Emil O W Kirkegaard

Standard disclaimer: I'm a journalism major who likes horsing around with R. A Real Social Scientist(TM) would probably set a lot of these up differently, don't take them as gospel.

Anyhow, my usual approach was to control the interaction of gender & age (polynomial) for one regression, and then add education to for the second. Upshot: Birthplace gaps usually a lot smaller after you control education, though some do still exist.

Regardless, HUGE effects for education in everything I looked at, even controlling birthplace. I think we should let people prove merit in other ways too bc too many people go to college, but that's an obvious way to improve immigrant outcomes.

Anyway, the results tables are too big to tweet, so I just dumped 'em into a Google Doc along with the code for the regression formulas. Comments/method suggestions welcome! https://docs.google.com/document/d/1lF4J5XmyZXVaqr90yybOEzdJ_xUSkMfPsHBOTpOIIrc/edit?usp=sharing …

My data only include immigrants, so an interesting question is which birthplace to use as the "reference." I went with Mexico because it's the biggest source -- but it's also an outlier on some metrics (eg very high employment rates for males and low income among employed)

Another tricky thing is that there aren't a ton of observations for many birthplaces, so you have to watch out for huge standard errors and just ignore those.

One possible improvement: I wish I'd had more time to mess around with the age control, both trying different polynomial degrees to see what made a difference and also experimenting with splines instead (which @KirkegaardEmil has suggested to me before).

And one last note: Yeah, I know the variable I coded as "Country" includes some birthplaces that are continents or other types of regions too. Sue me.