The Marshmallow Test: What the Replication Crisis Revealed About Willpower

One Marshmallow Now or Two Later — and What It Actually Predicts

Few psychology experiments have entered popular consciousness as completely as Walter Mischel's marshmallow test. The premise is elegantly simple: offer a child one marshmallow now, or two if they can wait 15 minutes. Mischel's Stanford studies in the late 1960s and 1970s found that children who waited longer grew up to have higher SAT scores, lower BMI, better educational outcomes, and greater life success by nearly every measure. The findings spawned a self-help industry promoting willpower training and spawned decades of follow-up research. Then, in 2018, a larger and more representative study largely failed to replicate the key long-term predictions — and the results tell a more complicated story about poverty, family stability, and what marshmallow-waiting actually measures.

The marshmallow test is not wrong. It is simply much more contextually constrained than four decades of popular interpretation suggested.

Mischel's Original Research

Walter Mischel and colleagues conducted the original delay of gratification studies at Stanford's Bing Nursery School, primarily with the children of Stanford faculty and staff — a homogeneous, high-socioeconomic-status sample. Children aged 3–5 were placed in a room with a marshmallow (or cookie, or pretzel) and told they could eat it immediately or wait for the experimenter to return and receive a second treat. Some children waited; many did not.

Mischel's longitudinal follow-ups tracked these children into adolescence and adulthood. A 1988 paper found that seconds of delay time at age 4 predicted SAT scores at age 17 with a correlation that seemed to rival standard IQ tests in predictive power. Later follow-ups reported associations with educational attainment, drug use, and even body mass index decades later.

Why the Original Sample Was Problematic

The Stanford Bing Nursery School sample was not representative of the broader population. The children were overwhelmingly white, highly educated middle-class and upper-middle-class — children whose family environments made trusting the experimenter's promise rational. A child from a stable, resource-rich household has good reason to believe that if they wait, the second marshmallow will appear. Waiting is the rational choice when promises are reliably kept.

A child from a less stable background may have learned through experience that promises are not always honored, that food is sometimes scarce and should be taken when available, and that waiting for future rewards that may not materialize is irrational. In this context, eating the marshmallow immediately is not a failure of self-control — it is adaptive risk management.

The 2018 Replication Study

Tyler Watts, Greg Duncan, and Haonan Quan at NYU and UC Irvine analyzed data from 918 children — over 10 times Mischel's sample — and found that while there was a correlation between delay time and age-15 outcomes, the effect was substantially reduced and largely statistically insignificant after controlling for family background, socioeconomic status, and early cognitive ability. The marshmallow test was measuring environmental stability as much as individual self-control.

What the Research Does Support

The replication findings do not mean that self-control is irrelevant or that delay of gratification never predicts outcomes. What the evidence supports with more confidence:

• Self-regulation skills are genuinely important for educational and occupational outcomes and can be developed through practice and supportive environments.
• Trust in the environment is a prerequisite for rational delay — children in less reliable environments are not demonstrating low self-control when they take the immediate reward.
• Poverty itself undermines executive function through stress, cognitive load, and resource scarcity — regardless of any individual personality trait.
• Interventions targeting environment (reducing food insecurity, creating stable routines) may be more effective than training children to wait longer.

Mischel's Response and Legacy

Walter Mischel, who died in 2018 shortly before the replication paper's publication, had himself cautioned against over-interpretation of his work. He consistently emphasized that delay of gratification was a skill that could be taught — he and colleagues developed "hot/cool" cognitive strategies for reframing the tempting object — and he never claimed it was a fixed, innate trait. The popular press interpretation of marshmallow waiting as a fixed character trait was a distortion of his more nuanced claims.

The marshmallow test's enduring value is as a measure of the conditions under which children develop trust, self-regulation, and executive function — not as a fixed predictor of individual destiny.