The Balkan Homicide Study: Research Design and Operationalization

This chapter provides indispensable insights into the BHS research design and its practical operationalization. The chapter’s leitmotiv is that there is no perfect empirical violence research – with each study we come a bit closer to revealing few of the many unknowns of (lethal) violence, while making valuable mistakes that open new lines of research. In that sense, the most meaningful way of handling the methodological and practical imperfections of the BHS is to be transparent and objective about the crucial “whys and hows” of its research design. After explaining the study’s two core objectives, the main methodological decisions and challenges will be presented. This includes various aspects of designing and using a unique instrument for data collection, sampling strategies, data representativeness, normative and statistical context, as well as field work and data analysis challenges. The chapter’s aim is to realistically depict all the methodological ups and downs of the BHS. It will equip readers with all the necessary information needed to arrive at own, potentially even divergent, conclusions on the study’s first findings.

After having provided an overall analysis of the state of art in European homicide research and its balkanization (Chap. 2), as well as the necessary Balkan-specific context and research setting (Chap. 3), we now turn to the specific methodological context the BHS findings are imbedded in. The broad scientific and cultural context enables one to understand the necessity and value of the BHS as a criminological research undertaking. The specific context allows for an informed and critical consumption of the study's data, its findings, as well as their interpretation and the first conclusions these lead to. Section 4.1 presents the study's core objectives, whereas Section 4.2 deals with the main methodological aspects of the BHS. Section 4.3 contains essential sample features and provides the normative and statistical context relevant for assessing the representativeness of the survey's sample. Section 4.4 discusses the key practical aspects of the BHS field work in light of their potential impact on the completeness and quality of the collected data.

Objectives
The BHS has two core objectives which simultaneously follow two separate lines of research. The first one deals with the social and normative construction of violence, whereas the second one investigates the empirical realities of violence in the Balkans.

Social Construction of Violence
How is violence socially constructed and normatively perceived? Is there a common normative understanding of violence throughout the region or are there considerable national differences? Who has the power to define violence and how is this reflected throughout the criminal justice process? Are there detectable factors that might help explain or even predict the outcome of such a definitional process? These are the lead questions we had in mind when designing the first line of inquiry for the BHS. In order to answer them, we collected a vast amount of procedural data within the case file analysis (Sect. 5.4).
Empirical data collection on procedural aspects proved to be rather challenging. Since the BHS aims at investigating the power to define violence (Chap. 1), it was clear from the very onset that it needs to cover not only those incidents that are finally adjudicated as (lethal) violence by courts but also all incidents initially defined as (lethal) violence by police and/or prosecution that consequently got redefined throughout the criminal justice process (and potentially dropped out). Ideally the BHS would therefore have tracked all relevant incidents from the police stage throughout the prosecution stage and eventually up to the court stage. However, based on prior experience, access to police files was assessed as extremely unlikely in all of the participating countries, whereas even the access to prosecution files proved extremely challenging in some of the countries. The first findings on the normative, hence social construction of violence, are presented in Sect. 5.4, although it has to be noted that many questions remain unanswered in terms of understanding the process of defining and redefining (lethal) violence.
Due to lack of access to police and prosecution files, it is not possible to comprehensively assess the power to define violence on the side of the police, as accomplished in prior violence studies (Sessar, 1981;Hess, 2010;Dölling, 2015). But even with these prior studies, caution is required, as there is solid empirical evidence on a considerable dark figure of homicides that remain undetected by medical doctors or pathologists, due to their incorrect initial classification of actual homicides as natural (unsuspicious) deaths. Estimates go as high as 1200 for such undetected homicides annually in Germany (Universität Rostock, 2017;Esanum, 2017) and 175 to 350 additional homicides per year in the Netherlands (Bijleveld & Smit, 2006, p. 196). European homicide research has only recently started to investigate the flow of homicide cases through the health and justice systems (Liem & Eisner, 2020;Liem, 2018), marking a valuable new line of research the BHS will also need to consider prospectively.

Empirical Realities of Violence
The BHS's second line of research focuses on the three mainstream questions commonly addressed by violence research: What are the (situational) characteristics of the incidents (Sect. 5.1)? Who are the offenders (Sect. 5.2)? Who are the victims (Sect. 5.3)? This obviously involves the victim-offender relationship, as well as the relationship between co-offenders and co-victims and numerous contextual aspects of the incident. The BHS is also interested in finding out what (lethal) violence in the Balkans actually looks like, particularly with the aim of understanding the Balkan-violence-paradox (Chap. 1) and empirically challenging the violent Balkan stereotype (Sect. 3.1).
The BHS objective is thus to detect possible protective traits in victims that survived violent incidents. This is achieved by searching for deescalating situational factors that might be useful in preventing lethal violence and by investigating potential violent traits in the offenders, as well as accelerators of deadly situations (Chap. 6). Due to these objectives, the BHS focuses on attempted homicides as well as completed ones, especially since there is said to be evidence that the characteristics of offenders of attempted homicides might be markedly different from those of completed homicide offenders. 1 Subsequently, it is reasonable to presume that the characteristics of attempted homicide victims might be markedly different from those of completed homicide victims as well. However, with regard to (lethal) violence, the BHS data analysis will clearly distinguish between attempted and completed homicides, unless indicated otherwise. Eventually, the BHS's first findings will provide for a first look at the empirical realities of (lethal) violence in the Balkans, in line with the BHS violence typology used for classifying all analyzed incidents according to the type of violence (situation and context), victim-offender relationship, motive, as well as particularly cruel, sexual, and affective perpetration ( Fig. 4.1).

Methodology
The following paragraphs deal with the basics of BHS's methodology. This includes a description of why and how the case file analysis was conducted, an explanation of the study's instrument, a discussion of the BHS violence typology, and concluding remarks on cautiousness regarding BHS data. 1 Smit et al. (2012, p. 18) argue that there obviously is a huge difference between attempted and completed homicides. They refer to Bijleveld and Smit (2006) as an example of prior research which shows that (in the Netherlands) the characteristics of offenders of attempted homicides were found to be markedly different from those of completed homicide offenders. Yet, Bijleveld and Smit (2006, p. 199) clearly state that due to pragmatic and conceptual reasons, only completed homicides were the object of their study. Eventually neither Bijleveld and Smit (2006) nor Smit et al. (2012, p. 18) provide empirical justification for excluding attempted homicides nor do they elaborate on the actual characteristics of offenders that are said to be "markedly different" or explain why attempted homicides might constitute a "criminologically distinct phenomenon."

Case File Analysis
In the region, studies on (special types of) violence cannot be conducted from official statistics such as those published by the national statistical authorities, 2 but have to be tailored to the questions at hand and thus need special data collection efforts. That is why the BHS was from its onset designed as a case analysis-based study. This also made particular sense in light of the fact that lethal violence is a rather rare criminal occurrence throughout the region, 3 while homicides may be generally considered only the "tip of the iceberg" of underlying crime (Liem in Chap. 2). As such it might best be investigated and captured by focusing in depth on a smaller and more recent sample, instead of looking at long-term trends or more general statistical data, which would by default block out the necessary level of incident details. Thus, looking at the BHS objectives, case analysis appeared to be the only meaningful research method able to collect most of the relevant data. Surely, one could have applied a mixed method approach and combined case analysis with, for example, media analysis of reports on (lethal) violence, interviews with criminal justice practitioners, and a broad variety of other methods, but this was simply not within the BHS's resources in terms of funds, scope, staff, or time. The study is original as much as it is explorative in nature since little, if any comparable, research exists in the region. It marks a solid starting ground for future research, which based on first BHS findings, will hopefully be able to look more specifically into various different features of (lethal) violence in the Balkans.

Research Instrument
At a very early stage of designing the BHS, it became clear that none of the countries will grant access to police case files. Therefore, the BHS is based on court and prosecution case files, in which the available information has been validated, but which (compared to police files) contain no information on unsolved homicides, drop-out cases that did not lead to a prosecutorial investigation (e.g., due to lack of evidence or unknown offender), or initial (attempted) homicides that were redefined as other types of violence (e.g., grave bodily injury with lethal consequence) by the prosecution. This focus on court and prosecution files is well reflected in the BHS questionnaire that was used for data extraction from the files, especially in terms of its overall structure and the type of variables. The starting point for the BHS questionnaire was a questionnaire designed and tested by Hans-Jörg Albrecht for the study of violence in Uruguay, which investigated the level of violence related to cannabis trade and how the cannabis market may lead to insecurity, all in the context of the Uruguayan model of cannabis legalization (CORDIS, 2017). This questionnaire was extensively broadened at the first study meeting by the initial BHS partners and thus adopted to the specific regional context. In hindsight, it would have been immensely helpful and surely much easier to simply adopt some or even all of the variables from prior homicide studies, as well. 4 However, this became evident only after the field work had already been conducted and is partly also a consequence of the "learning by doing" approach throughout the region (Sect. 3.2).
The questionnaire itself is divided into five main parts covering procedural variables, case variables, offender variables, victim variables, a descriptive case summary, and relationship variables which include two variable subsets: the victim-offender and victim-victim relationship (BHS Codebook, 2021). In total, the BHS questionnaire contains more than 200 variables and a descriptive case summary, while the BHS database in fact comprises five separate databases due to the different counting units the variables relate to (case, offender, victim, victim-offender relationship, and victimvictim relationship). The coding was done by the BC office and the Violence Research Lab in Zagreb and partly assisted by the MPICC's criminology department in Freiburg. In terms of the questionnaire length and complexity, one might very well describe it as an extremely impressive challenge, or more frankly speaking, as a practical and analytical nightmare. Depending on the court or prosecution case file complexity and "thickness," as well as the national researcher's proficiency in case analysis and the level of legal expertise, data collection per case lasted from approximately 30 minutes to 1 hour and 15 minutes, or on average for approximately 45 minutes.
The BHS was conducted in English language, meaning that both the questionnaire used during data collection (except for Romania) and the input language are in English (including Romania). 5 Although this minimized potential errors inherent to back and forth translation of English language questionnaires to/from different languages, it is safe to assume that the English language aspect had a minor impact on the interpretation of some variables during data collection. 6 In order to minimize the language impact and to consolidate the understanding of all the variables regardless of the normative differences in all the BHS countries, a special BHS workshop was held in November 2017. During this workshop, a data collection manual was jointly drafted. It in detail explained all the terms used in the questionnaire. This step was 4 Initial contacts to the European Homicide Monitor (EHM) and one of its lead researchers Marieke Liem had been made in an early stage of the BHS, but due to time and staff constraints, neither the BHS instrument could be adjusted to the EHM nor were particular variables from the EHM adopted to the BHS instrument. Coordination with comparable studies and validated instruments has meanwhile been achieved and should not only upgrade BHS's future methodology and data analysis but also enable the EHM to significantly broaden its scope toward Southeastern Europe. For more detail on the EHM, see Liem et al., 2013 or Liem &Pridemore, 2012 andLiem in Chapter 2. 5 For Romania, the language used for collection of data via the questionnaire was Romanian due to the assessed low English language proficiency of the national field researchers. All data collected via the questionnaires (in Romanian language) was then transferred into the database and in doing so translated into English language where relevant, for example the short case descriptions. 6 E.g., blood feud was occasionally interpreted as violence between blood relatives. The term conflict (prior to the incident) could be understood either broad or narrow, ranging from a verbal argument all the way to a physical hassle. also necessary because partners from some of the initially selected BHS countries could not participate in the study after all, so partners from other countries stepped in and needed to get acquainted with the study and its methodology. The workshop was also attended by a high-ranking prosecutor from the region who provided valuable expert insights on the type and quality of data contained in case files as well as research questions that would be particularly valuable to criminal justice practitioners.
As basis for the methodological finetuning of the questionnaire during this workshop in November 2017, the Croatian pilot study was used. Again, as a consequence of a hands-on learning approach, the Croatian pilot study was in fact the Croatian BHS study conducted in 2016 and 2017 on a full national sample "gone wrong." This became clear only after the data collection was already completed (paper & pencil style), and all of the data were entered into statistical databases. During data cleaning and preliminary analysis, we found that some crucial variables were missing in the questionnaire, and a considerable share of cases was surveyed unobjectively by one of the field researchers. Hence, the first BHS data collection in Croatia became the BHS pilot study and the BHS questionnaire was significantly improved before repeating the whole exercise in Croatia and starting data collection in the remaining BHS countries.

BHS Violence Typology
After the successful completion of data collection in six countries, all BHS partners, except for Kosovo, delivered their statistical databases to the BC office in Zagreb. For Kosovo, the filled-out paper questionnaires were delivered and data entry into statistical databases was conducted by BC staff. Data cleaning was conducted centrally by the BC office in Zagreb for all the databases and consecutively by two researchers. In the end, the BHS databases in their questionnaire-like form proved to be utterly user-unfriendly, even user-hostile, and successfully evaded any meaningful attempt for analysis. In a nutshell, due to having used a new research instrument and several counting units without beforehand having designed elaborate pathways for later-on data analysis, the BHS databases proved to be far too complex for meaningful analysis. So, for example, in order to identify cases of intimate partner violence and analyze these in terms of gender or age of the offender and the victim and the modus operandi, the relevant relationship variable from the victimoffender database was needed. However, the relationship databases (victim-offender, offender-offender, victim-victim) displayed a huge share of missing data on the one side, whereas, on the other side, the combining of the relationship variables with the relevant variables from the case-, offender-and victim databases simply proved to be too demanding and time-consuming. So, although the data needed for such kind of analysis was clearly contained in different databases, we did not manage to extract all the necessary variables and combine them correctly and meaningfully.
Many sleepless nights down the road were spent puzzling about a sensible solution. Having read through all the descriptive case summaries in an attempt to get at least a general sense about "what's in there," sheer despair led to the idea of an emerging BHS violence typology. Basically, by using the case descriptions (or when these were missing or incomplete by using specific variables to reconstruct the descriptions), a new set of key variables was constructed. Using these key variables, all cases were classified according to the type of violent incident, victim-offender relationship, and motive, as well as particularly cruel, sexual, and affective perpetration. This proved to be not only very user-friendly with respect to data analysis but also in line with the fact that the BHS does not work from a specific theoretical perspective nor does it aim for testing any specific theory on (lethal) violence.
Essentially being an exploratory study, the BHS was in itself designed to start from no theoretical base assumption. So, what better way is there than by taking what you have and after having looked at it in full detail to classify all of it into sensible larger categories/types. Instead of trying to fit all cases into predefined categories based on empirically weak or unfounded prior assumptions or prior research from other parts of the world, the collected data from the actual incidents was used as the basis for developing a corresponding typology -the BHS violence typology ( Fig. 4.1, BHS Typology, 2021). Methodologically speaking, the tailor-fit development of a study-specific violence typology that is in fact rooted in the study's own data appears far sounder and more appropriate, than using any other hypothetical typology, particularly in light of lacking prior research or data on (lethal) violence in the region.
As the BHS violence typology shows ( Fig. 4.1), only 3.6% of all cases were excluded from the overall sample due to not being a (lethal) violent incident at all (mainly false reports). In only 1.5% of all violence cases, the incident was perpetrated in a cruel manner, whereas in only 2.1%, there was a sexual component involved in the incident. In as much as 40.1% of all violence cases, the incident was categorized as an affective (non-premeditated) act of (lethal) violence. Incidents were categorized as affective if the offender acted impulsively and timely closely connected to a prior dispute or conflict with the victim without having had the time or opportunity to "cool down" or plan the violence beforehand. The share of 40.1% affective violence cases simultaneously means that the remaining 59.9% of cases are premeditated.
Looking at the type of violence (context and situation), as much as 36.7% of cases were other private violence 7 compared to 26.1% of other public violence 8 and only 7.6% of bar violence 9 . The remaining two most frequent types were thievery violence 10 with a share of 7.4% and 7.2% of separation violence 11 . Even after combining the remaining types of violence into larger categories (3.9% of self-justice openness of the setting indicates that the victim should feel less safe and be more on guard, while the offender has less control over unwarranted interruptions by others or potential witnesses. 9 The main characteristic of this type of violence implies a larger group of people in a "party atmosphere" which (usually) includes alcohol consumption, fun, and a generally relaxed leisure setting. Although commonly committed in a regular (night) bar, bar violence might also be committed at a home party or a street fair, or in any other location where the same atmosphere is present. If the incident takes place in front of/at the parking lot of a nightclub, it is also considered as this type, since the situational context remains the same. However, the mere location of the incident is not enough for determining this type of violence (e.g., the offender and the victim have an argument over a money debt while drinking in a bar and the offender kills the victim -the money debt makes this case more specific; hence, it is a case of enforcement violence, despite being located in a bar). 10 This type is characterized by the context and setting of trying/gaining financial profit through thievery. These cases usually pertain to robbery, burglary, or theft "gone wrong." The context of illegally gaining financial profit can be found either on the side of the offender or on the side of the victim, the latter indicating that the person being robbed might end up being the actual offender with regard to the violent incident, whereas the case motive would be (self)defense. The decision to combine robbery, burglary, and theft "gone wrong" within thievery violence is based on the finding that the difference between robbery, burglary, and theft cases cannot be determined clearly enough based on collected case file information. Basically, consistency and accuracy of "typing" were chosen over the "specialty" of the three different offenses, since all three unquestionably are thievery in nature when it comes to context and setting. 11 The main criterion for this type of violence is the offender's dissatisfaction over the fact that his/ her intimate relationship is ending or has ended. The main focus is on the "being broken up with." Thereby, the relationship between the victim and the offender is irrelevant -the context and setting counting unit: case; 0.1% missing data for type; 1.6% missing data for relationship; 0% missing data for motive); for more detail, see BHS Typology 2021 violence or 2.9% of crime-related violence), none of them made up a comparably large share as the previously listed five types. On a methodological note, it needs to be stressed that other private and other public types of violence due to lack of more detailed or conclusive information in the case descriptions could not be classified as any of the phenomenologically more specific types. This basically means that within this share of 62.8% of all cases, an unknown distribution of more specific violence types remains hidden.
In terms of the victim-offender relationship (focusing on the status of the victim toward the offender) of (lethal) violence, only 14.7% of incidents can be classified as stranger violence, whereas as much as 40.1% of all cases are domestic violence cases. This finding is in line with findings from previous studies which show that commonly in Europe (Liem & Pridemore, 2012), as well as in the United states (Timrots & Rand, 1987;Riedel, 1987), the vast majority of (lethal) violence relates to non-stranger violence, not to stranger violence. 12 The finding also makes sense with regard to the rather low levels of (violent) street crime in the Balkans (UNODC, 2008), since one would expect a much higher share of stranger violence in case of a higher prevalence of (violent) street crime in the Balkans (e.g., more robbery gone wrong cases). It needs to be highlighted that the BHS stranger category does not include those cases in which the relationship was unknown, but only those cases in which data indicated that the victim-offender relationship was one between strangers. The same applies for all the BHS violence typology categories -lack of data needed for clear categorization was categorized as missing data.
Returning to the victim-offender relationship and in light of the high share of domestic (lethal) violence cases, it will be interesting to take a closer look at this type of victim-offender relationship, particularly related to violence between intimate partners (19.8%). The same goes for (lethal) violence between friends and acquaintances (37.1%). The BHS violence typology distinguishes not only between the typical three broad categories of victim-offender relationships (intimates, acquaintances, and strangers), but also in much more detail captures the different degrees of intimacy between the victim(s) and the offender(s). Therefore, it will be possible to investigate this feature of (lethal) violence in much more detail. The BHS typology enables not only the testing of different variables connected to the victim-offender relationship but also allows for analyzing relevant motives and types of violence (Chap. 5).
Focusing on the motive of the offender, in almost half of the cases (47.4%), this remains unclear, 13 since the BHS violence typology applies a very restrictive are decisive, meaning that the victim of separation violence is not only the offender's (former) intimate partner but may also be the new or a previous partner. 12 For constructive criticism of the traditional concept of stranger violence and empirically founded suggestions for improvement, see Polk, 1993. 13 Unclear motive was assigned to a case whenever it was highly speculative to determine one of the following motives clearly: revenge, greed, (self)defense, vigilantism, jealousy, lack of due classification approach. It assigns (apparent, not actual) motive to a case only if this is rather straight forward. Such a restrictive approach is grounded on the fact that the exact determination of human motive ex post facto and based solemnly on case files, as well as its methodological construction in form of a variable, is highly dubious (at best). But even in the case of "determined," motive cautiousness is requested, since this is merely an apparent motive, not the actual motive, which in essence is only known to the offender, although even this intrinsic insight gets extorted by processes such as neutralization. Out of the clear motives, approximately half of them are revenge 14 (or 26.5% of all cases) and one-quarter is greed 15 (or 12.4% of all cases). As the next chapter will demonstrate, the BHS violence typology allows not only a meaningful data analysis but also enables to test the soundness of the briefly presented typology itself, by cross-checking the violence types with victimoffender relationships and violence motives, as well as numerous other BHS variables.

Cautious Use of BHS Typology and Data
The BHS violence typology was developed by analyzing and "typing" each case on the basis of its descriptive case summary and its relevant variables according to actual violence, violence type, victim-offender relationship, motive, as well as cruel, sexual, and affective perpetration. In the next step, all these initial different case types were combined into phenomenologically meaningful broader categories, up to the point where further broadening of the categories (new variables) would have led to losing the phenomenological specifics of the category itself. All "typing" and "categorizing" decisions were noted and the final typology was validated by an external researcher with a non-legal social science background who was not involved in the BHS or any of the "typing" and "categorizing" of cases, using only the descriptive case summaries and the written typology instructions (BHS Typology 2021). care, hate, or on request. Unclear motive was also assigned when it was not clear which single one of multiple motives is the dominant one or when the case involved too many offenders and/or victims to clearly identify one single main motive. 14 Revenge denotes a motive directed toward getting even with the victim for some kind of wrong that has been committed toward the offender. It is irrelevant whether "the wrong" is essentially banal in its nature or even occurred at all -the perception of the offender that he/she is being the "victim" of some sort of injustice is decisive. Revenge in some instances might seem to overlap with (self)defense, but the difference is that revenge is assigned when "the wrong" against the offender was not a criminal offense/misdemeanor, but rather something more banal (e.g., victim spilled a drink on the offender in a bar). The situation is similar with regard to vigilantism. 15 The offender is motivated by acquiring financial gain (e.g., money, drugs, land, and car) from his/ her actions, risking that by doing so he/she might harm someone. In cases such as debt collecting, if the offender is the debtor, the motive is greed since his/her main goal is to keep the money that is not rightfully his/hers. While contrary, if the offender is the creditor, he/she is motivated by revenge because his/her main goal is to retrieve something that is rightfully his/hers. The highest variance in classification decisions relates to the categories motive (19%), 16 type of violence (16%) 17 , and affective perpetration (13%) 18 . This result should come as no surprise. The variables motive, type, and affective are very tricky and highly sensitive constructs in terms of methodology, and thus realistically difficult to exactly determine, even if one witnesses a violent incident, let alone conducts the categorization based on case files or only descriptive case summaries. The remaining categories display a variance of less than 10% (violence: 1%; sexual: 0.6%; 19 cruelty: 2.6%; 20 relationship 8.4%). The overall 91.3% match in classifying the BHS cases is more than enough to justify a preliminary validation of the BHS violence typology as a methodologically sound and well-functioning analytical tool. Nevertheless, caution is needed with regard to both typology and data due to the "second-hand" nature of the source of data (case files), potential interpretation effects of researchers collecting the data (from the case files), and likely interpretation effects of researchers analyzing the descriptive case summaries. Thus, with regard to motive, type, and affectivity of the (lethal) violence incidents, caution is advisable, even though the BHS typology applies a very restrictive approach.
The presented violence typology ( Fig. 4.1), although aiming at highest possible phenomenological accuracy, primarily targets methodological consistency and clarity with regard to its classification criteria. This basically means that in case of the two aims conflicting, the typology opts for consistency and clarity of classification criteria, instead of capturing all phenomenological details, which are rather noted as missing data/unclear, than forcefully speculated into most likely categories. That is why the BHS violence typology is not to be understood as a general typology of (lethal) violence, but as a study-specific typology designed to capture, analyze, and present the study's first findings in a meaningful manner.
With regard to having introduced a new research instrument (despite its pilottesting on a full national sample, as well as its foundation in a previously used instrument), cautiousness is clearly in place. Thus, as with any comparable (lethal) violence survey based on case file analyses, cautiousness is advisable with regard to the aforementioned problem of a presumably significant dark figure of (lethal) violence. The same applies with respect to missing data, missing case files, as well as substantive and procedural criminal law differences between the different BHS 16 The variance largely relates to a more frequent usage of the category "unclear motive" and also indicates differentiation problems between the categories "vigilantism" and "revenge." 17 The variance can mostly be explained as a tendency to rather use the broader categories (e.g., other public or other private) than the more specific ones (e.g., honor killings and blood feuds, discrimination, separation, or neighborhood). In that sense, the classification is not incorrect, but rather less specific. 18 If including variation caused by more frequent 999 entries (unable to determine due to lack of information), instead of 0 entries (no), the share of different classifications is 23%. This essentially means that the independent validator was less likely to exclude the possibility that the incident might have been perpetrated in an affective manner. 19 Same explanation as provided in Fn. 18 applies accordingly, with a resulting variance of 6%. 20 Same explanation as provided in Fn. 18 applies, with variance in 8% of cases.
countries, but even within the same countries, due to normative changes (offense descriptions as well as sentencing ranges) over time. Due to its relevance for the study at hand, the missing data issue will be explored in more detail in Chapter 5 and in view of the different BHS counting units (see also Appendix). Finally, it needs to be stressed that neither statistical nor normative context analyses were possible for Kosovo, due to lack of data.

Sample
This section will briefly describe the BHS sample, including information on the study's main sampling decisions regarding covered time period, included/excluded offenses (Table 4.1), and scope of sampling within each country, for example, full national or regional sample (Table 4.3). In order to assess the study's representativeness, Eurostat data on officially registered completed homicide suspects is compared to corresponding BHS data ( Fig. 4.2). In order to assess the potential impact of the country-specific normative frameworks, relevant context data and analyses of national substantive criminal law provisions will be presented as well (Table 4.2). This section will be concluding with an overview of basic data collection features and main BHS sample characteristics (Table 4.3). More detailed descriptive analysis with particular focus on lethal and non-lethal violence will then be presented in Chap. 5.

Sampling
Due to being interested in the current features of (lethal) violence throughout the region, the BHS strived for a cross-sectional sample of finally adjudicated cases -a snapshot of more recent (lethal) violence in the Balkans. Therefore, the BHS covers all cases that were finally concluded at the prosecution stage and/or court level within a recent five-year period (except for N. Macedonia and Hungary with a three- Except for Romania, where we opted for a regional sample due to country size/ population and anticipated case load, the BHS aimed for a full national sample, in order to reach a target sample size of approximately 600 cases per country. Those countries with smaller populations (as expected) don't even come close to this sample size, but this would have been different if the initially agreed BHS countries could have been covered (Albania, Bosnia and Herzegovina, Serbia, and Turkey). Now, in terms of comparability of the collected data and ensuring that the BHS covers only those recent (lethal) violence incidents that had been committed (not finally concluded) in the targeted five-year period, we should have sourced case files according to the year the incident took place, not the year of final decision. This was however practically not feasible, since it would have implied that the registry clerks at all prosecution offices and/or courts pick out all the relevant cases by the year the offense was committed (in most countries this information is not even noted in the registries) and then out of these select only those that have been finally concluded, as case file analysis of ongoing proceedings was not possible. Due to this, the BHS sample includes (lethal) violence incidents dating back as far as 1986.
With regard to criminal offenses included, the BHS covers a broad range of (lethal) violence incidents summarized under the categories basic, privileged, and qualified homicide (Table 4.2), whereby it focuses exclusively on adult offenders. The decision to exclude cases involving juvenile offenders was based on practical reasons (difficult access to juvenile courts' case files), while balancing costs and benefits of including such cases in the sample (e.g., additional courts to cover for a very small number of cases). In the majority of BHS countries, the source of the case files (prosecution or court) corresponds to the stage of the criminal proceedings at which the case had been finally concluded (Table 4.3, column titled stage). However, in those instances when court case files could not be accessed or access to prosecution case files was simply far more convenient, case files were (also) sourced from the prosecution, but are in fact a "copy" of the concluded court case file kept in the prosecution's records (including data on the final outcome of the case). The field work for the BHS was largely conducted during 2018, except for the Croatian pilot study which mainly took place throughout 2017.

Statistical Context: BHS Representativeness
In order to assess how representative the BHS sample is with regard to officially recorded incidents of (lethal) violence in each of the covered countries, it was necessary to make more than only one compromise. The main question is which data from the BHS sample should be compared to which data and from what other (most relevant/reliable) source. Due to having included in the BHS sample attempts as well as negligent homicides and keeping in mind the considerable methodological differences in (attempted) homicide definitions even among relevant international sources (Smit 2012, p. 15), as well as in all six official national sources of homicide data, the answer to the aforementioned question is indeed complex and would deserve its own chapter.
In a nutshell, after having analyzed several national and international sources of homicide data and their methodologies, Eurostat data proved to be most suitable and thus available for five out of the six countries (Kosovo excluded). In terms of counting unit, the offender is used, and with regard to matching definitions of homicides, the BHS sample has been reduced to include only completed homicides, while the Eurostat count includes by default all of the following: murder, honor killings, serious assaults leading to death, death as a result of terrorist activities, dowry-related killings, femicide, infanticide, voluntary manslaughter, extrajudicial killings, and killings caused by excessive use of force by law enforcement/state officials. The years refer in both instances to the year in which the suspected offender was reported to the police (BHS) and thereby was counted as a suspect (Eurostat). Now, obviously the assessment of the BHS sample's representativeness has clear limitations, and it thus varies from one country to another due to targeted sample size and country population size, which explains the lower representativeness for Romania (where only a regional sample was collected) and extremely high representativeness for Slovenia (Fig. 4.2).

Sample
For the years 2011, 2012, 2013, and 2014, we observe that the total BHS sample is most representative. For 2012, the BHS sample covers one-third of all homicide suspects in five BHS countries as reported to Eurostat. Bearing in mind that the BHS is a casefile-based study aiming at a first snapshot of (lethal) violence in the selected countries, this is quite an achievement of its own. Even more if one considers that the overall representativeness of the total BHS sample as analyzed for Fig. 4.2 is under a considerable influence of the Romanian sample, or to be more precise, under the influence of Romania's population size, case load and the fact that a regional, not a national sample was obtained. If one were to exclude the Romanian sample, then the representativeness of the remaining 4-country BHS sample would amount to as much as 70% for 2012. Considering the BHS's explorative nature, as well as the far broader definition of homicide applied by Eurostat (which also includes serious assaults leading to death) and the impossibility to account for the representativeness of the overall BHS sample (including attempted homicides and the Kosovo sample), it is not possible to exactly assess the study's representativeness. 21 However, the assessment as presented here might very well be qualified as a "worst case scenario," or in other words present the lower range of the BHS's representativeness.

Normative Context: BHS Comparability
From a methodological perspective, the BHS's normative context is relevant not only with respect to sampling and offenses included/excluded (Table 4.   methodological challenges inherent to all comparative research and data collection that is based on or sourced from different national criminal justice agencies. These challenges relate mainly to varying offense definitions as well as their varying sentencing ranges, on the one side, and to their changes over time, on the other side, meaning that not even within the same country, the offense definitions or the sentencing ranges need to be consistent. Now, the longer the time frame in which the sampled incidents took place, the more frequent the variations in the relevant normative contexts, and thus potentially more impactful their influence on the findings and their interpretation. In case of the BHS, the longest time frame covered by the total BHS sample is 28 years (Table 4.3). Country-wise, we find the longest time frame covered for the Croatian BHS sample (25 years). During this time span, as many as four different penal codes were applicable in Croatia, not even counting all the in-between changes in single homicide-relevant criminal law provisions (definitions and/or sentencing ranges). In order to capture all the criminal law variations in the six BHS countries, while also considering these changes over time, a normative context analyses has been conducted. The main findings are presented in Table 4.2, whereby the sentencing ranges displayed provide the minimal and the maximal proscribed sentences throughout the entire period. 22 The presented findings show the complexity of the normative framework the BHS findings are embedded in, just as much as they demonstrate the scope of variations in homicide-relevant offenses and sentencing ranges throughout the different countries. Clearly, not even the category of "basic" homicide is a clear overlap in all surveyed countries, with variations being even more noticeable in case of "qualified" and "privileged" homicide offenses, both with regard to homicide definitions and sentencing ranges. Looking at the total BHS sample, the possible sentencing outcomes range from 5 to 20 years for basic homicide, 10 years to life imprisonment for qualified homicide, 0.08 to 20 years for privileged homicide, and 0.5 to 15 years for negligent homicide. In methodological terms, one might argue that such normative findings call for weighting of the national sentencing variables in order to enable comparisons of sentences imposed on offenders. However, if one considers that it is not only the individual sentence that "measures" a societies reaction to (lethal) violence but also the sentencing framework provided by criminal law itself, then a weighting of the national sentencing variables would (unjustifiably) distort the findings. Nevertheless, further in-depth research into this matter would be highly feasible, as it is a fundamental question not only for the BHS but also virtually for all comparative criminological research works that are based on criminal justice data and/or sources from different countries. 22 The normative analyses are limited to the special parts of the relevant national penal codes applied on the sampled incidents (offense definitions and sentencing ranges). If one would have included all the relevant provisions from all the applicable penal codes' general parts (e.g. concurrence, mitigation and remission of punishment, or the principle of lex mitior regarding the application of most lenient penal code), then for example in case of Croatia the death penalty as well as the maximum penalty of 50 years imprisonment would also need to be considered.

Sample Description
The BHS successfully sampled a total of 2073 (lethal) violence cases, which include a total of 2416 offenders and 2379 victims (excluding the pilot study). Out of these sampled cases and by using the BHS violence typology, a total of 1997 cases, including 2321 offenders and 2299 victims, were confirmed as actual cases of (lethal) violence. The total BHS dropout of 3.6% of sampled cases, involving 4.4% of offenders and 3.3% of victims, is a consequence of having sampled cases in the prosecution stage, which were finally concluded mainly as false reports and/or as dismissals (due to lack of evidence). Since there was no clear indication that these cases reflect actual incidents of (lethal) violence, they were not considered in the data analysis. The smallest sample sizes as expected relate to the smallest countries (Kosovo, N. Macedonia, and Slovenia), with Hungary, Romania, and Croatia all reaching the initially targeted sample size of approximately 600 cases. Basic data collection information and sample and BHS database characteristics are displayed in Table 4.3.

Field Work
Full disclosure and transparency on empirical research appear impossible in terms of methodology and research operationalization without providing at least a brief overview of the main features and misfortunes of the relevant field work. Notwithstanding the great overall success of the BHS, its modest resources, the quite challenging regional research setting, and limited prior experience in comparable empirical adventures on the side of most national partners are certainly some of the aspects of the field work that may have had an impact on the completeness and quality of the data. A first set of issues relate to the mainly voluntary engagement of the researchers conducting the case file analysis (paper & pencil style). These were mainly (PhD) students who got reimbursed for actual costs encountered due to their field work (travel and accommodation costs), while renumeration for their time/work was symbolic (if at all provided). In many instances, the completeness of the questionnaires and quality of data (especially the descriptive case summaries) depended on the interest and motivation of the single field researchers. Due to mainly having entered the data into statistical databases after the field work had already been conducted, there was only exceptionally a possibility to go back to the source case file for missing data or double-checking "strange" entries. Often times the success of field work also depended on the persistence and social skillfulness of the researchers on the spot, especially in those instances when access to court/prosecution case files was authorized by the relevant official in charge, but then denied or incomplete when showing up as agreed in the prosecutions'/courts' registries. Most data collection had to be conducted at the prosecution/court venues throughout each country, which put additional pressure on field researchers to be as quick as possible, particularly in those instances where registry clerks were not overly enthusiastic about the researchers' presence and the additional workload put on them by having to sort out (and put back) the requested case files. When it comes to interrater reliability, it must be mentioned that the Croatian BHS pilot study was a valuable experience that was shared with all BHS partners prior to their field work. In the pilot study, one of the field researchers consistently reinterpreted case file information from the perspective of the offender instead of collecting the information as documented in the case files. This was detected only after the whole data collection was completed and in the process of entering the data into the database. Based on this experience, all BHS partners were instructed to check the questionnaires periodically for "strange" entries. The vast majority of variables deals with factual information (e.g., dates, legal qualifications, weapons used, and sentence), so the effects of interrater reliability should be minimal in this regard. They are most likely to have had some impact on the short case descriptions and motives as initially covered by the questionnaire. However, tests or analyses checking the scope and possible impact of interrater reliability were not conducted. Since data cleaning and coding as well as classifying the cases in line with the BHS violence typology were conducted centrally at the BC office in Zagreb by two researchers, the impact of differences in ratings at this stage was minimized. Thus, the BHS violence typology ratings were checked independently by an external researcher not involved in the BHS or the designing of the typology -with a rather good result (Sect. 4.2).
Looking at the different BHS country samples, only for the Kosovo sample, additional cautiousness is advisable (see also Appendix). Thus, for Kosovo, the statistical and normative background analysis could not be conducted, and this was also the only country sample delivered to the BC office in Zagreb in paper-version, with limited feedback information from the national research partner.
In conclusion, would we have done the BHS field work differently looking back and could we have anticipated some of the misfortunes? Yes, of course. But would this have been possible within the given circumstances? No, clearly not, and even provided significantly more resources, experience, and staff, neither the completeness nor the quality of the collected data would have been significantly higher, since the majority of missing data is a consequence of the source case files' incompleteness and quality, which becomes most obvious when looking at the data in full detail (Chap. 5).
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.