Introduction

Criminal psychologists have long been involved in researching the longitudinal patterning of criminal behavior over the life-course, namely how and why it begins, develops, continues, and ends. While most studies have focussed on early onset as a key factor in a long, serious and escalating criminal career, there have been very few investigations of the mechanisms involved in delayed criminal manifestations and in late criminal onset. Previous longitudinal findings (Farrington 1986, 1991) have showed that offending is not predominantly an adolescent phenomenon. Individuals who commit offenses do not all begin at the same age and with the same patterns (Farrington 1989b; Piquero et al. 2003). Variations in the ages of antisocial onsets are central to understanding the causes of criminality development (Caspi and Moffitt 1991; Lahey and Waldman 2003; Lahey et al. 1999; Loeber and Farrington 2001a, b; Moffitt 1993). A comprehensive model of criminal careers must not only describe differences in the ages of criminal onsets but also explain why some individuals continue, why others improve or worsen, and why others desist from or never begin a criminal career. For the study of adult criminal onset to provide insights over and above those gained from the study of early onset, researchers have to search for factors that may act as a buffer against offending and delay the beginning of antisocial manifestations.

The rationale for this study is based on the widely accepted idea that early interventions may reduce the risk of late criminality. Identifying and tackling those factors involved in the delayed manifestation of criminal behavior may prevent the emergence of adult criminal onset, and may lead to a better understanding of how and why certain early factors may exert a prosocial influence in childhood and adolescence, while in adulthood they are more likely to encourage antisocial behavior. The challenge of this work is to empirically demonstrate the link between past and future behavior on the basis of certain individual differences that are likely to persist over time, despite their phenotypical (behavioral) variation. Thus, the emphasis is on both the process of change in behavior over time, and continuity among diverse criminal onsets (Farrington 2005).

In the Cambridge Study (described below), a shy and timid temperament tended to prevent a boy from becoming delinquent, and this was one of the very few instances in which an apparent adverse feature was associated with an encouraging outcome (West and Farrington 1977). Even though researchers (Kagan and Snidman 1991; Kagan et al. 1993) recognize the influence that shyness, anxiety, inhibition, and neuroticism have on behavior, there is no agreement about the direction of effect. Anxiety, nervousness and inhibition seem sometimes to be associated with an increased risk of antisocial problems and sometimes with a decreased risk (Lahey and Waldman 2005). Anxious and nervous children (e.g. being worried for no reason, being shy, weary, withdrawn or socially isolated, avoiding situations that cause nervousness, tenseness or a sense of inadequacy) do not get involved with others in daring or risk-taking activities, and so anxiety and nervousness have proximal protective effects. However, when anxiety and inhibition reflect negative emotionality, they tend to be positively associated with conduct disorders (Lahey and Waldman 2005). Inhibited children are likely to have high right frontal lobe activation (Fox 1991, 1994), and may be less able to express negative affect and less able to modulate their affect in general (Kagan 1997). Perhaps because of these difficulties in emotion regulation, these children may be withdrawn, socially anxious, and have few, if any, friends. These characteristics, that in early life may have had a buffering effect against antisocial influences, might tend to lose their protective power as individuals age (Kerr et al. 1997). This leaves individuals socially unprepared, ill-equipped, risk-exposed, vulnerable, or inexperienced to deal with external, stressful and antisocial influences (Rutter 2003). The main aim of this article is to investigate late onset criminal careers, to see how early they can be predicted, and to examine which early psychological characteristics may play a significant role in delaying antisocial onset until adulthood.

What is a Late Criminal Career?

We define a late criminal career as a pattern of antisocial and/or criminal behavior whose official onset (i.e. age of first conviction) occurs only in adult life, at the age of 21 or later. The age of 21 is considered as a suitable age cut-off for establishing late onset, and it marks, not only the legal, but also the psychological and social scope of adulthood. To our knowledge, this study is the first to have used this threshold of late criminal onset in adulthood. In most studies, late onset is before age 21. It is also the first time that a prospective longitudinal study has been used to investigate true late criminal onset. Very few studies have devoted their attention to late or adult antisocial onset (Eggleston and Laub 2002; Gomez-Smith and Piquero 2005). Some seemed more interested in emphasizing the proximal and social influences on a late criminal career (Laub and Sampson 2003). Others investigated the role of family relationships and the level of closeness to parents as differential predictors of late criminal onset compared with adolescent or chronic delinquency (Mata and van Dulmen 2007). Other researchers (Klevens et al. 2000) started to investigate the possibility of early psychological and social influences on future behavior, when comparing early versus late (adolescent) antisocial onset. Others (Krohn et al. 2001, p. 69) argued that “antisocial onset is continuously distributed over childhood and adolescence”, and that causal influences on early onset may be more powerful than, but not necessarily different from, causal influences on late onset. The differentiation between early and late starters is associated with the intensity of structural, psychological, and social deficits experienced by individuals and their families. The dynamic perspective of Thornberry et al. (2003) especially focuses on the cumulative effect of antisocial behavior on the individual’s ability to make successful transitions through life, rather than attempting to explore the processes that differentiate early and late criminal onsets.

The speculation that a delayed criminal onset could explain certain antisocial patterns of criminal careers has not come without debate (Gomez-Smith and Piquero 2005). There are in the literature at least four theoretical controversies that proffered explanatory assumptions of the unlikelihood of adult-onset criminal behavior, and almost reduced it to an unimportant topic in psycho-criminological research. This current study is aimed to shed some light on the significance of bringing adult criminal onset into the study of delinquency development, and to contribute to a wider understanding of the psychological and behavioral heterogeneity underlying early and late criminal onsets. To address these tasks also requires arguing against the theoretical controversies that see adult criminal onset as negligible.

The “apparent late onset” controversy. It is argued that late starters are not really different from early starters in their antisocial potential. It is just that their antisocial potential never took the phenotypical form, earlier in life, of illegal behavior. Researchers might be dealing with apparent late onset, because adult onset criminality is preceded by juvenile maladjustment, antisocial conduct, and/or major mental illness, as in the study of Elander et al. (2000). It is suggested that late onset offenders were previously antisocial but not criminal. Hence, it is not necessary to postulate a true late onset group (Moffitt 2006).

The “bond to society” controversy. It is suggested that adult crime is likely to occur when bonds to society are weakened, and therefore that it cannot be predicted by early childhood factors. Under the umbrella of the age-graded theory (Sampson and Laub 1993, 2005a), informal social control in adulthood (e.g. employment, marital attachment) “explains changes in criminal behavior over the life span, independent of prior individual differences in criminal propensity” (Laub et al. 2006, p. 315). Thus, even if there is cumulative continuity (i.e. delinquency continuing from adolescence to adulthood because of its negative consequences for future life opportunities), it is suggested that childhood delinquency has only an indirect, weak effect on adult offending through the attenuation of social bonds (Laub et al. 2006).

The “late blooming” controversy. It is suggested that there exists a late blooming group (Thornberry and Krohn 2003). Late bloomers have an unusually late upswing, with offending “increasing away from a near-zero level only in mid to late adolescence” (Thornberry and Krohn 2005, p. 186). The idea is that early and late starters seem to share the feature of being both off-time in their criminal onsets (Elder 1995). However, the main difference from other offenders lies in their upbringing; late bloomers are, at early ages, cocooned by their family, and by their supportive living and school environment, which provide them with a quite smooth transition through childhood and scholastic difficulties. Despite all the early protection, when late bloomers enter the adult world, they are incapable of responding adequately to social demands and strains, and “their deficits in human capital increase the difficulty of making successful transitions to adult roles” (Thornberry and Krohn 2005, p. 200). It is then concluded that while childhood and adolescent factors have become, later in life, relatively weak and less influential upon the individual’s behavior, deficits in human capital result in a serious disadvantage for acquiring employment, establishing an intimate relationship, and compensating for the influence of antisocial friends (Thornberry 2005).

The “all offenders are alike” controversy. It is suggested that individuals with an official adult onset follow the delinquent behavioral pathway like other offenders, apart from the fact that their earlier criminality simply went undetected by authorities or was not recorded. Different from the “apparent late onset” controversy, the focus here is on the delinquent behavioral pattern rather than on the antisocial potential. Recent studies (Kazemian and Farrington 2005) have suggested that it is important to use methods that are likely to maximize the validity of both official and self-report offending measures: official figures offer precise information about criminal events (e.g. the date of occurrence), while self-report data provide more details of the nature of offending (e.g. frequency, co-offending, motives, escalation etc.). Official and self-report data compensate for the limitations of each other, and both are useful in studying the development of individual criminal careers (Huizinga and Elliott 1986; Farrington 1989a; Farrington et al. 2003).

These four controversies have sometimes challenged the rationale for studying late onset offenders. The empirical concern is that late onset offenders may overlap with earlier onsetters in most of their characteristics, except for the age of official onset; they may simply have been undetected or unrecorded. The methodological concern refers to the mismatch between official and self-report data, with official data indicating late onset but self-report data perhaps suggesting an earlier onset.

Hypotheses

This study attempts to pose counterarguments to these controversies. The aim is to examine to what extent early adult criminality can be predicted. The main hypothesis tested in this study is that late criminal onset is influenced by early factors that delay antisocial and criminal manifestations, and put the individual at risk for a subsequent delayed criminal career. Following on from this assumption, four other hypotheses can be tested.

  1. (1)

    Adult criminal behavior can be predicted from childhood. Hence, a real late criminal onset is plausible and not unlikely.

  2. (2)

    There are psychological and temperamental factors (e.g. anxiety, inhibition, nervousness, shyness, neuroticism, and social isolation) that play a role in delaying offending until adulthood. Thus, there are early predictors of late onset.

  3. (3)

    Late onset criminal behavior is associated with childhood factors that early in life act as protectors against delinquency.

  4. (4)

    Adult onset and early onset criminal behavior differ in the constellation of risk factors that are present in childhood and adolescence, making adult onsetters different from early offenders, and more behaviorally similar to nonoffenders, early in their life. Thus, individual differences represent a key factor in the patterning of criminal careers.

Method

This paper uses prospective longitudinal data (official and self-reported) on the development of antisocial and criminal behavior in the life course (from childhood to adulthood) from the Cambridge Study in Delinquent Development (CSDD). The CSDD surveys the development of 411 South London boys, mostly born in 1953. The Study began in 1961. Those in the sample were followed prospectively and assessed from ages 8–10 to 48–50. Nine face-to-face interviews have been completed with them over a 40-year period, and the attrition rate has been extraordinarily low for such a long-term study (Piquero et al. 2007). At age 32, 378 of the 403 men still alive (94%) were contacted (Farrington 2003), while the figure at age 48 was 365 out of 394 alive (93%) (Farrington et al. 2006). The majority of the boys in the sample (n = 399) were chosen by taking all the boys who were attending 6 state primary schools within a one-mile (1.6 km) radius of the research office established for the purpose of this investigation. Twelve other boys were drawn from a local school for the educationally subnormal with the aim of attempting to make the sample more representative of the male population of the area. Most of the boys were of British origin (n = 357, 87%), and were being brought up by parents who themselves grew up in England, Wales or Scotland. Of the remaining 54 boys, 12 had a parent of West Indian or African origin; the other 42 boys were White but of non-British origin. Almost all boys were living in traditional two-parent families with both a father and a mother figure present in the home (Farrington 2003).

Sample

The sample used in the present study was divided into different groups depending on their official offending (conviction) status. Early starters (ES) (n = 129) were defined as those individuals with a criminal onset before age 21. Late starters (LS) (n = 35) were those offenders whose criminal career began at the age of 21 or later; nonoffenders (NO) (n = 236) were those who had no criminal record. The average age of first conviction for an early criminal career was 15.75, while for a late career it was 30.73.

Offenses were defined as acts leading to convictions (Farrington 2003; Farrington et al. 2006). Convictions were counted if they were for ‘standard list’ more serious offenses. Most recorded crimes were of theft, burglary, violence, vandalism, fraud, or drug use; all motoring offenses were excluded. Up to 1994, when most of the males were aged 40, criminal record searches were carried out in the Central Criminal Record Office (National Identification Service) at Scotland Yard in London. From 1995 onward convictions were recorded on the Police National Computer (PNC) and therefore searches were based on this database. The latest search of criminal records occurred in December 2004 (Farrington et al. 2006). According to criminal records, of the 404 males at risk of having a conviction recorded, excluding seven participants who emigrated early and were not subsequently searched, 167 of them (41%) were convicted. Most official offenders (n = 83) were first convicted between ages 13 and 17, and the 35 late onset offenders were spread over a time period of 30 years (between ages 21 and 50), with only six first offenders after age 35 (Farrington et al. 2006; see also Piquero et al. 2007). The group of late starters in this study is composed of individuals whose official and self-reported delinquency records, measured at ages 12–14 and 16–18, were congruent. Official record and self-report data were examined to ensure that the late versus early antisocial onsetters were correctly identified. Because juvenile involvement in antisocial minor acts is quite common during the adolescent years (Emler and Reicher 1995; Moffitt 1993, 2006), it was decided to remove from the analysis those cases (n = 3) of late onset offenders with high self-reported delinquency scores (in the top quarter) both at ages 14 and 18. On the other hand, those cases who self-reported delinquent acts only at age 14 (n = 3) or at age 18 (n = 5), but did not have any criminal record before adulthood, were retained in the analysis. One further boy was excluded from the analysis because he died at age 23 and hence did not really have the opportunity to become a late starter. This left a total of 400 boys in the analysis.

Procedure

At the basis of this investigation is the assumption that adult offenders may not just be offenders who managed to avoid being apprehended earlier. If this were the case, there should not be any differences between late onset and early offenders.

The research procedure was organised in two stages. In the first stage, early offenders (convicted up to age 20), late starters (first convicted at age 21 or later), and nonoffenders were contrasted to analyze whether adult offenders were exposed in childhood and adolescence to factors that may have had a buffering effect against delinquent influences, so as to prevent an early antisocial onset. In the second stage, attention was given to exploring why adult offenders started a late criminal career, while nonoffenders continued to remain unconvicted and early offenders were convicted early in life, and to identifying differences and similarities among offending onsets.

This article analyzes a wide range of childhood, adolescent and early adulthood risk factors, including standardized measures derived from participants themselves and their parents, teachers, peers, psychologists, social workers, and psychiatrists. Boys were interviewed and assessed between ages 8 and 48. Parents were also interviewed about once a year from when the boys were 8 until when they were 15. They provided details on matters such as boy’s daring or nervousness, family income, family composition, their employment history, child-rearing pattern including discipline and supervision, their (temporary or permanent) separation from them, and their history of psychiatric treatment. Boys’ teachers completed questionnaires when the boys were about 8, 10, 12 and 14, and furnished details on topics such as their troublesome and aggressive school behavior, restlessness, concentration, truancy, school attainment and disruptive behavior in class. Ratings were also obtained from the boys’ peers about their daring, dishonesty, troublesomeness, and popularity.

Study Validity

Numerous tests of validity were carried out on the data collected between ages 8 and 32, in most cases based on a comparison between interview data and external information from records, and between different sources of ratings (e.g. mothers, teachers, peers). For example, self-reported convictions were compared with official criminal records. Reliability checks were also made. Information about a given topic (e.g. age of leaving school) from different interviews was compared, as was information about the same topic in different stages of the same interview. The measures, tests of validity and reliability, and major findings have been reported in five books (West 1969, 1982; West and Farrington 1973, 1977; Piquero et al. 2007), and many comprehensive publications (see Farrington 1995, 1997, 2003).

Explanatory and Outcome Variables

Late criminal onset is the outcome variable measured in this study. A range of 81 predictors were the explanatory variables (see Tables 1, 3–5), divided into four main time periods: childhood (ages 8–10), adolescence (ages 12–14), the teenage years (ages 16–18), and adulthood (age 32). For each period, predictors were organized into categories: psychological/individual, school, family, socio-economic, social, and behavioral.

Table 1 Childhood predictors of criminal onsets
Full size table

Analytic Strategy

A set of 24 key dichotomous variables assessed in childhood (ages 8–10), 19 key variables assessed in early adolescence (ages 12–14), 19 in the teenage years (ages 16–18), and 19 in adulthood (age 32) were examined in relation to early offenders, late starters, and nonoffenders. The rationale behind this procedure was to compare the importance of different variables, and also to use a risk factor approach, which helps target prediction and intervention efforts (Farrington 2006, 2007; Sherman et al. 2002). In order to explore the extent to which these predictors were associated with criminal onset at different stages of individual development, and might have acted as delayers of antisocial manifestations in adult onsetters, a two-step analysis was carried out.

First, the odds ratio (OR) was calculated to identify which factors significantly predicted later criminal careers. The OR provides information about the existence, direction, and strength of an association between target and comparison groups regarding the likelihood of an event occurring (Farrington and Loeber 2000). Where odds ratios are higher than 1, people with that particular attribute have relatively higher odds of offending than those who do not have this attribute. The use of the OR produces a more encouraging view about the prediction, explanation, and prevention of offending (Farrington and Loeber 2000; Farrington and Welsh 2007).

Second, the predictors, which in the first analysis were identified as potential explanatory factors in the etiology of adult onset offending, were included in a series of logistic regression analyses. The rationale behind logistic regression is to build multi-variable explanatory models that best account for observed variation in the outcome variable and identify which explanatory variables have independent influences on it. To identify which of the risk factors were significant independent predictors of criminal onset, the variables were entered into a forward stepwise logistic regression model, and only those predictors with a significant or near-significant weighting in the equation were retained.

Results

Overall this study tested 81 variables, and 59 of them were statistically significant in distinguishing late offenders from nonoffenders or early offenders. The percentage of statistically significant results (59 out of 162, or 36.4%) was very much greater than the chance expectation of 5%, suggesting that very few results were attributable to chance. We considered using the Bonferroni correction but decided not to in light of the convincing arguments of Perneger (1998).

Childhood Predictors

Early factors significantly predicted differences among criminal onsets. Table 1 shows childhood features of criminal onset for the late starter (LS), early starter (ES), and nonoffender (NO) groups. The percentages of each group possessing each risk factor are shown. Late starters were significantly different from other groups in some aspects of their childhood lives. They tended to be more nervous, more impulsive, lacking in concentration or restless, and were less likely to be troublesome or daring.

Psychological/Individual and Social

Late starters, in comparison with nonoffenders, were more impulsive or clumsy (OR = 2.1, CI = 1.0–4.5), and had poor concentration (OR = 2.2, CI = 1.0–5.2). The risk factors of being nervous or withdrawn, and having few friends (i.e. social isolation) were quite exceptional in being associated with an increased likelihood of late onset. Those boys who were rated as particularly nervous or isolated were likely to become late starters; 36.4% of late starters versus 25.2% of nonoffenders were rated as nervous (OR = 1.7, CI = .80–3.7). Despite the finding not being significant (on a two-tailed test), this predictor deserves further investigation. When LS offenders were compared with ES offenders (17.6%), being nervous was significantly associated with adult onset (OR = .38, CI = .16–.88), so that a delay in the onset of a criminal career was more likely to occur when the boy was nervous, withdrawn, or inhibited in childhood. Table 2 shows how the proportion of those who were nervous was distributed as the age of criminal onset increased. It can be seen that those males first convicted between ages 26 and 50 were the most nervous at ages 8–10 (see Table 2). When late starters were compared with early starters, having few friends also significantly predicted late criminality (OR = .29, CI = .09–.95).

Table 2 Distribution of nervousness (at ages 8–10) by ages of criminal onset
Full size table

School

Unsurprisingly, late starters had lower verbal (OR = 2.1, CI = 1.0–4.6) and non-verbal IQ (OR = 2.3, CI = 1.1–5.1) in comparison with nonoffenders. They also were more likely to have low junior school attainment (OR = 2.7, CI = 1.2–6.1).

Family

Compared with nonoffenders, late starters were significantly more physically neglected (OR = 3.4, CI = 1.2–9.6), lived in poor housing (neglected accommodation, poor decoration and interior, inadequate and old furniture, etc.) (OR = 3.1, CI = 1.5–6.4), came from large sized families (i.e. boys with five or more siblings) (OR = 2.7, CI = 1.2–5.9), and came from disrupted families (i.e. a permanent or temporary separation from a parent up to age 10) (OR = 2.5, CI = 1.1–5.7). These family conditions may have contributed to the boys’ sense of social inadequacy and anxiety, leaving them poorly equipped to cope competently with adult life pressures and difficulties.

Socio-economic and Behavioral

No significant differences were found between late starters and nonoffenders on low family income, poor child-rearing, or criminal parents. Also, late starters and nonoffenders did not differ in daring or risk-taking, troublesomeness, or difficulty of disciplining. Significant differences were found for all these risk factors between late and early starters (see Table 1).

Adolescent Predictors

Table 3 shows adolescent (ages 12–14) features of the onset groups.

Table 3 Adolescent predictors of criminal onsets
Full size table

Psychological/Individual and School

Late starters were significantly more likely to be highly anxious adolescents compared with nonoffenders (OR = 2.8, CI = 1.0–7.9). It was also more likely that late starters had a low verbal IQ (OR = 2.0, CI = 1.0–4.5), and left school earlier (OR = 2.1, CI = 1.0–4.4) than nonoffenders. Late starters and nonoffenders did not differ in their daring or mental concentration. In comparison with early offenders, late starters were less likely to manifest teacher-rated aggressiveness (OR = 3.5, CI = 1.5–8.1).

Social and Behavioral

Late starters and nonoffenders did not differ in the risk of having delinquent friends, in self-reported delinquency and self-reported violence, in lying, hostility to police, or stealing outside their home. Late starters differed from early offenders especially in antisocial features such as self-reported delinquency (OR = 8.4, CI = 2.4–28.9), self-reported violence (OR = 3.2, CI = 1.3–7.8), having antisocial friends (OR = 4.1, CI = 1.6–10.5), and stealing outside home (OR = 3.5, CI = 1.1–10.7). In all cases, late starters were less antisocial in adolescence (see Table 3).

Teenage Predictors

Table 4 shows the significant teenage (ages 16–18) features of the onset groups.

Table 4 Teenage predictors of criminal onsets
Full size table

Psychological/Individual

The most significant individual difference between late starters and nonoffenders was that the late starters had a high level of neuroticism; 40% of late starters were highly neurotic, in comparison with 21.5% of nonoffenders (OR = 2.4, CI = 1.2–5.1). Late starters were also more likely than nonoffenders to express anti-establishment attitudes (OR = 2.3, CI = 1.0–5.1) and to report an erratic job history (OR = 2.3, CI = 1.0–5.5). Interestingly, they did not differ in their level of aggressiveness.

Late starters were also significantly less likely to express aggressive attitudes (OR = 5.7, CI = 1.6–19.7) in comparison with early offenders. Also, never having had sexual intercourse (based on self-reported information) was related to late criminal onset (OR = .15, CI = .05–.39); 35.3% of late starters in comparison with 7.3% of early offenders had never had sexual intercourse, which may be related to their level of anxiety and neuroticism, which in turn might have amplified their difficulty in social relationships (see Table 4).

Socio-economic and Behavioral

Late starters were more likely than nonoffenders to report an erratic job history (OR = 2.3, CI = 1.0–5.5), but no differences were found in the level of unemployment. Early starters were more likely to report high unemployment and to have unskilled manual jobs than late starters. Late starters and nonoffenders did not differ in having debts, heavy drinking, heavy gambling, self-reported delinquency or self-reported violence. Late starters and early offenders were significantly different in all antisocial features such as self-reported violence (OR = 25.5, CI = 3.4–192.1), self-reported delinquency (OR = 6.0, CI = 2.2–16.5), and heavy gambling (OR = 6.3, CI = 1.8–21.8).

Adulthood Correlates

While in adulthood some of the differences between late starters and nonoffenders intensified, in most cases the differences between late and early offenders decreased, as shown in Table 5.

Table 5 Adulthood correlates of criminal onsets
Full size table

Psychological/Individual

A significant difference was found between late and early starters in relation to hospital treatment for illness (OR = 5.2, CI = 1.2–23.0). Early starters had more hospital treatment.

Socio-economic

Late starters endured a higher level of unemployment (OR = 2.3, CI = .85–6.2) and poorer home conditions (OR = 2.2, CI = .92–5.2) than nonoffenders. Late starters differed from early offenders in having short duration jobs (OR = 3.2, CI = 1.0–9.8).

Behavioral

Late starters reported higher levels of self-reported delinquency (OR = 4.3, CI = 1.9–9.8) in comparison with nonoffenders. In comparison with early starters, late starters were less likely to be involved in heavy drinking (OR = 2.4, CI = 1.1–5.3) or self-reported drug use (OR = 2.4, CI = .90–6.2); no differences were found in self-reported offending.

Life Success

Late starters were more unsuccessful than nonoffenders in many aspects of their lives; 40.6% of late starters were leading unsuccessful lives, in comparison with 12.2% of non-offenders (OR = 4.9, CI = 2.2–11.1). No differences were found between late and early starter offenders in this. The life success index was a combined measure based on 9 criteria: unsatisfactory accommodation; unsatisfactory cohabitation; unsuccessful with children; unsatisfactory employment history; involved in fights in the last 5 years; substance use in the last 5 years; self-reported offenses (other than theft from work or tax fraud) in the last 5 years; unsatisfactory mental health (scoring five or more on the General Health Questionnaire); convictions in the last 5 years (see Table 5).

Regression Analyses

Different logistic regressions were carried out for each age group, and then all the most significant predictors for each age group (8–18) were included together in a final comprehensive logistic regression analysis. Tables 6 and 7 show the variables that were selected in the model in order of their strength of prediction (i.e. their contribution to the predictive power of the model), the change in the likelihood ratio chi-squared (LRCS), and the partial odds ratio (OR) in the final model.

Table 6 Logistic regression analyses for variables predicting adult onset in late starters (1) versus non-offenders (0)
Full size table
Table 7 Logistic regression analyses for variables predicting adult onset in early starters (1) versus late starters (0)
Full size table

Predictors of Late Starters Versus Nonoffenders

The most important ages 8–10 predictors of late starters were poor housing and low nonverbal IQ. At age 12–14, high anxiety (rated by teachers) was the only independent predictor of late offending. The most important predictors at ages 16–18 were high neuroticism (measured by the Eysenck Personality Inventory), and an unstable job record. The independent predictors at ages 8–18 were poor housing, high neuroticism, and high anxiety (see Table 6).

Predictors of Late Starters Versus Early Starters

Late starters were significantly less likely to have criminal parents and poor child-rearing but significantly more likely to be nervous and to have few or no friends at ages 8–10. At ages 12–14, aggressiveness and high self-reported delinquency were the strongest independent predictors of early offending. The most important predictors at ages 16–18 were self-reported violence, heavy gambling, involvement in antisocial groups, and having high debts (see Table 7). Late starters were significantly less likely to have had sexual intercourse. In all, the significant predictors for late criminal onset measured at ages 8–18 were nervousness and no sexual intercourse, while the significant predictors for an early criminal onset were self-reported violence, antisocial group activity, gambling, and being aggressive.

Discussion

In line with the empirical hypotheses tested in this study, our results suggest that adult criminal behavior can be predicted from childhood; that early psychological and temperamental factors are likely to play a significant role in delaying offending until adulthood; that late onset offending is associated with childhood and adolescent factors that early in life perform a protective function against delinquency; and that, early in life, adult onsetters are more likely to be behaviorally similar to nonoffenders, and psychologically and behaviorally dissimilar from early offenders. It is clear that late onset offenders, in this study, were interestingly and markedly different from early offenders, especially in childhood, in their level of nervousness, neuroticism and anxiety.

Childhood Predictors

Late starters (36.4%) tended to be disproportionally nervous at age 8–10 when compared with nonoffenders (25.2%), and when compared with early starters (17.6%). They were also more socially isolated, with few or no friends. These findings suggest that early in their life late starters were more similar to nonoffenders in some aspects, psychologically and temperamentally, which may explain why in adolescence and the teenage years they were likely to be socially well behaved. They also suggest that the inhibitory impact of nervousness on early criminal onset is likely to change in adulthood when it is combined with other factors (e.g. unstable job records) that promote criminality.

Overall, our nervous boys resemble the inhibited children in Caspi and Silva’s work (1995) who were shy and fearful already from age 3 and had difficulty in concentrating on tasks in novel settings. While in adulthood these children were characterized by an over-controlled, restrained behavioral style, and a nonassertive personality, the nervous and anxious children in our study were likely in adulthood to begin a criminal career. This discrepancy between the findings of the two studies is just an apparent one, and in fact it could be explained by the fact that in Caspi and Silva’s study the analysis was carried out for a period of 15 years (from ages 3 to 18), while in this study the longitudinal span included data from ages 8–10 to 48–50.

Social isolation may be a quite effortless behavioral preference for a nervous and shy child aged 8–10, and for an anxious adolescent. Given that peer pressure is associated with adolescent delinquency (Tremblay et al. 2003), having few or no friends may constitute a kind of defensive shield. However, it may become more difficult later in life to avoid offending, especially when the individual faces the adult world outside the family protective cocoon. These factors seem to have only temporarily reduced the levels of social impairment in those nervous and anxious children who later became offenders. Possibly isolated, neurotic and anxious individuals began offending in response to adult social stress, uncertainties and adversities. This is not to say that anxious or nervous children are destined for a life of crime in adulthood, but we can gain new insight into how children, with certain psychological or temperamental characteristics, could find themselves in high-risk situations later in their lives (Moffitt and Caspi 2007).

Another aspect that deserves attention involves the extent to which children who came from poor housing were more likely, in comparison with nonoffenders, to become adult offenders. Previous research (Murray and Farrington 2005) has already demonstrated the impact of problematic family conditions upon the individual criminal career, and not surprisingly, having a criminal parent or coming from a poor child-rearing environment more than doubles the risk of beginning an early criminal career. These results clearly suggest that family conditions and parental criminality are not just indicators of family maladjustments, but especially bestow specific risk on the life of children. The significance of these factors as predictors of offending has been demonstrated in numerous studies (Farrington et al. 1988a, b; Farrington and West 1995; Fergusson et al. 2007; Gulotta 2002, 2005; Lahey et al. 2002; Lösel 2002; Murray and Farrington 2005; Piquero et al. 2007; Rutter et al. 1998).

Adolescent Predictors

At ages 12–14, in comparison with nonoffenders, late offenders were highly anxious, had a low verbal IQ, and left school prematurely, in most cases without taking or passing any exams. Previous studies (Thornberry et al. 2003; Tremblay et al. 2003) have confirmed a significant association between academic failure and delinquency. Anxiety seems to characterize the adolescent years of late starters, which may have prevented these individuals from getting too exposed to antisocial and deviant peer influences.

Teenage Predictors

At ages 16–18 late starters were characterized by high neuroticism, which may have played an inhibitory and over-controlling role. They may have been quite similar to the cautious, socially feeble, and submissive inhibited children at the age of 18, described by Caspi and Silva (1995). Late starters were more likely than early starters to have difficulties in establishing intimate relationships, reporting that they have not had any sexual intercourse. This may be an important reason why their antisocial onset was delayed until at least age 21, because the early protective influence of nervousness and anxiety may later have been overridden by more reactive and externalizing responses to life circumstances. This result is supported by previous empirical investigations. For example, Armour and Haynie (2007), employing data on 7,297 adolescents participating in three waves of the National Longitudinal Study of Adolescent Health, reached the conclusion that those adolescents who had an early sexual debut, relative to their peers, were at higher risk of delinquency, compared to those whose debut was “on time” with their peers. Adolescents who had the latest sexual debut were the least likely to commit delinquency.

Adult Correlates

With increasing age, the differences between onset groups seemed to take a new course. Late starters appeared to become more similar to early starters in their anti-establishment attitude, in their level of high unemployment, in the quality of home conditions, in self-reporting offending, and in the degree of life failure (Farrington 1989b, 1991). Unsurprisingly, all these variables have been shown in previous studies (Farrington et al. 1988a; Farrington and Maughan 1999; Loeber et al. 2006; Sampson and Laub 2005b; Wikström 2005) to be robust correlates of an adult criminal career.

Evidence-Based Responses to the Four Theoretical Controversies Concerning Late Criminal Onset

The results of this study have offered evidence to support a delayed patterning in criminal careers, the likelihood of a late criminal onset, and the significance of early factors in the prediction of adult criminal careers. Our reading of the research findings with respect to different criminal onsets (late onset, early onset and nonoffending) leads us to believe that the identification of a late criminal onset is necessary to criminal careers research, and policy and prevention strategies. What is clear is that evidence is mounting that late criminal onset exists, and that it can be predicted early in the life-course.

The “Real Late Onset” Explanation versus the “Apparent Late Onset” Explanation. While early starters are similar to Moffitt’s life-course persisters (1993, 2006) in the sense that they began in childhood and became worse thereafter, late starters, in this study, emerged as a distinctive group, who, after socially acceptable behavior before adulthood, found themselves inadequately equipped to cope with adult life demands and adversities. These offenders’ psychological and temperamental makeup (e.g. nervousness, inhibition, anxiety, and social isolation) seems to have steered their early social path away from disruptive and delinquent pressures. However, life requires individuals to resist the effects of environmental risk experiences and danger; robustness in the face of hazards may derive from controlled exposure to risk and life strains rather than from their avoidance (Rutter 2006). Is a late onset group really necessary? We think so, because the late starters were relatively well-behaved before age 21.

The “Early Predictors of Late Onset” Explanation versus the “Bond to Society” Explanation. Rather than exploring who is more likely to begin a criminal career, and explaining why, social bonding theory focuses on why most individuals follow a law-abiding life. According to this theory (Laub et al. 2006), delinquent activity can be predicted by focusing on weak social bonds and defiance of conventions. The findings gathered in this study showed that, above and beyond social factors, individual and psychological aspects could trigger a delayed antisocial onset, once their protective influence faded away. Even though, as social bonding theory suggested (Sampson and Laub 2005a, b), salient life events and socialization experiences in adulthood may neutralize, to some extent, the influence of early factors (Loeber and Farrington 2001a, b), an alternative explanation for their findings is that some psychological traits (e.g. nervousness, anxiety, neuroticism, etc.) and some childhood life experiences (e.g. social isolation and lack of friends) may have still exerted an impact during adult life in having perhaps left the individual ill-equipped to cope with new difficulties (e.g. no sexual intercourse, school failure, unstable jobs). Therefore, early predictors are important.

The “Early blooming of Late Criminality”Explanation versus the “Late Blooming” Explanation. Individual differences in antisocial proneness emerge early in life and are likely to be stable throughout the life course. In line with previous studies (Loeber 1990), childhood and adolescent features were robust and stable predictors of late criminality. Contrary to the late blooming theory (Thornberry and Krohn 2005), these findings suggest that factors in the early life of adult offenders were serious enough to affect their later adjustment, even though their consequences were, originally, not antisocial.

The “Individual Differences” Explanation versus the “All Offenders Are Alike” Explanation. There is unlikely to be a dramatic difference between early and late onset offenders. We would expect that variations in risk factors would reflect the continuum of criminal onsets. Nevertheless, late starters do differ, as these findings show, in their psychological traits and temperament, in their upbringing, and in the ways they react antisocially to life (Farrington 1997, 2007).

Limitations of the Study

This study is not without limitations. It is based on a sample of working-class males from South London. It could be particularly informative for future empirical work to investigate whether the patterns of late female criminal careers are influenced by similar psychological factors (Kratzer and Hodgins 1999; Zara 2007). It could also be interesting to examine whether and to what extent a late onset is likely to occur in middle or upper class environments. It would also seem important to understand the connection between late onset and residual career length, and to explore whether there is an association between adult onset and crime specialization. It would also be interesting to compare findings from longitudinal studies on the development of criminal careers carried out in different countries, and examine whether the predictors of late criminal onset found in this study were replicated in different cultures.

Implications for Intervention

Since most disturbing antisocial behaviors do not emerge suddenly, and without any warning, an effective intervention agenda should plan the early prevention or treatment of those early risk factors that are likely to encourage the individual into a personal and social maladjustment trajectory later in life. For a work scheme to be convincingly sound, the targeted risk factors must be amenable to change, and, as suggested in evidence-based research (Farrington and Welsh 2003; Sherman et al. 2002; Welsh and Farrington 2006), any change is more likely the earlier the intervention occurs. Given that diverse strongest predictors of adult criminality in this study can be addressed (e.g. nervousness), kept under control (e.g. anxiety), or modified (e.g. not having had sexual intercourse), they imply possible targets for successful intervention. Hence, there is enormous scope for significant cost savings, both economically and in the quality of life, from early intervention policies.

Conclusions

Nervousness, social isolation, anxiety, and neuroticism seem to protect against offending under age 21 but not after age 21. More research is needed to explore the interaction between these psychological and temperamental factors, and adverse social circumstances, in producing adult offending. Further research is also necessary to investigate why the protective effects disappear with increasing age, and to establish the precise causal processes linking early psychological determinants with adult criminal behavior. As Moffitt et al. (2002) put it, when speaking about post-pubertal delinquency onset, late onset offending certainly “warrants intervention, mainly to prevent future acquisitive crimes and, in particular, drug- and alcohol-related problems” (p. 202). And paraphrasing the key principles of intervention (Loeber and Farrington 1998), we believe that it is never too late for early intervention.