Skill Accumulation and Learning by Doing
A distinct form of physician learning occurs through learning by doing, whereby repeated practice improves clinical skill and decision quality over time. Unlike belief updating about treatment effectiveness, skill accumulation reflects changes in a physician’s underlying productivity or ability, often tied to procedural volume and experience. This mechanism has been used to explain persistent differences in outcomes across physicians and hospitals, as well as the concentration of complex procedures among high-volume providers.
The literature on skill accumulation emphasizes that experience can affect both treatment choices and patient outcomes. Early empirical work documents strong volume–outcome relationships and productivity spillovers, suggesting that experience generates improvements that extend beyond the individual patient encounter (Chandra and Staiger (2007)). More recent work embeds learning-by-doing directly into dynamic models of physician behavior, allowing experience to interact with experimentation and patient selection. For example, Gong (2018) develops a structural model in which physicians simultaneously learn about treatment effectiveness and accumulate procedural skill, showing how these forces jointly shape treatment decisions and technology diffusion.
An important insight from this literature is that experience does not always lead to efficient learning. Skill accumulation may be distorted by biased beliefs or imperfect feedback, particularly when physicians misperceive their own ability. Comin, Skinner, and Staiger (2022) highlights how overconfidence in perceived skill can drive excessive adoption of new technologies, and how learning about this bias over time can generate both rapid diffusion and subsequent abandonment. Together, these papers underscore that learning by doing is a powerful but imperfect mechanism, with implications for productivity, technology adoption, and patient outcomes.
Potential papers for presentation today include:
- Gong (2018) — joint learning by doing and Bayesian updating in treatment choice
- Comin, Skinner, and Staiger (2022) — skill misperception, learning, and technology adoption