We introduce an experimental design that allows us to observe a proportion of decisions in a lying experiment. We vary the objective probability of observation and compare the differing equilibrium predictions of models, which could not be tested with existing experimental methods.
It is a well-known theoretical result that paying one random decision is not an incentive compatible payment procedure in ambiguity experiments. Recent work by Baillon, Halevy, and Li (2022 - Econometrica) has shown that using this payment procedure may distort decisions by up to 50%! They lay out the dilemma well. You can give a single decision problem and pay subjects for that single decision, which is incentive compatible but restricts the types of questions you can ask, or you can ask multiple questions but risk miscoding people as subjective expected utility maximizers when they actually have ambiguity averse preferences. In this project, we identify and apply a payment procedure that allows us to ask subjects multiple questions, while maintaining incentive compatibility for ambiguity models that satisfy consequentialism.
In a US population, we study lying behavior in an uncertain environment where karmic motivations may be relevant. We hypothesize that karmic motivations may not be as prominent in our population, so we implement a treatment where subjects are given a religious prime before participating in the task. If this prime makes religious identity more salient, then behavior should shift to being consistent with karmic motivations (which are common even in western religions). We additionally introduce a money prime and a competitive prime to test if these shift behavior to more lying without following the pattern consistent with a karmic motivation.
We test the concept of "Detectable vs Deniable Lies" (Tergiman & Villeval, 2023 - Management Science) in an individual decision setting. For this setting, we analyze equilibrium behavior of a model that considers a fixed lying cost and social image concern. We vary the probability of observation and the value of the deniable lie and compare our results to the comparative statics of the model.