In a four-year-old is left alone with a sweet and promised two for waiting. Years later the ones who managed to wait apparently did better in life, and that became a story about and willpower. In 2018 researchers ran it again on a large, representative sample. What survived?
Much less. The 2018 replication found the link between waiting and later success far weaker, and most of it disappeared once the mother’s education and the state of the family were taken into account. What looked like a trait of character was largely a reflection of a stable environment and of resources. The replication drew criticism of its own, which is in the card below.
The story that stuck from the marshmallow test is simple and charming. A four-year-old holds out against a sweet, and what happens in those three minutes predicts what their success will look like twenty years on. Character as destiny.
The weakness is in the sample rather than the finding. The original ran at a nursery on the Stanford campus, which is to say on a homogeneous group of academics’ children, and in small numbers. When every child comes from the same background, there is no seeing what background does.
The 2018 replication changed exactly that, with ten times larger and far more varied, controlling for the mother’s education and the state of the family. The link between waiting and later success came out much weaker, and most of it vanished once background entered the arithmetic.
The reading that follows is that what looked like character was largely a reflection of a stable environment and of resources. And there is a sequel to it: a child who has learned that adults’ promises hold gains by waiting, and a child who has learned otherwise is simply deciding correctly. It bears saying that the replication took criticism too, on the grounds that some of the it added contain self-control within them and so cancel the thing it meant to measure.
wat. Once the mother’s education and the family’s circumstances went into the arithmetic, a four-year-old’s willpower predicted almost nothing.
wat wat
It changes the practical conclusion: programs that try to teach children willpower on its own should be expected to do less than hoped, next to improving the conditions a child grows up in.
How do we know?
The analysis covered 918 children, with the subgroup at the center of the study numbering 552, roughly ten times the original from 1990, and it controlled for family background.
What is still unknown
The replication drew criticism of its own. Other researchers argued that some of the controls it added, early cognitive ability among them, contain self-control inside them and so cancel out the very thing it set out to measure.
Learn something?
Where to next
Does resisting one temptation really leave you weaker against the next?self-control
What happened when the Dunning-Kruger effect was run on random data?replication
What does the familiar version of the Milgram experiment leave out?replication