Skip Navigation


American Journal of Epidemiology Advance Access originally published online on June 14, 2006
American Journal of Epidemiology 2006 164(2):151-160; doi:10.1093/aje/kwj172
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
164/2/151    most recent
kwj172v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (1)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Fabio, A.
Right arrow Articles by Farrington, D. P.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Fabio, A.
Right arrow Articles by Farrington, D. P.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

American Journal of Epidemiology Copyright © 2006 by the Johns Hopkins Bloomberg School of Public Health All rights reserved; printed in U.S.A.

Original Contribution

Why Some Generations Are More Violent than Others: Assessment of Age, Period, and Cohort Effects

Anthony Fabio1, Rolf Loeber2, G. K. Balasubramani3, Jeffrey Roth4, Wenjiang Fu5 and David P. Farrington6

1 Department of Neurosurgery, Center for Injury Research and Control, School of Medicine, University of Pittsburgh, Pittsburgh, PA
2 Department of Psychiatry, School of Medicine, University of Pittsburgh, Pittsburgh, PA
3 Epidemiology Data Center, Graduate School of Public Health, University of Pittsburgh, Pittsburgh, PA
4 Jerry Lee Center of Criminology, University of Pennsylvania, Philadelphia, PA
5 Department of Epidemiology, Michigan State University, East Lansing, MI
6 Institute of Criminology, Cambridge University, Cambridge, United Kingdom

Correspondence to Dr. Anthony Fabio, Center for Injury Research and Control, Suite B-400, University of Pittsburgh Medical Center–Presbyterian, 200 Lothrop Street, Pittsburgh, PA 15213 (e-mail: fabioa{at}upmc.edu).

Received for publication September 19, 2005. Accepted for publication January 31, 2006.


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
Empirical longitudinal studies assessing why community-level violence rates change over time are lacking. Despite a wide-ranging literature, questions remain as to whether changes over time are due to factors occurring in specific periods (period effects) or individuals in successive cohorts (cohort effect). The objective was to assess the relative contribution of age, period, and cohort effects on violence trends. The authors assessed differences in self-reported violence between two cohorts of males (n = 1,009) from the Pittsburgh Youth Study, which tracked delinquency and risk factors from 1987 to 2000. The youngest cohort were aged 7–19 years, and the oldest cohort were aged 13–25 years. Yearly measures of violence were examined through generalized estimating equations. The oldest cohort reported higher levels of violence even after adjustment for age and major individual-level risk factors (odds ratio (OR) = 1.45, 95% confidence interval (CI): 1.17, 1.81) such as gang participation and drug dealing, as well as community-level factors (OR = 2.16, 95% CI: 1.65, 2.82). However, when period effects were included, cohort differences were rendered insignificant (OR = 1.23, 95% CI: 0.78, 1.94). The authors conclude that differences in the rates of violence over time may be attributed to changing social factors (period effects) and not to differences between the individuals (cohort effect) of cohorts.

cohort effect; crime; prospective studies; social change; social environment; violence

Abbreviations: CI, confidence interval; OR, odds ratio


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
Rates of violent crime in the United States have increased over the years, with a series of peaks or "epidemics" in different periods. The latest epidemic occurred during the early 1990s when the United States experienced dramatic increases in rates of violence. Although advances have been made in the explanation of age-related changes of an individual's risk of violence (1Go, 2Go), the empirical explanation of why violence rates vary over cohort and time period has lagged behind. A review of the literature demonstrates a staggering variety of theories based on contrasting perspectives (3Go). Of importance is the fact that these theories shift between period explanations (crack-cocaine (4Go, 5Go), gangs (6Go, 7Go), subculture of violence (8Go), and guns (5Go, 9Go)) and cohort explanations (abortion legalization (10Go–12Go), cohort size (13Go–17Go), and lead levels (18Go, 19Go)) with no conclusive explanation. Further, few theories have been rigorously tested, leaving little empirical evidence for prevention programs. Work has been limited in four major ways: 1) concentrating on a single explanatory factor, 2) examining cross-sectional but not longitudinal data, 3) not assessing individual-level data, and 4) ignoring the complexity of the contextual processes (not accounting for contextual relations leaves open the possibility of ecologic (20Go) or individualistic (21Go) fallacies). Major questions continue to exist about the relative effects of period and cohort on violence trends despite the attention paid to violence in research, public policy, and the press.

The contrasting and broad literature illustrates the need for identification of fundamental factors related to violence trends. Three fundamental factors have emerged: age, time (or period), and cohort. These are said to be elemental for understanding the temporal trends of health in general and of violence trends in particular (22Go). Age effects are specific to an age and consist of biologic processes related to aging, such as puberty or social processes, for example, the beginning of schooling. Changes of violence in relation to age have been described by Farrington (23Go) and others (24Go, 25Go) as the age-crime curve. Although specifics regarding the shape of the age-crime curve are not fully understood, this general shape has been shown to be consistent (26Go). Period effects pertain to factors transpiring during a specific period that affect all cohorts. An important aspect is that they are temporary. Once the social factor recedes or disappears, the risk also recedes or disappears. An example is the economy. During times of a weak economy, legitimate employment opportunities may decline, increasing the risk that an individual will participate in illegitimate employment opportunities such as drug dealing. This has been reported in detail by Blumstein and Wallman (4Go) and Blumstein et al. (5Go) to explain the crack-cocaine drug market. Cohort effects refer to phenomena particular to a specific birth cohort that, unlike period effects, remain with the cohort for life. In many respects, these can be considered as changes in the demographic makeup of a cohort. Abortion legalization is an example of a cohort effect. Donohue and Levitt (12Go) and Levitt and Dubner (27Go) theorize that legalized abortion leads to fewer "unwanted" babies, and unwanted babies are at an increased risk for criminality. Therefore, a cohort born after abortion legalization would have a lower percentage of high-risk individuals, which would lead to lower rates of violence. This decreased risk would persist through the cohort's lifespan, as the demographic makeup of the cohort would not be altered.

The purpose of this analysis was to investigate the fundamental factors (age, period, and cohort) in the variation of violence over time by assessing cohort differences within a longitudinal 14-year multiple-cohort study in Pittsburgh, Pennsylvania. We simultaneously assessed individual- and community-level factors longitudinally. The time of the study (1987–2000) occurred when community levels of violence dramatically increased and decreased in the nation and Pittsburgh. Importantly, the oldest cohort reported significantly higher rates of violence than did the youngest cohort at the same age, and the age-crime curve of the cohorts intersected with peak community violence at different ages. The immediate question that we addressed in this analysis is whether the different levels of violence between the cohorts can be explained by changes in community-level risk factors over time (period effect) or by differences between individuals in the cohorts (cohort effect).


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 References
 
Participants—Pittsburgh Youth Study
Participants were from the Pittsburgh Youth Study, which documented the development and risk factors of antisocial and delinquent behavior from childhood to early adulthood. Boys attending the first, fourth, and seventh grades in the Pittsburgh public school system (referred to as the youngest, middle, and oldest cohorts) were randomly selected. Eighty-five percent of the boys consented to participate in screening. From each cohort, the top 30 percent of boys with the highest rates of antisocial behavior were selected, along with an equal number randomly selected from the remaining 70 percent. (In the Pittsburgh Youth Study, data collection on the middle sample was discontinued after 3 years and is therefore not used here.) The randomization procedure resulted in a total sample of 1,009 boys (503 from the youngest cohort and 506 from the oldest). Each sample was followed up initially every 6 months and later every year for 14 yearly assessments (1987–2000). The youngest and oldest cohort members' median ages were 7–19 and 13–25 years. Although the Pittsburgh Youth Study oversampled high-risk boys, the data presented here were weighted to represent those of the public school system. A more detailed description of the Pittsburgh Youth Study can be found in the book by Loeber et al. (28Go).

Measures
The individual-level data came from self-reports from the youth supplemented by reports from parents and teachers. The validity of the measures used has been demonstrated and documented in detail elsewhere (28Go). Whether or not the participant reported violence during the year served as the dependent variable. Violence consisted of a positive response to gang fighting, strong-arming, attacking someone with a weapon or an intent to seriously hurt or kill, and rape or forced sex. This classification is based on the General Delinquency Seriousness Classification (28Go, 29Go). Information was acquired from the primary caretaker by use of the Child Behavior Checklist (30Go–32Go) and from the youth by use of the Self-Reported Delinquency Scale and the Youth Self-Report (33Go). These forms have been validated and used extensively to measure violence (34Go). The cohort effect was examined in the analyses as a dichotomous variable—oldest versus youngest cohort. The cohort could be defined by birth year or by grade in school. We chose to define our cohort by grade for several reasons. First, the samples of the Pittsburgh Youth Study were recruited and defined by grade. We felt that following the original methodology of the Pittsburgh Youth Study was important. Second, we felt that the socialization process of being in the same grade was more important than birth year.

To test for period effects, we grouped time into four categories: 1987–1990, 1991–1993, 1994–1997, and 1998–2000 (coded as 1, 2, 3, and 4, respectively). Our approach for this categorization was based on statistical and theoretical reasoning. Statistically, it was necessary to collapse the period measure of year into fewer groupings than yearly to help deal with the identifiability problem. Theoretically, our groupings were based on our hypothesis—understanding whether the differences reported between the cohorts could be explained by period effects. To accomplish this, we utilized current understanding of what would be a high and a low risk to operationalize our period effect. These cutoffs were chosen to correspond to the rise and fall of national and local violence trends.

Covariates were derived from the Pittsburgh Youth Study, the 1990 and 2000 US censuses, and the Uniform Crime Reports. Variables in the analysis are listed in table 1. These variables include individual-level (Pittsburgh Youth Study) and community-level (census and Uniform Crime Reports) data. The family's socioeconomic status was measured by the Hollingshead index of social status (35Go). Because of the shape of the age-crime curve, we included the quadratic form of age in the models. Following the crack-cocaine theory of Blumstein and Rosenfeld (36Go), we examined yearly measures of gun carrying, hard drug use, drug dealing, and gang participation. Drug dealing (37Go–40Go) and gang membership have been shown to increase the risk of violence (41Go). Hard drug use included hallucinogens, cocaine, crack, heroin, phencyclidine (PCP), and the nonmedical use of tranquilizers, barbiturates, codeine, amphetamines, and over-the-counter medications. Drug dealing was defined as 50 or more sales per year. Gang membership was collected by self-report using the Child Behavior Checklist (30Go–32Go) and the Youth Self-Report (33Go), beginning at age 15 years for the oldest sample and age 11 years for the youngest. For the youngest ages in which drug dealing and gang membership were not measured, we assumed a no response to minimize missing values. It is unlikely that many were miscoded given the young ages of the participants at this time. For instance, only six participants in the youngest cohort responded that they were dealing drugs the first year this question was asked.


View this table:
[in this window]
[in a new window]
  		TABLE 1. Variables considered in multivariate regression models, Pittsburgh Youth Study, 1987–2000

 
Community-level variables known or thought to be related to violence were included (table 1) (42Go). We assigned 1990 US Census data to the years 1987–1995 and 2000 US Census data to the years 1996–2000 for each neighborhood. Violent crime from the Uniform Crime Reports (43Go) was measured yearly for each neighborhood.

Analyses
To account for challenges in repeated measures, we used models computed from generalized estimating equations with the GENMOD procedure in SAS, version 8.01, software (SAS Institute, Inc., Cary, North Carolina) (44Go–46Go). The repeated subject was the array, each with up to 14 observations (i.e., 14 yearly assessments). The autoregressive working correlation matrix was used, as it was assumed that violence was more strongly correlated in closer follow-ups. (Analyses were also run with the independent working correlation matrix. Results were similar.) We identified initial models for further investigation through critically analyzing the data using descriptive techniques, our theoretical model, and prior literature. Our model-building algorithm was based on the work of Hosmer and Lemeshow (47Go). A modified blockwise approach was used, with a model containing only cohort and age, a model containing individual-level variables, a model containing community-level variables, and a model combining variables from both levels. Tests for period effects were performed separately for each model by adding the period variable into the model.

A common problem when trying to separate age, period, and cohort effects is the "identifiability problem" where a specific variable can be perfectly predicted by a linear combination of the remaining variables, resulting in no unique set of regression parameters (48Go–51Go). However, in our analysis, since 1) our cohort was based on school year (not age), 2) the age varied within each cohort, and 3) the grouping period comprised 4-year groupings, the linear dependency among age, period, and the dichotomized cohort effects no longer holds (52Go). To demonstrate the absence of the identifiability problem, we estimated eigenvalues for age, period, and cohort to check for linear dependency. All eigenvalues were positive (data not shown), demonstrating that linear dependency is not present (48Go). Additionally, a large correlation among any of the three covariates could cause multicollinearity in the model. However, the condition number, the ratio of the largest eigenvalue to the smallest, is not large and suggests that there is no collinearity problem. Further, three problems that occur with collinearity did not occur in our models: 1) large changes in coefficients when a variable is added/deleted; 2) wide confidence intervals, nonsignificant test statistics, unexpected algebraic signs; and 3) instability of the coefficient estimates between samples (42Go).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 References
 
Demographic characteristics of subjects
A total of 1,009 cases were in the study: 503 from the youngest cohort and 506 from the oldest. There were 14,126 available observations from all cases across all years with 4,592 missing observations (32.5 percent) in the final reduced model. Therefore, 9,534 observations were used in the analyses. All subjects contributed data for at least several years. Table 2 presents basic demographic information at baseline. Totals of 24 percent and 23 percent of the oldest and youngest cohorts, respectively, were from a family with low socioeconomic status, and 57 percent and 55 percent were Black with the remainder White. These factors were not statistically different across cohorts. A higher percentage of boys in the oldest cohort were held back in school at entry into the study (39 percent vs. 26 percent; p < 0.001). Further analyses are adjusted for this. Large differences in violence existed between cohorts. Figure 1 presents violence rates by cohort by year. The peak for the oldest cohort is 1987 (32 percent), and the peak for the youngest is 1990 (21 percent). (There are no data for earlier years.) Figure 2 presents the prevalence (per 100) of violence for the cohorts. At each age, the prevalence for the oldest cohort is higher than that for the youngest. The following generalized estimating equation regression models were run to attempt to explain these differences through a period or cohort effect.


View this table:
[in this window]
[in a new window]
  		TABLE 2. Basic demographics by cohort at baseline, Pittsburgh Youth Study, 1987–2000

 

Figure 1
View larger version (18K):
[in this window]
[in a new window]
  		FIGURE 1. Violence rates reported for the youngest and oldest cohorts, by year, Pittsburgh Youth Study, 1987–2000.

 

Figure 2
View larger version (17K):
[in this window]
[in a new window]
  		FIGURE 2. Violence rates reported for the youngest and oldest cohorts, by age, Pittsburgh Youth Study, 1987–2000.

 
Regression analysis
Tables 3, 4, and 5 present the results of regression analyses. In the cohort-only model (table 3), age and cohort were statistically significant, demonstrating that those in the oldest cohort were more likely to report violence (odds ratio (OR) = 1.73, 95 percent confidence interval (CI): 1.40, 2.14) after controlling for age. The second model (individual-level model) in table 3 consisted only of individual-level variables. All individual-level factors were significant. Carrying a gun (OR = 3.70, 95 percent CI: 2.70, 5.00), being a member of a gang (OR = 2.08, 95 percent CI: 1.45, 2.94), dealing drugs (OR = 4.17, 95 percent CI: 3.13, 5.88), or using hard drugs (OR = 2.63, 95 percent CI: 1.89, 3.70) was positively associated with violence. The effect of cohort remained significant (OR = 1.45, 95 percent CI: 1.17, 1.81), suggesting that differences between the cohorts could not be accounted for by these major individual-level influences. However, adding period into the model rendered cohort nonsignificant (OR = 1.02, 95 percent CI: 0.99, 1.51). The odds ratio for period decreased with succeeding periods. The odds ratio for the period 1987–1990 was 2.86 (95 percent CI: 1.47, 5.52) compared with the referent period of 1998–2000, 2.37 (95 percent CI: 1.43, 3.95) for the period 1991–1993, and 2.28 (95 percent CI: 1.50, 3.47) for the period 1994–1997. This is in accordance with the falling violence rates over those years.


View this table:
[in this window]
[in a new window]
  		TABLE 3. Generalized estimating equation regression model for individual-level risk factors and yearly self-reported violence, Pittsburgh Youth Study, 1987–2000

 

View this table:
[in this window]
[in a new window]
  		TABLE 4. Generalized estimating equation regression model for community-level risk factors and yearly self-reported violence, Pittsburgh Youth Study, 1987–2000

 

View this table:
[in this window]
[in a new window]
  		TABLE 5. Generalized estimating equation regression model for community- and individual-level risk factors and yearly self-reported violence, Pittsburgh Youth Study, 1987–2000

 
In the third model, the community-level model (table 4), the neighborhood factors of family poverty (OR = 2.07, 95 percent CI: 1.06, 4.05) and residential instability (OR = 2.86, 95 percent CI: 1.22, 6.67) were significant. The other neighborhood factors were not. The effect of adding period to the model was similar to that in the individual model. The effect of cohort was rendered insignificant (OR = 1.29, 95 percent CI: 0.86, 1.94), suggesting that differences in violence between cohorts may be explained by period effects. Again, the odds ratios for period successively decreased in accordance with the falling rates of violence over those years. The significant community-level factors remained the same (family poverty and residential instability), and the magnitude of the odds ratios did not change.

The final model (table 5) combines community- and individual-level factors eliminating nonsignificant variables to present the most parsimonious model. The effect of cohort remained insignificant (OR = 1.23, 95 percent CI: 0.78, 1.94) with period in the model. Period was again a strong correlate of violence. All of the individual-level variables remained significant correlates of violence, and the magnitude of the relations remained similar to that of the individual-level model. Significant community-level factors included the percentage of homicide by guns (OR = 2.66, 95 percent CI: 1.12, 6.28), median household size (OR = 1.28, 95 percent CI: 1.01, 1.64), and percentage unemployed (OR = 2.86, 95 percent CI: 1.28, 6.39).

To verify our model, sensitivity analyses were conducted by altering several dimensions of the analysis. First, in two separate analyses, we categorized period into either 2- or 3-year groupings. A second sensitivity analysis eliminated data for the early years in which questions on gang participation and drug dealing were not asked. These sensitivity analyses did not change the results, providing additional evidence of the appropriateness of the model.


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
References
 
This study is the first of its kind, as far as the authors are aware, that has simultaneously assessed age, period, and cohort on longitudinal changes of violence in individuals. We have adjusted for important longitudinally measured individual-level factors, as well as being able to capitalize on the timing of the Pittsburgh Youth Study, which occurred during the most recent violence epidemic. We found that the oldest cohort reported higher unadjusted rates of violence at each age. These differences in cohorts could not be explained by age, individual-level, or community-level factors in multivariate models. However, when period was included, differences between the cohorts were no longer significant. This provides evidence that period effects may explain in part the difference in violence between the cohorts in this study. We interpret this to mean that differences between cohorts in the Pittsburgh Youth Study cannot be explained by cohort effects—either inherent differences in the cohorts or different distributions of risk factors (e.g., gang participation). We do not assume causation. However, our sample is representative of Pittsburgh when adjusted for the sampling procedure of the Pittsburgh Youth Study, suggesting that the changing rates of violence in Pittsburgh between 1987 and 2000 were in part due to the changing period effects. In other words, these results provide evidence that changes in violence rates over time are due in part to changing social factors (period effects) and are not due to intrinsic differences between cohorts (cohort effect).

Previous studies have looked at period and cohort effects, but most explanations focus on single explanatory factors and suffer from the limitations described above. For instance, some authors exclusively assess relative cohort size, while others examine only relative cohort size and family structure (13Go–15Go, 53Go–55Go). The literature examining the effect of abortion on violence suffers from the same limitation (12Go, 56Go). This severely limits conclusions that can be made from the analyses, as no confounders or interactions between factors are accounted for. However, our results do support period-effects theories. Some of the more intriguing work has been on the relation between the crack-cocaine epidemic and violence. As the crack-cocaine epidemic reached its peak, Blumstein and Wallman (4Go) theorized that drug traffickers were recruiting youth who acquired guns for protection. In an empirical examination, Grogger and Willis (57Go) examined the introduction of crack-cocaine into 27 US metropolitan areas. Their results suggest that the arrival of crack-cocaine had significant effects on murder and aggravated assault. Cohort effects in trafficking offense rates were also observed; however, when age and period effects were included, the cohort effects were muted or disappeared. Another study found that, when a city experienced growth in crack-cocaine arrests, a similar growth in homicides followed (58Go). Fryer et al. (59Go) analyzed the relation between violence rates and a crack-cocaine index at the city and state levels and found a strong link between the index and homicide. The period of risk demonstrated in our data corresponds to several measures of crack-cocaine use. Additionally, we found a strong association between violence with gun carrying, drug dealing, and gang membership, corroborating the importance of these high-risk activities in the changing rates of violence over time.

Importantly, however, these major correlates of violence could not explain the difference between cohorts of the Pittsburgh Youth Study. Further, data suggest that the crack-cocaine market does not account for all of the changes in violence, particularly the drop starting in the early 1990s (3Go). Drug dealing has also been shown to be related to social and environmental factors specific to a period (60Go). If cohort effects existed, one might find that differences were due to differences in risk factors at the individual level. For example, if there was a "moral poverty," as has been suggested as a cohort effect (61Go), in the older cohort that led to higher levels of risk factors (such as drug use or gang membership) in the older cohort, we would expect to see these factors that were measured in our analysis explain away differences between cohorts. A similar argument has been made regarding abortion legalization and violence trends (12Go). We did not observe such a phenomenon. Even after we controlled for hard drug use, gang membership, drug dealing, and other individual-level variables, we found a significant difference between cohorts. These data suggest that the period effect is not explained by these major individual-level risk factors.

An important consideration in understanding changes in violence over time may be the shifting of the shape of the age-crime curve. In the Freiburg Cohort Study, police and court records of people born in the years 1970, 1973, 1975, and 1978 were examined (62Go). A log-linear analysis showed that period effects caused different shapes of the age-crime curves. This change may be due to the age at which a cohort intersects with peaks of high-risk period effects. For example, in our models in a secondary analysis, we found significant interactions between period and age (data not shown). It is possible that individuals at high-risk ages who intersect with high-risk periods begin initiation of violence earlier and/or desist later, extending the age-crime curve. Descriptive results of our analyses also support this (figure 2), as we see that the oldest cohort does not drop to the level of violence of the youngest cohort at age 19 years until 3 years later at the age of 22 years. This issue requires further study.

Our measure of a period effect was crude. The limitations in our analysis prevent the assessment of what these factors may be; consequently, it is imperative that future studies address this important question. It is likely that social and environmental factors (period effects) play a major role in how adolescents choose criminal paths. Our theory is guided by a developmental-ecologic model for understanding risk, which suggests that individual development, hence risk, is influenced by the ongoing structure of the social situations in which the individual lives or interacts (63Go). In particular, we hypothesize that violence differences between cohorts can be explained by period effects, and that individuals respond differently to these period effects. Previous work by Loeber (64Go) has described developmental pathways to problem behavior. If we consider the overt pathway, progression goes from minor aggression to physical fighting to violence. Although this model was developed for individuals, we contend that period effects have a significant impact on the progression from one level to another. We contend that social changes influence decisions made by youth that increase the risk of moving from one level to the next. For instance, gang membership can arise for societal reasons, such as poverty and unemployment (6Go). Drug dealing has also been shown to be related to period-specific social and environmental factors (65Go). The developmental ordering of these period factors needs to be further understood. Recently, several important papers have been published examining community effects in relation to youth development (66Go, 67Go). They have suggested that a full developmental model, including individual and social risk factors, is needed to explain the course of violent behavior. In light of our results and those of other papers, we underscore the particular importance of this.

The methodological, statistical, and theoretical approach that we have used here does not eliminate all aspects of the age-period-cohort identifiability problem. In fact, it is unlikely that any tools available to researchers at present would. One must consider our findings in light of this. Further, the Pittsburgh Youth Study consists only of males, and no conclusions can be made regarding the possible effects on females. However, given the limits of the current literature, we feel that we provide new and very intriguing evidence of the importance of period effects. We have been able to remove the statistical aspects of this problem and have utilized a data set that provides a unique and valuable opportunity to develop a model for violence trends based on a large amount of individual data from separate cohorts collected over many years.

In summary, our data suggest that the violence epidemic was due in part to specific social factors that affected individuals growing up during the period of the epidemic. These factors may have led to participation by individuals in high-risk activities, such as drug dealing or belonging to a gang, which increased their risk for violent crime. However, these individual-level variables could not explain the cohort differences. Our analyses also suggest that cohort effects do not exist, as they were rendered insignificant by the addition of period factors in each of the models. The next important steps are to assess what specific factors are driving the period effects. This information can then be used for more informed policy development. Knowing these factors can help in prediction of future increases or, if the factors are modifiable, they can be addressed to prevent future increases.


   ACKNOWLEDGMENTS
 
This study was supported in part by a training grant from the Centers for Disease Control and Prevention (1K01 CE000495-01), research grants from the National Institute of Mental Health (MH-73941 and MH-507780), and a research grant from the Office of Juvenile Justice and Delinquency Prevention (2001-JN-FX-K001).

Conflict of interest: none declared.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 

  1. Loeber R, Farrington DP, Waschbusch DA. Serious and violent juvenile offenders. In: Loeber R, Farrington DP, eds. Serious and violent juvenile offenders: risk factors and successful interventions. Thousand Oaks, CA: Sage Publications, 1998:13–29.
  2. Sampson RJ, Laub J. Crime in the making. Pathways and turning points through life. Cambridge, MA: Harvard University Press, 1993.
  3. Cook PJ, Laub JH. After the epidemic: recent trends in youth violence in the United States. Cambridge, MA: National Bureau of Economic Research, Inc, 2001. (NBER working paper no. 8571) (http://www.nber.org/papers/W8571).
  4. Blumstein A, Wallman J. The crime drop in America. Cambridge, MA: Cambridge University Press, 2000.
  5. Blumstein A, Rivara FP, Rosenfeld R. The rise and decline of homicide—and why. Annu Rev Public Health 2000;21:505–41.[CrossRef][ISI][Medline]
  6. Lawson G, Lawson A. Substance abuse in special populations. Gaithersburg, MD: Aspen Publishers, 1989.
  7. Meehan PJ, O'Carroll PW. Gangs, drugs, and homicide in Los Angeles. Arch Pediatr Adolesc Med 1992;146:683–7.[Abstract]
  8. Gottesman R, Brown R, eds. Violence in America: an encyclopedia. New York, NY: Scribner, 1999.
  9. Fagan J, Wilkinson D. Guns, youth violence and social identity in inner cities. In: Tonry M, Moore MH, eds. Crime and justice: a review of research. Vol 24. Youth violence. Chicago, IL: University of Chicago Press, 1998:105–88.
  10. Brown S, Eisenberg L. The best intentions: unintended pregnancy and wellbeing of child and families. Washington, DC: National Academy Press, 1995.
  11. Gruber J, Levine P, Staiger D. Abortion legalization and child living circumstances: who is the marginal child? Q J Econ 1999;114:263–91.[CrossRef]
  12. Donohue JJ, Levitt S. The impact of legalized abortion on crime. Q J Econ 2001;116:379–420.[CrossRef][ISI]
  13. O'Brien RM. Relative cohort size and age-specific crime rates: an age-period-relative-cohort-size model. Criminology 1989;27:57–78.[CrossRef][ISI]
  14. Steffensmeier DJ, Streifel C, Harer MD. Relative cohort size and youth crime in the United States, 1953–84. Am Sociol Rev 1987;52:702–10.[CrossRef]
  15. Steffensmeier DJ, Streifel C, Shihadeh ES. Cohort size and crime rates over the life course: the Easterlin hypothesis reconsidered. Am Sociol Rev 1992;57:306–14.[CrossRef]
  16. Easterlin RA. On the relation of economic factors to recent and projected fertility changes. Demography 1966;3:131–53.
  17. Easterlin RA. Birth and fortune: the impact of numbers on personal welfare. New York, NY: Basic Books, Inc, 1980.
  18. Nevin R. How lead exposure relates to temporal changes in IQ, violent crime, and unwed pregnancy. Environ Res 2000;83:1–22.[Medline]
  19. Stretesky PB, Lynch MJ. The relationship between lead exposure and homicide. Arch Pediatr Adolesc Med 2001;155:579–82.[Abstract/Free Full Text]
  20. Gordon RA. Issues in the ecological study of delinquency. Am Sociol Rev 1967;32:927–44.[CrossRef][ISI][Medline]
  21. Diez-Roux AV. Bringing context back into epidemiology: variables and fallacies in multilevel analysis. Am J Public Health 1998;88:216–22.[Abstract/Free Full Text]
  22. Schaie KW. Beyond calendar definitions of age, time, and cohort: the general development model revisited. Dev Rev 1986;6:252–77.[CrossRef]
  23. Farrington DP. Age and crime. In: Tonry M, Morris N, eds. Crime and justice: an annual review of research. Vol 7. Chicago, IL: University of Chicago Press, 1986:189–250.
  24. Hindelang MJ. Variations in sex-race-age-specific incidence rates of offending. Am Sociol Rev 1981;46:461–74.[CrossRef][ISI]
  25. Hirschi T, Gottfredson M. Age and the explanation of crime. Am J Sociol 1983;89:552–84.[CrossRef]
  26. Moffitt TE. Adolescence-limited and life-course-persistent antisocial behavior: a developmental taxonomy. Psychol Rev 1993;100:674–701.[CrossRef][ISI][Medline]
  27. Levitt SD, Dubner SJ. Freakonomics: a rogue economist explores the hidden side of everything. New York, NY: HarperCollins Publishers, Inc, 2005.
  28. Loeber R, Farrington DP, Stouthamer-Loeber M, et al. Antisocial behavior and mental health problems: explanatory factors in childhood and adolescence. Mahwah, NJ: Lawrence Erlbaum Associates, 1998.
  29. Wolfgang ME, Figlio RM, Tracy PE, et al. The National Survey of Crime Severity. Washington, DC: Bureau of Justice Statistics, US Department of Justice, 1985. (National Criminal Justice (NCJ) publication no. 96017).
  30. Achenbach TM. The child behavior profile. I. Boys aged 6–11. J Consult Clin Psychol 1978;46:478–88.[CrossRef][ISI][Medline]
  31. Achenbach TM, Edelbrock CS. The child behavior profile. II. Boys aged 12–16 and girls aged 6–11 and 12–16. J Consult Clin Psychol 1979;47:223–33.[CrossRef][ISI][Medline]
  32. Achenbach TM, Edelbrock CS. Manual for the child behavior checklist and revised child behavior profile. Burlington, VT: University of Vermont, 1983.
  33. Achenbach TM, Edelbrock CS. Manual for the youth self-report and profile. Burlington, VT: University of Vermont, 1987.
  34. Loeber R, Farrington DP, Stouthamer-Loeber M, et al. Antisocial behavior and mental health problems: explanatory factors in childhood and adolescence. Mahwah, NJ: Lawrence Erlbaum, 1998.
  35. Hollingshead AB. Four factor index of social status. New Haven, CT: Department of Sociology, Yale University, 1975.
  36. Blumstein A, Rosenfeld R. Trends in rates of violence in the USA. Stud Crime Crime Prev 1999;8:139–67.
  37. Hawkins JD, Herrenkohl T, Farrington DP, et al. A review of predictors of youth violence. In: Loeber R, Farrington DP, eds. Serious and violent juvenile offenders: risk factors and successful interventions. Thousand Oaks, CA: Sage Publications, 1998:106–46.
  38. Herrenkohl TI, Maguin E, Hill KG, et al. Developmental risk factors for youth violence. J Adolesc Health 2000;26:176–86.[CrossRef][ISI][Medline]
  39. Menard S, Mihalic S, Huizinga D. Drugs and crime revisited. Justice Q 2001;18:269–99.[CrossRef]
  40. Sampson RJ, Lauritsen JL. Violent victimization and offending: individual-, situational-, and community-level risk factors. In: Reiss AJ, Roth JA, eds. Understanding and preventing violence. Washington, DC: National Academy Press, 1994:1–114.
  41. Thornberry TP. Membership in youth gangs and involvement in serious and violent offending. In: Loeber R, Farrington DP, eds. Serious and violent juvenile offenders: risk factors and successful interventions. Thousand Oaks, CA: Sage Publications, 1998:147–66.
  42. Land KC, McCall PL, Cohen LE. Structural covariates of homicide rates: are there any invariances across time and social space? Am J Sociol 1990;95:922–63.[CrossRef]
  43. Federal Bureau of Investigation. Uniform crime reports: crime in the Unites States—1998. Washington, DC: US Government Printing Office, 1999.
  44. Diggle PJ, Liang KY, Zeger SL. Analysis of longitudinal data. Oxford, United Kingdom: Clarendon Press, 1994.
  45. Stokes ME, Davis CS, Koch GG. Categorical data analysis using the SAS system. Cary, NC: SAS Institute, Inc, 1995.
  46. Horton NJ, Bebchuk JD, Jones CL, et al. Goodness-of-fit for GEE: an example with mental health service utilization. Stat Med 1999;18:213–22.[CrossRef][ISI][Medline]
  47. Hosmer DW, Lemeshow S. Applied logistic regression. New York, NY: Wiley, 2000.
  48. Wothke W. Nonpositive definite matrices in structural modeling. In: Bollen KA, Long JS, eds. Testing structural equation models. Newbury Park, CA: Sage Publications, 1993:256–93.
  49. Mason KO, Mason WM, Wolfgang J, et al. Some methodological issues in cohort analysis of archival data. Am Sociol Rev 1973;38:242–58.[CrossRef][ISI]
  50. Rogers WL. Estimable functions in age, period and cohort effects. Am Sociol Rev 1982;47:774–87.[CrossRef][ISI]
  51. Holford TR. Understanding the effects of age, period, and cohort on incidence and mortality rates. Annu Rev Public Health 1991;12:425–57.[CrossRef][ISI][Medline]
  52. Yang Y, Fu WJ, Land K. A methodological comparison of age-period-cohort models: the intrinsic estimator and conventional generalized linear models. Sociol Methodol 2004;34:75–110.[CrossRef]
  53. O'Brien RM, Stockard J. Variations in age-specific homicide death rates: a cohort explanation for charges in the age distribution of homicide deaths. Soc Sci Res 2002;31:124–50.[CrossRef][ISI][Medline]
  54. O'Brien RM, Stockard J, Isaacson L. The enduring effects of cohort characteristics on age-specific homicide rates, 1960–1995. Am J Sociol 1999;104:1061–95.
  55. O'Brien RM, Stockard J. The cohort-size, sample-size conundrum: an empirical analysis and assessment using homicide arrest data from 1960 to 1999. J Quant Criminol 2003;19:1–32.
  56. Joyce T. Did legalized abortion lower crime? Cambridge, MA: National Bureau of Economic Research, Inc, 2001. (NBER working paper no. 8319) (http://papers.nber.org/papers/W8319).
  57. Grogger J, Willis M. The emergence of crack cocaine and the rise in urban crime rates. Rev Econ Stat 2000;82:519–29.[CrossRef]
  58. Cork D. Examining space-time interaction in city-level homicide data: crack markets and the diffusion of guns among youth. J Quant Criminol 1999;15:379–406.[CrossRef]
  59. Fryer RG, Heaton PS, Levitt SD, et al. Measuring the impact of crack cocaine. Cambridge, MA: National Bureau of Economic Research, Inc, 2005. (NBER working paper no. 11318) (http://www.nber.org/papers/w11318).
  60. Braga AA. Serious youth gun offenders and the epidemic of youth violence in Boston. J Quant Criminol 2003;19:33–54.[CrossRef]
  61. Bennett WJ, DiLulio JJ, Walters JP. Body count: moral poverty—and how to win America's war against crime and drugs. New York, NY: Simon & Schuster, 1996.
  62. Grundies V. The effect of period on the age-crime curve. A log-linear analysis of crime rates estimated by a study of four cohorts. Frieburg, Germany: Max Planck Institute for Foreign and International Criminal Law, 2000.
  63. Bronfenbrenner U. The ecology of human development: experiments by nature and design. Cambridge, MA: Harvard University Press, 1979.
  64. Loeber R. Questions and advances in the study of developmental pathways. In: Cicchetti D, Toth S, eds. Rochester symposium on developmental psychopathology. Vol 3. Rochester, NY: Rochester University Press, 1991.
  65. Saner H, MacCoun RJ, Reuter P. On the ubiquity of drug selling among youthful offenders in Washington, D.C., 1985–1991: age, period, or cohort effect? J Quant Criminol 1995;11:337–62.
  66. Tolan PH, Gorman-Smith D, Henry DB. The developmental ecology of urban males' youth violence. Dev Psychol 2003;39:274–91.[CrossRef][ISI][Medline]
  67. Sampson RJ, Raudenbush SW, Earls F. Neighborhoods and violent crime: a multilevel study of collective efficacy. Science 1997;277:918–24.[Abstract/Free Full Text]

Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?


This article has been cited by other articles:


Home page
 Inj. Prev.Home page
A Fabio, Z Yuan, S R Wisniewski, D B Henry, D P Farrington, J A Bridge, and R Loeber
Cohort differences in the progression of developmental pathways: evidence for period effects on secular trends of violence in males
Inj. Prev., October 1, 2008; 14(5): 311 - 318.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
164/2/151    most recent
kwj172v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (1)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Fabio, A.
Right arrow Articles by Farrington, D. P.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Fabio, A.
Right arrow Articles by Farrington, D. P.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?