In 1969, a psychologist named Walter Mischel ran a study at Stanford that would become one of the most cited — and most misrepresented — experiments in the history of developmental psychology. You probably know the broad outlines: a child is left alone with a marshmallow and told that if they can resist eating it until the researcher returns, they'll get two marshmallows. Some children wait. Some don't.
The results that followed — follow-up studies suggesting that the waiting children had better life outcomes decades later — turned the marshmallow test into a cultural touchstone. It was interpreted as evidence that self-control is a stable, predictive trait: you either have it or you don't, and if you have it young, you'll do better in life.
Recent replications have complicated this picture significantly. And the complications are more interesting than the original story.
What the Replications Found
A 2018 study by Tyler Watts, Greg Duncan, and Haonan Quan re-ran the marshmallow test with a much larger and more demographically diverse sample than the original.¹ The results: when researchers controlled for family background, socioeconomic status, and home environment, the predictive power of delay of gratification shrank dramatically — and for children from lower-income backgrounds, nearly disappeared.
This finding reframed the question. The original studies had used a small, relatively homogeneous sample of children from Stanford's on-campus preschool — a privileged population. When the test was applied more broadly, the ability to wait for the second marshmallow turned out to be heavily influenced by circumstances outside the child's control.
The ability to delay gratification may reflect trust in the environment as much as willpower — a child who has learned that promised rewards sometimes don't materialize has rational reasons to take the sure thing.
This is not a minor methodological footnote. It changes what the marshmallow test actually measures. A child who lives in an unpredictable household — where promises aren't kept, where food security isn't guaranteed, where the future is genuinely uncertain — may eat the marshmallow not because they lack self-control, but because experience has taught them that waiting isn't worth it.
Self-Control as Skill vs. Self-Control as Resource
A separate debate in the psychology literature concerns the nature of self-control itself. The dominant framework for many years was ego depletion — the idea that willpower functions like a muscle that gets tired with use. Make a series of decisions or resist a series of temptations, and your capacity to exercise self-control later in the day diminishes.²
This model generated a huge volume of research and intuitive resonance: it seemed to explain why people make worse dietary choices late at night, why judges give harsher parole rulings before lunch, why difficult workdays produce difficult home evenings.
But here too, replications have been mixed. Large multi-site studies have failed to reproduce many classic ego depletion effects consistently.³ The question of whether willpower is a depletable resource, a learned skill, or primarily a motivational state dependent on beliefs about willpower itself remains genuinely open.
What does seem robust: beliefs about self-control matter. Research by Veronika Job and colleagues found that people who believe willpower is unlimited show fewer signs of depletion than those who believe it is limited — even after performing the same effortful tasks.⁴ If this holds up, it suggests that the mental model you carry about your own capacity to regulate behavior partially creates the capacity.
What Actually Predicts Behavioral Outcomes
If neither the marshmallow test nor ego depletion is as clean as originally believed, what does predict the kinds of outcomes — academic achievement, financial stability, relationship quality — that the self-control literature has tried to explain?
The answer appears to be more contextual than originally thought. Researchers like Angela Duckworth, who has studied grit and self-control extensively, have noted that the relationship between self-regulatory capacity and outcomes is real but significantly mediated by environment, social support, and what psychologists call implementation intentions — specific if-then plans that reduce the need for in-the-moment willpower by automating responses to anticipated situations.⁵
In other words: self-control is less about white-knuckling through temptation and more about designing situations so that temptation requires less resistance. The person who removes junk food from their home isn't exercising more willpower than the person who must resist the snack cabinet every day — they're using less willpower because they've restructured their environment.
The Practical Upshot
The marshmallow test saga is a good case study in how psychological findings travel. A real, careful study gets simplified into a portable story ("delay gratification = succeed"). The story gets applied broadly. Then the replication comes and the story fractures — which gets reported as "self-control doesn't matter," which is also wrong.
What the body of research actually supports is more nuanced: self-regulation matters, but it is not a fixed trait you possess or lack. It is influenced by environment, by trust, by practice, by the design of your daily situations, and by the mental models you carry about your own capacity.
That's a harder story to tell in a single marshmallow. But it's a more honest one — and ultimately, a more hopeful one.
Sources ¹ Tyler Watts, Greg Duncan & Haonan Quan — "Revisiting the Marshmallow Test," Psychological Science (2018) ² Roy Baumeister et al. — "Ego Depletion: Is the Active Self a Limited Resource?" Journal of Personality and Social Psychology (1998) ³ Evan Carter et al. — "A Series of Meta-Analytic Tests of the Depletion Effect," Psychological Science (2015) ⁴ Veronika Job, Carol Dweck & Gregory Walton — "Ego Depletion — Is It All in Your Head?" Psychological Science (2010) ⁵ Angela Duckworth & James Gross — "Self-Control and Grit," Current Directions in Psychological Science (2014)
