan 1991, 1994; Pate & Hamilton 1992) because it is best suited for measuring both the outcome and process hypotheses. This experiment was designed to test whether the use of arrest produced a deterrent effect among suspects randomly assigned to receive this outcome relative to those receiving contact from police but no arrest. Eligibility requirements were: (1) responding police officer(s) determined that probable cause existed to arrest the offender for a misdemeanor spouse battery offense, (2) the offense was not a felony, (3) both the victim and suspect were present when police arrived, (4) the police did not believe there was immediate danger to the victim, (5) the victim was an adult female (at least 18 years old), (6) neither the suspect nor the victim attempted to assault responding police officers, and (7) there were no warrants, injunctions, or criminal protective orders against either the suspect or the victim in the case.

Data were collected from August 1987 to July 1989. For the first 13 months of the experiment, only incidents involving married and formerly married individuals were eligible to be included in the study. Following a legislative change expanding the arrest powers of police, data collection was expanded for the remaining months to include present and former unmarried cohabitants and dating partners.

Cases meeting the above eligibility criteria were randomly assigned an outcome when they were reported to police via the Metro-Dade Police Department's Computer Assisted Dispatch system. Responding police officers were instructed as to the whether the case fell into the “arrest” or “nonarrest” condition. Suspects randomly assigned to the arrest condition were to be handcuffed and transported to the local jail. The amount of time in custody varied widely, but average jail time was 14.6 hours. Offenders in the nonarrest condition did not receive any particular intervention. Victims in these cases were supposed to receive a brochure on their legal rights in the state of Florida. Officers were then to leave the scene. In 90% of the cases, the arrest and nonarrest conditions were delivered as randomly assigned. Researchers conducting the original experiment found a deterrent effect of arrest when revictimization was measured by victim reports, but not in official measures of repeat violence (for details, see Pate, Hamilton, & Annan 1991; Pate & Hamilton 1992).

From the Metro-Dade SARP Experiment, we used the victim interview data to examine how perceptions of police behavior may impact official reporting behavior, i.e., whether perceptions of police encourage or discourage future victim calls for police services. Victims were interviewed within several weeks of the “experimental incident” and again after six months. Of 907 eligible cases, 594 (65%) victims completed the initial interview. A similar share (385 of 594, or 65%) of these victims completed the six-month follow-up interview, accounting for 42% of the 907 originally eligible cases (Pate, Hamilton, & Annan 1994).

Sample Description

Because of the focus of the present research, we identified the subsample of victims who suffered some form of subsequent victimization that may or may not be reported to police. In both the initial interview and six-month follow-up interview, victims were asked whether there had been any form of revictimization since the experimental incident. Specifically, in both interviews victims were asked four questions about the behavior of the offender following the initial experimental incident, and these were classified into four categories of revictimization. Since the experimental incident, had the offender (1) hit, slapped, or tried to hurt the victim; (2) hit, slapped, or tried to hurt a family member; (3) made any threats to physically harm the victim, a family member, or damage property; and/or (4) actually damaged property?

Of the 594 interviewed victims, 396 did not answer yes to any of these questions in either the initial or the follow-up interview (or did not complete the follow-up interview), and we thus excluded them from our subsample, leaving 198 remaining cases. Of those, we excluded 18 cases because they were missing data on the dependent variable. Thus, the subsample we used in the present study consisted of 180 cases. Of the 180 cases in the revictimized subsample, follow-up interviews are missing for 46 women. We retain these cases because all had been revictimized before the initial interview and therefore had the opportunity to consider calling police for help in light of their experiences with police during the experimental incident.

One considerable restriction on the generalizability of our findings is the large amount of missing data that plagues the Dade County victim interview data. The question that arises relates to how well this group of located, cooperative victims represents the full group of 907 eligible victims. Victims who could not be located or refused to cooperate may be more seriously victimized, may live in greater fear of the offender, may be more likely to be homeless, or may experience household instability than those present in the sample. As a result, this study's sample may underrepresent high-risk victims. We found no significant differences between the demographic characteristics of victims who completed and those who did not complete the follow-up interview. In a comparison of the characteristics of the group of all 907 eligible victims to the 594 who completed the initial interview, we uncovered significant differences in victim employment status and race. Employed victims (likely those with a higher level of stability) were more likely to participate in the initial interview, as were blacks and Hispanics relative to whites. Given that race effects have been found in some victim reporting studies (Hutchison 1999), we understand that any race effects we uncovered in this study may be due to biases built into our sample. Yet given the exploratory nature of this study, the Dade County data provided a reasonable first test of the process and outcome hypotheses. Further, we are also encouraged by the assessment of the missing interviews in the five SARP experiments by Garner and Maxwell (2000), who assert that, despite their shortcomings, interview completion rates of these experiments are comparable, if not superior, to most other criminological research of this nature.

Of the 180 women in our sample, the majority were either black (41%) or Hispanic (23%) (See Table 1). Offenders and victims were from the same racial group in 87% of the cases. Victims ranged in age from 18 to 80, with an average age of 32. Most (69%) had achieved at least a high school diploma and were employed at the time of the experimental incident (63%), working on average about 40 hours per week. About one in three stated that they were not dependent at all on the offender for financial support, while a similar share stated that they were totally or very dependent on the offender. Most offenders (81%) were husbands or ex-husbands (4%) of the victim. Generally, the relationships were lengthy, with the mean length of relationship being nearly nine years. Nearly 90% of the victims were living with the offender at the time of the experimental domestic violence incident.

Table 1. Victim Sample Demographic Characteristics

¹ May not total 180 due to missing data.

² May not total 100% due to rounding.

For most of the women in the sample, the experimental incident was not the first time they had suffered victimization at the hands of this partner. In the six months prior to the experimental incident, about 65% had been hit, slapped, or hurt in some way. Also during this period, 20 (11%) reported that other family members had been hit or hurt, and a little more than half of these were children. About one in three women stated that the offender had damaged property, and 40% reported that the offender had issued threats to harm them, other family members, or to damage property. Overall, almost 75% stated that the offender had engaged in at least one of these four abusive behaviors in the six months prior to the experimental incident. Of these, 40% stated that police were called in response at least once. Most of these calls came from the victim herself (76%), with the remaining reported by another family member (11%), the offender (8%), or “someone else” (17%). Regarding the experimental incident, most women (70%) stated that they had been the one who called police. Thus, it is clear that the sample contains a group of women who were exposed to repeated victimization of various types and, on at least some occasions, most were willing to call police for assistance.

Outcome and Process Measures

According to the outcome hypothesis, reporting of future domestic violence incidents is dependent upon whether victims have received their preferred outcome from police. In the Dade County experiment, the two possible outcomes were arrest and nonarrest. In the initial interview, victims were asked whether they wanted police to arrest the suspect in response to the experimental incident. Recall that arrest and nonarrest were randomly assigned outcomes. We considered victims to have received their desired outcome when they answered yes to the arrest preference question and the offender was arrested. We also considered those who answered no and the offender was not arrested to have received their preferred outcome. Victims not receiving their outcome preference consisted of (1) those who answered yes to the arrest preference question, but the offender was not arrested, and (2) those who answered no, but the offender was arrested. We constructed a dichotomous variable where (1) indicated that the victim received her outcome preference and (0) indicated that she did not. We recognize that this is a rather simplistic method of measuring victim preference. Our intention was to test the hypothesis as it has been asserted in the literature. Among those arguing against mandatory arrest laws, the relationship between victim preference and official reporting is discussed almost exclusively in terms of this arrest/nonarrest dichotomy (e.g., Gelles 1993; Bowman 1992; Buzawa et al. 1992) rather than as a range of outcomes victims may desire.

We used a measurement model developed by Tyler and Lind (1992) to test the process hypothesis. While the Dade County data do not contain a complete specification of this model, several similar measures are available. In the initial interview, victims were first asked to rate how carefully police listened to their side of the story, from “very carefully” (4) to “not at all carefully” (1). This item is similar to Tyler and Lind's (1992)“consideration of views” measure. We took another indicator from victim ratings of how seriously the police officers took their situation, from “very seriously” (4) to “not at all seriously” (1). This is similar to Tyler and Lind's (1992)“dignity and respect” measure. The third indicator, similar to Tyler and Lind's “concern for needs” concept, was a dichotomous variable regarding whether victims felt the police officers really wanted to help (1) or did not (0). Using these three indicators, we created a procedural justice scale variable by computing the average of the standardized scores for each of the four indicators on each case (alpha=0.84).

Victims were also asked to rate on a four-point scale how satisfied they were with police, from very satisfied (4) to very dissatisfied (1). While victim satisfaction is important to procedural justice, for our purposes its meaning is not clear. Victims may feel very satisfied with police because they received fair treatment (process hypothesis) or because they received the outcome they wanted (outcome hypothesis). To examine whether victim satisfaction fits with the procedural justice indicators, we conducted a principal components factor analysis utilizing varimax rotation. The results revealed that all four indicators loaded together on a single factor (Satisfaction=0.878, Careful=0.862, Serious= 0.859, and Help=0.847). Despite these factor loadings, we have reasons to question the face validity of victim satisfaction as an exclusive indicator of procedural justice and not distributive justice, as articulated in the outcome hypothesis. Thus, we did not include victim satisfaction in the procedural justice scale item but examined it as an independent variable.

Dependent Variable

In a general sense, the dependent variable of this study is clear—whether the victim called police or not in response to revictimization. Most victims, however, suffered more than one type of revictimization, and many suffered repeated revictimizations of the same type. The initial interview data did not distinguish reporting in individual incidents but asked victims whether they reported any of the potentially numerous repeat victimizations. The follow-up interview asked victims to state whether they reported only the first of potentially many victimizations since the initial interview. Consequently, we created a single dichotomous variable indicating whether victims reported at least one revictimization incident occurring before either interview. We included type and frequency of revictimization as control variables.

Control Variables

Demographic Characteristics

While the previous research is mixed on whether race plays a role in domestic violence victim reporting, there is reason to include race/ethnicity as a control in our analysis. Importantly, race and ethnicity may influence procedural justice judgments (Tyler & Smith 1999). We included three dichotomous variables to represent each of the most prominent racial groups—black, Hispanic, and white. The null findings in previous research related to socioeconomic status overall may be because social class itself does not have a direct impact on victim reporting. It may be that a victim's financial dependence upon the offender has a direct impact and that socioeconomic status is related to victim reporting through financial dependence (Hart 1993; Black 1980). To take this argument into account, we included a variable that contains data on how victims rate their financial (in)dependence on a five-point scale, ranging from “totally dependent” (1) to “not at all dependent” (5) on the offender.

In general, other demographic variables have not been found to be important in the existing literature on domestic violence victim reporting. Because marital status is fairly invariant in the data set, we did not include marital status as a control variable. One variable that has been found to be important in the general victim reporting literature is age (Skogan 1984; Laub 1997). Given the paucity of studies that explore victim reporting in domestic violence cases, we included victim age in the analysis. Similarly, we included a continuous measure of relationship length to account for its potential influence on a victim's decision whether to call police.

Victimization History

To account for the nature and extent of victimization prior to the experimental incident, we included one variable that captured the frequency of each of the four types of victimization in the six months prior to the experimental incident. Some researchers have found that previous calls to police may be the most predictive variable in a model of victim reporting behavior (see Berk et al. 1984). Therefore, we created a dichotomous variable capturing whether the victim had reported any incident prior to the experimental incident to police (1) or did not (0).

Experimental Incident Characteristics

The eligibility requirement of the Dade County experiment required that the incident consist of a misdemeanor-level battery, resulting in either minor or no injury. Several characteristics of the experimental incident may be relevant for future victim reporting. Using Charlotte (North Carolina) SARP Experiment data, Hutchison (1999) found that both victim and offender alcohol and drug use influenced the frequency of police utilization. Thus, we included a variable containing a victim self-report of drug or alcohol use at the time of the experimental incident. To some extent, this variable also controlled for the possibility that victim perceptions of police behavior were altered due to substance use. We included a second variable capturing victim perception of the offender's alcohol or drug use prior to the experimental incident. The data did not permit controls for alcohol and drug use during subsequent victimizations. In addition, we included a control for whether the victim reported the experimental incident to police (1) or it came to police attention in some other way (0). Aside from victim outcome preference, a specific action (arrest or nonarrest) taken by police may have some influence over whether victims choose to report subsequent victimizations. Thus, we included a variable indicating whether police did (1) or did not (0) arrest the offender following the experimental incident.

The Dade County experiment contained a second experiment that assigned victims in the original experiment to receive (or not receive) follow-up services from a special police unit (Pate, Hamilton, & Annan 1991). While not focusing specifically on domestic violence, Conaway and Lohr (1994) found that police follow-up services were associated with official reporting of subsequent victimization. Thus, we included a dichotomous variable indicating whether victims responded that they were contacted by the special police unit following the experimental incident.

Revictimization

Previous studies using similar samples, i.e., victims who had suffered one or more previous victimizations (Hutchison 1999; Johnson 1990; Berk et al. 1984), found no relationship between victim injury and reporting. It may be that the pattern of offender behavior, rather than injury in individual incidents, has greater impact on the reporting decisions of women with more extensive victimization histories. Thus, we included five variables that captured the type and frequency of victimization since the experimental incident. Four of these variables contained the frequency of each victimization type since the experimental incident. An additional variable captured whether the victim had suffered some injury as the result of being hit or hurt at any point following the experimental incident. To account for the different time periods over which victims were interviewed, we included a variable capturing the number of days from the experimental incident to the last interview.

Models to be Estimated

Estimating and evaluating separate models for each hypothesis (hereafter referred to as the process and the outcome models) is an appropriate first step in this inquiry. Estimating a single model containing all variables seems to be “getting ahead” of the existing literature, as these hypotheses have yet to be discussed comparatively. One difficult issue relates to the number of control variables, 21 in all. We ruled out the potential problem of high multicollinearity among the independent variables using the tolerance statistic test (Menard 1995), but the sample size (180) limits the number of independent variables that can be placed in a single model (Neter, Wasserman, & Kutner 1989). Consequently, we employed a method similar to that used by Hutchison (1999) in his investigation of domestic violence victim reporting behavior. We grouped the control variables in a meaningful way and conducted separate logistic regression analyses for each set of variables. We then selected the significant control variables from these grouped models and placed them in two separate theoretical models. Because the purpose of control variables is to examine the impact of theoretical variables while holding other potential influences constant, this method is not ideal. It does, however, allow some consideration of a number of factors within the restrictions placed on the analyses by a relatively small sample size. Since our goal was to test the specific predictions of the process and outcome hypotheses, we applied one-tailed significance tests to theoretical variables.

Results

Of the sample of 180 women, 61 chose to report while 119 refrained from reporting one or more victimizations suffered during the follow-up period. The simple bivariate relationships among the variables provided some support, albeit modest, for the outcome hypothesis and no support whatsoever for the process hypothesis. In our sample, 102 (57%) stated that they had received their preference of outcomes, and 77 (43%) stated that they did not (see Table 2). Consistent with the outcome hypothesis, the results show that those victims who received their preference were more likely to report subsequent victimization (39%) than those who did not (26%). While the chi-square and gamma statistics were not significant, the bivariate relationship was in the predicted direction and approached significance (p<0.057). Similarly, those who did not receive their preference were more likely to refrain from reporting (74%) than those who received their preferred outcome (61%) following the experimental incident. Overall, however, the majority of women did not report their revictimization to police.

Table 2. Bivariate Relationship Between Reporting and Outcome Hypothesis Variable

^† p<0.10.

As shown in Table 3, we found that the procedural justice scale variable was in the direction opposite that predicted by the process hypothesis. Those who chose to report subsequent victimization had a lower average value for perception of procedural justice than those who did not. Similarly, the bivariate relationships between victim reporting and the three individual variables making up the procedural justice scale variable was not in the direction predicted by the process hypothesis. For example, among victims who reported subsequent victimization, nearly 30% stated that they felt police really wanted to help, listened to them very carefully, and took their situation very seriously. This compares to around 70% on each of these items among the victims who chose not to report subsequent victimization. As with the three indicators of procedural justice, 28% of the victims who felt very satisfied with police overall reported repeat victimization to police, but a much greater share (72%) feeling very satisfied with police chose not to report subsequent incidents.

Table 3. Bivariate Relationship Between Reporting and Process Hypothesis Variables

¹ This row indicates the mean and standard deviation (in standard scores) for the procedural justice scale variable.

^† p<0.10; ^*p<0.05.

Control Variable Models

The demographic control variable model contained race/ethnicity, financial dependence, age, and relationship length. The logistic regression analysis revealed none of these demographic variables to be significantly predictive of reporting (see Table 4). The victimization history control model contained five control variables capturing elements of victimization experience in the six months prior to the Dade County experimental incident. These variables were victim reports of the frequency with which the offender hit or hurt her, hit or hurt a family member, made threats, and damaged property. In addition, this model included whether victims called police to report any of these victimizations. We found no variables to be significant at the standard 0.05 level (see Table 4).

Table 4. Logistic Regression Control Variable Models

^† p<0.10;

^** p<0.01.

The third control variable model included variables related to the Dade County experimental incident that may influence the future reporting of domestic violence victims. Two of these variables captured whether the victim stated that she or the offender had been using alcohol or drugs before the experimental incident. Another captured whether the victim was the person who had notified police following the experimental incident. The two remaining variables captured activities related to the Dade County experiment—whether the offender was arrested following the experimental incident and whether the victim received follow-up services from the special police unit. As with the previous two control variable models, none of these variables were found to be significant (see Table 4).

The final control variable model captured the characteristics of revictimization(s) victims may have suffered between the experimental incident and the final victim interview. Specifically, this revictimization model contained victim reports of the frequency with which the offender hit or hurt her, hit or hurt a family member, made threats, and damaged property. Also included in this model was a variable indicating whether the victim suffered any injury as the result of any revictimization incident. An additional variable included in this model captured the length in days from the experimental incident to the last interview. The results of this model indicated that only the frequency of being hit or hurt during the follow-up period significantly increased the likelihood of victim reporting (see Table 4).

Outcome and Process Hypotheses Models

Because the only significant variable from the control variable models was the frequency of the victim being hit or hurt during the follow-up period, this variable is the sole companion of the theoretical variables in the process and outcome models. As shown in Table 5, the results of the logistic regression analysis support the outcome hypothesis. Victim preference of outcome was a significant predictor (employing a one-tailed test) of victim reporting behavior, and the relationship was in the predicted direction.¹ The lone significant control variable, the frequency of being hit or hurt, remained an important predictor of victim reporting behavior. However, the results of the logistic regression do not support the process hypothesis. The procedural justice scale variable was a significant predictor (employing a one-tailed test) but influenced reporting in a direction opposite that predicted by the process hypothesis (see Table 5). The frequency of being hit or hurt remained a significant predictor of reporting behavior.

Table 5. Logistic Regression Outcome Hypothesis and Process Hypothesis Models

^* p<0.05;

^† p<0.10;

^** p<0.01.

Given the exploratory nature of this research, the direction of the procedural justice variable warrants further investigation. For the procedural justice scale variable and each of the three individual procedural justice variables, ratings of police fairness were consistently lower among those who called police to report revictimization than those who did not. Why might those who feel more fairly treated by police refrain from using them again and those who feel unfairly treated by police make greater use of them? The answer to this question is not readily apparent. One possible explanation relates to social desirability and the timing of the revictimization. One could argue that nonreporting victims inflated their ratings as a form of compensation for “disappointing” the interviewers by not calling police in the subsequent incident(s) discussed in the initial interview. This explanation is unlikely because questions relating to police ratings came before questions about revictimization and reporting in the initial interview. The scripted introduction to the interview did not state that questions would be asked about subsequent reports to police. Thus, it is unlikely that nonreporting victims engaged in this type of inflation.

In the Portland, Oregon Domestic Violence Experiment, researchers found that victims rated the performance of responding police officers higher when these officers were followed by specialized officers providing follow-up services (Jolin and Clavadetscher 1995). It is possible that such a “halo” effect around the officers responding to the experimental incident was produced when victims received contact from Dade County's specialized police unit. This explanation is unlikely because a t-test revealed no significant difference between the mean ratings of victims who received and did not receive these services (t=0.973, p=0.332).

Another possibility is that victims receiving their preferred outcome are likely to rate police fairness higher than victims who do not. Those victims who stated that they received their preference of outcomes perceived a higher level of procedural justice (N=102, M=0.1021) than victims who did not (N=77, M=−0.1283). However, because victim preference is positively associated with reporting, a relationship between victim preference and procedural justice ratings cannot account for our unanticipated negative procedural justice outcome.

To explore other possibilities, we conducted an ordinary least-squares regression analysis using the procedural justice scale variable as the dependent variable. We incorporated several categories of variables that could influence procedural justice judgments, including demographics, victimization history, and characteristics of the experimental incident. In addition to the variables already discussed, we included an additional variable that captured whether the victim resided with anyone other than the offender. The data set did not identify these other residents, who may have been children, parents, friends, boarders, or other relatives. Victim satisfaction may also be an important predictor of procedural justice judgments and was also included in the model. We conducted diagnostics on the independent variables and ruled out high multicollinearity.

According to the linear regression analyses, procedural justice judgments were significantly related to several variables (see Table 6). Race and ethnicity were predictive, in that white and Hispanic victims perceived a significantly lower level of procedural justice from police than did black victims. In addition, victims who felt more satisfied with police perceived a significantly higher level of procedural justice than did victims who felt less satisfied.

Table 6. Linear Regression of Variables Associated With Procedural Justice Judgments

^* p<0.05;

^† p<0.10;

^** p<0.01.

For purposes of exploration, two other variables are noteworthy particularly because of the negative direction of their effects (although failing to meet a strict p<0.05 criterion). Victims who indicated less financial independence from the offender gave police higher fairness ratings than victims who felt more financially independent. In addition, victims who were injured as a result of revictimization perceived a lower level of procedural justice than victims were not injured from a revictimization incident. It is possible that these five variables, in combination, may be responsible for the negative association between procedural justice judgments and victim reporting. To explore this possibility, we estimated a final model. We included these variables with those already identified as predictive of victim reporting in the separate tests of the process and outcome hypotheses, after ruling out the existence of high multicollinearity (see Table 7).

Table 7. Logistic Regression Results of Process Hypothesis Model Containing Variables Associated With Procedural Justice Judgments

^* p<0.05;

^** p<0.01.

The analysis revealed that, even while controlling for the effects of other variables associated with procedural justice ratings, the counter-theoretical finding could not be eliminated.² Yet further exploration of the process hypothesis produced an interesting result for the outcome hypothesis. Controlling for the influence of procedural justice judgments, victim preference became a more important predictor of reporting behavior. The original outcome hypothesis model revealed that the odds of victim reporting were increased by 79% when victims received their preference of outcomes (see Table 5). The “combined” model containing both procedural justice and victim preference variables revealed that the odds of victim reporting were increased by 108% when victims received their preference of outcomes (see Table 7). This finding suggests that omitting the procedural justice variable, along with other variables related to procedural justice, had a suppression effect on the victim preference variable.

Discussion and Conclusions

The goal of this research was to conduct an exploratory investigation of the often-asserted link between domestic violence victims' perceptions of police behavior and their subsequent utilization of police in response to revictimization. Results indicate that victims were more apt to call police upon further victimization when police had acted in a manner consistent with their preferences (i.e., the outcome hypothesis was supported). Conversely, we found no support for the process hypothesis. Instead, we found that victims who felt unfairly treated by police were more likely to call them in the future, compared to victims who felt fairly treated. In exploring this surprising result, we found that victims who felt more satisfied with police rated procedural fairness higher than victims who felt less satisfied. Satisfaction was not related to victim reporting and did not explain the counter-theoretical findings.

Holding a host of factors constant, including preference, white and Hispanic victims perceived themselves to be less fairly treated by police than did black victims; yet once these differences were controlled, procedural justice was still negative and significantly related to victim reporting of subsequent victimization. Given that our missing data were patterned by race (whites were significantly less likely to participate in the initial interview than blacks and Hispanics), we are hesitant to speculate about race/ethnicity effects in our models. If, however, we have discerned real race differences, our findings are inconsistent with both procedural justice and distributive justice predictions (i.e., blacks should perceive less procedural justice than whites and procedural justice should positively affect victim reporting).

Why might black victims perceive greater procedural justice than whites? One possibility is that black women (especially in poor communities) rely more on police to intervene in domestic violence incidents than do whites (Miller 1989). Even though the black community may view police as racist and oppressive, police may be one of the few institutional resources that low-income black women have at their disposal (Rasche 1988). White victims, on the other hand, have more resources available to them, which lessen their dependence on police. Hispanics, due to cultural circumstances (including communication barriers, fear of immigration authorities, and religious teachings), may be less able to communicate with police and more afraid to report victimization to them (Bonilla-Santiago 1996).

Given that black victims are significantly more likely to perceive procedural justice in their encounters with police than white or Hispanic victims, what can account for the negative relationship between procedural justice and reporting? This too may also be race-related. Although black women may perceive higher levels of procedural justice than other victims, they may find themselves in a gender-race quandary, caught between loyalty to one's minority group (and partner) and self-preservation (Daly and Maher 1998; see also Richie's discussion of gender entrapment, 1996:61–3). Thus, the group most likely to perceive procedural justice in their encounters with police (blacks) are also not as likely to call police for help.

Clearly, there is a need for more research focusing on the process hypothesis. As noted, the present sample suffers from a considerable missing data problem, and the extent to which our findings are impacted is unknown. But at a minimum, this work calls into question untested assumptions about victim reporting behavior. If these counter-theoretical findings are replicated, domestic violence researchers will need to carefully consider its present use. Theoretical work should focus on why unfair, rather than fair, treatment by police may be positively associated with domestic violence victim reporting or whether this is a spurious relationship explained by other factors. This task may require further exploration of whether procedural justice is the best operationalization of fair treatment or if some other measurement is more applicable in the special case of repeat victims of domestic violence.

While our exploratory study supported the outcome hypothesis, the results suggest that its articulation in the extant domestic violence literature may be somewhat simplistic. Victim preference has been very narrowly defined according to arrest versus nonarrest dichotomies, but victims may be interested in a host of additional outcomes. And contrary to unicausal statements about victim preference, our findings suggest that other factors may influence reporting as well. One explanation for the simplistic articulation of both the process and outcome hypotheses is that they have largely been framed as part of a public policy debate about what law enforcement should do in response to domestic violence (e.g., Hart 1993; Stark 1993; Miller 1989). Some factors that may be related to victim reporting behavior cannot be influenced by law enforcement policy and thus have not been the focus of attention. The policy debate has largely centered on the impact of mandatory arrest policies that restrict police discretion in the use of arrest in domestic violence cases. Because this restriction bars consideration of victim preference, critics of mandatory arrest argue that victims are disenfranchised (e.g., Gelles 1993; Buzawa et al. 1992; Miller 1989). Supporters of mandatory arrest policies would counter with concerns about the vulnerability of victims (and their preferences) to coercion and danger posed by the offender (e.g., Stark 1993). In this view, a potential increase in future reporting gained by giving active consideration to victim preferences would be seen as an inadequate tradeoff for fewer arrests of domestic violence offenders. Clearly, policy implications stemming from the replication of our findings raise complicated issues.

The present study provides one of the first pieces of evidence on whether and how victim perception of police behavior, both toward victims themselves and on their behalf, may influence future reporting. While this investigation found evidence in favor of the outcome hypothesis and counter to the process hypothesis, this research is exploratory and tentative. Continued empirical examination is vital to the understanding of how police may influence the reporting behavior of domestic violence victims and whether such influences may account for counter-theoretical findings in other domestic violence research.