Compromise effect

Evidence

The basic idea: people choose the option for which they have the best reasons — and the middle can always be justified. With 372 students, Simonson (1989) found that 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 held up.

The justification account, however, only supports the neighbouring effect. Expecting to justify one’s choice to others markedly strengthened the attraction effect alone. For the compromise effect the difference pointed the same way but stayed below the threshold. Simonson’s summary names both effects together; the numbers from Study 1 support that for the attraction effect alone.

Three years later the effect got a different name — and a different explanation. Across 22 experiments, Simonson and Tversky (1992) recast it: the middle does not attract, the edges repel. In choosing, people weigh advantages and disadvantages against the other options, and disadvantages weigh more. The middle option has only small disadvantages, each edge option one large one. Under that name — extremeness aversion — most of the later research has run ever since.

As soon as one of the two axes is price, the aversion works one-sidedly. The authors call this polarization: the cheap, weak option loses; the expensive, strong one does not.

In these experiments a third tier therefore does not shift sales toward the middle but from the cheap option to the expensive one. The authors have no explanation for the asymmetry; they suspect quality describes the goal of a purchase while price is merely the means.

The meta-analysis separates two questions that practice constantly runs together. Neumann, Böckenholt and Sinha (2016) analyse 142 experimental comparisons from 22 papers and 426 effect sizes in all. They distinguish three ways of measuring — the values are log odds, a measure of differences in frequency where 0 means “no difference”.

That last row is the most practically important sentence in the paper. That the middle option is chosen more often than the edges is established across 142 comparisons. That adding a third tier raises the middle tier’s absolute sales is not — the authors call such cases “rather exceptions”. Yet that is exactly the expectation pricing tables are built on.

And the case typical of shops is the weakest one.

The authors work through two extremes: with utilitarian products and non-numeric quality attributes, the middle option is preferred over the edge option in 88 percent of cases; with nondurables and numeric price-quality figures, in only 29 percent. A pricing table made of euro amounts and numbers sits closer to the second case.

The direction of the “non-numeric” feature is notable: with the decoy effect, the effect disappears once attributes are no longer given as numbers. Here it becomes stronger. The two patterns respond in opposite directions to the same design decision.

The effect is not confined to purchase decisions — and that argues against its usual explanation. Trueblood, Brown, Heathcote and Busemeyer (2013) had participants compare rectangles and pick the one with the largest area. All three context effects appeared, the compromise effect among them: of 59 students in the analysis, 66 percent showed it. It is narrow, though — t(58) = 1.97, p = .054, just above the conventional threshold of .05; the authors hold a one-tailed test admissible because the direction was predicted.

The theoretical consequence weighs more than the number. Simonson and Tversky derive extremeness aversion from loss aversion: disadvantages weigh more than advantages. With rectangles, however, there are neither gains nor losses, only height and width. An explanation carrying across consumer goods, inference and perception therefore cannot rest on loss aversion.

Distinction from the decoy: with a decoy, a dominated option does the work; with the compromise effect, it is the middle position. Anyone who confuses the two builds a pricing table wrongly.

Limits

The three measures can contradict one another. Neumann and colleagues record that they not only yield different effect magnitudes but can lead to opposite conclusions about moderators. Anyone reading a single study should check which of the three it reports.

Some moderators rest on few cases. The authors write themselves that their aim is to show that design decisions matter; the individual effect sizes are to be read with care and as suggestive until further studies support them.

A large share of the variation stays unexplained. The full model accounts for 46 percent of the differences in absolute share, 34 percent in relative share and 62 percent in the middle-option share.

Little points to publication bias. The funnel plots show no sign of a selection effect; only for absolute share do unpublished data sit slightly lower (p = .086). The authors call that evidence weak themselves.

The samples are mostly students — but that was tested. In Simonson and Tversky about two thirds were business students at three US West Coast universities, 100 to 220 per experiment; some problems were repeated with executives. In the meta-analysis three quarters of the cases are pure student samples, and that very difference showed no significant effect on the outcome.

Not everything is hypothetical. Simonson and Tversky worked with paper towel samples participants could handle and with catalogue photographs; in several studies participants were told that randomly selected respondents would receive the item they chose, and in the pen study ten percent were actually paid out. Real purchase data from a shop still have not been collected.

People who know what they want are harder to move. The authors expect both context effects to be weaker once customers hold settled preferences — with a habitual brand, they write, the mechanism barely matters.

A follow-up study by Simonson tempers the justification account. A middle position does lower expected criticism, but it does not make the choice easier to justify: respondents also described people who always pick the middle as indecisive.

The perception study was checked from the accepted manuscript, not from the journal typesetting. Volume and page numbers are therefore missing from the source list.