This paper examines how individuals with limited attention allocate their information consumption across neutral and biased news sources. I develop a model in which agents rationally choose a portfolio of outlets to learn about an uncertain state of the world. While confirmation bias emerges as the benchmark outcome when only biased sources are available, the presence of a sufficiently precise neutral source fundamentally alters optimal information acquisition. Neutral sources do not merely moderate beliefs: they restructure the informational value of biased outlets. In particular, agents with weak prior beliefs may optimally combine neutral sources with sources opposing their prior views. This portfolio exploits an asymmetric complementarity between reliable neutral information and the possibility of immediate uncertainty resolution provided by opposing sources, leading to sharp posterior beliefs without requiring any intrinsic preference for disagreement.
Distorting Attention: Platform Design, Market Complexity, and the Balance Between Advertising and Fees
This paper studies how digital retail platforms jointly monetize transactions and user attention when consumers face limited cognitive capacity. Advertising generates revenue but also degrades the information environment by increasing the cost of learning about product quality. Embedding rational inattention into a platform design problem, I show that the profitability of advertising is non-monotonic in market complexity: it is limited in simple environments and counterproductive in highly complex ones, but peaks at intermediate levels where consumers actively acquire information. From a welfare perspective, the model delivers a novel implication: markets with higher advertising intensity can generate higher buyer surplus. A suggestive empirical exercise using variation across eBay categories shows patterns consistent with the model’s predictions.
Categorization and Reference-Dependent Decision-Making under Statutory Thresholds: Evidence from a Criminal Law Reform (R&R in JEBO)
Thresholds do more than determine which formal rules apply: by defining discrete categories, they also shape the cases against which decision-makers evaluate one another. This paper examines this phenomenon through the lens of a 2020 Czech criminal law reform that shifted the monetary thresholds defining theft offenses. The reform lowered statutory sentencing ranges for some categories, while in others it left the formal range unchanged but expanded the category to include more severe cases. Using administrative case-level data and a difference-in-differences design, I find that custodial sentences become shorter in both groups, with little change in the probability of imprisonment. The decline in sentence length despite unchanged statutory ranges is consistent with a reference-group effect: as more severe cases enter a category, previously typical cases become relatively less severe within the category. The findings are also consistent with a severity effect associated with downward shifts in statutory sentencing ranges. More broadly, the results show that threshold reforms can affect decisions not only through changes in formal rules but also by reshaping comparison sets, generating spillovers into formally unaffected categories. This has implications for the design of threshold-based policies and for empirical strategies that use formally untreated groups as controls.
Why No Tomorrow? Cognitive Costs, Beliefs, and Failure to Plan
This paper studies sequential decision-making under cognitive costs. Motivated by experimental evidence showing systematic failure to plan, I develop a model in which
agents optimally choose whether to inspect future opportunities. The model characterizes the value of information in a setting with indivisible consumption choices and a binding budget constraint. I show that agents may rationally ignore future opportunities when the expected value of information is low relative to cognitive costs, generating behavior that resembles impatience. The model yields closed-form expressions for continuation values and admits a tractable decomposition of the value of information. A calibrated version matches experimental patterns.
