"Revisiting U.S. Wage Inequality at the Bottom 50%" Oren Danieli, Tel Aviv School of Economics

Sep 7, 1:20 pm2:35 pm
Event Description

Oren Danieli, assistant professor at Tel Aviv University School of Economics, is a visiting research scholar in the Section during the 2020-2021 academic year.


While inequality at the top half of the wage distribution has been rising steadily since 1980, inequality at the bottom of the distribution has been unstable: it increased in the early 1980s, decreased in the 1990s, and then moderately increased since the 2000s. Several papers have argued that these trends are the result of a routine-biased-technological-change (RBTC). Models of RBTC predict a decline in middle-wages (“Wage Polarization”) which generates a decline in bottom-half inequality as occurred during the 1990s. However, these models cannot explain why inequality at the bottom 50% resumed growing. They also do not explain why specifically middle-wages declined when routine workers are dispersed across the entire bottom half of the wage distribution. Earlier decomposition exercises argued that RBTC cannot explain these wage trends in full.

I show that a small-yet-important refinement to the RBTC model can resolve all these puzzles. Instead of assuming technology replaces workers, I assume it replaces their skill. At first, skill-replacing RBTC (SR-RBTC) lowers wages for middle-wage workers since they have the highest skill among routine workers. Middle-wage workers then leave routine occupations. When SR-RBTC continues it reduces wages for the remaining routine-workers who are mostly low-wage, and inequality at the bottom 50% resumes growing. I use two non-standard empirical methods to test this model. Using an interactive-fixed-effect-model I find the return to skill sharply declined in routine occupations and the composition of routine workers became less skilled. Using “Skewness Decomposition” I find that the drop in inequality at routine occupations is the main driver of wage polarization. This was not captured with other decomposition methods as it violates the ignorability assumption that underlies them.