Identifying Physician Peer-to-Peer Effects Using Patient Movement Data

In this paper, we identify and quantify the peer-to-peer effects using physician prescription data and patient movement data between physicians. We categorize the movements into three types: 1) primary care physician (PCP) to specialist and back; 2) specialist to specialist; and 3) PCP to PCP. In-depth physician interviews and surveys reveal different reasons for these movements: PCP to PCP is purely patient generated; PCP to specialist is mostly physician generated; and specialist to specialist is a mix of patient and physician generated movements. We estimate a simultaneous equations model on these three types and find that on purely patient generated movement samples (PCP to PCP) the physicians have a significantly negative effect on each other’s prescription behavior due to observational learning and congestion effects. In contrast, on the PCP to specialist sample and specialist to PCP sample, we find that the specialist has a significantly positive effect on the PCP, but not vice versa. This suggests an opinion leader effect. The specialist to specialist is a mixed case and the effect is insignificant in most cases. Based on model estimates, we calculate the social multiplier to quantify the effect of opinion leaders on other physicians in the sample. We find focal specialists who are high prescribers are more likely to be opinion leaders