Other control variables have the expected sign and are consistent with the literature that uses
survival analysis to model welfare duration (e.g., Blank 1989 and Barrett 2000). Being younger,
African-American, never-married, without a high school diploma or GED, and having more children
at the start of the first welfare spell are all associated with longer duration of welfare receipt. Living
in a state with higher welfare benefits is also associated with longer duration. One variable that is
typically not included in the duration literature is the recipient's AFQT percentile ranking. This is
included because it proxies a taste for education and was shown to be associated with college
enrollment and graduation. The coefficient on AFQT percentile ranking is statistically significant in
all models, indicating that those with a higher AFQT ranking stay on aid for shorter durations. The
other proxy for taste for education, mother's highest grade, is not significantly related to time on aid.