A Preregistered Falsification Test of the Decision by Sampling Model and Rank-Order Effect
通过预注册实验检验决策抽样模型（DbS）的核心预测：操纵近期经历的属性分布能否改变人们对选项的主观估值。结果强烈反对DbS的预测，但发现情境化效应显著影响人们对相同赌注的估值。
Many social scientists have assumed that people’s preferences can be described by stable and coherent “utility” functions. This notion of stable utility functions has been challenged by cognitive psychologists who suggest that preferences are malleable and constructed in the moment, but neither camp has explained how the subjective valuations underpinning preferences arise. One influential attempt to do so is the Decision by Sampling (DbS) model, which suggests that a quantitative attribute’s (e.g., money sum’s) subjective value is its rank order in a momentarily activated memory sample. DbS thus implies that manipulating the recently experienced attribute distribution should change people’s subsequent valuations of that attribute: for example, from the typically assumed concave shape of the utility function to a convex shape. However, recent studies have pointed out methodological concerns in the evidence previously thought to support this prediction (and thus, DbS). In this preregistered study, we replicate the previous paradigm but address the methodological concerns to test if such a “rank-order” manipulation does change valuations. We derive qualitative predictions from DbS to verify that our conditions yield distinct predictions. We find strong evidence against the DbS’s prediction that a “rank-order” manipulation changes what options the participants select and how strongly they prefer the options. We also find extreme evidence in favor of a contextualization effect, implying that people value formally identical gambles differently depending on whether they cue a real-life setting or not. Although we encourage replication by independent laboratories, these results suggest that the DbS is falsified for this binary choice task. This paper was accepted by Yuval Rottenstreich, behavioral economics and decision analysis. Funding: This research was funded by the Marcus and Amalia Wallenberg Foundation [Grant MAW 2016.0132] and the Swedish Research School of Management and IT. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2022.03611 .