Appendix:

The primary analysis concerns the cohort-age-year-state employment rate and mean log real wages for college graduates. Employment rates are estimated from the monthly Current Population Survey (CPS). I construct a synthetic panel from data from each month from 1979 through 2019, aggregated to the calendar year. Log wages are computed from the 1979-2017 Merged Outgoing Rotation Group (ORG) files from the monthly CPS, using an algorithm adapted from Center for Economic and Policy Research (2018; see also Schmitt [2003]). For those paid hourly, I use the higher of the reported hourly wage (without overtime, tips, or commissions) and the ratio of weekly earnings (which in principle include overtime, tips, and commissions) to usual weekly hours. For non-hourly workers, I use the latter. When necessary, usual hours are imputed using actual hours last week or the mean by gender and full time/part time status. Hourly wages are converted to real 2015 dollars using the CPI, set to missing if below $1 or above $200, and logged.