State Space Abstractions in Rational Inattention Discrete Choice Models

• Applicant: Professor Dr. Thomas Otter
• Subject Area: Management and Marketing
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 561639001

Human decision-makers (DMs) face cognitive constraints, limiting their ability to process all available information in complex environments. This project focuses on discrete choice among multiple alternatives with many attributes while accounting for these cognitive limitations. DMs engage in adaptive and partial information processing, selectively focusing on certain attributes while ignoring others to manage cognitive effort. Understanding these frictions is essential for firms and regulators aiming to measure preferences accurately and design effective products, pricing strategies, or policies. Consider an analyst designing subsidies for electric vehicles (EVs). Consumers of EVs must evaluate multiple attributes - battery range, charging infrastructure, ownership costs, environmental impact, and potential subsidies. Given the cognitive burden, they prioritize salient aspects like price or brand, processing more complex attributes (e.g., the subsidy) only if the effort seems worthwhile and potentially incompletely. This adaptive behavior can lead to mistakes, such as recognizing a subsidy but misjudging its value. An effective policy requires models that predict consumer responses accurately by accounting for how DMs allocate attention selectively and adaptively. Rational inattention discrete choice models (RI-DCMs) have recently gained traction in economics and marketing by capturing exactly this type of adaptive, partial information processing. RI-DCMs model how DMs optimally allocate attention, balancing the cognitive costs of processing more information with the potential benefits. However, a major challenge in applying these models lies in the "large state space problem". RI-DCM assumes that DMs enumerate the set of possible configurations ("the state space") of alternatives and integrate over it. As the number of attributes and alternatives grows, this space grows exponentially, making the model less plausible as a reflection of human decision-making and analytically intractable in many empirical applications. This project addresses this problem by adapting abstraction techniques from the cognitive and computer sciences to reduce large state spaces making RI-DCMs computationally feasible. Thus, our objectives are as follows. First, we extend the RI-DCM to handle high-dimensional choice environments using state space abstractions. Second, we seek to provide empirical evidence that DMs use state space abstractions in discrete choice tasks and that different abstraction types can be identified from discrete choice data typical in marketing and economics. We view the further development of solution methods for the RI-DCM within high-dimensional contexts as the key challenge in our project and its key contribution, once met. Moreover, insights gained from how DMs handle large state spaces will also inform other models of decision-making under uncertainty, such as consumer search models involving belief formation over complex alternatives.

• DFG Programme: Research Grants