The Decoy Effect: 91 Attempts Produced 11 Effects, and With Real Products It Ran at Chance

Small popcorn $3. Large popcorn $7. Add a medium at $6.50 and suddenly the large looks like a bargain. You have seen this example - with popcorn, with Economist subscriptions, with software pricing tiers - presented as a law of human choice that works on you every time.

It is one of the most confidently asserted findings in consumer psychology, and it has one of the worst replication records.

What the effect is, and why it mattered

The technical name is the attraction effect or asymmetrically dominated choice. Adding a third option that is clearly worse than one of the existing two - dominated by it on every dimension - increases the share choosing the option that dominates it.

Huber, Payne and Puto (1982), in the Journal of Consumer Research, reported it. It mattered far beyond marketing, because rational choice theory says it should not happen: adding an option nobody picks should not change preferences between the options that were already there. A demonstration that it does was a demonstration that preferences are constructed rather than simply revealed.

For four decades it was one of the standard examples in behavioural economics.

Then two papers arrived in the same issue

In 2014 the Journal of Marketing Research published two papers, back to back, that between them reframed the entire finding.

Frederick, Lee and Baskin (2014), at 51(4), 487-507, made a specific structural argument. Their abstract:

“the authors contend that this phenomenon may be restricted to stylized product representations in which every product dimension is represented by a number (e.g., a toaster oven that has a durability rating of 7.2 and ease of cleaning rating of 5.5). Such effects do not typically occur when consumers experience the product (e.g., taste a drink) or when even one of the product attributes is represented perceptually (e.g., differently priced hotel rooms whose quality is depicted with a photo).”

Their proposed reason is that perceptual representations “do not support the sorts of comparisons that drive the attraction effect”. If you can see the hotel room, you stop doing arithmetic on ratings and start forming an impression, and the dominance relation that produces the effect dissolves.

Their conclusion: “they question the practical significance of the effect.”

Yang and Lynn (2014), at 51(4), 508-513, supplied the count. Their abstract opens with it:

“Ninety-one attempts to produce an attraction effect (involving a total of 23 product classes and 73 different decoyed choice sets) produced only 11 reliable effects-significantly fewer than expected given the statistical power of the studies.”

Eleven out of ninety-one. And their moderator analysis matches Frederick’s argument precisely: “the use of meaningful qualitative-verbal descriptions, as well as pictorial depictions, to differentiate choice options substantially reduced the size of those effects. Indeed, the authors found attraction effects at only chance levels using these types of stimuli.”

At chance levels. With pictures or meaningful descriptions - which is to say, with anything resembling how products are actually presented.

The original authors replied

This is why the episode is worth reading rather than just citing.

Huber, Payne and Puto (2014), at 51(4), 520-525, responded in the same issue, under the title “Let’s Be Honest About the Attraction Effect.” They do not fold, and their defence has a real point.

They ground the effect in its original purpose: it was designed as “a test of an important theoretical assumption from rational choice theory”, not as a marketing tactic. And they argue that “the finding of an asymmetric dominance effect remains robust because it holds when the conditions of the study are essentially replicated.”

Which is a genuine distinction, and also the whole problem. An effect that reliably appears when you reproduce the original abstract, numeric conditions and disappears when attributes are shown as pictures is a real laboratory phenomenon and a poor guide to shop design. They concede boundary conditions - factors that “mitigate (and amplify) the attraction effect” - while maintaining the underlying claim that “preferences are constructed” rather than revealed.

Both sides are right about different things. The theoretical demonstration stands: rational choice theory’s assumption can be violated. The practical claim does not: adding a decoy to a real pricing page is not a reliable lever.

What this changes about pricing pages

Three tiers with an obviously bad middle option is now standard practice on subscription pages, and it is almost always justified by this literature.

On the evidence, that justification is weak in a specific way. The condition under which the effect reliably appears is every attribute expressed as a number. A software pricing table - storage in gigabytes, seats as integers, price in dollars, features as ticks - is unusually close to that condition. It may be the one commercial context where the laboratory conditions are approximately met.

But two things follow, and neither favours the confident version. First, even in the stylised conditions the effect held in only 11 of 91 attempts. Second, the moment a real attribute enters - a screenshot, a description of what the tier is actually for, a photograph, anything perceptual - the moderator analysis says it drops to chance.

So the honest summary for anyone building or reading such a page: the decoy is a documented laboratory phenomenon, it is fragile, it depends on the choice being presented as pure numbers, and the confident popcorn story is a much stronger claim than the evidence supports.

The pattern is the same one that runs through this section. The nine-ending price turns out to be a signal that works on customers without a reference point, not a reading defect. Scarcity cues turn out to be mostly genuine and occasionally regulated. Removing purchase friction turns out to have a small and shrinking effect on spending. In each case the mechanism is real, narrower than advertised, and the confident version circulating online is not the version in the journals.

Limitations and disputes

• The original authors dispute this reading, in print, in the same issue, and their argument that the effect holds under replicated conditions is not unreasonable.
• Huber, Payne and Puto (1982) was not read in the original. It is described here as its own authors characterise it in their 2014 commentary.
• “11 of 91” is a count of attempts, not a formal meta-analysis, and includes both published and unpublished attempts across varied designs.
• We have not verified the Economist subscription example, the most-repeated illustration of the effect, and have not used it.
• Frederick, Lee and Baskin’s studies were read at abstract level; the number of individual studies within the paper is not printed here.
• None of the replication work tested live commercial pricing pages. The inference to software tiers in the section above is ours, drawn from the stated moderators, and is flagged as an inference rather than a finding.
• This is not advice about how to price anything.