The new @rubinreport book has a section on the gender wage gap.
It's a very tight and concise exposition, such that it has one the highest errors-to-sentences ratios I've ever observed.
Let's walk through it. 1/18
It's a very tight and concise exposition, such that it has one the highest errors-to-sentences ratios I've ever observed.
Let's walk through it. 1/18
Obviously the GWG has not been debunked by "countless" economists. I'm sure there are a couple of economists who have made claims that there is no wage gap, but I suspect that number is quite small. 2/18 
That a number is "aggregate" doesn't suggest that it is "pure spin". Also, I'm not sure where Rubin is pulling the 79% from, but usually the number is derived looking at full time workers (e.g. here): americanprogress.org/issues/women/r… 3/18
I assume this is referring to the CONSAD report shrm.org/hr-today/publi…. FWIW, Rubin's summary is inaccurate. CONSAD's fully saturated model finds a gap between 4.8% and 7.1%. 4/18
(FWIW, I find it odd how high this particular report is esteemed in the "GWG debunking community". Has anyone ever heard of CONSAD outside of this context?) 5/18
Here's the "meat" of Rubin's argument - that the differences between men and women's earnings disappear when you include controls for things like education, hours worked and major. 6/18
But this is based on a faulty conception of statistical controls. 7/18
Your major, the amount of hours you work, and how much education you pursue are all decisions you make on the basis of how they change your expected income (at least in part). 8/18
For example, if a woman's expected income is going to be less than a man's because of discrimination, you would expect her to change her labor supply and human capital decisions. 9/18
This is fairly basic economics - people respond to incentives.
If people are paid less, they might not work as hard or as long. They might invest less in things that would increase their earnings by a fixed percent. 10/18
If people are paid less, they might not work as hard or as long. They might invest less in things that would increase their earnings by a fixed percent. 10/18
This truism applies to women, being people. 11/18
@erinhengel has a wonderful demonstration of this in @causalinf's "Causal Inference" textbook. scunning.com/cunningham_mix… 12/18
Because people are making choices, controlling for occupations/hours worked/education can actually make your regressions *less* accurate 13/18
Rubin is right that statistical disparity doesn't *always* means discrimination. But he's not actually accounting for "life choices" - which would change in the presence of discrimination. 14/18
Fortunately, we do not have to rely on correlations on observational data in assessing whether there is discrimination against women in the labor market 15/18
There is a wealth of work using field experiments. Bertrand and Duflo have a paper summarizing the field: economics.mit.edu/files/11449 16/18
They find that: "this literature offers staggering evidence of pervasive discrimination against minority or
under-represented groups all around the world." 17/18
under-represented groups all around the world." 17/18
The claims @rubinreport makes about the gender wage gap in the book are based on misunderstandings of both economics and statistics.
He is making basic, textbook errors.
Going forward, he should check his claims with people who are well informed in both fields. 18/18
He is making basic, textbook errors.
Going forward, he should check his claims with people who are well informed in both fields. 18/18
Addendum: Meme version of the above argument:
