Educational Spending and Imprisonment Risk: The Role of Schools as Prison Gatekeepers

Richard Arum, New York University
Gary LaFree, University of Maryland at College Park

During the last thirty years of the twentieth century, incarceration rates in the United States increased to levels that were both historically and comparatively unprecedented (Austin and Krisberg 1998; Western and Beckett 1999). This study examines the possibility that individual incarceration risk during this period was directly linked to the amount of resources states spent on their secondary school systems. Most prior research on incarceration rates has explored the effects of macro-level economic, political, and demographic variables, including unemployment (Grant and Martinez 1997; Chiricos and Delone 1992), economic inequality (Garland 1990; Greenberg 1999), electoral cycles (Beckett 1997; Jacobs and Carmichael 2001), welfare spending (Greenberg and West 2001) and race relations (Jacobs and Carmichael 2001; Greenberg and West 2001). These studies have begun to provide an understanding of the structural parameters in which individual-level processes leading to incarceration may occur. However, individual level modeling of imprisonment rates has thus far been hampered by methodological limitations and theoretical uncertainty over the appropriate conceptual framework for such an analysis. By using previously unreleased U.S. census data that include prison inmates, we are able to examine the impact of educational spending in the states where individuals attended high school on their subsequent imprisonment risk as adults. While we could identify no prior research that uses census data to examine connections between state-level spending on education and individual imprisonment risk, there are sound theoretical reasons to expect such a link. Prior research (Gottfredson and Hirschi 1990; Sampson and Laub 1993) has demonstrated strong connections between educational variables, criminal behavior and criminal justice processing. Educational variables have also been shown (Card and Krueger 1996; Arum 1996) to affect outcomes such as occupational status and earnings, which are in turn likely to affect criminal propensity and contacts with the criminal justice system. Based on prior research, we develop three competing models to explain the expected impact of state-level spending on the individual risk of imprisonment. An educational resource model sees high school as a critical defining moment and predicts that educational spending will directly reduce individual incarceration risk. By contrast, a null hypothesis takes the position that the propensity to commit crime is relatively fixed at an early age and therefore educational spending will have little impact on criminal propensity or adult incarceration risk. In addition, we develop a gatekeeper model of educational attainment that focuses on the role of secondary education in sorting juveniles into high and low status occupational roles, as well as highly stigmatized roles such as those connected with imprisonment. Our analysis is based on an innovative new methodology developed by labor economists Card and Krueger (1992, 1996), who used U.S. Census data to test the effects of state-level educational resource investment on labor market outcomes by assigning state-level, cohort-specific educational characteristics to individuals based on the state in which they were born. To control for differences in the return to education in different states, they estimated rates of return to schooling only on men who were educated in one state and then moved to another. For example, their analysis identifies whether high school graduates currently living in New Mexico have similar incomes if they were educated in New York (where educational resource investment was high) or California (where investment was low). They added state-level "fixed effects" to control for any unmeasured heterogeneity within states. Using this estimation technique, Card and Krueger demonstrated that educational resources invested in a state where an individual was born were systematically related to later adult earnings. We use logistic regression with state-level fixed effects for both state of origin and state of residence to calculate the likelihood of incarceration for adults who no longer reside in their birth state. Our research makes use of three waves of previously unreleased micro-level data on state prisoners and local jail inmates from the 1970, 1980, and 1990 U.S. censuses. Because they were not necessarily located in the same state where conviction and prior residence occurred, we exclude all federal prisoners from our analysis. We merge 16% micro-level census data on state and local prisoners with 5% samples of publicly released micro-level data on non-institutionalized individuals from the 1980 and 1990 U.S. censuses and two 1% samples from the 1970 census. We re-weighted data from the various samples to adjust varying sample densities to represent the population accurately. Following Card and Krueger, we analyzed 4,521,252 interstate migrants between ages 18-60 (i.e., individuals who report differing states of birth and residence). In addition, we also analyzed a more restrictive sample of 440,165 recent interstate migrants that includes only individuals who lived in their state of birth five years prior but resided during the census in a different state. Our analysis focuses on three indicators of the role of educational experience on adult incarceration. First, we examine reduced form models in which effects of state-level educational resources (measured as student-teacher ratio and imputed on the basis of reports of state and year of birth) are assumed to occur solely by affecting average incarceration rates of individuals exposed to education in a given state (i.e., by affecting variation in intercepts associated with state of origin). Second, we examine an interaction between high school graduation and public school resource investment. We look at high school graduation separately since non-linearities in the relationship between education and incarceration occur most clearly with respect to this dichotomy. Finally, we examine a more conventionally modeled interaction between years of education and our imputed assignment of educational resource investment. Our analysis is thus designed to explore the theoretical models described above.

(Return to Program Resources)

Updated 05/20/2006