impacts based on the DCR model with time-constant coefficients when the unobserved
heterogeneity distribution has one, two and three mass points, respectively. Rows 4
through 6 present the estimated impacts based on the DCR model when the coefficient of
the duration until enrollment variable is allowed to vary over time according to a third-
order polynomial and the unobserved heterogeneity distribution has one, two and three
mass points, respectively. Comparing the one mass point estimates to the two and three
mass point estimates indicates that the estimated impact of delay does not change
substantially when the potential endogeneity of delay is accounted for. Comparing rows 1
through 3 with rows 4 through 6 shows that the estimated impacts of enrollment delay
are somewhat larger in models that allow the coefficient of the duration until enrollment
variable to be time-varying.
To account for potential selectivity into two-year versus four-year institutions
due to the unobserved determinants of the two-year, four-year choice being correlated
with the unobservable determinants of completing a four-year degree, Tables 5 reports
estimates derived from the CR2 model described by equations (1') and (2) above.13 The
model estimates in Table 5 are based on the CR2 model where the coefficients of all
13The model was estimated twice. In the first estimation, those who first enter four-year institutions are
censored in the enrollment duration and in the second estimation, those who first enter two-year
institutions are censored in the enrollment duration. In small samples, the two sets of estimates obtained
for the coefficients in (1') are not necessarily equal. For our data, the two sets of coefficients estimates for
(1') do not qualitatively differ. Thus, Tables 4 and 5 present the coefficient estimates for (1') only in the
case where those who enter two-year institutions are censored in the enrollment duration.
21