Authors’ Rejoinder to Respondents (Shulman,Steinberg,& Piquero,2013)

We appreciate Respondents’ (Shulman, Steinberg, & Piquero, 2013a) detailed comments. However, they misunderstand our method, raise “ecological fallacy” objections that do not apply to our population-level study, and counter with research that includes only limited ages and inadequate methods to assess socioeconomic factors in crime and risk taking by age.

Our 2011 and 2013 articles investigated—evidently for the first time—the interaction of age, race, and poverty level with criminal arrest rates. Traditional age-crime-curve research has not controlled for older versus younger ages’ substantial socioeconomic disparity. We attempted to fill that research gap with preliminary analyses suggesting that at least for Californians age 14-69, the age-crime curve is an artifact of the substantially lower socioeconomic status (SES) younger ages suffer in comparison with older ages. Where the demographics of adolescents and emerging adults (in California, 54% Latino or African American, average poverty rates of 17%) and middle-agers (65% non-Latino White or Asian, average poverty rates of 9%) can be standardized, adolescents and emerging adults have violent crime arrest rates equivalent to those of middle-agers.

In contrast, Respondents cite studies (as did we) that compare divergent offending rates by age within populations, compare divergent offending rates by SES across populations, and show that controlling for SES mitigates disparities in offending rates by race or cohort. But they cite only one reanalysis relevant to our articles: their own, just published article on SES limited to offending rates among ages 12 through 23 that contains serious validity issues, discussed later.

Respondents first object that we supplied insufficient information about the geographic unit used to determine poverty level, and their confusion accounts for most of their objections to our method. Contrary to Respondents’ assertions, we did not apply the aggregate, county-level poverty rate to all ages and races or “cluster” counties as diverse as Los Angeles and Shasta in a single poverty bracket. Rather, poverty and arrest rates were tabulated separately for each age-by-race-by-county cell. Within each of the 58 counties, 20 separate arrest rates and poverty rates were calculated—one for each of the five age groups times the four major races. As explained in the original article’s method, each cell represented a separate demographic unit consisting of a race-age-county population (e.g., Hispanic—18- to 19-year-olds in Fresno County; or Asian—40- to 69-year-olds in Orange County) whose rates of poverty and arrest were then calculated. Cells with fewer than 50 people per age-year were excluded because poverty rates are nonexistent or unreliable for small populations. The remaining 906 cells containing 99.7% of the state’s arrestees were used in our analysis. These cells display large economic disparities. For example, while no county has an aggregate poverty level below 5%, there are 8 counties in which Whites and 13 counties in which Asians ages 40 to 69 have average poverty rates of below 5%.

Respondents are correct that some of those arrested in a county do not live there. However, law enforcement data obtained for Santa Cruz County showed that even in a county with particularly large tourist, student, and migrant labor populations relative to its resident population, 91% of arrestees were county residents. Other counties’ residents arrested in Santa Cruz would be offset to some extent by Santa Cruz residents arrested in their counties. This supports the assumption that the large majority of those arrested in a county are residents of that county. ¹

Respondents mischaracterize our conclusion that higher arrest rates associated with environments of poverty may be best explained largely or wholly by factors associated with poverty—including differential offending by poorer populations, stronger policing of poorer populations, and other well-known factors—as an “ecological fallacy.” “The ecological fallacy consists in thinking that relationships observed for groups necessarily hold for individuals” (Freedman, 1999). A standard example of the ecological fallacy is a statement, such as “less guilty by reason of adolescence” (Steinberg & Scott, 2003, title), that argues for applying tendencies alleged to characterize adolescents as a group to assess the guilt of individual adolescent defendants. Respondents are inconsistent in their concern for ecological fallacy and behavior risk. While they repeatedly (and without citing a single example) accuse us of applying aggregate SES data to predict individual behavior, Respondents routinely apply aggregate concepts of age and developmental stage to predict individual adolescent and adult behaviors. They postulate that petty offenses are a major indicator of age-specific “adolescent risk taking,” but they do not similarly claim that the later peak in more serious offenses and behaviors such as homicide and illicit-drug abuse mortality constitute age-specific “adult risk taking.”

In contrast, nowhere in our articles is there any claim that individual criminal propensities can be predicted from population-level poverty rates. Nor do we argue for individually directed policies such as curfews or “stop and frisk” policies targeting poorer demographics or advocate for treating poorer defendants as “less guilty by reason of poverty.” In fact, our article seeks to challenge the systematic ecological fallacy underlying the many age-based policies that collectively restrict or punish all adolescents along with the few that benefit certain adolescents, which Respondents (Shulman, Steinberg, & Piquero, 2013b) express incredulity that our analysis would dispute. Our conclusion is solely population based and recommends macro-level remedies consistent with a lengthy, broad literature on poverty factors that our findings reconfirm.

Respondents’ criticism that arrests of 18- to 29-year-olds underreport their real crime rates is true but unfair. Arrests underreport actual crime rates for all ages. Campus crime volumes are certainly larger than arrests indicate, just as domestic violence volumes are much larger than arrests of adults’ ages 25 and older indicate. Domestic violence, which involves an older average age of offender (Criminal Justice Statistics Center, 1999), is both notoriously underpoliced and suffers a long history of being treated by researchers and policy makers as a lesser type of crime or omitted from discussion altogether (Robinson, 1997).

Respondents (Shulman et al., 2013a, 2013b) also miscite Sampson, Morenoff, and Raudenbush’s (2005) body of research, which they admit does not directly apply to our findings. The Sampson et al. (2005) study Respondents cite provides little more than a paragraph on the basic age-crime connection, noting that “the probability of violence accelerates in early adolescence for all groups [in the study], reaching a peak between the ages of 17 and 18 and then declining precipitously” (Sampson et al, p. 4). Causality for this connection is not addressed explicitly, though the study does demonstrate that “neighborhoods explain a large percentage of individual-level disparities”; in particular, neighborhoods with high levels of concentrated disadvantage as a result of entrenched poverty are especially criminogenic (Sampson et al., 2005, p. 6). Cohort studies involving self-reported offending across the life cycle encounter not only the validity issues we discuss below but also the same confound hampering other previous studies: Do individuals offend less as they age due to older age, or due to the fact that most people become richer as they age? Sampson’s larger work would suggest the latter, noting that the notion of spontaneous desistance from crime is contradicted by the reality that people cease routine criminal offending through “turning points” that arise as the result of “opportunities,” such as stable employment, marriage, or wealth accumulation—all of which are confounded with increasing age. In all cases, the environment—the neighborhood in Sampson’s unit of analysis—remains particularly critical in sustaining or deterring criminal activity (Laub & Sampson, 2006; Sampson, 2012).

That brings us to Respondents’ newly published study, which employs National Longitudinal Survey of Youth (NLSY) offending and SES data over time (Shulman et al., 2013a) along with continuous claims to have refuted our work. Unfortunately, the age range in Respondents’ study is limited to 12 through 23, so that no conclusions can be drawn about offending by the ages 24 to 69 that our article includes. Respondents’ article finds self-reported crime and index offending declines substantially from ages 15 to 23 before and after SES is controlled.

Aside from investigating only a truncated age range, Respondents’ study contains at least three devastating limitations. The first, generic to self-reporting surveys, is the conundrum of “honestly reporting one’s dishonesty.” The same pattern found by Respondents could result from either a real decline in offending or a decline in willingness to report offending by age and over time. A 23- or a 40-year-old who beats his children may not admit to violence like a 15-year-old who starts a schoolyard fight. The suspicion that adults underreport their crime is reinforced by Federal Bureau of Investigation (FBI, 2013) totals showing arrest numbers for adults in their 20s and 30s are much higher than, and for adults in their 40s, equivalent to, those of juveniles for major offenses like assault. According to adolescent-risk theories, arrest levels of adolescents and 40-agers for a universal offense like assault shouldn’t even be in the same galaxy. More striking still, FBI (2013) clearance reports show juveniles are overarrested; that is, they commit a substantially lower proportion of crimes than their arrest proportions would predict. For example, youths below age 18 accounted for 12.7% of violent index and 20.4% of property index arrests in 2011 but just 9.5% of violent crimes and 14.2% of property crimes cleared by arrest.

If older adults do define and acknowledge their crimes the same as do adolescents, then a paradox emerges: Teenagers are getting away with vast amounts of petty and serious crimes for which adults are getting arrested. This is counterintuitive, because, if nothing else, experienced (usually older) criminals should be more adept at avoiding arrest, especially for property and other lower level offenses, than novice (typically younger) ones. Furthermore, public health statistics tied to a variety of crimes, such as hospital emergency treatments and deaths from abusing illicit drugs, peak at ages well beyond teen years. Taken together, independent measures indicate that the self-report data Respondents rely on decidedly is not convergent with official records.

The second major problem is that Respondents’ (Shulman et al., 2013a) analysis confounds two very different time periods. Crime declined sharply among all ages from the mid-1990s (when NLSY Wave 1 data were collected from 10- to 17-year-olds) to the 2000s (when Wave 6 and later Waves data were collected from 19-24 year-olds). Violent victimizations by single offenders estimated by victims to be ages 12 to 17 in the National Crime Victimization Survey (NCVS) fell by 50% from 1996 (Wave 1) through 2003 (Wave 7), as did single- and multiple-offender victimizations by ages 12 to 20. In tandem, Part I arrest rates fell by more than 50% among 15- to 17-year-olds from 1996 to 2011, with arrest rates declining considerably faster for teenaged youths than for emerging adults (Figure 1). While 15- to 17-year-olds and 19- to 24-year-olds had approximately the same arrest rate in 1996, the arrest rate for 15- to 17-year-olds fell to 40% below age 19 to 24’s rate by 2011. Yet Respondents do not adjust their changes in offending from Wave 1 to later Waves to control for the general decline in crime concentrated in younger ages over their study period. Thus, they map not an “age-crime curve,” but a “time-crime trend.”

OPEN IN VIEWER

Figure 1.

Change in the proportion of population arrested for any offense, ages 10 to 24, United States, 2011 versus 1996 (FBI, 2013).

Finally, Respondents (Shulman et al., 2013a), using figures from our race-by-age-by-poverty table, omit the poverty factor, retain only the age and race factors, and reproduce the standard age-crime curve for each race individually. It is not clear what the purpose of their exercise is. Controlling solely for race is inadequate, as our table shows, because youths and young adults suffer considerably higher poverty levels than do older adults within every race. Respondents also omitted Asians, a population of 4.5 million. For unknown reasons, Respondents then presented the race curves inefficiently and inconsistently in three separate figures with different scales.

Figure 2 presents the age-crime curves by race in one figure using one scale to allow direct comparisons. The problem (even using unadjusted rates) is immediately evident: African American 40- to 69-year-olds, an age supposedly all but immune to perpetrating crime, have violent crime rates 1.66 times those of White 18- to 19-year-olds and 3.23 times those of Asian 18- to 19-year-olds. This reality is difficult to explain for theorists who have invested in biodevelopmental terminology to explain heightened offending as features of adolescents’ innate “structural and functional changes in the brain,” “heightened propensity for risk-taking,” and “traits (e.g., sensation-seeking, reward salience, and susceptibility to peer influence)” (Shulman et al., 2013a, pp. 858, 859). Respondents are miffed that we characterize such claims about adolescents as pejorative; imagine how pejorative they would seem if applied to explaining even larger discrepancies in offending by race! Indeed, racial disparities in crime are explained not as the result of innate qualities but of environments of poverty, disadvantage, and discrimination. Yet, Respondents overlook the fact that the systematic social disadvantages associated with the much higher rates of offending by African Americans and Hispanics relative to Whites and Asians also explain the higher rates of offending by younger ages relative to older ones. Racial and age-based disadvantages are not only strikingly similar but also overlap demographically.

OPEN IN VIEWER

Figure 2.

Unadjusted rates of violent crime per 100,000 population by age and race/ethnicity (poverty level not controlled), California, 2010.

Figure 3 corrects Respondents’ misapplication by reinstating the poverty interaction. The figure presents violent crime rates by age and race at standardized poverty levels of 12.5% to 17.5%, a bracket constructed to maximize representations of the five age groups and four races. ² When poverty is standardized, Latinos and African Americans show modest age-crime curves while Whites and Asians show offending peaks in the mid- to late 20s. For all races together, offending peaks in the late 20s, with 30-age rates higher than are found among age 18 to 19, and age 40 to 69 rates only marginally lower than for age 14 to 17. California’s corrected age-crime curve is a mildly inverted U, with lower rates below age 18, a peak in late 20s, and a decline after age 50.

OPEN IN VIEWER

Figure 3.

Adjusted violent crime arrests per 100,000 population by age and race (standardized to poverty levels of 12.5%-17.5%), California, 2010.

Respondents’ contention that race is a better unit of analysis than SES because race is an enduring quality does not challenge our findings. The explicit purpose of the present study was to not confine investigation to enduring, immutable traits. Whether investigating an enduring, immutable trait like race or a temporary, immutable trait like age, potentially temporary, mutable traits such as economic disadvantage are highly relevant. Temporary “student poverty” among 18- to 29-year-olds would have different effects than the poverty endemic to chronically disadvantaged 18- to 29-year-olds; still, landlords do not allow 19-year-olds to delay paying rent until they are 40 and likely to have more money. Prioritizing immediate over long-term concerns is a necessary feature of coping with poverty. Even with these complications, poverty levels and demographics remain strong predictors of 18 to 29 violence arrest rates.

The critical issue is not enduring versus temporary traits but immutable versus mutable ones. Attributing crime and other risks to factors associated with age versus those associated with poverty status raises very different issues of analysis and policy. Assigning an immutable trait, which resurrects discredited “biodeterminist” notions of the past, signals an unchangeable characteristic innate to the population and suggests policies directed at the population. A mutable trait reflects a potentially changeable characteristic that may be affected by policies directed at external factors influencing the trait, such as job development or antipoverty programs. Even Respondents’ limited examples show that low SES far exceeds young age in its impact on offending and risk-taking differentials across populations, and any “integrated theory” of adolescents (Shulman et al., 2013a, p. 858) that fails to pivotally incorporate the large SES differences between adolescents and older adults’ risks retreat into 19th century mythmaking.