Revisiting the decoy effect: replication and extension of Ariely and Wallsten (1995) and Connolly, Reb, and Kausel (2013)‎

ABSTRACT

The decoy effect refers to the phenomenon whereby an inferior, unpreferable option reverses people’s preferences and increases the choice share of a targeted option. In two pre-registered experiments with an Amazon Mechanical Turk (MTurk) sample (N after exclusion = 1,001), we attempted to replicate two experiments from the decoy effect literature. We failed to replicate the original results in Study 1. The observed effects were not in the predicted direction, and their sizes were trivial. We replicated the decoy effect in Study 2, yet with a much smaller effect size than in the original. In addition, we concluded inconclusive evidence for the central hypothesis of the original study that regret salience weakens the decoy effect. We found some indication for a weak reduction, yet our sample size did not provide adequate power to detect this difference. Extending the replication in Study 2, we tested whether making salient the low reversibility of decisions can have a similar impact as regret salience. We again found indication for an effect in the predicted direction, yet the effect was too weak to be detected given our sample size. We discuss potential reasons for the discrepancies between the original and the replication results, as well as the implications. All materials, data, and analysis codes are available at https://osf.io/vsbzk.

KEYWORDS:

• Decoy effect
• decision reversibility
• regret
• attraction effect
• replication

Acknowledgments

We thank Siu Kit Yeung and Jonathan Pettibone for valuable comments on earlier drafts of this article.

Authorship declaration

Qinyu designed the experiment, produced the study materials, wrote the data analysis code, conducted the analysis, and drafted the manuscript and the supplementary materials. Shiyuan did preliminary work in analyzing the target articles and producing the surveys. Gilad guided the replication efforts, supervised each step, and ran data collection.

Additional information

The current replication is part of the larger “mass pre-registered replications in judgment and decision-making” project lead by Gilad Feldman. The project aims to revisit well-known research findings in judgment and decision making (JDM) and investigate the replicability of these findings. Further details can be found on: http://mgto.org/pre-registered-replications.

Disclosure statement

The authors declared no potential conflicts of interests with respect to the authorship and/or publication of this article.

Supplementary material

Supplemental data for this article can be accessed here.

Notes

1. Judgment and decision-making researchers use “decoy effects” to refer to a class of phenomena where preferences for two competing options are reversed due to the presence of a third option. The third option can be an asymmetrically dominated decoy, as mainly discussed in this article, a compromise decoy (an extreme option that makes its target look like a compromise between it and the other option; Simonson, 1989), or a phantom decoy (an option that dominates its target but is unavailable for choosing; Highhouse, 1996). Sometimes, however, researchers – particularly those whose expertise lies elsewhere – equate the decoy effect with the attraction or asymmetric dominance effect or at least make no explicit distinction (Reb et al., 2018; Stoffel et al., 2019). Also, the decoy effect almost always refers to the attraction effect in popular media. In this article, we did not distinguish between the two since our studies involved only the attraction effect. It also helped with the consistency in terminologies as we pre-registered our studies without making the distinction. Nonetheless, we would like our readers to be aware of the other decoy effects and the differences among them.

2. We deviated from this pre-registered plan due to a technical error. Please see the Participants section of Study 1 below.

3. Specifically, we did not select “as in SPSS” but proceeded with the default setting when using G*Power (Faul et al., 2007) to conduct power analyses for tests that involved repeated measures (Lakens, 2013).

Additional information

Funding

The authors received no financial support for the research and/or authorship of this article.