Testing Capacity-Constrained Learning
Andrew Caplin, Daniel Martin, Philip Marx · 2025
Read Testing Capacity-Constrained Learning by Andrew Caplin, Daniel Martin, Philip Marx (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-00195.txt.
Introduction
Andrew Caplin and co-authors devise in this 2025 paper the first broad test of learning models where the mind's intake pipe has a fixed width. Such models, from rational inattention to efficient coding, assume perception limits are set in advance. The team proves choice data fit this family exactly when they obey a no-improving-switches rule: no change of action or attention would have paid.
Checking past experiments where rewards for accuracy varied, they find people break the rule across classic tasks like judging color shares, shapes, and counts. Some tasks in the literature pass, though. The split, they argue, shows where fixed-capacity thinking holds and where it fails.
Catalog record
| CREATOR | Andrew Caplin, Daniel Martin, Philip Marx |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 76 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-00195
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Testing Capacity-Constrained Learning Thanks: We thank Federico Echenique, Mark Dean, and Gerelt Tserenjigmid for helpful comments. We also thank the Sloan Foundation for support under the “Cognitive Economics at Work” grant. Andrew Caplin, Daniel Martin, Philip Marx Abstract We introduce the first general test of capacity-constrained learning models. Cognitive economic models of this type share the common feature that constraints on perception are exogenously fixed, as in the widely used fixed-capacity versions of rational inattention ( sims2003implications ) and efficient coding ( woodford2012prospect ). We show that choice data are consistent with capacity-constrained learning if and only if they satisfy a No Improving (Action or Attention) Switches (NIS) condition. Based on existing experiments in which the incentives for being correct are varied, we find strong evidence that participants fail NIS for a wide range of standard perceptual tasks: identifying the proportion of ball colors, recognizing shapes, and counting the number of balls. However, we find that this is not true for all existing perceptual tasks in the literature, which offers insights into settings where we do or do not expect incentives to impact the extent of attention. Key words: Cognitive economics, rational inattention, efficient coding, subjective perception, Bayesian learning, incentives, experiments JEL codes: D60, D83, D91 1 Introduction Cognitive economics has emerged as a leading approach for understanding human cognition. Motivated by a long literature on subjective perception in cognitive science (e.g., weber1834pulsu ), this approach blends economic, psychology, and neuroscience methods to better understand the limits of human perception ( caplin2024data ). One of the main divides in cognitive economics is whether to treat the bounds on perception (“perceptual capacities” or “perceptual constraints”) as exogenously fixed or not. For example, in the rational inattention literature, sims2003implications assumes the bounds on attention are fixed, and matvejka2015rational assumes the bounds are elastic. In the efficient coding literature, woodford2012prospect provides versions of optimal coding with both fixed bounds and elastic bounds. 1 1 1 Fixed-capacity efficient coding has been applied to risky choice ( khaw2021cognitive , frydman2022efficient ), investing ( charles2024insensitive ), probability weighting ( frydman2023source ), belief updating ( ba2022over ), and play in games ( frydman2021coordination ). Whether or not perceptual capacity is exogenously fixed has substantial implications for the economic impact of bounded cognition. Most importantly, if perceptual constraints are exogenous, then changes in incentives will not alter the extent of attention, just how individuals decide to allocate their scarce attention. Whether incentives alter attention has become a key question for a growing literature on the role of cognition in economics (e.g., bronchetti2023attention ). 2 2 2 A halfway house between these two models is one in which there learning is endogenous but bounded, as in filiz2024information . The question of whether incentives impact the extent of attention also relates to an ongoing debate about whether incentivization matters in experiments. Various reasons have been given for incentivizing experiments (e.g., plott1986rational , smith1991rational ), but a long literature in psychology treats incentives as irrelevant to cognition, so it does not incentivize performance in experiments on perception and attention. However, this assumption has been challenged by papers showing the impact of rewards on visual perception in the psychology literature on psychometrics (e.g., pessoa2010embedding ). We offer a method for answering the question of whether perceptual capacity is exogenously fixed, and hence whether incentives impact the extent of attention, by providing a sharp test of capacity-constrained learning models, a class of models in which decision-makers choose among a feasible set of ways to learn. Because feasibility is an exogenous constraint, this model class covers exogenous perception constraints, including fixed-capacity versions of rational inattention ( sims2003implications ) and efficient coding ( woodford2012prospect ). We show that a single condition, No Improving (Action or Attention) Switches (NIS) , is both necessary and sufficient for capacity-constrained learning. NIS specializes the test of costly learning from caplin2015revealed by replacing their No Improving Attention Cycles (NIAC) condition and the No Improving Action Switches (NIAS) condition of caplin2015testable with a single new condition, which implies both NIAS and NIAC, but is stronger than both. In words, NIS states that wholesale switches of attention or actions, both within and across decision problems, should not improve utility. We apply our test of capacity-constrained learning to data from three existing experiments that vary the payoffs to being correct while performing visual perception tasks. We first look at experiments 1.2 and 2.2 of dean2023experimental , which vary the reward to correctly guessing the state in two tasks: the number of colored balls in a visual display (1.2) and the number of correct equations (2.2). Using the aggregated choices of participants, we find that NIAS and NIAC are satisfied for both experiments. However, NIS fails in both experiments, and the failure is statistically significant in Experiment 1.1 ( p<0.01 ). 3 3 3 One possible reason that the violation of NIS in experiment 2.2 is not statistically significant is that performance is already high (over 80%) at very low incentives, which leaves little room for improvement with incentives. Another possible reason is that there are only two incentive levels, so there are few inequalities to check. Importantly, NIS fails in a systematic way in both experiments: switches in attention in the direction of higher incentives always lead to improvements in expected utility. Next, we reexamine the data from the experiment of caplin2020rational , which varies the payoff for correctly guessing which shape appears the most frequently. We find that NIAS is satisfied for all four difficulty levels of the task, but that NIS is not satisfied for any of the difficulty levels ( p<0.01 for all four). Finally, we re-analyze the data from the experiment of dewan2020estimating , which varies the payoffs at two possible prize sizes ($10 and $20). For the first task, correctly guessing the number of balls in a display, we again find that NIAS and NIAC are satisfied for both prize sizes, but NIS fails for both ( p<0.01 for each prize size). As in the other experiments, we find that for a given prize size, switches of attention in the direction of higher incentives always lead to improvements in expected utility. However, for the second task of dewan2020estimating , which requires participants to judge the degree of an angle, we do not find strong evidence that NIS fails for either of the prize sizes ( p=0.45 and p=0.72 ). In addition, unlike the other experiments, switches of attention in the direction of higher incentives (for a given prize size) often do not lead to improvements in expected utility. These results suggest that certain task factors might dictate the suitability of capacity-constrained learning. Specifically, this model class might be more appropriate for tasks that are not responsive to additional effort. 4 4 4 The relationship between incentives and the effort required to complete judgment tasks is explored in camerer1999effects . This can occur because (1) even at low incentives, an individual chooses to exert enough effort to reach their perceptual limit, and (2) additional effort would not meaningfully improve perception beyond this limit. This combination of forces causes individuals to reach the same level of “irreducible uncertainty” at all levels of incentives. Another possibility is that uncertainty could actually