- The paper introduces a framework to identify preferences in discrete choice models without assuming completeness, using upper and lower utility bounds and defining a sharp identification region.
- It demonstrates that preferences can be partially identified from data even without the completeness assumption, and that completeness itself can be empirically tested and rejected.
- The research reveals that policies designed to increase an alternative's selection frequency might fail or even backfire under incomplete preferences, unlike in traditional complete preference models.
Identification of Incomplete Preferences: A Technical Exploration
The paper "Identification of Incomplete Preferences," authored by Arie Beresteanu and Luca Rigotti, introduces a novel framework to address the identification problem in discrete choice models under incomplete preferences. Traditionally, discrete choice models, which have played a pivotal role in applied economics since McFadden's seminal work in 1974, rest on the assumption of complete preferences—an axiom that assumes individuals always have the capacity to rank alternatives fully. Beresteanu and Rigotti challenge this traditional axiom by allowing for the possibility that individuals' preferences may be incomplete, particularly when only aggregate choice data is available. This approach presents a significant theoretical advance, as it accommodates a more flexible modeling of consumer behavior, where alternatives may not always be comparable.
The authors introduce a methodological innovation by defining a sharp identification region for such choice models. In their framework, consumer behavior is modeled using upper and lower utility bounds for each alternative, permitting the presence of non-comparable choices. The identification region established by the authors effectively places bounds on the probability distribution of these upper and lower utilities, yielding significant implications for econometric modeling and inference.
The robust findings of the paper dispel two prevalent misconceptions: firstly, that without the completeness assumption, data cannot elucidate the underlying preferences, and secondly, that completeness cannot be ruled out using observable behavior. By combining econometric and decision theory insights, Beresteanu and Rigotti demonstrate that preference parameters across individuals are partially identifiable, a result which, at a minimum, constrains the possible underlying preferences based on observed choice data.
An additional key contribution of the paper is the introduction of instrumental variables as a means to test and reject the hypothesis that all preferences in the population are complete. An instrumental variable, in this context, is defined as a variable that is independent of preferences but still exerts influence on choices. The existence of such a variable implies that it is possible for some individuals in the population to exhibit incomplete preferences, further refining the identification region.
Moreover, the paper examines policy interventions aimed at increasing the selection frequency of a particular alternative by adjusting its perceived utility. The authors provide a counterintuitive result: in settings where preferences are incomplete, such policies might not engender the anticipated outcome and could potentially lead to a decrease in the target choice’s selection frequency. This outcome is fundamentally unattainable within the paradigm of complete preferences.
The theoretical underpinnings are bolstered by an empirical application using data from the 2018 midterm elections in Ohio, focusing on judicial races where the candidates' party affiliations were not present on the ballot. The empirical analysis illustrates how actual voting data can be utilized to provide insights into voter preferences, particularly highlighting the role of ballot order as an instrumental variable that influences voter behavior, thus exhibiting the practical applicability of the theoretical framework.
Overall, the implications of this research are manifold. Theoretically, the findings elucidate the nuances of preference heterogeneity and augment the understanding of rational behavior within the context of incomplete preferences. Practically, the ability to identify and model preferences without the completeness assumption has significant ramifications for policy design and the evaluation of economic interventions. Future developments in AI could benefit from incorporating models accommodating incompleteness, potentially enhancing decision-making algorithms by allowing for more variance in preferences under uncertainty.
This paper signals a substantial shift toward models that better reflect real-world decision-making processes and broadens the scope for more nuanced analyses of consumer behavior in economics and beyond. As this line of inquiry progresses, it is poised to influence the design of econometric models in various domains, including market design, behavioral economics, and predictive analytics, where the assumption of complete preferences may be too restrictive.