What Did the Marshmallow Test Really Show?

The experiments were run by Walter Mischel at Stanford's nursery school in the late 1960s and early 1970s, and the marshmallow was one of several treats on offer; pretzels and cookies were used too. Mischel's original interest was not in predicting the future but in what children do while they wait. The ones who managed it did not stare the marshmallow down. They covered their eyes, sang, kicked the table, turned away, or pretended the treat was a cloud. The lesson he drew was that self-control is a matter of technique, of redirecting attention, rather than of gritting the teeth.

The famous follow-up came later and almost by accident. In the 1980s Mischel tracked down some of the original children, then teenagers, and found that the seconds they had waited correlated with their parents' ratings of their competence and with their scores on university entrance tests. Later follow-ups found associations with body mass index and other outcomes into adulthood. The sample was small, a few dozen children of Stanford academics, and Mischel himself was cautious about it. The public was not.

In 2018 Tyler Watts, Greg Duncan and Haonan Quan re-analysed data on more than nine hundred children from a national sample far more varied in income and background than a university nursery school, and followed them to age fifteen. The correlation between waiting and later achievement was there but was about half the size of the original, and most of it vanished once the researchers controlled for family income, the home environment and the child's early cognitive ability. Among children of the least educated mothers, waiting longer predicted almost nothing.

The reading that emerged is that the marshmallow test was measuring, in large part, the child's circumstances. A child from a stable home where promised treats reliably arrive has good reason to wait. A child who has learned that adults' promises are not kept, or that food may not be there later, is being sensible to eat now. A 2013 study made the point directly: children who had just seen a researcher break a promise waited a quarter as long as children who had seen one kept.

It would be wrong to say the test showed nothing. Self-regulation in childhood does predict later outcomes across many studies, and the ability to delay gratification is part of it. What the replication removed was the idea that a single measure at four fixes a child's fate, and the idea that the trait is a fixed quantity inside the child rather than a skill shaped by what the child has learned to expect. Mischel's own emphasis, on the strategies the waiting children used, has held up well and is the part that can be taught.

The story also became a case study in how a finding travels. The original papers were careful; the summaries were not; the video clips of squirming toddlers were irresistible; and a study of a few dozen privileged children became a policy argument about character education. The 2018 result was widely reported as a debunking, which overstated it in the other direction. The honest summary is a small effect, mostly explained by environment, with a technique inside it worth keeping. What each stage of the research found:

• •1970s: children who distract themselves wait longer; self-control is a strategy
• •1980s and 1990s: in a small sample, waiting time correlates with later achievement
• •2010s: trust in the adult strongly affects how long a child waits
• •2018: in a large diverse sample, the link is half as strong and mostly explained by background

The practical conclusions point away from testing children and towards changing what they can rely on. Keep promises, so that waiting has a track record of paying off. Teach the tricks the successful waiters invented on their own: look away, do something else, think about the treat as a picture rather than a taste. And treat a child who grabs the marshmallow as a child responding to a world, not as a diagnosis. A four-year-old's decision about a sweet is a decision about whether the future can be trusted, and that is something adults can do a great deal about.

If waiting reflects what children have learned to expect, then children raised with different routines should find different things easy to wait for, and research has found exactly that. A study comparing young children in Japan and the United States found that Japanese children waited longer when the reward was food, while American children waited longer when it was a wrapped gift. In Japan, waiting until everyone is served before eating is an everyday habit; in the United States, waiting to open presents is the practised routine. Each group was better at the kind of patience its culture rehearsed.

Social setting matters as well. Experiments have found that children wait longer when told that other children in their group usually wait, or when they know a partner's reward depends on them holding out too. None of this means self-control is an illusion. It means that a single score from one afternoon mixes together a child's skill, their trust in the adult, the familiarity of the situation and the expectations they bring with them. That is why researchers now prefer several measures of self-regulation, taken over time, to one sweet on a table.

The same tension shows up in adult economics, where it is studied as discounting, how much less we value a reward because it arrives later. A typical experiment asks people to choose between a sum of money now and a slightly larger sum later. Many people take the smaller amount today rather than wait a week for a little more. Yet when both options are pushed a year into the future, so the choice is between a sum in 52 weeks and a slightly larger one in 53, most happily choose to wait the extra week. The delay is identical; only its distance from now has changed.

Economists call this present bias, and it helps explain why people mean to save, exercise or study tomorrow and then do not when tomorrow becomes today. The practical answers echo the strategies of the children who waited. Remove the temptation from sight, as some children covered their eyes. Decide in advance, while the reward is still distant, and lock the decision in: automatic transfers into a savings account, or a phone left in another room during revision. These commitment devices work because they let the calmer, far-sighted version of you make the choice before the immediate reward is sitting on the table.

Findings like these are reported as correlations, a number between minus one and one that measures how closely two things rise and fall together, and it is worth seeing what a modest one means in practice. Take a correlation of 0.3 as an illustration. Squaring it gives 0.09, which statisticians read as the share of the variation in one measure that the other accounts for: about nine percent. The other ninety-one percent of the differences between teenagers' test scores would come from everything else, including schools, health, families, friends and plain chance.

Now imagine using the test to predict which children will end up in the top half of their class at fifteen. With a correlation of 0.3, sorting four-year-olds into waiters and grabbers would put them on the right side of the line roughly six times in ten, against five in ten from tossing a coin. That is a real effect, worth studying across thousands of children, and close to useless for judging any one child. The same arithmetic applies to most measures in psychology, which is why a result that matters for policy can still say little about the person in front of you.

A common assumption is that children raised with screens, streaming and on-demand everything must be worse at waiting than their grandparents were. A study published in 2018, led by the psychologist Stephanie Carlson, tested that idea by comparing results from versions of the delay test given to American preschoolers in the 1960s, the 1980s and the 2000s, choosing studies whose procedures were similar enough to compare. The researchers also surveyed hundreds of American adults about what they expected, and most predicted that children today would wait for less time than children in the past.

The opposite turned up. Children tested in the 2000s waited on average about two minutes longer than those tested in the 1960s, and about a minute longer than those in the 1980s. The study could not say why, but the authors pointed to changes that might plausibly help: far more children now attend preschool, where taking turns and waiting are practised daily, and scores on tests of abstract reasoning have risen over the same decades. The result fits the wider picture. Patience at four is not a fixed trait slowly being worn away by modern life, but something that responds to how children are raised and taught.

The marshmallow test found that young children who could wait for a bigger reward used attention-shifting strategies, and a small follow-up suggested they did better later in life. A large 2018 re-analysis found the link much weaker and mostly explained by family background and trust, so the surviving lesson is that self-control is a learnable technique shaped by whether waiting has ever paid off.