A2: Advantages of the dataset for the paper’s purposes
In addition to containing information on each individual’s performance in end-of-grade tests in the third and eighth grades and years of education completed by age 20, this dataset has a number of useful features for studying the impact of school entry laws on educational outcomes and teenage fertility.
First, unlike previous studies, this dataset uses individuals’ administrative school records up to age 20 rather than self-reported age and education data from birth certificates, which are not only higher quality but are also more representative of individuals’ final educational attainment since the majority of girls who give birth at school ages eventually return and graduate from high school (Mott and Marsiglio 1985; Upchurch and MacCarthy 1990). Comparing data from administrative school records and data from birth certificates for mothers aged 18 or below in this dataset, I find that while actual age and self-reported age on birth certificates are almost always identical (99.2 %), there is substantial disagreement between administrative records of grades attended and self-reported years of education. In particular, administrative records of number of grades attended exceed self-reported years of education for 40.3 % of observations (and fall below self-reported years of education for another 19.7 %, possibly due to private schooling or over-reporting).
Second, this dataset excludes girls who did not attend in-state public school and are hence more likely to have moved out of state. Since births to these girls are more likely to be unobserved, excluding these observations from the sample reduces potential downward bias on birth probabilities. Moreover, data on teenage childbearing outcomes are available at the individual rather than the cohort level. Hence, this dataset allows for more unbiased and precise estimation of the impact of school entry age on early childbearing.
Third, this dataset includes all women from six birth cohorts who attended public school in North Carolina between third grade and age 15 rather than only women who have had a live birth. Hence, the analysis of the impact of school entry age on educational success in this paper applies to a relatively broad socioeconomic spectrum.
Fourth, this dataset contains detailed information about characteristics at the individual and school and school district levels, allowing for a more effective search for heterogeneous treatment effects.
A3: Lower compliance rates among women born before the cutoff date
Among women born up to 60 days before the cutoff date, 77.5 % start third grade at age 8 together with the majority of their birth cohort, with 21.7 and 0.8 % starting at ages 9 and 10, respectively. There are two plausible reasons for why so many women born just before the cutoff date start third grade at age 9 rather than at age 8. First, parents may feel that their children are still too young to start kindergarten. Second, women born just before the cutoff date tend to have much weaker school performance and are hence more likely to be held back a year in kindergarten, first, or second grade.
Using nationally representative data, Bedard and Dhuey (2006) find that the first reason accounts for 57.4 % of students born up to 30 days before state cutoff dates who enter third grade at ages 9 or older, while the second reason accounts for 42.6 % (in their sample, 41.4 % of students born before the cutoff date enter third grade at higher-than-expected ages, which is substantially higher than in this paper, possibly due to their narrower window and the inclusion of boys, who are more likely to be held back).
On the other hand, statistics on retention rates between kindergarten and third grade in North Carolina public schools suggest that the second reason is likely to be more important. According to the Kindergarten Readiness Issue Group (2003), the probability of being retained between kindergarten and third grade rose from around 9 % in 1991–1992 to around 17 % in 2001–2002, possibly reflecting the large increase in the proportion of students from Hispanic immigrant families during this period. Since only 11.6 % of women in this dataset start third grade at higher-than-expected ages, a majority of them are likely to have been retained at an early age.