Violence against women is a major social problem and human rights violation, and femicides are the most severe form of it. According to the UN statistics, a third of all femicide victims globally (34%) are murdered by their intimate partners (UNODC 2019).Previous studies reveal that judges tend to perceive violence between partners as more acceptable than other types of violence. This results in lenient sentences in comparison with violence outside intimate relationships (Black 1976; Dawson 2004, 2006). There are different explanations of this pattern; most of them are rooted in theories of focal concerns and bounded rationality. These theories posit that judges take into account three key issues: offender culpability, public protection, and the practical constraints of the court (Steffensmeier et al. 1998). But in conditions of limited time and information, they rely on stereotypes to make their decisions about these issues (Albonetti 1991). In the case of intimate partner homicides, one of the stereotypes is that such crimes imply some degree of victim involvement, mitigating the defendant’s culpability (Dawson 2006; Dawson & Sutton 2017).Current empirical studies show that perceptions of the victim’s lifestyle and behavior may vary for male and female observers, but the direction of these differences is still under discussion. While many studies show that it is more common for men than for women to blame female victims (Bongiorno et al. 2020; Bryant & Spencer 2003; Davies et al. 2009; Grubb & Harrower 2009; Schult & Schneider 1991; Ulybina 2020; Witte et al. 2006; Yamawaki et al. 2009), some research identifies no association between the observers’ gender and perception of victim’s behavior (Abrams et al. 2003; Girard & Senn 2008; Hillier & Foddy 1993; Kern et al. 2007; Krahe 1988; Landström et al. 2016; Persson et al. 2018; Qiet al. 2016). Using the example of the victim’s history of alcohol abuse, this paper aims at contributing to this field of research by answering the following questions: RQ1: Does the victim’s abuse of alcohol affect sentencing in intimate partner femicides? RQ2: Does the effect of the victim’s abuse of alcohol differ for male and female judges in its direction?We focus exclusively on cases of female victims, as the perception of alcohol abuse by women is more strongly associated with a violation of traditional gender norms (Iwamoto et al. 2018). In many societies, including Russia, women’s alcohol consumption is viewed as deviant and inappropriate, which makes it a more potent trigger for victim-blaming attitudes (Sims et al. 2007). Therefore, the social meaning and evaluative load of alcohol abuse are not symmetrical across genders, justifying our decision to limit the analysis to female victims.

The novelty of the present study is twofold. First, it is the first empirical research on intimate partner violence in Russia, based on nonreactive court decision data1. Most of the academic studies of domestic and intimate partner violence are devoted to American, Canadian, or Australian contexts with little attention to other countries because of the general lack of available criminal data (Garcia-Moreno et al. 2003; Johnson et al. 2010; Stöckl et al. 2013; Vassileva & Delpeuch 2021). Russia may be an interesting case for the international audience: according to the UN Office on Drugs and Crime, it has one of the highest rates of intimate partner femicides worldwide (UNODC 2023). Other particular characteristics include the absence of any legal framework related to gender-based or domestic violence, and the ongoing debates regarding the corresponding law. Second, previous research on domestic and intimate partner violence and perceptions of the victim’s role is mostly based on vignette experiments describing hypothetical situations and involving students as participants (e.g., Braden-Maguire et al. 2005; Bryant & Spencer 2003; Luginbuhl & Mullin 1981; Pavlou & Knowles 2001; Summers & Feldman 1984; van der Bruggen & Grubb 2014; Witte et al. 2006). Such a design does not reflect the practices and attitudes of real judges who, in comparison with students, have specific qualifications, experience of dealing with similar crimes, and whose decisions directly affect people’s lives. In this study, we use unique data extracted from texts of court verdicts on actual crimes. With such methodology, our example continues the recently emerged stream of sentencing research that uses text mining2 approach (see Pina-Sánchez, Grech, et al. 2019; Pina-Sánchez, Roberts et al., 2019).The paper is structured as follows. The first section is devoted to the literature review on factors contributing to victim blaming in violence against women and the role of alcohol abuse in such attitudes; this section ends with the statement of hypotheses. The second section describes the process of text data collection, preprocessing, and analysis.The third section presents the results of the analysis in correspondence with the proposed research questions and hypotheses. The final section discusses the obtained results, the research limitations, and ideas for further studies.

Theoretical and empirical background

The role of the victim’s behavior in case outcomes is a subject of the literature devoted to the victim-blaming topic. One of the most frequently used frameworks on this issue — namely, the just world theory — assumes that individuals tend to believe that people get what they deserve, i.e., people’s actions lead to fair consequences (Lerner 1980). In case of crimes, this framework posits that people are victimized because they did not prevent the victimization or, in particular, instigated the crime to happen (Gamache 2006). Another relevant framework is the discounting principle proposed by Kelley (1973) as a part of the causal attribution processes. According to this principle, “the role of a given cause in producing a given effect is discounted if other plausible causes are also present” (Ibid., p. 113). In regard to crimes, this principle means that the victim’s behavior may be considered an additional cause for the crime, which reduces the amount of blame assigned to the aggressor.Indeed, victims’ behavior prior to the crime has been shown as one of the key factors contributing to the victim blaming in cases of violence against women (Witte et al. 2006). Thus, studies demonstrate that if the victim’s confrontation or verbal provocation predates the domestic battering, observers are prone to see the batterer as less responsible for the violence (Harris & Cook 1994; Kern et al. 2007; Pavlou & Knowles 2001; Rhatigan et al. 2011; C. Stewart et al. 2012; Witte et al. 2006). A sizable body of literature originating from Wolfgang’s work (1957) examines the so-called “victim-precipitated homicides”, i.e., homicides that were a result of victims first using physical force, weapons, etc. However, not only the direct use of violence but also other characteristics of the victim’s behavior may be associated with less responsibility put on the perpetrators. For example, if jealousy becomes a motive for battering, assuming that a woman gives a reason for a man to be jealous, more blame is assigned to her (Delgado et al. 1997; Pavlou & Knowles 2001). In fact, even minor things such as being late for the meeting may be associated with the increased blame to the victim for the consequent violence (Gamache 2006). Reviews show that the better the victim and perpetrator know each other, the more the victim is blamed (Gravelin etal. 2019; van der Bruggen & Grubb 2014), that is why in cases of intimate partner violence victim’s own responsibility for the violence becomes one of the most widespread stereotypes defining judges’ decisions (Dawson & Sutton 2017).Besides, because such conflicts imply a history of previous relationships between a perpetrator and a victim, some attributes of these relationships also lead to different attributions of victim blame. Among these attributes, for example, are the victim’s attempts to leave the abusive relationships. In cases when a prior history of violence is relevant for the family, perpetrators receive harsher sentences when there were attempts by the victim to leave the family. On the contrary, staying in such relationships increases the risk of future violence and, as a result, is perceived as a victim’s responsibility and leads to lenient punishment for perpetrators (Dawson 2003; Mahoney 1991; Rapaport 1991). However, the amount of blame given to a victim because of the previous behavior may vary depending on the crime type and severity (Koepke et al. 2014). In this regard, homicides become an interesting type of crime to examine since it implies the most severe of possible consequences for the victim (her death), which may outweigh all the prior circumstances.

In the literature, blaming of female victims is strongly associated with attitudes towards women in general. More conservative attitudes correspond to the higher level of blame to female victims (Pavlou & Knowles 2001; Stewart et al. 2012; Viki & Abrams 2002; Yamawaki & Tschanz 2005). As the theory of ambivalent sexism states, women “are rewarded with benevolent care if they embrace traditional roles” (Koepke et al. 2014, p. 447). In this regard, drinking women may be seen as those who violate gender norms (Kantor & Straus 1989). This gendered asymmetry is particularly relevant in the Russian context, where traditional norms surrounding femininity and modesty persist strongly (Bobrova et al. 2010). As a result, alcohol consumption by women is likely to be interpreted not just as a behavioral flaw, but as a moral failing or indication of deviant character. By contrast, alcohol consumption by men is often socially tolerated or even expected, making it a less salientfactor in judgments of victim blame. Hence, focusing on female victims allows us to analyze how gendered expectations and stereotypes shape judicial decisions in femicide cases.Women’s intoxication was shown to be perceived as justification or provocation for rape and sexual assault. Gravelin et al. (2019) in their review found that most of the studies that manipulated intoxication level illustrate that a female rape victim in a state of intoxication receives more blame in comparison with the victim being sober. Besides sexual crimes, studies show that drinking women are also more frequently exposed to domestic violence (see Devries et al. (2014) for a review). Kantor and Straus (1989) explain this by suggesting that intoxicated women become more verbally aggressive. In comparison with men, for women such aggressiveness does not correspond to the traditional gender roles and, as a result, women “lose the protection afforded by other traditional gender role norms, such as ‘never hit a woman’” (Ibid., p. 185) and thus legitimate the violent response.One of the possible mediators in the association between the described factors and victim blaming is the observer’s gender, which is the main focus of our study. The current empirical evidence of gender differences is rather contradictory. In a number of studies, no association was found between the gender of the observer and the level of victim blaming (Abrams et al. 2003; Girard & Senn 2008; Hillier & Foddy 1993; Kern etal. 2007; Krahe 1988; Landström et al. 2016; Persson et al. 2018; Qi et al. 2016). Other studies indicate that male respondents tend to blame female victims more than female respondents do (Bongiorno et al. 2020; Bryant & Spencer 2003; Davies et al. 2009; Grubb & Harrower 2009; Schult & Schneider 1991; Ulybina 2020; Witte et al. 2006; Yamawaki et al. 2009). This pattern is usually explained by the defensive attribution theory, claiming that with the decreasing similarity between an observer and a victim, the amount of blame increases (Shaver 1970). In this regard, men may sympathize with the perpetrator of intimate partner violence and place responsibility on female victims, especially in the presence of the provocative behavior related to intoxication. Likewise, women are prone to support female defendants or victims as they may imagine themselves in similar circumstances. As Gravelin et al. (2019) show, no studies in their review indicate that women put more blame on female victims than men.There is also a body of literature specifically on the role of gender in the judiciary; studies in the area also show inconclusive results. Some find no difference in judges’ behavior in domestic violence cases (Hunter 2008), but there are studies indicating that female magistrates in lower courts more often show courtesy to the defendants, compared to male judges (Anleu & Mack 2017). Some argue that these disparities show different gender-colored expectations in professional behavior of judges (Thornton 1996; Rackley 2012).As it was mentioned before, most of the described results are based on experimental studies utilizing vignettes (for critique of this approach see Stewart & Maddren 1997; van der Bruggen & Grubb 2014). Research on the role of victims’ behavior and lifestyle that involves real data on crimes and reflects actual court practices is scarce, and our study aims to fill this gap. In accordance with the described theories, we propose two hypotheses corresponding to the raised research questions:H1: Femicide offenders whose victims abused alcohol in the past will get more lenient punishment than those whose victims did not do it. This hypothesis stems from the just world theory or the discounting principle mentioned earlier.H2: Male judges will give more lenient punishment to femicide offenders whose victims abused alcohol in the past in comparison with female judges. This hypothesis is based on the defensive attribution theory mentioned earlier.H3: The mitigating effect of victim’s alcohol abuse on sentencing is explained by pre-homicide victim’s provocative behavior: when controlling for provocation, the coefficient for alcohol abuse loses statistical significance.This hypothesis is based on the assumption proposed by Kantor and Straus (1989) that a victim’s abuse of alcohol leads to verbal provocation and the provocation itself leads observers to put blame on the victim. The test of this hypothesis allows us to conclude which groups of factorsmatter more in femicide sentencing: long-term background factors orsituational triggers.

National context

Recent independent research demonstrated that Russia has the highest rate of intimate partner femicides in developing countries: 53% of all female homicides, which happened in 2011-2019 in Russia, were intimate partner homicides (Zhuchkova et al. 2021). In Russian society, attitudes towards domestic and intimate partner violence, which in its most serious form results in family-related homicides, are contradictory. The special law against domestic violence has been submitted to the government of the legislature at least three times since 1994, but still has not been adopted in the country. While national surveys demonstrate strong public support1 for the law against domestic violence, the official state discourse declares the values of the patriarchal family and remains in opposition to this law, considering family relations a private matter (Muravyeva 2021).

Activities of NGOs, independent media, and some politicians have brought domestic violence in Russia to the spotlight; however, the project of DV-prevention law was strongly criticized by conservative actors, specifically, the Moscow Patriarchate of the Russian Orthodox Church (Lukovitskaya 2023). Federal policy documents recognize the problem; however, they shift the discussion to the role of traditional values and gender roles, ignoring programs proposed by the UN or HRW (ibid). Moreover, in 2017, domestic violence was decriminalized, and the cor-responding article was relocated from the Criminal Code to the Administrative Code with only an up to $150 fine, 15 days of arrest, or 120 days of community service2. This caused a significant increase in demand for non-government initiatives, providing help to domestic violence victims (Bakin 2018).

Due to the absence of a legal framework related to domestic violence in Russia, it is not possible to find any reliable official disaggregated data on such crimes. Official statistics that are publicly available do not contain any indicators that crimes were committed by family members or intimate partners. These statistics are also aggregated by years, regions, crime types, etc., making it impossible to conduct an analysis on the crime-level data1. Statistics that are available upon request to the police or the court institutions usually use a very limited notion of intimate partners (e.g., cohabitants would not be included in this category). Moreover, certain authors doubt the reliability of official DV statistics due to misclassification of criminal cases, which was admitted by the Russian Ministry of Internal Affairs (Lukovitskaya 2023).

Even in cases when victims try to use judiciary channels, they face problems on every step of the way. Out of those asking for help from the police, only 4% were satisfied with the result (Ibid.). Cases of domestic violence often tend to be mitigated before going to court, but even if they doreach the court, existing measures lack prevention potential: there are no protective measures in case of stalking or harassment, and nonphysical (psychological or economical) domestic violence is unpunishable under the current Criminal Code (Glebova et al. 2021). Even after passing all thoseobstacles, the victim may face gender discrimination in court. Russia is a part of CEDAW (Convention on the Elimination of all Forms of Discrimination Against Women) and declares gender equality in the Constitution, but in domestic violence cases courts tend to be conservatively biased:victims face bureaucratic difficulties when forced to pursue the case on their own, get mocked in the media or even publicly and denied shelter for the time of judicial procedure (Chamusco 2017).

Previous studies of domestic violence and intimate partner femicides in Russia are limited due to several reasons discussed above: the sensitive nature of the topic, poor availability and quality of the data, and overall inconspicuousness of domestic violence in political discourse. Existing literature underlines long-term effects of domestic violence on victims and observers (Borisov et al. 2020), deproblematization of DV in the Russian media (Kazun et al. 2022), and specifics of domestic violence in Islamic regions of Russia (Sirazhudinova 2017; Gabaraeva 2021). However, most empirical research relies either on qualitative methods with limited potential for generalization or touches upon general legal documents, giving little knowledge about the current state of DV and femicides in Russia.

Data and methods

We consider texts of court decisions an alternative source of the relevant data. According to the Russian law, courts are mandated to publish such texts either on their websites or on the state automated system (SAS) Pravosudie [Justice]. There are some exceptions when verdicts are notpublished, namely, in cases related to treason, sexual assault, or minors. SAS Pravosudie contains cases from all regions except Moscow, while Moscow cases are located on the Moscow city court’s website. These texts include the detailed description of the committed crime, investigation results, witness testimony, victim’s statements, etc. Because of the comprehensive and semi-structured nature of such texts, the variables necessary for the analysis may be extracted from these texts using text mining techniques.Crimes that may involve intimate partner homicides belong to the articles 105 (“Murder”), 107 (“Homicide committed in a state of temporary insanity”), and 111 part 4 (“Intentional infliction of a grave injurywhich has involved the death of the victim by negligence”) of the Russian Criminal Code. For our research, we chose to include only cases of Article 105, Part 1. This part of the article implies that there was only one victim and one offender of a murder, the victim was not in a help-less state and not in a state of pregnancy, the murder was not committed with special cruelty and was not attended by rape or violent sexual actions. All the listed and some other aggravating circumstances are presented in Part 2 of Article 105, which was not included in our analysis.We intentionally limit our analysis to Part 1 of Article 105 as it makes the data more homogeneous and, therefore, substantially eases the further preprocessing and analysis of text data. The prescribed punishment for this crime is from 6 to 15 years of imprisonment.The Russian Criminal Code contains a list of factors that mitigate or aggravate the punishment and are applicable to all articles if these factors are not explicitly assumed by the article. The mitigating circumstances include offenders’ age under 18, their responsibility for infant children, the fact that offenders pleaded guilty, the illegality or amorality of the victim’s behavior, and some other factors. As the Criminal Code posits, judges may also take into account other factors not included in the list. The mentioned amoral behavior of the victim may potentially include the victim’s history of alcohol abuse; however, this category is not explicitly defined in the Criminal Code, so it remains the judges’ discretion to consider this factor as the amoral behavior. The mitigating factors are taken into account only if aggravating factors are absent. Because most of the general aggravating factors are assumed by Part 2 of Article 105, for the chosen Part 1 of this article, such factors include recidivism of offences, commission of a crime with the use of weapons, and others. The recidivism in Russia is of different degrees, which also define the type of correctional facility where criminals are sent. The perpetrator’s alcohol intoxication may be considered an aggravating factor “depending on the nature and degree of public danger of the crime, the circumstances of its commission, and the personality of the perpetrator” (Article 63 of the Criminal Code).This research relies on a unique set of quantitative data extracted from the texts of court decisions. We used the web scraping technique to download all available texts of court decisions for Article 105, Part 1 of the Criminal Code of the Russian Federation (“Murder”) from the SAS Pravosudie and the Moscow city court’s website. The initial data set included 23816 verdict texts on homicides committed in 2013-2019 in Russia.We first selected the cases of male-to-female homicides. In the Russian language, endings of words define the gender to which a word belongs. For example, the word подсудимый [podsudimiy] describes a male defendant, the word подсудимая [podsudimaya] describes a female defendant. Using these and similar expressions, we detected those cases where a female victim was murdered by a male defendant (N=3380). Then, we distinguished those cases that were intimate partner homicides. For this purpose, we used an open-source machine learning model that automatically predicts the type of victim-offender relationships based on the text of court decisions with an accuracy level of 94% (Zhuchkova 2021). This model was used to distinguish cases of intimate partner homicides (N=1478). Only these 1478 cases of male-to-female intimate partner homicides were used in the subsequent analysis.To automatically extract the values of the variables, we used a set of rules comprised of specific phrases and regular expressions (sequences of special characters)1. Appendix 1 demonstrates the examples of how the length of imprisonment2, the victim’s abuse of alcohol, and gender of a judge were defined. Using this approach, we automatically extracted the dependent variable (length of real incarceration in months), independent variables of interest (victim’s abuse of alcohol, and gender of a judge), and all the control variables for future regression models: the degree of recidivism, the degree of guilty plea, and the presence of other aggravating and mitigating circumstances. Unfortunately, since many personal characteristics are anonymized in the texts of court decisions, some of the key control variables, such as the age or occupation of victims and offenders, were not included in our analysis.

To extract the variable indicating the presence or absence of victim’s verbal provocation we defined phrases that describe insulting, swearing, arguing, etc. by a victim and assigned those texts in our sample that contained these phrases as cases with verbal provocation (see Appendix 2 for the examples that demonstrate how these phrases might look like in a text; see Appendix 3 for the full list of the regular expressions used to detect verbal provocation in verdicts).To answer the proposed research questions, we first conducted preliminary bivariate analysis by building and comparing confidence intervals for the average length of imprisonment (in months) and then employed linear regression models with this variable as the dependent variable.Four models were built. In the first model, the independent variable of interest is the presence or absence of the victim’s alcohol abuse. In the second model, the independent variable of interest is the interaction between this variable and a judge’s gender. The third and fourth models included the victim’s verbal provocation as a main effect and an interaction term with the judge’s gender. The set of control variables comprises year, the degree of recidivism, guilty plea, and the presence of additional aggravating and mitigating circumstances. The degree of recidivism and the guilty plea were distinguished from other mitigating and aggravating circumstances as they were the most prevalent factors mentioned in the verdicts and the easiest to extract from the texts. All these variables, except for year, are legal factors that may directly affect the term of imprisonment. The year was included in the analysis to control the possible change in the dynamics of the average length of sentences for intimate partner homicides that may be anticipated because of the gradual change in overall attitudes towards such crimes in Russian society observed in recent years. Inference comes from the variance-covariance matrix clustered at the court level to address the possible commonalities of error terms (for more, see Arellano 1987).The average length of imprisonment for the examined cases is 108 months (9 years; SD=21.7 months). Out of all 1478 cases included in the analysis, the victim’s abuse of alcohol is detected in 27% of cases, the victim’s verbal provocation is detected in 36% of cases. Cases are distributed for judges of different genders equally: 50% are considered by male judges, 50% are considered by female judges. Forty percent of defendants have no prior criminal records, 39% have prior records which in Russia are not considered recidivism, 7% have a record of conviction for an intentional crime committed earlier (“recidivism”), 6% of crimes are considered “dangerous recidivism”, and 8% of crimes are considered “especially dangerous recidivism”. In 54% of cases, defendants fully admit their guilt, in 23% defendants partially admit their guilt, and in 23% defendants do not admit their guilt. Aggravating circumstances are observed in 56% of cases, while mitigating circumstances are observed in 92% of cases.

Results

Summary statistics depending on the judges’ gender (see Table 1) indi- cate that there are no significant differences in these variables between male and female judges1, except for the length of imprisonment, which is higher for male judges (109 months for male judges and 107 for female judges), the presence of aggravating circumstances, which are also more common for cases judged by males (60% for male judges and 53% for female judges), and the presence of victim’s verbal provocation, which is mentioned more often in cases considered by female judges (31% for male judges and 41% for female judges).

Table 1Summary statistics by judges’ gende

Length of real incarceration, months (t = 2.24, p = 0.025)

Victim’s abuse of alcohol, % (χ2 = 0.98, p = 0.321)

Victim’s verbal provocation, % (χ2 = 17.37, p = 0.000)

Recidivism degree, % χ2 = 4.81, p = 0.307)

Guilty plea, % (χ2 = 0.25, p = 0.880)

Other mitigating circumstances, % (χ2 = 2.77, p = 0.096)

Other aggravating circumstances, % (χ2 = 5.57, p = 0.018)

The preliminary statistical analysis of differences in sentences was conducted by comparing the confidence intervals built for the average length of imprisonment. In regard to the first research question, these intervals were calculated for cases with and without the victim’s abuse of alcohol. In regard with the second research question, they were calculated for four groups of cases: 1) cases with no abuse of alcohol considered by a female judge; 2) cases with abuse of alcohol considered by a female judge; 3) cases with no abuse of alcohol considered by a male judge; 4) cases with abuse of alcohol considered by a male judge. The average lengths and their 95% confidence intervals for the corresponding two questions are presented in Tables 2 and 3. If the boundaries of the confidence intervals overlap, it means the lengths of imprisonment for the given groups do not differ statistically. At first sight, the results of the bivariate analysis show that cases with and without the victim’s abuse of alcohol differ significantly in terms of the sentence length. However, when taking into account a judge’s gender, it has been found that only one group of cases stands out among others: cases with the victim’s abuse of alcohol considered by female judges, for which the average sentence length is significantly lower than for others.

Table 2Average sentences for cases with and without the victim’s abuse of alcohol

Table 3Average sentences for cases with and without the victim’s abuse of alcohol considered by female and male judges

Nevertheless, in such an analysis, all other factors that possibly may affect the sentence length are not taken into account. Because of that, the identified difference of cases with the victim’s abuse of alcohol considered by female judges might be explained by other circumstances of the crime not related to the alcohol abuse itself. In order to control these other factors, regression analysis was employed.The results of the regression analysis are presented in Table 4. The previously identified pattern is reproduced after controlling for other factors as well. Hence, all other things being equal, the victim’s abuse of alcohol significantly decreases the length of imprisonment (p=0.000) when taken into account, regardless of the judge’s gender. However, when considered separately for male and female judges, the victim’s abuse of alcohol decreases the term of imprisonment only in cases with female judges (p=0.000). It is also worth mentioning that the victim’s abuse of alcohol, apparently, becomes the factor that distinguishes the differences between male and female judges’ decisions in general, since the main effect of the judge’s gender remains non-significant in both models.

Table 4Results of linear regression analysis (DV — length of real incarceration in months)

Note. The reference categories: no prior crime records, guilt is not admitted, no aggravating circumstances, no mitigating circumstances, male judge, no victim’s abuse of alcohol, year 2013. Standard errors are clustered at court level.

In regard to our hypotheses related to the effect of alcohol abuse itself, both of them have not been supported by our data and analysis. We expected both male and female judges to give more lenient punishments in cases where victims experienced alcohol abuse and assumed that for male judges, this effect would be stronger. According to our results, the victim’s abuse of alcohol affects sentences only in cases considered by female judges. It is also worth mentioning that the effect of alcohol abuse is comparable and, in some cases, even greater than the effect of legal factors. Thus, the presence of other aggravating circumstances, except for recidivism, increases the sentence, on average, by five months, while the victim’s abuse of alcohol in combination with a female judge decreases the sentence, on average, by more than six months.To explain the potential mechanism why a victim’s abuse of alcohol affects sentences, we also tested the third hypothesis that the victim’s abuse of alcohol leads to verbal aggression. As our data show, victims with a prior history of alcohol abuse, indeed, more frequently verbally provoke their partners in comparison with those who do not have such a history. Thus, in 41% of cases where victims abuse alcohol, the verbal provocation is mentioned, while in cases with no alcohol abuse, this percentage is 34% (χ2=5.54, p=0.019). However, the provocation does not seem to make a crucial contribution since (1) it is seen that more than a half of victims who abused alcohol did not provoke their partners, and (2) the very variable of provocation has no significant effect when including in the previous regression models neither as a separate variable, nor as an interaction with judge’s gender or alcohol abuse, and all other effects do not change significantly (see Table 5). This result indicates that judges tend to take into consideration the victim’s lifestyle rather than just behavior prior to or during the crime. Table 5Results of linear regression analysis with provocation variable included (DV — length of real incarceration in months)

VariableModel 1Model 2 Model 3BSEpBSEpBSEpVerbal provocation &alcohol abuse      -2.512.21.257Previous crimerecords which arenot consideredrecidivism2.301.11.0372.301.10.0372.251.11.043Recidivism14.452.78.00014.442.82.00014.292.80.000Dangerousrecidivism14.052.57.00014.052.57.00013.922.59.000Especially dangerousrecidivism20.722.28.00020.722.29.00020.742.28.000Partial guiltyplea-.151.59.926-.151.58.925-.171.58.913Full guilty plea-3.981.31.002-3.981.31.002-4.041.32.002Presence ofmitigatingcircumstances-2.701.62.096-2.701.62.096-2.691.62 .097Presence ofaggravatingcircumstances5.251.19.0005.251.19.0005.311.20.000Year 2014-.411.89.831-.401.89.831-.381.89.843Year 20153.391.75.0533.391.76.0553.311.77.061Year 20163.961.72.0223.961.72.0213.931.72.023Year 20175.761.80.0015.761.81.0015.741.81.002Year 20186.041.94.0026.041.95.0026.001.92.002Year 20191.351.83.4621.351.83.4591.341.83.466N  1478  1478  1478R2  .181  .181  .182Adjusted R2  .171  .170  .171 

Note. The reference categories: no prior crime records, guilt is not admitted, no aggravating circumstances, no mitigating circumstances, male judge, no victim’s abuse of alcohol, no verbal provocation, year 2013. Standard errors are clustered at court level.DiscussionThe obtained results contradict the proposed hypotheses and demonstrate that female judges, rather than male judges, are prone to take into account the fact that a female victim of intimate partner homicide abused alcohol and adjust sentences based on this characteristic. It was shown that the presence of such a characteristic decreases the length of imprisonment in cases considered by female judges. The identified effect also contradicts most of the previous studies devoted to the victim-blaming issue and the defensive attribution theory that claims that observers rely on their own similarity with the victims when (not) blaming them.Another theory underlying this research is the just world theory. According to this theory, people believe that bad things happen to bad people. It may be assumed that such a belief intensifies in cases of crimes that are more relevant for observers (Gravelin et al. 2019). Although we did not expect to obtain such significant disparities for male and female judges, some previous research confirms that the just world beliefs are more common for women, especially in cases of sexual crimes that involve much more women as victims than men (Sinclair & Bourne 1998).In regard to intimate partner violence, this explanation is also relevant as it is much more likely for women to be victimized by their partners than for men (Dobash et al. 1992). Kristiansen and Giulietti (1990) received similar to ours results in their scenario-based study and proposed a resembling explanation why women blame female victims by supplementing the just world theory with the notion of perceptions of control. They posit that, although it may be equally necessary for both men and women to believe that the world is “just” and everyone gets what they deserve, it may be more important for women to feel control over their own likelihood of victimization. If so, female judges tend to blame female victims to maintain such control and increase their confidence of not being the victim in the future whereas male judges may feel their own ability to prevent theirvictimization. Consistent with these assumptions, Janoff-Bulman (1982) showed that the more female observers blame rape victims, the more these observers are confident that they will not face the rape in the future.Finally, the observed effect may be explained not by gender of judges but rather by their attitudes toward women and their perceptions of gender roles and norms. In this regard, Simonson and Subich (1999) show that the effect of observers’ gender disappears when controlling for gender roles attitudes. In case of intimate partner violence, the victim’s alcohol consumption may be perceived as women’s violation of gender roles. If so, our results provide evidence to assume that for female judges traditional attitudes are more common than for male judges. Although our data lack indicators on such attitudes, we may test whether judges interpret victim’s abuse of alcohol as “amoral behavior” that, according to the Russian Crimi- nal Code, may be used as a mitigating factor. It turns out that only in 26% of cases with victim’s abuse of alcohol the judge mentions “the amorality of victim’s behavior” in a resolution section of the verdict, and these cases are distributed by male and female judges almost evenly: male judges mention this mitigating factor in 28% of cases with victim’s alcohol abuse, female judges mention it in 24% of cases with victim’s alcohol abuse, and these differences are non-significant (χ2=0.91, p=0.341). Therefore, at least explicitly most judges do not classify women’s alcohol abuse as amoral behavior.One notable pattern in the data is that female judges were more likely to consider cases in which the victim’s verbal provocation was mentioned (41% vs. 31%; see Table 1). This difference may reflect two distinct but interrelated mechanisms. First, it is possible that female judges are more often assigned or choose to adjudicate cases involving intimate partner violence, especially if such cases are perceived as requiring more sensitivity or empathy. Second, female judges may be more attentive to details of the relationship dynamics and thus more likely to include or emphasize instances of verbal provocation in the written verdicts. Both mechanisms would result in a higher proportion of such mentions in cases decided by women. Alternatively, this discrepancy could reflect a selection bias at the court level or institutional routines in case assignment, which can only be partially accounted for in our data. Future research should investigate the internal logic of case distribution and framing of victim behavior in judicial texts. ConclusionThis research aimed at detecting differences in how victims’ abuse of alcohol affects sentencing in intimate partner femicides considered by male and female judges. To answer the proposed questions, we used data extracted from the texts of court decisions on 1478 intimate partner femicides that occurred in Russia in 2013-2019. In comparison with previous research, our study uses data on actual crimes that reflect real practices of judges.Our analysis revealed that, all legal factors being equal, the victim’s abuse of alcohol decreases the sentence length in intimate partner femicides, but this effect is common only for female judges. The identified effect is comparable and, in some cases, is even stronger than the effects of legal factors. The data also demonstrate that specifically women’s abuse of alcohol, but not the consequent verbal provocation, is associated with lenient sentences.This study has several limitations. First and foremost, we rely on the data that are automatically extracted from the texts of court verdicts, and this approach imposes some restrictions on the quality of the data. Thus, we assume that we may have missed some phrases that are relevant for the victim’s abuse of alcohol or verbal provocation, which means not all relevant cases may have been assigned as such; furthermore, we may have inaccurately extracted the values of other variables, i.e., our data may have some “noise”. However, the regression coefficients obtained in this paper were quite stable and were reproduced in different research that included various samples of the initial data, which means that the negative consequences of this “noise” may be neutralized by the strong patterns observed in our data. Second, again, because of the nature of the data, we did not include some important control variables in our analysis that explain the observed variance in the length of imprisonment and potentially may nullify the identified effect of the victim’s abuse of alcohol. Nevertheless, the texts of court decisions are currently seen as the only available source of detailed crime-level data for empirical legal studies in Russia. Third, we only detected the mentioned effect, but our data lack evidence to explain the underlying mechanism of why female judges tend to reduce the sentence for male defendants if their victims abused alcohol, so we may only hypothesize what mechanisms are involved here. To answer such kinds of questions, further research, including interviews with current or former judges or with the quantitative assessment of their attitudes towards women and intimate partner violence, is needed. Finally, we only account for previous alcohol abuse, leaving out cases where alcohol intoxication was present during the crime (either from the aggressor’s or the victim’s side). Additional analysis could provide valuable insights about the intersection of previous use of alcoholic substances and intoxication in time of the crime.