Learning Under Uncertainty
An important feature of medical decision-making is that physicians often operate under substantial uncertainty about treatment effectiveness, side effects, and patient-specific responses. Much of clinical knowledge is acquired outside of formal training, through experience, feedback from outcomes, and exposure to new information. As a result, treatment decisions reflect not only preferences and incentives, but also evolving beliefs shaped by incomplete and noisy signals.
The literature on physician learning provides a framework for understanding how doctors update beliefs over time and how uncertainty influences treatment choice. Learning models formalize the idea that physicians face trade-offs between exploiting treatments they believe to be effective and experimenting with alternatives to gain information. These dynamics can generate persistence in practice styles, delayed adoption of new technologies, and heterogeneity in care even among observationally similar physicians.
We introduce this literature using Ching, Erdem, and Keane (2013), which reviews the development of empirical learning models in consumer and physician behavior over the past two decades. Building on foundational work such as Erdem and Keane (1996), the paper highlights how dynamic learning models have improved our understanding of decision-making under uncertainty, while also emphasizing key empirical challenges. In particular, it discusses the difficulty of distinguishing learning from other sources of persistence, such as habit formation or switching costs, and the importance of directly measuring information, beliefs, and expectations. This framework provides a common language for the remainder of the module, where we examine specific objects of learning and their implications for treatment decisions and technology adoption.
Potential papers for presentation today include:
- Ching, Erdem, and Keane (2013) — a synthesis of learning models and their empirical challenges
- Chan, Narasimhan, and Xie (2013) — a physician-focused model of learning about treatment effectiveness and side effects