Bridging between Hypothetical and Incentivized Choice

长期以来，在用于市场研究的典型离散选择实验中收集的选择的假设性一直是一个令人担忧的问题。由于这些实验中的选择对受访者来说或多或少是无关紧要的，从这些数据中收集的推论可能缺乏有效性。最近的市场营销研究确实表明，基于适当激励的受访者的选择校准的模型的预测有效性提高(例如，Ding等人，2005；Ding 2007，Ding等人，2009；董等人，2010)。然而，与标准假设设置相比，进行所谓的激励对齐离散选择实验更费力，成本也更高。因此，这项提议的目标是开发一个基于模型的框架，在假设的离散选择实验的数据和激励对齐的离散选择实验的数据之间简单易行地架起桥梁，目的是节省数据收集工作和成本，从而与预测和激励选择的目标保持一致。简而言之，该框架利用一定的不变性假设来融合来自假设离散选择实验的大量数据和来自独立实验中收集的激励对齐离散选择实验的相对较小的数据量，但在相同的人群中，目的是为了模拟该人群中的激励选择。该框架与经济学理论一致，假设了一组不变的(深度的)偏好参数，但明确说明了假设和激励一致的设置之间的不同决策努力。不同的决策努力被概念化为既影响认知加工的量，也影响被选择进行加工的信息集。因此，决策努力的数量可能会实质性地改变选择概率和结果，即使潜在的、深层次的偏好参数是不变的。在操作上，我们建立在数学心理学中发展的基于过程的选择模型，特别是最近提出的依赖泊松种族模型。我们的衔接框架将有助于在与足够多的受访者进行激励匹配实验非常昂贵甚至高得令人望而却步的情况下进行有效的推断。

The hypothetical nature of choices collected in typical discrete choice experiments for market research has long been a source of concern. Because choices in these experiments are more or less inconsequential for respondents, inferences gleaned from this data may lack validity. Recent research in marketing indeed demonstrates increased predictive validity of models calibrated based on choices by properly incentivized respondents (e.g. Ding et al.,2005; Ding2007, Ding et al., 2009; Dong et al., 2010). However, conducting so called incentive-aligned discrete choice experiments is more effortful and costly compared to the standard hypothetical setting. The goal of this proposal therefore is to develop a model based framework that parsimoniously bridges between data from hypothetical discrete choice experiments and data from incentive-aligned discrete choice experiments for the purpose of conserving on data collection effort and cost, however, in keeping with the goal of predicting to incentivized choices. In a nutshell, the framework leverages certain invariance assumptions to fuse a large amount of data from hypothetical discrete choice experiment and a relatively much smaller amount of data from incentive-aligned discrete choice experiment collected in independent experiments, but in the same population, for the purpose of simulating incentivized choices in this population. The framework assumes, in line with economic theory, a common set of invariant ('deep') preference parameters, but explicitly accounts for differential decision effort between the hypothetical and the incentive-aligned setting. Differential decision effort is conceptualized as both affecting the amount of cognitive processing, as well as the information set selected for processing. As a consequence, the amount of decision effort may materially change choice probabilities and outcomes, even if underlying, deep preference parameters are invariant. Operationally, we build on process-based choice models developed in mathematical psychology and specifically the recently proposed dependent Poisson race model. Our bridging framework will facilitate valid inferences in situations where it is very or even prohibitively costly to conduct incentive aligned experiments with a sufficiently large number of respondents.