Expectation Bias: How What You Expect Changes What You Get

An expectation arrives before the experience and becomes part of it. The usual objection is that people are only being polite, or consistent, when they say the expensive wine was better, and the strongest evidence against that objection is about timing. Leonard Lee, Shane Frederick and Dan Ariely gave pub customers beer that had a few drops of balsamic vinegar in it. Preference for it was highest among those who tasted it blind, and the drop was statistically reliable only for the group told about the vinegar before drinking, not for the group told immediately after tasting. If the information were simply an unpleasant fact weighed against the taste, the moment it arrived would not matter.

The same conclusion comes from the other direction. Hilke Plassmann and colleagues scanned people while they tasted wines labeled with prices and found that activity in the medial orbitofrontal cortex, an area that tracks how pleasant an experience is, was higher when the price was higher, along with the reported pleasantness. The authors noted the obvious worry, that some participants might think it rude to report preferring the cheap wine. Eight weeks later, tasting the same wines with no prices attached, the same people reported no differences, which shows the price information was doing the work. That does not settle the worry on its own, since removing the prices removes the social pressure too; the brain measurements are what argue the pleasantness was experienced and not only stated.

Remove the label and the advantage goes with it. Robin Goldstein and colleagues pooled more than 6,000 blind tastings and found the correlation between price and rating small and negative for drinkers without wine training, which puts the enjoyment in the scanner on the price rather than the liquid. Hilke Plassmann and Bernd Weber later showed that the size of the price effect differs a great deal between individuals, so it is not a fixed feature of taste.

The same mechanism operates on measurements, which gives the bias a second home in research methods. In a review published in the American Journal of Physics, Monwhea Jeng collected cases in which published values of physical constants clustered around the accepted figure and then shifted once a new value was established. Nobody has to be dishonest for this to happen: a reading at the edge of a scale has to be called, an outlier has to be classified as an error or a result, and an expectation decides the close calls. That is why blinding exists.

The paper that put the effect in a scanner is Hilke Plassmann, John O'Doherty, Baba Shiv and Antonio Rangel's "Marketing actions can modulate neural representations of experienced pleasantness," published in the Proceedings of the National Academy of Sciences in 2008. Twenty people, 11 of them men, aged 21 to 30, tasted wine through tubes while lying in an MRI scanner. They were told they were sampling five different Cabernet Sauvignons, identified by their retail prices, and that the study was about the effect of degustation time on flavor.

There were only three wines. Two of them were presented twice, once at their real retail price and once at a false one: one wine appeared as a 90 dollar bottle and as a 10 dollar bottle, the other as 45 dollars and as 5 dollars. Both wines were rated as more pleasant at the higher price, with a reported significance below 0.001, and reported pleasantness correlated with the stated price at 0.59. Ratings of taste intensity, which the price should not touch, did not move.

Two features of the design matter for how much weight the result can carry. The brain data showed higher activity in the medial orbitofrontal cortex and nearby prefrontal regions in the high-price condition for both wines, which is the paper's evidence that the price changed experienced pleasantness and not only what people said. And the behavioral follow-up, run eight weeks later outside the scanner with the same wines and no price information, found no differences in reported pleasantness, which rules out the wines having actually differed; because it removes the prices, it does not separately rule out a reporting effect.

The sample was 20 people, small by the standards now expected of imaging work, and the result should be read alongside the pub experiment of Lee, Frederick and Ariely and Baba Shiv, Ziv Carmon and Dan Ariely's 2005 finding that people who bought a discounted energy drink sold as an aid to mental acuity solved fewer puzzles than people who paid the full price for the same drink. The three studies measure different things, and they agree that the label gets into the experience.

Any label that arrives before an experience can shape it. The rows below are illustrations, not findings from the studies cited on this page.

Expectations cannot be switched off, so every counter works by separating the label from the thing.

• Blind the comparison. A plain cup, an unmarked bottle and an anonymized script are the whole method, and the reason trials are run double blind.
• Decide the rule before the reading. Write down what counts as a positive result, how borderline cases will be classified and what you expect to find, before the data arrive.
• Have someone else take the measurement. A person who does not know which answer is convenient makes the close calls differently.
• Try the cheap one first. If the plain version is as good on its own, the gap you felt was in the label. If it is not, you have learned something about the product.
• Separate enjoyment from judgment. For a meal or a bottle, the expectation is part of what you paid for and it is fair to enjoy it. For a measurement, a grade or a diagnosis, it is a source of error.

Three nearby terms are used almost interchangeably, and the differences are about who holds the expectation and where it acts.

• It is not experimenter's bias, the wider set of ways a researcher's beliefs tilt a study, from the design and the exclusions to the analysis and what gets published. Expectation bias is one route into it, operating at the moment an observation is made.
• It is not the observer-expectancy effect, in which the researcher's expectations reach the participants through cues and change their behavior. There the expectation travels; here it stays with the person doing the judging.
• It is not the placebo effect, the improvement that follows an inert treatment. The two are close relatives, since both run on expectation, but the placebo effect concerns a change in a condition or symptom, while expectation bias concerns a change in a judgment or a reported experience.