III. Methodological Implications of the Invariance Argument
Steffensmeier and colleagues (Steffensmeier, Allan, Harer, & Streifel, 1989) noted disagreement about the strength and consistency of the relationship between age and crime. One of the main methodological points of argument stems from Hirschi and Gottfredson’s (1983) assertion that the age–crime curve is invariant across time, place, individual characteristics, offense type, and so on. Many criminologists have addressed this argument and contend that the claim of invariance is overstated. In summarizing the debate, Tittle and Grasmick (1998) found evidence of invariance only when considering the general mathematical form of the curve; in other words, the aggregate age– crime curve looks similar across different places, times, types of individuals, and offense types. All curves share the same general unimodal pattern of rising rates to a peak in late adolescence and declining rates through adulthood. However, Tittle and Grasmick (1998) commented that the claims of parametric and individualistic invariance have been conclusively rejected.
In terms of parametric invariance, studies examining the specific properties of the age–crime curve (i.e., median, mean, skewness, and kurtosis) have found variations (see Farrington, 1986, and Tittle & Grasmick, 1998). For example, using both the Uniform Crime Reports and self-report data, Steffensmeier and colleagues (1989) found that whereas the general shape of the aggregate curves are similar, specific parameters of the curve do vary. They noted in particular that, over time, the peak of the age–crime curve has shifted to younger ages and the curve has become steeper. In his study of the specific components of the curve, Farrington (1986) also concluded that the age–crime curve is not invariant. Although Gottfredson and Hirschi (1986) dismissed these variations as unimportant, others find them to be substantively interesting and worth attention (Blumstein, Cohen, & Farrington, 1988).
The claim of invariance has also been rejected when the age–crime curve is considered at the individual level. Tittle and Grasmick (1998) noted many individual deviations from the modal age–crime curve. Blumstein and colleagues (1988) argued that because the aggregate age– crime curve is capturing prevalence (e.g., the proportion of the population of a given age that engages in crime), it only appears to be invariant. When one looks at the relationship between age and crime at the individual level, however, one can see a great deal of variation. Farrington (1986) also cited individual variation in offending trajectories, which have been replicated in more recent research (Nagin & Land, 1993).
This highlights an important methodological debate in terms of whether the relationship between age and crime is due to prevalence (i.e., participation rates) or incidence (i.e., individual patterns of the frequency of offending). Differences in prevalence or participation rates of offending imply that the general shape of the age–crime curve appears because involvement in offending varies by age group. In other words, a larger proportion of the adolescent population participates in offending, while the proportion of the population involved in offending declines for older age groups. A consideration of incidence or frequency of offending, on the other hand, suggests that the age–crime curve is an aggregate representation of individual differences in the number of crimes committed at various ages. In other words, individuals commit a larger number of offenses during their adolescent years and reduce the frequency of their offending as they age. It is now generally agreed that the relationship apparent in the aggregate age– crime curve is due to prevalence (Farrington, 1986; Moffitt, 1993). Beginning with the work of Nagin and Land (1993), studies have continued to demonstrate variation in trajectories of the frequency of offending. Hirschi and Gottfredson (1983) also agreed, and they further argued that differences in prevalence (i.e., distinguishing accurately between offenders and nonoffenders) are the most important consideration in criminology.
A. Age, Crime, and Criminal Careers
Other methodological debates surrounding the relationship between age and crime largely stem from the issue of the invariance of the age–crime curve. The claim of individualistic invariance has been rejected, and some researchers, recognizing that there are individual variations in the age–crime relationship, have considered offending in the context of a criminal career. Blumstein et al. (1988) referred to the criminal career as “the longitudinal sequence of offenses committed by an offender who has a detectable rate of offending during some period” (p. 2). The criminal careers perspective looks at the relationship between age and crime at the individual level and addresses such components of a career as onset, persistence, and desistance. Onset refers to the initiation of criminal behavior, and some researchers have focused on age of onset as an important element of the criminal career. In particular, research has examined whether individuals who initiate their offending early in life are more likely to become long-term or high-rate offenders. Persistence refers to the continuation or duration of an offending career, and desistance refers to the termination of that career. Although Blumstein et al. argued that there is no reason to expect any particular pattern or tendency within criminal careers, they suggested that the inquiry as to the presence of certain patterns (e.g., escalation in seriousness, specialization in particular types of crime, etc.) is open to empirical investigation.
The criminal careers perspective does not present any particular theoretical model; instead, it is a methodological and empirical strategy that separately considers participation in offending from the frequency of offending among active offenders and allows for the possibility that various theoretical perspectives may be important in explaining different components of the criminal career. Gottfredson and Hirschi (1986), on the other hand, contended that the appropriate comparison for criminology is between offenders and nonoffenders (i.e., prevalence or participation) and that theories that are capable of distinguishing these two groups are adequate without needing to explain different components of a career. For Gottfredson and Hirschi, questions of the incidence or frequency of offending are irrelevant to the understanding of criminal behavior, and the causes of crime are the same regardless of age or criminal career component.
B. Longitudinal Research in Criminology
The age–crime curve and the criminal career perspective both imply long-term processes at work. For a variety of reasons, longitudinal studies, which involve repeated measurement over time, have emerged as preferable to cross-sectional studies (i.e., measurement at one point in time) in the study of criminal careers. Blumstein and colleagues (1988) argued that understanding dynamic patterns of offending, whether one is looking for variation or stability, virtually requires longitudinal data, and they pointed to the inadequacies of cross-sectional data in studying criminal careers. Steffensmeier et al. (1989) likewise argued that there are many social factors that vary by age and that may provide an explanation for the shape of the age–crime curve. Greenberg (1985) pointed out that the impact of dynamic factors will be underestimated in cross-sectional analyses depending on the stability or instability of the variable over time. The distinction between longitudinal and cross-sectional data may be most important when causal ordering is unclear. For example, whereas social control theories propose that weakened social bonds will lead to criminal behavior, it is also plausible that involvement in offending may weaken social bonds (Greenberg, 1985). By measuring social factors and criminal behavior at various points in time, longitudinal studies are better able to ensure the appropriate temporal ordering necessary to demonstrate causation instead of just correlation.
Another important consideration in choosing between cross-sectional and longitudinal research was raised by Farrington (1986), who argued that cross-sectional research easily confuses period, cohort, and age effects. Period effects refers to the impact of living in a particular historical period. Regardless of age, individuals who live through particular periods (e.g., World War II) may experience the same events or social conditions. Cohort effects may be more easily confused with age effects in that individuals in the same cohort (e.g., born in the same year) may be exposed to similar life experiences (e.g., the Baby Boomer generation). In contrast, the term aging effects refers to those social conditions that may vary by age (e.g., maturational reform, changes in peer networks or social bonds, etc.) and would affect individuals regardless of cohort or period. Farrington argued for the necessity of using multicohort longitudinal studies to truly distinguish these period, cohort, and age effects.
Gottfredson and Hirschi (1986, 1987), however, have taken issue with many of these arguments. Hirschi and Gottfredson (1983) argued that if the age–crime curve is the same for everyone, then no special techniques are necessary to understand this relationship. They argued that good cross-sectional studies (e.g., true experiments) are capable of answering the same questions as longitudinal designs, especially considering that the timing of crime and social events is not ambiguous. Because temporal ordering should not be a major problem when studying criminal behavior, according to this perspective one major argument in favor of longitudinal designs is rejected. They also discounted the need to disentangle age, period, and cohort effects, because “crime cannot cause age, period, or cohort” (Gottfredson & Hirschi, 1987, p. 588), and they argued that the attention paid to these issues has distracted criminology from more substantive, policy-relevant concerns.
Gottfredson and Hirschi (1987) also contended that the correlates of crime uncovered by longitudinal research are the same as those reported by cross-sectional research, concluding that longitudinal studies have merely confirmed results from cross-sectional studies. They cautioned that longitudinal studies are far more expensive, inefficient, and time consuming than cross-sectional studies, providing no added value to the study of crime. Other researchers have pointed to additional difficulties of longitudinal research, including the possible confusion of testing and maturation or aging effects and the high levels of attrition (i.e., dropping out) of high-rate chronic offenders from longitudinal studies over time (Brame & Piquero, 2003).
This debate is far from settled. Gottfredson and Hirschi (1987) contended that the invariance of the age–crime curve means that nothing of value has been learned from longitudinal studies and that it is more important to distinguish offenders from nonoffenders, regardless of age. On the other hand, researchers who are interested in the incidence of offending (i.e., frequency) and criminal careers argue that the curve varies a great deal at the individual level and requires longitudinal data to truly understand the patterns. Sampson and Laub (1995), for example, suggested that although differences between offenders are important, differences within individuals over time are just as important to understand. Scholars, especially those in the criminal careers or developmental/life course traditions, are increasingly turning to individual-level, longitudinal designs (Brame & Piquero, 2003). This research continues to find evidence of varying criminal career patterns and to explore whether different social factors may account for these different patterns (Nagin & Land, 1993; Steffensmeier et al., 1989).
IV. Theoretical Implications of the Age–Crime Relationship
Debate over the age–crime curve also has significant implications for criminological theory. Hirschi and Gottfredson (1983) claimed that the age–crime curve is invariant, that the causes of crime are the same at all ages, and that no existing social theory is capable of explaining the curve. Traditional criminological theories, such as differential association and social control, have tended to focus on explaining crime during the adolescent period, which represents the peak of the age–crime curve. Although this is to be expected, given that the bulk of delinquent and criminal activity occurs during these ages, Greenberg (1985) argued that crime does not just level off following the transition to adulthood; instead, it consistently declines, which suggests the need for theoretical attention to the entire life span and to the decline and desistance from offending in addition to onset.
Hirschi and Gottfredson (1983) noted that traditional theories have often been judged by their ability to explain the patterns apparent in the age–crime curve. For example, theories are criticized as being able to explain the onset of criminality, leading to the peak of offending, but not desistance. The failure to explain all aspects of the age–crime curve is often taken as a fatal flaw for theories. Hirschi and Gottfredson argued, however, that a theory that adequately distinguishes offenders from nonoffenders at a particular age (e.g., adolescence) may not necessarily account for the aging-out effect. Because aging out and desistance from crime occur consistently for all groups, the failure to explain desistance should not be used to discount a theory, especially considering that no existing theory, in their opinion, is capable of providing an adequate explanation.
Hirschi and Gottfredson (1983) also argued that, in the absence of a sufficient theoretical explanation, the remaining conclusion is that age has a direct effect on crime independent of other social factors and incapable of being explained by any existing social theories. This would seem to imply some sort of biological explanation, and they referred to a process of maturational reform, which occurs pervasively for all offenders, as an explanation of desistance. Farrington (1986) also suggested that maturational reform reflects some biological forces, noting age-related variation in physical strength and skills. Again, according to Gottfredson and Hirschi (1986), because the age–crime curve is invariant, and because aging out of crime occurs similarly for everyone, attempts to explain these patterns with social forces, which are assumed to vary, is futile. These authors ultimately concluded that the correlation between various social factors and crime is spurious, calling into question all existing criminological theories.