Gang membership, gender, and sexual behavior in and outside a romantic relationship

Abstract

Prior research consistently finds that gang youth, compared to nongang youth, are more involved in risky behaviors such as violence and drug sales. Less attention has been given to comparisons in sexual behavior. While research demonstrates that gang membership is associated with risky sex and many gang members indicate that sex is a motivation for joining a gang, prior research is limited in its ability to account for self-selection into a gang, variations in involvement across gender, and different forms of sexual activity. This research addresses these limitations by using Add Health data and propensity score matching to examine the relationship between gang membership and sexual behavior inside and out of a romantic relationship as well as how this relationship differs by gender. While findings indicate that gang membership increases the likelihood of sexual intercourse, nonromantic sex, and the number of nonromantic sex partners, no unique gender differences were identified.

Introduction

Research consistently finds that gang membership increases the prevalence and frequency of adverse behaviors. While most of this research has focused on the relationship between gang membership and interpersonal violence (Melde and Esbensen 2013; Pyrooz et al. 2016; Thornberry et al. 1993) or victimization (Fox 2013; Melde, Taylor, and Esbensen 2009; Taylor et al. 2007), some research has explored the impact of gang membership on other risky behaviors such as drug dealing, police contact, and truancy (Decker and Van Winkle 1996; Esbensen and Carson 2012; Tapia 2011; Wiley, Carson, and Esbensen 2017). Decker and colleagues (2013) note, however, that there is a need for more research on the relationship between gang membership and non-criminal behaviors, especially those behaviors that might “pull” youth to the gang. Sexual activity may fall into this category given certain dynamics that are common to gang involvement. For instance, some youth report joining a gang for reasons that deal with peer socializing, such as “to party” or “to hang out,” reflecting the important role that gangs play in facilitating social relationships (Decker and Curry 2000; Esbensen and Winfree 2013; Sanchez-Jankowski 1991). These gatherings or periods of unstructured socializing often account for a lot of a youth’s time in the gang (Klein 1995; Lauger 2012), as well as offer an opportunity to interact with gang or nongang associates of sexual interest. Indeed, some male youth indicate that the opportunity to “access” or “impress” females was a primary reason for joining the gang (Curry 2000; Decker and Van Winkle 1996; Padilla 1992; Palmer and Tilley 1995). In general, prior literature finds a link between gang membership and sexual behavior, which is not surprising given that sexual encounters typically commence in adolescence, or around the time youth are likely to join a gang (Sanders, Lankenau, and Jackson-Bloom 2009).

Sexual behavior is a part of normal adolescent development, but research suggests that such behavior can be problematic when it occurs with nonromantic partners. Nonromantic sex has been identified as a risky behavior (Fortenberry 2003; Manning, Longmore, and Giordano 2005) in that it increases the likelihood of sexually transmitted infections and unwanted pregnancies (Ford, Sohn, and Lepkowski 2001; Manning, Longmore, and Giordano 2000; Norris et al. 1996). Increased involvement in risky behaviors, such as nonromantic sex, can further ensnare gang members (e.g., Moffitt 1993) and exacerbate the long-term consequences of gang involvement. Additionally, nonromantic sexual activity requires less commitment than sex with a romantic partner; therefore, regularly engaging in this sexual behavior may limit the development of the social skills necessary to form and maintain romantic relationships in adulthood (Furman and Simon 1999). This is a particularly notable consequence for gang youth as prosocial relationships with significant others can help youth break away from the gang (Carson 2018; Decker and Pyrooz 2011; Moore 1991).

Gang youth may be more likely than nongang youth to engage in sex with both romantic and nonromantic partners. However, existing research suggests that there may be important variations in this relationship based on gender. This research often identifies a strong double standard in which gang members encourage males to be sexually active, but judge females harshly if they engage in such behavior, especially if these encounters involve a nonromantic partner (Campbell 1984a; Miller 2001; Quinn et al. 2019). While certain processes such as “sexing-in” at initiation may increase the likelihood of nonromantic sex partners, Miller’s (2001) work indicates that more control is exercised over the sexual behavior of females once in the gang, serving as a “gendered sexual code” that questions the “respectability” of females, but not males, when they become sexually involved with nonromantic partners (Miller 2002; Quinn et al. 2019; Schalet, Hunt, and Joe-Laidler 2003). Understanding the relationship between female gang membership and sexual behavior is important given the risky nature of nonromantic sexual activity and that precocious entry into pregnancy is one potential consequence of gang involvement for females (Thornberry et al. 2003).

While previous research has been invaluable in establishing a link between gang membership and sexual activity, this work is limited in a few important ways. First, prior research has relied on samples drawn from a single city or composed of only males or females. This limits the generalizability of prior work as findings from the Centers for Disease Control and Prevention (CDC) indicate that sexual behaviors vary across regions, race/ethnicity, as well as gender (Ethier, Kann and McManus 2018). Second, the reliance on cross-sectional data limits the ability of prior research to account for self-selection into a gang. Prior research demonstrates that self-selection processes are present in gang joining especially for offending and victimization (Fox 2013; Melde and Esbensen 2013; Melde, Taylor, and Esbensen 2009; Pyrooz et al. 2016; Taylor et al. 2007; Thornberry et al. 1993). It is likely these selection processes extend to sexual behavior as well especially given that male gang members often cite “to get girls” as a motivation for gang joining. Finally, prior work is limited by its ability to determine whether sexual behavior occurred with a romantic or nonromantic partner. In this study, we improve upon the existing research connecting gang membership and sexual behavior by empirically examining this relationship among a national sample of male and female youth using two waves of data. The use of these data will vastly improve our understanding of this relationship and allow us to empirically compare the sexual activity of male and female gang members to identify possible differences in this behavior with romantic and nonromantic partners.

Gender differences in sexual behavior among gang members

Self-report research consistently finds that females account for 25 to 55 percent of gang members (Esbensen and Carson 2012; Estrada et al. 2016; Klein and Maxson 2006; Pyrooz 2014b; Thornberry et al. 2003). While males remain more deeply embedded in gang life (Decker et al. 2014), research also indicates that female gang members are not just on the periphery of gang life (see Esbensen et al. 1999; Peterson et al. 2001; Panfil and Peterson 2015) or mere “look-outs,” but take an active role in the gang by selling drugs, committing a crime, and participating in fights. However, the gang experience for females remains highly gendered and laced with misogyny (Joe and Chesney-Lind 1995; Peterson and Panfil 2014). The gendered treatment of female gang members is especially apparent in the sexual double standard described by qualitative researchers (Miller 2001; Moore 1991; Moore and Hagedorn 1996; Quinn et al. 2019). It is considered normal, even encouraged, for male gang members to have multiple sexual partners (Dickson-Gomez et al. 2017; Quinn et al. 2019). For male gang members, sexual prowess is associated with increased feelings of masculinity and members can face ridicule for refusing a sexual opportunity (Dickson-Gomez et al. 2017; Messerschmidt 1995; Portillos 1999). The sexual behavior of females, whether with romantic or nonromantic partners, is much more controlled by the gang (Miller 2001).

Female gang members are often more harshly stigmatized for engaging in nonromantic sexual activity. This stigmatization can come at the hands of both male and female gang members (Cepeda and Valdez 2003; Messerschmidt 1995; Miller 2001; Quinn et al. 2019). For instance, some research suggests that out of concern for their sexual image, female gang members socially distance themselves from female peers considered too promiscuous, a behavior that risks being viewed as nothing more than a “ho” or “slut” among male gang members (Cepeda and Valdez 2003; Hunt and Joe-Laidler 2001; Kolb and Palys 2016). Females who are “sexed in” to the gang especially risk such a label. The literature suggests that sexual initiation is reserved for females who cannot withstand other means of gang entry such as being “jumped in” (Miller 2001; Quinn et al. 2019), and females who are jumped in view their sexed-in counterparts in a derogatory manner (e.g., hoodrat) and consider them less central or loyal to the gang (Cepeda and Valdez 2003; Miller 2001; Portillos 1999). Additionally, once a female goes through a sexual initiation process, they are often expected to continue to have sexual relations with male members (Cepeda and Valdez 2003; Peterson and Panfil 2014; Quinn et al. 2019). This is viewed as a way to prove their loyalty to the gang and increase group cohesion (Dickson-Gomez et al. 2017; Hunt and Joe-Laidler 2001). Research further suggests that while females risk their gang affiliation if they decline the advances of male members (Dickson-Gomez et al. 2017; Miller 2001), they do not view their sexual initiation as forced. For example, a respondent in Quinn and colleagues’ (2019:157) study was asked how she felt during her sexual initiation, and she stated that “[I was] Disgusted, but it was like, whatever, I’m doing it.” The research suggests that female gang members are particularly mindful of perceptions surrounding their sexual reputation in the gang, and how this reputation is shaped by their participation in nonromantic sex and by their method of initiation.

Research also suggests that a double standard exists with romantic partnerships. Male gang members may engage in sexual activity with romantic partners (Cepeda and Valdez 2003; Messerschmidt 1995) in and outside of the gang (Miller 2001). In Miller’s (2001) study, for instance, the practice of dating outside the gang was viewed as acceptable by male members, but female members who sought relationships with males from rival gangs were viewed as being disrespectful and risked increasing tensions between the two gangs. Research also suggests that gang females face challenges when becoming romantically involved with nongang peers. Gang membership may be more stigmatizing for females given that gang involvement has been traditionally viewed as a male behavior, making gang-involved females “less attractive to boys outside the gang” and thereby narrowing their “dating options” (Miller 2002:191). Female gang members may also have to deal with male members being especially possessive (Moore 1991), which likely makes it difficult to sustain a romantic relationship with a nongang peer when such partnerships form. Even when females are romantically involved with a male in their own gang, they must still contend with a gang environment in which males consider it acceptable, even preferable, for other males to have multiple sexual partners. Such infidelity on the part of gang boyfriends places gang females at increased risk of contracting a sexually transmitted infection (Cepeda and Valdez 2003). Additionally, when the infidelity of gang boyfriends involves fellow female members, it can result in intra-gang violence among females in the gang (Campbell 1984b). Collectively, these dynamics suggest that even though female gang members, unlike males, are expected to engage in “serial monogamy” (Miller 2001, 2002), they encounter an environment in and outside the gang that is inclined to disrupt the formation and duration of romantic relationships.

Sexual experiences within the gang context have major implications for females who are already at risk. Not only are they at risk for sexually transmitted infections and unplanned pregnancy (e.g., Thornberry et al. 2003), but these experiences can affect future sexual and romantic relationships. For instance, it may be difficult for females with a history of gang involvement to develop intimacy and trust in romantic relationships. Additionally, research from Wesche and Dickson-Gomez (2019) suggests that inequitable treatment of females in the gang is related to intimate partner violence and forced sex.

Review of empirical literature

Several empirical or mixed-methods studies have addressed the prevalence or correlates of sexual behaviors among only current or former gang members. Studies involving a sample composed entirely of current or former gang members often employ a public health approach, in which the extent or potential drivers of risky sexual activities are assessed among youth known to be more engaged in adverse behaviors (Sanders and Lankenau 2006). Most of this research involves either male or female gang members or involves no descriptive or multivariate analyses that consider possible differences by gender when both male and female gang members are in the sample. This includes a study by King and colleagues (2013) who surveyed males and females aged 14 to 18 in eight detention centers in Georgia, with each youth indicating that he or she was a member of a gang (see also Brooks et al. 2009; Sanders et al. 2009, 2013). Such research consistently finds that a sizeable percentage of gang youth report being sexually active and at least occasionally involved in risky sexual behaviors. King et al. (2013) found, for instance, that nearly 35% of gang youth in their sample reported having sex with two or more people at least once.

Some studies involving only current or former gang members have assessed the prevalence of certain sexual behaviors by gender. This includes a study by Wesche and Dickson-Gomez (2019) who surveyed a community-based sample of 281 male and female gang members aged 14 to 19 in a Midwestern city. Nearly 60% of the males and 70% of the females who recently had sex reported vaginal intercourse without a condom in the prior month, and 5% of all males and females reported ever having “sex with someone when they said they did not want to.” Overall, Wesche and Dickson-Gomez (2019, p. 651) found modest differences between male and female gang members in the prevalence or frequency of various sexual behaviors (e.g., group sex prior 30 days), although female gang members were more likely to report being the victim of “forced sex” and “gang rape” in their lifetime.

Additional research has compared the prevalence and/or predictors of sexual behaviors among a sample of gang and nongang youth. Some of this research has involved only male or female samples, such as Voisin and colleagues (2004) study of the relationship between gang membership and sexual activity among detained males aged 14 to 18. They found that a history of gang involvement was strongly associated with sexual behaviors that increase the risk of a sexually transmitted disease (e.g., had sex while high on drugs or alcohol). Other studies involving a single-gender sample with both gang and nongang youth have also found a relationship between gang membership and sexual activity, including Harper and Robinson’s (1999), Wingood et al.’s (2002), and Voisin et al.’s (2014) research with African American female adolescents.

Other studies have involved a sample of males and females in which both gang and nongang youth were represented. Such research includes a study by Voisin et al. (2008) who surveyed nearly 600 male and female youth in detention centers throughout Georgia. They found no direct relationship between lifetime gang membership and an index measure of “HIV risk behaviors,” which included items that inquired about recent sex without a condom or while high on drugs or alcohol. However, this research involved no gender-specific analyses, which was the case in a study by Voisin and Neilands (2010). They surveyed more than 500 African American high school students in a large Midwestern city and found a significant direct effect of lifetime gang membership on an index measure of risky sexual behaviors for males but not females.

Each study mentioned above generally involved a cross-sectional design in which gang membership and sexual behaviors were measured at one point in time among a selective or non-representative sample. Such studies are unable to account for the possibility that more sexually active or risk-prone youth are more likely to join a gang, as the evidence generally indicates with antisocial and risky sexual behaviors (Bendixen, Endresen, and Olweus 2006; Higginson et al. 2018). In cross-sectional studies, therefore, observed differences between gang and nongang youth, or gang males and gang females, may simply reflect differences that predated gang involvement. Longitudinal studies offer an opportunity to account for such pre-gang differences and to better isolate the potential facilitative effect of gang membership on sexual behaviors. We are aware of only a few longitudinal studies that have assessed the relationship between gang involvement and a sex-related outcome. Minnis and colleagues (2008) completed a two-year follow-up study of mostly Latina (female) adolescents from a San Francisco neighborhood. They found a significant relationship between a male partner’s gang involvement and pregnancy but not between a female’s own gang involvement and pregnancy. Several large longitudinal studies of youth (e.g., Denver and Pittsburgh Youth Study) have also included measures of sexual activity, but only the Rochester Youth Development Study (RYDS) has directly connected gang membership and a dependent measure of sexual behavior. Lanctot and Smith (2001) found that being a gang member significantly predicted early sexual behavior among RYDS females, and Thornberry and colleagues (2003) found that gang membership was associated with teenage parenthood among RYDS males and males.

Current study

The current study addresses the following research questions:

1. Does gang membership facilitate sexual behavior with romantic and nonromantic partners?

2. Does the relationship between gang membership and sexual behavior vary based on gender?

By answering these questions, this study adds to the existing literature connecting gang membership and sexual activity in a number of ways. First, most of the existing research is cross-sectional and/or has generally involved non-representative samples of mostly at-risk youth such as detained or inner-city adolescents. The totality of this research indicates that gang youth are more sexually active than nongang youth, but to what extent this relationship may be attributable to selection (i.e., a pre-gang difference) or facilitation (i.e., a during or in gang difference), or to unique aspects about the sample, remains unclear. We address these limitations by drawing on a national sample of youth and by controlling for selection effects using propensity score matching with two waves of data. Second, the existing research has not typically examined whether the sexual behaviors of interest occurred in or outside a romantic relationship. This may be related to prior research focusing on sexual behaviors that are risky regardless of whether they took place with a romantic or nonromantic partner. We address this limitation by exploring the impact of gang membership on romantic and nonromantic sexual activity. This differentiation is important to understand because nonromantic sex is associated with more adverse consequences (e.g., sexually transmitted infections and unplanned pregnancy; see Fortenberry 2003). Finally, prior empirical research typically focuses on one gender or fails to formally examine gender variation in the relationship between gang membership and sexual activity. Qualitative research, however, suggests that the relationship between gang membership and sexual activity varies by gender due to a sexual double standard in the gang. Males are free to participate in sexual activity, whether romantic or otherwise, while the sexual behavior of females is more controlled.

Methods

Data

These questions are addressed with two waves of data from the National Longitudinal Study of Adolescent to Adult Health (Add Health). The Add Health study initially identified 80 high schools from the Quality Education Database (QED; N = 26,666 U.S. high schools), which was considered “the most comprehensive list of high schools available” (Tourangeau and Hee-Choon 1999:2). Each of these 80 high schools had the 11th grade and enrolled more than 30 students, and they were selected from the QED sampling frame after being stratified by region, urbanicity, type (public, private, or parochial), size, and racial composition, resulting in a representative sample of U.S. high schools based on these characteristics. Certain middle schools that “fed” these high schools were also selected, generating a combined sample of 132 middle and high schools. Students attending these schools were then stratified by grade and sex, and roughly 200 students per high school and feeder school pair were selected to participate in an in-home interview. Nearly 21,000 students in the “core” and “special supplemental” samples completed the first (Wave I) in-home interview in 1995. Adolescents in grades 7 through 11 at Wave I was interviewed again roughly one year later in 1996 (Wave II; see Harris 2013). Youth who completed the first and second in-home interview (N = 14,736) served as our initial sample.

This initial sample of youth was used to impute missing data, although some cases were excluded from our final analysis sample after imputation (n = 3,654). This included youth who were missing data on the Wave II independent (gang membership) and dependent (sexual behavior) variables (see von Hippel 2007). Additional youth were excluded if they lacked the Wave II sampling weight that accompanied the Add Health data. This sampling weight was applied in some analyses. A final group of youth was removed from the sample based on information from the third in-home interview, which took place roughly five years after the second in-home interview when the participants were young adults with an average age of nearly 22 years old. One question asked of Wave III participants was whether they “ever belonged to a named gang,” although the section in which this item was asked (section 26) began by indicating that “The next questions are about things you may have done in the past 12 months.” Roughly 16% of the participants who answered the Wave III gang question responded in the affirmative. This percentage is three times greater than the percentage who reported “yes” to being “initiated into a named gang” in the past year at Wave II (see below), when the Add Health participants were in adolescence and at an age in which gang involvement is more prevalent (see Pyrooz and Sweeten 2015). Most participants therefore likely answered the gang question at Wave III assuming a recall period of “ever” and not the “past 12 months.” Similar to DeLisi et al. (2009) and Watkins and Melde (2016), we excluded youth from the sample if they reported “yes” to the gang membership question at Wave III but “no” to the gang membership question at Wave II. Such youth may have been current or former gang members at Wave II, but they would be coded as nongang youth if retained in the sample. The final analysis sample was reduced to 11,082 youth after removing these cases. Roughly 17% of these youth (n = 1,850) had missing data on at least one of the Wave I control variables, although no control variable had more than 5% of cases with missing data. Missing data were imputed using the ice command and multivariate chained equations in Stata (Royston 2007). Ten datasets were generated, and these imputed data were used in the analyses.

Variables

Dependent (wave II)

We examine five dependent variables that collectively address the prevalence and frequency of sexual behavior in and outside a romantic relationship after the first in-home interview. Each of these variables was generated from questions posed during the second in-home interview and are similar to measures used in prior studies (e.g., Manning et al. 2005). More specifically, adolescents were asked if they “ever had sexual intercourse,” and, if yes, the month and year of their “very first” and “most recent” sexual encounter. Sexual intercourse was defined as “when a male inserts his penis into a female’s vagina.” Sexual intercourse is coded one if a youth reported yes to the intercourse question and indicated that the date (month/year) of his or her very first or most recent sexual encounter occurred after the first in-home interview. Youth who reported no sexual intercourse or indicated that it occurred before the first in-home interview were coded zero.

Subsequent questions as part of the second in-home interview permitted the opportunity to identify whether sexual activity between waves involved a romantic or a non-romantic partner. Youth were asked to identify up to three persons they “had a romantic relationship with” in the last 18 months. For each romantic partner, youth were asked if they had “sexual intercourse” with this partner and, if so, whether it involved inserting his/your penis into her/your vagina. Further questions were asked about the date (month/year) of the first and most recent of these sexual encounters. Romance sex is coded one if a youth reported sexual intercourse with at least one romantic partner after the date of the first in-home interview and coded zero otherwise. Romance sex partners range from zero to two and indicate the number of romantic partners a youth reported sexual intercourse with between waves. The maximum number could be three, but fewer than 60 youth in the analysis sample (less than .5%) reported sexual intercourse with three romantic partners, and among these youth, only eight reported gang membership at Wave II. Given these sparse numbers, we combined these youth with the larger group of youth who reported sexual intercourse with two romantic partners.

Youth were also asked if they “had a sexual relationship with anyone” beside a romantic partner since the first in-home interview. Nonromance sex is coded one if a youth reported such a relationship and coded zero otherwise. Youth who reported in the affirmative were then asked the number of nonromantic partners they had a sexual relationship with between waves. Nonromance sex partners measure the number of persons a youth reported such a relationship with since the first in-home interview. Most youth reported no such relationship (82%) or one to four partners (15%), whereas a small percentage of youth in the analysis sample (less than .5%) reported a sexual relationship with more than 10 nonromantic partners. We coded these few cases as equal to ten and therefore the values on the nonromance sex partners variable range from zero to 10.

Independent (wave II)

Gang membership was inquired about in the second in-home interview when adolescents were asked “have you been initiated into a named gang” in the past 12 months. It is possible that some adolescents who answered no to this question were current or former gang members who joined their gang more than one year ago or who were never initiated. As mentioned, we used a question from the third in-home interview in an effort to identify and remove such youth from the sample. Nearly 5% (n = 527) of youth in our analysis sample answered yes to the gang membership question at Wave II. This percentage is in line with the prevalence of recent gang membership found in other studies involving larger school- or community-based samples of adolescents (e.g., Gottfredson and Gottfredson 2001; Pyrooz and Sweeten 2015).

Control (wave I)

A considerable number of factors across multiple domains (e.g., individual, peer, family, school, and neighborhood) are associated with an increased risk of gang involvement (Howell and Griffiths 2018), and some of these factors are also related to sexual behavior during adolescence (see Kotchick et al. 2001). We account for more than 25 control variables at Wave I that are consistent with measures found in prior studies that address correlates of adolescent gang and sexual involvement. These control variables include Wave I measures of our dependent variables dealing with sexual behavior. Appendix A lists and describes each of these Wave I control variables.

Analyses

The nonrandom selection of youth into gangs, specifically the greater likelihood for higher-risk youth to join, compounds efforts to isolate the effect of gang involvement on an outcome. Increasingly researchers have turned to propensity score analysis to address this selection bias and to estimate the impact of gang involvement on various attitudinal and behavioral measures (e.g., Melde and Esbensen 2011; Pyrooz 2014a), including scholars who have utilized data from the Add Health study (e.g., DeLisi et al. 2009; Watkins and Melde 2016). We also use propensity score analysis to assess the effect of gang membership on our measures of sexual behavior. Our analyses were shaped by the work of Green and Stuart (2014) and Wiley et al. (2017). Green and Stuart (2014) offer general guidance on testing for moderation effects using propensity scores, and Wiley et al. (2017) apply this guidance to examine whether gang membership moderates the relationship between arrest and measures of deviance.

We first estimated a propensity score for each youth in our analysis sample using logistic regression in which Wave II gang membership was the outcome and the Wave I control variables were the predictors. As suggested by Dugoff et al. (2014), we also included the Wave II sampling weight and “strata” dummy variables (i.e., four regions of the U.S.) that address the original Add Health research design. This logistic regression model was estimated for the total sample and then separately for males and females, resulting in a “joint” and “gender-specific” propensity (i.e., probability of membership) score for each youth (see Green and Stuart 2014). We found that these two scores were highly correlated within gender (r > .90) and that good covariate balance was achieved regardless of whether we matched gang and nongang youth by their joint or gender-specific score. We elected to present results from our analyses matching the gender-specific scores given the findings of Green and Stuart (2014).

The gender-specific propensity scores were then used to match gang and nongang youth within gender (i.e., males to males and females to females) using three different methods: 1-to-1, 1-to-10, and 1-to-many (kernel) matching. Different matching methods were considered in order to assess the sensitivity of our findings (see Apel and Sweeten 2010; Guo and Fraser 2015). Each approach imposed a common support restriction in which cases with a propensity score outside the overlapping region between the gang and nongang youth were excluded. Furthermore, our 1-to-1 and 1-to-10 matching allowed for replacement (i.e., a nongang youth could be matched to more than one gang youth) and required matched gang and nongang youth to have propensity scores within a range of .024 for males and .015 for females. These values are equal to one-quarter of the standard deviation of the propensity score for males (SD = .098) and females (SD = .062), a recommended threshold (Rosenbaum and Rubin 1985). Kernel (1-to-many) matching offers the advantage of retaining possibly every case in the control group, but this method may introduce bias by retaining more control cases that are poorer matches, although untreated individuals “are weighted by their distance in propensity score from treated individuals within a range, or bandwidth, of the propensity score” (Garrido et al. 2014:1710). We considered the bandwidth values of .06 (the default) and .02 using the biweight kernel function. The choice of kernel function is not viewed as important as the choice of bandwidth (Caliendo and Kopeinig 2008). We present results from kernel matching involving the smaller bandwidth (.02), which produced a better covariance balance between the gang and nongang youth while only reducing the sample size by three control cases. Each matching method was performed using the psmatch2 module in Stata 14.1 (Leuven and Sianesi 2003).

Weights generated from each matching procedure were then used to estimate a series of regression models that accounted for the complex survey design of the Add Health study using the svyset command in Stata 14.1 (see Chen and Harris 2020). Using this command, each treatment case or gang youth, regardless of the matching procedure, was assigned a weight of one, whereas the weight assigned to each matched control case or nongang youth differed by the procedure. However, the summed weights of selected control youth always equaled the total number of treatment youth retained in 1-to-1, 1-to-10, and kernel matching (n = 527 gang youth). These matching weights were then applied to generate an estimate of the average treatment effect on the treated (ATT) in our regression models. The ATT represents “the average effect for individuals who actually received the treatment” of gang membership (Dugoff et al. 2014:286). The second set of weights were also utilized in our regression models that multiplied the matching weight by the Wave II sampling weight to provide an estimate of the population ATT (PATT), which can be generalized to the broader Add Health target population. In contrast, our ATT estimates applying only the matching weights are specific to the sample (i.e., SATT; see Dugoff et al. 2014). The conclusions drawn from our analyses were not affected by the matching procedure (1-to-1, 1-to-10, or kernel) or weight (SATT or PATT) applied. For brevity, therefore, we present findings from the matching procedure that resulted in the best covariant balance (kernel) using the SATT weights, which have been used more often in prior studies than PATT weights.

Results

Descriptive and balance statistics for the covariates

Table 1 displays covariate descriptive and balance statistics for the matched and unmatched samples by gang membership. A total of 537 youth (4.8%) in the unmatched analysis sample (N = 11,082) self-reported gang membership at Wave II, and roughly one-third of these gang youth were females (n = 183). The summary statistics for the unmatched sample clearly indicate that gang youth were more involved in adverse behaviors (e.g., drug use and delinquency) at Wave I than were nongang youth. Gang youth were also more likely to report sexual intercourse with a romantic (34%) or nonromantic (34%) partner than were nongang youth (19% and 15%, respectively). The column labeled “% Abs Std Dif” displays a statistic that addresses the extent to which the variables are “imbalanced” across the control and treatment groups. For each continuous variable, this statistic reflects a percentage calculated by dividing the absolute mean difference between nongang and gang youth by the averaged standard deviation for both groups. A slightly different formula was used for the dichotomous variables (see Austin and Mamdani 2006). These standardized percentages are comparable to an “effect size or Cohen’s d” (Green and Stuart 2014: 777), with scores in excess of 10% suggesting “meaningful imbalance” (Austin and Mamdani 2006:2086). More than 80% of the variables (32 of 39) exceed this 10% threshold in the unmatched sample, with an average standardized difference of nearly 33%.

Table 1.

Descriptive and balance statistics for the unmatched and kernel matched samples by gang membership.

	Unmatched sample					Kernel matched sample
	Nongang youth		Gang youth		% Abs Std Dif	Nongang youth		Gang youth		% Abs Std Dif
	(n = 10,545)		(n = 537)			(n = 10,542)		(n = 527)
Variable	Mn/Prop	SD	Mn/Prop	SD		Mn/Prop	SD	Mn/Prop	SD
Male (0, 1)	.47	–	.66	–	39.75	.65	–	.65	–	.00
Age	15.80	1.58	15.76	1.56	2.57	16.07	1.54	15.74	1.56	21.28
White (0, 1)	.55	–	.34	–	44.07	.33	–	.34	–	1.42
Black (0, 1)	.18	–	.21	–	6.89	.22	–	.21	–	1.71
Hispanic (0, 1)	.16	–	.32	–	36.88	.31	–	.31	–	.88
Other race/ethnicity (0, 1)	.11	–	.14	–	9.96	.13	–	.14	–	1.54
First generation (0, 1)	.09	–	.07	–	4.24	.07	–	.08	–	.78
Both parents in home (0, 1)	.55	–	.42	–	25.47	.41	–	.42	–	1.60
Household number	3.62	1.61	3.99	2.04	20.15	3.95	2.06	3.96	2.02	.55
Same residence (0, 1)	.54	–	.46	–	15.37	.46	–	.46	–	1.09
Parental permissiveness	5.06	1.54	4.89	1.67	10.77	4.88	1.72	4.91	1.65	1.39
Maternal attachment	4.71	.53	4.64	.59	11.79	4.65	.60	4.64	.59	1.65
Maternal involvement	4.00	1.94	3.84	1.99	8.03	3.82	2.00	3.85	2.00	1.35
Grade point average	2.82	.76	2.31	.75	67.06	2.32	.74	2.33	.75	1.19
School bond	3.77	.84	3.53	.94	27.41	3.54	.93	3.55	.92	.91
School engagement	4.05	.76	3.50	.99	63.15	3.53	.96	3.52	.97	.95
Depression	.58	.39	.75	.44	42.76	.75	.44	.75	.44	.05
Thought about suicide (0, 1)	.12	–	.26	–	34.66	.24	–	.25	–	2.13
Negative future outlook	1.58	.57	1.82	.69	38.09	1.79	.66	1.81	.69	2.30
Impulsivity	2.20	.62	2.29	.67	13.57	2.27	.64	2.29	.66	3.58
Drug use	5.92	11.74	14.12	16.93	57.15	13.03	17.48	13.64	16.49	3.60
Friend drug use	2.22	2.51	4.06	2.93	67.58	3.99	2.92	4.00	2.92	.41
Delinquency	.19	.30	.62	.58	97.98	.56	.57	.59	.54	5.58
Victimization (0, 1)	.15	–	.47	–	74.44	.47	–	.47	–	.59
Witnessed violence (0, 1)	.10	–	.39	–	70.82	.36	–	.37	–	2.54
Safe in neighborhood (0, 1)	.90	–	.81	–	24.40	.82	–	.81	–	2.23
Neighborhood density	4.48	8.46	6.54	11.22	20.98	6.68	11.66	6.56	11.29	1.03
Neighborhood disadvantage	−.06	.96	.16	.96	22.51	.18	1.07	.15	.97	2.37
Ever sex (0, 1)	.31	–	.55	–	49.36	.55	–	.55	–	1.55
Recent sex (0, 1)	.26	–	.47	–	43.23	.47	–	.46	–	1.27
Romance sex (0, 1)	.19	–	.34	–	33.53	.33	–	.33	–	.73
Romantic partners	.21	.47	.38	.59	31.97	.37	.58	.37	.59	1.57
Nonromance sex (0, 1)	.15	–	.34	–	46.32	.35	–	.34	–	2.49
Nonromance partners	.43	1.66	1.30	2.82	39.01	1.33	3.16	1.25	2.74	2.69
Northeast (0, 1)	.15	–	.18	–	9.77	.20	–	.19	–	2.94
Midwest (0, 1)	.26	–	.18	–	17.85	.18	–	.18	–	.43
South (0, 1)	.36	–	.28	–	17.66	.28	–	.29	–	.66
West (0, 1)	.23	–	.35	–	26.49	.34	–	.34	–	1.47
Wave II sampling weight	1400.6	1268.7	1279.8	1252.2	9.59	1309.4	1240.3	1291.7	1253.2	1.42
Mean % abs. std. dif.	–	–	–	–	32.90	–	–	–	–	2.10
Sum of matching weight	–	–	–	–	–	527		–	–	–

Notes: SATT weights from kernel matching are used to generate means and proportions for matched samples.

Abbreviations: Mn: mean; Prop: proportion; SD: standard deviation; % Abs Std Dif: percentage absolute standardized difference.

Table 1 also displays the same statistics for the matching procedure that resulted in the best covariate balance, which was kernel matching. The SATT weights generated from this method were applied to produce the statistics for the matched sample. Ten gang youth and three nongang youth were excluded due to the propensity score common support and/or bandwidth restriction. Table 1 indicates that only the standardized difference for age exceeds 10% after the SATT matching weights were applied. Indeed, the mean standardized difference dropped from 33% in the unmatched sample to less than 2% in the matched sample.

Descriptive statistics for the dependent variables

Table 2 displays proportions and means for the five dependent variables measured at Wave II. These statistics are displayed by gang membership and gender for the unmatched sample, as well as for the kernel-matched sample applying the SATT weights. The distributions in the unmatched sample indicate that gang members (57.9%) were more likely than nongang members (34.4%) to report sexual behavior at Wave II, including with romantic and nonromantic partners, and that these gang-nongang differences in prevalence and mean number of sexual partners were more pronounced among males. After weighting, the gang-nongang differences for the total sample and by gender were substantially reduced for each outcome. In fact, there is little to no difference by treatment condition in the prevalence of romantic sex or the mean number of such partners among the total sample. The matched samples indicate that, interestingly, males report a slightly reduced likelihood of romantic partner sex under the treatment condition of gang membership relative to the counterfactual condition of no membership (39.1% gang and 38.3% nongang). For females, however, romantic partner sex remains slightly more prevalent under the treatment condition.

Table 2.

Descriptive statistics (means) for the Wave II dependent variables by gender and gang membership.

	Total		Males		Females
Dependent variable/sample	Nongang	Gang	Nongang	Gang	Nongang	Gang
Sex (0, 1)
Unmatched sample	.344	.579	.325	.602	.360	.536
Matched sample	.516	.571	.533	.591	.485	.533
Romance sex (0, 1)
Unmatched sample	.275	.402	.232	.393	.313	.421
Matched sample	.395	.395	.391	.383	.403	.418
Romance sex partners (0-2)
Unmatched sample	.309	.480	.263	.460	.349	.519
Matched sample	.468	.474	.472	.452	.461	.516
Nonromance sex (0, 1)
Unmatched sample	.148	.367	.176	.412	.124	.279
Matched sample	.283	.361	.315	.406	.222	.275
Nonromance sex partners (0-10)
Unmatched sample	.381	1.279	.507	1.568	.270	.721
Matched sample	.853	1.228	.995	1.496	.583	.720

Notes: SATT weights from kernel matching were used to generate means and proportions for total, male, and female samples.

Gang-nongang differences are more evident for nonromance sex and sexual intercourse more generally among the weighted samples. For instance, 36% of youth under the treatment condition in the matched sample reported nonromance sex compared to 28% under the counterfactual condition. This gang-nongang difference for nonromance sex is more pronounced for males. Overall, the descriptive findings in Table 2 suggest that gang membership increases the likelihood of sexual behavior for youth who join and that this facilitating effect holds for females regardless of whether it involves romantic or non-romantic partners. For males, the gang membership effect is limited to nonromance sex and the number of such partners, although this relationship is the most pronounced in Table 2.

Regression results

Figure 1 displays estimates from regression models that regressed each Wave II-dependent variable on gang membership (a estimates) and then separately on an interaction term of gang membership and gender (b estimates). Every model included age as a control variable given its standardized difference in Table 1, and the main effects of gang membership and gender were also modeled when estimating the interaction effect between these two variables. Figure 1 displays regression estimates from models applying the SATT weights generated from the kernel matched sample. We provide additional estimates in Appendix B from kernel and 1-to-1 matching applying both SATT and PATT weights. In total, 10 separate regression models are represented in Figure 1. Logistic regression models were estimated for the binary outcomes (i.e., sex, romance sex, and nonromance sex), ordinal regression models were estimated for the number of romantic sexual partners (ranging from 0 to 3), and negative binomial models were estimated for the number of nonromantic sexual partners (ranging from 0 to 10). The estimates (black dots) in Figure 1 represent the odds ratios for the binary and ordinal outcome variables and the incident rate ratio for the measure of nonromantic sexual partners. The vertical lines indicate the 95% confidence interval of these estimates.

Figure 1.

Regression estimates of the effect of the gang and gang x gender (male) on Wave II outcomes. Notes: Ten separate regression models represented in the figure. Black dots indicate odds ratio or incident rate ratio from regression models that regressed each sexual behavior on gang membership and gang member x gender (male), and vertical lines represent 95% confidence interval of these estimates. Estimates generated using SATT weights from kernel matched sample and using svyset command in Stata. Abbreviations: IRR: incident rate ratio. *p < .05 two-tailed test.

These estimates indicate that gang membership is significantly associated with sexual intercourse, nonromantic sex, and number of nonromantic sexual partners. For each of these outcomes, the odds or incident rate ratio for gang membership is roughly 1.5. This implies that compared to the counterfactual of no membership, the expected odds of reporting sexual intercourse or nonromance sex are 50% greater ([odds ratio – 1] × 100) under the treatment condition for those youth who ultimately reported gang membership. The incident rate ratio indicates that these youth have an expected rate of nonromance sexual partners that is 1.5 times greater under the treatment condition.

Once again, these estimates for gang membership (a models in Figure 1) are adjusted for the effect of age, which was also included as a control variable in the interaction models (b models in Figure 1). The interaction models also include the main effects of gang membership and gender. None of the estimates for the gang membership and gender interaction term is significant, although the direction of these estimates aligns with the findings in Table 2. For instance, these estimates indicate that under the treatment condition of gang membership, the expected odds of romance sex among males are smaller than among females, while the expected odds of nonromance are greater for males. In general, though, there is no compelling evidence in Figure 1 that gang membership does more to facilitate sexual behavior among males. In fact, the interaction estimates in Figure 1 suggest that the expected odds of sexual intercourse are slightly greater among females under the treatment condition.

Discussion

Gang members participate in more risky behaviors than nongang youth, including sexual behaviors. This research expanded our knowledge on the sexual behaviors of gang members compared with nongang youth by accounting for selection into a gang, possible variation in sexual behavior by gender, and different forms of sexual partnerships. Our results indicate that gang membership does increase the likelihood of sexual intercourse, particularly nonromantic sex and the number of nonromantic sex partners. These findings are consistent with the prior literature, but also advance that literature by more systematically controlling for selection into a gang and by relying upon a nationally representative sample. In combination with prior research, our findings provide support to the idea that, unlike other motivations for joining a gang such as protection (Melde, Taylor, and Esbensen 2009) and to make money (Augustyn, McGloin, and Pyrooz 2019), access to “girls” and casual sex does appear to increase when youth join a gang.

Our second research question asked to what extent the effect of gang membership on sexual behavior varies based on gender. Our results, while not reaching significance, suggest that the relationship between gang membership and sexual behavior may work differently for male and female gang youth. While it appears that the effect of gang membership on sexual intercourse is stronger for females, our results suggest that this effect is due to female gang members experiencing an increase in romantic sex compared with their male counterparts. This could be evidence of the sexual double standard found in qualitative research, specifically the expectation of serial monogamy among females (Miller 2001, 2002), in that the sexual encounters of female gang members may be controlled by fear of being labeled a “slut” or “hoodrat” by both male and female gang peers. This argument is especially convincing in light of the finding that male gang members experience a greater increase in the number of casual sex partners than their female counterparts, although this relationship was not significant.

These findings could be an artifact of the variation in motivations for joining a gang across males and females. While some males indicate that they join a gang to hook up with females, research suggests that female contact with gangs can be the result of a relationship with a significant other in the gang. Drawing on Schalet and colleagues (2003) work, our findings indicate that gang females lean slightly toward the discourse of sexual respectability due to their involvement with romantic partners. With that said, given the lack of significant gender findings, it may be that gang females are split between sexual respectability and sexual autonomy as suggested by Schalet and colleagues’ (2003) findings. Regardless of the underlying mechanisms, gang-involved females who maintain a steady partner are still at risk for sexually transmitted infections due to the high-risk sexual activities of their partners (see Cepeda and Valdez 2003).

These findings should be considered with certain limitations in mind. First, we did not account for some group- and individual-level factors identified in the literature as possibly related to the sexual behavior of female gang members. Such factors include the gender composition of the gang. The literature suggests that females are more likely to experience a strong double standard when a member of a predominantly male gang (Miller 2002; Panfil and Peterson 2015). This double standard is defined by subjecting females to more restrictive expectations about their sexual behavior, especially sexual behavior that occurs outside a romantic relationship. The literature suggests in particular that females are more likely to enforce such gendered expectations in mostly male gangs (Miller and Brunson 2000; Sutton 2017). Research also suggests that race or ethnicity may shape the level of agency that females have in the gang. For instance, Peterson (2012) indicated that while the available evidence is limited, research suggests that African American females may have more behavioral latitude than Latina females in mixed-gendered gangs. Another factor not accounted for in this study is prior sexual abuse and its relationship with both gang membership and sexual involvement. Some research has found that a history of sexual abuse is more prevalent among female gang members (Joe and Chesney-Lind 1995), suggesting it may be a stronger risk factor for gang membership among females (Belknap and Bowers 2016). The available evidence also indicates that prior or childhood sexual abuse is a significant predictor of later sexual activity, including more risky behaviors such as unprotected sex (Abajobir et al. 2017). Prior sexual abuse may also mediate the strength of association between gang membership and sexual behavior among females in particular. Future research should also consider how the method of initiation shapes subsequent sexual experiences among females in the gang. The existing research suggests that females sexed into the gang are viewed more derogatively than females who are jumped into the gang or initiated some other way. Finally, we encourage future research to look at these research questions using longer-term follow-ups. Much of the prior research points to longterm (e.g., post-gang membership) consequences of risky sexual behaviors as part of gang membership. In particular, future research should consider outcomes such as the number of partners, domestic violence, and pregnancy/child health outcomes.

Overall, the results of this research provide a deeper understanding of the sexual behaviors of gang members by looking at different types of sexual behaviors (e.g., romantic versus nonromantic) as well as by exploring possible gender differences. Our results suggest that gang members and those seeking to leave the gang would benefit from interventions that address risky sexual behaviors (e.g., casual sex and condom use; see also Dickson-Gomez et al. 2017; Wesche and Dickson-Gomez 2019) and promote healthy relationships (Voisin et al. 2008; Wesche and Dickson-Gomez 2019). While this type of programming would be suitable for both male and female gang members, our results indicate that females may benefit from programs that foster norms to promote women as equal romantic and sex partners (e.g., sexual subjectivity; Schalet, Hunt, and Joe-Laidler 2003).

Biographies

Adam M. Watkins is an Associate Professor in the Criminal Justice Program at Bowling Green State University. His research interests include gangs, juvenile delinquency, and program evaluation. His recent publications have appeared in the Journal of Criminal Justice, Journal of Crime and Justice, and Children and Youth Services Review.

Dena C. Carson is an Assistant Professor within the Paul H. O’Neill School of Public and Environmental Affairs at Indiana University-Purdue University Indianapolis. Her research interests include youth violence, victimization, gangs, and delinquent peer groups. Her recent publications have appeared in Criminal Justice Behavior, Journal of Crime & Justice, and Justice Quarterly.

Appendix A.

Control variables used in the creation of the propensity score for gang membership.

Wave 1 variable	Coding description
Age	Age of respondent in years
Race	Dummy variables for white, black, hispanic, and other race/ethnicity
First-generation	Coded 1 if not born in the U.S. and 0 otherwise
Both parents in the home	Coded 1 if both biological parents present in-home and 0 otherwise
Number in household	Number of residents in the household
Same residence	Coded 1 if reside in the same residence as 5 years ago and 0 otherwise
Parental permissiveness	Summed 7-item parental permissiveness scale
Maternal attachment	Averaged 2-item maternal attachment scale
Maternal involvement	Summed 10-item maternal involvement scale
Grade point average	Coded as the average 4-point GPA for English, history, math, and science
School bond	Averaged 3-item school/student bond scale
School engagement	Averaged 3-item school engagement scale
Depression	Averaged 19-item depression scale
Thought about suicide	Coded 1 if seriously thought about suicide past 12 months and 0 otherwise
Negative future outlook	Averaged 3-item negative future outlook scale
Impulsivity	Averaged 4-item impulsivity scale
Drug use	Summed 3-item drug use scale
Friend drug use	Summed 3-item peer drug use scale
Delinquency	Averaged 15-item delinquency involvement scale
Victimization	Coded 1 if jumped/assaulted and 0 otherwise
Witnessed violence	Coded 1 if saw someone shot/stabbed and 0 otherwise
Safe in neighborhood	Coded 1 if usually feel safe in neighborhood and 0 otherwise
Neighborhood density	Persons per square mile in responden’s census tract
Neighborhood disadvantage	A standardized score of socioeconomic status for responden’s census tract
Ever sex	Coded 1 if ever had sexual intercourse and 0 otherwise
Recent sex	Coded 1 if sexual intercourse prior 12 months and 0 otherwise
Romance sex	Coded 1 if sexual intercourse with a romantic partner and 0 otherwise
Romantic partners	Number of romantic sexual partners prior 12 months (range 0 to 3)
Nonromance sex	Coded 1 if the sexual relationship with nonromantic partner and 0 otherwise
Nonromance partners	Number of nonromantic sexual partners prior 12 months (range 0 to 10)
Region	Dummy variables for Northeast, Midwest, South, and West
Sampling weight	Add Health sampling weight at Wave II

Appendix B. Regression estimates of the effect of the gang and gang x gender (male) on Wave II outcomes.

Notes. Forty separate regression models represented in the figure. Black dots indicate odds ratio or incident rate ratio from regression models that regressed each sexual behavior on gang membership and gang member x gender (male), and vertical lines represent 95% confidence interval of these estimates. Estimates generated using both ATT and PATT weights from kernel and 1-to-1 matched samples and using svyset command in Stata.

Abbreviations: IRR: incident rate ratio; SATT: sample average treatment on treated; PATT: population average treatment on treated.

*p < .05 two-tailed test.