Additional control variables are program-related measures: eligibility for matched withdrawals during
the period of data collection, length of participation in SEED, and length of participation squared
(which is included to consider the nonlinear relationship between length of participation and the
dependent variables).
We run additional models to check the robustness of our findings: (1) a model based on an
alternative outlier diagnosis tool, dfits (Belsley, Kuh, & Welsch, 1980), (2) a model including an
indicator of whether a participant lives independently without a caregiver, and (3) a model that uses
a categorical instead of continuous measure of age. We also run the basic model with different
sample compositions: (1) a sample excluding those who dropped out of the program, and (2) a
sample including only participants under 18 years old. Results from these additional analyses are not
substantially different from those reported in this study.3