Trendspotting: Millennials Drive the Secular Decline in Crime Rates

NewsWire: 5/24/21

• A new study tracking U.S. crime since 1980 confirms that Millennials have committed far less crime than Gen Xers or Boomers did as teens and young adults. The author argues that recent declines in crime have been mostly due not to crime reduction efforts, but to age and cohort effects. (Journal of Quantitative Criminology)
• NH: The near-term trend in U.S. crime rates is always a matter of uncertainty and speculation. While rates of violent crime jagged upwards in the summer and fall of 2020 (see "Homicides Spiked in 2020"), they now appear to be falling again.
• The direction of the long-term trend in crime, however, leaves no room for doubt. Over the last thirty years, America has experienced one of the most dramatic declines in criminal violence in its history. It hardly matters which measure you look at--the FBI's reported rate (based on police reports) or the BJS's rate (based on victimization surveys). Nor does it matter whether you're looking at violent crimes or property crimes.

• However you look at it, we're talking about a 50% to 75% reduction in the crime rate. The homicide rate for large cities (population over 250K) is back to where it was in the mid-1960s. Homicide nationwide is back to where it was in the early 1960s and late 1950s--in other words, near an all-time historical low.
• Such a massive trend has naturally prompted social scientists to ask: Why did this happen? Yet their research, while inspiring hundreds of reports and articles, has failed to generate many persuasive explanations.
• IMO, the best--and still largely unexplored--explanation is generational. Millennial cohorts born since 1982 grew up committing and experiencing significantly less violence than earlier-born Xers and Boomers had at the same age. As these cohorts entered their early teens in the mid-1990s, that's when the national crime rate decline began. And as they aged into their late teens and 20s during the late Clinton and GW Bush years, that's when the decline accelerated. Later-born Millennials and Homelanders have pushed these age-specific rates down even further.
• The idea of focusing on youth cohorts seems pretty logical when you reflect on the fact that the vast majority of crimes (especially property crimes, where rates have declined the fastest) have always been committed by young people under the age of 25 or 30. In other words, when we talk about the drop in overall criminality in America, we are mostly talking about the drop in youth criminality in America. So it makes sense to think carefully about what made these new youth cohorts different from those who came along before them.
• Two obstacles, however, have stood in the way of this obvious line of inquiry.
• One is conceptual. A cohort-based or "generational" approach just seems beyond the imagination of many criminal researchers, who would rather stick with the variables they're familiar with, like unemployment rates, police per capita, incarceration rates, conviction rates, punishment severity indexes, and so on. They may agree that it's plausible that cohorts with different experiences in early childhood--like much lower rates of sexual or physical abuse and lower rates of abandonment or running away from home--might commit less crime as they grow older. But plausible or not, criminologists just aren't used to thinking about such broad dynamics.
• The other obstacle is statistical. Most criminal researchers start with crime rates by year and by age. Year is important because that's how they track "period effects" (like employment, police, arrests, weather, etc.) and their impact on crime over time. Age is important because everyone knows that crime is tightly correlated with age--positively before about age 19 and negatively thereafter. So the first thing researchers do is isolate the "age effect" over the entire sample so they can then estimate all of the period effects.
• That makes it challenging to deal with birth cohorts as a separate variable. In effect, the equation becomes under-identified. You've got two independent variables, year and age, and you're trying to assess the independent effect of a third driver, birthyear, which can't easily be disentangled from the other two.
• I think I've now set the stage for this very important article by William Spelman at the LBJ School of Public Affairs at UT Austin. (See this press release for a quick synopsis.)
• In the first half of the paper, Spelman summarizes the statistical literature on crime and points out how researchers time and again have either ignored the cohort effect or found crude ways to minimize it. Spelman then lays out his own solution to the identification problem, which is (and here I'm simplifying a bit) to use yearly data across 25 different states as an independent means of estimating the period effects. This leaves him free to estimate both the age and cohort effect on the national data.
• In the second half of the paper, Spelman presents his results. He starts by showing the separate impact of age, period, and cohort for each of the UCR's seven "index crimes." These are: homicide, forcible rape, aggravated assault, robbery, burglary, larceny, and auto theft.

• Note that the vertical axis shows the relative crime commission rate, log scaled and with a mean of zero. (So for example +1 is roughly 10% above the mean.)
• Like everybody else, Spelman finds that the crime rate is tightly linked to the age of the perpetrator--though by definition this linkage doesn't change over time. He also tracks over a dozen period variables, by definition linked to year, and finds that these all pushed modestly in the expected directions. An improving economy, more police, and busier prisons during the 1990s, for example, tended to pull the overall crime rate downward.
• The big news, though, is in the link between birthyear (cohort) and crime rate. According to his estimate, a later birthyear almost always translates into a lower propensity for crime. What's more, the effect becomes steep after the 1980 birthyear--and steeper still after the 1995 birthyear. 
• In the next set of charts, Spelman isolates the independent effect of each of these effects on the average index crime rate in each year from 1980 to 2015. He estimates the age effect, for example, by fixing the period and cohort effects at their 1980 values and only changing the age composition of the population moving forward. Ditto for the other two effects.

• In concept, all three effects should sum to the actual change in the index crime rate, which (as shown here) fell by some 60% since 1980. In practice, it sums to a bit more due to interaction effects. Overall, Spelman concludes that the age effect (that is, an aging population) contributed about 10 points to this decline; all of the period effects contributed about 20 points; and the cohort effect contributed about 30 points--or about half of the total decline. Also, the cohort effect has become more important over time, accounting for nearly all of the crime rate fall since 2000. That's when the first-born Millennials were reaching their late teens and early 20s.
• Spelman's primary mission is to use state-of-the-art statistical methods to isolate the cohort effect on crime. And in this I believe he has been successful. He does not try to explain why the cohort effect is as large as it seems to be. He acknowledges that this is a separate task.
• He mentions that there are a couple of highly heterodox theories about the crime decline that are linked to a cohort effect. One, popularized by Freakonomics author Steven Levitt back in 2005, suggests that it is linked to rising abortion rates before 1980. (IMO, this theory has lost credibility after 30 straight years of declining abortion rates.) Another is that it is linked to declining levels of childhood exposure to lead.
• My own view is that the decline reflects broader changes in the Millennial (and now Homelander) family and school environment that have pulled these cohorts toward greater risk aversion, tighter conformity with rules, and closer relationships with their parents.
• In any case, Spelman ends his article with a provocative suggestion. For the purposes of forecasting, he says, it doesn't really matter why the cohort effect works, only that it does work. And because it's so powerful, we can already get a good glimpse into the future by looking at the criminal behavior of children today. That's how reliable the cohort effect is. "Keeping a close watch on youth arrest rates," he writes, "might... be useful as an early warning system: If burglary arrests for 10–12 year-olds increase significantly, we can reasonably expect burglary rates as a whole to increase within the next five to seven years."
• Recent national data on arrests of 10- to 12-year-olds are not available. But, in the spirit of Spelman's suggestion, let me show some recent trends in all "juvenile arrests" (arrests under age 18) through 2019.

• Notice how the downward trend under age 18 (reflecting later-born cohorts) is much steeper than the downward trend over age 25 (reflecting earlier-born cohorts, still including many Xers and even Boomers).
• Alternatively, you might want to look at arrests under age 18 as a share of total arrests and compare two years: 2019 versus 2000.

• Overall, the under-18 share of all arrests has fallen by slightly over 50%. For no type of offense has it risen.
• My takeaway from all of this: The cohort effect is still alive and well and working in our favor. Late-wave Millennials and Homelanders appear to be even less crime-prone than earlier-born Millennials were at the same age. That's why arrests are increasingly skewing toward older Americans. It's also why prisons are adding geriatric wards for the burgeoning number of incorrigible Xers and Boomers even as the incarcerated population under age 25 shrinks. (See "The Politics of Falling Crime" and "Growing Old Behind Bars").
• America faces plenty of challenges over the next decade. But you can probably cross a new youth-driven crime wave off your immediate worry list.

To view and search all NewsWires, reports, videos, and podcasts, visit Demography World. For help making full use of our archives, see this short tutorial.