Childhood Hunger and Adult Crime: The Impact of the Food Stamp Program
Barr and Smith (2021) #JHR_Threads explainer
This week’s #JHR_Threads will review a forthcoming article in the Journal of Human Resources by Andrew Barr and Alexander Smith entitled “Fighting Crime in the Cradle: The Effects of Early Childhood Access to Nutritional Assistance”.
Economists love to say “there is no such thing as a free lunch”. We often use it to describe the opportunity cost of scarce resources, but it is also literally true, and therefore hunger and poverty are usually positively correlated. This is because without income and work, there can be no trips to the grocer. And without ingredients, there can be no meals. And without regular meals, children eke out a level of consumption so small, they grow up malnourished and live below the biological minimum level needed for child development. Poverty, through malnourishment and stunted child development, can can make life feel hopeless, and hopelessness can make desperate choices appear best.
Andrew Barr and Alexander Smith have produced an exemplar study that plausibly shows that the Food Stamp program, by dramatically improving the development of cohorts through increased nutrition, caused a sizable decline in birth cohort crime at the onset of early adulthood. This paper adds to a growing body of research that shows early childhood interventions can have developmental ramifications so large, they may change a person’s entire life trajectory and in so doing, society itself.
Malnutrition, Food Stamps and American History
I was born in a small town in Mississippi in 1975. Though now half a lifetime has passed, I can still remember a childhood filled with awe and wonder. My days were filled with trips to the library or the local pharmacy where I’d take a stack of comic books off the rack and read for hours stories about super heroes and villains. Though my town was very poor, so long as the tires on my bicycle had air, I felt rich.
My father worked hard to provide our family with the money my mom needed to buy groceries and prepare meals. We would occasionally eat out, such as after church every Sunday at a local cafeteria, but mostly it was my mom’s delicious Southern cooking that kept us alive. We were not rich, except by the measures of human history itself, where I was wealthier than anyone my ancestors had ever seen or dreamt of.
But while I grew up privileged by my dad’s stable employment, and my mom’s skill in the kitchen, many people in my town had not been so lucky. Men and women, now adults, had endured the hardships of poverty and racism in their youth, and, as I learned from Barr and Smith (2021), that poverty included extreme malnutrition and food insecurity.But I did not know about food insecurity because I had never been food insecure. I had never gone hungry before.
But many Americans in the 20th century in particular knew intimately what hunger felt like. Many people in my community, only a few years before I was born, had grown up in such horrifying poverty that almost no meals would be consumed on a regular basis. Pregnant mothers would skip meals regularly, or eat small simple dishes such as rice, just so that they could feed the others in the family. Richard Wright, the great African-American novelist, once wrote that he was so poor growing up he would drink water until his stomach was full just to stave off the sharp pains of hunger.
Hunger is caused by poverty but it is not so easy for many of us to see it due to the segregation of class and race and economic inequality more generally. My parents likely had never been invited to visit a poor friend at night, and thus never witnessed first hand the paucity of food at dinner time. Without witnessing our neighbor’s suffering, we can be lulled into imagining that our lives are representative of others plus or minus some epsilon. Humans are not born with an understanding of variance, and when it extends beyond the support of their own experience, they may struggle to imagine what lies at the left tail of poverty. And when we do not know first hand what lies at the left tail of poverty, we will likely simply ignore it.
But that ignorance became hard to maintain when CBS aired “Hunger in America” in 1968. This documentary showed Americans a level of malnutrition in their neighbors’ lives so severe that it included death and starvation. Viewers would see with their own eyes that in certain pockets of their own country, poverty was so severe that babies would perish from lack of food.
Government reports began to appear around the same time showcasing just how bad food insecurity was in the shadows of the country. Barr and Smith (2021) write about these reports:
“In a nation-wide study of hunger, the Citizens’ Board of Inquiry into Hunger and Malnutrition in the United States estimated that at least 10 million people were suffering from hunger and malnutrition. A team of doctors reported to congress that the diets of children in many impoverished areas rarely contained food other than bread. In Mississippi, they estimated that half of the two thousand children they observed were below the third percentile in weight, and in some counties they found the prevalence of anemia from malnutrition was above 80 percent.” (my emphasis, Barr and Smith, 2021, page 7).
The situation was bleak and prompted legislative responses by the federal government. A so-called “war" on poverty was declared under President Lyndon Johnson and in 1964, he signed the Food Stamp Act which substantially expanded the fledgling food stamp program that Kennedy had piloted a little earlier. The Food Stamp program allowed American families to purchase vouchers which could be redeemed at grocery stores for food at extremely steep discounts. The program was adopted and spread across the country like wildfire through high demand areas.
Research by economists such as Hilary Hoynes, Diane Schanzenbach, Doug Almond and others found that the program led to improvements in adult outcomes such as reduced obesity and diabetes. The program appeared to have been a success in improving the health capital of cohorts exposed to its transfers. But Barrs and Smith’s study is not about diabetes or obesity. Rather, it is about crime, and specifically violent crimes committed in adulthood.
Childhood Nutrition and Crime
The economics of crime is a somewhat mature field in microeconomics dating back to Gary Becker’s seminal 1968 article as well as Jeremy Bentham’s 19th century work. Most of the work in the economics of crime is typical for the microeconomics from the mid to late 20th century in that it focuses on the rational and voluntary individual decision to engage in crime. Naturally, if criminals are rational and weigh costs and benefits, then focusing on penalties for deterrence are natural choices for public policy. Early work in this area would explore the relative merits of policing and prisons searching for causal elasticities that could support incentive-based crime fighting policies. Evidence for deterrence has been found in several carefully identified studies, such as Italy’s collective pardons, but the elasticities discovered are often so small it casts some doubt on whether we can primarily lean on them for crime control.
But a separate literature explored whether childhood environments might be responsible for changing crime in adulthood. In a famous study by John Donohue and Steven Levitt, abortion legalization was suggested as at least partly responsible for the large, secular declines in crime that began in the early 1990s. But this theory was questioned and has since been more or less dropped by social scientists as an explanation for shifts in American crime rates. More promising explanations have focused on lead exposure and removal. But very little work, save a couple of small RCTs, have suggested that nutrition might be responsible for adult crime.
That has changed recently, though, in the last few years. Jill Carr and Analisa Packham, in a series of papers, present evidence that SNAP benefits can impact adult crime and domestic violence, but their work has tended to emphasize the program’s scheduling characteristics, not in utero and childhood development itself. Barr and Smith are unique in this pantheon of crime papers because of their focus on the Food Stamp Program’s nutritional benefits as opposed to the rational calculation of crime itself by adults. By providing nourishment and alleviating the sharp negative effects of poverty on the body’s development which can increase broadly defined human capital stock, something like a Food Stamp Program might reduce adult crime, not by changing the incentives adults face, but rather by changing the adult altogether.
North Carolina and its grand experiment
When I teach students about program evaluation, I ask them to imagine they had unlimited wealth and power. What social experiment would they design if they wanted to evaluate the causal effect of a program? This thought experiment often leads them to imagine a randomized experiment. What if we randomized, for instance, in utero and childhood food consumption large enough so as to cause an increase in early childhood development. If one of the outcomes we were interested in was adult crime, then simply comparing those randomly assigned food security to the comparison group would yield the average treatment effect of food security on crime since food insecurity would’ve been assigned independent of all potential outcomes.
The purpose of such thought experiments is not to suggest that such a social experiment should be conducted, though randomized experiments to evaluate public policy are always welcome. Rather, the purpose is to suggest a possible natural experiment to the student — a non-experimental situation where some individuals have been assigned food security, and some haven’t. Rarely are such treatment assignments caused by randomness, but randomness is not the only trick in our bag for estimating causal effects. There are alternatives to randomness that one might exploit for estimating causal effects such as smoothness, exclusion, parallel trends and strict exogeneity.
Barr and Smith (2021) lean on strict exogeneity as their identifying restriction by exploiting the gradual adoption of the Food Stamp Program in North Carolina in the 1970s as counties enrolled residents differentially over time. They then link those county level Food Stamp rollout data to administrative crime records from the North Carolina Department of Public Safety for crime convictions as well as the FBI’s Uniform Crime Reports for crime arrests.
The Public Safety administrative records are available from 1972 to 2015 for the entire state of North Carolina. These data cover all individuals convicted of a crime in the state. The data contains rich individual characteristics, such as the type of crime, whether it was a felony, whether it was violent, the person’s name, date of birth, gender, race, and most important for this study, the county of the individual’s birth. Combining this information with National Center for Health Statistics data on birth counts, they are able to construct conviction rates for birth month cohorts of individuals born in North Carolina from 1972 to 2015.
As said, the birth month cohort conviction rate is based on the number of crimes committed for a cohort combined with the NCHS birth counts. But it may be easier to understand if you are walked through its calculation. If we wanted to calculate a birth month cohort conviction rate for people born in Wake County in January 1965, we would simply divide the number of convicted individuals born in Wake County in January 1965 by the total number of individuals born in Wake County in January 1965.
To supplement their analysis, the authors also collected data from the FBI’s Uniform Crime Reports arrest records. While the NC Public Safety data contains exact birth counties, the UCR data only contains the county and month in which a person of a particular age was arrested for an offense. Analysis on auxiliary datasets was done to show that movement out of state and even out of one’s birth county was low enough to justify the inclusion of the FBI UCR records, though keep in mind that the relative strengths of the Public Safety are such that the FBI data is really more of a robustness check. Both datasets were used to calculate birth month cohort conviction rates for violent and property crimes. Summary statistics of these calculations are shown here below.
Their main estimation equation is a linear model of the following form:
The Food Stamp measure (FS) uses the fraction of months from conception to age 5 that Food Stamps were present for birth month cohort t in county c, which is the same definition used by Hoynes, Schanzenbach and Almond (2016). This model is estimated with county fixed effects and time fixed effects and controls for several baseline covariate trends.
The gamma coefficient on the Food Stamp measure is causal insofar as strict exogeneity holds. They interpret strict exogeneity to mean “conditional on birth county and birth cohort fixed effects, Food Stamp availability is uncorrelated with other factors that would lead a particular birth cohort to be more or less likely to commit crime” (p. 15). They present the main results from this OLS regression in Table 2, which I reproduce below.
The point estimate on “any crime” is -0.013. To aid interpretation, they suggest dividing 1.3 by 5.75 where the 0.75 corresponds to the 9-month in utero period (footnote 30). This means that each additional year of Food Stamp Program availability in early childhood, in utero to age 5, is associated with a reduction in the likelihood of any criminal conviction by age 24 by 0.23 percentage points. This effect is statistically significant both adjusting for within-county serial correlation as well as randomization inference with 1,000 simulated randomized treatment assignments.
These effects are more precise for violent crime than they are for property crime, though each point estimate is contained within the confidence intervals of the other. The effect of Food Stamp Program availability in early childhood on violent crime is closer to a 0.09 percentage point reduction by age 24. The second row of Table 2 focuses on felony conviction and finds that the most precise effects come from violent felonies, though property crimes are similar in magnitude and sign. Table 2 constitutes the main results of the paper. The rest of the paper focuses on heterogeneity analysis, threats to validity due to endogenous adoption and falsification exercises.
Testing for endogenous adoption is challenging since usually it is unobserved selection that we are concerned about. And unobserved selection means that Barr and Smith cannot include such covariates in their baseline X-trends by definition. One would need for these unobservables to still have explanatory power even after conditioning on a wide set of observables. A lot of background variation is absorbed by these fixed effects and baseline trends, but perhaps unobservables would make it through still. Instead of searching for that missing X-factor, they exploit logic and falsifications. They explain the approach they take here with the following reasoning:
“If … endogenous policy implementation were occurring, we would expect to see some strong association between county characteristics and the timing of adoption. Hoynes and Schanzenbach (2009) and Almond, Hoynes and Schanzenbach (2011) argue convincingly that this was not the case and that the rollout of the FSP was largely dictated by funding limits.” (page 18).
But in an appendix table, they nonetheless explore this possibility themselves by regressing county characteristics onto FSP timing and aside from counties with larger Black populations in 1960 rolling out FSP earlier than others, county characteristics explain little of the variation in FSP timing.Falsifications were also employed in this project by ingeniously examining FSP exposure’s effect on individuals born outside of NC who now live in NC counties. But these effects are small, positive and imprecise thus suggesting the original results may in fact be causal.
Overall, the preponderance of evidence produced in this document makes a convincing case that exposure to Food Stamp Programs in utero to age 5 caused an economically and statistically significant decline in crime by age 24. Using measures of the cost of violent crime, they conclude that simply by reducing crime later in life, the benefits of the program far exceed the cost of administering the program itself.
This study fits alongside other studies that show early childhood interventions have benefits that extend into adulthood and sometimes even beyond the scope of the original program itself. This possibility should, if nothing more, cause us to consider the possibility that sometimes the unintended consequences of poverty programs select on the upside of external benefits, and not merely the usual pessimism of economist worry about perverse incentives.
But more to the point, this study adds one more link in what has now become a very long chain finding that the social costs of poverty, including hunger itself, are gigantic. Poverty winds back around to the rest of us through heightened crime and violence, not merely due to economic tensions, though those as well, but also through the biological warping of the mind, body and preferences of people as they progress into adulthood. If love for one’s neighbor is not enough an obligation to motivate consideration of policies aimed at reducing poverty, then perhaps consider efficiency due to external social benefits from reduced hunger.
Food. Insecurity. These two words sound strange to me when said side by side even now. I first learned the phrase over a decade ago, and I found it jarring because to me, insecurity was an emotion based on a nexus of thoughts brought upon by the harsh realities of social competition. But it is a sign of economic privilege that when one things of insecurity, one thinks primarily of social anxiety.
They also consider that the cause of their results was other War on Poverty programs, but as these were adopted at the state level for the most part in uniform, they think this can’t explain their results based on within county variation in food stamp exposure.
They also examine issues related to inter-state migration and inter-county migration using other data sources, but find little evidence to suggest that this could be driving their results.