Time Will Tell: Recovering Preferences When Choices Are Noisy
当选择具有随机性时，显示偏好分析通常依赖于随机效用模型。然而，若不对效用噪声的分布做出假设，便无法推断偏好。我们证明，利用反应时数据可以克服这一困难。只需对反应时分布施加一个简单条件，即可在无须分布假设的情况下保证选择揭示偏好。经济学与心理学中的标准模型所生成的数据均满足该条件。在对称噪声或费希纳（Fechnerian）噪声下，我们得到了更强的结果：反应时可用于揭示偏好，或在样本外预测选择概率。我们的工具应用简便，并能取得出色的预测精度。
英文原文
When choice is stochastic, revealed preference analysis often relies on random utility models. However, it is impossible to infer preferences without assumptions on the distribution of utility noise. We show that this difficulty can be overcome by using response time data. A simple condition on response time distributions ensures that choices reveal preferences without distributional assumptions. Standard models from economics and psychology generate data fulfilling this condition. Sharper results are obtained under symmetric or Fechnerian noise, where response times allow uncovering preferences or predicting choice probabilities out of sample. Application of our tools is simple and generates remarkable prediction accuracy.