The level of attrition in domestic violence: A valid indicator of the efficiency of a criminal justice system?

This article studies the process of attrition through a follow-up of all cases of domestic violence registered by the police forces of one Swiss canton in the first half of 2012 (N = 592) as they pass to the prosecution and the court stage of criminal justice proceedings. The results show that the attrition rate found in Switzerland (80 percent) is lower than the rate usually found in the United Kingdom. This rate is explained by the fact that domestic violence is usually treated by academics as a homogeneous construct, but it is in fact composed of a large variety of offences and, in practice, the vast majority of those that are reported to the police would not entail a custodial sentence.


Introduction
The term 'attrition' finds its roots in the Latin attritio, which means 'to rub' (Merriam-Webster, 2017). Gradually, the word took different meanings, keeping however the image of the gradual diminishing or weakening of something by abrasion or friction. In the field of criminology, attrition has been defined as 'the process whereby cases drop out of the criminal justice system at one of a number of potential points of exit from that system' (Lea et al., 2003: 583). Instead of using the concept of 'drop out', some authors refer to the 'loss' or the 'filtering out' of cases throughout the criminal justice system (Heiskanen et al., 2014: 151;Jehle, 2012). Criminologists have been applying the concept of attrition mainly since the publication, in 1967, of the report on The Challenge of Crime in a Free Society by the US President's Commission on Law Enforcement. That report illustrated the attrition process through a representation of the criminal justice system as a funnel that received 2,780,180 index crimes in 1965, of which only 6.5 percent led to a sentence (President's Commission on Law Enforcement, 1967: 8-9). A 'sieve' would be a better metaphor, however, because the mesh blocks some particles, whereas a funnel only slows their passage and, in the end, allows them all to pass (Killias et al., 2012: 344).
The fact that the criminal justice process has the structure of a sieve means that only very few of the crimes committed end up with their perpetrator being incarcerated. In that perspective, the literature on punitiveness considers that a low incarceration rate per crime may reflect the limited ability of a system to solve crimes, but it may also reflect a low level of punitiveness, something that is usually seen by criminologists as a positive characteristic of a criminal justice system (Blumstein et al., 2005). For example, Finland, and the Nordic countries in general, have repeatedly been presented as a model for the rest of the world because of their limited use of incarceration (Lappi-Seppälä and Tonry, 2011;Pratt, 2008). In contrast, the literature on attrition, and in particular that studying attrition in domestic violence (DV) offences, tends to focus on low incarceration rates per crime as indicators of the inefficiency of the system (see, for example, Hester, 2006;Walby et al., 2011).
Attrition should be operationalized through research designs that follow offences, offenders or cases throughout the criminal justice system and estimate the percentage that are disposed of at each stage of that system. This percentage is commonly known as the 'attrition rate' and corresponds to the percentage of offences (or offenders, or cases) that do not reach a specific stage (for example, prosecution, court or prison) of the criminal justice system. Nevertheless, there is a risk of confusion in the specialized literature because sometimes the attrition rate is presented as the percentage of offences that effectively reach one of these stages (Walby et al., 2011). Technically, the latter should be considered not as an attrition rate but as an indicator known as 'certainty of conviction', when it refers to the percentage of offenders known to the police who have been convicted, and as 'certainty of punishment', when it refers to the percentage of offenders known to the police or convicted who have been incarcerated (Blumstein et al., 2005). For example, in the year 2005, in eight West European countries, there were 35 persons convicted of homicide per 100 persons known to the police for that offence (Aebi and Linde, 2012b), which corresponds to an attrition rate of 65 percent and a certainty of conviction of 35 percent. Moreover, the indicator of the certainty of conviction is sometimes referred to as the 'conviction rate' (Council of Europe, 2016;Walby et al., 2017) or the 'conviction ratio' (Jehle, 2012).
This article analyses the level of attrition, from police files to prosecution services and court records, for all DV cases recorded by the police forces of the canton of Vaud, Switzerland, during the first six months of 2012 (N = 592). 1 According to the classification used in Swiss criminal justice statistics (Zoder, 2012), DV is defined in this article as physical and psychological violence committed between intimate partners (current or ex-husbands or boyfriends/girlfriends, within a year of the breakup of the relationship). In particular, the article tries to answer the following questions: What is the rate of attrition? How can that rate be explained? Is that rate comparable to the one observed in similar studies conducted in Europe? The answers provided by the empirical research will also lead to a more general question: Is the rate of attrition a valid indicator of the efficiency of a criminal justice system in dealing with domestic violence? The validity of an indicator is defined as its capacity to measure in an effective way the phenomenon under study (Aebi and Linde, 2012a). In that perspective, the European Commission for the Efficiency of Justice (hereafter CEPEJ) does not provide a specific definition of 'efficiency' (see CEPEJ, 2014b), but it was created to improve the judicial protection granted by the European Convention on Human Rights, which, in Article 13, guarantees the 'right to an effective remedy' to everyone whose rights and freedoms are violated (Johnsen, 2012). Ideally, this would imply reversing the situation to the one that existed before this violation, but in practice it usually means compensating the victim and punishing the offender. In particular, the literature on DV quoted above puts the accent on the punishing of the offender through a conviction.
The first part of the article discusses the operationalization of the concept of attrition and presents an overview of the main results of contemporary European studies on that subject. This is followed by a short presentation of key aspects of the Swiss criminal justice system, which will help the reader understand the findings. The latter are presented after a section that introduces the sample used in this research, as well as its methodology. The discussion places these findings in the context of the previous research on attrition presented at the beginning. Finally, the conclusion provides the answers to the questions posed above, highlights the limits of this study, and includes suggestions for further research.

Previous research on attrition in domestic violence
Attrition rates can be calculated for all offences or for specific offences and can refer to the cases disposed during the whole criminal justice procedure or at specific stages of it. In the UK, for example, Ratcliffe (2011) estimated that 41 percent of the crimes identified through the Crime Survey for England and Wales were reported to the police, 29 percent were recorded by them, 2.1 percent reached the courts, 1.5 percent led to a suspect being found guilty, and 0.4 percent ended up with a custodial sentence being imposed, which represents an overall attrition rate of 99.6 percent. In a comparative perspective, Jehle (2012) studied attrition from the police to the court stage for a group of six European countries in 2006, and found attrition rates of 70 percent for robbery, 83 percent for assault and 84 percent for rape. One particularity of these studies is that they operationalize attrition rates at the macro level, because they compare the total number of crimes identified at each stage of the criminal justice system in a given year. A weak point of that approach is that the offenders identified by the police in one year are not necessarily sentenced or sent to prison during the same year. Researchers have acknowledged this limitation but continue using this kind of analysis, accepting as a premise the hypothesis that, unless a country introduces a major change in its criminal law or its statistical counting rules, attrition rates do not vary radically from one year to the next. In Western Europe, that hypothesis was corroborated by research that studied trends in attrition rates from 1990 to 2006 (Aebi and Linde, 2012b).
An additional weak point of the macro-level approach is that there is no control over the reclassification of offences that may take place across the different stages of the criminal justice system. For example, Gregory and Lees (1996) found that 20 out of the 185 cases of rape and sexual assault recorded by two London police stations from September 1988 to September 1990 received a different classification at the judicial level. In particular, two were reclassified into a more serious crime and 18 were downgraded to a less serious crime (Gregory and Lees, 1996). The main reason for reclassification is related to the fact that the police intervention takes place in the context of an emergency and, as a consequence, the police officers cannot easily measure the real extent of the damage and the intention of the perpetrator. Even if the police files include the information collected during the whole police investigation, some elements can be known only during the justice proceedings. 2 At that point, the victim can also bring up new facts that imply additional offences. As a consequence, we included in this study a specific analysis of the reclassification of police recorded offences by the prosecutors and the judges.
An alternative way of operationalizing attrition rates consists in selecting and following throughout the criminal justice system a sample of offences, offenders or cases, thus adopting a micro-level perspective. For example, in a study that showed a wide diversity in the levels of attrition in reported cases of rape in 11 European nations, Lovett and Kelly (2009) included case-tracking samples, which consisted in a follow-up of samples that included 70-90 rape cases per country. Similarly, in the UK, several studies followed samples of cases of rape known to the police and found attrition rates that were usually above 90 percent (Bunting, 2008;Gregory and Lees, 1996;Kelly et al., 2005;Lea et al., 2003). In Switzerland, Aebi (2006) followed a sample of 900 drug addicts participating in the Swiss heroin prescription programmes in the 1990s. He found attrition rates of 96 percent from self-reported offences to police interpellations, and of roughly 99 percent to court convictions, for the offences committed during the six months preceding their admission to the programme (Aebi, 2006: 108).
This article follows that micro-level approach and therefore we focused our review of the literature on European studies that used a similar methodology to measure attrition in DV. To the best of our knowledge, such studies have been conducted only in the UK and often put the emphasis on attrition due to the withdraw of the complaint. Hence, Hester et al. (2003;Hester, 2006) followed 869 DV incidents recorded by the Northumbria (England) police in 2001. They observed that 222 (26 percent) of them resulted in an arrest, 60 individuals were charged with a criminal offence (27 percent of those arrested and 7 percent of the incidents), 31 individuals were convicted (52 percent of those charged, 14 percent of those arrested, and 4 percent of the incidents), and 4 of the convictions involved custodial sentences (13 percent of the convictions and 0.5 percent of the incidents). This corresponds to an attrition rate of 96.5 percent from police recorded incidents to persons convicted, which rises to 99.6 percent if one considers only the persons convicted and given a custodial sentence. Analysing the results, Hester (2006) considered that attrition in DV cases was explained by the lack of evidence for a conviction by a court and the difficulties for the victims to follow a trial. Similarly, an analysis conducted jointly by the UK Crown Prosecution Service Inspectorate and the Inspectorate of Constabulary (HMIC and HMCPSI, 2004) followed a sample of 463 incidents recorded by six police forces in March 2003. The results showed that 118 (25 percent) of the incidents resulted in a crime being recorded because in most cases the victims had withdrawn their complaints. These incidents led to 71 arrests being made and 25 offenders being charged. Finally, 13 offenders were convicted, which corresponds to an overall attrition rate of 97.2 percent. At the same time, the study followed another sample of 418 cases processed by the Crown Prosecution Services (that is, cases that had passed the police stage of the criminal process) and finalized in March 2003. Among these cases, 210 led to a conviction, of which 186 related to guilty pleas and 24 to convictions after trial (HMIC and HMCPSI, 2004). This corresponds to an attrition rate of 55.5 percent from the prosecution to the court level, including guilty pleas.
Focusing on the court level, Cook et al. (2004) followed 216 DV case files processed by Specialist Domestic Violence Courts during three months of 2003 at five different sites (Cardiff, Derby, Leeds, West London and Wolverhampton). They found that 18 percent of the cases dropped out of the system before trial, because they had been either withdrawn (11 percent) or discontinued (7 percent). The cases that reached the court led to 69 defendants (32 percent of the total sample) pleading or being found guilty (Cook et al., 2004). Robinson and Cook (2006) conducted supplementary analysis of the data and found that half of the victims had decided to retract. They conclude that retraction is one of the causes of attrition, but it is not necessarily synonymous with a failure of the justice system (Robinson and Cook, 2006). Retraction was also studied by Barrow-Grint (2016), who followed 50 cases registered by the Thames Valley Police in 2014 and found out that, in most cases, the withdrawal of the complaint had taken place during the five days that followed the violent episode. Summarizing the results of most of these studies (Hester et al., 2003;HMIC and HMCPSI, 2004;Cook et al., 2004 ) and one that followed a different research design (Hester and Westmarland, 2005), Hester (2006) considered that the patterns of attrition were 'depressingly similar' (Hester, 2006: 81).
At the same time, even if theoretically the level of attrition in DV cases should be lower than the average because the victim knows the perpetrator and therefore the police can easily identify him or her, empirical research has not corroborated that hypothesis. In practice, the level of attrition in DV is similar to and sometimes higher than that observed for offences in general in the UK (HMIC and HMCPSI, 2004).

The Swiss criminal justice system in a European perspective
Switzerland has a Roman law criminal justice system, and the Swiss Criminal Code (SCC) makes a distinction between contraventions (sanctioned with a fine), misdemeanours (which carry a custodial sentence not exceeding three years or a monetary penalty) and felonies (which carry a custodial sentence of more than three years). The difference between a fine (whose maximum amount is 10,000 Swiss Francs) and a monetary penalty is that the second is considered as an alternative to a custodial sentence (diversion) whereas the fine is a sanction in itself (arts. 10 and 103 SCC). The Swiss Federal Code of Criminal Procedure (SCP) introduced in 2011 abolished the position of investigating judge and replaced it by the prosecutor, who leads the criminal investigation and can impose custodial sanctions of up to six months. As a consequence, currently the vast majority of convictions are not pronounced by a court but are based on orders imposed by the prosecutor, which are 'not a judgment of first instance, but an offer to the parties for an out-of-court settlement of the criminal case or a proposed judgment' and the defendant can only accept or refuse by raising a written objection within three days, something that seldom happens (Gilliéron, 2014: 226-8). In particular, according to the national data collected by the Swiss Supreme Court for the CEPEJ, out of the 499,312 cases received by the public prosecution services in 2012, only 2.1 percent were brought before a court, and the rest ended up with different types of orders imposed by the prosecutor (CEPEJ, 2014a: 274).
When the defendant has accepted responsibility or the latter has been established, the prosecutor issues a penal order (art. 352 SCP), which can consist of a fine, a monetary penalty (up to 180 daily penal units), a community service sanction (up to 720 hours), or a custodial sentence of no more than six months. Apart from that, prosecutors can also impose three other types of order that lead to the temporary or permanent closing of the proceedings: suspension order, no-proceedings order, and abandoning of the proceedings order. Thus, the public prosecutor may suspend an investigation for up to six months if, for example, the offender or his or her whereabouts are unknown, a private settlement proceeding or another type of proceeding is ongoing and it seems appropriate to await their outcome, or a decision on the substance of the case depends on how the consequences of the offence develop (Suspension order, art. 314 SCP). This possibility is particularly relevant in the case of DV because the SCC foresees that, in the case of common assault, repeated acts of aggression, threats or coercion between partners or expartners (within a year of separation or divorce), the public prosecutor or the court may suspend the proceedings for six months at the request of the victim (Discontinuation of proceedings, art. 55a SCC). If the victim does not revoke his/her request within the six months, the prosecutor shall order the abandonment of the proceedings (Abandoning proceedings order, art. 319 SCP). 3 In practice, the effect of this combination of the suspension order with the abandoning proceedings order could be compared to the effect produced by the withdrawal of the complaint that still exists in other countries, such as the UK. Finally, the public prosecutor shall rule that no proceedings will take place if, on the basis of the complaint or the police report, it is established that the elements of the offence concerned or the procedural requirements have clearly not been fulfilled, if there are procedural impediments, or if there are other reasons (that is, the offender is seriously affected by the consequences of his/her act or he/she has proceeded to a reparation of the loss or damage) for waiving the prosecution (No-proceedings order, art. 310 SCP).

The operationalization of domestic violence in police files
The identification of DV in police statistics is particularly problematic because the latter usually follow a classification based primarily on the type of offence recorded and not on the relationship between offender and victim. For example, studying long-term trends in DV is quite difficult because of the lack of reliable data on basic indicators -such as the number of female victims of intimate partner violence -until the beginning of the 2000s, when DV became a criminal policy priority in Western Europe (Tonry, 2014). In Switzerland, the issue was solved at the federal level in 2009 with the introduction of a new statistical system for police recorded offences. 4 Assuming that, from a practical point of view, it is unrealistic to ask police officers to register the personal link between victim and offender for every intervention, the Swiss Federal Statistical Office (OFS) established a list of offences for which such ties must be systematically recorded. This list contains offences that typically can take place in the context of DV, including all kinds of intentional homicide, assault and battery, rape and sexual assault, as well as sexual harassment, threats, insults, coercion, endangering life, abduction, defamation, and misuse of telecommunications installations (for a full list, see Zoder, 2012). When one of these offences takes place, the police records whether it took place between partners or former partners (within a year of the breakup of the relationship), between parents and children, or between persons linked by other family ties. This was the criterion used for extracting the database for this research, but restricting the cases to the ones involving partners and former partners. It must be mentioned, however, that the main DV offence is often accompanied by other transgressions of the law -identified in this research as ancillary offences -that are not in the OFS list but that are also recorded by the police as part of the same case (see Table 1 in the Appendix).

The counting unit
Research on attrition faces a classic methodological problem caused by the fact that the counting units used at different levels of the criminal justice system are not necessarily the same (for example, the prisoner in prison statistics, the conviction or the person convicted in court statistics, the case in prosecution statistics, and the number of offences and of suspects in police statistics). In several of the studies quoted throughout this article, the authors start with the number of offences recorded by the police (see, for example, Hester, 2006) or revealed in crimes surveys (see, for example, Ratcliffe, 2011), which they later compare with the number of persons convicted or imprisoned. Nevertheless, the passage from the offence to the person as a counting unit remains problematic because a suspect may be accused of several offences, a sentence imposed by a court on an offender usually includes more than one offence, but all these offences can lead to only one person going to prison. In this research, we had access to an original police database in such a way that it was possible to use four counting units (case, offence, offender and victim) at the level of police files, and follow three of them (case, offence and offender) in prosecution and court records. The attrition process was measured using the case as the counting unit.

The sample
The sample used in this research is based on all the cases (N = 592) of DV recorded by the police between 1 January and 30 June 2012. The time period considered was decided taking into account that a period of six months provides a sample that is larger than the ones used in similar studies (see the previous section) and that, according to the length of the criminal justice proceedings in Switzerland, the vast majority of the cases would have been closed by the second semester of 2014, when the data collection took place (see below).
Researchers were granted access to an anonymized police database, which included the characteristics of the case (for example, date, place and offences recorded), the victim and the suspect (for example, age, gender, country of birth and professional activity). Personal data protection legislation did not allow the extraction of information on offences and victimizations committed before the period of reference, which could have been useful for the study of recidivism, and the database did not allow establishing whether foreign victims and suspects had the status of permanent or non-permanent residents.
The link between police and court files was established case by case because there is no common database between the two institutions. Researchers consulted the physical files stored in the five different prosecutorial offices and four different tribunals of the canton and collected the required information, which was added directly to the original database.
All the cases registered by the police (N = 592) were identified at the justice level; 23 cases had been filed for lack of evidence, 2 had been transmitted to the justice authorities of other states and 18 were still being processed and, as a consequence, could not be accessed (see Figure 3). The low number of cases that had not been closed at the time of the data collection (n = 18) corroborates that the time period chosen for the study was appropriate. Thus, the justice database includes 549 cases. In some of the analyses presented in this article, the number is slightly lower because there were some missing data regarding specific variables.

Findings
The profile of domestic violence cases Figure 1 presents the distribution of the offences recorded by the police and the justice authorities. In the case of the justice authorities, the classification of the offence corresponds to the one made by the prosecutors -in the cases that did not reach the courts -or by the courts. Figure 1 shows that roughly one-third of the offences processed by the criminal justice system correspond to battery only causing pain (art. 126 SCC), which constitutes a contravention because it is sanctioned with a fine. Insults (art. 177 SCC) and threats (art. 180 SCC) represent almost half of the offences (nearly one-fourth each) processed by the criminal justice system. Both are considered as misdemeanours. In the case of insults, the sanction is a monetary penalty that cannot exceed 90 daily penalty units, and in the case of threats the sanctions range from a fine to a custodial sentence not exceeding three years. Lastly, common assault (art. 123 SCC) represents 5.5 percent of the offences recorded by the police, but it increases to 11.6 percent when prosecutors reclassify the offences. Common assault also constitutes a misdemeanour and carries the same range of sanctions provided for threats. For these four contraventions and misdemeanours, which represent almost 90 percent of DV offences, the code foresees that the offender is prosecuted ex officio if he or she commits the offence repeatedly against her or his current or previous partner, within a year of the separation or divorce (art. 126 SCC).
The rest of the offences include the ones that represented less than 5 percent of the total number of offences and are presented in Table 1 in the Appendix. It can be seen that the most serious offences are included in this group, which includes mainly murder, rape, serious assault and other sexual offences. Table 1 also gives a general overview of the reclassification of offences by the judicial authorities, which are analysed in the following section.

The reclassification of offences
It can be seen in Table 1 in the Appendix that the passage from the police level to the justice level implies an increase of 24 percent in the number of offences (from 1132 to 1408). At the same time, it implies a decrease of 4 percent in the number of cases (from 592 to 567) due to the 2 cases transferred to another jurisdiction and the 23 that were filed for lack of evidence.
In order to study the specific phenomenon of the reclassification of offences, a random subsample of 245 cases was selected. For each case, the most serious offence according to both sources of data (police and justice authorities) was identified and then compared. The seriousness of the offence was established according to the sanction provided for in the SCC.
The analysis shows that a reclassification took place in 55.8 percent of the cases. In two-thirds of them, the reclassification led to an aggravation of the original offence (that is, the offence recorded by the police was reclassified as a more serious offence by the prosecutors); in the remaining one-third, the reclassification led to a less serious offence than the original one. Table 1 presents several examples of the aggravation of the classification. In the case of battery only causing pain there were 351 cases in the police files but 455 at the judicial level, threats grew from 278 to 312, common assault from 62 to 163, and rape from 8 to 17. The most common reclassification involved the passage from battery only causing pain to common assault. Table 1 shows that common assault represented only 5.5 percent of the offences recorded by the police but increased to 11.6 percent of the offences dealt with by the prosecutors and the courts.
The increase in the number of offences shown in Table 1 implies that the prosecutors and the courts reclassify some offences, but also introduce new ones. The main offence that was introduced by them was damage to data, which consists in altering, deleting or rendering unusable data that is stored or transmitted electronically (art. 144 SCC). This offence was introduced in 49 cases; while the classification of misuse of telecommunications installation, which was used by the police in 29 cases, was not kept by the prosecutors and the judges. At the same time, on several occasions the judicial authorities introduced the classification of attempts, especially for the most serious offences. That was the case for murder (n = 2 cases), rape (n = 2), serious assault (n = 1), unlawful entry (n = 1) and indecent assault (n = 1). They also introduce a case of negligent assault.

The number of offences included in each case of domestic violence
According to police records, almost 60 percent of the cases of DV included more than one offence. In particular, two offences were recorded in 35 percent of the cases, three in 23.1 percent of them, and four in 1.2 percent of them. The rest of the cases (40.7 percent) included only one offence. Figure 2 presents this distribution of the number of offences by case and allows a comparison with the distribution observed at the justice level.
The figure provides an additional illustration of the introduction of new offences by the prosecutors and the courts. For example, the police did not identify a single case with more than four offences, whereas, at the level of the judicial authorities, 6 percent of the cases included five or more offences. The percentage of cases with four offences also increased from 1 percent at the police level to 8 percent at the justice level, while the cases with three offences remained more or less stable. In contrast, the cases including only one or two offences decreased from 41 percent to 31 percent and from 35 percent to 31 percent respectively.

Attrition pattern
The way in which the 592 cases registered by the police were dealt with by the criminal justice system is presented in Figure 3. The vertical (shadowed) path illustrates the attrition process as it has been classically studied. Figure 3 shows that 96 percent of all cases recorded by the police were processed by the justice authorities, and that 12 percent of them led to a sanction imposed by a prosecutor (penalty order) and 8 percent to a sanction imposed by a court. Thus, all in all, 20 percent of the cases registered by the police led to a conviction, which means that the overall attrition rate was 80 percent.
The horizontal paths show how the rest of the cases were diverted during the proceedings. At the police level, 0.3 percent were transmitted to the justice authorities of another canton and 4 percent were closed for lack of evidence. At the justice level, the court acquitted the suspect in 0.8 percent of the cases, while the prosecutors filed 72 percent of the cases through different types of orders: 2 percent by a suspension order, 8 percent by a no-proceedings order, and 62 percent by an abandoning of the proceedings order. Finally, 3 percent of the cases were still being processed in mid-2015. If one makes the hypothesis that the distribution of the cases transferred to another canton (0.3 percent) and the ones pending (3 percent) is similar to that of the cases closed, it can be estimated that roughly 21 percent of all cases could lead to a conviction and the attrition rate would thus be 79 percent.

Discussion
The overall attrition rate observed in this research is 80 percent, which means that 20 percent of the cases of DV known to the police led to a conviction. From an abstract point of view, that level of attrition could be considered high and deserves an explanation. As mentioned in the introduction, when attrition rates are calculated as we have done in this article, the criminal justice system resembles a sieve in which only a few cases reach a conclusion. The risk of this kind of analysis is to consider the criminal proceedings as a linear process and conclude that the standard for efficiency would be to have each crime reported to the police, who will identify the perpetrator and charge him/her before the court finds him/her guilty and imposes an appropriate prison sentence (Bryden and Lengnick, 1997). In this perspective, cases in which the suspected offender was innocent or was diverted during the process will be considered as 'lost', 'filtered out' or 'dropped out' when in fact the system has processed them differently. In Europe, research has shown that the main actors in the attrition process are the police and the prosecutors, who can take a lot of case-ending decisions (Jehle, 2012;Jehle et al., 2008). In particular, the police act as a filter when they have discretionary powers that allow them to transmit or not the cases to the prosecutors on the basis of the seriousness of the offences, the degree of evidence, or the withdrawal of the complaint by the victim. In a similar way, prosecutors can act also as a filter between the police and the courts by imposing a sanctionusually in the form of a penal order -that does not necessarily require the approval of the court and may not appear in conviction statistics, thus artificially reducing the level of attrition. Our empirical research has clearly illustrated the weight of these interventions at the prosecutor stage in Switzerland, and our review of the literature showed that they play a major role at the police stage in the UK.
In that comparative perspective, an attrition rate of 80 percent could be considered as relatively low because the rates found in the UK are usually above 90 percent (Hester, 2006;HMIC and HMCPSI, 2004). However, comparisons based on data produced by the criminal justice agencies are particularly problematic because such data are influenced by legal, statistical and criminal policy factors -such as the definitions of offences and characteristics of the justice system, the statistical counting rules applied when collecting the data, and the interest in pursuing particular offences -which have a stronger influence than substantial factors, which relate to the real frequency of an offence (Aebi, 2010;Von Hofer, 2000). For example, in this research, 96 percent of the police recorded cases were transmitted to the prosecutors, whereas in the UK, according to the research conducted by HMIC and HMCPSI (2004), only 25 percent of the cases known to the police resulted in a recorded crime, mainly because the victim had decided to withdraw the complaint. As we have seen, the difference comes from the fact that, under the current Swiss legislation, the victim cannot withdraw the complaint at the police level, but she/he can ask the prosecutor to suspend the procedure for six months and, after that period, the prosecutor can directly order the abandoning of the proceedings. In this research, 62 percent of the cases were filed through an abandoning of the proceedings order.
The attrition rate observed in this study is also lower than the one observed at the level of the whole Swiss criminal justice system. In the latter, we have seen that the attrition rate for 2012 was 97.9 percent, as only 2.1 percent of the cases reached a court (CEPEJ, 2014a), whereas, in the cases of DV followed in this research, that percentage was multiplied by more than four as 8.8 percent of the cases reached a court. In addition, 90 percent of the DV cases that reached the courts (45 out of 50) ended up with a conviction.
Another interesting result is that, in more than half of the DV cases (55.8 percent), the prosecutors and judges proceeded to a reclassification of the most serious offence recorded by the police, which led to an aggravation of two-thirds of them and an attenuation of the other third. Prosecutors and judges also added new offences and introduced the concepts of 'negligence' and 'attempts'. Hence, the passage from the police to the prosecution and court stages of the criminal justice system implied a consistent increase in the number of offences recorded and their seriousness. This finding casts some doubts on the validity of attrition studies based on data collected at the macro level, which do not take into account these variations. In this research, the influence of the reclassification could be controlled because the counting unit was the case, and each case had been followed throughout the criminal justice system in order to establish the attrition pattern. In contrast, most macro-level studies of attrition are based simply on the number of offences of the same kind treated at the different stages of the criminal procedure.

Conclusion
Any conclusion made on the basis of the analysis presented in this article must start by highlighting the limitations of police recorded offences for the study of the real magnitude of any sort of crime. The cases of DV identified by the police represent only a fraction of the ones in effect committed. As a comparison, according to the Swiss survey on DV -which was part of the International Violence Against Women Survey or IVAWS (Johnson et al., 2007) -only 27.6 percent of the victims of DV reported the case to the police, 6.9 percent filed a formal complaint, and 3.4 percent obtained the offender's conviction (Killias et al., 2005: 84). Taking that limitation into consideration, in this article we analysed the way in which the criminal justice system dealt with the 592 cases of DV registered by the police of the canton of Vaud, Switzerland, during the first semester of 2012, with the aim of answering three research questions.
The first question was: What is the rate of attrition? Our findings show that 20 percent of the cases led to a conviction. In particular, 12 percent of the cases led to a conviction imposed by the prosecutor in the form of a penalty order, and 8 percent led to a conviction imposed by a court. This means that the attrition rate is 80 percent.
The second question was: How can that rate of attrition be explained? Our findings show that the main reason is that the criminal justice system is seldom confronted by felonies or very serious offences, which could lead to a custodial sentence of more than six months. In particular, roughly 90 percent of the DV offences processed by the criminal justice system corresponded to one contravention (battery only causing pain) and three types of misdemeanours (insults, threats and common assault). Battery only causing pain and insults are sanctioned with a fine, while threats and common assault can lead to a fine or a custodial sentence not exceeding three years.
The third question was: Is that rate of attrition comparable to the rate observed in similar studies conducted in Europe? As we have seen, such comparisons are not straightforward but, in general, the attrition rate found in Switzerland seems lower than that observed in the UK, which is the only country where it was possible to identify similar studies. Moreover, the attrition rate found in this research for DV offences is also lower than the Swiss general attrition rate (that is, the attrition rate for all offences).
However, criminologists seem divided as to how to interpret the level of attrition of a criminal justice system. As we saw in the introduction, the literature on punitiveness usually considers a high level of attrition (that is, a low level of convicted or incarcerated offenders) to be positive because, instead of reflecting a poor level of efficiency of the system, it could be reflecting a low level of punitiveness. In contrast, the literature on DV usually considers a high rate of attrition to be a negative outcome. A suitable example of this approach can be found in the following statement: 'This article explores the process of attrition, where domestic violence cases fail to make it through the criminal justice process and do not result in criminal conviction' (Hester, 2006: 79;emphasis added). 5 The reasons for these divergent interpretations seem related to the extreme sensitiveness of the topic of DV -which has been responsible of major increases in police recorded assaults in Western highly developed countries since the 1990s (Tonry, 2014) -but also to the way in which the concept of DV is defined. In that context, DV is usually treated as a homogeneous construct, when in fact the concept covers a huge variety of offences, ranging from the less serious ones provided for in the criminal code to homicide, the most serious one. We have seen that the attrition rate varies according to the type of offence, and the specific case of offences with a DV component is no exception to that rule. By combining all these offences in a single construct, the average rate will necessarily be influenced by the less serious offences, which are the majority and for which the attrition rate is generally low. Indeed, some of these offences, in particular the ones for which a petty fine is provided for (for example, insults or battery only causing pain) usually never reach a court in a system were prosecutors can impose imprisonment sentences up to six months.
An attrition rate of 80 percent can only be understood when it is placed in that context. The problem stems from the fact that, in most fields of research, a 'success' rate of 20 percent would seldom be considered as positive. For example, a medical treatment that heals one out of five patients would immediately lead to the conclusion that 'more research is needed'. The key issue is thus the definition of efficiency. In that perspective, it is unrealistic to consider a criminal justice system as efficient when each recorded offence leads to a conviction. As we have seen, the criminal justice system does not fit the profile of a linear process, and many of the legislative reforms introduced in Europe since the 1990s were introduced with the explicit purpose of multiplying the bifurcations during that process. Thus, with the aim of reducing the length of the proceedings, many countries introduced plea bargain or accelerated trials or, as in Switzerland, extended the discretionary power of prosecutors. At the same time, all around Europe, alternative ways of solving conflicts were introduced, restorative justice has been encouraged, and community sanctions and measures were developed in order to reduce the use of imprisonment.
In our opinion, the use of an overall rate for attrition in DV creates confusion in public opinion and should be avoided. Instead of referring to attrition in DV, it would be preferable to refer to attrition in DV homicides, DV serious assaults, and so on and, whenever possible, to compare such rates with the ones found for the same offences in non-DV cases. Unless such changes are introduced, our answer to the fourth research question of this article is that the level of attrition in DV is not a valid indicator of the efficiency of a criminal justice system.
Consequently, we consider that further research on the topic of attrition in DV should be conducted by analysing each type of offence separately and comparing the results with those obtained in non-DV cases. It would also be necessary to avoid using concepts such as attrition rate, certainty of conviction, certainty of punishment, conviction rate and conviction ratio interchangeably. efficiently with this phenomenon. This form is filled in by the police officers after a DV intervention, in such a way that, if the police are called to intervene again at the same address, a specific squad can be mobilized. However, when the offence is a very serious one, such as homicide or rape, the police follow the specific protocol provided for these offences and the form is not necessarily filled in. As a consequence, research on DV conducted in the canton of Vaud during the first decade of the 2000s (Jaquier, 2010) usually did not include the most serious offences and its findings cannot be compared with those presented in this article. 5. In a similar perspective, Lovett and Kelly (2009) define attrition as 'the process by which cases are discontinued, and thus fail to reach trial and/or result in a conviction' (Lovett and Kelly, 2009: 17; emphasis added).