1. Introduction
‘Why has violence declined?’ – no less than the answer to this laconic question is what American psychologist Steven Pinker promised to give in his highly acclaimed study The Better Angels of Our Nature published in 2011. He described ‘the most important thing that has ever happened in human history’, namely, that ‘violence has declined over long stretches of time.’Footnote 1 Accordingly, the German translation boldly announced in its subtitle that it presented ‘A New History of Humankind’ (‘Eine neue Geschichte der Menschheit’). Pinker's work broadly sweeps across millennia, from anthropological pre-historic discoveries to the present. At the same time, it addresses violence in all its manifestations, from interpersonal violence like murder and manslaughter to state-sponsored violence in the form of brutal punishments and torture, to interstate violence in the form of war. Pinker's argument gained widespread enthusiastic support, including from prominent figures such as Bill Gates and Peter Singer, but also encountered scepticism and opposition. Critics focused either on his narrow definition of violence or on the quality, scope and heterogeneity of his theoretical approach.
His data, on the other hand, appeared to be solid to most reviewers.Footnote 2 Thus, Pinker could confidently respond to a phalanx of critical social scientists that his arguments were not ‘ideological’ but rather ‘empirical’, noting that the ‘diverse datasets showing historical declines in violence […] are well accepted by the scholarly communities who study them.’Footnote 3 On his homepage, however, there is a shift in emphasis. The rhetorical question as to what led him, as a psychologist, to write a work of history is answered with reference to his professional expertise: ‘Actually, I'm an experimental psychologist. Better Angels concentrates on quantitative history: studies based on datasets that allow one to plot a graph over time. This involves the everyday statistical and methodological tools of social science, which I've used since I was an undergraduate.’Footnote 4
As the following discussion will show, Steven Pinker very much reduces the complexity of a field of social-historical research which investigates the long-term development of violence from the Middle Ages to the present on an empirical basis by means of quantification. A large part of the debate revolves around homicide rates which are used to count violent deaths across the centuries. Using statistical methods, this approach claims that we can measure the levels of violence in human societies and thus capture far-reaching societal transformations.
However, the data underpinning the decline of violence thesis has not been subjected to a satisfactory level of critical scrutiny. Our re-analysis of the source material will show that the data is far less reliable than is often suggested. We argue that many broadly accepted conclusions are consequently unsafe and that, for empirical reasons, many key assertions prove to be untenable. This is not to condemn quantitative analyses altogether. Quantification can undoubtedly provide valuable information about historical developments, but hypotheses based on such analyses have to take the limits and the historical context of the data more seriously.
The following discussion begins with a brief account of the rise of the decline of violence thesis in the social sciences in the second section (2). The principal section consists of a step-by-step examination and re-evaluation of the sources and data sets that seem to support the claim of a steady decline of violence (3). As a final step, we present a summary of the results outlining consequences for future research (4).
2. Decline of violence – the career of a concept
For several decades now, a seemingly infallible measure has informed the history of criminal violence dealing with long-term trends in violent acts: the homicide rate, i.e. the number of killings per 100,000 residents. The underlying assumption is that killings are quite consistently punished and documented offences with a comparatively low rate of underreporting, unlike the much larger and more diffuse field of violent crimes in general. By and large, this presumption also applies to the pre-statistical era prior to the late 1700s or 1800s when authorities began to compile and publish crime data systematically. Moreover, court and criminal records have been kept in an increasingly consistent and sequential manner since the late medieval period. Thus, it should be possible to count homicides over the course of time. Furthermore, as will be discussed below, those numbers are usually presented in relation to population numbers, and homicide rates then indicate the overall occurrence of lethal violence from a comparative perspective. In 1981, criminologist Ted Robert Gurr was the first to sum up the fragmented findings from historical studies on violent crime in England, projecting the results onto a timeline.Footnote 5 Two years later, Gurr's approach inspired Lawrence Stone's oft-quoted summation ‘that medieval English society was twice as violence-prone as early modern English society, and early modern English society at least five times more violence-prone than contemporary English society.’Footnote 6 More studies followed, including Cockburn's exceptional long-term investigation on homicides in Kent based on dense, serial sources.Footnote 7 By the mid-1990s, the decline of violence thesis (hereafter ‘decline thesis’) constituted the heart of a modernisation theory reaching beyond the English example and claiming validity for at least the entire Western world. An anthology edited by acclaimed international criminal historians entitled The Civilization of Crime represented an important milestone in this field of research. This volume claimed the view that interpersonal physical violence had decreased from medieval times until today was a matter of broad consensus.Footnote 8 Pieter Spierenburg's contribution to this collection is worth emphasising here for two reasons.Footnote 9 Firstly, his serial analysis of crime and violence in early modern Amsterdam provides major evidence of the decline thesis outside England. Secondly, Spierenburg, one of the most important Dutch students of the late Norbert Elias, used the theory of the civilizing process as a closed conceptual framework to explain the decline thesis.Footnote 10
In 1996, however, many historians deemed the decline thesis controversial, several of whom contributed to The Civilization of Crime. Cultural historians, who in particular rely on micro-level historic methods, criticised both the reliability of the sources and the conceptual framework as replete with flaws. Moreover, countless experiences of war and mass murder in the ‘Age of Extremes’ (Eric Hobsbawm) seriously query whether there has ever been a steady decline in violence.Footnote 11 Attempts to start a more in-depth discussion a few years later ran quickly aground.Footnote 12
At the same time, criminologist Manuel Eisner was collecting quantitative data on homicides from criminal history research on various European regions for the purpose of a ‘re-analysis.’Footnote 13 The unprecedented scope and density of his data made his studies quickly the central reference work in the field of violence studies. More importantly, the way in which he presented his data backed the decline thesis and remained widely unopposed. Even if some historians referred critically to the limited validity of those figures calculated from premodern sources or emphasised the rise of interpersonal violence in Western societies since the 1960s, they still acknowledged the stark contrast in homicide rates between premodern and modern times.Footnote 14 Eisner, although conceding that there are some problems regarding the source material, stressed that the thesis of a long-term decline in homicide rates is built on empirically firm ground: ‘Accordingly, the quantitative data discussed above should not be regarded as precise measurements. However, they show that the huge amount of sophisticated historical work done over the past 30 years has resulted in a remarkable convergence as to a number of secular patterns in lethal violence that cannot plausibly be interpreted as a result of systematic distortion.’Footnote 15 In a series of articles, Eisner has since expanded his data set and argumentation without fundamentally changing his framework.Footnote 16 To an extent, the final result of his work can be seen in Figure 1.
Eisner, European homicide trend, 1200–2010.
Source: History of Homicide Database.Footnote 18
Eisner's synthesis was adopted as a key reference by the principal scholars undertaking historical research on violence. Unsurprisingly, this is also true for Pieter Spierenburg's A History of Murder published in 2008.Footnote 17 Released in the same year, A History of Violence penned by French historian Robert Muchembled also followed this line of reasoning. ‘From the thirteenth to the twenty-first century, physical violence and brutality in human relationships were on a downward trajectory all over Western Europe’, reads the first sentence of his book. Muchembled's reference to Manuel Eisner states categorically: ‘The figures are accepted by all the specialists’.Footnote 19 From this perspective, Steven Pinker's best-selling work seems to mark the final triumph of the decline of violence paradigm.Footnote 20 As far as the confined historical debate about homicide rates is concerned, the psychologist can indeed assert that the decline thesis is widely accepted by historians.Footnote 21
On closer consideration, however, this assertion quickly proves to be hardly convincing. Many of the criticisms that will be systematically examined below have already been raised in numerous publications. Weighty objections to the over-simplified master narrative of the decline thesis have recently been formulated by Peter King. In his detailed comparative studies, for instance, he was able to prove a decades-long rise in homicide rates in some areas of northern England and Scotland in the 1820–40s.Footnote 22 Furthermore, the decline thesis is of very little help for research on violence in the United States, where historical developments have consistently been described as exceptional.Footnote 23
A critical attitude toward the decline thesis is common among historians who prefer a more cultural-qualitative approach. Yet, some pivotal works of this field of research do not even mention quantitative perspectives. One historiographical overview in a major interdisciplinary German volume on violence, for example, completely fails to address these approaches.Footnote 24 This is not merely remarkable, but a very serious problem: disregarding important parts of the discussion only deepens the divide between different approaches and artificially inflates the success of the decline thesis. Historians like Spierenburg or Muchembled, arguing on a macro level, at least attempt to integrate the micro level of social practice, such as the premodern culture of honour, into their models. By contrast, works that argue from a cultural historical point of view often fail to discuss the decline of violence paradigm beyond sweeping criticism.Footnote 25 Attempts to take a balanced position between the two poles are rare.Footnote 26 Thus, the field of historical research on violence is divided into two opposing camps. On the one hand there is an interdisciplinary mainstream that rests upon empirical mass data and supports modernization theories. On the other hand, there are cultural historians who prefer qualitative approaches and who are sceptical toward any modernisation theory.Footnote 27
Such a polarization not only characterises the historiographical debate, but also shapes research into contemporary violence.Footnote 28 Ignoring the challenge posed by alternative perspectives and approaches serves to stifle debate and prevent critical analysis or reassessment of the foundations of different theories. Straightforward assertions of the dogma associated with certain schools of cultural studies, that master narratives of modernization have outlived their usefulness, cannot replace the critical debate that we seek to promote. We will limit ourselves in the following to the late medieval and early modern periods, which are pivotal to the arguments in favour of a long-term decline in violence. In addition, we will centre upon England, still the paradigmatic case, upon the German territories and the Nordic countries; other areas will be included as appropriate. The European colonies as well as the fledgling United States are excluded, as our discussion follows the protagonists of the decline thesis and their focus on Europe. Yet, in view of the pertinent research on North America, additional critical objections could be raised against the validity of the decline thesis.
3. A critical reanalysis
The following analysis starts with Manuel Eisner's database of European homicide rates and highlights its underlying historical sources (3.1). The next section discusses critical objections to the validity and reliability of these sources (3.2). Even though these objections cast doubt on the decline thesis, further issues need to be addressed in order to fully understand its problematic empirical foundation: first, the often-patchy source material and its sometimes arbitrary and heterogeneous analysis (3.3), second, the uncertainty of the population estimates (3.4) as well as, third, issues arising from the combination of homicide counts and population figures (3.5).
3.1 The data and its sources
The starting point of our analysis is Manuel Eisner's database, which is being continuously expanded, and was made accessible to us in July 2016.Footnote 29 According to Eisner, the ‘premodern homicide data set’ included 390 ‘observation points’ in 2003 and 823 such ‘local estimates’ in 2014. They were extracted from an impressive total of 115 historical case studies.Footnote 30 The local estimates differ from each other significantly regarding their spatial and temporal distribution. This is, for one thing, due to the varying number and quality of available sources, but also because only a fraction of these sources has been taken into account by scholars so far. The database is divided into European regions and the results are displayed in tables or graphs. For ‘Netherlands/Belgium’, the database contains about 225 observation points, for ‘England’ about 185, while other regions contain considerably fewer.Footnote 31 Although their temporal distribution varies substantially, some clear patterns can be observed. The data for the late medieval era (late twelfth to fifteenth centuries) is appreciably more fragmented than for the early modern period (sixteenth to eighteenth centuries). The extremely uneven availability of data from late medieval times is reflected by the fact that there are some dramatic gaps, most notably for England: virtually no source material survived for the period from c. 1350 to 1559 that could be used for calculating homicide rates. Toward the end of the fifteenth century, some coroner's inquests do begin to occur but they are too fragmentary to serve as the basis for homicide rates.Footnote 32
These first observations direct attention to the dataset's sources. Contemporary aggregate statistics did not exist until the mid-1700s. In 1749, Sweden-Finland introduced pioneering population statistics that can be used to reconstruct homicide rates.Footnote 33 Genuine crime statistics followed toward the end of the century.Footnote 34 For Middlesex, Criminal Registers are available from 1791 onwards, for the rest of England and Wales from 1805. Only since 1827 have annual crime statistics been published in France. In 1829, the grand duchy Baden became the first German territory to follow this example.Footnote 35 For the pre-statistical era, therefore, researchers are dependent on judicial sources that can be used to compile statistics. This material is extremely heterogeneous and its validity must be carefully contextualized.
The most prominent group of sources are registers of official post mortem examinations for sudden deaths. In England, inquests were conducted by coroners, whose reports furnished the basis for indictments. These formal indictments, complemented by a few coroner's inquest reports, form the basis for the densest series in Eisner's database. They were published in 1991 by Cockburn for Kent and span the period from 1560 until 1981. In general, coroners’ inquests are valuable sources. Nonetheless, there are distorting factors. For example, if a suspicious death was reported, it depended on the coroner's and the jury's verdicts whether a case would appear as a homicide, thus affecting any retrospectively compiled statistics. Deficient medical knowledge among coroners could lead to unreliable results, as could the widespread practice of bribery.Footnote 36 Conflicting interests between coroners, almoners and local communities could also influence the recording of violent crimes. To what degree these circumstances exercised control over coroners’ work also varied depending on the time period, which again could affect how many documents survived for each period.Footnote 37
The system of post mortem examinations was known in other European regions as well, but not practiced as consistently as in England and certainly not as systematically recorded. A remarkable exception is Amsterdam, whose homicide rates have been introduced in Pieter Spierenburg's research as the second paradigmatic case after Kent. Inquests from the Dutch metropolis have been passed down from as early as the sixteenth century. Until now, they have only been examined for the period 1524–1565 and for the years 1560, 1570, 1580 and 1590, for which Spierenburg calculated homicide rates of approximately 21 to 28.Footnote 38 From around 1600, records of inquests run dry. Sources allowing a long-term study of violent deaths are not available again until 1666. After that date body inspection records, with few exceptions, reach into the nineteenth century. These records only document the medical cause of death, for instance lethal injuries, unfortunately omitting any circumstances and whether the death was the result of a natural cause, an accident, a suicide or a homicide. Therefore, Spierenburg himself, guided by the listed injuries, decided whether a death is counted as a killing and therefore included into the homicide rate or not. This method may reliably count a homicide in cases of knife wounds to the back, but quickly reaches its limits when injuries result from dull objects, single bullets or lacerations, which may just as easily suggest a suicide or an accident.Footnote 39 For this reason, homicide rates based on post mortem records should be treated with caution, the more so as they contain anomalies that do not fit into the master narrative of a steady decline.Footnote 40
In Germany, the only source of this type stems from Cologne, where the municipal police created a Totenbuch, a registry of unusual deaths.Footnote 41 It included the cause of death, thus permitting a more precise breakdown of violent killings. Despite covering the period from 1557 to 1728, it was only kept consistently enough to warrant evaluation in certain periods over these years.Footnote 42 Sweden-Finland apparently carried out inquests beginning in the second half of the seventeenth century, but they have not been thoroughly evaluated, nor have the parish burial records, available from 1722 on, in which homicides were also recorded.Footnote 43
The specific form of death registers varies from region to region, causing numerous assessment and interpretation problems. This applies even more to other crime and judicial sources, which can be demonstrated by the example of the Holy Roman Empire.Footnote 44 Municipal authorities were the first to introduce records on criminal cases at an administrative level, followed by territorial sovereigns and dukes. Cities such as Stralsund in the north or Nuremberg in the south kept proscription books (Achtbücher) from the beginning of the thirteenth century. Their contents are heterogeneous. They contain, first of all, the names of perpetrators who had fled or ignored multiple summonses and whom the authorities wanted to force to appear in court by means of proscription. Secondly, they record both temporary and permanent banishments from the city due to criminal violations. Finally, the proscription books contain prisoners’ truce declarations (Urfehden), i.e. formal proclamations of prisoners to forgo revenge against the city or other parties involved. These early sources therefore do not result from a need to document crimes and sanctions as such, but rather from very specific legal requirements. It was usually more important to record the names and punishments than the corresponding crimes, which, moreover, were hardly standardised. As time passed, these proscription registers not only became more detailed, but also new forms of records occur. Account books, for example, provide information on income through fines and criminal justice expenses, such as payments to the executioner. In some cities, like Regensburg, special registers recorded fines paid by violent offenders, which gives insight into everyday violence in the late Middle Ages. At other places, authorities began to list offences of those sentenced to corporal punishment (Urgicht- und Malefizbücher). Overall, these trends in administration led to a comprehensive documentation of judicial activities and detailed records of interrogations and witness testimonies in at least some places.
Pre-statistical source material is similarly fragmented for the Kingdom of Sweden, including the area today belonging to Finland. The first useful evidence related to homicidal violence is found in so-called notebooks (tänkeböcker) dating from the sixteenth century on. They served in part as municipal court transcripts and contain comprehensive information about the criminal cases heard there. Individual examples date back to the medieval era, but the survival rate is very poor.Footnote 45 Another important source type is the relatively consistently kept fine rolls (saköreslängder), available from around 1540 onwards.Footnote 46 Since manslaughter was generally punished by paying a wergild (mansbot) and only rarely resulted in execution,Footnote 47 fine rolls can certainly be employed to analyse homicides. However, they rarely include details regarding the circumstances of the crimes.Footnote 48
The source material improves significantly for the first half of the seventeenth century when courts of appeal (hovrätter) were established in several cities of the realm, for example in Stockholm (Svea hovrätt, 1614), in the Finnish city of Turku (Åbo hovrätt, 1623) and in Jonköping (Göta hovrätt, 1634).Footnote 49 These courts of appeal served as the highest judicial authority after the monarch until the establishment of the supreme court (Högsta domstolen) in 1789 and were concerned with supervising the lower courts. Another central task was to review the rulings for capital offenses in their jurisdiction and, if necessary, to reduce or increase the sentence.Footnote 50 This meant that municipal and rural courts essentially lost their right to impose death sentences and corporal punishment and were forced to record trial procedures in writing.Footnote 51 Of the three courts of appeal named above, however, only the records from the Göta hovrätt have survived in their entirety.Footnote 52 Still, their introduction constituted great progress, as court records are at least partially available in serial form from that time on.Footnote 53 For most of the early modern period, the source material used in Finnish and Swedish research is, therefore, mainly composed of municipal notebooks, fine rolls and records of courts of appeal. We have to bear in mind, though, that especially the first two types of these sources vary much in quality and local survival, notwithstanding the early centralisation of the Kingdom of Sweden.Footnote 54 As mentioned above, from 1749 onwards, the Swedish census also included information on cases of death and about their particular causes.Footnote 55
These are just a few examples of the source material's diversity and fragmentation. Were we to include other regions in our analysis, the picture would become even more multifaceted. Instead, we will now consider to which extent the selectively used heterogeneous sources provide valid and comparable data for calculating homicide rates.
3.2 The heterogeneity of the sources – basic objections
The heterogeneity of the sources alone renders a comparison of homicide rates over different periods of time difficult. First of all, every source reflects a certain segment of ‘unnatural deaths’, depending on norms, conventions and administrative practices. No relevant sources are available prior to the late thirteenth century. Laws and practices of criminal justice varied greatly among the different European judicial systems. A general, and problematic, tendency is that ‘manslaughter’ in medieval Europe did not constitute a criminal act which demanded corporal or capital punishment.Footnote 56 In Sweden-Finland, for example, in addition to the wergild recorded in the fine rolls, sanctioning was also prevalent as extra-judiciary compensation paid by the perpetrator or their family. This practice survived well into the early modern period.Footnote 57 However, these ‘private’ compensations, which were usually exorbitant and often approximated the cost of a medium-sized farmhouse, do not necessarily appear in the sources.Footnote 58
The omission of some homicide cases aside, many issues still remain with the recorded cases. Uncertainties often begin with the question of what exactly has been counted by the authorities: deaths or victims (as in inquisition records), or perpetrators/defendants as in most other types of judicial sources. The decision whether to count either victims or offenders or defendants has far-reaching consequences, since the corresponding figures can differ significantly.Footnote 59 In addition, there are the various categories of criminal acts. Should just manslaughter be considered, or (premeditated) murder as well? Are accidental killings or deaths occurring in the context of feuds and vendettas also to be included in the statistics (and what would the modern equivalent for the purpose of comparison be)? What effects did infanticides have on the total number of homicides? Sometimes they are reflected in the homicide rates,Footnote 60 more often they are explicitly excluded on the basis of the variability of their prosecution in time and space.Footnote 61 Sometimes, however, sources do not differentiate between infanticides and other homicides, making the exclusions of infanticides impossible. And how should we handle lethal violence in the context of robberies and burglaries?Footnote 62 Quite often, information about how such crimes were prosecuted is missing.
Clearly, the categorization of crimes greatly affects the calculation of homicide rates. There are, for instance, different opinions as to whether the verdict of English coroners’ inquests have always been transferred to the indictments without alterations. Some findings indicate that killings declared unambiguously as accidents in the coroner's inquest were nevertheless tried as murder in court and therefore contribute to the homicide rate.Footnote 63 Another example are cases of traffic accidents. From the end of the eighteenth century, in England and Scotland, killings caused by careless driving, especially of carts, increasingly appeared in court documents as manslaughter, whereas in modern statistics, these cases of involuntary manslaughter would not be included.Footnote 64 There are similar circumstances to be found in Sweden-Finland, where, until the legal reforms of 1734, the category dråp (manslaughter) also encompassed assaults and accidents resulting in death regardless of the offender's intentions. According to Arne Jansson, the later narrowed definition of manslaughter contributed significantly to the decline of homicide rates in Stockholm in the second half of the eighteenth century.Footnote 65 The decrease in violence in this phase was therefore, at least partially, simply a consequence of changes in judicial parameters.
More fundamental are objections based on the availability and the use of weapons and the state of medical knowledge and expertise. As James Cockburn discovered for Kent, deaths in most lethal violent crimes of the sixteenth and early seventeenth centuries resulted from stabbings. Since the seventeenth century, guns and other devices have become more prevalent. Cockburn regards the fact that in the mid-nineteenth century most killings were the result of beatings rather than shootings as clear evidence for how the broad availability of guns and especially knives and swords contributed further to the high homicide rates in the medieval and early modern period.Footnote 66 Moreover, most violent deaths in the early modern period occurred as a result of long-term consequences of injuries like blood loss or infections. In the late twentieth century, many of these victims would undoubtedly have survived. A comparison over long stretches of time would thus have to include categories from modern crime statistics such as ‘attempted murder’ and ‘aggravated assault’ in addition to manslaughter and murder.Footnote 67
Another bias that systematically distorts comparisons of homicide rates over time is the demographic structure. Young men between the ages of 16 and 30 are most frequently responsible for violent crimes in nearly all societies and periods. The substantial spatial and temporal differences in age and gender structures between premodern and modern times clearly influence the homicide rates. The same applies to the different demographic structures of rural areas. Using age-standardised homicide ratesFootnote 68 or rates that only include the adult population (in order to account for extreme fluctuations in the number of children in a given society)Footnote 69 can at best be a solution when sufficient information about demographic structures is available. For the early modern period and especially for medieval times this is not the case. We know very little about population structures, not to mention short-term factors influencing demographics such as migration,Footnote 70 epidemics, and other crises.
Proponents of the decline thesis have acknowledged most of these problems, discounting them, however, as insufficient to justify any general opposition to the long-term decline in violent deaths. As idiosyncratic as individual cases may be, the results of a large number of studies that point broadly in the same direction, can be considered reliable, according to the underlying assumption.Footnote 71 Detailed engagement with conflicting opinions is rare. One of the exceptions is the remark of a sociologist working on the phenomenon of present-day violence and attempting to defend Manuel Eisner's data against criticism. Wolfgang Knöbel asserts that:
there is, naturally, a large proportion of unreported cases and many factors that falsify the ‘true’ number of crimes. Those who present these arguments, however, have to explain why the data should be distorted systematically in one direction only, that is, only toward a reduction in violence. If this is not plausibly shown – and to my knowledge no one has tried this yet – then the trend cited here cannot be contested.Footnote 72
Several assumptions form the basis of this claim. One of them is the belief that, given the large amount of data, possible source problems cancel each other out. A second one is that the dark figure for homicides was quite high and the available figures consequently represent only low estimates. This is based on the assumptions that the criminalisation of fatal violence was a gradual process and that there was a higher number of unreported cases in premodern times than in the modern era. Such arguments could indicate that a critical re-assessment of the sources does not weaken the decline thesis in any case, but rather makes it stronger. At most these arguments are valid until further challenges are considered. We will, therefore, turn to substantially more problematic issues concerning the collection and collocation of data in the following sections.
3.3 Taming the sources – constructing evidence
Our first general objection emerges from the fact that serious gaps in the available source material are sometimes treated in a methodologically questionable manner. When considered closely, the ‘number of homicides per year’ for the most part appears to be a profoundly artificial figure, the result of quite arbitrary calculations differing from source to source. First of all, a basic differentiation must be made between individual data points and a series of data including figures from several consecutive decades. A series seems more valuable because, all other things being equal, developments can be observed for a certain geographical unit in a certain period of time.
In terms of serial sources, the situation is highly diverse. For that reason, we might divide the premodern era pragmatically into the medieval period until around 1500 and the early modern period, starting from 1500. As a basic principle, the availability and usability of records is much worse for the medieval than for the early modern period, which can be explained in part by an upswing in civil government and administrative record keeping. This is especially evident in Sweden-Finland, where there are hardly any records available for periods before 1500. For medieval England the situation is similarly poor, particularly (as mentioned above) for the years from ca. 1350 to 1559. No source material is available here to construct homicide rates. Furthermore, such data that underlies research on violence for the thirteenth and the first half of the fourteenth centuries is mainly based on the analysis of eyre rolls (circuit court records) and gaol delivery rolls.Footnote 73 But these sources have only survived in fragmentsFootnote 74 and provide merely individual data points rather than longer-term series.
The data set for medieval Germany looks completely different.Footnote 75 From a total of 93 data points, more than half date back to before 1500, which is undoubtedly due to the fact that German medieval sources are more often edited and are thus more accessible to modern evaluation. For this period, Eisner's data set contains extremely heterogeneous figures, ranging from high numbers, between 60 annual cases of manslaughter per 100,000 inhabitants (Freiburg im Breisgau) and 90 (Krakow), to more moderate numbers like those in Regensburg (25), Nuremberg (20) or Gorlitz (19), and many low figures, between 10 and 13, as in various Hanseatic cities in the north or even 6.2 in Konstanz. These heterogeneous findings are suggestive mostly because the numbers tend to be lower, at around 10, in the early modern period. Closer examination of the data reveals many pitfalls. Usually, a standard data point is supposed to reflect a 10-year period. Due to the fragmentary sources, this is more often than not impossible. Therefore, longer intervals are frequently summarised in one data point, as is the case for the cities of Liegnitz (19 years), Regensburg (25 years), and Nuremberg (99 years), rendering any internal differentiation invisible.Footnote 76
Eisner treats available serial data sequences for late medieval Germany rather carefully. For Augsburg, there is a data sequence from 1338 to 1399.Footnote 77 In the first decades, it yields high homicide rates of 23 to 35. For the last two decades of this sequence, the figures indeed decrease dramatically to 8.6. In Eisner's published graphs, these figures are not differentiated, but rather part of an arithmetic total average. Consequently, intriguing questions about a potential decline of violence in the fourteenth century are not examined. There is a similar situation with Martin Schüßler's problematic analysis of Nuremberg's proscription records. He calculatesFootnote 78 homicide rates for those years where killings are recorded and arrives at figures of at least 20 and up to 65. All years that lack records on homicides were left out under the premise that those records were carelessly documented, lost, or in any case obviously incomplete. However, many of the years in question did contain sufficient data for other crimes to make these assumptions questionable. If these years are included and only those without entries of any kind are excluded, then the picture is completely different. The homicide rates for the ten-year intervals then amount to numbers between 1.5 and 12.5, or on average of 7.5 over an 80-year period. Manuel Eisner, who has evidently not followed Schüßler's method here, derives a homicide rate of 20 for Nuremberg for almost 100 years.Footnote 79 Taken together, these procedures show the explicit or implicit tendency of some researchers to consider the highest available data as the most ‘realistic’.Footnote 80 Obviously, high numbers for the late Middle Ages serve as a backdrop that helps corroborate an important step toward ‘civilization’ in the early modern period. However, as these examples clearly show, the data for the medieval era are overall too narrow, heterogeneous and ambiguous for any reliable assertions about the development of lethal violence.
For the early modern period, the figures for Germany are hardly much better. So far, data only exist for the Duchy of Braunschweig-Wolfenbüttel and for the imperial cities of Cologne and Frankfurt am Main.Footnote 81 A ‘Report of All Criminal Cases’ (Repertorium aller Kriminalsachen) in Braunschweig-WolfenbüttelFootnote 82 provides data points for a few ten-year periods, the first of which starts in 1590 and is conspicuously low with 5.6, while the others, starting in 1600, rise to almost 18 and then sink once again. We also lack the information to assess to what degree this development could have been influenced by specific historic conditions such as internal conflicts or economic tensions. In respect of the cities, Cologne is especially important for an inter-regional comparison to Amsterdam and Kent due to the existence of the Totenbeschaubücher. Unfortunately, the records for Cologne were not as consistently kept nor with equal care and attention as in Amsterdam or Kent (Table 1).Footnote 83
Registered violent deaths in the city of Cologne (time periods with dense records)
Source: Totenbeschaubücher, Cologne. The number of 40,000 residents is the basis for calculation for the years until 1620. For the last row, 50,000 inhabitants were used; Gerd Schwerhoff, Köln im Ancien Régime (Köln, 2017), 45.
The most robust figures for Germany are those which are based on post mortem examinations and they remain within a quite moderate range. They do not vary dramatically and stay roughly at a same level until around 1700. The figures for Frankfurt, on the other hand, are reliable at best only for the eighteenth century. They are based on Joachim Eibach's analysis of a sample of the Criminalia. Starting in 1701, the homicide rate was 7.5, quite similar to Cologne, then reached a peak of 11 in the middle of the century, before sinking in the second half to 2.Footnote 84
Available data for the early modern period in England are more homogeneous. Sources stand out positively in terms of a comparative pan-European analysis of homicide rates. Starting in 1559, serial data deriving from assize records and indictments have been preserved for the counties of the Home Circuit (south-east England), and from about 1650 for other regions as well. Though, if one takes a closer look, gaps also appear in the English records.Footnote 85 Sharpe and Dickinson, for example, had to completely exclude the first half of the eighteenth century when they quantified homicides in Cheshire, since the source material was too patchy.Footnote 86 The assize records for Surrey, another well-researched county, are also fragmentary for the period between 1660 and 1730. As a consequence, Beattie's corresponding homicide rates, calculated in 20-year intervals, are based on records from only six or seven very unevenly distributed years, which may be only years with exceptionally high (or low) homicide counts.Footnote 87
In the case of Sweden-Finland, various regions have been at least cursorily examined in terms of their level of violence, although not all of them have been included in Eisner's database. Some of the areas that are missing show low homicide rates, thus indicating a relatively low level of violence for the sixteenth century, for example Jonas Liliequist's findings for northern Sweden. According to his calculations, the homicide rates in the three Norrland provinces of Helsingia, Ångermanland and Medelpad never exceeded the comparatively low rate of 4 during the sixteenth century. The rather moderate or even extremely low figures for some Finnish provinces displayed in Table 2 have not been taken into account either. Probably not intended by Eisner, this ‘gap’ again seems to confirm researchers’ general tendency to prioritise high homicide rates.Footnote 88 Even though the diverse sample of Scandinavian regions appears to be promising in terms of a closer examination, this does not hold up under closer scrutiny, since only few regions provide enough material to reconstruct serial homicide rates that would actually qualify for a credible long-term analysis. By contrast, only a very limited amount of data can be gathered. As a result, often only one to three short term homicide rates per geographic area exist. Unsurprisingly, then, corresponding findings are not only fragmentary and heterogeneous, but seem remarkably arbitrary at times. Comparing the small city of Vadstena in Östergötland and the Åland Islands (Finnish: Ahvenanmaa) offers an admittedly extreme example. While Vadstena's homicide rate was, according to Karonen, astronomically high (100) in the early 1540s,Footnote 89 the figure for the Åland Islands presented by Heikki Ylikangas, was less than 1 between 1537 and 1569, which is markedly low even by today's standards.Footnote 90 Well-founded assertions about long-term developments can hardly be made with data that paint such a fragmentary and disparate picture.Footnote 91 Ylikangas’ table of homicide rates for several Finnish regions illustrates pars pro toto the heterogeneity of the findings (Table 2).
Homicide rates in Finnish provinces and regions in the sixteenth centuryFootnote 93
But there are also exceptions: some cities possess relatively comprehensive data sequences. Source material from Turku (Åbo), then the largest city in Finland and the second largest in the kingdom, as well as from the small Swedish cities of Arboga and Vadstena stretches back to the medieval period and covers multiple decades or even centuries.Footnote 92 Unsurprisingly, the most abundant sources are from Stockholm. Boasting over 30,000 residents by the time the Swedish Empire had grown into a great power (stormaktstiden), it was by far the largest city in the early modern north.Footnote 94 Homicide rates for Stockholm can be calculated from 1475 to 1773 (with a few interruptions), thus encompassing a period of almost 300 years.Footnote 95 The overwhelming majority of the other regions, especially rural ones, are far from providing such comprehensive source material.
All in all, what seems to be an extensive set of data at first sight, reveals itself as a hodgepodge of the most varied, fragmentary information. In addition, the few cities mentioned above are clearly overrepresented, despite the country's vastly rural structure. The figures for the city of Stockholm, which was already atypical due to its size, particularly distort the picture as we have to bear in mind that in early modern Sweden-Finland over 95 percent of the population lived in the countryside.Footnote 96 Moreover, even where there is (partially) sequential data, its figures are subject to severe fluctuations and do in no way indicate a clear reduction in violence. While homicide rates in Stockholm, Turku and Arboga truly seem to diminish from around 1600 onwards, they climbed to a record high of 120 in mid-seventeenth-century Vadstena.Footnote 97
Without question, the source material is anything but flawless. Moreover, two (frequently combined) interpretive approaches add to the set of problems. First, records yielding low figures are often declared incomplete, which allows scholars to exclude them without further comment. Second, researchers tend to favour high figures. Schüßler's calculations for medieval German cities, mentioned above, constitute a drastic example. Similar instances can be found for other regions, such as in Hammer's study on Oxford. In order to calculate homicide rates, Hammer used coroners’ rolls, which he only assessed for the very limited period of six years. The first and the last year yielded significantly lower homicide counts than the rest of the years. This is why Hammer considered this finding a result of incomplete records. As a consequence, he left these two years out and used only those four years with high killing counts (1343/4–1346/7) for his calculation of a homicide rate. In combination with Oxford's small population of approximately 6,000, this resulted in a rate of 110.Footnote 98 This number has been adopted, for example, by Gurr and Stone without reference to its problematic origins.Footnote 99
This is not an isolated case, as the persistent suppression of low figures results from the idea that the recorded homicide cases always represent a minimum number – the higher the homicide counts, the closer they supposedly correspond to the ‘real’ number of killings. Figures that seem too low are therefore skipped over with the interpretation of incomplete source material, even when there is no clear evidence to support such a presumption. This leads to a systematic bias in favour of higher figures and higher homicide rates for premodern centuries. Somewhat paradigmatic for this approach is Spierenburg's assertion that ‘court records’ were only reliable as a source when no ‘body inspection records’ were available and
if the court records yield relatively high homicide rates. […] No conclusion can be drawn from underreported figures if they are low in a period when we had expected them to be high. If, on the other hand, the rates in early periods, even though underreported, are extremely high, they definitely indicate a trend.Footnote 100
To paraphrase this approach: It all adds up to trends and expectations, not to say expected trends.
3.4 How many people? Reconsidering the dynamics of population figures
It is hardly surprising that counting premodern crimes from fragmented and selective sources, in our case violent killings, is only one of many challenges. In order to produce actual homicide rates these killings have to be viewed in relation to the corresponding population count. Here lurks another source of error at least as significant as deficient crime data. As we will show, population estimates are imprecise to such an extent that they can in no way serve as a solid basis for calculations.
According to modern criminology models, the rate is always calculated as ‘killings per 100,000 residents’, a figure that in today's context marks the lower boundary of a major city. This alone gives rise to several questions. A very small number of cases can be transformed into monstrous rates if modern standards are applied. An example of this is Kasimir, a small satellite city of Krakow, where 21 killings in 31 years results in a rate of 45 killings per 100,000 residents.Footnote 101 Similar cases are, as mentioned above, Oxford, with its approximately 6,000 residents, resulting in a homicide rate of 110 (for only four years)Footnote 102 and Vadstena, which reached a rate of 120 in times of declining population, when only ca. 400 residents (previously more than 1,000) remained.Footnote 103 Statistically such a procedure may not be problematic, and may even seem unavoidable for the purpose of standardisation. Evidently, one runs the risk of distorting the proportions for the period around 1500, when there were only four cities in Europe (Paris, Naples, Milan and Venice) that even reached a population of 100,000.Footnote 104
Yet this is a comparatively minor issue. The real problems are much more fundamental. Arriving at exact population figures for a specific place at a specific time is a difficult, sometimes almost insurmountable, task. Above all, what is true about crime statistics also applies to population statistics – they are products of modernity. To be exact, they only began to evolve around a time when European authorities implemented bio-political principles (in Foucauldian terms) that saw ‘human resources’ as a crucial factor for any authority. In Germany, this development dates from the period after the Thirty Years’ War, whereas in most parts of Europe eighteenth century population policy was of the essence.Footnote 105 As far as the preceding centuries are concerned, the introduction of church records, in which christenings, marriages, and burials were registered, marks a rough turning point. With the help of those records’ data and complex mixed methods like the estimation of the ratio of births, the estimation of average household sizes and the estimation of total living residents, modern historical demography can determine the population of a certain place at a specific time.Footnote 106 Nonetheless, from medieval times until the sixteenth century only intermittent census-like data, which had been collected mainly for fiscal purposes, can be used to estimate population figures. These data usually contain only the number of houses or heads of households, making extrapolations of the total number of residents necessary. Probably the most prominent example is the English ‘Domesday Book’, an account of property taxes dating back to the end of the eleventh century. It enjoys the status of an early pioneer in this respect.
Population estimates for premodern Europe are necessarily always burdened with enormous uncertainty from the outset. This difficulty is, however, inversely proportional to its consideration by representatives of the decline thesis. Most meta-studies do not disclose in detail their sources of underlying population figures in homicide rates, though one may assume that these are the figures of the original case studies.Footnote 107 Matching criminal history data and population figures is an always tricky and often even inappropriate enterprise in a number of cases. Only in very few and exceptional instances are both pieces of data available for the same place and time. More often the population data stems from an earlier or later period than the criminal records. If researchers are able to trace back longer sequences of homicide counts, spanning decades or even centuries, for reasons of convenience they often believe population estimates to be constant over a longer period,Footnote 108 even though they represent only a certain point in time.Footnote 109 Alternatively, they claim a linear growth of population. Neither strategy reflects our knowledge of premodern demographics. Population developments could change drastically in a short period of time, for instance as a result of migration or crises such as war and epidemics.
Let us consider a few examples. Although James Given is clearly aware of it, the calculation of homicide rates in his classic study on thirteenth-century England is very problematic. First, he multiplied the number of heads of the households contained in the Domesday Book by 5 to account for family members and servants. In a second step, he multiplied the result again by a factor of 2.5 in deference to an assumed population growth within a period of 150 years.Footnote 110 Given then used the final result of this strikingly oversimplified procedure to calculate homicide rates.
In contrast, recent studies have revealed many difficulties in calculating medieval population figures. They have formed new and more modest estimates, which sometimes diverge markedly from earlier ones.Footnote 111 The need to revise population numbers is not limited to medieval times but affects the early modern period as well. The widely accepted figures from Wrigley's and Schofield's studies have not yet been entered into Eisner's database, but do reveal some surprises.Footnote 112 Eisner still uses the older estimates deriving from case studies. Striking examples are Hertfordshire and Sussex, both of which were examined by Cockburn in the late 1970s, who estimated a population of not more than 12,000 for Hertfordshire and approximately 28,000 for Sussex in the late sixteenth century. Wrigley's figures, in comparison, are much higher for this period: around 58,000 residents for Hertfordshire and 103,000 for Sussex.Footnote 113 Obviously, the consequences for the corresponding homicide rates are dramatic. Hertfordshire's rate sinks from 4.4 to 1.4 for the late sixteenth century and from 17.4 to 3.6 for the early seventeenth century; Sussex's from 10.1 to 3.0 and 8.9 to 2.4, respectively.Footnote 114
Another issue is the fact that population estimates are only available for certain points in time, whereas homicide rates refer to decades or even longer periods, and this mismatch has to be dealt with. The first option is to take the same population figure for longer periods of time. This is what Given did, when he used only one figure to calculate multiple homicide rates between 1202 and 1276.Footnote 115 Eisner takes population estimates from 1290 to calculate homicide rates for 1300–1348. Needless to say, within these six decades significant demographic changes could have happened, even though (and this is the main problem) there is no reliable data available.
Another, no less problematic option for scholars presenting homicide rates is to extrapolate population figures by themselves. These often simple calculations are almost always based on an assumption of linear growth. This is, for instance, evident in Cockburn's data sequence on Kent. For the period from 1571 to 1751, he consistently adds 2,600 people per year to the population. As a matter of fact, this also means that within nearly two centuries there must have been a continuous decrease in the overall population growth rate.Footnote 116 Such a result widely contradicts existing information on demographic development in premodern England.Footnote 117
Another stark example is the city of Rome. During the sixteenth century, as Peter Blastenbrei's research has revealed, this city has been ‘among the most violent places of all in the modern period’. Blastenbrei's argument is mainly based on ‘the barbers’ reports’. In sixteenth-century Rome all barber surgeons and other ‘professionals’ who were active in the healing arts were required to report suspicious injuries to the authorities.Footnote 118 Blastenbrei's overview of known killings in selected years confirms the general picture of a violent-prone age, but he himself expressly excludes the possibility of calculating any crime rates due to many uncertainties. Among these contingencies is the remarkable fact that Rome's demographic curve reached dramatic highs and lows over the course of just a few years. The population almost doubled between 1559 (45,000 to 50,000) and 1563 (80,000), while it almost tripled in the next decade from about 50,000 in 1570 to approximately 140,000 in 1575/80. Nonetheless, homicide rate calculations for Rome were attempted in Eisner's data set, where it fluctuates between one and four.
More examples exist for the German speaking regions. For a period of 60 years, from 1338 to 1399, the same population figure (17,000) was used to reconstruct Augsburg's homicide rates, despite the fact that the Black Death did not spare the city in 1349. A rough estimate supposes that around a third of the population fell victim to the plague. For 1396, a different study reckons there were only about 12,000 residents left.Footnote 119
These are just a few examples that could easily be extended. They all point to fundamental problems with historical demography when it comes to the history of violence. A reliable calculation of homicide rates in premodern times is, for this very reason, almost impossible. It is certainly less of an issue that estimating population figures for the pre-statistical era is very demanding and time-consuming. More problematic is the fact that complex calculations are based on various, sometimes apodictic assumptions when at the same time population numbers are available only for single years and not for broader time spans. Thus, calculating homicide rates on this basis is more or less educated guesswork. Given the fact that defenders of the decline thesis barely take any of the problems of historical demography into account, the confidence in the figures they use seems largely unwarranted.
3.5 Vague multiplied by ill equals …?
So far, we have seen that the figures for both violent deaths and population estimates for the premodern era are extremely inaccurate and uncertain. There is little doubt that homicide rates must be less reliable then, since they combine both figures. This impression is only strengthened when we scrutinise how historical statistics are related to their sources. Let us return to source issues once again.
The tacit premise for homicide rates calculations is to assume that ‘cases’ and population numbers each relate to a common spatial unit. Therefore, so the argument goes, by extrapolation of a number X of violent deaths related to a number Y of residents a homicide rate Z can be calculated. For modern administrative units this is supposed to be a usually unproblematic operation. In premodern times, however, spatial boundaries of jurisdictions were in no way consistently clear. Again, we encounter England's exceptionalism. Assize districts were mainly drawn according to the counties, so finding an analytical locus is no major obstacle. Some regions, however, also possessed a special jurisdiction which is often disregarded in studies on criminal history. Among these are the nine cities of the Cinque Ports in Kent and Sussex, which did not come under the jurisdiction of the assize judges until the seventeenth century.Footnote 120 In addition, perpetrators who fled over county borders represented a challenge to law enforcement. In particular, the counties bordering on the City of London tend to post heightened case figures.Footnote 121
In Sweden-Finland, the introduction of courts of appeal in the seventeenth century brought – at least theoretically – clear jurisdiction for capital punishment in every part of the country, even though the transgression of intrastate boundaries also posed problems.Footnote 122 As far as Germany is concerned, it is in no way certain that either killings or people involved in an alleged homicide could be traced to the same place where they had to stand trial. The jurisdiction for capital punishment in larger cities often extended far beyond their walls or district boundaries. It was also imposed on perpetrators who had carried out their crimes elsewhere. For example, in the fourteenth century Augsburg's city council punished any crimes that happened within a certain circuit that constituted a sphere of civic peace far beyond the city's settlements. In 1373, five men were banished for a killing in Zusmarshausen, a village 30 kilometres away. In Nuremberg in 1392, a ropemaker's servant was expelled for life from the city, because ‘he stabbed a woman who died in Bamberg’ (more than 50 kilometres away). Most likely, these types of cases were not always explicitly recorded and necessary critical considerations have not yet found their way into the databases.Footnote 123
Coroners’ reports and related sources that served to document the victims of violent crimes in a certain region are more reliable in regard to jurisdictions and case numbers. The cases recorded in Cologne's Totenbuch der Gewaltrichter (criminal court death registry) can be clearly localised within the city walls. The jurisdictions of the imperial city officeholders were intramural only. This does not mean, of course, that the victims (and/or perpetrators) came from the ranks of city dwellers. A large city like Cologne was a theatre of communication and interaction for countless guests and travellers, such as boatmen, carters, grocers and soldiers. Members of these trades appear disproportionately frequently in criminal court records. Whether they lived in Cologne or whether they were only passing through is often hard to ascertain. The records list many people in relation to their origin such as ‘a Dutchman’, ‘a Frenchman’ or ‘a man from the neighbouring city of Neuss’, whose names remained unknown to the authorities and who were most certainly not from Cologne. Neither was Frenchman Jan de Remichampani, who was attacked and killed by a group of fellow countrymen.Footnote 124 This bitter truth of Cologne's records also applies to the ‘body inspection records’ in Amsterdam, which is a paradigm for the decline thesis.Footnote 125 Primarily, these problems affected larger cities, which were in multiple ways attractive: not only residents, but also (and perhaps even in greater numbers) visitors and guests were drawn into lethal conflicts. Additionally, cases of murder were processed quite rigorously by a more advanced urban justice system, perhaps with a higher chance of conviction than in rural jurisdictions. Considering that cities are overrepresented in the databases, because they were more likely to produce sources that provide a basis for serial analysis, another bias becomes evident. This distortion of results is balanced out in modern, comprehensive criminal statistics. For the premodern era it paints, however, a false picture of reality by suggesting that cities, in particular, could represent a greater region.
Ignoring these issues and instead representing homicide rates graphically in a comparative perspective raises even more problems. As can be seen in Figures 2 and 3, conflating inconsistent data, such as for England and Germany/Switzerland, might produce interesting graphs. In the end, these are indeed little more than problematic depictions of dubious results.
Eisner, England: local estimates and national series.
Source: History of Homicide Database.Footnote 127
Eisner, Germany and Switzerland: local estimates and national series.
Source: History of Homicide Database.Footnote 129
Each data point in these charts represents a homicide rate whose validity sometimes differs drastically in terms of content, geography and time period compared to others in the same graph. The underlying data is a patchwork of homicide rates for small cities, wide-ranging areas and various times. The data for sparsely populated regions is not per se less valid, but those regions must at least be examined over longer periods to compensate for dramatic fluctuations caused by unique events.Footnote 126 This is not always done. Because of inherent issues of the source material, it is more often than not impossible to evaluate longer periods or even trace homicide rates in usual 10-years intervals. There is, however, no weighting of the data. The graphs depict all homicide rates as being equally reliable which in no way reflects the messy reality behind these figures.
Using these charts, proponents of the decline thesis count on the visual power of suggestion. Statistical operations, like standard deviation calculations, are not applied to test the validity and reliability of the findings. This is hardly surprising. Due to the source problems, that we have discussed above, such operations seem impossible for all data until far into the modern era. Some US researchers use the capture-recapture method in order to compensate for missing sources and gaps in records. By comparing evidence of homicides deriving from two different sources, such as court records and newspapers, an estimate of the total of reported homicides can be calculated.Footnote 128 Yet, this method is ineffective for the earlier periods where no suitable alternative sources parallel court records.
4. Beyond objections: perspectives for future research
Let us now briefly summarise our observations. First, the sources that provide information about homicides are extremely heterogenous and fragmented at least until the nineteenth century. If source material survived at all, then sometimes victims are counted and sometimes suspects. Due to varying judicial frameworks (criminal laws, definitions of offences), different kinds of violent deaths are represented. As a consequence, the sources are vague about whether even the object of interest, that is homicides, remains constant throughout the centuries.
Second, the sources are riddled with gaps. Robust serial data exist only for a few places or regions. Researchers are left to extrapolate of the available material in order to shed light on shorter periods or even only individual years. In addition, there are clear signs that high figures are readily accepted, while the absence of killings is attributed to missing source material rather than to entertain even the possibility that no homicides occurred during the time in question.
Third, such erroneous data, drawn from the source material, are then interrelated with extremely doubtful population figures. As we have shown, these figures are even more difficult to determine for the pre-statistical age than homicide counts. At best, well-founded estimates are available, but usually only for single years, which by no means ought to form the basis for calculations for longer periods. Yet, the opposite has been done regularly. Arbitrary extrapolations, such as the assumption of a constant linear growth between two time periods, also lack any methodological or empirical foundation.
Fourth, in an equally arbitrary manner and without any statistical weighting, these data are transformed into reference points, chronologically arranged and graphically presented. Thus, the construction of historical homicide rates represents a combination of two immensely dubious series of data. The resulting error is aggravated by the fact that any assumption that homicide numbers and population figures might refer to a common spatial unit is hardly convincing, because the boundaries of the respective jurisdictions were often uncertain. Taken together, the doubts outlined in this article, both methodological and source-related, lead to an unambiguous conclusion: proving a long-term decline of violence in Europe since the late medieval era is an ambitious yet so far unsuccessful project.
From the beginning, the decline thesis was closely associated with the theory of a civilizing process as described by sociologist Norbert Elias. Manuel Eisner's work is no exception in this regard, even though he carefully seeks to expand Elias’ theory with other concepts.Footnote 130 Yet again, it is Steven Pinker who vulgarises historical research arguing that ‘the tenfold-to-fiftyfold decline in [European] rates of homicide’ is a synonym for the ‘civilizing process’, since Norbert Elias ‘attributed this [sic] surprising decline to the consolidation of […] large kingdoms with centralised authority and an infrastructure of commerce.’Footnote 131 Taking a closer look at Elias' concept of a civilizing process, however, would reveal various problematic implications of empirical, methodological and theoretical nature. These issues, which are beyond the scope of this paper, are widely known and have been discussed repeatedly.Footnote 132 Nevertheless, Elias’ concept forms an important argument for claiming both, a process of modernization and a decline of violence. Sometimes the concept even enables researchers to integrate contradictory empirical findings into the overarching narrative. Within a process of civilization that encompasses the history of mankind, short periods of ‘de-civilization’ never contradict the big picture.Footnote 133 From such a perspective, although the premises are dubious, there is evidently no need to take any obstacles outlined in this paper seriously.
Therefore, a critical deconstruction of historical narratives, even cherished ones, is necessary. Such an approach may prima facie appear unattractive and even disenchanting because it seems to narrow the horizon of what we think we know. The purpose of this article was not to argue the case for a shrugging capitulation to the poverty of sources in earlier centuries. Instead, we insist on a necessary ‘spring-cleaning’ that paves the way for new questions and findings. Our concerns are intended to free researchers from the burden of an overly simple and one-dimensional interpretation that has blocked our view and impeded many interesting and promising lines of questioning. Some of these perspectives will be put forward in the concluding part of this article.
It is important to note that the empirical rejection of the decline thesis does not mean we cannot trust figures and quantitative data in general. They remain, in our view, central resources for historical research into violence. But reliability matters. Scattered and unreliable evidence, which characterises especially medieval sources, seems largely useless for interregional comparisons over time. More consistent, valid and reliable data, on the other hand, which become increasingly available for the early modern period, remain valuable and, within limits, meaningful. This is the case for data on the English county of Kent, on Amsterdam or on Stockholm, and, to a more modest degree, on the German imperial cities of Cologne and Frankfurt am Main. Given a thorough interpretation of the regional and local contexts of these data, they can also be used for comparative purposes, if only with caution.
A diachronic comparison, for one thing, actually does seem to reveal a decline of violence, which is, however, specifically restricted to the seventeenth and eighteenth centuries. This is indicated by research on England (Sussex, Cheshire and London) and by Joachim Eibach's study of Frankfurt am Main.Footnote 134 In Sweden-Finland, the city of Stockholm, in particular, seems to support this paradigmatic finding and the tendency is also perceptible in Amsterdam.Footnote 135
It seems promising to see this tendency toward a decline of violence within this shorter period first of all as an intermediate-term trend, which compels us to look for plausible explanations regarding the specific developments in urban centres. For instance, Joachim Eibach has outlined a process of a gradual containment of violence, that at first glance again borrows strongly from Elias’ ‘civilization of manners’. He also makes use of Jürgen Habermas’ narrative of the coffee house as the core of a bourgeois public sphere in the eighteenth century, which helps him to describe the cultural appropriation of bourgeois standards in the development of a new mainstream culture of town-dwellers including the increasing use of courts to resolve conflicts and uphold orderly forms of social interaction. In this story, the eighteenth-century urban middle-class effectively marginalised violence among their ranks. Blunt violence then remained an issue for the fringes of society, whereas new distinct practices of class-specific forms of violence, like duels, were on the rise.Footnote 136
While bearing in mind the critical approach taken in this paper, we might use Eibach's proposal to raise questions and develop new perspectives. We need to cautiously compare the value of any available source material for other cities and keep the obstacles described above in mind. This also means taking the sometimes contradictory developments in the history of violence seriously. We should then scrutinize the socio-spatial topography of violence and its transformation within different urban settings, but also within a comparative framework of rural and urban areas, more systematically. Moreover, following Eibach (and other researchers), we need to consider the differentiation of distinct violent practices as well as the question of what opportunities legal conflict resolution actually offered individual social groups in order to be relevant.
Either way, offering plausible theses for diachronic comparisons remains a challenge. The dramatic differences in synchronic comparison, however, have been unaccountably neglected thus far as a result of the dominant decline narrative. Remarkable differences are observable in both the level and rate of violence. Southern Europe – Italy and Spain, for example – seems to have comparatively high homicide rates well into the twentieth century. The quality of the source material is, however, even worse here than in the regions discussed above. Therefore, it seems risky to venture any assertions at all about developments in the level of violence in these countries.
This does not prevent defenders of the decline thesis from expanding their narrative into these regions as well. As ‘peripheral’ areas where processes like industrialisation started later, they seem to fit seamlessly into the overall picture, according to which modernisation inevitably leads to less violence. Both on a national and regional level, so the argument goes, there would be clear differences between rural regions, backwards in terms of modernisation and showing high homicide rates, and urban, industrial centres with lower levels of violence.Footnote 137 Nonetheless, the consistently observed inverse developments toward more violence in ‘advanced’ countries in the nineteenth century are at odds with this assessment. As Peter King has shown, homicide rates rose, sometimes dramatically, from the beginning to the middle of the nineteenth century precisely in industrial centres of northern England, Wales and Scotland. In these cases, rapid urban growth, induced by industrialisation and migration, led to a higher level of violence than in rural regions.Footnote 138
These findings illustrate (once again) the limited significance of comparative observations of long-term processes, in particular if larger regions are compared and significant differences within these regions are neglected. If comparisons are at all possible and make sense, then only among data sequences culled from smaller regional units. For such large-scale, interregional comparisons, the manifold factors that potentially influence the data must be carefully included and accounted for. Among these are social and economic changes as well as periods of political instability, social unrest, war and post-war scenarios as well as epidemic diseases.Footnote 139
The priority must be a comprehensive analysis of individual cases within their respective social contexts. The guiding principle, in terms of an appropriate historical method, is to understand the particularities of each region as extensively as possible instead of prematurely reducing it to its lowest common denominator, that is a supposed typical profile of violent behaviour in a specific era at a given time.Footnote 140 Enough interesting questions and problems are awaiting us. Why, for example, did violent crime throughout Finland apparently increase steadily from around 1750 onwards, while in Sweden (or in Frankfurt) a downward trend seems to occur at the same time?Footnote 141 And how exactly are we to understand the milieu and behaviour of those wild packs of youths who wreaked havoc in Southern Ostrobothnia (Etelä-Pohjanmaa, western Finland) for almost a century? These ‘knife fighters’ (Finnish: puukkojunkkarit) provoked an unprecedented wave of deadly attacks and battles starting in the late eighteenth century. Only after a hundred years did these violent outbursts finally come to a halt.Footnote 142
These few examples indicate the horizon of a broad variety of open questions. Instead of simply repeating a master narrative of a universal long-term decline in violence, future research could benefit from pursuing such questions. The arguments presented here can only be a starting point.
Acknowledgements
We are grateful to Matthias Bähr, Martin Dinges and Thomas Sokoll for their comments on earlier versions of this paper and to Manuel Eisner for his exemplary collegiality in granting us quick and easy access to his database. We also thank most sincerely Guido Mehlkop for sharing his perspective as an empirical social scientist with us and Andrew Wells for his patience and forbearance as he carefully proofread our text. Finally, we would like to thank the anonymous reviewers for their helpful comments.
Open access
