Why Does a Great First Visit Disappoint
A great first visit is partly the place and partly the evening. The evening does not repeat, a business book once mistook the same arithmetic for competition, and expectation adds a second cost that has nothing to do with luck.
Answer to the question
Why is the second visit worse than the first visit?
Partly chance. A great first visit mixes a place's usual quality with that evening's good luck, and the luck does not return, so the second visit lands nearer the usual level on average. Expectation adds a separate cost, because the first visit raises the standard the second must meet. Neither requires the place to have got worse.
Key takeaways
- Galton (1886) found that the adult children of unusually tall parents were on average less tall than their parents, and the children of short parents less short. He called it regression towards mediocrity.
- Hotelling's 1933 review argued that the convergence Secrist found among firms was a product of grouping them on an extreme year, not evidence of competition.
- Kahneman (2011, chapter 17) reports Israeli Air Force instructors who saw praised landings followed by worse ones and criticised landings followed by better ones. Coin throwers who received no feedback showed the same pattern.
- Oliver (1980) modelled satisfaction as a function of expectation and of the gap between expectation and outcome. His field study used a flu inoculation, so its application to a second visit is borrowed, not tested.
A great first visit is rarely matched by the second.
The diner usually blames the restaurant. A new chef, tired staff or a cost cut gets the credit. Some of that may be true. Most of the letdown needs none of it, because a great first visit is made of two things and only one of them returns.
What Does a Visit Measure?
A visit records two things at once: what the place is usually like, and what happened that evening. Who was on shift, whether the table was ready and how the kitchen's timing fell all belong to the second.
A first visit good enough to be called great has probably had more than its share of them. The usual quality comes back for the second visit and the good fortune stays behind, so on average the second visit lands nearer the middle. The building does not have to change for that to happen.
The word for this pull is Galton's, and it covers any pair of measurements that are imperfectly related. A first visit and a second visit are two measurements of the same place. The same drift appears among customers, and [Why Your Best Customers Don't Stay Best](https://thewhyweproject.com/articles/why-your-best-customers-dont-stay-best) traces it in heavy buyers.
Daniel Kahneman gives the cleanest demonstration in chapter 17 of Thinking, Fast and Slow. He was lecturing Israeli Air Force flight instructors on the evidence that rewarding improvement works better than punishing mistakes, and one of the most experienced instructors disagreed. The instructor said that after he praised a cadet for a smooth landing the next landing was usually worse, and after he shouted at a cadet for a rough one the next was usually better. The observation was accurate and the inference, that praise harms and shouting helps, was wrong.
Kahneman answered with coins. Each instructor turned his back to a chalk target on the floor and threw two coins at it without looking. Most of those who did best on the first throw did worse on the second, and most of those who did worst improved. Nobody had been praised or shouted at, and the pattern appeared anyway.
"Regression to the mean has an explanation but does not have a cause," Kahneman writes. People look for a cause regardless, and a restaurant offers several. Any of them may be true. The pattern would appear without them. A disputed case is [Is the Dunning-Kruger Effect Real](https://thewhyweproject.com/articles/is-the-dunning-kruger-effect-real), where part of a famous chart is what a noisy measure produces by itself. The same chapter applies the point to the Sports Illustrated cover jinx. An athlete reaches the cover after an exceptional season, probably with a nudge from luck, and a less exceptional season is the likelier sequel.
Regression to the mean has an explanation but does not have a cause.
Who Mistook Regression for Competition?
In 1886 Francis Galton reported that the adult children of unusually tall parents were, on average, less tall than their parents.
In "Regression towards Mediocrity in Hereditary Stature" he used the word for what he saw in heights.
In 1933 Horace Secrist published, through Northwestern University's Bureau of Business Research, a book on firms across many industries. He sorted them into groups by their performance in an early year and followed each group forward.
The top group slid toward the average and the bottom group rose toward it.
Secrist argued that the best-performing firms of one year had drifted toward the average later while the worst had climbed, and he attributed the pattern to competitive pressure from new entrants.
Harold Hotelling reviewed the book for the Journal of the American Statistical Association in 1933 and argued that the convergence was a product of how the groups had been formed, not evidence about competition. A group picked for an extreme result in one year carries that year's chance into the selection. Chance does not repeat on schedule. Sort the same firms by their later results and read the record backwards, and the groups converge in that direction as well.
Do Three Accounts Compete?
Regression is one explanation among three, and they combine rather than compete. Regression concerns the visit and expectation concerns the diner, which is why their effects add. Survivorship concerns the record, and it decides which second visits anyone sees.
| Account | What the second visit is judged against | What it predicts |
|---|---|---|
| Regression | The place's usual quality | A lesser visit on average, with the place unchanged |
| Expectation | The memory of the first visit | A lower rating for the same quality of visit |
| Survivorship | Only the second visits that happen | A record in which declines outnumber improvements |
The size of the regression depends on how much of a visit is luck, and for a given restaurant that share is unknown. Kahneman's shorthand is that success is talent plus luck, and that great success is a little more talent and a lot of luck. On that formula a great first visit is the likeliest kind to have borrowed from chance. A merely good one has borrowed less and should fall less.
The diner's forecast goes wrong in the same direction. In experiments reported in 1972 at the Oregon Research Institute, people's numerical predictions stayed as extreme as the impression they rested on, with no pull toward the average. Kahneman and Tversky later set this inside their account of prediction by representativeness.
Say a diner scores a first visit nine out of ten and expects nine again. A second visit that earns seven then feels like a decline, when the forecast was the error. Kahneman's own golf example gives the correct shape: a player who shot 66 on the first day against a par of 72 should be expected to do well on the second, but less well.
Where Do Recoveries Go?
The arithmetic is symmetric. A poor first visit is also partly bad luck, and the second should on average be better. Kahneman's coin throwers showed both halves. The half rarely seen with restaurants is the improvement. This is reasoning rather than a measured result: few people book a second visit to a place that disappointed them, so the recovery that regression predicts for that group mostly happens out of sight.
What reaches the record is the group that returned after a great first visit, and in that group the pattern runs one way. Survivorship bias is the name for a record that holds only the cases that continued. A ratings page is built from such a record, and [Why We Trust Online Ratings Anyway](https://thewhyweproject.com/articles/why-we-trust-online-ratings-anyway) looks at what readers make of it.
A comparison would separate regression from real decline. Record second visits for poor first visits as well as great ones, whether or not the diner chose to return. Regression predicts that the great group falls and the poor group rises toward the same middle, by amounts set by how much luck a visit carries. A fall in the great group with no matching rise in the poor one points to something else.
Expectation survives all of that. Richard Oliver proposed in 1980 that satisfaction is a function of expectation and of disconfirmation, the gap between what happened and what was expected. His field study followed consumers and non-consumers of a flu inoculation, so the model is borrowed here, not tested on diners. Applied to a second visit it adds a cost that needs no luck at all.
Even if every visit measured a place exactly, a diner who was delighted once judges the next visit against that memory. The disappointment is real in the diner, whatever the arithmetic says about how that first evening happened to fall. In Oliver's terms, the second visit is scored on its distance from that memory as well as on its own quality.
Common questions
Why is the second visit worse than the first visit?
Partly chance. A great first visit mixes the place's usual quality with that evening's good luck, and the luck does not return, so the second visit lands nearer the usual level on average. A second cause is expectation: the first visit raises the standard the second has to meet. Neither requires the place to have got worse.
Does a poor first visit improve on the second?
On average the same arithmetic says it should. It is seldom seen because few people return to a place that disappointed them. That is an inference from how the pattern works, not a measured restaurant result.
Does this mean the restaurant has not changed?
No. Regression says a letdown is expected even when nothing has changed. It cannot show that nothing has. Telling the two apart takes second visits recorded for poor first visits too, set against the great ones.
Sources
Galton, F. (1886). 'Regression towards mediocrity in hereditary stature.' Journal of the Anthropological Institute of Great Britain and Ireland, 15, 246-263.
Secrist, H. (1933). The Triumph of Mediocrity in Business. Evanston, IL: Bureau of Business Research, Northwestern University.
Hotelling, H. (1933). Review of 'The Triumph of Mediocrity in Business' by Horace Secrist. Journal of the American Statistical Association, 28(184), 463-465.
Kahneman, D. (2011). Thinking, Fast and Slow. New York: Farrar, Straus and Giroux, chapter 17.
Tversky, A. (1972). 'On the psychology of prediction.' Oregon Research Institute Research Bulletin, 12(4). University of Oregon Scholars' Bank.
Kahneman, D., & Tversky, A. (1982). 'On the psychology of prediction.' In D. Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment under Uncertainty: Heuristics and Biases (pp. 48-68). Cambridge University Press. doi:10.1017/CBO9780511809477.005
Oliver, R. L. (1980). 'A cognitive model of the antecedents and consequences of satisfaction decisions.' Journal of Marketing Research, 17(4), 460-469. doi:10.1177/002224378001700405