Abstract
This paper illustrates how one can deduce preference from observed choices when attention is not only limited but also random. In contrast to earlier approaches, we introduce a Random Attention Model (RAM) where we abstain from any particular attention formation, and instead consider a large class of nonparametric random attention rules. Our model imposes one intuitive condition, termed Monotonic Attention, which captures the idea that each consideration set competes for the decision-maker's attention. We then develop revealed preference theory within RAM and obtain precise testable implications for observable choice probabilities. Based on these theoretical findings, we propose econometric methods for identification, estimation, and inference of the decision maker's preferences. To illustrate the applicability of our results and their concrete empirical content in specific settings, we also develop revealed preference theory and accompanying econometric methods under additional nonparametric assumptions on the consideration set for binary choice problems. Finally, we provide general purpose software implementation of our estimation and inference results, and showcase their performance using simulations.
Introduction
In classical economic theory, the principle of "revealed preference" suggests that an individual’s choices directly reflect their underlying desires. If a consumer selects a burger over a salad when both are available, it is traditionally inferred that the burger is preferred. However, this logic relies on the assumption of full attention. In reality, decision-makers are often overwhelmed by options, leading to choices based on a limited subset of available information. A consumer might choose the burger simply because they failed to notice the salad on the menu.
Existing models of limited attention have historically been either deterministic—assuming certain options are always ignored—or parametric, relying on specific mathematical formulas to describe distraction. This paper introduces the Random Attention Model (RAM), a nonparametric framework designed to infer preferences when attention is both limited and stochastic. By moving away from rigid psychological assumptions, RAM provides a robust method for identifying preferences from observed choice probabilities.
The Random Attention Model (RAM)
Monotonic Attention
The core of the RAM framework is the Monotonic Attention assumption. This condition posits that options compete for a decision-maker's limited focus. Formally, it suggests that if an item is removed from a menu that the person was not attending to anyway, the probability of them noticing their current choice should not decrease. In this view, items occupy "slots" in the mind; reducing the number of items reduces the competition for these slots, potentially increasing the attention paid to remaining options.
Revealed Preference and Regularity Violations
Standard choice models adhere to "regularity," the idea that adding a new option should never increase the probability of selecting an existing one. RAM, however, explicitly allows for and utilizes violations of regularity. If adding a "distractor" item $c$ to a menu increases the frequency with which item $a$ is chosen, it reveals that $a$ is preferred to the items from which the consumer was previously distracted.
Under RAM, a choice rule is compatible with the model if and only if the revealed preference relation is acyclic. This means a decision-maker cannot be found to prefer $a$ to $b$, $b$ to $c$, and $c$ back to $a$. This acyclicity serves as the fundamental litmus test for the model's validity in a given dataset.
Econometric Framework and Implementation
To bridge the gap between theory and empirical application, the authors translate these logical conditions into a system of moment inequalities. This allows researchers to test whether a dataset is compatible with RAM and to identify the set of preference orderings consistent with the observed data.
To facilitate this, the authors developed the ramchoice R package. This software implements the estimation and inference results, making the framework accessible for use with real-world data such as consumer scanner data or digital advertising click-streams.
Performance and Key Insights
Identification and Efficiency
Using simulated choice data (e.g., a decision-maker with a fixed preference $a_1 > a_2 > a_3 > a_4 > a_5$ choosing from 26 different menus), the authors demonstrated that RAM can successfully reject incorrect preference orderings even with relatively small samples (50–400 observations per menu).
Furthermore, the moment inequality approach is computationally efficient. In a universe of six items (720 possible orderings), the RAM method identifies preferences in approximately 7 minutes. In contrast, alternative methods like the "J-test" (projection residual) can take over 90 minutes to evaluate just a single ordering.
The Power of Binaries
The research highlights that adding minimal structure to how individuals choose between just two items—referred to as the "$\phi$-Attentive at Binaries" assumption—drastically increases the model's power to identify the true underlying preference ordering.
Conclusion
The Random Attention Model serves as a "big tent" framework, mathematically encompassing several previous models (such as Logit Attention) as special cases. By treating distraction as a signal rather than noise, RAM avoids the "misleading inference" trap where researchers might incorrectly assume a consumer dislikes an item they simply did not see.
While the current model is optimized for single decision units or homogeneous groups, future work aims to extend the framework to handle heterogeneous populations and more complex search behaviors, such as the impact of item positioning on digital interfaces.
Summary updated