where Povratejt is the poverty rate for all persons under age 65 in division j in year t.
Following Figures 2–4, we control for macroeconomic cycles with the unemployment rate, uratejt and use the real median weekly wage ln(medwagejt) to control for overall income and growth in the economy. As above, our measure of inequality is the ratio of the median weekly wage to the 20th percentile of the weekly wage,ln ( p50/p20)jt.
Our growth and inequality measures are both specified in logs, and weekly wages are constructed by dividing annual earnings by weeks worked.
The model also controls for division fixed effects j and year fixed effects t.
This effectively purges our estimates from omitted variables bias resulting from variables common to all regions that are changing over time (such as changing rates of female headship) or fixed differences across geographic areas (such as differences in immigrant shares) that might also influence the poverty rate.