ABSTRACT
Using public information from birth certificates and prison records from Florida, we adopt a reduced-form approach to estimate the effect of relative age on the probability of incarceration in adulthood (until age 30–40). We use a Regression Discontinuity Design around the cutoff date for Kindergarten enrollment (Sept. 1). We find strong evidence of relative-age effects among black males convicted of drug-trafficking offenses. For that group, being born after Sept. 1 decreases the probability of incarceration by roughly 12–20%. We find weak or no relative-age effects for other groups and types of offenses.
Disclosure statement
No potential conflict of interest was reported by the author.
ORCID
Pablo A. Peña http://orcid.org/0000-0002-1327-5412
Notes
1 There are 1,386 unique offense descriptions. To classify them into types, we parsed those descriptions searching for specific terms. The details of the classification are available from the author upon request.
2 The variable RELATED in the U.S. Census and the ACS includes the category for institutional inmates (1301).
3 There is no appeal to manipulate a birthdate to send a child later because redshirting is always possible.
4 Manipulation in the running variable in an RDD is usually tested using the method proposed by McCrary (Citation2008). However, in this setting we don't expect an entirely smooth density of birthdates around the cutoff because of weekends and holidays. In particular, such a test would be affected by Labor Day, which is the first Monday in September. We deem more appropriate a parametric version of the same test based on Equation (1).
5 For the years 1968–1972, the Natality Data from the National Vital Statistics System only contains a random sample of 50% of all birth certificates from Florida. To compute the total number of births on birthdates in those years, we multiply the number of births by two.
6 For people born around Sept. 1 of 1975, 1976 and 1977, the relevant cutoff dates were Dec. 1, Nov. 1 and Oct. 1, respectively. We didn't include those cohorts to avoid the presence of other cutoff dates in the comparison group that could introduce a confounding factor. Although we could selectively restrict the windows of analysis to exclude those cutoff dates, we would lose comparability across windows.
7 The same reasons make unappealing the use of an RDD to estimate the effect of the Jan. 1 cutoff among those older cohorts.
8 Fifteen other states had a Sept. 1 cutoff applicable to people born in 1978–1988. Texas and Utah had a ‘start of the school year’ cutoff, which roughly is the start of September. Utah later applied a Sept. 2 cutoff.