Predicting adolescent cognitive and self-regulatory competencies from preschool delay of gratification: Identifying diagnostic conditions.

185 children (103 girls, 82 boys) of the original 653 Bing Nursery School subjects; mean age about 4.5 at the delay task, mean 18 years 3 months at follow-up

Stanford university community, United States; upper-middle-class preschoolers.

The marshmallow paper, and the design details are the story. The famous correlations come from ONE experimental cell - rewards exposed with spontaneous ideation - containing 35 children. The predictor was not randomised across cells for predictive purposes. Attrition is severe and non-random - questionnaires were mailed to 506 parents, about 404 probably received them, and 185 responded, a 46% return rate, with responders having delayed 62 seconds longer than non-responders. SAT scores were parent-reported for 94 parents and then matched against test records. Personality-rating subsamples fall to 12 to 35 children, and one highlighted correlation of r = .73 rests on 13 children.

In the diagnostic exposed-rewards cell, delay time correlated with SAT Verbal at r = .42 and SAT Quantitative at r = .57, n = 35, with reported confidence intervals of .10 to .66 and .29 to .76. In the obscured-rewards cell the same correlations were -.12 and -.31, and in the two suggested-ideation cells they ranged from -.40 to -.21. The result on which a generation of policy rested is therefore a cell-specific correlation in 35 children with confidence intervals spanning most of the possible range.

High. An observational predictive correlation in a homogeneous university-community sample, with no control for family background or early ability.

Watts, Duncan and Quan 2018 obtained roughly half the bivariate size in a ten-times-larger and far more varied sample, with the association falling to non-significance under full controls and behavioural outcomes showing essentially no association at all. Benjamin et al. 2020 followed the original cohort to mid-life and found the preschool delay measure predicts none of eleven capital-formation outcomes.