The decoy effect in pricing tables: what the research shows

When Pro sits next to Enterprise

Imagine a pricing page with three plans.

Pro is worse than Enterprise in every row but costs almost the same. Compared with Basic, however, Pro is not clearly worse, because it offers more users. Research calls exactly this constellation asymmetric dominance. Pro is the decoy, Enterprise the target, Basic the alternative.

1982: a third option nobody wants

In 1982 Huber, Payne and Puto showed something that ought not to happen. Put a third offer beside two others, one that is clearly inferior to one of them, and that one gets chosen more often — although nothing about it has changed.

Classical decision theory says this cannot occur: an option nobody takes anyway should not shift the choice between the rest. That is precisely why the paper became the starting point of research on the attraction effect.

153 students chose between two or three alternatives across six product classes — cars, restaurants, beer, lotteries, film, television sets — each described on two numeric axes.

The last line is the most awkward: after two weeks practically nothing of the shift was left.

The authors were more cautious than those who came after them. Their results were limited in scope, they wrote, because the choice was restricted to three alternatives on two dimensions; for more complex tasks they expected a weaker effect. For a practical test they suggested catalogues, which could be experimentally supplemented with decoys.

That test was a long time coming. When it came, it turned out differently than hoped.

In pricing the idea is tempting in its simplicity: build a tier so that it looks poor next to the plan you want to sell. Whether that works in practice has been disputed for years.

The debate: lab versus everyday life

The first big counter-finding came in 2014. Frederick, Lee and Baskin tested the effect with things people can actually perceive: drinks that were tasted, facial tissues that could be touched, apartment views shown as photographs. 38 studies in total.

In the 27 studies in which at least one attribute could be directly experienced, they found not a single meaningful attraction effect. They did find a meaningful effect in the opposite direction: for bottled water the target brand’s share fell from 70 to 52 percent as soon as the decoy stood beside it. The authors call this a repulsion effect — the decoy repels rather than attracts.

What decides it is the presentation, not the product. Their television study shows this on one and the same trio of sets. When picture quality was given as a rating (8.0 versus 5.5 versus 3.5), the target option’s share rose from 33 to 57 percent. When the same three sets were shown as photographs instead, the effect disappeared. Across five runs with abstract matrices of numbers it appeared four times.

A second group arrived at the same picture independently. Yang and Lynn (2014) made 91 attempts across 23 product classes.

For scale: chance alone delivers 5 percent. Had the effect been as large as in the older meta-analysis, roughly 29 percent would have been expected. Yang and Lynn note that published attraction effects with non-numeric presentations exist which they did not test.

The original authors objected — more narrowly than the title suggests. Huber, Payne and Puto (2014) do not dispute the boundary conditions. They call the gamble experiments, in which the probabilities were given as numbers, a successful replication, and say they had seen some of the null results in their own experiments. Their objection concerns the value of the effect for theory: the 1982 paper, they write, was a demonstration meant to test an axiom of rational choice theory — not a proposal for marketing practice.

For a pricing page a different sentence of the reply matters most. The authors consider asymmetric dominance rare in today’s market, for two reasons: real offers usually have many attributes rather than two numeric ones, and completely dominated options disappear from an assortment over time.

To this comes a null result of Huber’s own. In a commercial data set, 586 people each made 20 choices among three offers. He identified some 4,000 sets with asymmetric dominance — and, despite extensive analyses, found no gain for the dominating option. His explanation: people who repeatedly assess comparable offers develop settled decision rules and choose in about twelve seconds. That is exactly what reading a pricing table involves.

Three conditions from the reply apply directly when building a table.

One aside in the reply fits the compromise effect: in the laboratory asymmetric dominance is the more robust of the two, while in retail one more often observes successful use of the compromise effect.

An older meta-analysis narrows the effect in yet another way. Heath and Chatterjee (1995) pooled 92 comparisons. The target option’s share rose markedly in one constellation only: when the decoy was dominated and the target was the higher-quality brand. There it was +19.3 percentage points. In every other combination the changes lay between −1 and +4 percentage points and were not meaningful.

And even that pattern does not hold for everyone. Their own experiment with 1,044 participants:

For a pricing page that means the same layout can push a price-sensitive audience down rather than up.

That leaves the argument that pricing tables are a special case. They consist of exactly such numerically comparable rows: price, number of users, storage. Their structure would therefore sit closer to laboratory conditions than a shelf of everyday products — so runs the obvious conjecture.

The closest matching single finding argues against it. Frederick and colleagues twice tried to reproduce the famous Economist subscription example — a plain three-row pricing table — once with 515 and once with 2,003 people. Both times they found no attraction effect. In Yang and Lynn the same table likewise produced no meaningful effect. The authors qualify their own null result: a substantial share of respondents chose the decoy option itself, which complicates interpretation.

So the structural argument does not serve as evidence. It remains a conjecture.

What market data show

Only in 2020 was the effect examined in real sales data. Wu and Cosguner analysed the daily prices and sales of a large US online diamond retailer — February to September 2011, restricted to round stones between $2,000 and $20,000.

Their model separates two things that are otherwise conflated: how often a dominance relationship is noticed at all, and how strongly it works once it is. The two differ sharply.

So the decoy works strongly — but rarely. Across all segments it raised the retailer’s gross profit by 14.3 percent. Those detection rates are model estimates, not measured shares: the authors observe prices and sales, not individual searches.

The study is more than a correlation. The authors test two obvious alternative explanations and reject them: a pure reference-price effect predicts the opposite pattern, and search costs alone do not statistically account for the observed price dispersion. It is still not a randomised experiment.

One detail of the setting matters more for practice than any effect size. The retailer did not build the decoy structure at all. It applies the same markup of roughly 18 to 20 percent to every stone’s wholesale price, and those prices come from dozens of suppliers. The dominance relationships arose as a by-product.

So the study shows that a decoy works when it is noticed. Not that a pricing table built deliberately that way achieves the same.

Practical cases

Not to be confused: decoy and compromise effect

In many SaaS tables the middle tier is the plan you want sold, not the decoy. Then the compromise effect is more likely at work: with three options, people often pick the middle one. Simonson (1989) tested this with 372 students across five product classes. The same option was chosen on average 17.5 percentage points more often when it sat in the middle rather than at an edge; in 9 of 11 comparisons the difference was significant. Again, though, this was measured on hypothetical choices with two numeric axes. And the more recent literature on that pattern narrows it considerably — for pricing tables in particular. What is tested there and what is not is set out on the compromise effect page.

A side finding from the same study is interesting for B2B. Participants who expected to justify their choice to the group showed a stronger attraction effect. Simonson’s explanation: dominance is the only argument that holds up even when you do not know the evaluators’ preferences. For the compromise effect the difference pointed the same way but stayed below significance. And the effect can be overridden: in the beer category it was precisely the participants under pressure to justify who chose the higher-quality brand, even when the cheaper one was dominating.

Mixing up the two effects leads to a badly built table. Ask before designing: which tier should be bought, and what makes it attractive, the dominance of a neighboring tier or its middle position?

Ethics check: nudge or sludge?

Limits and criticism