282 — Revisiting the Marshmallow Test: A Conceptual Replication Investigating Links Between Early Delay of Gratification and Later Outcomes

The world-famous “Marshmallow Test” compared a young child’s ability to delay eating a marshmallow (for a reward) with a longitudinal tracking of their academic and behavioral performance later in life (such as SAT scores). A four-year-old who waited before eating a marshmallow, according to this seminal study, appeared to have a greater chance of achieving higher SAT scores, or performing better in school. What makes the difference between these students and, say, a student who eats the marshmallow right away?

The authors of this paper sampled data from the NICHD SECCYD dataset, polling the “marshmallow performance” of children of mothers who had not completed college (in order to control for premature truncation of recording in the other cohort, as well as to focus on a more commonly-considered “at-risk” population).

The findings are fascinating: When this famous study is reproduced (with slight variation), the authors found that the data for this new population (n=552, which is roughly an order of magnitude larger than the original study) only matched vaguely with the original conclusions: The effects on the larger population were only half as pronounced as on the original dataset, and was reduced even further after controlling for family background and other similar variables.

This suggests that the original conclusions were likely detecting these covariates (like home environment) rather than detecting an attribute of the child. Though there was still a real effect, I wonder how much of it is explainable by other (potentially unmeasured) variables, and how much of it is truly child-dependent.