Choice-induced preference change and the free-choice paradigm: A clarification

Abstract

Positive spreading of ratings or rankings in the classical free-choice paradigm is commonly taken to indicate choice-induced change in preferences and has motivated influential theories as cognitive dissonance theory and self-perception theory. Chen and Risen [2010] argued by means of a mathematical proof that positive spreading is merely a statistical consequence of a flawed design. However, positive spreading has also been observed in blind choice and other designs where the alleged flaw should be absent. We show that the result in Chen and Risen [2010] is mathematically incorrect, although it can be recovered in a particular case. Specifically, we present a formal model of decision making that satisfies all assumptions in that article but implies that spreading need not be positive in the absence of choice-induced preference change. Hence, although the free-choice paradigm is flawed, the present research shows that reasonable models of human behavior need not predict consistent positive spreading. As a consequence, taken as a whole, previous experimental results remain informative.

Changes in preferences reported after choices are informative, not merely statistical artifacts.

“…The null effect arises because SoA will be positive on trials where the differences in prechoice ratings (dV1) are low, but negative on trials where dV1 is high. The positive SoA for low dV1 trials has been explained in previous studies (Alo ´s-Ferrer & Shi, 2015;Chen & Risen, 2010;Izuma & Murayama, 2013;Lee & Daunizeau, 2021), but in brief, it is due to the fact that the expected postchoice value difference conditional on the choice will always be greater than the unconditional expected postchoice value difference. The negative SoA for high dV1 trials has to do with the boundedness of the rating scale.…”

“…This directly provided a measure of postchoice difficulty (dV2), as for prechoice difficulty. We deterministically set (stochastic choices are considered in the ) the chosen option on each trial as the option with a higher “true value.” Note that by eliminating noise from the choice process, we maximize the opportunity for SoA to arise (Alós-Ferrer & Shi, 2015; Izuma & Murayama, 2013). Finally, we calculated SoA in the usual way: the post- minus prechoice ratings of the chosen options minus the post- minus prechoice ratings of the rejected options.…”

Transient value refinements during deliberation facilitate choice.

“…Specifically, if the value ratings that participants report are assumed to be noisy measures of the true underlying subjective values, and the choices that participants report are assumed to align with their true preferences, then regression to the mean can sometimes cause the ratings of choice pair options to spread apart from pre- to postchoice. Chen and Risen provided a mathematical demonstration that positive SoA is always predicted even when preferences are stable, and other authors have further explained the finding conceptually and via computer simulation (Alós-Ferrer & Shi, 2015; Izuma & Murayama, 2013).…”

“…The basic idea is that, under the assumptions that ratings are noisy estimates of true preferences and choices reflect true preferences, postchoice ratings of choice option pairs will be (on average) farther apart relative to the corresponding prechoice ratings. It has been demonstrated via simulation that the statistical observation holds when the difference in prechoice ratings is sufficiently small (Alós-Ferrer & Shi, 2015; Izuma & Murayama, 2013; Lee & Pezzulo, 2022; see also ). However, many studies have since shown that SoA still occurs even after controlling for such statistical artifacts.…”

Does choice change preferences? An incentivized test of the mere choice effect

“…However, it has been recently shown that the experimental paradigm which has guided the development of this literature for over 50 years is regrettably flawed. It contains a statistical bias that can result in apparent preference change even if participants have stable preferences (Chen and Risen, 2010 ; Izuma and Murayama, 2013 ; Alós-Ferrer and Shi, 2015 ). Although some improved designs have been proposed (e.g., Alós-Ferrer et al., 2012 ), how the effect of tradeoff choices in economically-relevant domains could be studied remains an unresolved issue at the time of writing.…”