The Effects of the Kalamazoo Promise Scholarship on College Enrollment and Completion

Source: Authors’ calculations from KPS and Kalamazoo Promise administrative data.

Notes: Authors’ calculations of characteristics of KPS graduates for classes of 2003–2013 (excluding alternative education). Eligibility calculated based on Promise rules. “Before” represents the cohorts 2003–2005; “After” represents cohorts 2006 through 2013. DD represents the difference between eligibles after and before the Promise and ineligibles after and before the Promise. Standard errors robust to heteroskedasticity are in parentheses.

↵Notes: Standard errors robust to heteroskedasticity are in brackets. *p < 0.10, **p < 0.05, ***p < 0.01. Outcome timing is since high school graduation. Regressions include dummies for after the Promise, individual (pseudo-)eligibility, sex, race/ethnicity, free/reduced-price lunch status, and high school of graduation- by-graduation year. In Column 1, prior to 2006, eligibility is based on Promise rules had the Promise been in effect (rules-based eligibility). After 2006, eligibility is taken from administrative records from the Promise (observed eligibility). Column 2 always uses rules-based eligibility. Column 3 rescales the Column 2 coefficient by the first-stage coefficient. This first-stage coefficient is obtained, for the post-Promise period, by regressing observed eligibility on rules-based eligibility (and the same covariates as above) and is equal to 0.694; see Appendix Table A.5, Column 1. The first-stage F-test statistic is 1,266. The mean of the dependent variable is for eligible population in the pre-Promise period. The p-values are from a Hausman test of equality of the OLS and IV estimate. Sample size is 5,415.

↵Notes: Standard errors robust to heteroskedasticity are in brackets. *p < 0.10, **p < 0.05, ***p < 0.01. Outcome timing is since high school graduation. Regressions include dummies for after Promise, individual (pseudo-) eligibility, sex, race/ethnicity, free/reduced-price lunch status, and high school of graduation-by-graduation year. In Column 1, prior to 2006, eligibility is based on Promise rules had the Promise been in effect at the time (rules-based eligibility). After 2006, eligibility is taken from Promise administrative records (observed eligibility). Column 2 always uses rules-based eligibility. Column 3 rescales Column 2 by the first-stage coefficient. This first-stage coefficient is obtained, for the post-Promise period, by regressing observed eligibility on rules-based eligibility (and the same covariates as above) and is equal to 0.694; see Appendix Table A.6, Column 1. The first-stage F-test statistic is 1,266. The mean of the dependent variable is for eligible population in the pre-Promise period. The p-values are from a Hausman test of equality of the OLS and IV estimate. Sample size is 5,415.

↵Notes: Each cell represents a point estimate and the standard error from a separate regression. Standard errors robust to heteroskedasticity are in brackets, *p < 0.10, **p < 0.05, ***p < 0.01. In row labeled OLS, prior to 2006, eligibility is based on Promise rules had the Promise been in effect at the time (rules-based eligibility). After 2006, eligibility is taken from Promise administrative records (observed eligibility). Row labeled RF always uses rules-based eligibility. Row labeled IV rescales the RF estimate by the first-stage coefficient. This first-stage coefficient is obtained, for the post-Promise period, by regressing observed eligibility on rules-based eligibility (and the same covariates as above) and is equal to 0.694; see Appendix Table A.6, Column 1. The first-stage F-test statistic is 1,266. The mean of the dependent variable is for eligible population in the pre-Promise period. The p-values are from a Hausman test of equality of the OLS and IVestimate. Sample size is 5,415. KVCC stands for Kalamazoo Valley Community College, WMU stands for Western Michigan University, MSU stands for Michigan State University, UM stands for University of Michigan–Ann Arbor, Flagships stands for either MSU or UM, and K stands for Kalamazoo College.

↵Notes: Standard errors robust to heteroskedasticity are in brackets, *p < 0.10, **p < 0.05, ***p < 0.01. p-value is obtained using a placebo-regression permutation inference described in the text. Regressions include district-by-year proportions of students to teachers, students eligible for subsidized lunch, African-American students, and Hispanic students. For each observable, we also include the proportions of missings. The regressions control for district fixed effects, year-of-graduation time effects, and districtspecific linear time trends. Observations are weighted by the number of district graduates. The mean of the dependent variable is for the control districts in the pre-Promise period. The control districts consist of the Michigan Middle Cities Education Association (MCEA) districts described in the text.

↵Notes: Standard errors robust to heteroskedasticity are inbrackets. *p < 0.10, **p < 0.05, ***p < 0.01. Timing is since high school graduation. All regressions include dummies for after the Promise’s introduction, individual (pseudo-)eligibility, and graduation year. Other controls are sex, race/ethnicity, free/reduced-price lunch status, and high school of graduation- by-graduation year (except when subgroup is restricted on one of these dimensions). The estimates are calculated using rules-based eligibility with the coefficient of interest rescaled by the first-stage coefficient. This first-stage coefficient is obtained, for the post-Promise period, by regressing observed eligibility on rules-based eligibility (and the same controls as listed above). The F-test statistic is from the first stage. The income groupings pertain to whether the student is eligible for free/reduced price lunch or not. The race groups are white non-Hispanic versus other groups. The mean of the dependent variable for each group is calculated over the eligible population in the pre-Promise period.

↵Notes: Standard errors robust to heteroskedasticity are in brackets. *p < 0.10, **p < 0.05, ***p < 0.01. Outcome timing is since high school graduation. Regressions include dummies for after the Promise, individual (pseudo-)eligibility, sex, race/ethnicity, free/reduced-price lunch status, and high school of graduation-by-graduation year. The mean of the dependent variable is for eligibles in the pre-Promise period. In Column 1, prior to 2006, eligibility is based on Promise rules had the Promise been in effect at the time (rules-based eligibility). After 2006, eligibility is from Promise administrative records (observed eligibility). Column 2 always uses rules-based eligibility. Column 3 rescales Column 2 by the first-stage coefficient. This first-stage coefficient is obtained, for the post-Promise period, by regressing observed eligibility on rules-based eligibility (and the same covariates as above). The first-stage coefficient is 0.685 at the four-year horizon and 0.707 at the six-year horizon; see Appendix Table A.6, Columns 4 and 5. The first-stage F-test statistic is 819 at four years and 555 at six years. The p-values are from a Hausman test of equality of the OLS and IV estimate. Sample sizes are 3,869 at four years and 2,905 at six years.

↵Notes: Standard errors robust to heteroskedasticity are in brackets. *p < 0.10, **p < 0.05, ***p < 0.01. p-value is obtained using a placebo-regression permutation inference described in the text. Regressions include district-by-year proportions of students to teachers, students eligible for subsidized lunch, African-American students, and Hispanic students. For each observable, we also include the proportions of missings. The regressions control for district fixed effects, year-of-graduation time effects, and district-specific linear time trends. Observations are weighted by the number of district graduates. The mean of the dependent variable is for the control districts in the pre-Promise period. The control districts consist of the Michigan Middle Cities Education Association (MCEA) districts described in the text.

↵Notes: Standard errors robust to heteroskedasticity are in brackets. *p < 0.10, **p < 0.05, ***p < 0.01. The category of college is based on the first college attended within the first 12 months after high school graduation. “Most selective” denotes the most competitive colleges, none of which are Promise-eligible. “Highly selective” includes University of Michigan–Ann Arbor and Kalamazoo College, “very selective” includes Michigan State University, and “selective” includes Western Michigan University. See the note to Table 3.

↵Notes: The coefficient estimates represent the difference between eligible and ineligibles for the indicated year. Standard errors robust to heteroskedasticity are in brackets. *p < 0.10, **p < 0.05, ***p < 0.01. Outcome timing is since high school graduation. Regressions include dummies for after the Promise, individual (pseudo-)eligibility, sex, race/ethnicity, free/reduced-price lunch status, and high school of graduation-by-graduation year. The regressions use rules-based eligibility with the coefficient of interest rescaled by the first- stage coefficient. This first-stage coefficient is obtained, for the post-Promise period, by regressing observed eligibility on rules-based eligibility (and the same covariates as above); see Table A.5. The F-test statistic is from the first stage.

↵Notes: The table compares the baseline standard errors estimates (in parentheses) to standard errors clustered by high school × cohort (in brackets). All regressions include dummies for after the Promise’s introduction, individual (pseudo-)eligibility, and graduation year. Other controls are sex, race/ethnicity, free/reduced-price lunch status, and high school of graduation-by-graduation year.

Notes: The table compares the IV baseline estimates (Panel A) to IV estimates obtained without controlling for demographics (Panel B). Standard errors are in brackets. *p < 0.10, **p < 0.05, ***p < 0.01 All regressions include dummies for after the Promise’s introduction, individual (pseudo-)eligibility, and graduation year.

↵Notes: The table compares the baseline estimates (Panel A) to estimates of samples restricted to students enrolled in KPS at the time of Promise announcement and to the classes of 2003–2008 (Panel B). (All ineligible students from the classes of 2009 or later either had interrupted KPS enrollment or entered the district after the Promise was announced.) Standard errors are in brackets. *p < 0.10, **p < 0.05, ***p < 0.01. All regressions include dummies for after the Promise’s introduction, individual (pseudo-)eligibility, and graduation year. Other controls are sex, race/ethnicity, free/reduced-price lunch status, and high school of graduation-by-graduation year.

↵Notes: Regressions are estimated for the post-Promise period, that is after 2005, and control for year dummies. The differences in sample sizes are due to different timing since high school graduation. Standard errors robust to heteroskedasticity are in brackets. *p < 0.10, **p < 0.05, ***p < 0.01. The F-test statistic is from the first stage.