Decoy effect

The effect began as theory-testing, not as a sales tactic. Huber, Payne and Puto (1982) built it to violate regularity — the axiom that adding an option cannot raise the chance of picking one already in the set — by adding a third option dominated by the target but not by the competitor. It raised the target's share. Their own retrospective admits the first experiments worked for some choice sets and not others, and that they gradually identified the sets that worked. The decoy's share was held to roughly 2% across those studies, which is how you know the dominance was seen: a third tier that takes more than that is not being read as dominated.

The replication record is the part the pricing-page advice leaves out. Yang and Lynn (2014) ran 91 attempts at an attraction effect across 11 experiments and 23 product classes. Eleven produced a significant shift toward the target — 12.1%, more than chance alone, and far fewer than their power analyses predicted if the effect were as large as Heath and Chatterjee's (1995) meta-analysis reported. What separated hits from misses was how the options were described: 9 of 37 numbers-only scenarios worked, against 2 of 54 that distinguished options with a picture or a meaningful word such as a brand name, and the mean share shift was 8.0% against 1.1%.

Frederick, Lee and Baskin (2014) found the same split across 38 studies. Four of their five fully numeric, matrix-style tests produced an attraction effect; of the 27 where at least one attribute had to be seen, tasted or touched, none did. Printed as numbers, winning probabilities lifted the target gamble from 14% to 28%; drawn as shaded wheels they moved nothing, and a 4,033-respondent rerun reproduced both results. A numeric picture-quality rating took the target television from 33% to 57%, while the same choice shown as photographs pushed it the other way, 53% down to 35%, and tasting dilute grape Kool-Aid beside the full-strength version cut the full-strength share from 53% to 39%. If your comparison is carried by images, a dominated neighbour is about as likely to cost the sale as win it.

The Economist subscription story is the most repeated piece of evidence and the weakest. Ariely's Predictably Irrational opens with it: add a $125 print-only option that nobody picks and share of the $125 print-and-web bundle rises from 32% to 84%. Those figures come from his own MBA class. The published version of the stimulus is Kivetz, Netzer and Srinivasan (2004), who put it to 29 students in one classroom and 30 in another and report 43% against 72%. Frederick and colleagues tried twice to reproduce it — picnickers, then 2,003 survey respondents — and found no attraction effect either time; in the larger sample a substantial fraction simply bought the print-only “decoy.” It has never been run as a field experiment, so there is no published pricing-page result to copy.

Huber, Payne and Puto (2014) defended the finding and then conceded the practical point. The effect, they hold, “remains robust because it holds when the conditions of the study are essentially replicated,” and was never offered as a tool for marketing practice. They now “suspect that the asymmetric dominance effect occurs rarely in the marketplace today, at least in its strict regularity-violating form”: real products carry messy attributes, so almost nothing is unambiguously dominated, and dominated products do not survive because nobody buys them. What they say merchants succeed with is the compromise effect — flanking a target with a dearer and a cheaper option. Their sharpest evidence against themselves comes from a commercial conjoint data set: across nearly 4,000 asymmetrically dominated sets the dominating option gained no share, respondents having settled into a rule and deciding in about 12 seconds. Their nearest field evidence is a 1999 supermarket study by Doyle and colleagues, where identical cans of Heinz baked beans priced slightly higher raised the cheaper target's share in real purchases — one display, in one shop.

Recent well-powered work with money on the line finds the effect, and finds it small and two-sided. Dumbalska, Li, Tsetsos and Summerfield (2020) paid participants to pick the best-value rental from three photographs, each with a price and a star rating calibrated to that participant's own valuations. Attraction and compromise decoys were both significant, and so was a “repulsion” decoy that dominates the other target; all three moved preference by about 3 to 5 percentage points, with the attraction and repulsion halves roughly symmetric. Across four preregistered perceptual studies, Spektor, Kellen and Hotaling (2018) “generally observed large repulsion effects, with an attraction effect being observed only under a very specific set of circumstances,” with the display format deciding the sign.

The underlying claim — that people value an option against the set in front of them rather than against a fixed internal scale — is about as well supported as anything in the field. The tactic built on it is not: it works mostly where every attribute is a number, and the people who discovered it think it is rare in real markets. Make the comparison legible and point it at the plan you want chosen, rather than betting a launch on a dominated tier.