Abstract
This study investigates the effect of being an outlaw biker on criminal involvement in Denmark. Using a unique dataset, 297 outlaw bikers are matched on various background characteristics with 181,931 control individuals and effects are estimated in difference-in-difference regressions. This approach reduces the risk of selection bias and helps isolate the effect of affiliation on criminal involvement. The results suggest that affiliation with an outlaw motorcycle club may increase involvement in overall crime, specifically property crime, drug crime, and weapons crime. Results regarding violent crimes are inconclusive. It is concluded that an outlaw biker affiliation may increase criminal involvement.
Introduction
Outlaw motorcycle clubs are defined as clubs not registered with a nationwide motorcycle association, that is, the American Motorcycle Association (Dulaney, 2005; Wolf, 2008: 4). This includes clubs such as the Hells Angels MC (HAMC) and Bandidos MC (BMC). These clubs opened chapters in Denmark in 1980 and 1993, respectively. Although both clubs are part of the Danish nationwide motorcycle association Biker Foundation Denmark, they are still considered outlaw motorcycle clubs (OMCs). This is because Biker Foundation Denmark is not a member of the Federation of European Motorcyclists’ Associations – the European equivalent of the American Motorcyclist Association. In the current study, the term ‘outlaw biker’ (OB) is used to refer to members of both HAMC and BMC, and their affiliates and support groups.
OMCs have been proclaimed a new and growing type of organized crime, which adds further meaning to the adjective ‘outlaw’ (Barker, 2004, 2007; Barker and Human, 2009; Grascia, 2004; McDermott, 2006; Quinn and Koch, 2003). Furthermore, the curbing of their crime has become a European police priority (Europol, 2011). Although outlaw clubs are found throughout the globe, no previous study has been able to determine the causal relationship between OMC affiliation and criminal involvement. The causal relationship between gang membership and criminal behaviour, however, has been studied extensively. These studies tend to indicate that, although gang members come from a select group with above-average criminal tendencies, becoming a gang member further increases the likelihood of committing more serious and/or more frequent criminal acts (for a review see Krohn and Thornberry, 2008; for more recent studies see Bendixen et al., 2006; Haviland et al., 2007; Melde and Esbensen, 2011). The current study focuses on whether OMC affiliation facilitates criminal involvement.
The study is based on data concerning 753 male OMC members and their affiliates. These data were gathered by police units between July 2001 and June 2009 and are combined with background information from the national statistical archive, Statistics Denmark.
Among the 753 OBs, 297 are matched individually and exactly with non-OB controls on various background characteristics, including crime patterns prior to OB affiliation. 1 Of the 456 unmatched OBs, 76 are unmatched for reasons that may limit the conclusions of this study. Reasons for the inability to match the remaining 380 OBs, however, pose no selection problems.
The effects of OB affiliation on general criminal involvement and on involvement in specific types of crime are estimated using difference-in-difference (DiD) regressions. DiD regression is designed to reveal the independent effects of variables of interest while holding other variables constant and accounting for time-stable sources of selection bias.
The situation in Denmark is especially interesting from a European standpoint because of the unparalleled presence of OMCs in Northern Europe (Von Lampe, 2008: 13). The current study may therefore offer an example of what a strong OMC presence implies for national crime levels and control policy.
Theoretical perspectives
Social relationships matter in the commission of crime. Not only do peers in adolescence (Reiss and Farrington, 1991; Van Mastrigt and Farrington, 2009) and marital relationships (Sampson et al., 2006) matter, but even relationships to parole officers seem to matter (Andersen and Wildeman, 2014). Gang relationships also matter in a causal sense. Here I describe why links with a formal group such as an OMC might matter theoretically.
Gang researchers consider OBs to be distinct from members of youth and street gangs (Klein, 1995: 22). Nonetheless, gang studies have a great deal to offer OB research, especially on the relationship between group membership and criminal involvement. Three hypotheses on the relationship between gang membership and criminal involvement have been tested (Melde and Esbensen, 2011: 515–16; Thornberry et al., 1993: 57–9; Thornberry et al., 2003: 97–100).
The selection hypothesis is a ‘kind of person’ explanation, which argues that gang members commit crime because gangs consist of certain individuals who already have strong and relatively stable criminal tendencies prior to gang joining. The selection hypothesis predicts the same level of crime committed before, during and after membership in a gang.
The facilitation hypothesis is a ‘kind of group’ explanation, which argues that gang membership increases the individual tendency to commit crime. The facilitation hypothesis predicts the same level of criminal involvement both before and after joining a gang but an elevated level during active gang membership.
The enhancement hypothesis is a combination of the selection and facilitation hypotheses.
The three hypotheses do not suggest in the detail the causal mechanisms at work. In the following, more elaborated theories are categorized and described under each of the three general hypotheses.
Few theories hypothesize pure selection effects. Although Travis Hirschi concludes that delinquent boys are more likely to join gangs, he acknowledges that this may subject them to group processes resulting in increased criminal involvement (Hirschi, 1969: 159 and 161). These processes are, however, not an explicit part of his social control theory.
Some theories accommodate the facilitation effect while not addressing the selection effect. Edwin Sutherland postulates that the drives, rationalizations, attitudes and techniques of committing crime are learned in a process of communication within intimate personal groups (Sutherland et al., 1992: 89). Edwin Lemert concentrates on how subgroups differ from the established norms and on the interplay between conventional society’s reactions to normative transgressions and subgroups’ reactions to those reactions – including potential increases in criminal behaviour (Lemert, 1951). Albert Cohen states that some types of behaviour are highly regarded within delinquent subcultures precisely because they are considered so disreputable by the larger society thus making these behaviours more likely while in the subculture (Cohen, 1955: 68). According to Ragnar Hauge, youths not closely integrated in a gang, but wanting to be so, are at particular risk of exaggerated notions concerning preconditions for gang membership. This includes exaggerated notions concerning the gang’s overall level of criminal involvement, which they seek to match in an effort to become gang members (Hauge, 2001: 87). James Short and Fred Strodtbeck describe how status-threatened gang leaders may respond with gang violence in order to improve their standing (Short and Strodtbeck, 1965).
Other theories accommodate both selection and facilitation effects, thus supporting the enhancement hypothesis. According to Lewis Yablonsky (1962: 154) and Nathan L. Gerrard (1964: 369), gang membership neither creates sociopathic personality nor a hatred that finds its expression in violent acts, but rather ‘intensifies and consolidates it, and fosters its untrammelled expression’ (Gerrard, 1964: 369). Michael Gottfredson and Travis Hirschi argue that low self-control explains both criminal tendencies and self-selection into gangs (1990). The idea is that individuals with low self-control ‘who have difficulty making and keeping friends tend to end up in the company of one another’ (Gottfredson and Hirschi, 1990: 158). These individuals are additionally drawn towards gangs because they ‘use groups to facilitate acts that would be too difficult or dangerous to do alone (such as robbery)’ (Gottfredson and Hirschi, 1990: 158–9). Ronald Akers acknowledges both a selection and a facilitation effect, stating: ‘A peer “socialization” process and a peer “selection” process in deviant behavior are not mutually exclusive, but are simply the social learning process operating at different times’ (Akers, 2009: 56). In a similar vein, Robert Sampson and John Laub conclude that, although individual characteristics are important in predicting behavioural stability, experiences in adolescence and adulthood (presumably including gang membership) can independently amplify or suppress involvement in crime (Sampson and Laub, 2005: 16).
The enhancement hypothesis corresponds well with findings from previous gang studies (for a review, see Krohn and Thornberry, 2008). Taken together, recent studies continue to support the enhancement hypothesis (Bendixen et al., 2006; DeLisi et al., 2009; Haviland et al., 2007; Melde and Esbensen, 2011).
In the current study, only the facilitation hypothesis is tested explicitly. Selection cannot be addressed explicitly because the matching procedure is designed to remove group differences between OBs and non-OB controls. However, the exceptionally high level of criminal involvement among future OBs prior to OMC affiliation implicitly confirms the selection hypothesis.
The facilitation hypothesis would predict a reduction in criminal involvement upon cessation of affiliation. This, however, cannot be tested because dates of disassociation are unavailable.
Previous studies
The subculture of OMCs, including their criminal pursuits, has been described on numerous occasions (Bay, 1998; Dulaney, 2005; Quinn and Forsyth, 2009; Thompson, 1991; Wolf, 2008). Danner and Silverman (1986) offer an overview of classic qualitative studies of OMCs, whereas Quinn (2001) offers an overview of more recent literature. OMC subculture is generally described as not being a radical departure from street corner subculture (Wolf, 2008: 57). OMCs place a high value on motorcycling (and mechanical skills), on fellow OBs in general and especially on the bikers within one’s own club (Bay, 1998; Quinn, 2001: 384–5; Wolf, 2008: 105). These points are combined with a strong emphasis on traditional masculinity, including being honourable, tough, strong (though not necessarily in physical terms), independent and non-submissive (Wolf, 2008: 132). The subculture is reported to be male dominated (Wolf, 2008: 162). Women definitely appear in the subculture, although they seem to do so in submissive ways (Wolf, 2008: 131–62). In general women are considered a threat to a club. They are able to weaken a club member’s ties to his club because of challenging emotional ties and because OB clubs offer an exclusive masculine identity (Wolf, 2008: 134). Given this normative mindset, OB clubs tend to segregate themselves socially, but cannot do so economically (Wolf, 2008: 258). The OB lifestyle is simply too expensive. Given these norms, some degree of deviance is unsurprising.
Few studies have explicitly examined the relationship between OBs and crime in a manner that allows general casual inferences. Some studies solely rely on criminal events reported in the news media (for example, Barker, 2007; Barker and Human, 2009). Two previous studies utilized longitudinal observations, but do not compare their OB subjects to control groups (Alain, 1993; Tremblay et al., 1989). Another study compares incarcerated OBs to a control group, but lacks longitudinal observations (Danner and Silverman, 1986). A more recent study is based on communication networks linking HAMC members in Quebec and their affiliates (Morselli, 2009). Yet this study does not cover subjects during the period prior to their involvement in outlaw subculture. Seen as a whole, the aforementioned studies indicate a clear and statistically significant association between OMC affiliation and heightened criminal involvement, but are unable to explicate the causal relationship between the two. Proper causal inference requires longitudinal observations on the OBs before and after OMC affiliation and a comparable control group.
Data
Data for the current study come from the Danish National Police and Statistics Denmark. In June 2009 the National Police provided a list of 1146 individuals suspected of involvement in organized crime. Of these, 753 were OBs and 393 were street gang members. The data are from the police’s intelligence database (PID) where police record individuals related to OB and gang crime. Data for each individual include name, national identification number, club affiliation and PID date (that is, the date an individual is first recorded in the PID). The information is typically gathered by regular police patrol units and police squads specialized in OBs and street gangs, who report it to the National Intelligence Center, where each piece of information is evaluated in terms of investigative value by authorized personnel and then recorded in the PID depending on that evaluation.
Among the 753 OBs, 562 individuals have precise PID dates ranging from July 2001 to June 2009. The remaining 191 have PID dates prior to July 2001, though the exact PID dates were misplaced when police replaced an earlier intelligence programme. The Danish PID data are unique in that they include information on the date at which an individual is first recorded as being affiliated with a given group.
The National Police consider affiliation with an outlaw motorcycle club to be sufficient reason for recording a particular individual in the PID (National Police, 2009: 17–18; Klement et al., 2011: 6). This is based on the police’s assumption that affiliation with an OMC implies serious involvement in violent and/or drug-related crime. Determining whether suspected street gang members should be PID-recorded is somewhat more complicated. First, simply identifying street gang members is more difficult because they generally do not display associated symbols as openly as OBs, at least in Denmark. OBs are often required to do so at mandatory runs and/or parties, which the police sometimes have under surveillance. Furthermore, street gang members have to be suspected of serious violent and/or drug-related crime in order to be PID-recorded. OBs are therefore more likely than gang members to make it into the PID. Registration in the PID presumably increases police surveillance, the consequences of which are taken into account and discussed below.
This study rests upon the assumption that the PID dates are trustworthy estimates of the onset of OB affiliation. If affiliation actually occurred substantially earlier than recorded in the PID, then assumptions about the causal mechanisms at play would be incorrect. However, crime among OBs is a top priority in the Danish Police. In the 2013 annual report of the National Police, crime among gang members and OBs was rated with 20 percent priority (National Police, 2014: 12). No other area has a higher priority, though crime in disadvantaged neighbourhoods and crime targeting ordinary citizens have received the same level of priority. Furthermore, the police keep street gang members and OBs under tight surveillance, as reflected in more or less sensational descriptions (Brügger and Thomassen, 2009; Facius et al., 2005; Fischer et al., 2010; Frich, 2011) and as reported in the news media – which occasionally cover the registration and investigation of participants at OB parties (Damløv, 2014; Local Eyes, 2012, 2013).
Formal tests validating the PID dates does not seem possible. However, a recent US study using multiple measures found that official data on gang homicides can be sufficiently valid for research, especially in cities with specialized police gang squads (Decker and Pyrooz, 2010). A similar study in Denmark has so far not been carried out. Yet, given the presence of specialized police squads in Denmark, the high police priority placed on OB crime and the resulting tight surveillance, the PID dates are considered trustworthy estimates.
The national identification numbers in the PID make it possible to link additional individual-level data from the national statistical archive, Statistics Denmark (described in Lyngstad and Skardhamar, 2011). Statistics Denmark was able to provide extensive information on each individual in depersonalized, anonymous format including date of birth (and death, where applicable), gender, country of origin, and information on criminal history, sanctions, imprisonment, education, employment, and immigration and emigration for individuals residing in Denmark during the period 1980–2010. Statistics Denmark makes these data available under strictly controlled conditions. One of these conditions is that only aggregated results can be published.
Table 1 contains information on the affiliations of OBs and their support groups, and on inclusion and exclusion from this study. Table 2 provides reasons for OB and support group exclusion.
Included and excluded OBs, percentages (n = 753).
Grounds for OB subject exclusion, percentages (n = 456).
HAMC has two kinds of support group. One consists of motorcycle clubs and the other consists of AK81. 2 BMC support groups are grouped in one category.
An individual is categorized as an OB if he is affiliated with an outlaw motorcycle club or an organization that supports such clubs – even if he does not ride a motorcycle. This is sometimes the case for ‘hang-arounds’ in outlaw clubs and for members of support groups. The argument for their inclusion is that affiliated individuals striving to become regular members may play a role in the subculture’s crime pattern.
For all OBs the average age of PID registration is 28 years (the PID age). Yet this varies by group: HAMC (31); HAMC support groups (37); AK81 (23); BMC (30); BMC support groups (24). 3 Note also that age 28 is remarkably different from when US youths become part of street gangs, which is typically between the ages of 13 and 15 (Maxson, 2011: 163). All registered OBs are male and 97 percent of them have parents whose country of origin is Denmark (Klement et al., 2011). This means that the majority are most likely ethnic Danes.
Accurate PID dates are a requirement for inclusion in the study since it is otherwise impossible to distinguish between crime committed before and during OMC affiliation. OBs affiliated with non-Danish chapters and/or residing outside Denmark during the observation period may suffer incomplete criminal history data. Since the Danish age of criminal responsibility is 15 and criminal history records are available only from 1980, data are complete only for individuals born in 1965 or later. Since criminal history data are available only through 2010, individuals first registered in the PID in 2009 or later are excluded. Only a few exclusions are made owing to an absence of registry information on OB subjects or an inability to identify a suitable matched control. All of the OBs were 15 years of age when first recorded in the PID. A small number died within two years of their PID registration and are excluded.
Methodology
A fundamental problem of general causal inference is one of counterfactuals, which would ensure that treated and non-treated units are perfectly comparable (Morgan and Winship, 2007). Given the non-existence of counterfactuals, the most trusted way of ensuring that treated and non-treated units are comparable is by using randomized controlled experiments. However, randomized controlled experiments are neither foolproof nor the only way to examine causal relationships (Berk, 2005; Sampson, 2010). Furthermore, experiments on group membership and crime are difficult.
The current study uses a two-step procedure to estimate the effect of OMC affiliation on criminal involvement. The purpose of both steps is to render treated and non-treated individuals comparable. The first step consists of precisely matching individual OBs to non-OB control subjects identified in Statistics Denmark’s registry data. OBs and non-OB controls will occasionally be referred to as treatment and control subjects, respectively. The second step consists of estimating the effect of OMC affiliation on criminal involvement using DiD regression. The advantage of matching is that it increases comparability between treatment and control subjects and helps the model adhere to the underlying assumptions of the DiD approach (Cochran and Rubin, 1973; Morgan and Harding, 2006; Rubin, 1973). 4
Although DiD’s multivariate properties render it capable of estimating the independent effects of observed variables, its primary advantage lies in its ability to control for time-stable, unobserved differences between treatment and control subjects (Lechner, 2010). Bias caused by any time-constant differences between treatment and control subjects, for example geography, physiology, intelligence or mental illness, is therefore removed. Furthermore, studies indicate that matching combined with regression is superior to matching or regression alone (Rubin and Thomas, 2000). 5
A DiD regression requires two data points in time because individual values on the dependent variable at T1 are subtracted from individual values on the dependent variable at T2. In the current study, T1 is two years prior to the subject’s PID date whereas T2 is two years after the PID date. Time-stable unobserved individual characteristics are likewise subtracted from each other, which results in no bias caused by these sources. In effect, each individual acts as his own control and only within-individual differences over time are compared between individuals. DiD regression is explained in more detail in Appendix A.
This study includes some independent variables only at T1 in order to avoid post-treatment confounders. For example, OB status might affect not only criminal involvement but also educational attainment and employment. Educational attainment and employment are therefore included only at T1. Readers should note, however, that changes in educational attainment are relatively rare at age 28, which is the average age of PID registration.
As previously mentioned, registration in the PID presumably increases police surveillance, implying potentially different levels of police surveillance between OBs and control subjects. In an effort to take this into account, cases of possession of minor amounts of illegal drugs are used as a proxy for police proactivity. This seems like a good proxy because it typically requires close proximity between police and offender. In the DiD regressions the proxy is dealt with like any other time-varying independent variable.
Being longitudinal, the design accounts for both cross-sectional differences and differences over time. Furthermore, since PID dates allow for the measurement of criminal involvement both before and during OB affiliation, causal order can be specified.
Matching
Across the period July 2001 to June 2009, in which the 307 OBs (including 10 unmatchable OBs) were PID-recorded, the yearly average PID date is 14 August. This date is taken as an artificial PID date for potential control individuals, who do not have PID dates otherwise.
Control individuals with matching characteristics to the 307 OBs on the variables birth year, PID year (the year an OB is recorded in the PID), country of origin, number of overall convictions, number of convictions for property crimes, number of convictions for violent crimes, age at criminal onset, and lifetime number of days sentenced to prison were selected on the basis of 14 August 2001 through 2008. The pool from which the matching control individuals were selected included a representative sample of 12,000 individuals with no criminal record prior to their artificial PID date provided by Statistics Denmark. Using the OBs’ registered PID dates and the control individuals’ artificial PID dates, information on the following variables was derived: highest educational achievement, employment status, age at criminal onset, number of days sentenced to prison as well as number of convictions for property crimes, violent crimes, drug crimes, weapon crimes, pimping, sex crimes, and all types of crimes. For every unique combination of birth year, country of origin and PID year, a potential match can be selected only once as a control subject. Besides that, the maximum number of controls per subject was limited only by the number of suitable matches. This resulted in the same control individual in some cases being the control individual of more than one OB.
Controls that had died or were not residing in Denmark between their 15th birthday (the age of criminal responsibility) and two years after their artificial PID date were not included in the pool of potential controls. PID-recorded street gang members were likewise excluded. Controls were excluded if they had more overall convictions, a greater number of days sentenced to prison, a later age of criminal onset or a higher level of educational attainment than OB subjects. Furthermore, the potential control pool did not include females because all registered OBs in the PID are male.
The matching resulted in the selection of 181,931 unique control individuals. It was possible to find exact matches to one or more controls for only 297 of the 307 OB subjects. Ten OBs were unmatchable because of their unique criminal careers, that is, a considerable number of violent crimes coupled with an unusually low number of property crimes and late age of criminal onset. The reuse of the 181,931 unique control individuals amounts to 352,045 non-unique control units.
The OBs and controls differ on various characteristics when match variable categories are not taken into account. OBs are in general more unemployed and less educated than their controls. On average, their criminal onset is at an earlier age and they are in general younger than their controls. These differences are all highly significant.
Table 3 shows, without taking match variable categories into account, average overall number of convictions, number of convictions for specific crime types and number of days sentenced to prison at three periods: two years before the PID date; on the PID date; and two years after the PID date.
Average numbers of convictions and days sentenced to prison (N = 352,342).
Note: Levels of significant results from t-tests comparing bikers and controls within each of three time points are indicated in the upper cell (relating to Bikers).
p < .05; *** p < .001 (two-tailed t-test).
On average, OBs have been convicted more frequently than controls regardless of crime type. The only exceptions are sex crimes and pimping. OBs also tend to have been sentenced to more days in prison than controls. Tables 3 and 4 make clear that OBs are a highly select group in terms of criminal involvement. At the average PID age (28 years as mentioned), the average OB will have been convicted of 18 crimes, including two drug crimes and two violent crimes.
Average birth year, PID age and age at criminal onset distributed across match variable categories.
Note: Average PID year is not shown because each PID year was coded separately, resulting in no variation between the OBs and their controls.
p < .05 (two-tailed t-test).
When match variables are taken into account, differences between the OBs and their controls are reduced. Table 4 shows the average birth year, PID age and age at criminal onset distributed across applied match variable categories. Despite accounting for match variable categories, Table 4 still shows significant differences in 3 out of 13 cases concerning birth year, PID age and age at criminal onset.
Table 5 shows the average number of convictions, by category, as well as the average number of days of sentenced prison time two years before the PID date, on the PID date and two years after the PID date.
Average number of convictions, by category, and average number of days of sentenced prison time, by match variable category.
Note: Levels of significant results from t-tests comparing bikers and controls within each of three time points are indicated in the upper cell (relating to Bikers).
p < .05; ** p < .01; *** p < .001 (two-tailed t-test).
Two years before the PID date, 8 out of 19 comparisons across OBs and controls differ significantly from one another. On the PID date itself, 6 out of 19 comparisons differ significantly, and two years after the PID date 13 out of 19 comparisons differ significantly.
In all eight comparisons showing significantly different proportions two years before the PID date, it is the control group that has the higher average. However, the control group has the higher average in only 1 of the 13 comparisons yielding significant differences two years after the PID date. It seems as if the control group is more involved in crime than OBs before the PID date and less involved than OBs after. This raises the concern that OBs and their controls may be following different criminal trajectories at the group level, and that a pre-treatment dip among the OBs may be present. A pre-treatment dip signals a drop in the dependent variable for one of the two groups prior to treatment, that is, the PID date, which may lead to biased estimates (Heckman and Smith, 1999: 317). The possibility of such a drop is considered below.
In spite of individual exact matching, significant differences in PID date still exist between OBs and their controls. Additional adjustment is therefore required in order to reduce the potential for selection bias.
Results
The 46 unstandardized DiD regression estimates in Table 6 show the effect of an affiliation with an outlaw motorcycle club on criminal involvement. Estimates are shown for convictions for overall crime, specific types of crime and different combinations of control variables in two specific ‘differencing periods’. The dependent variable in Differencing Period A is the sum of the number of convictions during the two-year period prior to the PID date subtracted from the number of convictions during the two-year period after the PID date. The dependent variable in Differencing Period B is the sum of the number of convictions registered during the two-year period two to four years prior to the PID date subtracted from the number of convictions registered during the two-year period just before the PID date. Since a case of possession of minor amounts of illegal drugs is used as a proxy for police proactivity, this type of offence is not included in the dependent variables in the regressions below covering all types of convictions and drug convictions. Drug convictions cover more serious drug crimes such as selling and trafficking illegal drugs. The number of convictions for pimping and sex crimes was so small that they were excluded from the analysis.
Unstandardized DiD regression estimates (N = 352,342).
p < .01; *** p < .001 (two-tailed Wald-test).
As Table 5 indicates, it is possible that a pre-treatment dip in criminal convictions, that is, the dependent variable, exists among the OBs. Differencing Period B is designed to allow examination of this possibility. If the estimates obtained in Differencing Periods A and B are similar, the influence of a pre-treatment dip can be ruled out. The estimates for Differencing Period B should therefore not be interpreted as estimates of an effect, but rather as a method of validating the estimates obtained in Differencing Period A.
Of the 46 regressions, 36 have control variables (all rows except 1 and 6 in Table 6) covering age, PID year, age at criminal onset, employment at PID year, highest educational attainment at PID year as well as a police proactivity proxy. In addition to these control variables, there are various combinations of additional control variables used in the 26 regressions for which output is shown in rows 3 through 5 and 8 through 10. These additional variables adjust for the number of prior convictions – both overall and in specific crime categories. The reason some of the match variables are also used as control variables is because they are more fine-meshed coded as control variables than as match variables.
In rows 1 and 6, the treatment variable (affiliated or not) is the only independent variable included in the models. In rows 2 and 7, all control variables are included in the models except number of convictions in general and for specific types of crime. In rows 3 and 8, all control variables are included in the models except number of convictions in general. In rows 4 and 9, the models are identical to those in rows 3 and 8 with the exception that the overall number of convictions is now included and variables representing convictions for specific types of crime are dropped. In rows 5 and 10, all available control variables are included. 6
Since use of a particular crime type control variable is unnecessary when estimating the effects on crime in general, 4 out of 10 of the estimates for crime in general are not reported (rows 3, 5, 8 and 10 under the column ‘All types’). DiD regression outputs for convictions in general and convictions for drug crimes in row 4 are provided as examples in Appendix B. All other DiD regression outputs are available upon request.
All the estimates in Differencing Periods A and B are positive and highly significant (with the exception of violent crime in Differencing Period B), suggesting that affiliation with an outlaw motorcycle club leads to an increase in crime in general and in all crime types, which effectively rules out the possibility of a pre-treatment increase. For each column, the effect size is smaller in Differencing Period B than in Differencing Period A. This is logical because the effect of being affiliated has not taken place in Differencing Period B. Estimates for violent crime in Differencing Period B are the exception, being negative and significant, indicating a pre-treatment dip. Within the 10 combinations of type of conviction and Differencing Period, the estimates are rather similar regardless of whether or not control variables are used and if so regardless of the combination of control variables. This is probably because the two groups are already matched.
In almost all of the regressions the police proactivity proxy is positive and significant, as you would expect based on the conventional wisdom that increased police attention increases the rate of arrest and ultimately convictions.
The various combinations of control variables have different advantages and disadvantages. Identification of the best specified model is debatable. However, when considering the effect of affiliation on overall convictions and specific types of convictions in Differencing Period A, all models appear to be robust in terms of effect size and significance, which makes an elaborate prioritization of these redundant.
Discussion
Applying a robust methodology to a unique dataset containing a diverse assortment of individual-level variables, this study indicates that becoming affiliated with an outlaw motorcycle club causes an overall increase in criminal involvement, specifically involvement in property crimes, drugs crimes and weapons crimes. The estimates obtained are noteworthy in size and highly significant. Affiliation with an OMC facilitates criminal behaviour. An increase in violent crimes is also observed, but this may be due to a pre-treatment dip, that is, a drop in crime prior to affiliation. Given this, questions as to the effects of affiliation on violence remain inconclusive. The reason behind the pre-treatment dip is unclear. Furthermore, it is clear that OBs differ from the general population in terms of criminal involvement even prior to affiliation. On average, an OB has already been convicted of 18 crimes when PID-recorded at a mean age of 28. In addition to the facilitation effect, a selection effect therefore seems clearly present. The result is thus consistent with the enhancement hypothesis and with earlier findings from gang studies.
PID registration almost certainly results in heightened police surveillance, which may explain at least some of the increase in criminal involvement among those registered. However, the matched control subjects have criminal histories just as extensive as the OBs, which means that they should also be subject to a high degree of police attention. The fact that the facilitation effect is observed across different crime types suggests that the results are caused by OMC affiliation as opposed to increased surveillance. Furthermore, the potential difference in the level of surveillance is at least reduced if not eliminated by using cases of possession of minor amounts of illegal drugs as a proxy for police proactivity in the study’s regressions. This makes it less plausible that heightened surveillance is the driving factor behind the significant results.
There are a few issues to consider when evaluating whether the results of this study can be generalized to the broader population of OBs. Ten OB affiliates are excluded from the analysis owing to an inability to find proper controls. However, it seems very unlikely that the criminal trajectories of the 10 excluded OBs would substantially alter the study’s results. Generalization to the broader population of OBs is far more seriously threatened by the fact that the study includes only individuals who were OBs at the time that the PID information was provided and only those residing in Denmark from age 15 until two years after PID registration. One cannot assume that excluded OBs, including non-resident OBs, are similar to the remaining OBs in all other respects. Both characteristics imply selection. Furthermore, only OBs born in 1965 or later and PID-recorded between July 2001 and December 2008 are included in the analysis. The bikers included in the study are therefore the youngest possible subsample of those in the data. They can thus be considered a reasonably representative sample of future OBs in Denmark.
The results of this study have a number of implications. From a methodological standpoint, the identification of a pre-treatment dip suggests that it might be important to check for dips or increases prior to treatment, as these may have considerable consequences for results. This said, prior studies using matched control groups have rarely conducted these important analyses. Relatedly, another methodological lesson of the study is the importance of matching treated and non-treated subjects at more than one point in time in order to ensure similarities in criminal trajectories among the two groups.
The total absence of females in the OB sample is striking. This implies that masculinity plays an important role in OB culture and does so in such a way that it excludes females. However, since this study does not shed much light on the issue, it remains to be studied further. Another implication is that the enhancement effect is relevant to a broader spectrum of groups than only the youth and street gangs in which it has previously been studied, whose members generally have a lower average age than OBs (Maxson, 2011: 163). From a more general theoretical standpoint, the study’s results support neither a pure selection nor a pure facilitation mechanism, but rather a more complex enhancement process. These results lend support to diverse theories such as Gottfredson and Hirschi’s (1990) self-control theory, Akers’ (2009) social learning theory, and Sampson and Laub’s (2005) age-graded theory of informal social control. Both selection and social relations seem to matter in the commission of crime. However, when and how they matter is less clear. A way to advance general criminological theory would be to test more than one theory simultaneously in an effort to identify the exact mechanisms at work in the selection and facilitation.
Footnotes
Appendix A
DiD regression is expressed by the following equation:
where Δ represents the difference between T1 and T2, y represents the dependent variable, β represents the effect of Δx, x represents the independent variable of theoretical interest, γ represents the effect of Δz, z represents a vector of control variables, δ represents the effect of control variable q, which is included only at T1, and ε represents the error term.
Appendix B
Table 7 provides DiD regression outputs concerning the effects of OMC affiliation on convictions for crime in general and for drugs crimes, respectively, both measured within the context of Differencing Period A.
Acknowledgements
An essential part of this study’s data was provided by the Danish National Police. The study was co-financed by the Danish Ministry of Justice and the University of Copenhagen. Insights gained during a visit to the University of California, Irvine, had important implications for the study. I’m grateful for grants that made this visit possible from the Oticon Foundation, the Foundation of the 8th of July, and Knud Højgaard’s Foundation. For advice of various types, I am indebted to Lin Adrian, Susanne Ditlevsen, Peter Fallesen, John R. Hipp, Anders Holm, Kristian Karlson, Britta Kyvsgaard, Dave Sorensen and Torben Tranæs.