THE ECONOMICS OF INSTITUTIONAL DISCRIMINATION: A PRIMER
Much has been written recently about the failure of economics to study institutional discrimination. But, I believe there are important contributions to improve upon. Here is a survey (with a bias for my own work -sorry) 1/
Much has been written recently about the failure of economics to study institutional discrimination. But, I believe there are important contributions to improve upon. Here is a survey (with a bias for my own work -sorry) 1/
In the last JEP, sociologists Small and Pager (pubs.aeaweb.org/doi/pdfplus/10…)
claim that economists do not study implicit biases and institutional discrimination. @WSpriggs in an open letter adds that economists should study race as a social construct bit.ly/37x0k1x 2/
claim that economists do not study implicit biases and institutional discrimination. @WSpriggs in an open letter adds that economists should study race as a social construct bit.ly/37x0k1x 2/
Economists have tools to study these issues and have done so. The critics focus on the old (but textbook-popular) notion of statistical discrimination modeling (exogenous) stereotypes (=group averages), which they dislike because it places the burden of change on the victims 3/
While (possibly biased) stereotypes are an important feature of human psychology and IMHO should be studied, the critics ignore the lesser-known literature studying stereotypes as equilibrium outcomes, originating not from Phelps (1972) but from an idea in Arrow (1973) 4/
Coate & Loury (AER, 1993) generate endogenous stereotypes as equilibria with self-fulfilling prophecies (with ex-ante identical groups). Inequality arises as a social norm and there is nothing any agent can do to change group averages. But, this may not satisfy the critics... 5/
...because in C&L groups are as if in separate islands, so it
is possible to "blame the victims", for their failure (as a group) to coordinate on the "good" equilibrium. Whether this can be viewed as a prototype of institutional discrimination is up to personal interpretation 6/
is possible to "blame the victims", for their failure (as a group) to coordinate on the "good" equilibrium. Whether this can be viewed as a prototype of institutional discrimination is up to personal interpretation 6/
But Moro and Norman (JET 2004) show that if the technology has complementarities there are equilibria where the dominant group has a vested interest to keep the other group subordinated. See Section 6.2 andreamoro.net/perm/papers/a_… 7/ 
Intuition: when blacks' increase their human capital, the marginal product of skilled whites decreases. So I think this is closer to the notion of institutional discrimination: not only hard to break social norms, but with incentives not to. Also, small groups fare worse. 8/
If you dislike stereotypes Mailath, Shaked, and Samuelson AER 2000) derive similar results in a model where search frictions (not incomplete information) are the driver of the
inequality. Lang (QJE 1986) does it with language frictions. 9/
inequality. Lang (QJE 1986) does it with language frictions. 9/
What about race as a social construct? Fang (AER 2001) extends the model allowing for "endogenous" group formation. He calls it cultural activity but it could be an endogenous race (I guess the world wasn't ready for that interpretation) 10/
What about the historical legacy of past discrimination?
Several papers attempt to model these outcomes in a dynamic setting, showing how current stereotypes can arise from past injustice and then persist (Blume, EJ 2006, Fryer JPUBE 2007, and other more recent ones). 11/
Several papers attempt to model these outcomes in a dynamic setting, showing how current stereotypes can arise from past injustice and then persist (Blume, EJ 2006, Fryer JPUBE 2007, and other more recent ones). 11/
But how important are these factors empirically? Empirics about this is very hard and I feel the recent complaints about the use of race dummies fail to appreciate the difficulty of studying the complexity of these questions. 12/
A few papers have attempted to estimate these models to address various questions, showing that, in principle, it can be done: Moro (IER 2003), Fang(IER 2006), Antonovics (2004). Most of you won't be satisfied (I am not either) but better people could improve! 13/
Conclusion: I welcome the push to direct research towards preferred goals, but accusing scholars of bias or worse is not the best approach, especially if you ignore or misinterpret parts of the literature. Nor is accusing those using race dummies of bias as I see on twitter 14/
A race dummy has no bias - it may not measure what you think it measures, but the interpretation is there in the equation for you to see. You can't make up your own interpretation and then accuse the econometrician. 15/
Nor it is good to count papers with the word "racism" to conclude that the profession suffers from white supremacy. The word racism isn't used much because it's very hard to measure. And so is discrimination: there is lots of work and there is much more to do! Do it! 16/
Refs: for a brief intro to the theory and its different strands, my Palgrave dictionary entry "Statistical Discrimination". For a more comprehensive survey, my article with Hanming Fang on the Handbook of Social Econ. All on my webpage. Tag your work if I forgot to include 17/17
Short follow-up after a brief conversation with @SandyDarity that clarified something on my mind. This thread may confirm the critics' fears that the econ lit is biased towards certain explanations. They may be right..but we need an honest discussion that includes all we have +
My point was to a) show that there is the possibility, in some models, of finding a common basis with the models they have in mind... e.g. his stratification model, of which a formalization could in principle help with empirical work +
b) show that someone has thought about these issues, I hope in a way that partly addresses their concerns against the classic (phelps-style) statistical discrimination model (I left out the small reduced-form literature on this: I agree results are thin) +
and finally c) I admitted that the empirical evidence regarding the models I talk about is very thin, mostly methodological, which to me means there's room for somebody smart to do it. I don't know if it goes anywhere before anybody tries. /end
