Working Papers

"Physicians as Persuaders: Evidence from Hospitals in China"

Abstract: I apply a Bayesian persuasion framework to study physician-patient interaction in medical treatment decisions. This approach generates new policy-relevant insights by explicitly modeling how patients incorporate partially self-serving, though more informed, physicians’ recommendations into their treatment decisions. I estimate the model using a large claims database from China, exploiting independent sources of random variation in physicians’ and patients’ financial incentives. For a diagnosis where surgical treatment is somewhat discretionary, physicians respond to changes in reimbursement more strongly when patients have more generous insurance coverage. Moreover, nearly half of the patients who received surgery would not have done so were they fully informed. Such misdirection from physicians is greater when patients are better insured, indicating a new channel of moral hazard: Besides the well-documented channel of reduced patient price sensitivity, lowering patient coinsurance rate could potentially increase healthcare use through the additional channel of increased susceptibility to physicians’ misdirection. 

"Physician Learning and Defensive Medicine" (with Hanming Fang, and Ming Li) 

Due to concerns about potential medical malpractice litigation, physicians engage in defensive medicine. But how do their perceptions of the risk of medical malpractice litigation change when they experience one?  We combine a large claims database from China with the universe of malpractice litigations to provide new evidence of defensive medicine: First, physicians' mental suffering alone, with little financial loss concern, can cause them to behave more conservatively after an adverse event, ultimately negatively impacting patient health outcomes. This finding holds implications for tort reforms and malpractice insurance. Second, the effect of malpractice litigation spills over to other departments within the same hospital. Third, such an effect is short-lasting, steadily rebounding in four weeks, consistent with the notion that physicians overreact to salient malpractice risk, although the actual risk may not change.

We show that these facts are consistent with a Bayesian learning model where physicians have short memory and put more weight on more recent events. We further find novel evidence supporting this interpretation: hospitals or specialties with many lawsuits respond similarly to those with fewer lawsuits. In addition, if a patient's death occurs shortly after the physician experiences a lawsuit, the physician responds by behaving less conservatively, despite becoming more conservative in face of a patient's death otherwise. 

"Welfare Cost of Adverse Selection in the U.S. Consumer Credit Market" 

Abstract: I test for the existence of adverse selection and quantify the resulting inefficiency in the U.S. consumer credit card market. By analyzing unique data from a randomized balance transfer market experiment, I show that conditional on price, higher risk types, both in terms of observable and hidden default risk, are more willing to take up an offer. Furthermore, conditional on taking up an offer, higher risk types are more willing to accept an offer with a higher price. Adverse selection is more severe in the group with a lower FICO score range. By leveraging the exogenous interest rate variations in the market experiment, I estimate an average welfare loss of at least $12 per U.S. credit card holder. In addition, the FICO score captures 90 percent of the welfare loss due to adverse selection.

Publications

Price dispersion in an online health marketplace: Evidence from telemedicine services in China (with Terence C. Cheng, Hongqiao Fu and Winnie Yip). Economics Letters, 2024. [pdf] [Published version]