Does the Criminal Justice System Treat Domestic Violence and Sexual Assault Offenders Leniently?

We examine whether men who physically assault their female partners or who commit sexual assault receive more lenient treatment than offenders who commit other types of assaults. Analyses of the National Violence Against Women Survey do not support these hypotheses. Rather, they suggest that women who assault their male partners are particularly likely to avoid arrest. In addition, both men and women who assault partners are more likely to avoid conviction than other offenders. Evidence suggests that there has been an increase over time in rates of arrest and conviction for partner violence.


Introduction
Some scholars and activists have criticized the criminal justice system for being too lenient in its response to assaults on wives and female partners (e.g., Belknap, 1995;Buzawa & Buzawa, 2002;Byrne, Kilpatrick, & Howley, 1999;Ferraro, 1989;Koss, 2000;Oppenlander, 1982;Stalans & Finn, 1995). Police officers are said to be hesitant to make arrests in these cases for Richard B. Felson is Professor of Crime, Law, and Justice and Sociology at The Pennsylvania State University. In his current research, he examines race and regional variation in violence and the use of firearms. Other recent work is concerned with the role of alcohol and drugs in violent crime. His latest book is Violence and Gender Reexamined (American Psychological Association, 2002). Paul-Philippe Pare is assistant professor in the Sociology Department at the University of Western Ontario. He received his Ph.D. from the Pennsylvania State University in 2006. His current research interests include cross-national criminology, the sociology of the criminal justice system, and the sociology/social psychology of violence. Correspondence to: Richard B. Felson, The Pennsylvania State University, Crime, Law, and Justice and Sociology, 1012 Oswald Tower, University Park, PA 16802-6207, USA. E-mail: rbf7@psu.edu various reasons, including a belief that domestic violence is a private matter, tolerance of marital violence, sexist attitudes, and a lack of confidence that victims will assist in prosecution (e.g., Belknap, 1995;Saunders & Size, 1986). Prosecutors have been described as reluctant to file charges and pursue strict sentences against violent husbands (Byrne et al, 1999). Criminal justice officials, in general, have been criticized for blaming the victim instead of the offender (e.g., Hart, 1993). Because of these processes, it is asserted, the men who commit these assaults are likely to avoid arrest and criminal prosecution. As a result, women who are assaulted by their male partners are less likely than other victims to receive protection from the criminal justice system.
These criticisms have had a major impact on the criminal justice system. The critique of police handling of domestic violence influenced the passage of mandatory and proarrest laws in the mid-1980s and 1990s (e.g., Bachman & Saltzman, 1995;Sherman, 1992). Whether these changes have affected the likelihood of arrest and prosecution is unclear. According to Buzawa and Buzawa (2002), mandatory and proarrest laws have had only a minor impact on police behavior (see also Dugan, 2003;Felson & Ackerman, 2001;Ferraro, 1989).
Scholars and activists have also criticized the criminal justice system for its lenient treatment of sexual assault offenders, particularly those who victimize people they know (e.g., Allison & Wrightsman, 1993;Belknap, 2001;Searles & Berger, 1995). They argue that police, prosecutors, judges, and jurors are overly skeptical of testimony from rape victims and that they tend to blame them for the crime rather than the offender (e.g., Belknap, 2001;Stanko, 1985). The police are reluctant to arrest, prosecutors are reluctant to prosecute, and judges and juries are reluctant to convict.
The rape reform movement influenced state legislatures in the 1970s to introduce laws designed to make conviction for sexual assault easier (Horney & Spohn, 1991). These laws eliminated the requirement that the victim resist her attacker in order to demonstrate lack of consent, eliminated the rule requiring corroboration of the victim's testimony, and placed restrictions on the introduction of evidence of the victim's prior sexual conduct. The effects of rape reform laws on conviction rates have been minimal according to some research (Horney & Spohn, 1991;Polk, 1985).
An examination of the leniency question requires a comparison of legal outcomes for cases involving different types of offenses, offenders, and victims. A comparative approach is important because offenders in general often avoid legal sanctions. It is important to know whether they avoid legal sanctions more often for certain crimes than others. We examine three legal outcomes: arrest, conviction, and incarceration. We attempt to determine whether these outcomes depend on the gender and the social relationship of the offender and victim, and whether the offense involves a sexual or physical assault. Before reviewing prior research, we discuss various theoretical perspectives that are relevant to the leniency question.

Theoretical Perspectives
Some scholars interpret leniency toward offenders who assault their partners or commit sexual assault as a form of gender discrimination (e.g., Belknap 1995;Buzawa, Austin, & Buzawa, 1995;Dobash & Dobash, 1979). They argue that these crimes are not prosecuted vigorously because the victims are typically women, the offenders are typically men, and women have a lower status than men. 1 From a conflict perspective, victims who are members of groups with low status have less legal recourse than victims who are members of higher-status groups. Victims also receive less consideration when the offender is from a higher-status group, particularly when they have a higher status than the victim (Black, 1976).
According to another formulation of the gender discrimination thesis, only women who violate gender roles are subject to discrimination by the criminal justice system (Chesney-Lind, 1978;Visher, 1983). Those who conform to gender roles are treated well, and may even be given preferential treatment as a reward. They receive chivalrous treatment for being compliant, while women who violate role expectations get worse treatment. The process could affect the consideration given to female victims as well as the treatment of female offenders. 2 When female victims conform to gender roles, the criminal justice system may give them special protection and treat those who offend against them more harshly. On the other hand, it may be lenient toward offenders who assault women who are violent, who use alcohol or drugs, or who are promiscuous, since they are violating gender roles. A statistical interaction between the gender and deviant behavior of the victim is implied.
Alternatively, it may be that the criminal justice system treats offenders who target deviant victims less harshly regardless of gender. Officials may perceive deviant victims as less credible witnesses, as more blameworthy, and as less deserving of sympathy (Elwork, Sales, & Suggs, 1981;Frohmann, 1991;Myers & Hagan, 1979;Sales, 1981). From this perspective, the deviant behavior of the victim has only main effects.
Gender discrimination is not the only explanation for why domestic violence might receive different treatment than other forms of violence. The treatment of domestic violence may reflect the nature of the crime rather than the gender of the participants. From the point of view of focal concerns theory (see, e.g., Steffensmeier, Ulmer, & Kramer, 1998), sentencing for offenses reflects three major factors: judgments of the blameworthiness of the offender; the desire to protect the community; and practical constraints and consequences. 3 These 1. This is analogous to the argument that offenders who commit street crimes are treated more harshly than offenders who commit "white collar crimes" because the former are more likely to be poor and from minority groups. 2. One could argue that any woman who has committed a crime has violated gender roles regardless of whether she otherwise conforms. An examination of the effects of the victim's gender may therefore be a better test of this hypothesis. 3. We are not able to test the focal concerns approach since we have no measures of these concerns. We use the approach only to organize the discussion. factors may also apply to arrest and conviction. First, the police and the courts may assign less blame to domestic violence offenders if they view the offense as a "crime of passion," or if offenders are more likely to be intoxicated and they treat alcohol as a mitigating factor (see, e.g., Dawson, 2004;Warner, 1996). They may assign less blame to offenders who are responding to the victim's infidelity or other provocation (Warner, 1996). Regarding their interest in protecting the community, criminal justice officials may consider domestic violence a private matter and less of a threat to public safety than assaults on strangers (see, e.g., Dawson, 2004;Pleck, 1987). They may consider an offense by anyone the victim knows as less of a threat to the community than an offense committed by a stranger. Finally, there may be practical constraints and consequences that lead the criminal justice system to treat domestic cases more leniently. The evidence may be weaker if the victim refuses to testify against the offender or if the only other witnesses are children (Warner, 1996). The criminal justice system may also be lenient because they believe that incarceration will produce family hardship. For example, the victim and children may suffer financially if the offender is arrested or incarcerated.
Leniency toward offenders who sexually assault people they know may also reflect these focal concerns. Court officials may assign more blame to the victim, and less blame to the offender, if they believe that the victim's behavior played a causal role. Their response to these crimes may reflect their response to any type of crime involving people who know each other: they may consider these offenders less of a threat to the community than offenders who assault strangers. Finally, it may be that these sexual assaults are difficult to prosecute because of practical constraints and consequences associated with the nature of the crime. According to Bryden and Lengnick (1997), there are special evidentiary concerns for sexual assaults involving people who know each other. There is often a lack of physical evidence, an absence of witnesses other than the victim, and ambiguity regarding the issue of consent (see also Felson, 2002).
To address these issues, it is necessary to disentangle the effects of the victim's and the offender's gender, victim-offender relationship, and crime type when examining legal outcomes. The discrimination perspective suggests that the gender of the offender and victim affects legal outcomes. On the other hand, it may be that those who assault intimate partners are less likely to be arrested than other offenders regardless of gender. Or it may be that offenders who assault anyone they know receive more lenient treatment than those who assault strangers, regardless of gender and regardless of whether the offense involves sexual behavior.

Prior Research on Legal Outcomes
We first review past research on the effect of the victim's relationship to the offender on legal outcomes. We then discuss the literature on gender effects, the victim's deviant behavior, and differences in legal outcomes for sexual and physical assault.

Victim-offender relationship
The evidence regarding whether the police are more likely to make an arrest for partner violence than other forms of assault is unclear. Some studies suggest that the police show greater leniency in cases of partner violence (Avakame, Fyfe, & McCoy, 1999;Black, 1971;Buzawa et al., 1995;Fyfe, Klinger, & Flavin, 1997). For example, Avakame and Fyfe (2001), using the National Crime Victimization Survey, found that men who assaulted their female partners were less likely to be arrested than men who assaulted other women. On the other hand, some studies find that the police are more likely to make an arrest for partner violence (Klinger, 1995;Smith, 1987;Smith & Klein, 1984), and some show no difference (Feder, 1998;Oppenlander, 1982).
It is important to examine the full range of victim-offender relationships when studying the tendency for the police to make an arrest. The police may respond differently to violence when it is committed by partners, other family members, other nonstrangers, or strangers. According to Black (1976), the closer the relational distance between adversaries, the less likely the criminal justice system will intervene. In addition, they may use more discretion when the offense is a misdemeanor. A recent study based on the National Crime Victimization Survey incorporated these features (Felson & Ackerman, 2001). It found that in cases of minor physical assaults, the police were less likely to make an arrest when the suspect was an intimate partner than when the suspect was an identifiable stranger. However, the police were more likely to make an arrest when the suspect was an intimate partner than they were when the suspect was someone else the victim knew. The authors suggested that the greater tendency to arrest violent partners might reflect the impact of mandatory arrest laws for domestic violence.
Most research on conviction and other legal outcomes does not distinguish between partners and other non-strangers. An exception is Dawson's (2004) study of legal outcomes for homicide cases in Toronto, Canada. She found that offenders who killed intimate partners were just as likely to be convicted as other offenders. However, prior to 1984, they were more likely to receive a reduced charge and a lighter sentence. After 1984, there was no evidence of leniency toward partner violence. Finally, she found that offenders who killed anyone they knew received shorter sentences than those who killed strangers. Felson and Ackerman (2001) also examined the effects of gender on arrest, controlling for victim-offender relationship and other variables. They found that the police were more likely to arrest male offenders than female offenders. They were also more likely to make an arrest when the victim was male, but only when they controlled for the fact that male victims were less likely to sign complaints. Smith (1987) found that arrest rates were higher when the disputants were both males than when the assault involved a man and woman. Finally, in a study of hypothetical scenarios involving domestic violence, police officers indicated that they would be more likely to arrest men than women and that they thought male victims were more responsible for the incident than female victims, and more in control of their actions (Finn & Stalans, 1997).

Gender
Gender also has been shown to affect sentencing. Approximately half of the fifty studies reviewed in a meta-analysis showed more lenient treatment for female offenders than male offenders, about one-quarter showed inconsistent effects, and one-quarter showed no effects (Daly & Bordt, 1995). Many studies have found that female offenders are less likely to be incarcerated than male offenders, controlling for other relevant variables (e.g., Farnworth & Teske, 1995;Steffensmeier, Kramer, & Streifel, 1993;Ulmer & Kramer, 1996). A number of studies also show that offenders who are violent toward women are treated more severely (Baumer, Messner, & Felson, 2000;Beaulieu & Messner, 1999;Bureau of Justice Statistics, 1995;Curry, Lee, & Rodriguez, 2004;Dawson, 2004;Glaeser & Sacerdote, 2000).

Victim's deviant behavior
We are not aware of any research that examines whether the effect of the victim's deviant behavior on the response of the legal system depends on the victim's gender. A number of studies have shown that the conduct of the victim does have a main effect on legal outcomes. With some exceptions (e.g., Albonetti, 1986Albonetti, , 1991, this research suggests that prosecution and conviction are less likely, and punishment is less severe, if the victim provoked the offender (Baumer et al., 2000;Kruttschnitt, 1985;LaFree 1980;LaFree, Reskin, & Visher, 1985;Myers, 1979Myers, , 1980Spears & Spohn, 1997;Stanko, 1981;Williams, 1976;Wolfgang, 1967). The literature also suggests that prosecution is less likely if there is evidence of disreputable conduct by the victim at the time of the offense (e.g., drinking, using drugs, engaging in criminal behavior) (Albonetti, 1986;Baumer et al., 2000;Frohmann, 1991;Spears & Spohn, 1997;Stanko, 1981;Williams, 1976). On the other hand, the evidence is inconsistent as to whether a victim's misconduct affects legal outcomes in sexual assault cases (Horney & Spohn, 1996;LaFree, 1980;LaFree et al., 1985).

Sexual vs. physical assault
The evidence is unclear as to whether prosecution is less likely for sexual assaults than physical assaults. Arrest rates and clearance rates are lower for sexual assault than for physical assault (Bureau of Justice Statistics, 1993;Federal Bureau of Investigation, 2000). On the other hand, conviction rates are higher for rape than for other assaults (Brereton, 1997;Bureau of Justice Statistics, 1993;Galvin & Polk, 1983;Steffensmeier, 1988). However, Myers and LaFree (1982) found that, once evidentiary strength was controlled, the rates of prosecution and incarceration were similar for defendants accused of sexual assault and other crimes. Finally, Walsh (1984) found that offenders who commit sexual assault are punished more severely than offenders who commit other felonies. Tjaden and Thoennes (2000) examined the response of the legal system to assaults by intimate partners using the National Survey of Violence Against Women. In bivariate analyses, they found that men were much more likely than women to be prosecuted for assaulting their partners. Sexual assaults were just as likely to be prosecuted as physical assault, although the conviction rates appeared to be slightly lower; no significance test was performed, however.
In sum, the literature on the effects of social relationship, gender, and crime type on legal outcomes is conflicting. The most consistent findings are that female offenders are treated more leniently than male offenders and that offenders are treated more leniently if victims have engaged in deviant behavior. The evidence is unclear as to whether those who assault partners or engage in sexual assault are treated more leniently. However, most of these studies do not disentangle the effects of gender and social relationship, do not examine a variety of social relationships, or fail to control adequately for the characteristics of the offense. In addition, none of the studies examines the effects of social relationship, gender, and crime type in a single design. It may be, for example, that offenders are treated more leniently when they know the victim in any way, regardless of whether the victim is their partner, and regardless of whether they commit sexual or physical assault.

Current Study
In this research, we examine leniency by determining whether offenders were arrested, convicted, and incarcerated. Based on the literature, we make the following hypotheses: H1: Offenders who assault partners, particularly men who assault female partners, are treated more leniently than other assault offenders. H2: Offenders are treated particularly leniently when they sexually assault people they know. H3: Offenders are treated particularly leniently when they assault women who are drinking or using drugs or were the first to use violence.
Our equations include measures for the gender of victim and offender, whether the victim was sexually or physically assaulted, and whether the offender was a partner, a family member, someone else known, or a stranger. A multivariate approach allows us to disentangle the effects of these interrelated variables and to examine statistical interactions. Hypotheses about lenient treatment of assaults by male partners and sexual assaults involving nonstrangers imply statistical interactions between gender and social relationship. On the other hand, we may observe mainly additive effects. For example, it may be that offenders who commit sexual assaults may be less likely to be convicted than offenders who commit physical assaults, regardless of their relationship to the victim. Following Black's (1976) ideas about status and relational distance, we may find main effects of gender and victim-offender relationship.
Importantly, we include extensive control variables related to the seriousness of the assault and its context. For example, we control for whether offenders injured the victim, whether they were armed, whether they had assaulted the victim before, whether the incident was victim-precipitated, and whether victims feared for their lives. These controls allow us to address the possibility that the legal system treats women's violence against their partners more leniently than men's because it is less serious or because it has a special context. Prior research suggests, for example, that male offenders cause more injuries than female offenders, regardless of their relationship to the victim (Felson & Cares, 2005) and offenders are more likely to be arrested if they injure the victim (Avakame & Fyfe, 2001;Felson & Ackerman, 2001).
Finally, we examine trends over time in legal outcomes for different crimes. By examining statistical interactions between time period and victim-offender relationship, we can determine whether responses to assaults on intimate partners or sexual assaults have changed over time. Because of the increased public attention to these crimes and the changes in the legal system we make the following hypothesis.
H4: The legal treatment of offenders who assault intimate partners or commit sexual assaults became more severe in the 1980s and 1990s.

Methods
Our analyses are based on incidents of physical and sexual assaults reported to the police in the National Survey of Violence Against Women (and Men). The survey is based on data collected in 1994-1996 from a nationally representative sample of 8,000 women and 8,000 men, age 18 and over (see Tjaden & Thoennes, 2000). A computer-assisted telephone interview was conducted with each respondent that included information about their experiences with violence as an adult.
Respondents were asked about incidents of physical assault during adulthood (although some reported assaults occurred before they were 18). Specifically, they were asked whether anyone had: thrown something at you that could hurt you? Pushed, grabbed or shoved you? Pulled your hair? Slapped or hit you? Kicked or bit you? Choked or attempted to drown you? Hit you with some object? Beat you up? Threatened you with a gun? Threatened you with a knife or other weapon besides a gun? Used a gun on you? Used a knife or other weapon on you besides a gun?
Respondents were asked about incidents of sexual assault since childhood. Specifically, they were asked Has a man or boy every made or tried to make you have sex by using force or threatening to harm you or someone close to you? Has anyone, male or female, ever made or tried to make you have oral sex by using force or threat of harm? Has anyone ever made or tried to make you have anal sex by using force or threat of harm? Has anyone, male or female, ever put fingers or objects in your vagina or anus against your will by using force or threats?
Respondents could report up to six physical assaults by different offenders and six sexual assaults by different offenders. If a victim was assaulted multiple times by the same offender, only the most recent incident was recorded. Incidents involving same-sexed partners were eliminated from analyses, because their number is insufficient for analysis (n = 30) and because their inclusion complicates the analyses of gender by partner interactions. We also excluded 259 cases with no information on the assault. To preserve sample size, we treated cases with missing data on most independent variables by constructing missing data dummy variables. For example, respondents who did not report the date of their victimization are coded 1 on the "missing decade" dummy variable and zero on the other decade variables. When less than 1 percent of the cases were missing, we omitted them (n = 417). After all of these deletions, our sample included 8,143 incidents (6,400 physical assaults and 1,743 sexual assaults). Our analyses of arrest are based on the 1,963 cases that were reported to the police. Our analyses of conviction are based on 568 cases that involved arrest. Finally, our analysis of incarceration is based on 337 cases that involved a conviction. 4 Probit regression is used to predict the three binary dependent variables. We treat the incident as the unit of analysis. Because many respondents reported more than one incident, the assumption of independence of errors across observations is violated. We use Stata's (2003) Cluster adjustment to address this issue.

Measurement
Victims were asked what the police did in response to the crime. If they indicated that the offender was arrested, taken into custody, or charged with the 4. In analyses not presented we: (1) used a Heckman selection regression to control for the nonrandom selection of cases at each stage of the legal process (Heckman, 1979;Greene, 2003); (2) included controls for whether victims waited more than 24 hours before reporting the incident to the police, and the victim's race/ethnicity, current education and income, and age at the time of the incident. In each case, the results were similar to those we present. Results are available on request. crime at any time, the incident was coded one on arrest (and zero otherwise). If the offender was charged, victims were asked "What happened to these charges? Was he/she convicted, pled guilty, acquitted, or were the charges dropped." If the respondent indicated that the offender was convicted or pled guilty, the incident was coded as one for convicted. On the other hand, the incident was coded zero if the respondent was acquitted, the charges were dropped, or the offender was arrested but not charged. If the offender was convicted, the respondent was asked "Did this conviction result in his/her being sent to jail or prison?" If respondents answered yes, the incident was coded as one for incarceration (and zero otherwise).
The coding of the independent variables is described in Table 1. The principal independent variables of interest include the relationship between the offender and victim, the gender of the offender, the gender of the victim, the type of assault, and the decade the offense was committed. Victim-offender relationship is coded as either partner (spouses, former spouses, partners, and former partners), other family (parents, children, uncle, aunt, cousin, brother, sister, "in-laws"), other known (i.e., friend, date, and acquaintance) or stranger (including people known by sight only; the reference category).
Type of assault is coded as rape (i.e., sexual assault with penetration), other sexual assault (including attempted rapes and assaults where it could not be determined whether there was penetration), and physical assault (the reference category). We distinguish rape from other sexual assaults because it is a more serious crime and because it is more likely to produce physical evidence.
Our main measures of seriousness are whether offenders injured the victim and whether they were armed. Respondents were asked "Were you physically injured during this incident?" and "Did he/she use a gun, knife or other weapon during this incident?" We also include two characteristics of the incident that are associated with seriousness: whether victims feared for their lives and how often the offender had assaulted the victim before. These variables are based on answers to the questions: "Did you believe you or someone close to you would be seriously harmed or killed during this incident?" and "How many different times has he/she done this to you?" The measures for deviant behavior by the victim include whether the victim was under the influence of alcohol or drugs during the incident, and whether the victim precipitated the incident. Victims were asked "Were you using drugs or alcohol at the time of this incident?" and "Who was the first to use or threaten to use physical force during this incident? Was it you or the other person?" We also include measures of substance use by offenders (based on victims' reports) since research shows it can increase the risk of arrest (Avakame & Fyfe, 2001;Engel, Sobol, & Worden, 2000). 5 In the incarceration model, we included a measure of whether the offender pled guilty, since this might have enabled some offenders to avoid incarceration (e.g., Steffensmeier et al., 1998). 6 Results Table 1 provides descriptive statistics for the dependent variables, while Table  2 provides descriptive statistics for the explanatory variables. Among incidents reported, we see that almost half (48.3 percent) of the offenders were arrested. In the sample of incidents that resulted in arrest, 63.6 percent of the offenders were convicted. Of those that resulted in conviction, 67.8 percent of the offenders were incarcerated. Most of the offenders (88.4 percent for reported incidents) and more than half the victims were men. 7 These percentages did not vary much by sample. As one moves to later stages of the criminal justice process, the percentage of incidents involving partners declines.
In Table 3, we present our final models. They include all additive effects and all statistically significant interactions, but omit the missing data dummy variables. In Table 4 we present all the interaction terms that we investigated in our initial equations. To avoid redundancy we do not present the additive terms. The interaction terms were entered either individually or in a block of variables (see Table 4). The interaction term for substance use is based on whether the victim was using either alcohol or drugs during the incident. A minor incident involves a misdemeanor: the offender was unarmed and did not injure the victim.
Because of an insufficient number of cases, we did not have the statistical power to examine some statistical interactions in the sample of arrested and convicted offenders (see Table 4). Because of concerns about statistical power 5. If the victim or the offender were using both alcohol and drugs during the incident, the information was coded as using drugs. 6. Approximately 38 percent of defendants pled guilty. 7. Analyses not presented reveal that 13 percent of the incidents reported to the police involving partner violence (75 out of 583) were committed by a female offender. in detecting other interactions, we used a less stringent significance level (.10) in our initial models. In two instances, we observed borderline interactions, but they became significant at the conventional level (.05) in the final models, where they were entered individually, not as part of a block of variables.

Arrest
The results from the equation for arrest are presented in columns 1 and 2 of Table 3. They show that the offender's relationship to the victim is unrelated to arrest. We do observe a strong gender × partner interaction, but it does not support the hypothesis that men who assault their female partners are particularly likely to avoid arrest (see Figure 1). Rather, it suggests that it is women who assault their male partners who are particularly likely to avoid arrest. For example, the predicted probability that a woman will be arrested for assaulting her male partner is about .15, compared to about .38 for a man assaulting his female partner. 8  Table 2. The probability equals the return of the standard normal cumulative distribution of z, where z = (b0 + b1x1 + b2x2 + …). Estimates are based on incidents involving: female victim (reverse for female partner), male offender (reverse for female partner), physical assault, no alcohol or drug use by victim or offender, no weapon, physical injury, during the 1980s, no prior incident.
We also observe a significant statistical interaction between whether the incident involved a partner and whether it occurred in the 1990s. As predicted, arrests for partner violence increased in the 1990s (see Figure 2). This evidence suggests that the institution of mandatory arrest laws had an effect. We see no evidence that arrests for partner violence significantly increased in the 1980s, however (see Table 4). In addition, we find no evidence suggesting that incidents involving sexual assaults were more likely to result in arrest after 1980.
8. It is difficult to interpret the relative strength of probit coefficients because they are not easily transformed into odd ratios. We indicate the strength of some our effects using predicted probabilities. The interaction terms involving decade and type of assault are statistically insignificant (see Table 4).  Table 2. Estimates are based on incidents involving: female victim, male offender, physical assault, no alcohol or drug use by victim or offender, no weapon, physical injury, no prior incident.
We also observe an interaction between the seriousness of the offense and the offender's relationship to the victim. The results suggest that, for minor offenses, the police were less likely to arrest offenders who assault their partners and others they know than those who assaulted strangers. A similar pattern is observed for offenders who assault other family members, but the interaction term is smaller and not statistically significant. We observe no support for the hypothesis that offenders are particularly likely to avoid arrest if they assault deviant women. We did not observe statistical interactions between the victim's gender and either substance use or victim precipitation. We do find main effects of alcohol use: offenders were more likely to be arrested if they were drinking or using drugs, and they were less likely to be arrested if the victim was drinking.
The evidence also indicates that the police are more likely to make an arrest for rape and sexual assault than for physical assault. We observe no support for the hypotheses that offenders who sexually assault nonstrangers or who assault victims who are drinking or using drugs are particularly likely to avoid arrest.

Conviction
The results in Table 3 (columns 3 and 4) show a strong main effect of partner, and minor × partner and decade × partner statistical interactions. The results indicate that offenders are much less likely to be convicted if they assault a partner (vs. a stranger), particularly if the assault is minor. 9 In addition, 9. Analyses not presented show that the coefficient for partner violence is significantly larger than the other known category and the other family category.   Table 2. The probability equals the return of the standard normal cumulative distribution of z, where z = (b 0 + b 1 x 1 + b 2 x 2 + …). Estimates are based on incidents involving: female victim (reverse for female partner), male offender (reverse for female partner), physical assault, no alcohol or drug use by victim or offender, no weapon, physical injury, during the 1980s, no prior incident.  Table 2. Estimates are based on incidents involving: female victim, male offender, physical assault, no alcohol or drug use by victim or offender, no weapon, physical injury, no prior incident.
convictions for partner violence increased in the 1980s and 1990s (see Figure 3). For example, before 1980, the predicted probability that a male offender would be convicted for assaulting his female partner was about .25, compared to almost .75 for assaulting a female stranger. In the 1990s, the corresponding predicted probabilities were both just above .60.  Table 2. Estimates are based on incidents involving: female victim, male offender, physical assault, no alcohol or drug use by victim or offender, no weapon, physical injury, no prior incident.
We see no evidence of an interaction between the victim's gender and deviant behavior. Offenders who assault women who were drinking or on drugs, or who were responding to a violent provocation are just as likely to be convicted as offenders who target conforming women. We do observe a main effect of alcohol use by victims, however: Offenders are less likely to be convicted when victims were drinking during the incident.
We do not find support for the hypothesis that offenders are less likely to be convicted if they sexually assaulted someone they knew (see Table 4). Nor do we observe evidence of any main effects of sexual assault: offenders accused of rape or sexual assault were just as likely to be convicted as offenders accused of physical assault. Finally, offenders who used a firearm were more likely to be convicted that those who did not.

Incarceration
The results (Table 3, columns 5 and 6) suggest that offenders who assault people they know are much less likely to be incarcerated than those who assault strangers. The coefficient for other family members is of borderline significance  Table 2. Estimates are based on incidents involving: female victim, male offender, physical assault, no alcohol or drug use by victim or offender, no weapon, physical injury, no prior incident.
(p = .08), but note that the number of cases is quite small. Offenders who assault their partners are slightly less likely to be incarcerated than those who assault family and others they know, but the differences do not approach statistical significance (analyses not presented). It is possible that we do not have the statistical power to detect these differences. Consistent with the literature, male offenders are much more likely to be incarcerated than female offenders. Offenders who commit rape are also more likely to be incarcerated than those who commit physical assaults. We do not observe the predicted interactions. We observe no support for the hypotheses that offenders are more likely to avoid incarceration if they sexually assault people they know or if they assault a woman high on alcohol or drugs (see Table 4). We do observe one unexpected interaction: offenders who committed minor assaults against people they knew were particularly likely to be incarcerated.
The control variables also have some effects. Offenders are more likely to be incarcerated if they were using drugs at the time of the offense, or if they caused the victim to fear serious injury or death. Surprisingly, the effects of the offender's use of a weapon and injury to the victim are not statistically significant. The effect for gun use, however, is mediated by the level of fear and the guilty plea variables. When these variables are left out of the equation, offenders who used a gun were more likely to be incarcerated (b = .501; p = .02). Finally, offenders are much less likely to be incarcerated if they pled guilty.

Discussion
Our results suggest that the criminal justice system is not particularly lenient toward men who assault their intimate partners or who sexually assault people they know. We do observe some interesting patterns, which we discuss in detail below. We first consider the evidence regarding arrest and then consider the evidence regarding the response of the courts.

Arrest
Our evidence suggests that the police show leniency toward offenders who assault their partners under two conditions. First, the police are particularly unlikely to arrest women who assault their male partners. We suspect that the pattern reflects the focal concern of protecting the community, particularly the victim. The police assume that the man can protect himself from his female partner and that a woman's violence is not dangerous unless she assaults someone other than her partner. It may also be that the police are chivalrous toward female offenders unless they are a threat to outsiders.
Second, the police are less likely to arrest offenders who engage in minor assaults against their partner or other people they know (vs. strangers). Felson and Ackerman (2001) have reported a similar pattern with data from the NCVS. It may that the police are more likely to consider minor forms of violence between people who know each other as less of a threat to the community. 10 It is noteworthy that their lenient treatment is not restricted to intimate partners.
We also find evidence that the likelihood of arrest for partner assault increased in the 1990s. These findings suggest that mandatory arrest laws did have their intended effect on the response of the police to partner violence. By the 1990s, the police were more likely to arrest offenders who assaulted their partners than other offenders. Thus, our results suggest that the police became more punitive toward men who assaulted their partners, whereas they had treated them like other offenders earlier. They continued to be lenient toward women who assaulted their male partners, but not as lenient as before.
Mandatory arrest was a policy instituted because of the belief that the police were not arresting enough violent husbands (and male partners). An evaluation of whether they were (or are) too lenient in these cases depends on one's expectations. One might conclude that the police were not too lenient toward violent husbands using a standard of equal justice. On the other hand, there are other possible standards that might lead one to prefer a more punitive response to violent husbands than violent wives. For example, one might argue that the police should arrest violent husbands more often than violent wives, even if they commit the same offense. Violent husbands tend to be more dangerous than violent wives: they are usually bigger and stronger, and they tend to assault with greater frequency (Felson & Cares, 2005). One could argue that protecting victims is the most important focal concern in cases of partner violence. Protecting victims and equal justice might be competing goals.
The evidence is not consistent with the idea that arrest is positively associated with the relational distance between the victim and offender, as suggested by Donald Black (1976). The likelihood of arrest is no greater for assaults involving strangers than assaults involving other family members or other people who know each other. However, there may be offsetting effects. The availability of an identifiable suspect is an important determinant of the police's ability to make an arrest (Greenwood, Chaiken, & Petersilia, 1977;Felson & Ackerman, 2001). Non-strangers are much easier to identify and locate than strangers, and partners and family members are more likely to be present when the police arrive (Buzawa et al., 1995). Therefore, any reluctance to arrest offenders whom the victim knows may be offset by the fact that it is easier to make an arrest when the offender is someone the victim knows. In other words, perhaps domestic violence offenders would be treated more leniently were it not for the fact that they are so easily identified and located. 11 10. Note that we cannot determine whether police behavior reflects administration policy or the preference of individual officers. 11. Our results may be affected by evasive actions by the offender to avoid arrest and conviction. For example, offenders who commit more serious offenses may be more concerned about prosecution and therefore more likely to take steps to avoid it. Perhaps this factor offsets some of the effects of injury and firearm use on legal outcomes.
The evidence does not support the hypothesis that offenders who sexually assault people they know are able to avoid arrest. The police show no greater leniency in cases of sexual assault than in cases of physical assault, regardless of the victim-offender relationship. In fact, offenders who commit sexual assault are more likely to be arrested than those who commit physical assault. The analysis of interactions by decade indicates that the relative likelihood of arrest for sexual assault did not change in the 1980s or 1990s.
Finally, the evidence does not support the hypothesis that offenders who assault women who violate gender roles are treated more leniently. We observe no evidence that offenders who assault intoxicated or violent women or who sexually assault intoxicated victims receive more lenient treatment. Offenders are less likely to be arrested when victims are drinking, but the effect does not depend on gender. The evidence supports the literature cited earlier suggesting deviant victims are perceived as less credible witnesses, as more blameworthy, and as less deserving of sympathy. In the language of the focal concerns perspective, the deviant behavior of the victim affects judgments of blameworthiness.

Conviction and Incarceration
Offenders who assaulted their partners before the 1980s were much less likely to be convicted than other offenders, regardless of gender. Leniency in conviction, however, largely disappeared in the 1980s and 1990s. This change supports the idea that public concerns regarding partner violence affected the response of the courts.
We did not find that offenders who assaulted partners were particularly likely to avoid incarceration, although it is possible that we did not have the statistical power to detect it. We did find that offenders convicted of assaulting anyone they knew were less likely to be incarcerated than offenders who assault strangers. We can think of three possible explanations for the pattern: (1) assaults on non-strangers are less likely to be viewed as a threat to the community than assaults against strangers; (2) victims are more reluctant to cooperate with the prosecution or are less credible as witnesses when the offender is someone they know; (3) court officials assign less blame to offenders who assault people they know because they are more likely to view these offenses as crimes of passion.
We find no evidence supporting the hypothesis that offenders who sexually assault people they know are particularly likely to avoid conviction. The results do not support the idea that these types of offenses are particularly difficult to prosecute because of evidentiary concerns or gender discrimination. Nor do we observe main effects of crime type: offenders who have committed sexual assault are just as likely to be convicted as offenders who have committed physical assault. Once convicted, offenders who commit rape are more likely to be incarcerated than those who commit physical assault. Note that higher incarceration rates for rape are expected because the law dictates more severe sanctions for rape than for most forms of physical assault. Finally, we observe no evidence that changes in rape laws increased the likelihood of conviction and incarceration for sexual assault during the 1980s or the 1990s.
In general, the evidence does not support the idea that the response of the courts depends on the gender of the offender or victim. The courts are no more likely to convict male or female offenders, or offenders who victimize men or women. While we find that the courts are more likely to incarcerate male offenders than female offenders, this may reflect the fact that men are more likely to have criminal records while women are more likely to have family obligations (Steffensmeier, Kramer & Streifel, 1993). Finally, we find no evidence that offenders are treated more leniently when they assault women who have engaged in deviant conduct during the incident. The evidence does not support the notion that the criminal justice system withdraws chivalrous treatment when female victims have violated gender roles.

Limitations
In studies of discrimination, one must always be concerned about omitted variables that might be associated with both the independent and dependent variables. In this study, we were not able to control for the quality of evidence, or the offender's demeanor, social-demographic characteristics, and prior record of violence against other victims. We would argue that, with the exceptions we have already mentioned, these variables are unlikely to be related to our independent variables. In addition, in analyses not presented, we controlled for the victim's age, race, education, and income, and doing so did not affect our results. Perhaps we would have found more lenient treatment of female suspects by the police if we controlled for demeanor, since evidence suggests that female suspects are more likely to be disrespectful to police officers than are male suspects (Engel, 2003).
Finally, our analysis of social change is limited by the fact that we rely upon a retrospective analysis-some crimes occurred far in the past. Because of the nature of our data, we used the decade in which the offense occurred-a fairly crude measure of change over time. In general, our conclusions about time trends must be more tentative than our other conclusions because of the possible impact of the victim's memory. However, physical and sexual assaults are memorable events, particularly those that are reported to the police. We think it is reasonable to assume that most respondents remember objective facts such as their relationship to the offender, the gender of the offender, the decade the crime occurred, and the legal outcome. Of course, respondents were given the opportunity to say that they "don't know" in response to questions if they did not remember.
In conclusion, our results demonstrate the importance of a comparative approach in studying legal outcomes. These comparisons are important to understanding the response of the criminal justice system to sexual assault, and violence against women or intimate partners. Most research studies these offenses in isolation. Only a comparative approach reveals if the criminal justice system's response to these crimes is special or not.