Is Rising Returns to Scale a Figment of Poor Data?

While using detailed firm-level data from the private business sector, this study identifies two empirical puzzles: (i) returns-to-scale (RTS) parameter estimates rise at higher levels of data aggregation, and (ii) estimates from the firm level suggest decreasing returns to scale. The analysis shows that, although consistent with rising estimates, the Basu-Fernald (1997) aggregation-bias effect does not drive this result. Rather, rising and too low returns-to-scale estimates probably reflect a mixture of random errors in factor inputs. It turns out, in fact, that a 7.5-10 percent error in labor (hours worked) can explain both puzzles.