So, as Mankiw argued in his first post, omitted variable bias does seem to explain the relationship between income and test scores.
MIT's David Cesarini sends me the above chart with this note:Dear Professor Mankiw,
I prepared a graph which I think nicely illustrates the simple point you made in your blogpost on the spurious association between SES and test scores. The graph is attached. The dataset is comprised of a large sample of men born in Sweden between 1955 and 1970 who took an IQ test at conscription, at the age of 18. Income is measured as the biological father's income in the 1970 census.
The red line is the average measured IQ of the non-adoptees, plotted against the biological father's income decile. The blue line shows the same relationship for adoptees and their biological fathers. The patterns are remarkably similar,even though the biological fathers of adoptees did not raise these children.The fact that the biological father's income is almost an equally strong predictor of a child's test scores even when the biological father was not present in the household clearly suggests that most of the association between income and test scores does indeed arise because of omitted variable bias. Of course, an important caveat here is that it is quite likely that non-random assignment of adoptees may explain some of the similarity between the two lines.
I hope you find this useful.
Wednesday, 2 September 2009
Test Scores and biological father's income
Earlier I noted The least surprising correlation of all time. This is a graph from Greg Mankiw's blog showing that kids from higher income families get higher average SAT scores. Now Mankiw has blogged again on this topic. He writes,