Decomposing an effect with a categorical IV

[Laura Freds]

clear all
sysuse nlsw88
gen wagemedian = (wage > 6)
gen black = race == 2

The relationship between wage and black (versus not black) is given with the regression

reg wagemedian black

It seems that industry matters. There is a relationship between black and industry. There is also a relationship between wage and industry.

 reg black i.industry reg wagemedian i.industry

How can I isolate the effect black on wage in net of the sorting between black and industry?

[Fei Wang]

Laura, controlling for the industry dummies would partial out their effects - It's exactly what a multiple linear regression is used for.

reg wagemedian black i.industry

A more advanced strategy is to regress wagemedia on black after matching on industry. Or simply regress wagemedia on black with each industry's subsample and average all the industry-specific effects. A possible code would be

kmatch em black industry (wagemedian)

where kmatch is from SSC.