Concetti Chiave
The optimal "de-biasing" procedure for a survey designer to estimate the true public reception of a product, given strategic and boundedly-rational respondents with varying levels of information and cognitive hierarchy.
Sintesi
The paper considers a survey design problem where respondents have different levels of rationality and strategic behavior:
Type 0 (Honest-Nonstrategic Respondents): These respondents provide truthful information based on their actual opinions about the product.
Type 1 (Level-1 Strategic Respondents): These respondents wish to influence the survey outcome correlated with their attitudes. They assume all other respondents are Type 0 and that the estimator (designer) is only aware of Type 0 respondents.
Type 2 (Level-2 Strategic Respondents): These respondents have a higher level of strategic thinking and behave as the best response to a mix of Types 0 and 1, assuming the designer perceives the responses as coming from a distribution of these lower types.
The survey designer, aware of these respondent types and their true statistics, aims to optimally (in the Bayesian sense) estimate the unbiased scores that reflect the true public reception of the product.
The authors model this problem using the strategic quantization framework, which is a special case of the information design problem in Economics. They focus on quadratic distortion measures and provide a gradient-descent based algorithm to compute the optimal classifier implemented by the boundedly rational Type 2 respondents.
The numerical results show the impact of various parameters, such as the cognitive hierarchy parameter, bias variance, and correlation between the state and bias variables, on the receiver's (designer's) estimation distortion. The results also compare the receiver's distortion for different types of senders, including non-strategic, fully rational, boundedly rational, and partially-strategic.
Statistiche
The paper does not contain any explicit numerical data or statistics. It focuses on the theoretical modeling and algorithmic design aspects of the problem.
Citazioni
The paper does not contain any striking quotes that support the key logics.