Closed manncz closed 7 months ago
2e65fdd fixes the issue, so for each pair, only one alpha is calculated, which we use to interpolate between the two estimation methods for both v1 and v2. Therefore, the interp methods no longer depend on the ordering of the data. We tested that giving the same dataset, in different orders of units within each pair would return the same estimates.
The
p_interp
(p_ols_interp
andp_rf_interp
) depended on the ordering of the data given (aka which unit was labeled "1" or "2"). Therefore, if data was fed in, that had all of the treated units first and all of the control units after, p_loop returned "NA" for the treatment effect.