When The Conditions Are Essentially Replicated

This is the first article in the series about a technique rather than a bias, and it is here because the technique is sold to small businesses constantly and because the research behind it had a public argument that almost nobody who sells it has read.

The Verdict, Stated First

Five claims, in descending order of confidence.

One. The effect is real in the laboratory and its conditions are narrow. That is not our reading against the original authors; it is their own published position, and the challengers' too.

Two. It appears with numerically represented options and largely not otherwise, on the challengers' finding as quoted by the original authors in their reply.

Three. It can reverse. The decoy can pull share away from the option it was meant to promote, which converts a free trick into a two-sided bet.

Four. There is a serious counter-argument and it cuts the other way. A later study reports the effect much stronger when decisions are binding rather than hypothetical, which is a real criticism of the replications.

Five. The commercial case is thinner than either side's argument. On our own arithmetic a five percent tier shift is wiped out by a 3.75 percent conversion drop, and detecting a three point shift needs about 7,500 buyers split across two versions.

Our Grades For These Claims

Applying the scheme from the first article in this series.

Grade A for the 2014 exchange, whose four papers we can cite precisely and two of whose abstracts we obtained verbatim, including body text of the original authors' reply from a copy hosted at one of their own universities.

Grade B for the challengers' specific findings, which reach us mostly through the original authors' characterisation of them and through a replication-focused website.

Grade A for the counter-finding on binding choices, from an abstract obtained verbatim from the publisher.

Grade A for our own arithmetic, which is standard, on entirely invented commercial parameters.

Our position: this is the rare case where the dispute was conducted openly and both sides published in the same issue, and a reader who takes the trouble can simply read what each side conceded.

A Note On Method

Everything here is verified to August 2026.

We obtained the original authors' 2014 reply as an abstract from the publisher, corroborated by a university research portal, plus body text from a copy hosted at one author's own university[1]. We did not obtain the full reply.

We did not obtain either challenge paper. Their findings reach us through the original authors' characterisation[1] and through a replication-focused website[2], flagged at every use.

We obtained the 2015 counter-study's abstract verbatim from the publisher[3]. We did not obtain the paper.

We did not obtain the 1982 paper, the 1989 paper, the 1995 meta-analysis, or the third 2014 commentary, and report all from citation records[4].

All arithmetic is ours and every commercial parameter in it is invented, which we flag at each use.

This article discusses research on choice. It is not pricing, marketing or commercial advice.

The Effect

What is being claimed, stated plainly before any dispute about it.

You have two options and buyers split between them. You add a third that is clearly worse than one of the two and not clearly worse than the other. The claim is that adding it raises the share going to the option that dominates it, which standard choice theory says should not happen.

Three observations, ours.

The technical name is asymmetric dominance, and the decoy is asymmetrically dominated: beaten by one option on everything, not beaten by the other on everything.

The theoretical violation is the point of the original work. Adding an option nobody chooses should not change how the others split, and the finding was that it does.

And the commercial promise follows only if the effect is large, reliable and survives outside a questionnaire. Those three are separate questions, and the 2014 exchange is about the third.

It is worth being explicit that the first two are not settled either, and this article will not settle them. We obtained no effect size for the attraction effect from any source, which for a series that usually reports one is a conspicuous absence and reflects our failure to obtain the meta-analysis rather than its non-existence.

The 1982 Paper

The source.

Huber, J., Payne, J. W., and Puto, C. (1982), Adding Asymmetrically Dominated Alternatives: Violations of Regularity and the Similarity Hypothesis, Journal of Consumer Research, 9(1), 90–98[4].

We did not obtain it.

Three observations, ours.

The title announces a theoretical purpose rather than a commercial one. Violations of regularity is a statement about axioms of choice, not about menus.

The authors' own 2014 reply confirms that framing, describing the effect "in its historical context as a test of an important theoretical assumption from rational choice theory."[1]

And that distinction runs through everything below. A result can be decisive against a theory and useless as a tactic, because the theory needs the effect to exist and the tactic needs it to be large and reliable.

The Business Version

How the finding travels, and where the distortion enters. Ours.

Four observations.

The version a small business owner receives is a rule: add a third tier nobody buys and more people will choose the expensive one. No conditions attached.

That version is free, which is the source of its appeal. A decoy costs nothing to list, requires no discounting, and promises revenue from customers you already have.

Anything free and effective would already be universal, which is the first thing to notice. Pricing pages are the most tested surface in commerce, and a costless lever with a reliable payoff would not still be presented as a secret.

And the popular version drops every qualification the researchers themselves attach. The rest of this article is those qualifications, taken mostly from the people who found the effect in the first place.

The 2014 Challenge

What happened, and the structure is unusually clean.

In one issue of a marketing journal, four papers appeared: Frederick, S., Lee, L., and Baskin, E. (2014), The Limits of Attraction, Journal of Marketing Research, 51(4), 487–507, DOI 10.1509/jmr.12.0061; Yang, S., and Lynn, M. (2014), More Evidence Challenging the Robustness and Usefulness of the Attraction Effect, 51(4), 508–513, DOI 10.1509/jmr.14.0020; Simonson, I. (2014), Vices and Virtues of Misguided Replications: The Case of Asymmetric Dominance, 51(4), 514–519, DOI 10.1509/jmr.14.0093; and Huber, J., Payne, J. W., and Puto, C. P. (2014), Let's Be Honest About the Attraction Effect, 51(4), 520–525, DOI 10.1509/jmr.14.0208[1][2].

Four observations, ours.

Two challenges and two responses, published together and consecutively paginated. That is a journal deciding an argument should happen in public rather than over years.

The responders are the original 1982 authors and a leading researcher on the related compromise effect. Nobody was speaking for them.

That matters more than it sounds for how a reader should weigh this article. Most disputes this series reports have to be assembled from citing papers, with each side's position reaching us through someone else's summary of it.

Here the concessions are the researchers' own words about their own work, published under their own names in the same issue as the criticism. Where we quote a limitation of the effect below, it is usually the discoverers stating it, which is the strongest form this kind of evidence takes.

The titles carry the temperature. "Let's Be Honest" and "Vices and Virtues of Misguided Replications" are not neutral, and the exchange was clearly felt.

And this series has repeatedly had to reconstruct disputes from citing papers. Here the field staged it in one place, which makes it unusually checkable by anyone.

Numerically, But Not Otherwise

The core finding of the first challenge, quoted by the original authors in their reply.

"Specifically, FLB report that an attraction effect is found 'when stimuli [are] represented numerically, but not otherwise' (p. 488)."[1]

We did not obtain the challenge paper and report this quotation of it by the people it was aimed at, which we regard as unusually reliable sourcing for a critical claim.

Four observations, ours.

The condition is representation, not product category. Options laid out as numbers on two attributes produce the effect; the same choice presented any other way does not.

A replication-focused website describes the range tested: "When products were described qualitatively, when images were shown, when attributes were more complex than two numeric scales, the predicted shift toward the dominating option either failed to appear or sometimes reversed."[2] This is not an academic source, flagged here and at every use.

And the commercial reading is immediate. A pricing page with images, feature lists and qualitative descriptions is the "not otherwise" condition, not the numeric one.

And Sometimes It Reverses

The clause that changes the decision entirely.

The same website reports that the shift "either failed to appear or sometimes reversed (the decoy could pull share away from the option it was supposed to enhance)."[2] Non-academic source, and we did not obtain the paper to check the strength of this claim.

Four observations, ours.

A tactic that sometimes does nothing is a waste of effort. A tactic that sometimes does the opposite is a different kind of thing, and has to be evaluated as a gamble.

Most advice about decoys treats the downside as zero, on the reasoning that nobody buys the decoy so it cannot cost anything. That reasoning ignores both the reversal and the added decision.

We compute the consequence below, and the conclusion is stark. If working and backfiring are anywhere near equally likely, the expected shift approaches zero while the cost of the extra option is paid in full.

And we would flag our own sourcing here as the weakest in the article. A reversal reported through a website, describing a paper we did not read, is thin evidence for the article's most consequential claim.

The Second Challenge

What the companion paper added.

The original authors record: "Yang and Lynn report many other studies that also fail to obtain an attraction effect."[1]

Three observations, ours.

This is a different kind of evidence from the first challenge. Not one careful demonstration of boundary conditions, but an accumulation of failures across studies.

The two together are harder to dismiss than either. One says here is exactly when it works; the other says here are many times it did not.

And again, this reaches us through the original authors' own summary of the paper criticising them, which is about as unfriendly a witness as a critical claim can have.

The Original Authors Respond

The reply, and it is more measured than its title.

Its abstract: "Frederick, Lee, and Baskin (2014) and Yang and Lynn (2014) argue that the conditions for obtaining the attraction effect are so restrictive that the practical validity of the attraction effect should be questioned. In this commentary, the authors first ground the attraction (asymmetric dominance) effect in its historical context as a test of an important theoretical assumption from rational choice theory."[1]

Then the defence: "Drawing on the research reported by scholars from many fields of study, the authors argue that the finding of an asymmetric dominance effect remains robust because it holds when the conditions of the study are essentially replicated."[1]

And the close: "Next, the authors identify some of the factors that mitigate (and amplify) the attraction effect and then position the effect into a larger theoretical debate involving the extent to which preferences are constructed versus merely revealed."[1]

Three observations on that closing sentence, ours, because it names the stake and most readers will skip it.

Constructed versus merely revealed is the real question underneath. If preferences are revealed, they exist before the menu and the menu only displays them. If they are constructed, the menu participates in making them.

That is a much larger claim than a pricing tactic, and it is why the original authors resist the practical framing. Their finding was evidence in a debate about what a preference is, and its usefulness for selling subscriptions was never the point.

And it explains the tone of the exchange. Being told that a result central to a theoretical question lacks practical validity is a category complaint, which is roughly what the title of their reply is objecting to.

Essentially Replicated

The sentence this article is named after, and it deserves close reading. Ours.

Four observations.

The defence is that the effect "holds when the conditions of the study are essentially replicated." That is true and it is a much narrower claim than the one being defended against.

Nobody disputed that the effect appears when you reproduce the original conditions. The challenge was that it does not appear otherwise, and this sentence does not contest that.

So the two sides agree on the empirical position and disagree about what to call it. One says robust, meaning it reproduces under its own conditions. The other says restrictive, meaning it does not survive leaving them. Both are correct.

And for a business reader the disagreement is irrelevant, because the shared finding answers the question. A pricing page is not the conditions of the study essentially replicated, and neither side claims otherwise.

What That Concession Means

Reading the defence as a business owner rather than as a researcher, because the two readings differ. Ours.

Four observations.

For a scientist, "it holds when the conditions are essentially replicated" is a complete defence. The finding was a claim about choice under specified conditions, and it reproduces under those conditions. Nothing more was ever promised.

For a practitioner it is the opposite of a defence. The whole commercial proposition is that the effect leaves the laboratory, and this sentence says nothing about whether it does.

So both sides of the 2014 exchange are answering a question the business reader is not asking. The researchers argue about theoretical robustness; the seller of the tactic asserts field efficacy; and nobody has produced evidence for the second.

And that gap is where this series has repeatedly found trouble. A finding is established narrowly, travels widely, and the qualifications drop off in transit, which the sixty-ninth article documented in a different literature and which is visible here in real time because the researchers wrote it down.

And They Concede More Than That

A passage from the reply's own body, which we obtained from a copy hosted at one author's university.

"As such, the FLB and YL articles contribute to the existing literature by showing not only that the attraction (asymmetric dominance) effect replicates but also that there are moderators of the effect that both decrease and increase the size of the effect. However, the authors go on to argue that the conditions for obtaining the attraction effect, when it is found, are so restrictive that the practical validity of the effect should be questioned."[1]

Four observations, ours.

The clause "when it is found" is the original authors' own, inside their description of the objection, and it concedes that finding it is not automatic.

They also credit the challengers with a contribution rather than dismissing them, which is worth noting given the title of the piece.

The framing they resist is practical validity, not the empirical picture. Their objection is to the conclusion drawn, not to the observations.

And that is the cleanest possible outcome for a reader in our position. When the discoverers and the challengers agree on the facts and differ on the verdict, a third party can take the facts and form their own, which is what the arithmetic below does.

The Counter-Finding

The strongest argument against the challenges, and it did not come from the original authors.

A 2015 paper in a marketing journal: "Researchers have recently strongly questioned the robustness of the attraction effect, according to which adding a decoy option to an existing choice set affects consumers' choice behavior. Tying in with this debate, we identify the persistent use of hypothetical choices in the domain to be a major shortcoming in attraction effect research."[3]

And its result: "In an experiment on the attraction effect with a realistic choice setting that fosters external validity, we manipulate the choice framing by contrasting hypothetical choices with binding choices that entail economic consequences. We find the attraction effect to be much stronger when decisions are binding, underlining the effect's usefulness as a marketing tool."[3]

We obtained the abstract and not the paper, so we report no effect size, sample or design detail.

Hypothetical Versus Binding

Why that finding matters more than its length suggests. Ours.

Four observations.

The criticism is aimed at the entire literature, including the original work. If hypothetical choices are the problem, then the 1982 studies share it.

And the direction is inconvenient for the challengers specifically. The replications used hypothetical choices too, so a finding that the effect strengthens under real stakes weakens the conclusion that it lacks practical validity.

This is the correct shape for a criticism to take, and we would rank it above the exchange it comments on. Rather than arguing about whether the effect exists, it identifies a systematic gap between every study and the situation everyone cares about.

But we would hold it at its weight, which is one study. One paper reporting a moderator does not overturn a literature, and the phrase "underlining the effect's usefulness as a marketing tool" is a conclusion drawn by authors who found what they were looking for.

A Meta-Analysis Exists

A source we can name and did not obtain, which is a real gap in this article.

Reference lists identify Heath, T. B., and Chatterjee, S. (1995), Asymmetric Decoy Effects on Lower-Quality Brands: Meta-Analytic and Experimental Evidence, Journal of Consumer Research, 22(3), 268–284[1].

Three observations, ours.

The title indicates a moderator we have not otherwise discussed: that decoy effects may differ for lower-quality brands, which would be directly relevant to a firm positioning against a premium competitor.

It predates the 2014 exchange by nineteen years, so it cannot speak to the representation finding that the exchange turned on.

And we obtained neither its abstract nor its effect size, which means this article reports a contested literature without the one source that would quantify it. We would rather say that than write around it.

The Reply We Could Not Read

A gap in this article that a reader should know about, since it is one of the four papers in the exchange.

Simonson, I. (2014), Vices and Virtues of Misguided Replications: The Case of Asymmetric Dominance, Journal of Marketing Research, 51(4), 514–519, DOI 10.1509/jmr.14.0093[2].

We obtained the citation and nothing else.

Four observations, ours.

The author is a major figure in this literature and the title indicates a methodological objection to how the replications were conducted, which is a different line of defence from the one we do have.

The word "misguided" suggests the argument is that the replications tested the wrong thing, and if that argument is strong it would weaken this article's conclusion.

We flag it rather than passing over it because a reader is entitled to know that one of the four voices in the exchange is missing from our account, and it is a voice on the side we did not land on.

And the word "virtues" is in the title too, which suggests it is not a wholesale dismissal. We cannot tell you what it concedes, which is exactly the problem with reporting an argument you have only half read.

What Actually Survives

Our reading, stated directly.

Five statements.

The effect is real under its own conditions, and the original authors defend it on exactly that basis: it holds when the conditions of the study are essentially replicated.

Those conditions are numeric two-attribute representation, on the challengers' finding as quoted by the original authors themselves.

Outside them the effect often vanishes and sometimes reverses, on sourcing we grade as the weakest in this article.

The whole literature may understate it, because a 2015 study reports the effect much stronger when choices are binding rather than hypothetical.

And nobody on either side has demonstrated it working on a commercial pricing page, which is where it is universally recommended.

What A Decoy Tier Must Achieve

Turning the question into arithmetic, because the qualitative dispute cannot settle a commercial decision. Our own calculation, and every parameter here is invented.

Adding a third option is meant to shift buyers up a tier. It also adds a decision, and some buyers respond to an extra decision by making none.

The gain per buyer is the share shifted up times the margin gap between tiers. The loss is the fall in conversion times the average margin.

On an invented average margin of $400 and an invented gap between tiers of $300, the gain per buyer is: at a 2 percent shift, $6.00. At 3 percent, $9.00. At 5 percent, $15.00. At 8 percent, $24.00. At 12 percent, $36.00. At 20 percent, $60.00.

The Break-Even Conversion Loss

The number that decides it. Ours, same invented parameters.

The fall in conversion that exactly cancels each gain above: 1.50 percent, 2.25, 3.75, 6.00, 9.00 and 15.00 percent respectively.

Four observations.

A five percent shift up a tier is cancelled by a 3.75 percent fall in conversion. That is a very small amount of added friction to be confident you have avoided.

The general rule is simple and worth carrying. Break-even conversion loss equals the shift times the ratio of the tier gap to your average margin, so a firm whose tiers are close together needs the shift to be nearly free of friction.

And almost nobody measures both. A test that reports tier mix and ignores total conversion cannot answer this question, and that is the test most firms run when they try this.

The parameters are ours and the shape is not. Change the margin and the gap and the numbers move; the requirement to measure both quantities does not.

One case is worth separating out, because it is where the tactic is most defensible. A firm whose tier gap is large relative to its average margin has a wider tolerance for friction, since the gain per shifted buyer is bigger.

A software business selling at ten dollars and a hundred dollars a month is in that position, and a trades business quoting eight hundred against a thousand is not. The same tactic has a break-even conversion loss ten times larger in the first case than the second, which is why advice written for one category travels badly to the other.

The Backfire Arithmetic

What the reversal does to the expected value. Our own calculation.

Suppose the tactic works with some probability, does nothing with another, and backfires by the same magnitude otherwise. On a five percent effect:

Works 70 percent, nothing 20, backfires 10: expected shift 3.00 percent.

Works 50, nothing 30, backfires 20: 1.50 percent.

Works 40, nothing 30, backfires 30: 0.50 percent.

Works 33, nothing 34, backfires 33: 0.00 percent.

Four observations.

At an even three-way split the expected shift is exactly zero and the conversion cost is still paid in full, so the tactic has negative expected value.

The probabilities are entirely ours and no source supplies them. What the literature supplies is only that reversal occurs, not how often.

Which is precisely the problem. To evaluate this as a gamble you need the reversal rate, and nobody has published one, so a firm adopting the tactic is taking a bet whose odds are unstated.

And that asymmetry is worth noticing. The advice is always given as though the downside were zero, when the literature it rests on reports a downside and does not quantify it.

Two things would change our reading here and neither exists as far as we could find. A published reversal rate would turn this from an unquantified gamble into an ordinary decision under risk, and we would then simply compute it.

And a field study on a live commercial page would settle the question the entire 2014 exchange leaves open. The technology to run one has existed throughout the period of the dispute, which is itself worth noticing.

Could You Even Detect It?

The seventy-fourth article's question, applied here. Our own calculation, standard power analysis on an invented 30 percent baseline tier share.

Buyers needed in total, split across two versions, to detect a shift at 80 percent power and a five percent level:

A 2 point shift: 16,788. A 3 point shift: 7,526. A 5 point shift: 2,754. An 8 point shift: 1,100. A 12 point shift: 500.

Four observations.

Detecting a three point shift needs about seven and a half thousand buyers. A firm with a few hundred sales a year cannot run this test in any useful timeframe.

So most businesses adopting this tactic are in the fourth cell of the previous article's grid. Practising a technique in an environment that cannot tell them whether it works.

And the natural response makes it worse rather than better. A firm that adds a decoy and sees revenue rise will attribute it to the decoy, having no way to separate it from the season, the market or anything else that moved.

Two points on that, because it is the mechanism by which this becomes conventional wisdom. The attribution runs one way only. A firm whose revenue falls after adding a tier will look for a cause in the market, not in the tier, because the tier was supposed to be free.

So the feedback a firm receives is filtered by what it expected, which is the sixty-sixth article's structure and the seventy-fourth article's wicked environment operating together on the same decision.

And that combination is worth naming because it is self-sealing. An environment that cannot supply evidence, plus an attribution rule that only fires in one direction, produces a belief no amount of practice will revise. The tactic then survives on testimony, which is what it currently runs on.

Which produces the specific outcome the seventy-fourth article predicted. Confidence in the tactic grows with use and accuracy about it does not, and that is how a contested laboratory finding becomes universal business advice.

The Example Everybody Quotes

A note on the demonstration that carried this idea into business writing. Ours.

Four observations.

Most owners who know this idea know it through a magazine subscription example: three options, one apparently pointless, and a large share choosing the expensive bundle. We are deliberately not reproducing the figures, and the reason is the point of this section.

That demonstration is a classroom exercise with students choosing hypothetically, which is the exact design the 2015 counter-study identifies as the field's major shortcoming. It is not a field result and was never presented as one by its author.

It also has the properties the challengers say are required. Three options, presented numerically, on a small number of attributes, which is the condition where the effect is agreed to appear.

So the famous example is evidence for exactly the narrow claim and not for the broad one. It demonstrates the effect under laboratory conditions and is quoted to justify a tactic outside them, which is the transit failure this article is about.

Your Pricing Page

The application. Ours, untested, and not pricing or marketing advice.

Four points.

The condition under which the effect is agreed to hold is numeric representation on two attributes. A pricing page with feature lists, images and qualitative descriptions is not that.

If you want to test it anyway, the arithmetic says what to measure. Tier mix and total conversion, in the same experiment, because a shift of five percent is cancelled by a conversion drop of under four.

And the power calculation says whether you can. Below a few thousand buyers per version you will not distinguish a realistic effect from noise, and running it anyway produces a belief rather than a finding.

One route around the volume problem is worth mentioning, since it is the only one we can see. The test does not have to run on your own traffic if the structure is common to a trade. Several firms with the same tier shape, running the same two versions and pooling the counts, reach a usable sample where none of them could alone.

We are not aware of anyone having done this and we would read the result with interest. It is the cheapest available answer to a question the research literature has left open for twelve years, and it needs a trade association rather than a laboratory.

Which leaves a defensible position for a small firm. Do not adopt it as a rule, do not attribute revenue changes to it, and do not pay anyone who presents it as settled.

What To Do Instead

Because a reader who came for pricing help should not leave with only a debunking. Ours.

Four points.

The strongest finding in this whole area is not about decoys at all. It is that presentation format changes choice, which both sides of the 2014 exchange agree on and which is the actual content of the numeric-representation result.

That points at something a small firm can act on. Making your options comparable on a small number of explicit attributes changes how they are chosen, independent of any decoy, and it is also just clearer.

The sixty-eighth article's finding is more useful for most owners than anything here. Near the optimum, the payoff surface on a price is nearly flat, which means a few points of price matter less than closing the sale.

And the sixty-fourth article's finding on fairness constrains all of it. Customers hold a norm about what may be passed through and what may not, and a tier structure that reads as manipulation risks the punishment the sixty-eighth article priced.

Two further things a small firm can do that rest on better evidence than the decoy does. Reduce the number of options rather than adding one, since the mechanism the challengers identify, that added complexity breaks the effect, is also a reason added complexity may cost conversion outright.

And make the middle option genuinely the best value rather than apparently so. That requires no behavioural finding to justify, survives a customer working out the arithmetic, and does not depend on a literature whose own authors describe its conditions as narrow.

The Honest Position

Where we land, stated as a position rather than a verdict. Ours.

Four observations.

We are not saying the effect is not real. The people who found it, the people who challenged it, and a later study that criticises both agree it exists.

We are saying the commercial claim is unsupported in a specific way. No study any of these parties cites demonstrates the effect on a real pricing page with real money, and the closest thing, the binding-choice study, criticises the whole literature for that gap.

Our own arithmetic then makes the practical case worse. A modest shift is cancelled by modest friction, the reversal is unquantified, and most firms cannot run the test.

And the reason to publish this rather than pass over it is the asymmetry in how it reaches people. The tactic is presented to small businesses as free and settled, and it is neither, by the account of the researchers who discovered it.

One thing we would not claim, because this series has been careful about it elsewhere. Nothing here shows that any particular firm's tier structure is failing, and a business currently using three tiers has no reason to change on the strength of this article alone.

What it has reason to do is different and cheaper. Stop treating the structure as a proven lever, measure conversion alongside mix if the volume permits, and stop paying for advice that presents a contested finding as a rule. That is a change in confidence rather than in the page.

What To Do

Read what the original authors actually claim. Their defence is that the effect holds when the conditions of the study are essentially replicated, which a pricing page does not do.

Check the representation. The condition both sides agree on is numeric two-attribute presentation, and feature lists with images are the condition where the effect was not found.

Measure conversion and tier mix together. On our own arithmetic a five percent shift is cancelled by a 3.75 percent conversion drop, so a test that reports only mix cannot answer the question.

Price the downside rather than assuming it is zero. The literature reports the effect sometimes reversing and nobody has published how often, so this is a gamble with unstated odds.

Check whether you have the volume. Detecting a three point shift needs roughly 7,500 buyers split across two versions on our own figures.

Do not attribute revenue changes to it. Without a test you cannot separate the tactic from the season, and the belief will strengthen regardless.

Take the general lesson rather than the tactic. Presentation format changes choice, which both sides agree on, and that is worth more than a decoy.

Discount anyone selling this as settled. Two challenges and two responses were published in one journal issue in 2014, and the responses concede the conditions are narrow.

The Limits Of This Analysis

Several caveats matter, and the sourcing here is uneven in a way worth stating up front. This article discusses research on choice and is not pricing, marketing or commercial advice; the applications are our own reasoning and untested. Everything is verified to August 2026. We did not obtain either of the two challenge papers, which are the empirical basis for most of this article; their findings reach us through the original authors' characterisation of them, which we regard as reliable for a critical claim, and through a replication-focused website, which is not an academic source and is flagged at every use. The claim that the effect sometimes reverses, which is the most consequential statement in this article, rests on that website alone, describing a paper we did not read; we grade it the weakest thing here and have said so in the body. We did not obtain the original authors' 2014 reply in full, only its abstract from two sources plus body text from a copy hosted at one author's university. We did not obtain the 1982 paper, the 1989 paper, the third 2014 commentary, or the 1995 meta-analysis, which means this article discusses a contested literature without the one source that would quantify it. The 2015 counter-study is reported from its abstract, so we have no effect size, sample or design detail, and it is one study whose authors draw a conclusion favourable to what they found. All arithmetic is ours and every commercial parameter in it is invented: the average margin, the gap between tiers, the baseline tier share, and above all the probabilities of working, doing nothing and backfiring, which no source supplies and which drive the entire expected-value conclusion. A reader who believes the reversal is rare should redo that table with their own figures.

Related Reading

Behavioural Finance Twenty Years Is Not Evidence Why a firm that cannot run the test will believe the tactic works anyway. Behavioural Finance The Rule Your Customers Already Know The constraint any tier structure has to survive. No. 76 The Bias That Won Fourteen optimisation models across seven datasets, and none consistently beat splitting evenly. The data needed for optimisation to win was around 250...

Frequently Asked Questions

What is the decoy effect?

Adding a third option that is clearly worse than one of two existing options, and not clearly worse than the other, is claimed to raise the share choosing the option that beats it. Standard choice theory says adding an option nobody picks should not change how the others split, which is why the 1982 finding mattered theoretically.

Does it replicate?

Under its own conditions, yes, and that is the original authors' own defence: it holds when the conditions of the study are essentially replicated. The challengers found it appears when stimuli are represented numerically and not otherwise. Both sides agree on the observations and differ on what to call the result.

Can it backfire?

Reportedly yes. A replication-focused website describes the predicted shift sometimes reversing, with the decoy pulling share away from the option it was meant to enhance. That is the weakest-sourced claim in this article and also the most consequential, because it turns a supposedly free tactic into a gamble.

Is there an argument the other way?

Yes, and a good one. A 2015 study argues the persistent use of hypothetical choices is a major shortcoming across the whole literature, and reports the effect much stronger when decisions are binding and carry economic consequences. That criticism applies to the replications and to the original work alike.

What would it need to earn on my pricing page?

On our own arithmetic with invented figures, a five percent shift up a tier is cancelled by a 3.75 percent fall in conversion. The general rule is that break-even conversion loss equals the shift times the ratio of your tier gap to your average margin, and almost nobody measures both quantities in the same test.

Could I test it myself?

Probably not. On our own power calculation, detecting a three point shift in tier mix needs about 7,500 buyers split across two versions. A firm with a few hundred sales a year cannot run that, and running it anyway produces a belief rather than a finding.

So is the whole thing nonsense?

No, and we would not say so. The effect is real, it was theoretically important, and everyone in the dispute agrees it exists under its conditions. What is unsupported is the commercial claim: no study any party cites shows it working on a real pricing page with real money, and the closest thing criticises the literature for exactly that gap.

References

1. Huber, J., Payne, J. W., & Puto, C. P. (2014). Let's Be Honest About the Attraction Effect. Journal of Marketing Research, 51(4), 520–525, DOI 10.1509/jmr.14.0208. Publisher record reproducing the abstract in full, corroborated independently by a university research portal giving the same abstract text, volume, issue and page range: on Frederick, Lee and Baskin (2014) and Yang and Lynn (2014) arguing that the conditions for obtaining the attraction effect are so restrictive that its practical validity should be questioned; on the authors grounding the attraction or asymmetric dominance effect in its historical context as a test of an important theoretical assumption from rational choice theory; on the authors arguing, from research reported by scholars in many fields, that the finding remains robust because it holds when the conditions of the study are essentially replicated; and on the authors identifying factors that mitigate and amplify the effect before positioning it in a larger debate about whether preferences are constructed or merely revealed. Together with a copy of the same commentary hosted at the first two authors' own university, reproducing body text: that Frederick, Lee and Baskin report an attraction effect is found when stimuli are represented numerically but not otherwise, citing page 488; that the two articles contribute to the literature by showing not only that the effect replicates but that there are moderators which both decrease and increase its size; that the authors of those articles go on to argue the conditions for obtaining the effect, when it is found, are so restrictive that practical validity should be questioned; and that Yang and Lynn report many other studies that also fail to obtain an attraction effect. The same copy carries a reference list confirming Frederick, S., Lee, L., and Baskin, E. (2014), The Limits of Attraction, Journal of Marketing Research, 51(August), 487–507; Heath, T. B., and Chatterjee, S. (1995), Asymmetric Decoy Effects on Lower-Quality Brands: Meta-Analytic and Experimental Evidence, Journal of Consumer Research, 22(3), 268–284; and Hedgcock, W., and Rao, A. (2009), Trade-Off Aversion as an Explanation for the Attraction Effect, Journal of Marketing Research, 46(February), 1–13. Note: the publisher's record plus a copy hosted at two of the authors' own university. Our principal source, and the route by which the challengers' central finding reaches this article, quoted by the researchers it was aimed at. journals.sagepub.com
2. Replication-focused website entry on the decoy effect, recording that the first replication paper was Frederick, Lee and Baskin (2014), The Limits of Attraction, Journal of Marketing Research, 51(4), 487–507, DOI 10.1509/jmr.12.0061; that the authors attempted to demonstrate the attraction effect across many product categories and conditions; that their headline finding was that the effect, robust in the original paradigm with numeric two-attribute stimuli, largely disappeared when realistic stimuli were used; that when products were described qualitatively, when images were shown, and when attributes were more complex than two numeric scales, the predicted shift toward the dominating option either failed to appear or sometimes reversed, with the decoy pulling share away from the option it was supposed to enhance; and that the lab paradigm and the application target are different choice environments. The same entry confirms the citations for Yang, S., and Lynn, M. (2014), 51(4), 508–513, DOI 10.1509/jmr.14.0020; Simonson, I. (2014), Vices and Virtues of Misguided Replications: The Case of Asymmetric Dominance, 51(4), 514–519, DOI 10.1509/jmr.14.0093; and Huber, Payne and Puto (2014), 51(4), 520–525. Note: a replication-focused website, not an academic source, flagged at every use. The claim that the effect sometimes reverses rests on this source alone, describing a paper we did not obtain; we grade it the weakest evidence in this article. atticusli.com
3. Publisher record for a 2015 paper in Marketing Letters on what matters in attraction effect research when choices have economic consequences, DOI 10.1007/s11002-015-9394-6, reproducing its abstract: on researchers having recently strongly questioned the robustness of the attraction effect, according to which adding a decoy option to an existing choice set affects consumers' choice behavior; on the authors identifying the persistent use of hypothetical choices in the domain as a major shortcoming in attraction effect research; on an experiment with a realistic choice setting fostering external validity, in which the authors manipulate choice framing by contrasting hypothetical choices with binding choices entailing economic consequences; and on their finding the attraction effect much stronger when decisions are binding, underlining its usefulness as a marketing tool. Note: the publisher's record. We obtained the abstract only, so no effect size, sample or design detail is reported; this is one study, and its authors draw a conclusion favourable to what they found. link.springer.com
4. Reference lists carried on two peer-reviewed journal articles about the attraction effect, confirming Huber, J., Payne, J. W., and Puto, C. P. (1982), Adding asymmetrically dominated alternatives: violations of regularity and the similarity hypothesis, Journal of Consumer Research, 9(1), 90–98; Huber, J., Payne, J. W., and Puto, C. P. (2014), Let's be honest about the attraction effect, Journal of Marketing Research, 51(4), 520–525; Frederick, S., Lee, L., and Baskin, E. (2014), The limits of attraction, Journal of Marketing Research, 51(4), 487–507; Doyle, J. R., O'Connor, D. J., Reynolds, G. M., and Bottomley, P. A. (1999), The robustness of the asymmetrically dominated effect: buying frames, phantom alternatives, and in-store purchases, Psychology and Marketing, 16(3), 225–243; Herne, K. (1999), The effects of decoy gambles on individual choice, Experimental Economics, 2(1), 31–40; and Gomez, Y., Martinez-Moles, V., Urbano, A., and Vila, J. (2016), The attraction effect in mid-involvement categories: An experimental economics approach, Journal of Business Research, 69(11), 5082–5088. Note: reference lists carried on peer-reviewed articles; citations only. We obtained none of the works named, including the 1982 paper this article is about. link.springer.com

This article discusses research on choice and is not pricing, marketing or commercial advice. Neither challenge paper nor the 1982 original was obtained; the challengers' findings reach this article through the original authors' characterisation and through a replication-focused website which is not an academic source. The claim that the effect sometimes reverses rests on that website alone and is graded the weakest evidence here. The 1995 meta-analysis was not obtained, so this article discusses a contested literature without the source that would quantify it. All arithmetic is the authors' own and every commercial parameter in it is invented, including the backfire probabilities that drive the expected-value conclusion.