Abstract: Cannibalism in the animal kingdom is widespread and well characterized, whereas the occurrence of human cannibalism has been controversial. Evidence points to cannibalism in aboriginal societies, prehistory, and the closely related chimpanzees. We assembled a non-comprehensive list (121 offenders, ~631 victims) of cannibalistic homicides in modern societies (since 1900) through internet-searches, publications, and expert questioning. Cannibalistic homicides were exceedingly rare, and often sex-related. Cannibalistic offenders were mainly men and older than offenders of non-cannibalistic homicides, whereas victims were comparatively young. Cannibalistic offenders typically killed manually (stabbing, strangulating, and beating) rather than using a gun. Furthermore, they killed more strangers and fewer intimates than conventional offenders. Human cannibals, similar to cannibalism in other species, killed and ate conspecifics, occasionally vomited and only rarely (2.5% of victims) ate kin. Interestingly, cannibalistic offenders who killed their blood relatives had more severe mental problems than non-kin-cannibals. We conclude that cannibalistic homicides have a unique pattern of murder methods, offenders, and victims.
Discussion
Cannibalistic homicides were very rare, often violent, manual and sex-related crimes. Victims were younger and offenders were older than in conventional homicide offenses. Cannibalistic offenders only rarely consumed kin and most who did suffered from serious mental problems.
Our data set consists of 121 offenders with approximately 631 victims. This is a very large number of victims, but note that we are dealing with the entirety of easily accessible cannibalism cases in modern societies since 1900. The case numbers in the US and Germany might be of particular relevance, because, in these two countries, we made a special attempt for a complete coverage of cases. Unsurprisingly, cannibalistic homicide is an exceedingly rare crime, accounting for a minute fraction of homicides. For the US in the period 1960–2018, we estimated the fraction of cannibalistic homicides, being the number of cannibalistic homicides divided by the total homicides [Uniform Crime Reporting (UCR) Program Federal Bureau of Investigation, 2020], to be 0.01%.
The cannibalistic homicides described here had many hundreds of victims but included only one cannibalized neonate. The two other children 2 years or younger killed by cannibalistic offenders were not cannibalized. Since in non-cannibalistic homicides the killing of children 2 years or younger is quite common (accounting for ~2.6% of homicides as reported in FBI data), it appears that this victim population is largely missing in cannibalistic offenses, i.e., we would have expected >15 cases. From a biological perspective, the absence of cannibalistic neonaticide in humans is surprising. Human neonaticide shares many features of cannibalistic neonaticide in rodents and lagomorphs (Sawin et al., 1960; DeSantis and Schmaltz, 1984). In humans, neonaticide, as defined by the killing of a neonate on the day of its birth by his/her own mother, incidence varied from 0.07 to 8.5 per 100,000 births (Tanaka et al., 2017). This behavioral pattern is strongly promoted by maternal stress and intrusion by novel partners or detection of predators. Thus, even though neonaticide has been suggested to be a prototypical “biologically” predetermined behavior (Hausfater and Hrdy, 2017), it still shows differences between rodents and humans.
Our data suggest kin-avoidance in cannibalistic homicide. About 97.5% of the victims were non-kin, a statistical difference to occurrence of kin-homicides in conventional homicides (Figure 6). The consanguinity of victim-offender pairs was very low. This conclusion agrees with findings from two other large-scale studies on cannibalistic homicides (Rajs et al., 1998; Lester et al., 2015). A diverging result with a large fraction of kin-cannibalism was only observed in one study with only five cases (Raymond et al., 2019). We think these discrepancies reflect different sampling of cases, i.e., while our internet search sampled mostly cases by news coverage, Raymond et al. (2019) focused on psychiatric patients. Indeed, in our sample the cannibalistic offenders with the most obvious mental problems killed more kin (Figure 6D). This reduced kin avoidance in offenders with mental problems might not be limited to cannibalistic homicides. Offenders of conventional homicides in Scotland targeted three times more kin when they had a psychiatric diagnosis at the time of trial. The fraction of kin victims for offenders with a psychiatric diagnosis at the time of trial was 30.9%, compared to a kin victim rate of 11.3% for mentally healthy offenders (Gillies, 1976).
The fact that mental problems can reduce kin-selectivity in cannibalistic homicides does not argue against the existence of strong anti-kin-ingestion mechanisms, whether such mechanisms are a result of nature, nurture, or a combination of both. In rats, there are kin-responsive neurons in the lateral septum (Clemens et al., 2020), which might point to a possible biological basis of kin recognition at least in rodents. Humans use stable psychosocial cues to distinguish different types of kin from non-kin, and the cues used for kin detection depends on the specific dyad (Lieberman et al., 2007; Tal and Lieberman, 2008; Antfolk et al., 2018; Billingsley et al., 2018). One such cue is early co-residence between purported siblings as suggested by Westermarck (i.e., the hypothesis of Westermarck, 1922). It might be that this system of kin detection is affected in offenders with a psychiatric diagnosis and thereby prevents the offenders from recognizing and avoiding kin.
Spitting out conspecifics or parts of them is a characteristic behavior of cannibalistic animals and humans. We found five such throwing up /spitting out events, which may not sound like a lot in 631 victim cases, but our documentation is not comprehensive and it is important to keep in mind that many humans may eat a thousand meals a year without any one such throw up event. In sticklebacks, a strongly cannibalistic fish species, spitting out of newly hatched fish has been carefully documented and referred to as “testing” (Bruijn and Sevenster, 1982). Similarly, the cannibalistic spadefoot tadpoles have been seen to nip at and spit out conspecifics (Pfennig et al., 1993), a behavior referred to as “tasting”. Our observations on humans and a review of the animal literature suggest, however, that it is very unlikely that spitting out meat from cannibalistic events indeed reflects a sensory discriminatory behavior or gustatory “tasting”. The reasons, why we reject the tasting/testing hypothesis are the following: (i) Stickleback, spadefoot tadpoles, and humans have potent non-gustatory kin-discrimination mechanisms. In stickleback (Mehlis et al., 2008) and in salamander tadpoles (Pfennig et al., 1994), such mechanisms are olfactory in nature and in humans and apes (Parr and de Waal, 1999) such mechanisms are presumably visual; (ii) when tested intadpoles by nare occlusion, olfactory mechanisms were necessary for kin discrimination, whereas the remaining gustatory mechanism (after nare occlusion) were insufficient for kin-discrimination (Pfennig et al., 1994); (iii) human gustatory kin-discrimination appears highly implausible from our screening of cannibalism reports. On numerous occasions, human meat was sold (and probably eaten) by cannibals as ostrich, pork, horse, or tenderloin, mostly without any customer complaints reported. For instance, J.R.M. sold meat of his victims as special barbecue meat in a stand next to his trailer. He mixed the meat together with pork, which he claims tastes very similar to human meat, so that nobody could tell the difference (Serial killer who “cut up victims and sold them as BBQ” dies, 2017). If humans cannot discriminate human meat from beef, how could humans or animals taste kin? (iv) “Testing” behavior in stickleback is strongly dependent on the state of satiation (Bruijn and Sevenster, 1982); it is not obvious why a sensory discriminatory behavior should strongly depend on the state of satiation; and (v) in interviews with human cannibalistic offenders, they do not report that kin had a bad taste. We propose an alternative explanation for cannibalistic spitting-out: We suggest that this behavior is driven by internally generated repulsion and reflects a conflict between kin-protective, anti-cannibalistic drives, and predatory/consumptive systems. This explanation makes sense, because (i) we do not assume gustatory kin-discrimination, (ii) it is consistent with strong dependence on internal variables like hunger, and (iii) it fits with reports from cannibalistic offenders. In our explanation we explain a contradictory behavior (spitting out, a reversion of the decision to ingest), by an internal contradiction (preying vs. protection of potential kin).
Forensic awareness could be another explanation for the kin-avoidance observed in our study. Accordingly, cannibalistic offenders would avoid eating kin in order to escape prosecution, which is conceivably more likely for kin offenses. While we think forensic awareness is an important consideration, we do not think this hypothesis can fully explain our data. In particular, we do not see any evidence for a differential “forensic awareness” of cannibalistic and conventional offenders. Instead, many cannibalistic offenders enjoyed the celebrity emanating from their deeds; indeed, cannibal P.K. sends a letter to the press detailing the whereabouts of the grave of one of his victims, an action not speaking to forensic awareness. Also, other considerations do not align with the forensic awareness hypothesis. Offenders of cannibalistic crimes who showed forensic awareness were excluded from this study. For example, it is thought that the “confession” of cannibalism by E.K. is the result of forensic awareness, and he is, therefore, excluded from this study on the grounds of not enough evidence for cannibalism (Supplementary Data Sheet 3). However, we cannot exclude that other offenders still included in our study had a similar tactic unknown to us. It could be that offenders with forensic awareness will target strangers more often than family to dismiss any suspicions which might fall on them. In that case, one might argue that a reduced mental health also likely reduces the forensic awareness and that that could explain our finding that offenders with reduced mental health targeted kin more often. While we can not exclude this theory, we deem it unlikely because (i) this would then also be true for non-kin relatives such as intimate partners, and this is not the case (Figure 6E), and (ii) we excluded cases with suspicions of forensic awareness, and in first-hand reports by cannibals included in this study, they claim to really kill and eat for the purpose of pleasure, and (iii) cannibalistic offenders, who had enough forensic awareness to never be detected, are also not included in this study for the obvious reason that they are unknown. According to the theory above, uncaught cannibalistic offenders would target strangers more often than family, and those cannibals are missing from our study. Thus, if this is true, we are missing more strangers victims but we are not missing as many kin victims, in which case the actual effect size would be even larger than reported here.
War and hunger related cases of cannibalism appear to share features of cannibalistic homicide identified here. In war crimes, it is often the enemy which is cannibalized (Tanaka, 2018). In hunger-related cannibalism there are also indications of kin-avoidance. In the famous Uruguayan Air Force flight 571 incident, where starving victims of the plane crash fed on meat from deceased co-passengers to survive, it is said that at least some survivors made efforts to avoid eating kin (Arijón, 2008). Specifically, when one of them could only eat his kin for survival, he decided to cross the Andes to search for help instead (Parrado and Rause, 2013), which is the journey which eventually led to their rescue. Internally generated disgust is evident in such cases, i.e., the plane crash victims were appalled by eating human meat.
To achieve a correct interpretation of the results reported here, it is necessary to consider the limitations of our data set. Cannibalistic homicide is a secretive crime and our study is based on second hand and potentially distorted information. Thus, despite the best of our efforts, some errors are inevitable and we expect that our data set may contain mistakes about homicides, offenders, victims, and victim-offender relationships. The case of cannibalistic offender J.K. provides a warning. No less than five putatively innocent people were arrested for his crimes (three of whom committed suicide and a fourth was wrongly convicted of murder). Furthermore, some homicide offenders may give a calculated “confession” of cannibalism in order to use an insanity plea, as is suspected to be the case with E.K. (who for this reason has been excluded from this study, see Supplementary Data Sheet 3). In anthropology, the mere existence of human cannibalism in, for example, aboriginal cultures has been questioned on grounds of sensationalistic reporting in travelogs (Arens, 2004). Does this mean that one can dismiss our evidence about cannibalistic homicides, because these crimes seldom have eyewitnesses or video evidence and, without exception, incidents have been hyped by the press? We think the answer is a resounding no. The hundreds of cases documented in our report leave no room for doubt. We acknowledge, however, that our worldwide collection of cases is subject to a variety of sampling biases.
Limitations of Our Approach
The value of the data provided in our study might be limited by the following weaknesses:
1. The secretive nature of cannibalistic homicide. Cannibalistic homicide is nothing that can be openly performed in modern societies. The vast majority of cannibalistic homicides described here have been performed secretively and often there have been considerable efforts by offenders to destroy evidence. Hence, this paper largely describes deeds that nobody witnessed and cannibalism – the defining characteristic of what is talked about here – has rarely been directly established.
2. The need for estimates. Given the secretive nature of cannibalistic homicide, many details of the offenses described here can only be estimated. Many of these estimates are based on excellent evidence such as confessions, post-mortem examinations, or strongly suggestive circumstance (a child’s hand cooking in salted water on the stove), but nevertheless the evidence remains inferential.
3. Sampling biases. Perhaps the biggest problem of the data presented here is that our “news-based” search for cannibalistic incidents is subject to sampling biases. In particular, the data presented here will be biased towards particularly newsworthy cases with high victim numbers or gruesome case details.
4. Outright distortions. Sensationalism in the press might also lead to inflated presentation of evidence and distortion of the facts.
5. Reliance on second hand information. Our data set relies largely on web and newspaper reports and less so on official documents (original verdicts, interviews, letters and notes from the cannibal offender, autobiographies by offenders, and the like). Hence, verification of incident details is largely indirect.
6. Unintentional mistakes. Our data set is very large and unquestionably contains mistakes. Such mistakes would happen less in official documents double-checked and based on primary crime evidence but are unavoidable in our type of analysis.
7. Language barriers. We decided for a worldwide search of cannibalism cases, which allowed us amassing a large sample, but led us to encounter language problems. In several instances we used tools such as Google Translate to check local news sources. Such tools are powerful, but imperfect.
Strengths of Our Approach
1. Numbers. The biggest strength of the worldwide internet-based search for cannibalistic homicides is the sheer number of cases it returns.
2. Substantial coverage. The news coverage of cannibalistic cases is substantial. In a contemporary society, it is therefore simply unlikely that a cannibalistic homicide is not covered, unless (a specific detail of) a case is prohibited to be covered by the press.
3. Rich detail. Our data set contains very detailed information about cannibalistic homicides.
4. Good estimates. A lot of the evidence presented here is inferential, but many of the estimates presented here are good estimates. In modern societies an immense amount of effort is made to clear up homicides. Thus, a lot of the estimates about the deeds of serial murderers presented come from police investigators, who spent years of their life chasing the perpetrators; such detailed police work entitles to estimates about the deeds of offenders.
5. Documentation effort in internet-data sets. A lot of our searches and results rely on pages like murderpedia.org and lists of cannibalistic incidents. Many of these data sets do not comply with strict scientific standards of referencing; nevertheless, it would be a mistake to underestimate the effort and expertise that went into aggregating these data sets.
6. Access to rare cases. Cannibalistic homicides are rare and if one wants to investigate subclasses of such cases (i.e., very rare cases) the comprehensive search strategy is indispensable.
7. Informal cross validation. Whenever possible we informally cross-validated data sets against each other.
Our results indicate that cannibalistic homicide is a distinctive offense with a special pattern of murder methods, a strong relation to sexual acts, distinctive patterns of victims and offenders, and unique victim-offender relationships. There is a seemingly high amount of kin-avoidance in such crimes, in particular, if offenders do not suffer from serious mental health problems. We suggest that kin-avoidance and spitting out of conspecifics might be triggered by internally generated disgust against kin-ingestion.