Fig. 4.

Treatment effect estimates by survey mode.

Notes: Coefficient estimates for treatment effect according to (1) by crop and by survey mode. Error bars represent true 95% confidence intervals for estimation using phone survey data. For estimation using in-person survey data, 95% confidence intervals are shrunk by $\sqrt{N^{InPerson}/N^{Phone}}$ to represent the hypothetical confidence interval had the in-person survey had the same number of respondents as the phone survey. Top production values are Winsorized to the 95th percentile independently by crop and by mode before regression estimation.