Criminal Careers in the Making? Characteristics and Trajectories of Juveniles Charged With a Sexual Offense

Sexual Offending as Part of a Criminal Career

As noted by Lussier and coworkers (2021) criminal career research focuses on a person’s behavior, that is, the sexual offending itself. Clinical research on sexual offending has mostly focused on individual (cognitive, emotional, relational) characteristics related to what is assumed to be a stable and fixed propensity to commit sexual crimes. Approaches that focus on deconstructing the behavior in question (sexual offending) into its constituent parts (Lussier & Cale, 2013) represent an important addition to the knowledge base. Criminal career research focuses on the longitudinal sequencing of offending patterns (cf. Blumstein & Cohen, 1987), and has provided a set of parameters (onset, frequency, continuity, diversity, desistance, etc.) to break down the development of offending over time into parts that can be subject to empirical investigation (see Piquero et al., 2003).

If adolescent and adult sex offending are distinct phenomena, we need to know more about what explain their occurrence at different developmental stages. This research seeks to identify essential differences between the presumably small group of people who continue a criminal career from an early onset sexual offense charge and the much larger group who do not continue a criminal career. How do they differ in terms of life circumstances known to influence the chances of criminal behavior over the life-span (Blokland & Nieuwbeerta, 2005)? As noted, clinical research investigating sex crime perpetrators are often based on what Blokland and Nieuwbeerta (2005) categorize as static theories, operating from an assumption of a fixed and more or less stable propensity for offending. In the case of sexual offending, the presence of sexual deviancy constitutes a central element in this propensity. Importantly, the criminal propensity (sexual deviancy) is assumed to influence the development in other life domains (such as education, work, and marriage), while influence the other way around is not expected. Dynamic theories, such as Sampson and Laub’s (1993) theory of informal social control, allow for influence from life circumstances on criminal behavior. Kruttschnitt and colleagues (2000) investigated the role of formal and informal social control in the discontinuity of crime among sex offenders with results indicating that job stability, although not marriage, influences desistance from sex offending. Typology theories represent a third set of life course theories of crime, in which the influence from life circumstances is assumed to be different among different “types” of offenders (most notably adolescent-limited and life course persistent (Moffitt, 1993)). Typology studies based on sex offending samples have provided insight into different aspects of the criminal career. As shown by Cooley’s (2022) qualitative investigation of the desistance process among persons convicted of contact sexual offending, the subsequent trajectories of those who do not reoffend can be further subdivided into a desisting and a non-reoffending pattern. The desisting pattern was characterised by cognitive transformations from an offending to a non-offending identity. The non-reoffending pattern involved behavioral changes, but neither the cognitive nor affective transformations assumed necessary to be considered a desister from sex offending.

A criminal career approach to sexual offending does not mandate a particular theoretical perspective or framework. Rather, studies investigating the progression of sexual offending over time and its antecedents may be informed by different theoretical perspectives, pertaining to the behavioural dimension in focus.

Age of Onset - When Does Sexual Offending Start?

Crime onset can be operationalized in different ways, with potentially large impact. When based on self-report, age of onset of sexual offending is on average seven years younger than what appears from officially registered onset (criminal justice or police data), a gap which is considerably more pronounced for people who committed sex offenses with child victims than those who offended against adults (Mathesius & Lussier, 2013). Also, when based on official data, age of onset differs between sex offender types; perpetrators with adult victims are generally charged for a first offense in their late twenties/early thirties, while those with child victims are usually in their thirties/forties (Lussier & Cale, 2013). Within the group of perpetrators with child victims, age of onset varies, as crimes with extra-familial victims are initiated earlier than crimes with intra-familial victims (Marshall et al., 1991; Smallbone & Wortley, 2004). Age of the child victim also seems to play a role; those offending against significantly younger victims are often younger at sexual offending onset than those with victims at their own age (Groth, 1977; Hendriks & Bijleveld, 2004). Currently, the literature suggests that there is a continuum of ages of onset of sexual offending which is broader than the 8–16 year age range often observed in the general criminological literature (Farrington, 2003), and which needs to be better identified and understood (Lussier et al., 2021).

Characteristics of Young Perpetrators of Sexual Offending

According to Seto and Lalumiere's (2010) review and meta-analysis of variables associated with generalist versus specialist theories of sexual offending, adolescent sexual offending is not sufficiently explained as a manifestation of general antisocial tendencies. Overall, general criminal risk factors such as criminal history, antisocial associates, and substance abuse were less often present in adolescent perpetrators of sexual offenses, while atypical sexual interests and childhood abuse experiences were more often present, when compared to adolescent perpetrators of other crimes. Results from adolescents in the Pathways to Desistance study indicated that adolescent perpetrators of sexual offenses were largely similar to other adolescent perpetrators on general criminogenic risk factors (Fanniff et al., 2017). When differentiating between early (12 years of age or less) and adolescent onset of sexual versus non-sexual offending, Rosa and colleagues (2020) concluded that early onset sex offending is characterised by a specific kind of risk, in which sexual abuse plays a central role, along with familial adversity (including parental imprisonment) and neuropsychological adversity (such as ADHD, irritability). In a prospective study looking at childhood risk factors for onset of sex offending, Lussier and coworkers (2015) investigated a set of socieconomic indicators such as source of income, parental education, ethnicity, and neighbourhood characteristics. Their findings indicate that living in a poor neighbourhood, exposure to socioeconomic adversity, behavioural problems, and more extensive juvenile delinquency are among the risk enhancing factors. Accumulation of risk across domains seems to be an important aspect of early onset sex offending.

Persistence and Diversity – How do Criminal Careers Progress from an Early Onset?

It is a well-supported finding from the general criminal career literature that early-onset offending is associated with a more persistent, frequent and diverse pattern (e.g., Piquero et al., 2003). In the sexual offending field, this observation has transformed into an inaccurate expectation that today’s young (adolescent) perpetrators are tomorrow’s adult perpetrators of sexual violence (Lussier et al., 2021). This has been partly backed by findings from early studies among adult offenders indicating that adolescent-onset of sexual offending was common (e.g., Abel et al., 1987), while later evidence indicate that this is more an exception than a rule (Marshall et al., 1991; Smallbone & Wortley, 2004), although findings are not consistent across studies (Långström, 2002; Miner, 2002). As discussed by Lussier and colleagues (2021), the idea of continuity in sexual offending from adolescence into adulthood is a strongly held assumption which has influenced policies in the sexual offending field from the 1990s up to the present. In a recent meta-analysis of 158 studies of adolescent perpetrators of sexual offenses, Lussier and coworkers (2024) warned against applying adult strategies for prevention among adolescent perpetrators, as recidivism into sexual crimes was unlikely (weighted pooled mean = .08). Further, the study emphasised the more general criminogenic needs present in the group of adolescent perpetrators of sex offenses, evident in their considerably higher rates of recidivism into general crime (weighted pooled mean = .44).

Although several studies have challenged the assumption of high transition rates between adolescent and adult sexual crime (see Nisbet et al., 2004), the idea lingers on. Part of the reason behind this is the observation that adult antisocial behavior is most often preceded by a history of childhood antisocial behavior, and, as sexual offending may be operationalized as a form of antisocial behavior (Seto & Barbaree, 1997), the continuity expectation in sex offending follows from this. However, it has also been observed that most antisocial children do not go on to become antisocial adults, so continuity and discontinuity exists in parallel, a fact which is often referred to as Robins’ continuity paradox (Robins, 1978). Lussier and Blokland (2014) investigated the validity of Robins’ continuity paradox in sex offending using data from the Dutch 1984 birth cohort. Their results did not support the existence of continuity in sexual offending from adolescence through adulthood but rather indicated that juvenile and adult sexual crime perpetration are two distinct phenomena. In their sample, only 4.5% of those with an adult sexual offending record had a sexual crime related history from adolescence, and only 3% of those with one juvenile sex offense went on to adult sex offending. The results of Chouinard-Thivierge and coworkers (2022) further attest to the lack of continuity in adolescent sex offending. Adolescents who had been adjudicated for a sex offense as juveniles (12–17 years of age), were less frequently involved in offending during early adulthood, both when measured by official data and by self-report. Although the idea of persistence in sex offending (coined by the phrase “once a sex offender, always a sex offender”) has been repeatedly empirically refuted (Harris & Hanson, 2004; Jennings et al., 2015; Zimring et al., 2007, 2009) it lingers on in spite of the fact that continuity in sex offending seem to be “(..) unusual, atypical, and rare (..)” (Lussier et al., 2021, p. 96).

What are the Risk and Protective Factors Influencing Continuity/Discontinuity in Sex Offending?

From studies of general offending trajectories, the role of both formal and informal social control, such as having a job and a non-criminally involved partner emerge as well documented desistance-supporting factors (Humphrey et al., 2024; Zedaker & Bouffard, 2017). In terms of risk enhancing factors, the systematic review and meta-analysis conducted by Yohros (2023) supports the role of adverse childhood experiences on youth (general) reoffending. A similar impact from childhood adversity on continuity in sexual offending has not been observed (Zgoba et al., 2023).

In this study, we focus on the onset of sexual offending by investigating the social and criminal trajectories of juveniles indicted of a sexual crime before the age of criminal responsibility (age 15). From that point of departure, we investigate offending patterns according to persistence and diversity, focusing on the dynamic influence of life circumstances on these career parameters. The current paper has three aims. First, we investigate whether juveniles charged with any sex offense in the period 1997–2005 differ from juveniles charged with either violent offenses or other crimes (non-sexual and non-violent combined) on a set of social indicators. Secondly, we track the sample’s criminal records for a period of 22 years, exploring what kinds of offenses are committed by those within the crime type groups defined above who have a subsequent criminal record, and thirdly, whether any of the social background indicators can distinguish between recidivists and non-recidivists.

Setting

The age of criminal responsibility varies markedly between countries, as does the range of sanctions available for the youngest perpetrators. In Norway, perpetrators younger than age 15 can be charged, but the case is dropped from further legal prosecution due to lack of criminal responsibility. When closing the case due to lack of criminal responsibility, the police can decide to transfer the case to the child protection services (according to §71b in the Criminal Procedure Act). The child protection services then decide if and what kind of further measures are needed. The criminal charge and the reason for closing the case (perpetrator not criminally responsible) remain in the police registers with details on the charged perpetrator. Information related to if and how other service providers follow-up on the case is not included in the police register.

Study Sample

The sample used for analyses is established from the population registers and the police registers retrieved from Statistics Norway. The sample used in the current study consists of the total population birth cohorts aged 15 during the period 1997–2005 (N = 388,814), of which 19,559 had criminal charge before their 15^th birthday. Since relatively many juveniles commit crimes at some point during adolescence, the risk of recidivism should be compared with the risk of offending in this age group, thus we include also those with no offences before age 15 (n = 369,255). Those with a recorded crime before age 15, are sorted into the following groups:

(a) Any sex offense covered by regulations in Chapter 26 of the Norwegian penal code. This includes offenses in three broadly defined categories: unwanted sexual behaviours (involving exposure to pornography, indecent exposure), unwanted sexual acts (such as being touched, being forced to masturbate someone) and unwanted penetration/intercourse (including rape).

In the case of a person having committed multiple types of crimes, these categories are applied hierarchically so that those classified as non-sexual violent offenders have not been charged with any sexual offences, and other offenders have no sexual or violent offences.

Measures

Family income (in total) – defined as household income from work, measured in Norwegian currency at the juvenile’s age 15. Adjusted for consumption units according to the EU scale. ¹

Family type – defined as households with children, containing either one or two adults (parents).

Place of residence – defined as juvenile living in Oslo (the capital city) at age 15 or not.

Parents’ educational attainment – defined as the highest attained level of education achieved by either parent at the juvenile’s age 16.

Immigrant background - defined according to the definition used in the population register, as having immigrated themselves or being the descendant of two immigrants (parents born outside of Norway).

Parental criminal involvement – defined as any criminal charge raised against the mother/father when the juvenile is 10–15 years old.

Recidivism - defined as a new, recorded offence, committed after the initial charge. As we are interested in types of offence committed when recidivating, our outcome measure is re-arrest, as this includes information on all offences, not only the primary/index offence. The police data include dates for both when the crime was solved, reported and committed. In all analyses, we use the date when the crime was committed. Thus, the time to recidivism is less affected by e.g., investigation time, only if the crime is solved within the observation window.

Analyses

Recidivism analyses often focus on whether an offender re-offends during a given period, but this can be problematic when there is varying observation time such as in our data. Thus, we use survival analysis to take varying exposure time into account. We use Kaplan-Meier plots to show the baseline risk of recidivism as a function of time (months), conditional upon still being at risk. We use proportional hazard models (Cox regression) to take other characteristics into account. In Cox regression the outcome is the risk of an event at time t, conditional on being at risk at that time, which is denoted the hazard rate. Thus, the regression coefficients are comparison of hazard rates for groups, and the exponent of the regression coefficients, exp (β) are interpretable as hazard ratios (HR).

One disadvantage of survival analysis is that its main purpose is for analysis of time to a single event or competing risk. In our context, the types of recidivism are not mutually exclusive for several reasons. One can commit multiple types of crimes at the same occasion, and one might commit multiple repeated crimes of varying types. For this reason, we estimate separate models with time to selected recidivism types. We first run Cox regression models where the outcome is time to a new crime of any type. We then proceed to repeat the analysis for the outcome of any sexual crime, while ignoring any other types of crimes. Finally, we repeat the analyses with the outcome of any non-sexual violent crime, ignoring any other types of crimes.

For each outcome, we present two Cox regression models. The first only includes the characteristics of main interest: criminal history before age 15. The second model includes several basic control variables: sex, type of family household, parental educational level, immigrant background, resident in major city, and whether at least one parent has a criminal record. These controls capture basic social background characteristics which are correlated with early onset of crime. The estimates for the variables capturing criminal history before age 15 are thus likely reduced when taking these into account.

While it is common to measure recidivism as a single event in this way, criminal career research has for a long time emphasized the need to distinguish between persistent offenders and more time-limited patterns of offending. These sequences of repeatable offending are sometimes summarised using latent class growth curves (Nagin & Land, 1993), which cluster individuals with similar offending patterns into a specified number of groups and estimates their average trajectory as a nonlinear function. One of the disadvantages of that method is that it is primarily suited for studying one kind of outcomes at a time. In our case, it is more relevant to study the sequence of alternating types of crimes, in particular sexual, violent and non-violent crimes. To do so, we use sequence analysis, which has gained popularity in life course research (Liao et al., 2022) and have a similar goal of summarising trajectories into a manageable number of groups using nonparametric methods.

A sequence are events distributed over discrete time intervals, such as years. For one individual, the sequence might consist of sexual crime, violent crime or other types of crimes, with periods of no crime in between. With ten time periods, a sequence for an individual might i.e., look something like {N-N-O-N-V-O-N-N-S-N}, where each position represents either if no crime (N) occurred in that time interval, or if violent crime (V), sexual crime (S), or other crimes (O) occurred. A non-recidivist would accordingly only have N’s at all periods.

The individuals’ sequences are then reduced into clusters of people with similar offending patterns over time. We use the hierarchical agglomerative clustering algorithm with Ward’s aggregation criterion (Müllner, 2013). Each individual is then classified into the cluster he/she resembles the most. Sequences with lack of variation gives little substantive sense and leads to computational problems. Thus, the clustering of sequences is only done on those who have recidivated more than once. The final groups for analysis contain the non-recidivist and single recidivists in addition to the clustered sequences. We then further explore the risk factors of criminal careers using these groups as the outcome variable in multinomial logistic regression models, where the exp (β) is interpretable as an odds ratio (OR).

Sequence analysis is a data driven exploratory approach, but there are choices to be made by the researchers that are likely to affect the results, but there are no clear-cut rules for these choices. First, one has to decide on how the individual sequences are constructed. We use yearly periods because finer time intervals of relatively rare events yield sequences which are mainly non-offending and gives little variation to analyse. Since persons might commit multiple crimes in the same year, we construct the outcome similarly as the initial categories of criminal history before age 15: having committed (1) any sexual crime, (2) any non-sexual violent crime, (3) any non-violent crime, or (4) any other crime excluding sexual or violent crimes. Thus, a period with sexual crime does not rule out that other crime types have been committed as well, but any period of “other crimes” do not contain violence or sexual violence. Sequence analysis is mainly meaningful for the part of the sample where there is variation over time. Additional technical details are provided in the Appendix.

Finally, we use the clusters from sequence analysis as outcome variable in a logistic regression analysis, using the same predictors as in the survival analysis.

Data has been provided from Statistics Norway, reference no. 21/3982 in accordance with Norwegian regulations. Data Protection Impact Assessment (DPIA) has been created in agreement with the Data Protection Official at the University of Oslo and filed with Norwegian centre for research data, project no 436491. The authors take responsibility for the integrity of the data, the accuracy of the data analyses, and have made every effort to avoid inflating statistically significant results.