I propose a model of skill-replacing routine-biased technological change (SR-RBTC). In this model, technology substitutes for the use of skill in routine tasks, in contrast to standard RBTC models, which assume that technology replaces the workers themselves. The SR-RBTC model explains three key trends that are inconsistent with standard RBTC models: (1) why specifically middle wages declined even though workers in routine occupations are dispersed across the entire bottom half of the wage distribution, (2) why middle wages stopped declining while technological change continued, and (3) why there is no substantial decline in the average wage of workers in routine occupations. I derive two new testable predictions from the model: a decrease in the return to skill and a decrease in skill level in routine occupations. I use an interactive fixed-effects model to confirm both predictions. Since SR-RBTC violates the ignorability assumption required by standard decomposition methods, I introduce a “skewness decomposition” to show that SR-RBTC is the main driver of bottom-half inequality trends.

This paper develops a method to estimate workers’ outside employment opportunities. We outline a matching model with two-sided heterogeneity, from which we derive a sufficient statistic, the “outside options index” (OOI), for the effect of outside options on earnings, holding worker productivity constant. The OOI uses the cross-sectional concentration of similar workers across job types to quantify workers’ outside options as a function of workers’ commuting costs, preferences, and skills. Using German micro-data, we find that differences in options explain 20% of the gender earnings gap, and that gender gaps in options are mostly due to differences in the implicit costs of commuting and moving.

The validity of instrumental variable (IV) designs is typically tested using two types of falsification tests. We characterize these tests as conditional independence tests between negative control variables—proxies for unobserved variables posing a threat to the identification—and the IV or the outcome. We describe the conditions variables must satisfy in order to serve as negative controls. We show these falsification tests examine not only independence and the exclusion restriction but also functional form assumptions. Our analysis reveals conventional applications of these tests may flag problems even in valid IV designs. We offer implementation guidance to address these issues.

Economists have become increasingly interested in studying the nature of production functions in social policy applications, with the goal of improving productivity. Traditionally models have assumed workers are homogenous inputs. However, in practice, substantial variability in productivity means the marginal productivity of labor depends substantially on which new workers are hired--which requires not an estimate of a causal effect, but rather a prediction. We demonstrate that there can be large social welfare gains from using machine learning tools to predict worker productivity, using data from two important applications - police hiring and teacher tenure decisions.