The marshmallow test is one of the most repeated stories in popular psychology. A small child is left alone with a treat and told that waiting will earn a second one. The children who held out, the familiar version goes, grew into teenagers and adults who scored higher, coped better, and generally did well, as if a single act of willpower at four had set the course of a life.
A 2018 study tested that promise against a much larger and more varied group of children, and the promise mostly did not hold. Concentrating on children whose mothers had not completed college, the researchers found that an extra minute of waiting at age four predicted only about a tenth of a standard deviation more achievement at age fifteen. That link was roughly half the size reported in the original work, and it shrank by about two-thirds once the researchers accounted for the child’s family background, early cognitive ability, and home environment.
What the replication actually did
The original results came from a study by Shoda, Mischel, and Peake in 1990, which reported strong correlations between a preschooler’s ability to delay gratification and both achievement and behaviour in adolescence. The sample was small and highly selective, drawn from children in the Stanford University community.
The 2018 team, led by Tyler Watts with Greg Duncan and Haonan Quan, ran what they called a conceptual replication. Rather than re-stage the exact experiment, they drew on the long-running NICHD Study of Early Child Care and Youth Development, which had measured children’s delay of gratification at 54 months and then followed them for years. Their analysis sample included 918 children, and the subgroup they focused on, children whose mothers had not finished college, numbered 552, about ten times the size of the original 1990 study.
That difference in scale matters. A larger and more economically mixed sample makes it possible to ask whether an apparent effect of willpower is really an effect of the circumstances that shape willpower in the first place.
The choice to focus on children of less-educated mothers was deliberate, and it cuts against the suspicion that the researchers were stacking the deck. The full study sample skewed toward advantaged families. The subgroup of children whose mothers had not completed college was, by the authors’ account, more comparable with a nationally representative sample of American children. In other words, the group where the marshmallow effect looked weakest was also the group that looked most like the country as a whole.
Why the story travelled so well
Part of what made the marshmallow test so sticky is how clean it looks. A child, a treat, a timer, and a single number: seconds waited. From that one number flowed a tidy moral that fit decades of advice about character and success, the idea that the capacity to wait is a stable trait, visible early, that quietly shapes the rest of a life.
A finding that simple is easy to repeat and hard to qualify. The original correlations were real, but they came from a small and unusual group of children, and a strong correlation in a narrow sample can look very different once a wider, more varied population is measured. That is the gap the 2018 study set out to probe, and it is why the size and makeup of the new sample do so much of the work.
The link did not vanish, but it thinned
It would overstate the findings to say the marshmallow test was debunked outright. Among children of less-educated mothers, waiting longer at age four was still associated with somewhat better achievement at fifteen. The point is how much of that association survived scrutiny.
Before controls, the correlation was already only half the size of the original studies’ headline numbers. After the researchers adjusted for family background, early cognitive skills, and the home environment, roughly two-thirds of even that smaller link disappeared. What remained was a faint association rather than the strong, life-shaping signal of the popular telling.
The team also noticed that most of the predictive value sat at the very start of the waiting period. Most of the variation in adolescent achievement came from whether a child could wait at least 20 seconds, not from the heroic minutes of resistance that make the test memorable. A child who could hold out briefly looked much like a child who held out for the full delay.
The part the famous version leaves out
The behavioural promise fared worst of all. The original story implied that early self-control predicted not just test scores but a calmer, better-adjusted adolescence. In the replication, associations between delay time and measures of behaviour at age fifteen were much smaller and rarely reached statistical significance.
In other words, the strand of the marshmallow legend most often repeated by parents, that a patient four-year-old becomes a well-behaved teenager, is the strand the larger study supported least.
What this does and does not prove
This is a single conceptual replication, and it should be read as one. Because it used a different sample and a different delay-of-gratification measure than the 1990 study, it is not a perfect like-for-like rerun, and the authors are careful to frame it as a replication and extension rather than a verdict. A follow-up analysis by other researchers later argued that the two studies’ results, read carefully, are less far apart than the headlines suggested.
The findings are also correlational. The study can show that early delay of gratification predicts later outcomes more weakly once background is accounted for; it cannot prove that teaching a child to wait would, or would not, change that child’s future. Self-control may still matter. What the data undercut is the specific, tidy claim that a few minutes of waiting at four reveals a trait that determines how the next decade goes.
And the controls themselves carry the real message. The reason the link thinned so much is that delay of gratification is bound up with the resources around a child: a parent’s education, the cognitive skills already in place by age four, the stability of the home. A test that looks like it measures character may be measuring circumstance.
What is left to sit with
The marshmallow test endures because it offers a clean moral: wait, and you will be rewarded. The larger study does not erase that idea so much as complicate it. A child’s patience at four still says something, but much of what it says is about the world that child was already living in.
That is a less quotable finding, and a harder one to put on a poster. It asks whether we have been admiring willpower when we were looking, much of the time, at advantage.
