Table 1: Estimates of Learning By Doing (1) (2) Panel A. Weekly Data Estimated learning rate, B -0.289* -0.335* (0.007) (0.017) Time trend 0.007* (0.002) Observations 47 47 R2 0.961 0.969 Panel B. Daily Data Estimated learning rate, B -0.306* -0.369* (0.006) (0.014) Time trend 0.001* (0.0002) Observations 224 224 R2 0.931 0.943 Notes: Column (1) in both panels shows estimation results for In (D,) = a + B In(E,) + 8, , where D, is the average defects per car in time period t and E, is production experience up to that point: cumulative number of cars produced before the current period. Column (2) in both panels shows estimation results for In(D, ) = a + B In(Et ) + y * t + Et. Heteroskedasticity-robust standard errors are in parentheses. * Significant at the 5 percent level. Table 1 shows the results of estimating these specifications with our sample. Panel A shows results from specifications using weekly data (average defect rates over the week and production experience at the week's outset); Panel B shows results using daily observations." (p. 655)4.5 + 3.5 Natural log of average defects per car this week LO 6 8 10 12 Natural log of cumulative production of cars in previous weeks Figure 2: Like Figure 2 on p. 656 of Levitt et al. (2013), except that it shows the weekly data (instead of the daily data). This figure corresponds exactly to Table 1, Panel A, Column (1)