Garbage in, garbage out at Heritage and AEI?

Jason Richwine of the Heritage Foundation and Andrew Biggs of the American Enterprise Institute are at it again (following up on an earlier study for the Business Roundtable), claiming that government workers—in this case teachers—are grossly overpaid. EPI  and others have expended much ink on this topic, and forthcoming EPI research will address some of the latest claims in greater detail (though maybe Jon Stewart said it all in his message to teachers about “the greed that led you into the teaching profession”).

But one of the key arguments Richwine and Biggs make is so sloppy, it should only take a blog post to rebut: the claim that “teachers exhibit low cognitive abilities compared to other college graduates” and that once you take this into account teachers suffer no wage penalty. Since all employers would love to be able to accurately assess the skills of prospective employees, it’s amazing that such a tool, if it exists, isn’t in widespread use. The miracle tool turns out to be the Armed Forces Qualification Test, which Richwine and Biggs refer to as an IQ test. Here’s what the AFQT actually tests:

  • general science
  • arithmetic reasoning
  • word knowledge
  • paragraph comprehension
  • numerical operations
  • coding speed
  • auto and shop information
  • mathematics knowledge
  • mechanical comprehension
  • electronics information

Is it really surprising that a future kindergarten or high school history teacher would score lower on this test than a future engineer or army officer?  There are many other issues one can raise about the AFQT score, but that will have to wait for a later time.

But even if the AFQT score contained important information about teaching ability, Richwine and Biggs aren’t content to add this measure to their statistical model to explain wages as economists normally do.

Teacher wage regressions with education and Armed Forces Qualification Test (AFQT)
Key controls included Teacher wage effect (%) R-Squared
Row Education AFQT
1 Yes No -12.6* 0.31
2 Yes Yes -10.2* 0.33
3 No Yes 0.6 0.29
*Significant at 95 percent level

That’s because adding this variable doesn’t change the basic story, which is that teachers’ earnings are significantly lower than those of similar college grads, even those with the same AFQT scores.

See the results in their table. In regressions with the traditional specification (i.e., the variables included as controls) they find teachers earn 12.6 percent less than comparable workers (see row 1). In their next specification, they add the AFQT score, thus controlling for comparable education and AFQT score (which they mistakenly refer to as IQ). Their results show that teachers earn 10.7 percent less than other workers with comparable education and AFQT scores. That means that including the AFQT score seems to reduce the teacher penalty (actually, they do not provide the statistical information to judge whether there is a statistically significant difference between these two estimates) but in no way eliminates it. So, how do Richwine and Biggs reach the conclusion that there is no teacher wage penalty? They say:

“The wage gap between teachers and non-teachers disappears when both groups are matched on an objective measure of cognitive ability rather than on years of education.”

Richwine and Biggs take this as their most important bottom-line finding and it is based on a regression, row 3, with no control for education. This is JUNK science plain and simple. If you asked any labor market economist if they could have only one predictor of wages available to them, the overwhelming choice would be to use the education level of a worker. Ask yourself, do you expect two people with the same AFQT score to earn the same amount if one has a college degree and the other has not completed high school? If not, then one needs to control for education level. That is, there is every theoretical/conceptual reason why education should be included in these wage regressions and there is no basis for excluding it just because you include another variable representing a test score. There certainly was not any empirical test offered, such as showing that education was not statistically significant once you included the AFQT score. Richwine and Biggs do not present the basic details of their regressions, such as the coefficients and standard error for each of the variables, but it is almost certainly the case that the education controls in row 2 are economically and statistically significant in a regression that also includes the AFQT measure.

Their claim that the teaching wage penalty is zero should be discounted completely. Their “evidence” only shows that teachers do not make more, or less, than others with the same test scores when the “others” being compared to have much lower education (since teachers have much higher education than the average worker). That’s not much of a compliment to the wages teachers earn. This exercise by Richwine and Biggs is nothing more than generating a result you wish to find even though you violate basic economic thinking and avoid the empirical testing (as in the removal of the education controls) that is the norm in professional analysis.

Check out EPI research on the teacher pay penalty and the updated analysis and watch this space for an upcoming blog on teacher benefits, which Richwine and Biggs claim are worth as much as teacher salaries. In the meantime, you may want to read this DailyKos blog from a teacher inviting Richwine and Biggs to join him in the public schools. We can give Richwine and Biggs a pass on the value of their research if they want to enjoy these lavish perks themselves.