Control Samples

Control samples were collected from studies of abuse, bullying, and video game violence exposure prevalence with male samples (acknowledging that there were a small number of females in the mass homicide database, they were eliminated from further analysis to avoid confounding). The control study related to childhood abuse exposure was González and colleagues (2015). This study focused on 2,928 men in the general public to determine if there were indirect or direct pathways to violence and health issues in adulthood related to childhood maltreatment. The sample consisted of men aged 21 to 34, 73.8% of whom were White, 11.1% of whom were Black and 15.2% of whom were Asian/South Asian. In this sample, 6% reported experience childhood physical abuse. For bullying, a sample by Wong and Schonlau (2013) was used. This study used data from the National Longitudinal Study of Youth, a database that began in 1996 employing youth aged 12 to 16, putting respondents in their mid-20s. Surveys for this study were taken 10 years later in the 7th wave in 2007. This sample examined the life outcomes of just under 9,000 individuals. Bullying data was available for 4,510 males. For this sample, childhood bullying prevalence among males was 22%. Lastly, regarding violent video game exposure, we used a sample provided by Olson and colleagues (2007). Data were gathered from 1,254 participants in the form of a self-reported survey. This study focused on the play patterns of boys and girls to determine a correlation between gender and playing violent video games (Olson et al., 2007). The average age of participants was 12.9 and 90% were White. This sample was younger than the others, though this seemed appropriate for assessing adolescent violent game use, and samples of prevalence among adult males are rare. Among the boys, prevalence of violent video game exposure (defined as having played an M-rated for “mature” game) was 67.9%. Descriptives of all samples are included as Table 1.

Procedures

The statistical analyses that were used to examine our hypotheses were three one sample t-tests, which allowed the mass homicide database with proportions of individuals experiencing the three predictor variables in the three control studies. For the mass homicide perpetrator database, each participant was coded as either having experienced or not having experienced each of the three predictor variables (1 or 0). This allowed comparison with a criterion value based on the proportion of individuals experiencing the predictor variable in the three control samples (e.g., .06 for physical abuse, .22 for bullying and .679 for violent game exposure).

Results

The first hypothesis was that mass homicide perpetrators have experienced more childhood trauma (or abuse) compared to individuals from the general populace. This hypothesis was analyzed by using a one sample t-test comparing the mass homicide perpetrators against the analyzed proportions of childhood trauma (or abuse) in the control sample. For this, the childhood trauma variable was used, and perpetrators were coded as having experienced abuse if abuse was reported by any family member. First, it is worth noting that data were available for only 64 of the perpetrators. Among them, experiencing of abuse was consistent (92.2% had experienced reported abuse). After examining the descriptive statistics, a one sample t-test was used to the hypothesis related to abuse: t(63) = 25.491, p < . 001. The effect size was Cohen’s d = 3.19. The results suggest that abuse was more likely to be reported for mass homicide perpetrators as compared to self-reported data from individuals in the general populace. This outcome supports this hypothesis. One possibility is that the missing data may cause an overrepresentation of abuse history among perpetrators. This was reexamined by making the assumption that every missing value was non-abuse. Rerunning the t-test still finds the association to be significant t(172) = 7.774, p < .001. Thus, even if every missing data point indicated an absence of abusiveness, our hypothesis would still be supported. Given that the data related to abuse was discrete, there was little clear basis from which to use imputation methods to address missing data.

The second hypothesis was that mass homicide perpetrators have experienced more childhood bullying compared to individuals from the general populace. This hypothesis was analyzed using a one sample t-test comparing the mass homicide perpetrators against the prevalence of childhood bullying in the control sample. First, the proportion of mass homicide perpetrators with data on bullying (n = 155) had a 19.4% prevalence rate for bullying. A one sample t-test can be used to compare the mass homicide perpetrators to the control sample: t(154) = -.831, p = .407. As such our hypothesis in this area was rejected. Our data suggest that reports of bullying experienced by mass homicide perpetrators does not differ from self-reported bullying by men in the general populace.

The third hypothesis was that mass homicide perpetrators have experienced playing more violent video games compared to individuals from the general populace. Cases of perpetrators who committed their crimes before a time likely to have the opportunity to play highly violent games (1992) were eliminated from analysis. This was because extreme violence in games was generally less common before this time, with a rise of fighting and shooter games beginning in roughly 1992. Specifically, the databased coded whether the perpetrators were reported either to have played violent video games or definitively had not played violent video games. However, in some cases, perpetrators were known to have played video games, but the content of the games was unknown. Individuals in this last category were excluded from analysis as violent game exposure could not be confirmed. This left 111 perpetrators, among whom 21.6% were reported to have played violent video games. A one sample t-test was then used: t(114) = -11.790, p  < .001. The effect size was Cohen’s d = -1.15. The results found that mass homicide perpetrators experienced playing fewer violent video games rather compared to individuals from the general populace. This result was opposite of the hypothesis.

Discussion

The issue of the etiological origins of mass homicides remains under considerable debate. We sought, by examining a preexisting database of mass homicide perpetrators, to examine this issue related to experiences with childhood abuse, bullying, and exposure to violent video games. Our results were mixed for evidence to support common beliefs linking these predictors to mass homicides.

The first hypothesis focused on whether experiencing of childhood abuse would be predictive of mass homicide perpetration. Our analysis supported a hypothesized correlation. The results clearly found that the mass homicide perpetrators as children had experienced trauma (or abuse) in higher proportions than a control sample of males. This finding is correlational, of course, but does suggest some degree of familial transmission of violence across generations, which may relate to mass homicide. By any metric of interpreting effect size (e.g., Cohen, 1992), this effect size is quite large indeed, frankly, quite unusually so for social science research. This is reflective of the near ubiquity of reported abuse among the sample of mass homicide perpetrators, whereas abuse is much less common in the general population. As such, this appears to be a particularly strong predictor variable.

The second hypothesis focused on whether experiencing childhood bullying would relate to mass homicide perpetration. However, we found no significant difference in proportion of mass homicide perpetrators who experienced bullying compared to a control sample of males. It may be that childhood bullying possibly had less of an impact on future violence than society has come to believe. It is possible that childhood bullying may play more of a role in youthful shooters, and the United States Secret Service and United States Department of Education (2002) analysis suggested that youthful perpetrators commonly perceived being bullied. It is also possible that these perpetrators may be more sensitive to bullying or interpret ambiguous slights as bullying. More data on this is clearly needed.

The third hypothesis focused on whether mass homicide perpetrators had experienced more exposure to violent video games than a control sample of males. In fact, our analysis suggested the opposite: that mass homicide perpetrators experienced fewer violent video games than control males. The effect size was moderate in size, not nearly as large as that for abuse, though still non-trivial. This appears to be consistent with data from the United States Secret Service and United States Department of Education (2002) report. The theory for why this hypothesis wasn’t supported was because violent video games do not influence a child/adult’s mind into becoming a violent or aggressive person. Increasingly, evidence suggests that violent games have little impact even on minor aggression. As such, links between games and mass homicides are largely mythical (Fox & DeLateur, 2014). As to why an inverse relationship may, in fact, exist, we don’t conclude violent games necessarily prevent mass homicides. There are several possible explanations. One possibility is that, indeed, violent games reduce stress or provide a release for youth in crisis. Another is simply that violent game playing is normative among young males, as evidence by the high proportion in the Olson et al. (2007) study. Thus, any deviation from normative behavior, however circumstantial, may be something of a warning sign. In this sense, if playing violent video games is “normal” for youth (Olson, 2010), then not engaging in the behavior might be more common among struggling individuals than engaging in the behavior is, even if the behavior may cross moral lines among older adults. As another interesting example, though historically adults have worried about teens having sex, more recently, some scholars have argued that teens delaying sex is a worrying sign of delayed adulthood due to technology use and other factors (South & Lei, 2021).

Broader Implications

For decades, policy makers and scholars have wondered over the potential causes of mass homicides and how they might be prevented. Any complex behavior such as mass homicide undoubtedly has multiple causal factors. Our current study is correlational; however, correlation is a necessary precursor for causality and, as such, can point policy makers toward phenomenon what may help us to understand what elements may or may not help us to prevent future mass homicides.

With this in mind, policies that target reducing violence in the home may be particularly fruitful. Naturally, most people who experience violence in the home don’t go on to become mass homicide perpetrators, nor even other violent criminals. However, family violence is one risk factor for violent crime in general and, as we see in the current study, mass homicide specifically. Trying to predict whom, based on their family history, would continue on to become a mass homicide perpetrator is unlikely to be fruitful, yet prevention and intervention strategies that target family violence broadly may help to reduce all forms of violent crime, including mass homicide.

By contrast, and somewhat a surprise to us, anti-bullying programs may be less helpful in this realm. The experience of bullying is fairly widespread, though it has been decreasing in recent years (Finkelhor, 2020). Thus, it’s predictive value may be limited. It may be more the case that mass homicide perpetrators are injustice collectors viewing themselves as victims of societal ill-treatment more than, in reality, experiencing it to any degree different from the average youth or adult. It is possible that, once again, prevention programs broadly applied may, nonetheless, be helpful, although we observe that historically the success of such programs has been modest (Ferguson et al., 2007).

What we can most definitively say is that pursuing policies to reduce exposure to violent video games, particularly as they conflict with the First Amendment in the US, or free speech values elsewhere, are unlikely to be of much help. Our data were correlational, so we don’t necessarily endorse the idea that violent games are cathartic in reducing the incidence of mass violence among those who play them although, broadly speaking, the release of highly popular violent games is generally associated with concomitant reductions in societal violence (Beerthuizen et al., 2017; Cunningham et al., 2016; Markey, Markey & French, 2015). Societal attention to the issue of video games is most likely to act as a distractor and, as such, pull the attention of policy makers and the public away from intervention and prevention efforts that actually may help.

Limitations

As with any study, ours has limitations. First, our data comprises the use of several secondary datasets. Although this approach can be useful, such datasets inevitably involved different methods and approaches in assessing their constructs. This can inevitably cause issues related to data management that can influence the reliability and validity of comparisons made across datasets. Each dataset also carries individual strengths and limitations, such as the use of self-report in the control samples as opposed to the use of archival psychological autopsy type data for the mass homicide database. Second, it is possible that definitions of terms may differ from the database of mass homicide perpetrators to the control samples, causing some unreliability in comparisons. Third, our study considered only a limited number of predictors. Naturally, other issues may also play an important role in the etiology of mass homicide. By contrast, our analyses benefit from being preregistered, reducing the potential for false positive results due to researcher expectancy effects. Case controlled examinations, such as this one, have their limitations and it might be possible for some to overestimate the relationship between child abuse and mass homicide (and, again, we caution that most individuals exposed to child abuse do not go on to commit serious crimes), but this approach can be an excellent complement to studies of the general populace among whom mass homicides are exceedingly rare.

Conclusions

The frequency of mass homicides appears unlikely to abate in the near future. Much of the discussion around such acts focuses on policy issues such as those related to gun control and mental health. We believe these are worthwhile. A fuller understanding of etiological factors involved in the developmental pathway toward mass homicide can also be worthwhile. Studies that are preregistered and use a standardized approach can be particularly valuable. We hope that our own analysis provides one small step in better understanding which variables do and do not correlate with mass homicide perpetration.