Abstract
The current study examines the offender, victim, and crime characteristics between solo perpetrators and team perpetrators of serial homicide. Cases on 1,137 solo perpetrators and 254 team perpetrators were collected from the Consolidated Serial Homicide Offender Database. Results showed team perpetrators were more likely to be older than those who committed serial homicides alone. Offenders who never confessed their crimes were less likely to participate in teams. In terms of victim and crime characteristics, team perpetrators were more likely to target employees or customers, have a lower victim count, and were more likely to exhibit sadistic behaviors compared to solo perpetrators. Theoretical and practical implications from this study are discussed.
Introduction
Although numerous studies on the characteristics of homicide offenders have been conducted, the topic of multiple perpetrators has received less attention. According to the U.S. Department of Justice’s (2011a) report, one in five homicides in 2008 involved multiple offenders, representing approximately 20% of all homicides. This was an increase from 11.3% in 1980. Younger offenders represented the largest age group involved in multiple-perpetrator homicides (37.5% of 14–17 years old), followed by young adults (27.5% of 18–24 years old), and adults (13.7% of above 25 years old). Serious crimes such as homicides and stranger homicides present various challenges to police investigators (Dauvergne & Li, 2006) which can be compounded by the involvement of multiple perpetrators.
The majority of early multiple-perpetrator studies focused on juvenile offenses in terms of co-offending behavior for non-violent crimes (e.g., Gold, 1970; Shaw & McKay, 1931). Wolfgang (1958) published one of the first multiple-perpetrator studies on serious and violent crimes which outlined the extensive descriptive information on the offender, victim, and crime characteristics. Since then, several research studies on homicide cases have suggested a delineation in characteristics between solo and multiple perpetrators (e.g., Cheatwood & Block, 1990; Clark, 1995; Gunn et al., 2014; Juodis et al., 2009; Park & Cho, 2019; Roscoe et al., 2012; Woster, 2020). In sexual homicide studies, results showed that while there were some shared characteristics between solo and multiple perpetrators, there were also apparent differences between the two types of offenders, such as motivation (Higgs et al., 2019), victim characteristics, modus operandi, and detection avoidance behaviors (Clarkson et al., 2020).
The consensus of past studies indicated that solo and multiple perpetrators of homicide do differ in characteristics. Although informative, these studies were limited in terms of small sample sizes ranging from 40 to 175 cases (e.g., Cheatwood, 1996; Clark, 1995; Juodis et al., 2009; Park & Cho, 2019). Further, these studies were mainly exploratory and were focused on single homicides, consequently neglecting serial homicide events. Research has shown that homicide offenders with two or more victims were different from offenders who only killed one victim in terms of victim-offender relationship, psychological disorder, and modus operandi (Fox & Levin, 1998; Harbort & Mokros, 2001; Sturup, 2018). In an area where the subjects of interest are rare (e.g., less than 1% of all homicides in the United States (US) were serial in 2010 (U.S. Department of Justice, 2011b), scholarly studies of serial homicide offenders have generally been idiographic or case study driven (Delisi & Scherer, 2006).
The low base rate of serial homicide events is further complicated by the variations in its definition, data collection methods, and data sources (Yaksic, 2015). Thus, our aim is to address the literature gap in both dimensions: (1) solo versus multiple perpetrators of homicide, and (2) serial homicide. The main focus of this study is to add to and expand on the current knowledge by empirically investigating the differences between solo and multiple perpetrators of serial homicide based on information collected from the Consolidated Serial Homicide Offender Database (CSHOD).
Background
Multiple-Perpetrator Offending
Crimes committed by more than one perpetrator, or co-offending, have received notable attention in criminological research, particularly within juvenile and property offenses (Tillyer & Tillyer, 2015). The majority of research has generally found that younger offenders were more likely to commit crimes in groups (Carrington, 2009; van Mastrigt & Farrington, 2009), although the age effect disappeared for violent crimes (Andresen & Felson, 2012). Further, based on an instrumental perspective, Weerman (2003) suggested that individuals chose to co-offend because it was easier, more profitable, and less risky compared to committing the crime alone.
Nonetheless, much research has shown that incompetence, betrayal, and risk of arrest were higher when individuals co-offend for certain types of crime (Bouchard & Nguyen, 2010; Lantz, 2020; Tillyer & Tillyer, 2015). Consistent with Erickson’s (1973) “group hazard hypothesis,” committing crimes in groups “increased the likelihood of official detection and reaction (e.g., apprehension, arrest, court appearances, and so on)” (p. 128). From the field of social psychology, studies which employed the concept of game theory (e.g., Prisoner’s Dilemma Game) has suggested that individuals were less likely to cooperate based on uncertainty and the fear of being exploited (e.g., Insko et al., 1990).
First conceptualized in the 1950’s by Albert W. Tucker (as cited in Sowden & Campbell, 1985) the Prisoner’s Dilemma Game has demonstrated that individuals often act strategically when making decisions that will benefit them the most (Flood, 1952; Luce & Raiffa, 1957). In the game, two players were asked to imagine they have been arrested for a crime committed together. In separate rooms, players were given the choice to implicate their partner in exchange for leniency. In experimental studies, results revealed that the majority of players (e.g., 75%) chose not to cooperate with their partners when playing the Prisoner’s Dilemma Game (e.g., Evans & Crumbaugh, 1966; Oskamp & Perlman, 1965).
The Prisoner’s Dilemma Game has been used and applied in social psychology and criminology research to explain co-offending behaviors. For example, a social experiment involving undergraduate students revealed that levels of distrust and cooperativeness were negatively correlated (Insko et al., 1990). That is, the more students distrust the other group, the less likely they were to cooperate in the simulated game. In contrast, a study of street youths in Vancouver and Toronto revealed that as the level of adversity increases (e.g., lack of food or shelter), youths were more likely to engage in collaborative behaviors for criminal corporation (McCarthy et al., 1998). This suggests that there is an interaction between distrust and perceived adversity to which influences self-preservation and individualistic behaviors (i.e., non-cooperation).
Despite considerable research on co-offending, much of the focus was on non-violent crimes, such as property offenses, while homicide events received less attention. Of the few studies that examined co-offending in homicide, the distinction between two groups of offenders were aptly labeled as solo and multiple perpetrators of homicide, instead of the term co-offending.
Solo Perpetrator Versus Multiple Perpetrators of Homicide
Comparative research on solo and multiple perpetrators of homicide has revealed several differences in characteristics between the two groups. In a study of homicide data in Baltimore, Cheatwood and Block (1990) compared the characteristics between juvenile and adult perpetrators of homicide. Although the focus of the study was not on multiple perpetrators, the authors found that homicides committed by more than one perpetrator were more likely to involve juveniles who had prior criminal history and were more likely to use guns to carry out their crimes (Cheatwood & Block, 1990).
One of the first studies to investigate the descriptions of homicides committed by solo perpetrators versus multiple perpetrators analyzed crime data collected from nine US cities (Clark, 1995). The victim, offender, and situational characteristics between the two types of perpetrators were compared and results showed that homicides committed by multiple perpetrators were more likely to involve victims who had a history with illicit drugs, strangers, those younger than the perpetrators, and were of the same sex as the perpetrators. Additionally, multiple perpetrators were more likely to be under 18 years old, have prior criminal history, and used different types of methods to inflict death. Homicides involving multiple perpetrators were also more likely to occur in places of employment and in tandem with felony circumstances (e.g., robbery) (Clark, 1995).
Efforts to replicate the study of solo versus multiple perpetrators of homicide outside the US have been conducted by several researchers. Juodis et al. (2009) examined official file information from two Canadian federal penitentiaries to identify the differences of offender, victim, and crime characteristics between solo and multiple perpetrators of homicides. In a sample of 125 incarcerated male offenders, results showed that offenders involved in multiple-perpetrator homicides were younger than those involved in solo-perpetrator homicides (26.85 years vs. 31.41 years). Further, multiple perpetrators of homicides were three times more likely to be characterized by an instrumental element (i.e., motivated by external goals) and had acquaintance victims, compared to solo-perpetrator events. In contrast, solo perpetrators were more likely to strangle their victims, targeted female victims who were current or intimate partners, and were three times more likely to engage in sadistic violence (Juodis et al., 2009).
Similar findings were found in a sample of convicted individuals in England and Wales. In addition to exploring the differences in demographic and criminological characteristics between solo and multiple perpetrators of homicide, Roscoe et al. (2012) included the offenders’ clinical aspects into their study. The authors found that offenders who participated in multiple-perpetrator homicides were more likely to be younger, unemployed, unmarried, from an ethnic minority, have previous violent history, and were more likely to receive a guilty verdict compared to solo perpetrators (Roscoe et al., 2012). In terms of the offenders’ clinical history, results showed that solo perpetrators were more likely to suffer from severe mental illness compared to offenders who were involved in multiple-perpetrator homicides.
In South Korea, Park and Cho (2019) found that solo perpetrators of homicide were more likely to target female victims and were more likely to be arrested at the scene within 24 hours compared to those who were involved in multiple-perpetrator homicides. Furthermore, results showed that multiple perpetrators were more likely to target stranger victims, prepared and brought a weapon to the crime scene, premeditated their crimes, used restraints on victims, and removed evidence from the crime scene (e.g., move victim’s body) when compared to solo perpetrators of homicide (Park & Cho, 2019). Overall, results from previous studies suggest that there are significant differences between solo and multiple perpetrators of homicide, leading to the proposition that separate theories are needed for each group of offenders (Cheatwood, 1996).
Non-Serial Versus Serial Homicide
While informative, current existing research on the two different types of offenders were examined for non-serial homicide events, leaving an information gap on serial homicides. Serial homicide has a broad-based impact on the community and society; however, little is known about these types of crimes (Rossmo, 2000). In addition to its low base rate (e.g., less than 1% of all homicides in 2010 [U.S. Department of Justice, 2011b]), inconsistencies in the definition of serial murder further complicated efforts to study this type of crime (Yaksic, 2015; Yaksic et al., 2019). Although scholars and practitioners have debated over the definition of serial murder and its inclusion criteria, the most widely used definition was proposed by the FBI where serial murder is defined as “the unlawful killing of two or more victims by the same offender(s), in separate events.” (U.S. Federal Bureau of Investigation, 2008, p. 9).
Comparative studies have shown differences in characteristics between non-serial and serial homicide events. Harbort and Mokros (2001) examined a sample of 61 individuals convicted of serial homicide in Germany to 750 non-serial offenders and found that serial murderers were more likely to have personality disorders, targeted strangers, had an instrumental and sexual motive, and were more likely to have cerebral anomalies when compared to non-serial murderers (Harbort & Mokros, 2001). In the US, Kraemer et al. (2004) reported that serial murderers were often sexually motivated, targeted strangers, premeditated their crime, and strangled their victims compared to solo murderers.
Similar results were found among serial murderers in Sweden; serial murderers were more likely to be diagnosed with personality disorders, targeted female strangers, planed their crime, strangled their victims, and were sexually motivated (Sturup, 2018). From a criminal career perspective, exploratory results have shown that serial murderers were more likely to have persistent and extensive criminal records, an indication of antisocial behavior (Campedelli & Yaksic, 2021; Delisi & Scherer, 2006). In effect, several typologies and classification systems have been developed to understand serial murders (e.g., Fox & Levin, 1998; Hickey, 2002; Holmes & De Burger, 1988; Ressler et al., 1988). Despite considerable research on non-serial versus serial homicide events, information pertaining to dynamics of co-offending (i.e., solo vs. multiple perpetrators) in serial homicide is lacking within the criminological field.
To our knowledge, only two empirical studies were conducted which examined the differences between solo and multiple perpetrators of serial homicide. Gunn et al. (2014) compared 383 solo and 104 multiple offenders of serial homicide and found that those who killed with a partner were less likely to be psychiatrically disturbed but were more likely to be sexually deviant and displayed sadistic behavior. Furthermore, females who killed with a partner were more likely to have a history of sexual abuse, had parents with psychiatric disorders, and were sexually deviant (Gunn et al., 2014). A comparison study between 3,806 solo and 1,059 team serial homicide offenders showed that team offenders tended to have shorter killing careers, higher victim counts, and non-intimate killing methods (e.g., shooting) compared to solo serial offenders (Woster, 2020). Motivation factors were also found to be distinct for each offender type. Solo offenders were more likely driven by enjoyment, anger, and multiple motives whereas team offenders were motivated by financial gain and organized crime (Woster, 2020).
Preliminary findings from these studies suggest that there are meaningful differences between solo and team serial offenders which warrants further investigation. However, the fact remains that this phenomenon is still greatly understudied, particularly for serial homicides, and previous studies’ results are in need of replication and validation.
Current Study
Despite considerable research on co-offending, violent crimes committed by multiple perpetrators remain understudied. Some scholars have found that early co-offending may lead to more serious and violent criminal careers (Conway & McCord, 2002) however, other scholars have found that co-offending rates were significantly higher for non-violent offenses such as property offenses (Andresen & Felson, 2012; Reiss & Farrington, 1991). The inconsistencies of these results led to several, albeit limited, studies that aimed to uncover the characteristics of co-offenders by examining violent types of crime, specifically focusing on the characteristics between solo and multiple perpetrators of homicide.
Nonetheless, previous studies that examined the differences between solo and multiple perpetrators of homicide were limited in several ways. First, the majority of those studies consist of a small sample size, ranging from 40 to 175 homicide cases committed by multiple perpetrators (e.g., Cheatwood, 1996; Clark, 1995; Juodis et al., 2009; Park & Cho, 2019; with the exception of Roscoe et al., 2012). Second, data used in those studies were constrained geographically to a single city or at the most, a country. Third, and most importantly, the lack of research on multiple perpetrators of homicide is further compounded by the intricacies and complexity of serial crimes. The majority of existing studies did not specifically examine serial homicide events.
Although Gunn et al. (2014) and Woster’s (2020) studies provided preliminary findings on the differences between solo and multiple perpetrators of serial homicide, an in-depth analysis to uncover meaningful differences between the two groups is warranted. We aim to fill this gap in the literature by examining a broader set of data which include 102 countries from around the world in combination with a comprehensive set of offender, victim, and crime variables. Furthermore, since the majority of previous studies compared differences between solo and multiple perpetrators of homicide on a bivariate level, we aim to expand our analysis further by examining the probability of the characteristics on a multivariate level. In order to narrow down the focus on the broader distinction between solo and multiple perpetrators of homicide, our study will compare the differences between solo and two perpetrators of serial homicide (which we will refer to as team perpetrators of serial homicide).
Specifically, the current study seeks to answer the following research questions: (1) what are the characteristics that differentiate between solo and team perpetrators of serial homicide, and (2) which offender, victim, and crime characteristics predict the likelihood of participating in a team compared to a solo serial homicide?
Method
Data and Sample
This study relied on data from the CSHOD, a product of merged records from independently built databases maintained and contributed by scholars who are members of the Atypical Homicide Research Group (AHRG). The CSHOD is a filterable database designed to allow users to curate the data to fit their given inclusion and exclusion criteria and allows users to select the serial murder definition that best suits their research study. While the AHRG does not endorse any one particular serial murder definition over another, the CSHOD was designed to be as comprehensive as possible and, to that end, was built using the definition put forth by the FBI where serial murder is defined as “the unlawful killing of two or more victims by the same offender(s), in separate events.” (U.S. Federal Bureau of Investigation, 2008, p. 9).
The CSHOD is comprised of records obtained through: (1) scholarly journal articles, (2) court and prison records, (3) textbooks, (4) news articles, (5) contributions from law enforcement, (6) Freedom of Information Act requests, (7) Internet searches, (8) personal files, (9) social media, and (10) true crime books. A database coordinator is responsible for overseeing the process of verifying information. The validity of the data is maintained by consistent cross-checks of old and new information against daily Internet searches for the terms “serial killer,” “serial homicide,” and “serial murder.” Any new additions to the database are securitized by at least two coders and measured against at least two independent sources. Data is only input into the database after approval by the coordinator. Those that have downloaded the database for use in their studies also submit edits to the coordinator when errors are encountered.
While the CSHOD contains data on all subsets of serial murderers who victimize acquaintances, family members, intimate partners, and strangers in various settings (e.g., college campuses, medical or nursing facilities, private residences, public spaces, retail establishments, street dwellings, and truck stops) and scenarios (e.g., arguments, domestic violence, gangland homicides, home invasions, homeless killings, professional contracts, prison inmates, random attacks, robberies, and willing or unwilling sexual encounters), serial murderers must be responsible for at least two homicides, committed in separate events, to be included in the CSHOD.
These events occur across at least two locations, either in a rapid, spree-like sequence of minutes or a longer time span of days, weeks, months, or years and for a variety of motives: (1) anger (e.g., anti-government beliefs, arguments, domestic disputes, grievances, racial hatred, revenge, vigilantism), (2) convenience (e.g., eliminate witnesses, undesired relatives), (3) criminal negligence (e.g., depraved heart, over prescription of drugs, reckless indifference), (4) enjoyment (e.g., control, pleasure, power, thrill), (5) financial (e.g., access to spousal resources, burglary, criminal enterprise, debt collection, insurance fraud, professional contract, robbery), (6) attention (e.g., fame, celebrity), (7) avoid arrest (e.g., escape apprehension), (8) torture (e.g., inflicting pain and suffering), (9) mental illness (e.g., commanded by outside force, paranoia, schizophrenia), (10) multiple, (11) random attacks, (12) rape, or (13) unknown. Users of the CSHOD may sort by and remove any of the subcategories within the 13 overarching motive categories that do not fit their study aims. Data from the CSHOD has been used in recent studies that reviewed serial homicide time intervals (Yaksic et al., 2021), spree killing (Yaksic, 2019), geographical profiling (Salafranca Barreda, 2021), and careers (Campedelli & Yaksic, 2021).
The present study only considered cases of serial homicide perpetrators with three or more victims and homicides that occurred in at least two separate events over a period of time greater than 14 days. Serial homicide perpetrators who were part of a large organization (i.e., gang, drug, or criminal enterprise) and offenders who were accused, suspected, self-proclaimed, ordered others to kill, or whose status as a serial murderer is in doubt were excluded. Offenders without a designated motive and records with a large degree of missing information were excluded. Serial homicide perpetrators cases that occurred before the year 1949 were also excluded. Serial murderers that killed two victims were excluded from the present analysis given the controversy surrounding the inclusion of these offenders (Yaksic, 2018). Events must occur over a period greater than 14 days to abide by the parameters set forth in Schlesinger et al. (2017) and Yaksic (2019) where it was determined that serial murderers that kill their victims within an interval of 1 to 14 days could be considered spree killers. To avoid such confusion, the present analysis was limited to including “true” serial murderers, or those that conformed most closely to the literature base.
The present analysis focused on dyadic teams given the inherent difficulty in capturing the nuances of group dynamics among this population of serial offenders when more than two perpetrators are involved (e.g., determining the level of participation in each homicide). Offenders must have had a designated motive given the subjectivity involved in assigning a motive to a series of crimes when one is not readily apparent from established sources. Serial murderers that killed for financial reasons were narrowed from a list of multiple sub-motives to only those that did so to acquire goods of value through either burglary or robbery to ensure that homicides were not taking place between groups who were ordered by others to eliminate certain targets. The analysis was limited to serial homicide series that occurred after 1949 given concerns with the veracity of data before that time, issues that arose due to substandard record keeping, a lack of definitional consensus, and varied classification schemes for serial murders.
Application of these study parameters brought the total sample of solo perpetrator of serial homicides from 4,073 to 1,137 offenders. The sample of two-member team perpetrators of serial homicide was derived by applying these same exclusion criteria. There were 126 individuals who killed as part of a team that consisted of more than two people (29 trios, 6 quartets, and 3 quintets), 46 individuals (23 pairs) whose final homicide occurred outside the study timeframe, and 26 individuals with various affiliations that could not be verified were excluded. The application of these study parameters brought the total sample of serial team perpetrator serial homicides from 573 to 254 individuals (127 pairs).
Measures
The choice of variables has been guided by previous empirical work focusing on the comparisons between solo versus multiple perpetrators of homicide (see Clark, 1995; Juodis et al., 2009; Park & Cho, 2019; Roscoe et al., 2012). Several measures were used in the present study across three predictor variable categories (i.e., offender, victim, and crime characteristics). The age range for offenders spanned 13 to 19 years for teens, 20 to 64 years for adults, and 64 to 100 years for seniors. The age range for victims spanned <1 year for infants, 1 to 12 years for children, 13 to 19 years for teens, 20 to 64 years for adults, and 64 to 100 years for seniors. Victims were classified based on their role in society and/or where they were killed. “Street people” referred to sex workers, the homeless, drug abusers. “Hitchhikers” referred to individuals who requested a ride in a vehicle with a person unknown to them. “Sexual encounters” referred to johns, lonely hearts, consensual partners, and meetings originating in bars or parties. Victims that were “patients” referred to hospital and or ward patients, those who lived in nursing homes, or received home care. “Employees/customers” referred to interactions that occurred at either the offender’s or victim’s occupation. Victims killed “in their own residences” were overtaken while in a private dwelling. Victims who were “targeted in the street” encountered the offender outdoors and were killed in the same location. Victims who were “involved in criminal activity” may have been drug dealers or customers, pedophiles, gang members, inmates, or informants but none functioned as part of a large criminal enterprise nor were they ordered to engage in criminal activity by others before or at the time of their death and each were attacked and killed during either a burglary or robbery.
There were eight primary motivations that spurred offenders forward in the present study and their homicides were committed for the following reasons: (1) financial (i.e., to acquire goods of value), (2) attention (i.e., to garner fame by way of the news or entertainment media or impress others with their prowess), (3) avoid arrest (i.e., to escape apprehension), (4) enjoyment (i.e., personal gratification in the form of pleasure, leisure, power, or control), (5) anger (i.e., to quell feelings of rage), (6) rape (i.e., to attain sexual dominance), (7) torture (i.e., to bring about pain and suffering), and (8) mental illness (i.e., a by-product of a disease or defect). Serial homicide perpetrators could have engaged in the series for more than one of these reasons.
Analytical Strategy
In this study we used a two-step analytical process. First, we conducted bivariate analysis (i.e., Chi2 analysis for binary variables and Mann-Whitney U test for continuous variables) to determine the differences between the dependent and independent variables. The goal was to determine which variables are the most significant predictors to use in multivariate analysis through a purposeful selection approach (see e.g., Hosmer et al., 2013; Tabachnick & Fidell, 2018). However, we acknowledge that the exclusion of non-significant variables at the bivariate level does not necessarily mean that those variables are not important at the multivariate level, and that the results could potentially be due to suppression effects. Second, we tested at the multivariate level only the significant differences (p < .05) identified with bivariate. Specifically, we computed four generalized estimating equations (GEEs). The GEE method allows for data that have correlated responses to be modeled as a generalized linear model (Liang & Zeger, 1986; Zeger & Liang, 1986). Correlated data are used in longitudinal studies, clustering (Johnston & Stokes, 1997) and analysis of serial crimes (Ensslen et al., 2018; Hewitt & Beauregard, 2014).
The GEE is particularly appropriate for the data used in this research. Since the unit of account is the offender, team serial murderers were associated with victims and crimes with the same characteristics. As a result, it cannot be assumed that independence exists between each observation, and the current models need to take this dependence into account. The Quasi Likelihood under Independence Model Criterion (QIC) was analyzed to determine the appropriate correlation structure. The QIC allows for the selection of covariates and working correlation structure simultaneously (Cui & Qian, 2007; Pan, 2001). Results showed that the Independent correlation structure had the lowest QIC value thus, our GEE models were analyzed using this method.
The first three models identified the most predictive variables for each of the three blocks of variables (i.e., offenders, victims, and crime characteristics). Then, a nested GEEs analysis was conducted using only the significant variables from all the previous models. This analysis represents the final and best model. Multicollinearity was checked for the variables included in the multivariate analyses and no variance inflation factors (VIFs) were above 3.36 and no tolerance levels were below 0.41.
Results
Bivariate Results
Table 1 shows the bivariate analysis results between three predictor variable categories (i.e., offender, victim, and crime characteristics) and participation in a team perpetrated serial homicide.
Bivariate Analysis of Offenders, Victims, and Crime Characteristics (N = 1,391).
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Four out of six offenders’ characteristics were significantly associated with team perpetrator participation: (1) offender being a male (χ2 = 19.11, p < .001), (2) age at first kill (Mann-Whitney U = 125,611, p < .001), (3) offender never confessed his crimes (χ2 = 4.24, p = .039), and (4) difference between the number of suspected and convicted murders (Mann-Whitney U = 121,027, p < .001).
Four victim roles were significantly associated with the type of serial perpetrator membership. Specifically, victims were street people (χ2 = 7.58, p = .006), victims were in patient wards (χ2 = 5.50, p = .019), victims were employees/customers (χ2 = 38.73, p < .001), and victims were involved in criminal activities (χ2 = 15.28, p < .001). Further, the type of victim and victims’ gender were also significantly associated with team perpetrator membership. Solo perpetrators targeted female victims exclusively (40.02%, χ2 = 28.84, p < .001). Children were more likely to be victims of solo perpetrators (20.32%, χ2 = 15.16, p < .001) and team perpetrators were more likely to have adult victims (96.06%, χ2 = 4.08, p < .001). Generally, team perpetrators were more likely to have multiple types of victims (19.29%) compared to solo perpetrators (11.61%, χ2 = 10.82, p < .001), but tended to have lower average number of victims (n = 4.83) compared to solo perpetrators (n = 6.53, Mann-Whitney U = 101,548, p < .001).
In terms of crime characteristics, five offenders’ primary motivation were significantly associated with the type of perpetrator membership. Team perpetrators were more likely to kill for financial reasons (53.54%, χ2 = 94.55, p < .001) and torture (20.87%, χ2 = 14.02, p < .001), whereas solo perpetrators were more likely to be motivated by enjoyment (54.94%, χ2 = 45.76, p < .001), anger (13.10%, χ2 = 5.32, p = .021), and rape (45.03%, χ2 = 21.52, p < .001). Strangulation (17.15%, χ2 = 13.67, p < .001) was more likely associated with solo perpetrators, whereas shooting was more likely the choice of killing method associated with team perpetrators (40.16%, χ2 = 53.83, p < .001). Finally, the average number of days throughout the killing series was significantly longer for solo perpetrators (n = 4,702) compared to team perpetrators (n = 1,045, Mann-Whitney U = 90,304, p < .001).
Sequential GEE Results
Table 2 shows the multivariate analyses through a sequential GEE resulting in four estimated models. Goodness-of-fit for the models was assessed by the Corrected Quasi-Likelihood under Independence Model Criterion (QICC), in which QICC adds a penalty to the quasi-likelihood based on the number of parameters within a model (Hardin & Hilbe, 2003; Pan, 2001). The lowest QICC value was associated with the model with significant variables from all three categories (QICC = 1,060.49), representing the Final Model. Through sequential modeling, significant offenders’ characteristics variables at the bivariate level were analyzed in Model 1. Results showed two variables negatively predicted participation in the serial team perpetrator group. Male offenders (b = −1.23, S.E. = 0.21, p < .001), and offenders who never confessed their crime (b = −0.36, S.E. = 0.15, p = .018) were less likely to participate in the team perpetrator group. Conversely, older offenders were more likely to commit their serial crimes in teams (b = 0.16, S.E. = 0.03, p < .001).
Sequential Generalized Estimating Equations of Factors Predicting Participation in a Serial Team Perpetrator Homicide (N = 1,391).
Quasi-likelihood under Independence Model Criterion.
Corrected Quasi-likelihood under Independence Model Criterion.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Model 2 presents the results of victims’ characteristics predictors on serial team participation. Three victims’ roles significantly predicted membership in a team perpetrator of serial homicide. Specifically, offenders were less likely to participate in the team perpetrator group when they targeted victims who were street people (b = −0.33, S.E. = 0.28, p = .039), and victims who were involved in criminal activities (b = −1.26, S.E. = 0.48, p = .009). On the other hand, victims who were employees or customers were more likely to predict the offenders’ likelihood of committing homicides in teams (b = 0.96, S.E. = 0.24, p < .001). Additionally, victims that were exclusively female predicted less likelihood that the offender carried out his crime in teams (b = −0.69, S.E. = 0.20, p < .001). Offenders were also less likely to participate in teams when the average number of victims increased (b = −0.05, S.E. = 0.29, p = .015).
Two motivation factors significantly predicted an increase in the likelihood of offenders’ participation in teams (Model 3). Offenders were more likely to operate in teams when the motivation of the crime was financial (b = 0.88, S.E. = 0.29, p = .002), and torture (b = 1.52, S.E. = 0.23, p < .001). Only one kill method was significant: shooting (b = 0.74, S.E. = 0.17, p < .001). Further, the longer the duration of the series, the less likely that the offender participated in teams (b = −0.01, S.E. = 0.00, p = .014).
In order to maintain model parsimony, only significant variables from Models 1, 2, and 3 were analyzed in our final GEE model, representing the Final Model. Overall, all but one predictor coefficients maintained their direction and did not significantly differ in terms of magnitude and significance. The Final Model showed that male offenders and those who never confessed their crimes were less likely to commit crimes in pairs (b = −1.69, S.E. = 0.30, p < .001; b = −0.35, S.E. = 0.17, p < .001). In addition, older offenders were more likely to participate in serial team perpetrator homicide (b = 0.11, S.E. = 0.03, p < .001).
Two victims’ roles remained significant predictors: (1) victims were all employees/customers (b = 1.51, S.E. = 0.47, p < .001) and (2) victims were all involved in criminal activities (b = −1.13, S.E. = 0.48, p = .013). However, victims who were street people failed to predict offenders’ participation in serial team perpetrator homicide once offenders’ characteristics and crime characteristics were included into the model. The average number of victims and exclusively female victims remained significant predictors of lower likelihood in serial team participation, although the latter has decreased in significance level (b = −0.53, S.E. = 0.20, p = .035).
All four crime characteristic variables remained significant predictors of serial team perpetrator homicide participation. Financial and torture motivated crimes predicted an increase in likelihood that offenders committed their crime in teams (b = 0.99, S.E. = 0.19, p < .001; b = 1.46, S.E. = 0.23, p < .001). Additionally, shooting as a kill method decreased in significance level but remained a significant predictor in the likelihood of participating in serial team perpetrator homicide (b = 0.42, S.E. = 0.20, p = .011). The average duration of days in the series remained a significant predictor of a decrease in the likelihood of offenders participating in serial team perpetrator homicide, however, its significance has decreased slightly. In other words, as the duration of the series increased, it was less likely that the offender committed his crimes in pairs (b = −0.01, S.E. = 0.00, p = .045). By analyzing multiple factors of the criminal event, we were able to better identify significant factors which predicted the offenders’ participation in serial team perpetrator homicide compared to serial solo perpetrator homicide.
Discussion
Despite the considerable negative impact that serial homicide has on society, these types of crimes remain poorly understood (Dauvergne & Li, 2006; Rossmo, 2000). Existing studies on multiple perpetrators of homicide are limited in terms of their sample size and generalizability and are geographically restricted (see e.g., Clark, 1995; Juodis et al., 2009; Park & Cho, 2019). The current study addressed these issues by empirically examining the characteristics between solo and team perpetrators of serial homicide using homicide data from over a hundred countries. Additionally, multivariate analyses using GEE was conducted to identify factors that predict the participation in multiple perpetrators homicide.
Since the majority of previous studies that examined the differences between solo and multiple perpetrators of homicide were conducted on a bivariate level, we will briefly discuss our bivariate results in comparison with previous results. Table 1 revealed several significant differences in characteristics between serial solo and team perpetrators of homicide. Consistent with existing literature, perpetrators who committed homicide in teams were more likely to target victims who were younger and were primarily motivated by instrumental goals (Cheatwood & Block, 1990; Clark, 1995; Juodis et al., 2009; Park & Cho, 2019; Roscoe et al., 2012). Some studies have found that team perpetrators often targeted strangers (Clark, 1995; Park & Cho, 2019) whereas other studies have found acquaintances to be victims of team perpetrators (Juodis et al., 2009). Our results are mixed in this aspect: both stranger victims (i.e., victims were all street people) and acquaintances (i.e., victims were all employees/customers or on a patient ward) are more likely to be associated with team perpetrators of serial homicide. Given the overall mixed results from the current study, further analysis is warranted. We addressed this by conducting multivariate analyses.
Who Makes the Team?
The most interesting finding shows that older offenders are more likely to participate in teams. This contradicts previous studies in which multiple perpetrators of homicide were primarily characterized by younger offenders (Cheatwood & Block, 1990; Clark, 1995; Juodis et al., 2009; Park & Cho, 2019; Roscoe et al., 2012). There are several possible explanations to this contradictory result. First, the majority of past studies included younger offenders (e.g., 8 years old) whereas our youngest offenders were 16 years old. The exclusion of younger offenders in our sample may simply be the difference in the definition of “younger offenders” between the current and past studies.
Second, the offenders’ age from all previous studies have been aggregated into two groups either based on the sample’s median age (Juodis et al., 2009; Roscoe et al., 2012), or a cut-off legal age of 18 years old delineating juveniles and adults (Cheatwood & Block, 1990; Clark, 1995). Andresen and Felson (2010) cautioned against using aggregated age groups as it has been shown that changes in co-offending rates were highly sensitive to sample age ranges. For instance, sensitivity analysis has shown that changes in co-offending rates were also found within-teen variances (i.e., age 11–19) and were further amplified by the inclusion of different types of crime (Andresen & Felson, 2012). In fact, when offenders’ age was disaggregated by single year of age and by crime type, results showed that the ratio between property crimes and violent crimes co-offending rates were higher for offenders under the age of 17. However, the age effects disappeared for offenders over 17 years old (Andresen & Felson, 2012).
In other words, age may be a predictive factor for the likelihood of participating in team perpetrator homicide up until 17 years old. But as offenders get older, the likelihood of participating in a team perpetrator homicide may be influenced by different circumstances. Indeed, the aging out of youth offenders who committed residential burglary has been shown to be quicker compared to youths who committed commercial burglary (Andresen & Felson, 2012), suggesting that more complicated crimes require older co-offenders. This is in line with the results found in our Final Model in which older offenders were more likely to participate in serial team perpetrator homicides since it required more skills and experience to carry out this type of crime.
Another significant finding concerns offenders who never confessed to their crimes as they were less likely to participate in a team. In other words, solo offenders were less likely to confess to their crime. While solo offenders will not have the option to implicate a partner, the fact remains that these offenders were less likely to confess their crimes compared to offenders who operated in teams. Drawing from the context of the Prisoner’s Dilemma Game, it is possible that team perpetrators were less likely to cooperate with their partner (i.e., confess and implicate partner) due to fear of uncertainty, being exploited, and high levels of distrust (Evans & Crumbaugh, 1966; Insko et al., 1990; Oskamp & Perlman, 1965). We hypothesize that team perpetrators were more likely to confess as a means to implicate their partner, usually as a plea deal to obtain lesser charges for themselves.
Victims and Crime Characteristics of Serial Team Perpetrators
There are several victim and crime characteristics variables which reveal interesting results: victims who were employees or customers were more likely to predict that the homicide was perpetrated in teams. Similar to Clark’s (1995) study on lone versus multiple homicide perpetrators, victims of multiple perpetrators often had a drug history. Although Clark (1995) did not identify victims’ roles in his study, it could be hypothesized that team perpetrators were drug sellers who more likely attacked their customers due to a drug deal gone wrong. This notion corresponds with the financial motivation in our study, which has been found to predict an increase in likelihood for participating in serial team perpetrator homicide. Several studies have shown that multiple perpetrators of homicide were mainly characterized by the presence of another felony crime and instrumental motivations (Clark, 1995; Juodis et al., 2009; Park & Cho, 2019).
Although seemingly counterintuitive at first, our results, which revealed higher number of victims predicting less likelihood of serial team perpetrator participation, can be explained by the group hazard hypothesis (Erickson, 1973). According to this hypothesis, crimes committed in groups are more likely to result in detection and reaction from official authorities since there are twice as many opportunities for offenders to leave behind evidence. Support has been found for the group hazard hypothesis in illicit drug crimes (Bouchard & Nguyen, 2010), robberies (Tillyer & Tillyer, 2015), and homicide (Lantz, 2020). This finding is further corroborated by our results where the average duration of the series for team perpetrators was shorter than for solo perpetrators. Team perpetrators are often faced with the risk that their partners will betray them or jeopardize the criminal act due to incompetency. Thus, it is likely that offenders who participate in serial team homicides are detected and apprehended earlier compared to offenders who act alone, leading to less opportunities for claiming more victims.
Further, previous non-serial homicide studies have suggested that lone perpetrators were more likely to exhibit sadistic violence and severe mental illness (Juodis et al., 2009; Roscoe et al., 2012). However, mixed results were found in serial homicide cases: multiple perpetrators were more likely to exhibit sadistic behaviors but less likely to have psychiatric disorders (Gunn et al., 2014). Similarly, our results showed torture was more likely associated with participation in serial team perpetrator homicide. Juodis et al.’s (2009) study of Canadian homicide offenders found that while solo perpetrators were more likely to engage in sadistic violence, multiple perpetrators who scored high on the Psychopathy Check List often inflicted pain and suffering on their victims in order to achieve some instrumental goal (e.g., money, drugs, sex, or revenge). This suggests that the majority of the offenders in our serial team perpetrators sample are likely to have more psychopathic characteristics in which they inflict pain and suffering on victims primarily for instrumental purposes.
Limitations
The current study is the first to examine the differences in characteristics of serial homicide committed by solo versus team perpetrators in a global sample. Nonetheless, there are several limitations to this study. First, the current dataset lacked complete information on the relationship between partners. This information might have allowed for a deeper examination of the dynamics between the offenders, what brought them together to kill, and what factors influenced their decision-making process. Given that team perpetrators were more likely to confess and implicate their partners, future research should investigate the strength of those pair bonds and what relationships (e.g., familial, intimate, acquaintance, friend) are more likely to result in retribution through confessions. Similarly, the lack of differentiation in the data between perpetrators who killed in teams and three or more participants further complicates the understanding of group dynamics, in which the exact level of each serial homicide perpetrator’s participation in these series is not well documented and has therefore been understudied. Future research should collect this data to study the delineation between each serial homicide perpetrator’s contribution to the series and whether that outcome matched their initial desire or plan.
Second, our sample only included convicted perpetrators. The exclusion of suspected or undetected perpetrators confines our results to information on known perpetrators only and cannot be generalized to individuals who are still at large. Further, the classification of “solo” perpetrators in our sample may have committed the crime in a team, but the partner successfully eluded police.
Third, it was not possible to describe the psychological characteristics between the serial solo and team perpetrators, given that we did not have enough information pertaining to the mental state of the offenders. Psychopathic traits have been found to be related to solo perpetrators of sexual homicide, and among multiple perpetrators with instrumental motivations (Juodis et al., 2009). Our results suggest that the act of torture which was associated with participation in serial team perpetrator homicide may be an indication of higher psychopathic characteristics among the offenders. Future studies should examine the psychological characteristics by including the offenders’ psychopathy scores or mental health history in order to uncover the motivations behind violent behaviors such as torture.
Fourth, although our sample is comprised of serial homicide cases from 102 countries, it was not possible to control for cultural and societal differences across countries. There is a growing body of research on serial murderers from countries outside the US suggesting that the behaviors of serial murderers from those countries are similar to offenders from the US (Deepak & Ramdoss, 2021; Harbort & Mokros, 2001; Morton et al., 2010; Salfati et al., 2015; Sturup, 2018). Serial homicide may be a universal occurrence, but future research would benefit from adopting a more extensive international perspective to discern any potential cultural and societal differences.
Fifth, legal proceedings may vary among each country. The conviction of serial homicide perpetrators is dependent on the criminal justice system of that particular country. This includes the skills and experience of the police, the prosecution and defense team, and even the opinion of the judge. Circumstantial factors surrounding the homicide case may also influence whether the offender is convicted or released (e.g., fail to maintain chain of evidence for conviction). Thus, the current study is exploratory rather than hypothesis driven. Future studies should not only attempt to replicate the current findings using different datasets to further the development and understanding of serial homicide perpetrators but also control for the variability in policing strategies across nations.
Implications and Conclusion
On a theoretical level, results from this study contributes empirically to the gap in knowledge in both multiple perpetrators offending and serial homicide fields of research. Our analysis extends beyond descriptive and exploratory methods used by previous studies and results indicate that it is possible to use the characteristics of the offender, victim, and crime to predict membership in a team perpetrated serial homicide. Furthermore, the use of an expansive and representative sample of serial homicide perpetrators in this study allows for greater generalizability capabilities, especially in comparison to previous research that was restricted in sample size and geographical locations.
On a practical level, the ability to differentiate between a homicide committed by a solo serial homicide perpetrator and one carried out by a pair of perpetrators would be of great benefit to police as a crucial step in an investigation. For example, older offenders are more likely to participate in a serial team homicide since it requires more experience and knowledge to carry out a complicated crime, especially when it is financially motivated. When police are able to identify a financial motivation behind the homicide, coupled with victims who are employees/customers, then investigators will be able to narrow down the suspects list to serial homicide team perpetrators.
In terms of interviewing suspects, serial team perpetrators are more likely to confess or implicate their partner. This allows police investigators to interrogate suspects separately, possibly offering reduced sanctions if one of them confesses and implicates their partner. However, this strategy should only serve as an investigative tool and caution should be exercised to avoid false confessions and incriminations (Leo & Drizin, 2010).
The near ubiquity of the internet and its social media channels may hasten the formation of serial team perpetrator homicide as like-minded offenders find it easier than ever to commiserate with someone else with their world view. Until teams no longer provide serial homicide perpetrators with a dual sense of power, feelings of diminished responsibility, heightened levels of commitment, the appeal of interdependency, trust and emotional connectedness (Gurian, 2013), the allure of joining a deadly dyad may outweigh the risks inherent to such pairings and lead to further loss of life in the future.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
