Beyond Peer Homophily: Cross-Age Collaboration in Juvenile Co-Offending

Abstract

Most delinquent behavior occurs within age-homogeneous peer groups. Using incident-level data from the National Incident-Based Reporting System (NIBRS), this study reassesses the extent to which contemporary juvenile group offending reflects peer-only networks versus cross-age collaboration. Results show that while juvenile-only groups remain the dominant pattern, approximately one-third of co-offending incidents involve adult participants. Mixed-age groups are associated with group size, offense type, and situational context, and are especially common in serious offenses such as homicide, aggravated assault, and drug crimes. Mixed-age co-offending is also associated with greater offense severity, particularly higher odds of victim physical injury. These findings have important implications for the criminal justice system’s response to juvenile crime. While most juvenile offending diversion programs currently focus on interventions that counter peer influence and reduce the time spent with peers engaging in antisocial behavior, intervention strategies that also address the facilitating role of adult co-offenders may also be necessary.

1. Introduction

Understanding the social organization of offending has long been a central concern in criminological theory. Classic perspectives emphasize that criminal behavior is learned and sustained through social interaction with others (Sutherland 1947). Despite some important exceptions (Stolzenberg and D’Alessio 2008), research frequently finds that delinquency is a group activity and that these groups are typically composed of similarly aged peers (van Mastrigt and Farrington 2009). Studies using both self-report surveys and official crime data consistently show that juvenile offending reflects strong patterns of peer homophily (Reiss and Farrington 1991; Warr 2002).

Group offending is understood within this framework to be shaped primarily by horizontal social processes operating within adolescent peer networks, including status competition, reinforcement, modeling, diffusion of responsibility, and emotional amplification (Warr 1996, 2002). Delinquency is commonly viewed as emerging from the dynamics of adolescent social worlds rather than from hierarchical or intergenerational relationships because these processes occur within peer groups comprising individuals at similar developmental stages. This emphasis on age-homogeneous peer interaction has led many theoretical accounts to conceive juvenile offending as primarily embedded within peer-based networks.

This peer dominance thesis aligns with broader sociological principles of homophily, the tendency of individuals to associate with others who are similar in age, status, and background (McPherson et al. 2001). Network research consistently finds age homophily to be one of the strongest organizing principles of social ties, particularly during adolescence when schools and neighborhoods structure interaction patterns (McPherson et al. 2001). This situation creates conditions where peer pressure intensifies. Within criminology, differential association and social learning traditions have long posited that deviant behavior is transmitted through intimate peer groups (Akers 1998), while group process theories stress the situational intensification that occurs when adolescents act together.

Developmental and psychological research also suggests that susceptibility to peer influence is heightened during adolescence and that association with older or more experienced peers can accelerate involvement in risk-taking and delinquency (Steinberg and Monahan 2007). Experimental evidence further shows that the mere presence of peers increases adolescent risk-taking and reward sensitivity to a degree not observed among adults (Gardner and Steinberg 2005). A substantial literature on peer contagion and deviancy training further demonstrates that exposure to deviant associates, who are frequently older or more entrenched in antisocial activity, predicts escalation in delinquency, substance use, and violence, and that interventions that aggregate such youth can inadvertently amplify these behaviors (Dishion et al. 1999; Dishion and Tipsord 2011). Developmental accounts of the maturity gap similarly suggest that younger adolescents may emulate the antisocial conduct of older peers to assert autonomy and mature status (Moffitt 1993). This body of research indicates that cross-age association is not merely incidental to delinquency but may itself operate as a mechanism of influence, lending theoretical support for a closer examination of the age composition of co-offending groups.

Collectively, these perspectives converge on a horizontal model of delinquent organization. Juvenile offending groups are assumed to consist largely of similarly aged youth embedded in reciprocal peer networks. This assumption has become deeply embedded in both theoretical reasoning and policy discourse. Prevention strategies frequently focus on peer clustering, school-based interventions, and youth network disruption because they implicitly presume that adolescent crime is primarily sustained within age-homogeneous groups.

Yet several theoretical strands suggest that this horizontal framing may be incomplete. First, social network theory distinguishes between homophily and structural embeddedness (Borgatti and Foster 2003). While homophily predicts similarity among ties, network structures can also exhibit vertical integration, with actors bridging across age strata. In criminological network research, adult offenders often occupy brokerage or central positions within illegal networks, linking younger participants to resources, information, or markets (Morselli 2009). Even if peer influence dominates everyday delinquency, more instrumental or organized forms of offending may exhibit cross-age integration. Age homophily may characterize friendship networks, but offending networks may not mirror them perfectly.

Second, theories of intergenerational transmission (Thornberry et al. 2003) and criminal embeddedness (Hagan and McCarthy 1997) suggest that criminal opportunity structures often extend beyond adolescent peer worlds. Older offenders may provide access to weapons, vehicles, or specialized knowledge. Social capital within criminal contexts can be vertically distributed, with experienced actors occupying positions of coordination or mentorship (McCarthy et al. 1998). In such cases, co-offending may involve asymmetric role differentiation rather than symmetrical peer reinforcement (Morselli 2009). The presence of older participants introduces the possibility of facilitation, recruitment, or hierarchical organization. These mechanisms are not fully captured by purely horizontal peer models (Warr 1996).

Third, routine activity and opportunity theories imply that age composition may vary by situational context (Felson 2002). School-based or recreational settings may generate age-homogeneous groups, whereas street-based, commercial, or nighttime environments may facilitate broader age mixing. If opportunity structures differ across contexts, then the age composition of co-offending groups may also vary systematically. Serious or instrumental crimes, such as robbery or burglary, may require coordination, transportation, or access that younger adolescents lack independently, thereby increasing the likelihood of adult involvement.

These theoretical considerations do not contradict the core insight of peer homophily. Peer processes undoubtedly remain central to adolescent behavior. However, they raise an empirical question that earlier data could not fully address. Is juvenile co-offending overwhelmingly peer-homogeneous, or does cross-age collaboration represent a substantial and patterned component of contemporary offending?

The National Incident-Based Reporting System (NIBRS) provides a unique opportunity to examine this question. Unlike earlier Uniform Crime Reporting (UCR) statistics, NIBRS links individual offenders to specific incidents, enabling precise identification of the age composition of groups. This allows researchers to move beyond assumptions about peer similarity and directly assess whether co-offending groups are predominantly horizontal or whether vertical age structures are common. If a substantial proportion of juvenile co-offending incidents involve at least one adult participant, the implications are noteworthy (Stolzenberg and D’Alessio 2016). First, such findings would suggest that age homophily in offending is less absolute than commonly assumed. Second, such a finding would indicate that theories of juvenile delinquency must accommodate both horizontal peer processes and vertical network integration. Third, variation in cross-age collaboration across offense types, locations, and situational contexts would imply that group age structure is contingent rather than uniform.

It is important to recognize that identifying cross-age collaboration does not invalidate peer theory (Warr 2002). Instead, it refines its scope conditions. Peer reinforcement may explain many forms of expressive or situational delinquency, particularly in school-centered contexts. However, instrumental crimes, weapon-involved offenses, or late-night street crimes may reflect more complex age structures. A comprehensive theoretical account of juvenile co-offending must therefore integrate peer homophily and reinforcement mechanisms (horizontal influence), network embeddedness and brokerage (structural integration across age strata), intergenerational transmission and facilitation (vertical role differentiation), and situational opportunity structures (contextual variation in group composition).

The present study evaluates whether juvenile delinquency conforms primarily to a horizontal peer model or whether cross-age collaboration constitutes a substantial and theoretically meaningful component by directly examining the age composition of co-offending groups using incident-level data. Rather than rejecting the foundational insights of peer theory, this study reassesses one of its core empirical premises under new measurement conditions and in a transformed social context. Should our findings show that juvenile co-offending frequently involves adults, particularly in serious or weapon-involved offenses, then delinquency may be embedded not solely in adolescent peer cultures but also in broader, vertically structured networks. Such findings would call for theoretical integration, expanding peer-based frameworks to incorporate age-stratified network dynamics and opportunity-based differentiation. Understanding the prevalence and patterning of these structures is essential for advancing criminological theory and for designing interventions that accurately reflect the social architecture of juvenile offending.

Theoretical issues concerning the response of the criminal justice system to delinquent youth have significant implications for how courts, prosecutors, and diversion officials assess and address youthful offenders. The assessment of whether to use punitive sanctions, e.g., direct filing into adult court vs. community-based diversion programs, is increasingly influenced by the debate over which sanction(s) are most appropriate. However, these assessments are generally made without systematic data on the age distribution of those in the group(s) the juvenile was part of at the time of offense. If there exists a substantial portion of serious juvenile co-offending that includes adult members (especially older adults who facilitate/coordinate/co-lead other youths), then the assumption underlying traditional juvenile justice approaches needs to be re-examined. Diversion programs that attempt to disrupt same-age peer networks may be ineffective for juveniles embedded in offending structures that span multiple generations of offenders. On the other hand, decision-making processes regarding the transfer of juveniles to adult court that focus solely on the juvenile offender’s role in the crime being assessed may fail to recognize the criminal agency of adult co-participants, which may be relevant to both the nature and seriousness of the crime. Therefore, determining the extent and manner in which young and old individuals co-offend is not just theoretically interesting. It provides an empirical basis for developing more informed and appropriately directed system responses.

Hypotheses

Peer Homophily Hypothesis: The peer homophily perspective posits that juvenile offending groups are overwhelmingly composed of peers of similar age. If this horizontal model accurately characterizes contemporary offending, most juvenile co-offending incidents should involve only juvenile participants, reflecting age similarity and reciprocal peer influence. Accordingly, we first expect that the modal pattern of juvenile co-offending will consist of age-homogeneous groups composed exclusively of juveniles.

H1. 

Most juvenile co-offending incidents will involve only juvenile offenders.

Cross-Age Presence Hypothesis: However, if cross-age collaboration represents a meaningful component of group offending, a non-trivial proportion of juvenile co-offending incidents should include at least one adult offender. The presence of such mixed-age groups would indicate that vertical age integration is not merely incidental but a structured feature of a substantial minority of incidents. This hypothesis does not contradict the peer homophily hypothesis. It evaluates whether cross-age collaboration is substantively noteworthy rather than exceptional.

H2. 

A non-trivial proportion of juvenile co-offending incidents will involve at least one adult offender because of meaningful vertical age integration within offending groups.

Age Gap Distribution Hypothesis: Beyond simple presence, the magnitude of age differences within mixed-age groups provides a more precise test of peer homophily. If cross-age co-offending primarily reflects transitional overlap, such as 17- and 18-year-olds offending together, then the core logic of peer similarity remains largely intact. By contrast, a vertical rather than horizontal organization would be suggested if mixed-age incidents frequently involve larger age gaps.

H3. 

In mixed-age incidents, the age disparity between the youngest and oldest offender will exceed near-peer differences, consistent with structurally differentiated group membership.

Instrumental Crime Hypothesis: Theoretical expectations also imply that the age composition of offending groups should vary across offense types. Horizontal peer reinforcement models are especially applicable to expressive or situational offenses. These offenses would include minor assaults or school-based altercations, where status competition and emotional contagion may predominate. Conversely, opportunity and network perspectives suggest that more instrumental and serious crimes, such as robbery, burglary, and weapon-involved offenses, may be more apt to involve cross-age collaboration. This situation is particularly relevant if adult offenders furnish access to resources, coordination, or criminal capital.

H4. 

Mixed-age co-offending is theorized to be disproportionately concentrated in instrumental and more serious offense categories.

Contextual Differentiation Hypothesis: Situational context further conditions these expectations. Age-homogeneous peer groups are likely to emerge in environments structured around adolescents, such as schools or daytime recreational settings. Mixed-age groups, by contrast, may be more common in street-based, commercial, or late-night contexts where opportunity structures are less age-segregated.

H5. 

Mixed-age co-offending is more prevalent in nighttime incidents than in daytime incidents.

Severity Amplification Hypothesis: Finally, if adult presence reflects facilitation or role differentiation rather than purely horizontal peer reinforcement, mixed-age incidents may differ in severity. Controlling for offense type and group size, incidents involving both juveniles and adults may be more likely to involve weapon use and victim injury than incidents composed exclusively of juveniles. Such a pattern would suggest that vertical age integration carries substantive behavioral consequences, potentially amplifying harm or enabling more complex forms of offending.

H6. 

Crime incidents involving mixed-age co-offending are more likely to involve weapon use and victim injury than juvenile-only co-offending incidents.

These six hypotheses evaluate whether juvenile co-offending is best characterized as a predominantly horizontal peer phenomenon or whether vertical age integration constitutes a systematic and theoretically meaningful dimension of contemporary group crime. Rather than rejecting peer-based theories, these hypotheses test their scope conditions by examining whether age homophily remains the dominant organizing principle of juvenile co-offending in incident-level data.

2. Research Methods

2.1. Data and Analytic Strategy

This study uses incident-level records from the FBI’s 2023 NIBRS. NIBRS is the incident-based component of the FBI’s UCR Program and provides detailed information on each reported crime incident. This information includes the characteristics of offenses, victims, and offenders, as well as the incident context. We obtained the 2023 NIBRS extract files distributed through NACJD/ICPSR, which are provided in multiple segment-level files that can be linked to construct incident-level analytic records. The analytic sample is restricted to co-offending crime incidents.

Our analyses proceed in two stages to evaluate both the structure and consequences of co-offending age composition. First, we examine how incident characteristics are associated with co-offending group type using multinomial logistic regression. All co-offending incidents are included in the analysis to provide a comprehensive assessment of the age structure of group offending across different offense types. Second, we assess whether age structure predicts offense severity (firearm use and victim physical injury) for violent crimes using two binary logistic regression models. Offenses lacking firearm use and victim injury are excluded from these two analyses. The three dependent variables analyzed in this study are theoretically noteworthy because they allow us to evaluate whether vertically integrated co-offending groups differ from juvenile-only and near-peer groups in their situational context and offense severity, net of crime type, group size, jurisdictional characteristics, and temporal variation.

2.2. Dependent Variables

The dependent variable used in the multinomial model is the age structure of the co-offending group. Co-offending incidents were classified into three mutually exclusive categories based on offender age composition: (1) Juvenile-only group includes offenders under the age of 18. This group serves as the reference category. (2) Near-peer mixed group encompasses at least one juvenile offender (under age 18) and at least one adult offender aged 18–20. (3) Vertically integrated group includes at least one juvenile offender (under age 18) and at least one adult offender (age 21 or older). These distinctions are designed to separate transitional age overlap from meaningful vertical age integration. Prior research on peer homophily emphasizes age similarity as a defining feature of delinquent groups, while network and opportunity perspectives suggest that larger age disparities may reflect hierarchical or facilitative structures. Differentiating near-peer from vertically integrated groups allows a direct empirical test of whether observed effects are attributable to simple adult presence or substantively greater age stratification. This variable is also used as an independent variable in the two equations examining the severity of co-offending.

Two dependent variables are used to measure the severity of the co-offending incident. The first dependent variable is firearm involvement in co-offending crime, coded 1 if any firearm was used in the incident and 0 otherwise. Prior research on group processes suggests that collective offending may increase risk-taking and escalation. Network and facilitation theories further imply that older offenders may increase access to weapons or contribute to more instrumental forms of crime. Firearm use in the crime thus indicates the offense’s seriousness and potential for escalation. This variable is also used as an independent variable in the multinomial logistic regression analysis of violent offenses.

The second variable used to measure the severity of the co-offending incident is victim injury. Victim injury is coded 1 if the incident resulted in physical injury and 0 otherwise. Victim injury is commonly used in criminological research as a stable indicator of offense severity and harm. If vertical age integration reflects differentiated roles or increased instrumental coordination, incidents involving such groups may be more likely to result in physical harm. Similar to the firearm use variable, this variable also serves as an independent variable in the multinomial logistic regression analysis of violent offenses.

2.3. Independent Variables

Group size is operationalized categorically as two offenders (reference category), three offenders, four offenders, and five or more offenders. Group process research shows that the dynamics of collective offending change as group size increases, potentially affecting escalation, diffusion of responsibility, and weapon use. Additionally, larger groups are mechanically more likely to contain age variation. Modeling group size categorically allows for non-linear effects and ensures that any association between vertical integration and severity is not attributable simply to larger group composition.

Offense-type indicators are included in the analyses to account for differences in crime composition. Major categories (e.g., homicide, robbery, aggravated assault, burglary, motor vehicle theft, sexual assault, drug offense) are coded using NIBRS offense classifications, with larceny/theft serving as the omitted reference category. Prior research indicates that offense seriousness, instrumentality, and situational dynamics vary substantially across crime types. Because certain offenses have an enhanced proclivity to involve weapons, injury, or coordinated planning, controlling for offense type ensures that observed differences in severity are not simply artifacts of crime composition.

Incident context is captured with location indicators (street/outdoor/transit, retail/commercial, school/daycare, and other/institutional), coded relative to residence/home as the reference category. Routine activity theory suggests that crime patterns vary by place (Cohen and Felson 1979; Felson 2002), and prior peer-based research indicates that adolescent offending is often concentrated in school or neighborhood settings. If near-peer or vertical integration groups are more common in street-based or commercial contexts, location provides a mechanism linking age structure to opportunity environments.

A binary indicator distinguishes incidents occurring during nighttime hours (1 = 6:00 p.m.–5:59 a.m.) from those occurring during daytime hours (0 = 6:00 a.m.–5:59 p.m.). Prior research shows that co-offending and serious violence are more common during evening and nighttime periods, when guardianship is reduced and opportunity structures shift (Cohen and Felson 1979). Including a temporal control helps isolate whether age mixing reflects broader patterns of routine activities rather than structural age effects.

Two jurisdictional variables were also included as controls. Urbanicity was measured with indicators for urban and suburban jurisdictions, with rural jurisdictions serving as the reference category. Because NIBRS classifies agencies by the population they serve, this measure reflects the reporting agency’s urbanicity rather than a direct measure of each incident’s immediate setting. Regional variation was captured with indicators for the Northeast, South, and West, with the North Central region serving as the reference category. These controls help account for structural and geographic differences in co-offending that may be correlated with both age composition and offense severity.

To account for time-invariant differences across reporting jurisdictions, including demographic composition, policing practices, reporting conventions, and local crime environments, models incorporated police department controls. These controls attenuate the risk that estimated associations reflect cross-jurisdictional differences rather than within-agency variation. Standard errors were also clustered at the police department level to address non-independence of incidents within jurisdictions and to provide more conservative inference.

Table 1. Descriptive Statistics for Co-offending Incidents (2023).

Variable	N	%
Age composition
Juveniles only (reference)	68,495	66.4
Juveniles + young adults	19,464	18.9
Juveniles + adults	15,122	14.7
Group size
2 offenders (reference)	62,051	60.2
3 offenders	24,981	24.2
4 offenders	9995	9.7
5+ offenders	6054	5.9
Offense type
Larceny/theft (reference)	39,580	38.4
Homicide	599	0.6
Robbery	11,603	11.3
Aggravated assault	11,312	11.0
Burglary	9308	9.0
Motor vehicle theft	11,762	11.4
Sexual assault	2429	2.4
Drug offense	16,488	16.0
Location
Residence/home (reference)	24,292	23.6
Street/outdoor/transit	22,627	22.0
Retail/commercial	41,079	39.9
School/daycare	10,577	10.3
Other/institutional	4506	4.4
Time of day
Daytime (reference)	58,680	56.9
Nighttime	44,401	43.1
Urbanization
Rural (reference)	13,364	13.0
Urban	50,089	48.6
Suburban	39,543	38.4
Region
North Central (reference)	26,959	26.2
Northeast	8787	8.5
South	47,681	46.3
West	19,629	19.0

Note: N = 103,081 co-offending incidents. Region is missing for 25 cases and urbanization for 85 cases; percentages are based on the full sample.

reports variable coding and descriptive statistics for all measures included in the analysis.

3. Results

3.1. Descriptive Results

Table 1 reports descriptive statistics for the analytic sample of 103,081 co-offender incidents drawn from the 2023 NIBRS. A central implication of the peer homophily argument is that juvenile delinquency tends to occur within networks of similarly aged peers. A visual examination of Table 1 shows that data drawn from NIBRS are broadly consistent with this expectation. Approximately two-thirds (66.4%) of co-offending incidents involve only juveniles, indicating that age-homogeneous peer groups remain the modal structure of juvenile co-offending. This pattern aligns with prior research showing that age similarity strongly structures social interaction and peer networks during adolescence, reinforcing the idea that delinquency often emerges within horizontally organized peer groups.

At the same time, the findings suggest that peer homophily does not fully characterize the age structure of juvenile offending. Roughly one-third of co-offending incidents involve adult participants, including 18.9% near-peer mixed groups and 14.7% vertically integrated groups involving offenders aged 21 or older. These figures indicate that cross-age collaboration is neither rare nor incidental. The presence of near-peer groups suggests a transitional overlap between older juveniles and emerging adults, while vertically integrated groups imply more pronounced age stratification that extends beyond typical adolescent peer networks.

These data suggest that the peer homophily perspective captures a salient but incomplete dimension of juvenile co-offending. Most incidents occur within age-homogeneous juvenile groups, consistent with peer influence models. However, it appears that juvenile offending may sometimes be embedded in broader age-stratified networks rather than exclusively within peer groups because a substantial amount of the minority of incidents involve cross-age collaboration. Such findings refine rather than contradict peer homophily theory since both horizontal peer processes and vertical age integration may mold the social organization of juvenile crime.

Co-offending groups tend to be relatively small. Incidents involving two offenders accounted for 60.2% of the cases, while 24.2% involved three offenders, 9.7% involved four offenders, and 5.9% involved five or more offenders. With respect to offense composition, larceny/theft accounted for the largest share of co-offending incidents (38.4%), followed by drug offenses (16.0%), motor vehicle theft (11.4%), robbery (11.3%), aggravated assault (11.0%), burglary (9.0%), sexual assault (2.4%), and homicide (0.6%). In terms of situational context, retail/commercial settings were the most common location (39.9%), followed by residence/home (23.6%), street/outdoor/transit (22.0%), school/daycare (10.3%), and other/institutional settings (4.4%). Finally, 56.9% of incidents occurred during daytime hours while 43.1% occurred at night.

These descriptive results provide support for H1, H2, and H3. Consistent with H1, juvenile-only groups constitute the modal pattern because they account for nearly two-thirds of all co-offending incidents. The data also support H2 because approximately one-third of co-offending incidents involve adult participants. This finding indicates that cross-age collaboration is neither rare nor incidental but rather a structured and substantively meaningful feature of juvenile group offending. Lastly, the descriptive data furnish support for H3. Among near-peer mixed incidents, the mean age difference between the youngest and oldest offender was 2.7 years (SD = 1.4). In contrast, within vertically integrated incidents, the difference was 17.2 years (SD = 11.4). These results show that age disparities in vertically integrated groups substantially exceed transitional near-peer overlap.

3.2. Multinomial Logistic Regression Analysis

To assess the situational and offense correlates of co-offending age composition, we estimate multinomial logistic regression models predicting group type. The dependent variable distinguishes among three mutually exclusive categories: juvenile-only groups (reference category), near-peer mixed groups, and vertically integrated groups. Multinomial logistic regression is appropriate when the outcome variable is nominal with more than two unordered categories (Long and Freese 2014). All models use cluster-robust standard errors clustered at the police department level. Given the large sample size, interpretation emphasizes effect magnitudes (odds ratios) and the stability of associations across contrasts. Because statistical significance is readily attained with a sample of this size, we treat odds ratios near 1.0 as substantively negligible even when they are statistically significant, and we base our conclusions on the magnitude and consistency of associations rather than on statistical significance alone. The results for this model, which identifies whether vertically integrated co-offending is systematically associated with certain offense types, locations, or temporal contexts, are reported in

Table 2. Multinomial Logistic Regression Predicting Co-offender Age Composition (Reference Outcome = Juveniles Only), Clustered on Police Department.

Predictor	Juveniles + Young Adults OR	p	Juveniles + Adults OR	p
3 offenders	1.44	<0.001	1.70	<0.001
4 offenders	1.33	<0.001	1.30	<0.001
5+ offenders	1.16	<0.001	1.18	<0.001
Homicide	3.06	<0.001	3.91	<0.001
Robbery	1.25	<0.001	0.96	0.546
Aggravated assault	1.79	<0.001	3.00	<0.001
Burglary	0.96	0.377	0.76	<0.001
Motor vehicle theft	0.77	<0.001	0.51	<0.001
Sexual assault	0.72	<0.001	1.63	<0.001
Drug offense	3.39	<0.001	3.39	<0.001
Street/outdoor/transit	1.00	0.898	0.58	<0.001
Retail/commercial	1.08	0.034	0.90	0.008
School/daycare	0.14	<0.001	0.04	<0.001
Other/institutional	0.78	<0.001	0.57	<0.001
Nighttime	1.27	<0.001	1.00	0.898
Urban	0.82	<0.001	0.90	0.036
Suburban	0.97	0.393	0.95	0.224
Northeast	0.94	0.427	1.02	0.848
South	1.20	<0.001	1.18	<0.001
West	1.00	0.942	1.01	0.879

Note: Odds ratios (OR) are reported from a multinomial logistic regression with juveniles only as the reference outcome. Standard errors are cluster-robust (sandwich) clustered at the police department level. Predictors are binary indicators (1 = present, 0 = absent).

.

Several predictors are associated with higher odds of near-peer mixing relative to juvenile-only incidents. The size of the co-offending group exhibited a consistent positive association. Compared with two-offender incidents, incidents with three, four, or five or more offenders were more apt to involve young adult participation. Offense type was also differentiated between near-peer and juvenile-only groups. Relative to larceny/theft, near-peer mixing was substantially more apt for drug offenses, homicide, and aggravated assault. While there was a smaller positive association with robbery, robbery did not differ from larceny/theft for vertically integrated groups. In contrast, motor vehicle theft and sexual assault were less likely than larceny/theft to involve near-peer mixing. Burglary did not differ from larceny/theft in this contrast.

Situational context further distinguished near-peer incidents. Compared to residence/home, incidents occurring at school/daycare locations and other/institutional settings were markedly less likely to include young adults. Street/outdoor/transit locations did not differ reliably from residence/home, whereas retail/commercial locations showed a small but statistically significant increase in the odds of near-peer mixing once clustering was accounted for. Finally, nighttime incidents were more likely to involve young adults than daytime incidents, consistent with the expectation that age mixing may be tied to opportunity structures that intensify in the evening.

A related but distinct pattern emerged for vertically integrated groups. Group size again increased the likelihood of adult involvement. Compared to two-offender incidents, three-offender incidents were associated with substantially higher odds of vertical integration. So were four-offender and five-or-more offender incidents. Offense-type associations were pronounced and generally stronger than those observed for near-peer mixing. Relative to larceny/theft, vertical integration was strongly associated with homicide, drug offenses, and aggravated assault. Sexual assault also showed elevated odds of adult involvement. By contrast, burglary and motor vehicle theft were less likely than larceny/theft to involve adult participation. Robbery was not reliably different from larceny/theft in this contrast.

Location effects were especially strong for vertical integration. Relative to residence/home, school/daycare incidents were extremely unlikely to involve adults. Adult involvement was also less likely in street/outdoor/transit settings, retail/commercial settings, and other/institutional locations. These findings partially support H5. Nighttime incidents are more apt to involve near-peer mixed groups, consistent with the expectation that age mixing intensifies in opportunity environments during evening hours. However, H5 receives no empirical support for vertically integrated groups, as nighttime was not associated with adult involvement after controlling for covariates. This distinction suggests that the temporal patterning of age mixing varies with the degree of age stratification, with young adult participation tied to nighttime opportunity structures while older adult involvement is driven more strongly by offense type and group size than by time of day. The jurisdictional controls show a similar contrast. Net the other predictors, co-offending in urban jurisdictions was less likely than in rural jurisdictions to involve either young adult or adult participation. Incidents in the South were somewhat more likely than those in the North Central region to involve mixed-age groups, while other regional differences were small. Adding these controls did not alter the offense type, group size, or the contextual associations reported above.

Across both contrasts, age mixing is more likely in incidents involving larger co-offending groups and in serious violence and drug offenses, while school/daycare contexts strongly predict juvenile-only co-offending. Importantly, the temporal pattern differs by type of age mixing: near-peer mixing is more common at night, whereas vertical integration is not. Several associations also vary in direction across contrasts (e.g., sexual assault is less likely to involve young adults but more likely to involve adults), underscoring the value of distinguishing transitional age overlap from more pronounced adult–juvenile age stratification.

The multinomial results suggest that while juvenile-only co-offending remains the dominant pattern, mixed-age collaboration is neither random nor rare. Instead, it appears systematically associated with group size, offense type, and situational context. These findings support the broader theoretical expectation that juvenile co-offending is structured by both peer homophily and cross-age network integration, setting the stage for the subsequent analyses examining whether vertically integrated groups are also associated with greater offense severity.

These findings furnish support for H4. Mixed-age co-offending is disproportionately concentrated in serious and instrumental offense categories. This is particularly relevant for homicide, aggravated assault, and drug offenses. Robbery is associated with higher odds of near-peer mixing, but not with vertically integrated co-offending. By contrast, motor vehicle theft is less likely to involve either type of mixed-age group. Additionally, while near-peer groups show little association with burglary, vertically integrated groups are less likely to be involved in burglary offenses. This offense-type patterning is consistent with the theoretical expectation that cross-age collaboration emerges selectively in contexts where older offenders may provide capacity that younger participants lack independently.

Table 3. Model Fit Statistics for Multinomial Logistic Regression (Clustered on Police Department).

Statistic	Value
N	102,996
Clusters (police departments)	7571
Log-likelihood (final)	−83,698.1
Log-likelihood (null)	−89,377.4
McFadden R2	0.064

Note: McFadden R² is computed as 1 − (LL_final/LL_null). The model is estimated on complete cases; 85 incidents with missing region or urbanization are excluded.

reports model fit statistics for the multinomial regression. An assessment of Table 3 suggests that while age composition is shaped by many unobserved contingencies, the variables included in the multinomial logistic regression explain a modest but nontrivial portion of the observed variation. The final model outperformed the null model, with the log-likelihood increasing from −89,377.4 to −83,698.1 and the McFadden R² at 0.064. Pseudo-R² values of this magnitude are common in incident-level models, and they indicate that a substantial share of the variation in co-offending age composition remains unexplained by the measured predictors. We therefore interpret the model as identifying systematic associations among the included variables rather than as offering a comprehensive account of why particular incidents involve cross-age co-offending.

To convey the practical significance of these associations, average predicted probabilities were computed from the multinomial model. For a larceny incident, the predicted probability that the offending group is mixed-age is approximately 29%, comprising 16% near-peer and 13% vertically integrated. This figure rises to roughly 56% for homicide, divided almost evenly between near-peer (28%) and vertically integrated (28%) configurations, and to a comparable 56% for drug offenses (31% near-peer and 24% vertically integrated). Offense type, therefore, substantially increases the likelihood of cross-age collaboration. Mixed-age offending is especially common in serious and instrumental crimes. By contrast, group size has comparatively little effect on group composition. These quantities indicate that the offense-type associations are not only statistically reliable but also large enough to be substantively meaningful.

3.3. Logistic Regression Models

In the second stage of the analysis, we shift the focus to violent co-offending incidents. The dichotomous dependent variables in these two logistic regression analyses are firearm use and victim physical injury. Logistic regression is appropriate for dichotomous outcomes and models the log-odds of the dependent variable as a function of explanatory variables (Hosmer et al. 2013). The key independent variables are indicators for near-peer mixed (juveniles + young adults) and vertically integrated groups (juveniles + older adults). The juvenile-only group serves as the reference category.

Table 4. Descriptive Statistics for Violent Co-offending Incidents (2023).

Variable	N	%
Firearm use
None/other weapon (reference)	15,131	60.2
Firearm used	9991	39.8
Victim injury
None (reference)	13,442	57.1
Injured	10,110	42.9
Age composition
Juveniles only (reference)	15,993	61.6
Juveniles + young adults	5061	19.5
Juveniles + adults	4889	18.8
Group size
2 offenders (reference)	13,872	53.5
3 offenders	6623	25.5
4 offenders	3096	11.9
5+ offenders	2352	9.1
Offense type
Homicide (reference)	599	2.3
Robbery	11,603	44.7
Aggravated assault	11,312	43.6
Sexual assault	2429	9.4
Location
Residence/home (reference)	7294	28.1
Street/outdoor/transit	9573	36.9
Retail/commercial	5732	22.1
School/daycare	2120	8.2
Other/institution	1224	4.7
Time of day
Daytime (reference)	13,980	53.9
Nighttime	11,963	46.1
Urbanization
Rural (reference)	2297	8.9
Urban	16,233	62.6
Suburban	7396	28.5
Region
North Central (reference)	6181	23.8
Northeast	2984	11.5
South	10,625	41.0
West	6146	23.7

Note: Total N = 25,943 violent co-offending incidents; clusters = 3818 police departments. Percentages for age composition, group size, offense type, location, time of day, urbanization, and region are based on the full sample (N = 25,943). Firearm use percentages are based on valid cases (N = 25,122; 821 missing). Victim injury percentages are based on valid cases (N = 23,552; 2391 missing). Region is missing for 7 cases and urbanization for 17 cases.

shows that within this violent crime subsample. Severe outcomes are relatively common in violent juvenile co-offending, as firearms were used in 39.8% of valid cases and victim injury occurred in 42.9%. More importantly, the age-composition distribution in Table 4 differs from that of the full sample reported in Table 1. Juvenile-only groups remain the dominant form of violent co-offending, accounting for 61.6% of the violent subsample. By comparison, juveniles + young adults account for 19.5% and juveniles + adults account for 18.8%. Although juvenile-only groups remain the modal pattern, mixed-age groups together constitute 38.4% of violent co-offending incidents. Relative to the full sample, this pattern suggests that cross-age integration is somewhat more common in violent co-offending, but it does not displace juvenile-only groups as the dominant group form.

Table 4 also shows that robbery (44.7%) and aggravated assault (43.6%) account for the overwhelming majority of violent co-offending incidents, while sexual assault (9.4%) and homicide (2.3%) comprise smaller shares. Violent incidents were more concentrated in street/outdoor/transit settings (36.9%) than in residences/homes (28.1%), and nearly half occurred at night (46.1%). These distributions are consistent with the view that violent co-offending unfolds in settings more conducive to confrontation and escalation in public settings.

Table 5. Logistic Regression Models Predicting Firearm Use and Victim Injury in Violent Co-offending Incidents (Clustered on Police Department).

Predictor	Firearm Use OR	p	Victim Injury OR	p
Firearm use	—	—	0.25	<0.001
Juveniles + young adults	1.89	<0.001	1.21	<0.001
Juveniles + adults	1.07	0.177	1.27	<0.001
3 offenders	1.07	0.097	1.16	<0.001
4 offenders	1.07	0.223	1.30	<0.001
5+ offenders	0.94	0.255	1.61	<0.001
Robbery	0.06	<0.001	0.04	<0.001
Aggravated assault	0.05	<0.001	0.11	<0.001
Sexual assault	0.001	<0.001	0.008	<0.001
Street/outdoor/transit	0.83	0.051	1.04	0.528
Retail/commercial	0.90	0.053	1.06	0.202
School/daycare	0.12	<0.001	1.34	<0.001
Other/institutional	0.47	<0.001	1.26	0.005
Nighttime	1.34	<0.001	1.10	0.009
Urban	2.03	<0.001	0.91	0.320
Suburban	1.50	<0.001	1.07	0.307
Northeast	0.47	0.136	1.69	0.025
South	1.95	0.062	1.27	0.182
West	0.92	0.826	1.52	0.032

Note: Odds ratios (OR) are reported. Standard errors are cluster-robust (sandwich) and clustered at the police department level. Both models are estimated on complete cases, excluding incidents missing region or urbanization. Firearm model: N = 25,105; clusters = 3705. Victim injury model: N = 23,096; clusters = 3593. The victim injury model includes firearm use as a covariate. Consequently, cases with missing data on firearm use are excluded from that model.

presents the two logistic regression models predicting firearm use and victim physical injury in violent co-offending incidents. These models address the expectation that mixed-age violent incidents are likely associated with greater offense severity. The firearm model provides partial support for this expectation. Relative to juvenile-only incidents, violent incidents involving juveniles and young adults have significantly higher odds of firearm use. However, incidents involving juveniles and older adults did not differ significantly from juvenile-only incidents in firearm use. These results suggest that near-peer mixing with young adults may be especially relevant to firearm involvement, perhaps because young adults occupy a transitional position with greater access to weapons than juveniles but still substantial overlap in peer networks.

The victim injury model provides stronger support for the mixed-age severity argument. Compared with juvenile-only incidents, both juveniles + young adults and juveniles + adults are associated with higher odds of victim injury. Mixed-age violent incidents are not simply different in demographic composition. They are also more harmful in their consequences. Such findings are consistent with the position that cross-age co-offending amplifies the seriousness of offenses through greater experience, coordination, escalation, or asymmetries in influence within the offending group.

These findings offer partial support for H6. The hypothesis that mixed-age co-offending is associated with greater offense severity is confirmed for victim injury, where both near-peer and vertically integrated groups show markedly higher odds than juvenile-only incidents. However, H6 receives only partial support for firearm use because elevated odds are only confined to near-peer mixed groups. The absence of a significant firearm effect for vertically integrated groups is noteworthy and suggests that the severity-amplifying consequences of cross-age collaboration may operate differently depending on the degree of age stratification and the type of outcome examined. Other findings presented in Table 5 reinforce this assertion. Group size is positively associated with the likelihood of injury to victims. Three-, four-, and five-or-more-offender incidents all showed higher odds of victim injury than two-offender incidents. The pattern for firearm use was markedly weaker, and once urbanicity and region were included in the model, group size was no longer reliably associated with firearm involvement.

Offense type was also related to outcome severity. Relative to homicide, robbery, aggravated assault, and sexual assault showed markedly lower odds of firearm use and injury. Since homicide is the reference category, these estimates indicate that homicide is uniquely concentrated in firearm involvement and injury. This is consistent with the logic of the severity models and confirms that the outcomes behave as expected across violent offense categories.

The results for the injury model also show that firearm use is associated with lower odds of victim physical injury. Although this may appear counterintuitive, it likely reflects the distinction between the presence of firearms and nonfatal injury in crime reporting. Research on weapon effects in violent encounters suggests that the presence of a firearm reduces the likelihood of victim resistance and, paradoxically, lowers the probability of physical injury in a crime incident because victims tend to comply rather than resist an offender’s demands (Kleck and DeLone 1993). The presence of a firearm appears to function as a coercive tool that facilitates offender control over the situation while simultaneously attenuating the probability of victim injury.

Location and time-of-day effects are more modest but still informative. Offenders are less apt to use a firearm in school/daycare and other/institutional settings than in residences/homes, with a comparable but only marginally significant pattern for street/outdoor/transit settings. By contrast, victim injury is more likely in school/daycare and other/institutional settings. Crime incidents occurring at night are more likely to involve firearm use and slightly more apt to involve injury. Jurisdiction type is also informative. Incidents in urban and suburban jurisdictions have substantially higher odds of firearm involvement than those in rural jurisdictions, whereas victim injury is somewhat more likely in the Northeast and West than in the North Central region. These results generally align with the expectation that situational context shapes both the composition and seriousness of co-offending, although the effects are more moderate than those for age composition and offense type.

Average predicted probabilities were computed from the firearm and victim-injury models to express these associations on the probability scale. The predicted probability of firearm use rises from approximately 37% in juvenile-only incidents to 50% when young adults are involved, while remaining essentially unchanged at 38% when older adults are present. The predicted probability of victim injury rises more modestly, from approximately 42% in juvenile-only incidents to 45% with young adults and 46% with older adults. These quantities reinforce the odds-ratio results. Near-peer mixing is most consequential for firearm involvement, whereas the injury difference, although consistent across both mixed-age configurations, is small in absolute terms.

3.4. Supplementary Analyses

Two supplementary analyses were conducted in response to issues about the measurement of age composition and the internal structure of mixed-age groups. The first analysis examined whether the three-category scheme obscures meaningful age heterogeneity among co-offenders. For each incident, an age span was computed as the difference between the oldest and youngest recorded offender. As Panel A of

Table 6. Supplementary Analyses of Age Span and Group Composition (2023).

Panel A. Age Span by Age-Composition Category
Age composition	N	M span (years)	% within 2 years
Juvenile only	68,495	0.9	91.1
Juveniles + young adults	19,464	2.7	51.5
Juveniles + adults	15,122	17.2	0.0
Panel B. Composition Among Mixed-Age Violent Incidents
Group composition	N	% firearm use	% victim injury
Juvenile-majority (2 juveniles, 1 older)	1224	49.7	44.6
Older-majority (1 juvenile, 2 older)	1505	52.5	45.5

Note: Panel A reports the age span (oldest minus youngest offender) for all co-offending incidents (N = 103,081). Panel B reports unadjusted outcome rates for three-offender mixed-age violent incidents (N = 2729), holding group size constant. Age span and group composition are computed from the recorded ages of up to three offenders per incident. Adjusted odds ratios are reported in the text.

shows, the categorical scheme closely corresponds to the actual age distance. Juvenile-only incidents have a mean span of less than one year, with 91.1% falling within a two-year window. In contrast, vertically integrated incidents always exceed two years and average more than 17. A minority of juvenile-only incidents (8.9%) nonetheless span more than two years, including 1006 cases pairing an offender aged 13 or younger with one aged 17. This illustrates the within-category heterogeneity that a coarse classification cannot capture. When the firearm and victim-injury models were re-estimated with the continuous age span added, the age-composition categories retained their direction and significance (firearm odds of 2.07 for near-peer and 2.29 for vertically integrated incidents, and injury odds of 1.23 and 1.46, respectively), whereas the age span itself showed only a small association (firearm OR = 0.95 per year, p < 0.001; injury OR = 0.99, p = 0.03). The use of a categorical variable thus suggests that our substantive conclusions are robust to a finer age-distance measure.

Another ancillary analysis assessed whether finer age distinctions within the juvenile category carry behavioral consequences. Among violent juvenile-only incidents, those in which all offenders fell within two years of one another were compared with those spanning more than two years. The two configurations differed little in severity. Firearms were used in 34.9% of close-age juvenile incidents and in 30.1% of wider-span juvenile incidents. Similarly, victim injury occurred in 41.0% of close-age juvenile incidents and in 44.2% of wider-span juvenile incidents. Thus, re-anchoring group composition to the youngest offender’s age does not alter the substantive conclusion that the presence of adult co-offenders, rather than fine age gradations among juveniles, is the feature associated with greater offense severity.

The second supplemental analysis examined the internal composition of mixed-age groups, distinguishing configurations in which juveniles predominate from those in which juveniles are a minority. Because older-majority configurations can only arise in larger groups, holding group size constant isolates composition from group size. Thus, the analysis focuses on three-offender incidents where the contrast is cleanest. Among the 34,586 mixed-age incidents, three-offender groups were divided into 4355 juvenile-majority configurations (two juveniles and one older participant) and 5043 older-majority configurations (one juvenile and two older participants). A visual examination of Panel B of Table 6 shows that within the corresponding three-offender mixed-age violent incidents, older-majority groups have only slightly higher unadjusted rates of firearm use (52.5% versus 49.7%) and victim injury (45.5% versus 44.6%) than juvenile-majority groups. Once offense type and situational context were controlled for, composition was not associated with severity. Relative to juvenile-majority groups, older-majority groups showed no reliable difference in the odds of firearm use (OR = 1.11, p = 0.23) or victim injury (OR = 1.07, p = 0.45). These results suggest that the presence of cross-age collaboration, rather than the relative proportion of juveniles within a group, is the operative feature for offense severity.

4. Discussion

The findings from this study have salient implications for life-course and criminal career perspectives that seek to explicate the relationship between age and offending. Traditional interpretations of the age–crime curve typically emphasize individual developmental processes. Criminal offending rises during adolescence, peaks in early adulthood, and then declines as individuals assume adult social roles (Hirschi and Gottfredson 1983). Life-course criminology and criminal career models explain this pattern largely in terms of changes in self-control, social bonds, and key transitions such as school completion, employment, military service, marriage, and parenthood (Sampson and Laub 1993). While highlighting important developmental mechanisms, these explanations typically view criminal offending as an individual behavior. However, a substantial portion of crime, especially among juveniles, occurs in groups.

Before elaborating on the theoretical implications of these findings, we believe it is useful to summarize the empirical support for each hypothesis. H1 and H2 are fully supported in this study. Juvenile-only groups remain the dominant pattern while cross-age collaboration constitutes a substantial minority of incidents. H4 is supported, as mixed-age co-offending is systematically concentrated in both serious and instrumental offense categories. H3 is supported. The mean age gap in near-peer incidents was 2.7 years, compared with 17.2 years in vertically integrated incidents. This finding confirms that cross-age collaboration in the latter category extends well beyond transitional age overlap. Lastly, H5 and H6 receive partial support because nighttime is associated with near-peer interactions, but not with vertical integration. Additionally, while the firearm effect is limited to near-peer groups, mixed-age groups show higher odds of victim injury.

Overall, our results suggest that the age–crime relationship is likely influenced by the age structure of offending networks. Consistent with life-course perspectives emphasizing peer influence during adolescence, roughly two-thirds of juvenile co-offending incidents involve only juveniles. Adolescence is a developmental period in which peer networks expand and parental supervision declines. This situation amplifies the likelihood that delinquent behavior occurs within same-age groups. In this sense, the prevalence of juvenile-only co-offending aligns with the perspective that adolescent crime is embedded within age-homogeneous peer networks.

Nevertheless, the presence of near-peer and vertically integrated groups complicates a purely developmental interpretation of the age–crime curve. A substantial share of juvenile offending occurs alongside older individuals. This finding suggests that some portion of the observed relationship between age and crime likely reflects exposure to criminal opportunities created by older offenders rather than age-specific propensities alone. Because younger participants in mixed-age groups likely have greater access to resources, knowledge, and criminal opportunities than those in juvenile-only groups, it seems probable that cross-age collaboration is associated with entry into certain types of offending (McCarthy et al. 1998). Our findings are consistent with the possibility that vertically integrated co-offending groups constitute one context through which criminal behavior is transmitted across age cohorts, although longitudinal data are needed to establish this directly.

These dynamics also intersect with criminal career models that emphasize onset, persistence, and desistance (National Research Council 1986). Participation in vertically integrated groups places juveniles in proximity to more experienced offenders because of the positive relationship between age and prior criminal record (Brame et al. 2012), a pattern that may be associated with greater persistence in crime. Conversely, near-peer groups involving individuals aged 18–20 may reflect transitional networks linking adolescent delinquency to early adult offending. These patterns suggest that the age–crime relationship may partly reflect changes in the composition of social networks across the life course, rather than solely individual developmental change. Incorporating the age composition of offending networks into life-course frameworks may therefore provide a more comprehensive explanation of how criminal behavior emerges, evolves, and eventually declines over time.

The findings generated in this study also align with broader opportunity-based perspectives that emphasize the salience of situational and structural contexts in shaping patterns of co-offending crime (Stolzenberg and D’Alessio 2008). If juveniles frequently offend alongside older individuals in certain contexts, then part of the age–crime relationship may reflect exposure to criminal opportunities created by older offenders, rather than simply developmental changes in an individual’s propensity to offend. The present results extend this perspective by demonstrating that opportunity structures likely shape the age composition of offending groups. The multinomial analyses show that cross-age collaboration is not randomly distributed across incidents but varies systematically with group size, offense type, and situational context. For example, school-based incidents overwhelmingly involve juvenile-only groups, reflecting the age-segregated nature of school environments (McPherson et al. 2001). In contrast, serious violence and drug offenses are more likely to involve adult participants. Such a finding is broadly consistent with the expectation that certain criminal activities may require access to resources, experience, or networks more commonly possessed by older offenders. However, it should be noted that drug co-offending may partly reflect market-based transactional relationships rather than the facilitative peer network mechanisms theorized here (Moeller 2018).

These patterns indicate that the age structure of co-offending groups may itself be conditioned by the environments in which crimes occur. In this sense, age mixing within offending groups can be understood as another dimension of criminal opportunity structures, reinforcing the broader argument that crime patterns are shaped not only by individual propensities but also by the social contexts that structure criminal opportunities. Rather than being randomly distributed, cross-age collaboration appears to emerge in settings where opportunities for interaction between younger and older offenders are more likely.

These results provide noteworthy implications for criminal justice responses to juvenile delinquency. Given the current controversy over whether jurisdictions should continue to divert juvenile offenders from the adult criminal justice system and/or consider transferring juveniles to adult courts for certain crimes, this study provides insight into the types of diversion programs needed. Most diversion programs currently being implemented are rooted in the idea that juvenile offending occurs through associations with other juveniles. Thus, the focus is primarily on intervening to counter peer influence and on reducing time spent with peers who engage in antisocial behavior. However, a high percentage of juvenile co-offending cases involve adult participants. In these instances, juvenile participation may be facilitated by their relationships with adults who possess greater criminal capital, have greater access to resources, and may be positioned to exert greater influence over their younger associates. Consequently, focusing solely on intervening against peer influences in cases involving juvenile-adult co-offending may be insufficient. Nevertheless, we fully recognize that many peer-oriented diversion programs emphasize prosocial skill building, mentorship, and the avoidance of antisocial associates more generally, and that these components may carry over to relationships with older co-offenders as well. Our argument is therefore not that such programming is irrelevant to mixed-age offending, but that its effectiveness in these cases may depend on supplementing it with explicit attention to the facilitating role of adult co-participants.

In addition, charging and direct file/transfer decisions regarding juveniles often rely heavily on assessments of the juvenile offender’s level of culpability and amenability to treatment. Although such assessments can assist in identifying which juvenile offenders require additional rehabilitative services, they may fail to determine the extent to which adult co-participants contributed to the offense’s initiation, planning, or escalation. The concentration of vertically integrated co-offending in precisely the categories most likely to trigger transfer consideration, homicide, aggravated assault, and drug offenses, suggests that adult co-offender involvement may not be incidental and could be structurally important to the offense. To the extent that adult co-participants occupy positions of greater experience, they may play an important role in facilitating the commission of these offenses, although the present data cannot directly establish this process. A more balanced approach to addressing serious group-based juvenile offending would take account of the composition of the offending group (i.e., whether it contains both juvenile and adult members) when making decisions about charges, diversion eligibility, and disposition. Furthermore, instead of viewing adult co-participants as secondary actors in these situations, prosecutors could develop focused strategies to target adult co-participants. Such strategies could include clearly communicating to adult co-participants the potential consequences of continuing to associate with juvenile offenders. These types of strategies may prove successful in lessening serious group-based violence while also protecting the long-term developmental interests of the juvenile participants involved.

Limitations

Several limitations of this study need to be acknowledged. First, while NIBRS provides detailed incident-level information, it captures only crimes reported to law enforcement. It is plausible that some incidents involving juvenile offenders may go unreported or may not result in the identification of all participants, which could affect estimates of co-offending age structure. Our findings may thus be influenced by reporting practices and enforcement patterns that vary across jurisdictions. Additionally, while the 2023 NIBRS represents the broadest agency coverage to date, participation remains voluntary and non-random. Larger urban agencies and certain states are disproportionately represented, while rural and smaller jurisdictions are underrepresented. Such a situation might potentially be problematic because prior research has shown that urbanization influences co-offending behavior (D’Alessio and Stolzenberg 2010). Thus, the generalizability of our findings to non-participating jurisdictions cannot be assumed.

Second, while the data allow us to identify the age composition of offending groups, NIBRS does not provide detailed information on potentially salient factors like an offender’s socioeconomic status and prior criminal record. Data are also lacking on social relationships among co-offenders. One social relationship that may underlie some cross-age co-offending is gang involvement because organized gang activity can connect younger and older offenders within hierarchically structured networks (Morselli 2009). NIBRS provides only limited information on this dimension. The data element capturing the type of criminal activity and gang information is recorded for roughly 30% of incidents and is concentrated in particular offense categories, and explicit gang designations identify fewer than 1% of co-offending incidents in the present sample. Because gang involvement is both rare and unevenly recorded across offenses, it cannot be incorporated as a general contextual control. That said, the cross-age associations observed here across the full range of offense types are unlikely to be attributable solely to gang structures. Examining how gang membership conditions cross-age co-offending nonetheless remains an important direction for research using data with more consistent gang measurement.

Third, the classification of mixed-age groups into near-peer (ages 18–20) and vertically integrated (age 21 or older) categories, while theoretically motivated, involves age cutoffs that are to some degree arbitrary. The 18–20 boundary reflects the legal transition to adulthood and aligns with developmental perspectives on emerging adulthood, while the 21+ threshold captures more pronounced age stratification. Nevertheless, alternative cutoffs, such as 18–24 versus 25 or older, might yield somewhat different results (Royston et al. 2006). The sensitivity of our findings to these boundaries was not examined here. A related issue is that all offenders under 18 are treated as a single juvenile category, which does not capture age heterogeneity within juvenile groups. A 13-year-old offending with a 17-year-old is classified in the same way as two 16-year-olds, even though the former pairing departs from strict age homophily. Future research should assess whether the patterns observed here are robust across alternative age classification schemes.

Fourth, because the analysis focuses on incident-level characteristics rather than on individuals’ longitudinal criminal trajectories, we cannot determine whether juveniles who participate in vertically integrated groups exhibit different long-term criminal career patterns than those who offend primarily with peers. Relatedly, while the cross-sectional, incident-level data analyzed here allow us to document associations between the age composition of offending groups and offense characteristics, they do not permit us to directly examine the interpersonal mechanisms, such as facilitation, recruitment, mentorship, coordination, or hierarchical control, that may underlie these associations. Where we invoke such processes, we intend them as plausible interpretations consistent with the observed patterns rather than as empirically demonstrated mechanisms. In addition, our three-category classification captures the presence of adults but not the relative proportion of juveniles and adults within a group. Future research might consider distinguishing between predominantly juvenile groups that include a single older participant and adult groups that contain only one or two juveniles. Such configurations may reflect different organizational dynamics, like adult leadership of younger associates versus the peripheral involvement of a few juveniles in adult-dominated offending.

5. Conclusions

The extent to which juvenile co-offending is organized around peer connections versus the degree to which cross-generational collaborations are integral to the crimes committed by a group of juveniles is a central issue in how our justice system assesses and responds to juvenile offenders. Utilizing incident-level NIBRS data, we demonstrated that both forms of juvenile co-offending exist today. While most juvenile co-offending incidents involve peer groups of similar age, thereby supporting the concept of peer homophily, a significant percentage of incidents involve cross-generational collaborations comprising both juveniles and adult offenders. Furthermore, these mixed-age offender groups are not randomly distributed across offense types. Specifically, cross-generation collaborations occur at higher rates in larger groupings than in smaller ones, and in more severe forms of offending. Additionally, those violent incidents that included both juveniles and adult offenders were more likely to result in victims sustaining greater levels of physical injury. Thus, these results indicate that the organizational structure of juvenile crime cannot be explained solely by peer-network frameworks.

Criminal justice responses need to account for the diversity of such structures. Juvenile justice programs intended to divert youth away from the formal court process through the disruption of their peer networks may be inadequate for that proportion of youthful offenders who are involved in vertical integrations of co-offending. Similarly, decision-making regarding the transfer or direct filing of charges against youthful offenders, which focuses solely on the culpability of the individual youthful offender, may overlook the potential role(s) that adult co-offenders potentially play in shaping the seriousness of juvenile co-offending groups’ crime. Therefore, a more responsive and developmental approach to the juvenile justice system would recognize the age composition of co-offending groups as one factor routinely used when determining charges, diversion opportunities, and disposition options for youthful offenders. The termination of juvenile-adult co-offending relationships, specifically in cases where youths commit serious violent crimes or are involved in serious drug-related crime, should also be a focus of discrete and high-priority interventions. Future research employing a combination of incident-level data collection techniques, network analysis, and longitudinal methodologies will further elucidate how cross-generational collaboration affects the life-course trajectories of youthful offenders and how interventions employed by the juvenile justice system can be optimally calibrated to address this form of collaborative behavior.