Research

Working Papers

Beyond Bayes: Naive Frequentism and the Behavioral Impact of Redundant Information (with David Walker-Jones)

This paper presents a novel experiment investigating how individuals update beliefs when faced with repeated or redundant information, uninformative signals common in real-world communication but rarely studied in economics. While Bayes’ rule prescribes that such information should be ignored, we find it significantly influences behavior. These deviations from Bayesianism are not due to computational complexity: calculating posteriors is generally as simple as averaging two integers. Instead, our results suggest that many subjects rely on heuristics consistent with naive frequentism, interpreting repeated or redundant information as indicative of frequency, which helps explain its persuasive power and prevalence in real-world settings.

Cognitive Ability and Perceived Disagreement in Learning (with Piotr Evdokimov)  

Revise & Resubmit at AEJ:Micro

Do agents believe they agree more with others over time? This paper explores how individuals perceive the belief updating behavior of others and the resulting disagreement in a sequential experiment with public information. We uncover a persistent gap in the perception of disagreement as a function of cognitive ability. Higher cognitive ability correlates with less perceived disagreement, although the average subject underestimates the extent of actual disagreement. Information about a partner's cognitive ability only impacts perceived disagreement when the partner has a low test score. Our findings highlight the roles of overconfidence and cognitive projection in shaping these perceptions.

Individual vs. Social Learning: An Experiment (with Piotr Evdokimov)

We study the effectiveness of social learning when subjects can also learn individually. In the baseline treatment, a subject observes a sequence of private signals about an unknown state of the world and chooses an action to match the state. In the unilateral social learning treatment, a subject observes the lagged actions of another subject in the baseline condition in addition to her own private signals. In the bilateral treatment, two subjects observe each other's lagged actions in addition to their own signals. While beliefs and decisions become more accurate over time, social learning fails in the sense that accuracy is not improved by observing another subject's actions. The median subject extracts 1.6% of the available information from her partner's action in the bilateral treatment and about 20% of the available information in the unilateral treatment. We also find evidence of correlation neglect: subjects in the bilateral treatment treat the actions of others as if they do not depend on their own past histories.

Learning while Signaling in Markets (with Kane Sweeney)

Consider the case of a seller that signals its own product quality but faces uncertainty about the cost of signaling. If signaling reveals information about its cost which can be used in future trades, how does learning affect signaling incentives and the informativeness of signaling for buyers? We uncover a novel intertemporal trade-off: An increase in the probability of early signaling reduces the future value of signaling. Learning introduces endogenous noisiness in the interpretation of the signal which generates a dynamic adverse selection effect. We characterize signaling equilibria in which a positive option value of learning exceeds the dynamic adverse selection effect and induces some types to experiment with signaling. These equilibria are ranked with more experimentation leading to lower welfare.

Uncertainty in a Connected World

This paper studies network games with strategic complementarities when agents have only partial knowledge of, but some control over, how their actions are mapped into payoff-relevant outcomes. Uncertainty changes several predictions of standard network models. Equilibrium actions are positively related to in-degree network centrality but more central agents also show greater aversion to uncertainty. Higher systematic risk, and thus higher uncertainty, can generate local welfare improvements by decreasing aggregate risk in equilibrium. Agents that contribute the most to aggregate risk maximize the product between a measure of individual incentives to take up risk and their out-degree network centrality.

The Downsides of Managerial Oversight in Signaling Environments

This paper studies the interplay between a worker's signaling incentives and costly managerial oversight. The worker is privately informed about his ability. Ability affects the marginal return from effort, which is an observable choice. Contrary to the logic of oversight, the firm would benefit from committing not to acquire any information precisely when it is most uncertain about the worker's ability. Managerial oversight weakens incentives because signaling and oversight are strategic substitutes for the worker. Yet, effort decisions are distorted upward. The model explains unpaid overtime and presenteeism in the workplace while warning about the unintended consequences of common managerial practices.

An Agency Theory of Open Innovation

Open innovation is the practice of combining internal R&D efforts with external sources of knowledge. While project developers allocate effort between internal development and external knowledge processing, knowledge brokers search for relevant external ideas. When external knowledge is potentially complex to exploit, a firm faces a trade-off in the decision to "open up" its innovation process. This trade-off and the resulting degree of openness are affected by internal agency conflicts which determine the firm's ability to absorb external knowledge. The model provides several predictions regarding the optimal degree of openness which shed light on existing empirical evidence.

The Dynamics of Authority in Innovative Organizations

Why do innovative organizations often reallocate authority? I propose a simple theory in which innovation with new technologies generates an endogenous need for coordination among divisions due to cross-divisional externalities. A division manager privately observes the expected productivity of new technologies, which can be communicated strategically to headquarters. The organization has an advantage in coordinating technologies across divisions and can only commit to an ex-ante allocation of authority. I show that reallocations of authority can be an optimal (dynamic) organizational response to informational asymmetries. Generally, the decision to reallocate authority depends on the outcome of past innovation.

Centralize or Decentralize? How Productivity Shocks Affect Authority in Organizations

This paper identifies a channel through which productivity shocks may propagate within a multi-divisional organization and induce organizational change. Key ingredients are the presence of cross-divisional spillovers, strategic communication within the organization, and the risk-return trade-off associated with innovation. When the importance of spillovers is small and authority is decentralized, a shock that increases the productivity gap across divisions increases the relative performance of centralization over decentralization. Thus, a sufficiently large, and positive productivity shock may lead the organization to centralize authority. Centralization: i) helps to curb the innovative ambitions of the manager of the most productive division, which hurt spillovers; and ii) improves communication within the organization.

Delegated Experimentation

This paper studies situations in which a principal can only acquire information for a mutually relevant decision through a biased agent. The principal has no commitment power and transfers are impossible. We base our analysis on an exponential bandit model with one "risky" action and one "safe'' action.  We first characterize the unique, Markov-perfect equilibrium when the agent's effort choice is observable, and thus learning is symmetric. The chance to generate positive information about the risky action is balanced with the risk of producing negative information that makes the principal lean toward the undesirable safe action. In equilibrium, this tension produces a delay in information acquisition. However, when effort is unobservable, this delay can actually hurt the agent. Thus, when the agent's cost of experimentation is low, the agent ends up implementing the principal's optimal experimentation policy.