Productivity, place, and plants : revisiting the measurement

Why do cities differ so much in productivity? We document that most of the measured dispersion in productivity across US cities is spurious and reflects granularity bias: idiosyncratic heterogeneity in plant-level productivity and size, combined with finite plant counts. As a result, economies with randomly reallocated plants exhibit nearly as high a variance as the empirical economy. Stripping out this bias using our nonparametric split-sample strategy reduces the raw variance of place effects by about two thirds to three quarters. For new plants, about four fifths of the dispersion reflects granularity bias, and new plants' place effects are only imperfectly correlated with those of older plants. These US-based patterns broadly extend to the 15 European countries we study in internationally comparable firm-level data.