-
A robust measure of complexity
Authors:
Egor Bronnikov,
Elias Tsakas
Abstract:
We introduce a robust belief-based measure of complexity. The idea is that task A is deemed more complex than task B if the probability of solving A correctly is smaller than the probability of solving B correctly regardless of the reward. We fully characterize the corresponding order over the set of tasks. The main characteristic of this relation is that it depends, not only on difficulty (like m…
▽ More
We introduce a robust belief-based measure of complexity. The idea is that task A is deemed more complex than task B if the probability of solving A correctly is smaller than the probability of solving B correctly regardless of the reward. We fully characterize the corresponding order over the set of tasks. The main characteristic of this relation is that it depends, not only on difficulty (like most complexity definitions in the literature) but also on ex ante uncertainty. Finally, we show that for every task for which information is optimally acquired, there exists a more complex task which always induces less effort regardless of the reward.
△ Less
Submitted 15 January, 2025;
originally announced January 2025.
-
Belief identification by proxy
Authors:
Elias Tsakas
Abstract:
It is well known that individual beliefs cannot be identified using traditional choice data, unless we exogenously assume state-independent utilities. In this paper, I propose a novel methodology that solves this long-standing identification problem in a simple way. This method relies on the extending the state space by introducing a proxy, for which the agent has no stakes conditional on the orig…
▽ More
It is well known that individual beliefs cannot be identified using traditional choice data, unless we exogenously assume state-independent utilities. In this paper, I propose a novel methodology that solves this long-standing identification problem in a simple way. This method relies on the extending the state space by introducing a proxy, for which the agent has no stakes conditional on the original state space. The latter allows us to identify the agent's conditional beliefs about the proxy given each state realization, which in turn suffices for indirectly identifying her beliefs about the original state space. This approach is analogous to the one of instrumental variables in econometrics. Similarly to instrumental variables, the appeal of this method comes from the flexibility in selecting a proxy.
△ Less
Submitted 22 November, 2023;
originally announced November 2023.
-
Belief identification with state-dependent utilities
Authors:
Elias Tsakas
Abstract:
It is well known that individual beliefs cannot be identified using traditional choice data, unless we impose the practically restrictive and conceptually awkward assumption that utilities are state-independent. In this paper, we propose a novel methodology that solves this long-standing identification problem in a simple way, using a variant of the strategy method. Our method relies on the concep…
▽ More
It is well known that individual beliefs cannot be identified using traditional choice data, unless we impose the practically restrictive and conceptually awkward assumption that utilities are state-independent. In this paper, we propose a novel methodology that solves this long-standing identification problem in a simple way, using a variant of the strategy method. Our method relies on the concept of a suitable proxy. The crucial property is that the agent does not have any stakes in the proxy conditional on the realization of the original state space. Then, instead of trying to identify directly the agent's beliefs about the state space, we elicit her conditional beliefs about the proxy given each state realization. The latter can be easily done with existing elicitation tools and without worrying about the identification problem. It turns out that this is enough to uniquely identify the agent's beliefs. We present different classes of proxies that one can reasonably use, and we show that it is almost always possible to find one which is easy to implement. Such flexibility makes our method, not only theoretically-sound, but also empirically appealing. We also show how our new method allows us to provide a novel well-founded definition of a utility function over states. Last but not least, it also allows us to cleanly identify motivated beliefs, freed from confounding distortions caused by the identification problem.
△ Less
Submitted 30 November, 2022; v1 submitted 20 March, 2022;
originally announced March 2022.
-
Identification of misreported beliefs
Authors:
Elias Tsakas
Abstract:
It is well-known that subjective beliefs cannot be identified with traditional choice data unless we impose the strong assumption that preferences are state-independent. This is seen as one of the biggest pitfalls of incentivized belief elicitation. The two common approaches are either to exogenously assume that preferences are state-independent, or to use intractable elicitation mechanisms that r…
▽ More
It is well-known that subjective beliefs cannot be identified with traditional choice data unless we impose the strong assumption that preferences are state-independent. This is seen as one of the biggest pitfalls of incentivized belief elicitation. The two common approaches are either to exogenously assume that preferences are state-independent, or to use intractable elicitation mechanisms that require an awful lot of hard-to-get non-traditional choice data. In this paper we use a third approach, introducing a novel methodology that retains the simplicity of standard elicitation mechanisms without imposing the awkward state-independence assumption. The cost is that instead of insisting on full identification of beliefs, we seek identification of misreporting. That is, we elicit beliefs with a standard simple elicitation mechanism, and then by means of a single additional observation we can tell whether the reported beliefs deviate from the actual beliefs, and if so, in which direction they do.
△ Less
Submitted 24 December, 2021;
originally announced December 2021.
