TL;DR. The regret minimization framework is a one-question test: imagine yourself at 80, looking back, and ask which option you would regret more. Jeff Bezos described it in a 4 May 2001 interview as the tool that made leaving a Wall Street job to start Amazon “an incredibly easy decision”. The evidence behind the idea is real but narrower than the slogan. In the original 1994 studies, 75 percent of 60 telephone respondents regretted inactions more than actions, and 84 percent of 32 adults named an inaction as their biggest lifetime regret. A 2023 replication with 2,600 museum visitors found 49 percent, not 84, and a nationally representative US sample found no significant action gap at all (47.5 versus 52.5 percent). What that representative sample did find is that romance (19.3 percent) and family (16.9 percent) lead the list of regrets, ahead of education and career, which reverses the order found in convenience samples of students and the well educated. People also overestimate how much they will regret anything, so the 80-year-old you imagine is a direction finder, not a measuring instrument. The one experiment that tested acting versus not acting, Steven Levitt’s coin-toss study of more than 20,000 decisions, found that people who made a change on an important question reported being happier six months later. The CEO move is to use the framework only on one-way doors. The student move is to write the two sentences your 80-year-old self would say, then test them against the data below.
This article is a spoke of the personal decision stack, which sets out the full system for deciding well when AI gives you infinite options. That piece explains where a named framework sits in the stack. This one examines a single framework in depth: where it came from, what research says about its central claim, and how to run it without fooling yourself.
Where does the framework come from?
The source is a 4 May 2001 interview Bezos gave to the Academy of Achievement in San Antonio. He described telling his boss in 1994 that he wanted to leave to sell books on the internet. The boss took him on a two-hour walk and convinced him to think about it for 48 hours before making a final decision. Bezos then described the tool he used:
“The framework I found which made the decision incredibly easy was what I called, which only a nerd would call, a regret minimization framework. So, I wanted to project myself forward to age 80 and say, okay, now I’m looking back on my life. I want to have minimized the number of regrets I have.”
He added the claim that gives the framework its bite: “I knew that if I failed I wouldn’t regret that, but I knew the one thing I might regret is not ever having tried.” And he generalised it: “Most regrets, by the way, are acts of omission and not commission.” Leaving a Wall Street firm in the middle of the year meant walking away from the annual bonus, which he presented as the price of the decision.
He returned to the theme in his Princeton Baccalaureate remarks on 30 May 2010, which framed life as a series of choices rather than gifts: “I didn’t think I’d regret trying and failing. And I suspected I would always be haunted by a decision to not try at all.” The speech ends with the sentence most often quoted: “In the end, we are our choices.”
Three claims are packed into the framework, and each can be checked against research.
- In the long run, people regret what they did not do more than what they did.
- You can predict today what your 80-year-old self will regret.
- Once you run the projection, the decision becomes easy.
Claim 1: Do we really regret inactions more?
The claim comes from a line of research that began before Bezos used it. In 1982 Daniel Kahneman and Amos Tversky asked people to compare two investors who each lost the same amount of money, one by switching shares and one by holding. As reported by Gilovich and Medvec in 1994, 92 percent of respondents said the investor who acted would feel more regret. In the short run, action hurts more.
Thomas Gilovich and Victoria Husted Medvec, both at Cornell, then asked whether the pattern reverses over time. Their 1994 paper in the Journal of Personality and Social Psychology reported five studies.
| Study (Gilovich and Medvec, 1994) | Sample | Short-term result | Long-term result |
|---|---|---|---|
| 1a: telephone survey, Ithaca | 60 adults | not asked | 75 percent regretted inactions more |
| 1b: telephone survey, Chicago | 30 adults | not asked | 70 percent regretted their biggest failure to act more |
| 2: open-ended interviews | 77 people (10 professors emeriti, 11 nursing-home residents, 40 undergraduates, 16 clerical and custodial staff) | not asked | failures to act outnumbered actions 63 to 37 percent |
| 3 and 4: Dave and Jim scenario (one transfers university, one stays) | 80 undergraduates; then 34 and 42 | 76 percent said the one who acted regrets more at first | 64 percent (within subjects) and 62 percent (between subjects) said the one who stayed regrets more in the long run |
| 5: own regrets, past week versus lifetime | 32 adults | 53 percent named an action as the week’s biggest regret | 84 percent named an inaction as their biggest lifetime regret |
The 84 percent figure from Study 5 is the number that travelled into speeches and books. It rests on 32 people, 27 of whom chose an inaction. Three later data sets qualify it.
- A nationally representative sample found no action gap. Mike Morrison and Neal Roese surveyed 370 adult Americans by telephone in 2011 (207 women; response rate 20.5 percent). Action regrets were 47.5 percent of the total and inaction regrets 52.5 percent, a difference that was not statistically significant. What did replicate was duration: inaction regrets involved more elapsed time than action regrets (means 7.29 versus 6.44 on the study’s scale, d = 0.41). Inactions are not necessarily more common, but they linger.
- A 2,600-person replication found 49 percent. In 2023 Gilovich and colleagues repeated Study 5 at a Chicago science museum with 2,600 visitors between July 2021 and July 2022. In the short term, 58 percent (743 people) said an action troubled them more and 42 percent (541) an inaction. For “life to this point”, 51 percent (675) chose an action and 49 percent (641) an inaction, a split not different from chance. The direction of the shift over time replicated; the long-run majority for inaction did not. The authors call it a partial replication.
- Other samples land in between. A 2016 postal survey of 500 New Zealand adults found inaction regrets more frequent but action regrets slightly more intense. Daniel Pink’s American Regret Project, a weighted online panel of 4,489 Americans run through Qualtrics in June 2020 for his book and not peer reviewed, found 57.1 percent naming an inaction, 35.9 percent an action and 7.0 percent unsure.
The verdict on Claim 1: the tilt toward regretting inaction is real in most samples, but it is a lean, not a landslide, and it is strongest for the regrets that persist longest. Bezos’s “most regrets are omissions” is defensible as a statement about the regrets that stay with you. It is not a law that applies to every decision.
Claim 2: Can you predict what your 80-year-old self will regret?
This is where the framework is weakest, and the weakness is useful to know.
Daniel Gilbert and colleagues published four studies in 2004 under the title “Looking forward to looking backward”. Students overestimated how much more regret they would feel when they nearly won than when they clearly lost a contest. Subway riders overestimated how much more regret they would feel after missing a train by a narrow margin than by a wide one. The authors conclude that people are less susceptible to regret than they imagine, and that decision makers who pay to avoid future regrets “may be buying emotional insurance that they do not actually need”.
The general mechanism is the impact bias described by Timothy Wilson and Gilbert in 2005: people overestimate the intensity and duration of their emotional reactions to events. Two causes matter for the framework. Focalism makes you underestimate how much everything else in your life will dilute a single outcome. And people fail to anticipate how quickly they will make sense of what happened, which speeds emotional recovery, especially after negative events.
So the imagined 80-year-old is systematically too emotional, in both directions. The projected regret of not trying is probably overstated. The projected regret of trying and failing is probably overstated too. The framework survives this because it is comparative. You are not measuring the regret; you are asking which of two overstated feelings is larger. That comparison is more reliable than either estimate alone, but it means the framework answers “which direction” rather than “how much”.
Claim 3: What do people actually regret?
If the point is to avoid the regrets that persist, it helps to know what those are. Three data sets answer the question, and they disagree in a way that matters for anyone whose regret projection is mostly about career.
Neal Roese and Amy Summerville pooled 11 data sets from nine published articles in 2005: 3,041 participants who reported 4,054 regrets. Women were 73 percent of the sample (2,216 women, 819 men) and the well educated were overrepresented; the authors say so themselves. Morrison and Roese’s 2011 telephone survey is the only nationally representative sample. Pink’s 2020 panel is large and weighted to census estimates but was run for a book, with categories that do not map exactly onto the academic ones.
Table: The Regret Ledger, three samples compared (CEOtudent editorial synthesis)
| Life domain | Roese and Summerville 2005, pooled convenience samples (n = 3,041) | Morrison and Roese 2011, representative US sample (n = 370) | Pink 2020, weighted online panel (n = 4,489; not peer reviewed) |
|---|---|---|---|
| Education | 32.2 percent (rank 1) | 14.0 percent (rank 3) | 15.6 percent (rank 3) |
| Career | 22.3 percent (rank 2) | 13.8 percent (rank 4) | 15.1 percent (rank 4) |
| Romance or partners | 14.8 percent (rank 3) | 19.3 percent (rank 1) | 19.2 percent (rank 2) |
| Parenting | 10.2 percent (rank 4) | 9.0 percent (rank 6) | not a separate category |
| Family | 2.3 percent (rank 8) | 16.9 percent (rank 2) | 21.8 percent (rank 1) |
| Finance | 2.5 percent (rank 7) | 9.9 percent (rank 5) | 13.5 percent (rank 5) |
| Health | 1.5 percent (rank 9) | not reported in the top six | 6.6 percent |
| Education plus career combined | 54.5 percent | 27.8 percent | 30.7 percent |
| Romance plus family combined | 17.0 percent | 36.3 percent | 41.0 percent |
Sources: Roese and Summerville 2005, Table 3; Morrison and Roese 2011; Pink, Results of the American Regret Project. The 2005 and 2011 combined figures are the work-versus-love comparison the 2011 paper reports (54.5 versus 17.0 and 27.8 versus 36.3); the 2020 combined figures are CEOtudent sums of the published category shares. Pink’s “partners” category is placed in the romance row and his “family” category in the family row; the categories are similar but not identical, so read the 2020 column as a check on direction, not a third decimal.
The pattern is hard to miss. In convenience samples dominated by students and graduates, education and career account for more than half of all regrets. In the representative sample, that share halves to 27.8 percent, and relationships take the top two places. The panel data points the same way. Morrison and Roese also found that women were more likely to report romance regrets and men were more likely to report career regrets.
What does this do to the framework? Bezos used it on a career decision, and most people who quote it use it the same way. The data say the regrets that a representative adult carries into old age cluster around partners and family, not around jobs not taken. A career projection that ignores what the move does to the people at home is projecting the wrong variable. The framework is still valid; the input is usually incomplete.
Why do some regrets last and others fade?
Two findings explain which regrets you should be minimising.
Opportunity. Roese and Summerville argued that regret is most intense where opportunity is greatest, because an open door keeps the comparison alive; education tops their list because there is always another course. Denise Beike and colleagues countered in 2009 with the lost opportunity principle: across their studies, the most regretted outcomes were the ones that could no longer be changed and had never been closed off emotionally. Morrison and Roese’s representative sample sided with Beike: 62 percent of regrets were about low-opportunity situations and 38 percent about high-opportunity ones. In practice, the regrets that persist are the ones attached to doors that have already shut.
Age. Carsten Wrosch and colleagues studied 120 adults in Montreal (62 aged 19 to 35, 58 aged 55 to 89) and then 150 adults aged 18 to 85. Older adults perceived fewer opportunities to undo their regrets, and regret intensity predicted lower quality of life only among the older group. In 2012 Stefanie Brassen and colleagues reported in Science that healthy older adults showed reduced responsiveness to regret in a risk-taking task, compared with young adults and with older adults who had late-life depression; disengaging from regret looked like a resilience factor. Both findings cut the same way: the 80-year-old you imagine will, if things go well, be better at letting go than you are now.
Ideal versus ought. Shai Davidai and Gilovich found in six studies in 2018 that people’s most enduring regrets stem more often from gaps between their actual and ideal selves (hopes and aspirations) than from gaps between their actual and ought selves (duties and responsibilities), because people act faster to repair failures of duty. Cornell’s press release on the paper put the split at 72 versus 28 percent across listed regrets, with 76 percent naming an ideal-self regret as their single biggest; those figures come from the university release rather than the paper itself. The implication for the framework: the “did not try” regret Bezos feared is an ideal-self regret, exactly the kind the evidence says persists.
Does acting actually pay off?
Everything above is about how regret feels. One experiment asked whether acting on a hard decision leaves people better off.
Steven Levitt, an economist at the University of Chicago, built a website where people who could not decide let a coin decide for them, then followed up at two and six months. Over 20,000 coins were flipped in the year of data collection; 13,935 people completed the two-month survey and 8,159 the six-month survey. The most common important question was “Should I quit my job?” with 2,186 tosses, followed by “Should I break up with my significant other?” (1,686) and “Should I go back to school?” (1,203).
Three results matter here.
- People did follow the coin, which is what makes the experiment work. On important questions, 56.10 percent reported following the toss at two months and 55.82 percent at six months; on less important questions the rate was above 67 percent. Someone who got heads (make the change) was 26 percentage points more likely to have made the change at two months than someone who got tails.
- In the simple comparison, people who made a change on an important question reported happiness around one full point higher on a 10-point scale, nearly two-fifths of a standard deviation, than those who kept the status quo.
- In the causal estimate that uses the coin as the instrument, the effect of making a change was positive but not significant at two months and large by six months: for all questions, a gain of 0.476 points, more than a fifth of a standard deviation of within-person change, and for important questions about one standard deviation. For less important questions the effect was small, negative and not significant.
Levitt’s own summary is that “people may be excessively cautious when facing life-changing choices”. The limits are worth stating. Everyone in the sample was already torn enough to consult a coin, so the result applies to decisions at the margin, not to every impulse. Happiness is self-reported, and the paper documents some tendency for happy changers to answer the follow-up. Still, it is the only causal evidence on the framework’s core bet, and it points the same way Bezos did.
Where the framework fails
A framework that made one of the most successful business decisions in history looks easy to trust. Four limits keep it honest.
- Survivorship. The framework is famous because the person who used it succeeded. The 1994 decision cannot tell you how often “I would regret not trying” preceded a failed venture, because those stories do not get interviews. The right correction is a base rate; see base rates and the outside view.
- It is a Type 1 tool. In Amazon’s 2015 shareholder letter, filed with the US Securities and Exchange Commission on 5 April 2016, Bezos distinguished one-way doors (consequential and irreversible decisions that must be made methodically and slowly) from two-way doors, which are changeable and reversible and “can and should be made quickly”. Projecting yourself to 80 is heavy machinery. Using it on two-way doors produces exactly the slowness and risk aversion the letter warns against.
- It measures feeling, not outcome. A regret-free life and a good life are not the same thing. The framework says nothing about probability, cost, or the people affected. It needs a companion that does: thinking in bets covers the probability side, and a pre-mortem covers failure modes.
- It can be run too fast. Bezos’s boss asked for 48 hours. The research on slow thinking under pressure explains why that pause is part of the method, not a delay to it.
The Regret Minimization Worksheet (CEOtudent editorial framework)
The following procedure turns the one-question test into a six-step check that incorporates the evidence above. It is an editorial framework, not a validated instrument. Use it on the small number of decisions a year that are genuinely one-way doors.
| Step | What to do | Evidence it rests on | Output |
|---|---|---|---|
| 1. Classify the door | Ask whether the decision can be reversed within a year at a cost you could bear. If yes, stop here and decide quickly. | Amazon 2015 letter: Type 1 versus Type 2 decisions | One-way or two-way |
| 2. Write both sentences | Write what your 80-year-old self would say if you did it, and if you did not. One sentence each, in the past tense. | Bezos 2001; Gilovich and Medvec 1994 on long-run inaction regret | Two sentences |
| 3. Name the regret type | Mark each sentence as action or inaction, and as ideal-self (aspiration) or ought-self (duty). | Davidai and Gilovich 2018; Morrison and Roese 2011 on duration | Four labels |
| 4. Check the window | Ask whether this option will still exist in five years. A closing window raises the weight of the inaction sentence. | Beike and colleagues 2009; Wrosch and colleagues 2005 | Open or closing |
| 5. Discount the forecast | Assume both sentences overstate the feeling. Keep the direction, halve the intensity, and ask whether the comparison still has a clear winner. | Gilbert and colleagues 2004; Wilson and Gilbert 2005 | Clear or unclear |
| 6. Add the missing rows | Write the same two sentences for the people most affected (partner, family), since those are the domains a representative adult regrets most. | Regret Ledger above | Two more sentences |
Decision rule. Proceed with the action if, after step 5, the inaction sentence is still clearly heavier, the window is closing, and the step 6 sentences do not reverse the answer. If the comparison is unclear, the framework has told you something useful: this is not a regret decision, so switch to probability and cost. Record the result in a decision journal with the date, so that your 60-year-old self can check whether your 30-year-old self forecast well. And write your kill criteria before you walk through the door, because the same research that says you will overestimate regret says you will also underestimate how fast you adapt to a wrong turn.
The CEO move and the student move
A chief executive does not use the heaviest decision process on every item. The CEO move is triage: count the one-way doors you face this year, which for most people is somewhere between zero and three, and reserve the 80-year-old projection for those. Everything else gets a fast, reversible call. For the longer horizon those few decisions sit in, see how far ahead you should plan.
The student move is to treat your own regret forecasts as data. Keep the two sentences from every worksheet. After a year, reread them. The literature predicts that the intensity you wrote down will look exaggerated and that the inaction sentences will have aged worse than the action sentences. If your own records say otherwise, you have learned something about yourself that no study can supply. The second-order thinking exercises are a good companion for the rereading.
Frequently asked questions
Did Jeff Bezos invent the regret minimization framework?
He named it. The interview was given on 4 May 2001 and describes the 1994 decision. The research on regret over time, including the finding that inactions are regretted more in the long run, was published by Gilovich and Medvec in 1994 and 1995, and the economics of regret goes back to Bell and to Loomes and Sugden in 1982.
Is it true that most people regret what they did not do?
In the long run, more often than not, but by less than the famous numbers suggest. The original 84 percent came from 32 people; a 2,600-person replication found 49 percent, and a representative US sample found 52.5 percent with no significant gap. The stronger finding is that inaction regrets last longer.
What do people regret most?
It depends on who is asked. Pooled convenience samples put education first (32.2 percent). The only nationally representative sample puts romance (19.3 percent) and family (16.9 percent) first, with education third (14.0 percent) and career fourth (13.8 percent).
Should I make the change if I cannot decide?
The one experiment on this question found that people who made a change on an important decision, such as quitting a job, reported being happier six months later, with a causal effect of about one standard deviation. That evidence covers people who were already at the margin, which is the situation the framework is for.
How accurate are my predictions of future regret?
Not very. People overestimate how much regret they will feel after near misses and bad outcomes, and they underestimate how quickly they will make sense of what happened. Use the projection to compare options, not to measure feelings.
Sources
- Academy of Achievement. Jeff Bezos: Founder and Executive Chair, Amazon. Interview transcript, San Antonio, Texas, 4 May 2001.
- Bezos J. We are what we choose. Baccalaureate remarks, Princeton University, 30 May 2010.
- Amazon.com, Inc. 2015 Letter to Shareholders. Exhibit 99.1 to Form 8-K filed with the US Securities and Exchange Commission, 5 April 2016.
- Gilovich T, Medvec VH. The temporal pattern to the experience of regret. Journal of Personality and Social Psychology. 1994;67(3):357-365.
- Gilovich T, Medvec VH. The experience of regret: What, when, and why. Psychological Review. 1995;102(2):379-395. Abstract.
- Kahneman D, Tversky A. The psychology of preferences. Scientific American. 1982;246(1):160-173. As reported in Gilovich and Medvec 1994.
- Gilovich T, et al. A very public replication of the temporal pattern to people’s regrets. Royal Society Open Science. 2023.
- Morrison M, Roese NJ. Regrets of the typical American: Findings from a nationally representative sample. Social Psychological and Personality Science. 2011;2(6):576-583.
- Roese NJ, Summerville A. What we regret most… and why. Personality and Social Psychology Bulletin. 2005;31(9):1273-1285.
- Pink DH. Results of the American Regret Project. Survey of 4,489 Americans conducted with Qualtrics, June 2020; published with The Power of Regret, 2022. Author’s survey, not peer reviewed.
- Towers AJ, Williams MN, Hill SR, Philipp MC, Flett R. What makes for the most intense regrets? Frontiers in Psychology. 2016;7:1941.
- Gilbert DT, Morewedge CK, Risen JL, Wilson TD. Looking forward to looking backward: The misprediction of regret. Psychological Science. 2004;15(5):346-350. Abstract.
- Wilson TD, Gilbert DT. Affective forecasting: Knowing what to want. Current Directions in Psychological Science. 2005;14(3):131-134.
- Beike DR, Markman KD, Karadogan F. What we regret most are lost opportunities: A theory of regret intensity. Personality and Social Psychology Bulletin. 2009;35(3):385-397.
- Wrosch C, Bauer I, Scheier MF. Regret and quality of life across the adult life span. Psychology and Aging. 2005;20(4):657-670.
- Brassen S, Gamer M, Peters J, Gluth S, Buechel C. Don’t look back in anger! Responsiveness to missed chances in successful and nonsuccessful aging. Science. 2012;336(6081):612-614. Abstract.
- Davidai S, Gilovich T. The ideal road not taken: The self-discrepancies involved in people’s most enduring regrets. Emotion. 2018;18(3):439-452. Abstract, with figures from the Cornell University news release on the study.
- Levitt SD. Heads or tails: The impact of a coin toss on major life decisions and subsequent happiness. NBER Working Paper 22487, August 2016; published in The Review of Economic Studies. 2021;88(1):378-405.
- Zeelenberg M, Pieters R. A theory of regret regulation 1.0. Journal of Consumer Psychology. 2007;17(1):3-18.
- Bell DE. Regret in decision making under uncertainty. Operations Research. 1982;30(5):961-981. Abstract.
- Loomes G, Sugden R. Regret theory: An alternative theory of rational choice under uncertainty. The Economic Journal. 1982;92(368):805-824.
The Regret Ledger, the combined work-versus-love figures for the 2020 panel, the reading of the replication evidence and the Regret Minimization Worksheet are CEOtudent analyses built on the sources above. All other figures are reported as printed in those sources.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
