Notes: This table reports the estimated relationship between antibiotic prescribing as predicted by patient observable characteristics and the difference in average antibiotic prescribing between post- and pre-exit physicians for treated patients. A strong relationship would suggest that patients choose post-exit physicians based on observable characteristics that are predictive of antibiotic prescribing. We proceed in two steps. First, we predict post-treatment antibiotic prescribing to treated patients based on basic demographics, health, family background, education (as in Table 2), and including age-squared. We use a linear prediction model trained on data from never-treated patients and treated patients prior to the physician exit. Second, we regress predicted prescribing on the difference in average prescribing between post- and pre-exit physicians assigned to treated patients. The second-step regressions include calendar year fixed effects and fixed effects for the pre-exit physicians, with observations on the patient–year level. Standard errors are calculated using a bootstrap with 50 repetitions at the patient level. *p < 0.10, **p < 0.05, ***p < 0.01.

↵^aΔ_i denotes the difference in mean prescribing between patient i’s assigned sets of physicians and is estimated by , the average prescribing to untreated patients.