Clayton R. Featherstone

Assistant Professor of Business Economics and Public Policy
The Wharton School of the University of Pennsylvania





Boston Versus Deferred Acceptance in an Interim Setting: An Experimental Investigation
with Muriel Niederle
Games and Economic Behavior, 2016, Vol. 100: 353–375.

Can Social Information Affect What Job You Choose and Keep?
(Online Appendix)
with Lucas C. Coffman and Judd B. Kessler
American Economic Journal: Applied Economics, 2017, Vol. 9 (1) : 96–117.

What Matters for the Productivity of Kidney Exchange?
with Nikhil Agarwal, Itai Ashlagi, Eduardo Azevedo, and Ömer Karaduman
AEA Papers and Proceedings, 2018, Vol. 108: 334–340.

Liquidity Affects Job Choice: Evidence from Teach For America
(Online Appendix)
with Lucas C. Coffman, John J. Conlon, and Judd B. Kessler
Quarterly Journal of Economics, 2019, Vol. 134 (4) : 2203–2236.

finalist for the 2020 exeter prize ( for Research in Experimental Economics, Decision Theory and Behavioral Economics)

Market Failure in Kidney Exchange
(Online Appendix)
with Nikhil Agarwal, Itai Ashlagi, Eduardo Azevedo, and Ömer Karaduman
American Economic Review, 2019, Vol. 109 (11) : 4026–4070.

Working papers

Rank Efficiency: Modeling a Common Policymaker Objective

abstract: Policymakers often gauge matches using rank distributions—how many get their reported 1st choice, 2nd choice, etc. Formalizing this observation, I call an assignment rank efficient if its rank distribution cannot be stochastically dominated. Rank efficiency refines ordinal efficiency, and hence ex post (Pareto) efficiency. In addition, a class of linear-programming mechanisms seen in the field guarantee rank-efficient assignments. Policymakers can also attain rank efficiency simply by looking for local improvements that increase a natural objective. In a Harvard Business School match, such tinkering could have increased the number who get their first or second choice by 18 percent. Such gains suggest tinkering might be widespread, which magnifies rank efficiency’s importance as a descriptive concept. However, since rank efficiency and strategyproofness cannot coexist, tinkering gains may prove illusory. Nevertheless, tinkering need not undermine incentive properties entirely: when agents have little information and no outside options, truth-telling can be a best response.

A Theory of Information Nudges
with Lucas C. Coffman and Judd B. Kessler
(This paper is undergoing heavy revision. An older version is available by request.)

abstract: Nudge-style interventions are popular but are often criticized for being atheoretical. In this paper, we present a model of information nudges (i.e., interventions that provide useful—but imperfect—information about the utility of taking an action) based on Bayesian updating in a setting of binary choice. We use reduced-form and structural methods to conduct a meta-analysis of 73 information nudges in the economics and psychology literature and find that the sign and magnitude of their treatment effects vary in a way that is consistent with our model. We provide guidance for practitioners about the environments in which information nudges will positively affect a desired behavior and those in which they may backfire.

Why Do Some Clearinghouses Yield Stable Outcomes? Experimental Evidence on Out-of-Equilibrium Truth-Telling
with Eric Mayefsky and Colin D. Sullivan

abstract: In two-sided settings, market designers tend to advocate for deferred acceptance (da) over priority mechanisms, even though theory tells us that both types of mechanisms can yield unstable matches in incomplete-information equilibrium. However, if match participants on the proposed-to side deviate from equilibrium by truth-telling, then da yields stable outcomes. In the lab, we find out-of-equilibrium truth-telling under da but not under a priority mechanism, which could help to explain the success of da in preventing unraveling in the field. We then attempt to explain the difference in behavior across mechanisms by estimating an adaptation of the experience-weighted attractions (ewa) learning model (Camerer and Ho 1999). This adaptation contributes in two ways. First, it reparametrizes the original ewa model to separate the learning process into three distinct parts: non-experiential initial cognition, experiential learning, and how important the former is when it feeds into the latter. Second, it explores methods for fitting learning models to games with large strategy sets. (Most of the literature on learning models focuses on games with a single-digit number of strategies; in our setting, there are over 300.) Once we fit the adapted ewa model, we find that initial beliefs drive the difference in agents’ ability to find strategic equilibria, rather than differences in the learning process.

Works in Progress

Beyond Strict Preferences: The Value of Indifferences and Cardinal Information in Matching
with Judd B. Kessler and Karthik Tadepalli

abstract: In a Harvard Business School match, we show that eliciting indifferences increases the fraction of 1st-choice assignments by almost 11 percentage points (compared to a counterfactual match in which mbas are forced to submit strict rankings). Of course, the welfare impact of this improvement depends on the underlying (and unobserved) cardinal preferences. In a teacher-placement match in Chile, we use a novel, incentive-compatible method to elicit cardinal preferences from real teachers. We then compare the standard mechanism used in the field—random serial dictatorship (rsd) with strict preferences—to a version of rsd that allows for indifferences, and to the cardinal welfare max. Relative to rsd with strict preferences, the cardinal welfare max makes each participant almost $1,700 better off on average. Allowing indifference reporting accounts for almost $500 of this.

Does Liquidity Affect Job Choice in the Long Run: Evidence from Teach For America
with Lucas C. Coffman, John J. Conlon, and Judd B. Kessler

abstract: A limitation to our qje paper “Liquidity Affects Job Choice: Evidence from Teach For America” is that it only follows its subjects’ initial career choices. A natural follow-up question is how liquidity affects career choice several years later. To answer this question, we look to two sources of data. First, sufficient time has passed since our first field experiment that we can check to see who remained a teacher at the conclusion of their two-year stint with tfa. Second, we can track subjects even beyond their two-year stints with tfa by finding them via professional and social media platforms such as LinkedIn and Twitter. Although arduous, we find we are able to track the vast majority of out subjects in this way. Preliminary results suggest that the effects of our intervention stretch even to four years after initial treatment.