Sources
What the adjustment is doing
The headline "women earn X% of what men earn" is a ratio of medians across the whole workforce. Both sides of that ratio pool together very different jobs, hours, and experience levels. This page holds occupation fixed — same job title on each row — and shows two numbers side by side.
Statistically adjusted ratio. The same comparison after removing differences in the following observable characteristics between the men and women in that occupation:
Δ (adjusted − raw) is what the controls change. A large positive Δ means women in that occupation have observables that would predict lower wages (less potential experience, fewer hours), so removing those effects raises the ratio. A negative Δ means the opposite — women in that occupation are more selected on the covariates than men are.
By occupation
Loading…
Numbers
Raw JSON: salaries.json
Method
Sample: full-time (usual hours ≥ 35) wage-and-salary workers aged 25–64, non-agricultural, non-self-employed, from the Current Population Survey Merged Outgoing Rotation Groups (NBER extract), pooled across two years. Hourly wage is usual weekly earnings divided by usual hours. Allocated (imputed) earnings are dropped following Hirsch & Schumacher (2004). Observations are weighted with the CPS earnings weight.
For each occupation, the adjusted ratio is exp(β̂female) from a weighted least-squares regression of log hourly wage on the female indicator plus the listed controls. Confidence intervals in the table are 95% Wald intervals on the exponentiated coefficient.
The CPS weekly-earnings variable is top-coded (censored above a year-specific cap). The % top-coded column reports the share of each occupation's observations at the cap; where it is meaningfully large, both ratios understate the true dispersion at the high end.