Skip to main content
Category: Effects
Type: A higher choice share for an option once that option is the middle of the set
Origin: Itamar Simonson, Journal of Consumer Research (September 1989); the restaurant orders are Pia Pinger, Isabel Ruhmer-Krell, and Heiner Schumacher, IZA Discussion Paper 9648 (January 2016)
Also known as: A gain in share for the middle option
Quick Answer — Compromise Effect is a higher choice share for an option once that option is the middle of the set. Itamar Simonson named and tested the pattern in 1989. In one row of his classroom sets, a television at $299 with a picture quality of 80 was chosen by 23 percent of students when it was not the middle and by 48 percent when it was. A later cashier file, from one German restaurant, states a smaller average movement, about five percentage points, and that average is not the classroom pair.

​
What is Compromise Effect?

Compromise Effect is a higher choice share for an option once that option is the middle of the set.
An alternative would tend to gain market share when it becomes a compromise or middle option in the set.
That sentence is Itamar Simonson in 1989. The middle is defined by the options on the page. Add a dearer or a weaker option at one end, and an option that used to sit at the end can become the one between two others. Its share of choices can then rise relative to the option that was already there. The added option does not have to be worse than the middle on both attributes. A set can contain no dominated option and still show the pattern. Decoy Effect is the neighboring pattern in the same paper. There, one added option is dominated by one of the options already on the page, and that dominating option gains. The pattern on this page can run when every option remains undominated. Simonson’s account is that a person who is unsure of the weights will take the alternative that supplies the best reason. The middle can be described as a mix of both attributes. It can also be described as the choice least open to criticism when the audience’s tastes are unknown. Loss Aversion enters only as his sketch: if the middle is the reference, giving up one attribute to reach an extreme can weigh more than the gain on the other attribute. Prospect Theory is the 1979 curve behind that sketch. This page does not carry the 1979 gamble counts.

​
Compromise Effect in 3 Depths

  • Beginner: Three options in a row make the center one feel like the choice you can explain. The center is a position, not a proof that it fits you.
  • Practitioner: Before the three options are on the page, write the attribute you will still want if both ends disappear. Choose from that line, then look at the set.
  • Advanced: The move from 23 percent to 48 percent is one television row among 372 marketing students. The restaurant’s “about five percentage points” is a different average, in a different file. Neither number is a share for the next menu.

​
Origin

The name and the prediction are from Itamar Simonson, “Choice Based on Reasons: The Case of Attraction and Compromise Effects,” Journal of Consumer Research, volume 16, number 2 (September 1989), pages 158–174. The journal records the manuscript as received on 1 January 1988 and revised on 1 March 1989. The byline places him as assistant professor at the School of Business Administration, University of California, Berkeley. He thanks his dissertation co-chairs, Jim Bettman and John Payne, and committee members Joel Huber, Richard Staelin, and Julie Edell. Support came from the Center for Decision Studies and the Fuqua School of Business, both at Duke University. The attraction effect was already in print. Simonson cites Joel Huber, John Payne, and Christopher Puto (1982) for an added option that is dominated by one existing option and raises that option’s share. He then predicts a compromise effect in the other direction from the similarity, or substitution, effect in Huber and Puto (1983). Under similarity, a new neighbor takes more share from the option it resembles. Under the compromise prediction, the new neighbor can take relatively less from the adjacent option, because that adjacent option has become the middle. The pilot used 147 students in a marketing management class. In the high condition they were told that their choices would be evaluated in class and that they might be asked to justify them. In the low condition they were promised confidentiality. Simonson reports a compromise effect with a t-statistic of 4.27, significant at the 0.001 level, and a larger effect in the high condition, with a t-statistic of 2.19, significant at the 0.05 level. Study 1, the share table on this page, came next. A later paper gives the principle a second name. Itamar Simonson and Amos Tversky, “Choice in Context: Tradeoff Contrast and Extremeness Aversion,” Journal of Marketing Research, volume 29, number 3 (August 1992), pages 281–295, define extremeness aversion as a gain in attractiveness for an intermediate option and a loss for an extreme one. They write that the principle can explain earlier findings, including the compromise effect. Extremeness aversion is that 1992 principle. It is not the name of the share comparison on this page. This page does not carry the 1992 camera or computer shares.

​
Key Points

Four readings keep the middle tied to the set that created it. A share in Study 1 is a classroom choice. It is not a purchase, and it is not the restaurant average.
1

Place the option between two others

Simonson’s compromise sets use two attributes, with the range of each attribute printed so the numbers are readable without the other brands. In the television sets the attributes are price and picture quality. In the apartment sets they are distance and general condition. An option is a compromise when it sits between the others on those attributes. The comparison is its share in a set where it is the middle, against its share in a set where it is not.
2

Read the classroom pairs one row at a time

Study 1 asked 372 marketing students at three Southeastern universities to choose and to rate attractiveness during class. Simonson writes that shares were, on average, 17.5 percent larger when the brands were compromises than when they were not, and that nine of eleven differences were statistically significant. The television at 299withpicturequality80movedfrom23percentto48percent.Thetelevisionat299 with picture quality 80 moved from 23 percent to 48 percent. The television at 350 with picture quality 85 moved from 43 percent to 51 percent; that row is marked at the 0.10 level. The apartment at 6 miles with condition 75 moved from 50 percent to 66 percent. A mouthwash scored 90 and 50 on the two effectiveness scales moved only from 36 percent to 38 percent, and that row is one of the two without a significance mark.
3

Keep the justification result on its own study

The pilot’s high condition showed a larger compromise effect. Study 1’s logit puts a compromise indicator beside a utility built from the student’s own importance weights. The coefficients are 1.40 on utility, 0.63 on the compromise indicator, and 0.05 on the interaction with the high condition. The standard errors are 0.09, 0.08, and 0.11. The compromise coefficient has a t-statistic of 7.9. The interaction does not reach significance. The last column of the share table, the high-minus-low gap in the size of the increase, averages 2.6 percent. Simonson notes that the manipulation was weaker in two of the three universities, and that the interaction is significant at the 0.01 level in his own university alone.
4

Ask what reason the protocol actually used

Study 2 asked 100 psychology students to predict how a class would receive each choice. Being the compromise lowered the rated chance of criticism, with a t-statistic of −4.0. The same indicator did not make the choice easier to justify. That result is not significant, and the paper places the test above the 0.20 level. Sixty-five of the 100 said they would be more likely to criticize someone who tends to avoid the middle. Thirty-five said they would be more likely to criticize someone who tends to choose it. In Study 3, protocols that ended on a compromise were longer than the median length 71 percent of the time, against 32 percent for protocols that ended elsewhere. The reasons students spoke included “I’ll take the middle alternative” and “because it is a compromise.”

​
Applications

Use the set, the attribute, and the audience. Keep each percentage inside the count that produced it.

Write the attribute first

Before a short list appears, write the one attribute you would still pay for if the list had two options. A laptop’s battery, a flat’s commute, or a course’s schedule can be that line. The middle of a later list is a position someone can defend. It does not replace the line you wrote.

Name the audience before the reason

Simonson’s high condition told students the class might review the choice. The middle was easier to shield from criticism in Study 2, and it was not rated easier to justify. If you know the other person’s taste, use that taste. If you do not, notice that “it is the middle” is a reason, then check the attribute anyway.

Treat an added end as a share change

When a line gains a dearer or a cheaper end, the option that becomes the middle can gain share relative to the option that was already there. Read that gain as a change in the set. Do not read it as a test that the middle product is better made. The 17.5 percent figure is Simonson’s average across his categories. It is not a markup.

Split a long menu from a three-item card

A menu with many dishes, grouped and ordered by price, is the restaurant file below. Familiar dishes and a large set are the conditions that paper says can shrink the pattern. Do not export the classroom move from 23 percent to 48 percent onto a dish, a ballot, or a survey scale.

​
Case Study

Pia Pinger, Isabel Ruhmer-Krell, and Heiner Schumacher tested the pattern on orders rather than on a questionnaire. The working paper is IZA Discussion Paper 9648 (January 2016). The same text was published in the Journal of Economic Behavior & Organization, volume 128 (2016), pages 14–34. The numbers below are the discussion paper. The restaurant is a specialties restaurant in a rural part of northeastern Germany. The owner asked that the name and the exact place stay off the page. The owner also said that at most 25 percent of customers visit about once a month, and that the others are tourists on a day trip. Fifty-seven percent of the choices fell on weekends. The cashier file runs from 5 January 2002 through 29 May 2009. It holds 88,113 choices of a main dish. There were 21 menus. Each menu listed 24 to 30 main dishes, grouped into six categories: traditional, fish, venison, steaks, poultry, and vegetarian. Inside a category the dishes run from the cheaper to the dearer. The mean price of a chosen main dish is EUR 10.05. The table’s standard deviation is 2.63. The mean number of main dishes on a bill is 3.62. Bills with a single main dish are 7.13 percent of bills. The comparison they treat as closest to an experiment is a pure case: a category gains exactly one dish, at the cheap end or the dear end, and the other dishes keep their prices and their positions. They find nine such cases, covering 16.8 percent of the observations. Three add a cheaper dish. Six add a dearer dish. They assume price and quality move together, because the owner prices by cost plus a markup. That assumption is the owner’s account of the kitchen. It is not a taste test. Their introduction and their conclusion both describe an average increase of about five percentage points in the share of an option that had been at an end and then became a compromise. The same introduction says the strength varies by food category from zero to over one hundred percent. That range is on a percent scale. It is not a second reading of the five points. Table 3 prints a stricter object, the change in absolute share inside the category. That change is positive in eight of the nine cases. The positive values sit between 0.92 and 5.12 percentage points. One case is −4.26. The paper says only two of these absolute differences are significant. The table marks 1.69 points at the 0.05 level and 5.12 points at the 0.10 level. A second measure compares the new middle with the option at the far end of the old menu. Those differences are positive in every case but one, five of them significant, and the positive values run from 2.80 to 30.54 percentage points. The cases whose old menu had four dishes move more, which the authors set beside a 2004 laboratory result they cite and do not re-estimate here. In each of the nine cases, a dish that was already a middle option and then sat even closer to the center lost share. They put that loss at around five percentage points, significant at the 5 percent level in five of the nine cases. The dish-by-dish arithmetic for that loss is not shown. The lesson is a set of menu changes in one unnamed restaurant, an author’s average of about five percentage points, and an absolute-share table that is smaller and not uniformly significant. It is not the classroom move from 23 percent to 48 percent.

​
Boundaries and Failure Modes

Compromise Effect is the wrong name in three retellings. A classroom share becomes a restaurant’s sales. The middle of a price list becomes proof that the middle claim is true. An added dominated option is given this page’s pairs. The first boundary is the questionnaire. The 23 percent and the 48 percent are one television row in Study 1. Students were told to shop for themselves. They did not buy the television. The paper does not print a head count for each version of each set. Three versions and two justification conditions sit inside the 372. The average “17.5 percent larger” pools the categories. One calculator, with 24 functions and a 5 percent repair chance, has non-middle shares of 36 percent and 47 percent and a middle share of 45 percent. That row is not marked significant. An unavailable extreme can still create a middle: a calculator battery at 12 hours and a 2 percent corrosion chance moved from 40 percent to 66 percent, and one at 14 hours and 4 percent moved from 34 percent to 60 percent. Both rows are marked at the 0.05 level. The high-versus-low gap on those two rows is 0 percent. The second boundary is the reason, and whose view is being guessed. Study 2 separates a lower chance of criticism from an easier justification. The open answers call a middle choice safe, and also call it wishy-washy, too eager to please, and a settlement for mediocrity. Study 3 paid 23 first-year graduate students $5 to think aloud in a business-school laboratory. Two who did not keep speaking were dropped. The longer protocols support the claim that choosing the middle went with a harder, more elaborate decision inside that room. They do not score the decision as a better one. Simonson writes that a more thorough process, under accountability, does not by itself produce a better choice. The third boundary is the dining room. The five percentage points are the authors’ average for an extreme option that becomes a compromise. They are not the mean of the nine absolute changes, and they are not a forecast for the next menu. In the full-menu conditional logit, being last in the category cuts choice probability by about 20 percent, while being first and being in the middle are not significant. A later specification that instruments position still finds a negative effect at either end and a positive effect in the middle, each at the 1 percent level. Those coefficients are not printed here. The last-position pattern is significant for bills with more than one main dish, and it is not significant for bills with one. The single-bill estimates are imprecise, and the point estimates are not zero. Group size is the number of main dishes on the bill, not a count of people at the table. The restaurant’s name stays unknown. Choice Overload is a different claim, about the size of the set. The pattern named here is the share of the option that the set has made into the middle.

​
Common Misconceptions

Three retellings travel with the name. The middle share is treated as the right answer. A dominated extra option is given this page’s television. The classroom pair is treated as the restaurant.
A higher share says how often the middle was chosen in a set. It does not say the middle claim is the accurate one. Middle Ground Fallacy is the habit of treating the midpoint of two statements as the truth. A person who picks the center television because it is easier to defend is in the pattern on this page. A person who calls the center statement true because it sits between two others is in that fallacy.
A decoy, on the Decoy Effect page, is dominated by one of the options and raises that option’s share. A compromise set can have no dominated option. The new option sits beyond the old pair, and the option that becomes the middle gains relative to the option that was already there. Simonson treats the two patterns as different reasons. The 23 percent and the 48 percent belong on this page.
The 23 percent and the 48 percent are one television in Study 1, among options the students did not buy. The restaurant sentence is an average of about five percentage points, across menu changes at one unnamed restaurant from 2002 through 2009. Table 3’s absolute changes include a rise of 0.92 points and a change of −4.26 points. Those figures do not move onto a household, a chain, or the next election.
Compromise Effect sits beside a dominated extra option, beside the extra weight of a loss, and beside a curve that makes a loss and a gain different shapes. It also sits beside a large set, and beside the habit of calling the midpoint true.

Decoy Effect

An added option that one existing option dominates, raising that option’s share. A compromise set can contain no dominated option.

Loss Aversion

A greater weight on a loss than on a matched gain. Simonson uses that weight as one sketch of why a move off the middle is hard to defend.

Prospect Theory

The 1979 value curve behind that sketch. This page does not carry the 1979 gamble counts.

Choice Overload

A claim about the size of the set. The restaurant menu is large. The count on this page is the middle option’s share, not the pain of choosing.

Middle Ground Fallacy

Treating the midpoint of two statements as the truth. A higher choice share for a middle product is a different claim.

​
One-Line Takeaway

Write the attribute you would still pay for if both ends of the list disappeared, and use that line before you accept the middle because it is the middle.