Histogram of marginal effect and partial effect for top choice specialists and for everyone a PCP has ever sent patients to (collapse data to one row for top choice specialist or a handful of rows for any non-zero history pair, which we then can collapse to the referral level).
10th, 20th, 50th, 70th, 90th percentiles. SE is observed SDev/sqrt(nrows)
Sum (p_j*spec_qual) = ex ante expected probability of success (collapse to referral level): Compute difference in baseline and counterfactual and show histogram of differences across referrals
Consider counterfactual again without congestion effects, as a way to assess role of congestion
Histogram of marginal effect and partial effect for top choice specialists and for everyone a PCP has ever sent patients to (collapse data to one row for top choice specialist or a handful of rows for any non-zero history pair, which we then can collapse to the referral level).
10th, 20th, 50th, 70th, 90th percentiles. SE is observed SDev/sqrt(nrows)
Sum (p_j*spec_qual) = ex ante expected probability of success (collapse to referral level): Compute difference in baseline and counterfactual and show histogram of differences across referrals
Consider counterfactual again without congestion effects, as a way to assess role of congestion