← All blog articles · Play & Learning topic hub
Quick answer: Was the marshmallow test debunked? Not exactly. The famous finding shrank rather than vanished. In a larger, more varied group, waiting at age 4 was only modestly linked to achievement at 15, and the link shrank by about two-thirds once family background and early skills were taken into account (Watts and colleagues, 2018). Follow-ups into adulthood found little or no link (Sperber and colleagues, 2024). These studies describe group averages; they cannot predict your own child's future from one wait.
Key takeaways
- Smaller, not gone. In the 2018 replication, the link to age-15 achievement was about half the size first reported, and it shrank by about two-thirds after accounting for background and early ability.
- Weak link to adult life. Two preregistered adult follow-ups (analysis plans fixed in advance), one co-authored by original Stanford team members, found preschool waiting barely predicted later outcomes.
- Waiting depends partly on trust. In a small 2013 experiment, children waited about four times longer after an adult had kept earlier promises; a 2020 repeat found a smaller effect.
- Kids today are not less patient on this task. US preschoolers tested in the 2000s waited longer than those tested in the 1960s.
The original marshmallow test results
In the late 1960s and early 1970s, Walter Mischel and colleagues tested children at Stanford University's Bing preschool. A child could eat one treat now, or wait alone until the researcher returned and get two. In a 1989 review in Science, Mischel and colleagues reported that, in some versions of the task, preschoolers who held out longer were described years later as more capable teenagers, both academically and socially, and as better at handling stress and setbacks.
Two limits were easy to miss. The long-term findings were correlations, so they could not show that waiting caused anything. And although more than 600 preschoolers took part, the teenage analyses each used only 35 to 89 children, small and highly selective groups from the Stanford University community, as Watts and colleagues (2018) note.
The 2018 marshmallow test replication: smaller, not zero
Tyler Watts, Greg Duncan and Haonan Quan (2018) ran a conceptual replication (same question, new group, slightly different test) in a larger, more diverse US sample, the NICHD Study of Early Child Care and Youth Development, which the authors note is not nationally representative. They analyzed 918 children tested at about 4½ with a 7-minute maximum wait and followed them to age 15. Most analyses focused on the 552 children whose mothers had not finished college, partly because 68% of children with college-educated mothers waited the full seven minutes, leaving too little spread.
- A modest link survived. Each extra minute of waiting went with age-15 test scores roughly 0.1 standard deviation higher (roughly the 50th versus 54th percentile). That raw link was around half the size of the Stanford correlations.
- It shrank with controls. Accounting for family circumstances, the home and early thinking skills cut the link to about a third of its raw size; adding skills and behavior measured at 4½ left a small link that was no longer statistically significant.
- The big split came early. Most of the achievement gap tied to waiting was between children who gave in within 20 seconds and everyone else.
- Behavior barely moved. Links with behavior ratings at 15 were weaker still and mostly not statistically significant.
The replication has critics. Armin Falk, Fabian Kosse and Pia Pinger (2020) argued that the 7-minute cap squeezed differences between children more than the original studies did, pulling the correlation toward zero, and that some controls were not fixed before the test, so they may overlap with what it measures. Limiting controls to things settled beforehand moved their estimates back toward the Stanford figures. Laura Michaelson and Yuko Munakata (2020) ran their own preregistered analysis of the same data, planned before the Watts paper appeared. Using the original studies' simple correlations, they found significant links for three of five teenage outcomes (achievement, problem behavior and social skills). Only the problem-behavior link held once each child's age-4 level was controlled, and social support explained it better than self-control did. Watts and Duncan (2020) replied that the achievement link still partly replicated and that their controls probed what might explain it.
So the dispute is real: the teams agree a link exists but disagree about its size and how much reflects self-control rather than circumstances.
What happened when the children grew up
Two preregistered studies followed children into adulthood.
- The original Stanford children at midlife. Benjamin and colleagues (2020), a team that included Mischel and Shoda, surveyed 113 of the original preschoolers, a small share, in their late 40s about 11 self-reported, mostly financial outcomes, such as net worth, income, high-interest debt and education. Preschool waiting did not predict them (average correlation 0.02). By contrast, self-control questionnaires the same people filled in at 17, 27 and 37, combined into one score, tracked 10 of the 11 outcomes (average correlation 0.19), and the score worked just as well without the preschool wait.
- The replication sample at 26. Sperber and colleagues (2024) followed 702 children tested at 54 months in 1995 and 1996. Small raw correlations appeared with education (0.17) and body mass index (−0.17), but almost all links lost statistical significance once other factors were accounted for. The sample was 83% White, limiting how far the findings travel.
Keeping promises to kids: why trust changes how long they wait
The most useful finding for parents is experimental. Celeste Kidd, Holly Palmeri and Richard Aslin (2013) randomly assigned 28 children aged 3½ to almost 6 to two conditions. First, a researcher twice asked each child to wait for something better: a new set of art supplies instead of a jar of worn crayons, then better stickers instead of one small sticker. The researcher then either came back with them (reliable condition) or apologized and returned empty-handed (unreliable). Then came the marshmallow task, capped at 15 minutes.
Children in the reliable condition waited about 12 minutes on average, versus about 3 in the unreliable one. Nine of 14 children in the reliable group waited the full 15 minutes; only 1 of 14 in the unreliable group did. The authors' reading: a child's wait reflects a reasonable bet about whether a promise will be kept, not only self-control. A child who eats the first marshmallow may be making a sensible call.
Reliable · Average wait: About 12 minutes. The adult came back with the promised art supplies and stickers. Nine of 14 children then waited the full 15 minutes.
This was a small, single-lab study. A 2020 repeat with 60 children aged 3 to 5, run in their schools, found the same direction but a smaller effect, which differed between boys and girls (Moffett and colleagues, 2020). We are not aware of a large, preregistered replication. In a separate small experiment, preschoolers who saw an adult act untrustworthily toward someone else were less likely to wait the full time for that adult's promised reward (Michaelson and Munakata, 2016).
Are kids less patient today?
In a survey of 354 US adults, Stephanie Carlson and colleagues (2018) found that adults tended to expect today's children to wait less. The same study pooled original data from 840 US children aged 3 to 5 who did the same task, with a 10-minute maximum, in the late 1960s, the 1980s or the 2000s. All came from similar, middle to high socioeconomic backgrounds and were matched on age.
Waiting went up, not down: about a minute longer in the 2000s than in the 1980s, and about two minutes longer than in the 1960s. This comparison of existing data could not test the authors' suggested reasons, such as more children attending preschool and more public interest in self-regulation skills, and they call for studies in more diverse groups.
What research on delayed gratification in children does and doesn't show
Broad, repeated measures of childhood self-control do predict adult outcomes, unlike a single wait. In the Dunedin study, Terrie Moffitt and colleagues (2011) followed 1,037 New Zealand children born in 1972 and 1973 to age 32. Researchers, parents, teachers and the children rated self-control nine ways between ages 3 and 11. Lower childhood self-control predicted poorer adult health, weaker finances and more criminal offending, in steps across the whole range, even after accounting for IQ and social class. Among about 500 British fraternal twin pairs followed to age 12, the twin with weaker self-control at 5 was already more likely to have started smoking, to struggle at school and to behave antisocially. This, too, is correlational.
Put together, the evidence does not show that:
- One wait predicts a child's future. Both adult follow-ups found little or no link.
- The test predicts nothing. A small link with teenage achievement survived, and reanalyses argue it may be larger.
- Training a child to win this task changes their life. No study here tested it, and Watts and colleagues (2018) suggested boosting waiting alone, without broader skills, would likely do little.
- A child who grabs the treat lacks willpower. Trust shapes the choice (Kidd and colleagues, 2013; Michaelson and Munakata, 2016).
How to teach kids patience without testing them
No study here tested a home patience program, so treat these steps as reasonable applications, not proven methods. Whether rewards undercut interests kids already have is a separate debate, covered in do reward charts work? A related self-control idea, pretending to be a hard-working hero during a dull task, is covered in why pretend play is important.
Try this at home
- Skip the home marshmallow test. One wait says little about your child and turns snack time into a verdict. Avoid labels like “the impatient one.”
- Keep small promises exactly as stated. If you say “after dinner,” make it after dinner, and if plans change, say so early. Example script: “It’s raining, so the park moves to tomorrow morning. I’m writing it on the calendar.”
- Start with short waits you know you can honor. Set a timer for a minute or two, deliver exactly what you promised, and lengthen waits slowly over weeks.
- Make waiting easier, not a test of willpower. Mischel’s lab studied how children steer their attention while waiting; hiding the treat and offering something else to do apply that idea. Example script: “Let’s put the cookie on the high shelf and finish the puzzle before the timer beeps.”
- Pair the wait with the payoff. Example script: “First bath, then two stories.” Then read both stories, even on a tired night.
- Let older kids save toward something they chose. Keep the price fixed once set. See our guides to allowance by age and non-toy reward ideas.
Where a star chart can fit
Kidd’s study is a good reason to make rewards predictable; a chart is one optional way to do that. In Star Chart Kids, a free web app, you set the real-life rewards. When a child requests one, the stars are reserved until you approve and returned if you decline, and the app never takes earned stars away as punishment. No study has tested a chart as a patience program; what counts is that the adult follows through. Setup takes a username and parent passcode, with no email or payment. Prefer paper? See our reward chart guide.
More research explainers live in our play and learning hub.
This article is general information about published research, not medical advice. If you are concerned about your child's attention, impulse control or development, talk with your pediatrician.
Frequently asked questions
Was the marshmallow test debunked?
Not exactly. A larger 2018 replication found a real but much smaller link between preschool waiting and teenage achievement, which shrank sharply once family background and early skills were accounted for, though critics dispute those controls. Adult follow-ups found little or no link.
What did the marshmallow test replication find?
Watts and colleagues (2018) studied 918 children tested at about age 4½, focusing on 552 whose mothers had not finished college. In that group, the link with age-15 achievement was about half the original size and shrank by about two-thirds once family circumstances, early ability and home life were accounted for.
Are kids less patient today than in the past?
Not on this task. Carlson and colleagues (2018) found that US preschoolers tested in the 2000s waited about 2 minutes longer on average than those tested in the 1960s. The children came from middle to high socioeconomic backgrounds, so the result may not apply to all groups.
Should I worry if my child can't wait for a treat?
One wait tells you very little: preschool waiting time barely predicted adult outcomes, and children wait less when they doubt a promise. If you are worried about attention or impulse control across many settings, talk with your pediatrician.
How can I teach my child patience?
No home method has been tested against long-term outcomes, but sensible steps follow from the research: keep small promises exactly, start with short waits you can honor, move temptations out of sight, and avoid testing or labeling.
Sources and further reading
- Delay of Gratification in Children (Science (Mischel, Shoda & Rodriguez, 1989))
- Revisiting the Marshmallow Test: A Conceptual Replication Investigating Links Between Early Delay of Gratification and Later Outcomes (Psychological Science (Watts, Duncan & Quan, 2018))
- Re-Revisiting the Marshmallow Test: A Direct Comparison of Studies by Shoda, Mischel, and Peake (1990) and Watts, Duncan, and Quan (2018) (Psychological Science (Falk, Kosse & Pinger, 2020))
- Same Data Set, Different Conclusions: Preschool Delay of Gratification Predicts Later Behavioral Outcomes in a Preregistered Study (Psychological Science (Michaelson & Munakata, 2020))
- Controlling, Confounding, and Construct Clarity: Responding to Criticisms of “Revisiting the Marshmallow Test” by Doebel, Michaelson, and Munakata (2020) and Falk, Kosse, and Pinger (2020) (Psychological Science (Watts & Duncan, 2020))
- Predicting mid-life capital formation with pre-school delay of gratification and life-course measures of self-regulation (Journal of Economic Behavior & Organization (Benjamin and colleagues, 2020))
- Delay of gratification and adult outcomes: The Marshmallow Test does not reliably predict adult functioning (Child Development (Sperber, Vandell, Duncan & Watts, 2024))
- Rational snacking: Young children’s decision-making on the marshmallow task is moderated by beliefs about environmental reliability (Cognition (Kidd, Palmeri & Aslin, 2013))
- The influence of environmental reliability in the marshmallow task: An extension study (Journal of Experimental Child Psychology (Moffett, Flannagan & Shah, 2020))
- Trust matters: Seeing how an adult treats another person influences preschoolers’ willingness to delay gratification (Developmental Science (Michaelson & Munakata, 2016))
- Cohort effects in children’s delay of gratification (Developmental Psychology (Carlson and colleagues, 2018))
- A gradient of childhood self-control predicts health, wealth, and public safety (Proceedings of the National Academy of Sciences (Moffitt and colleagues, 2011))