A gradient of childhood self-control predicts health, wealth, and public safety

Dunedin - a complete birth cohort of 1,037 children born 1972-73 in New Zealand, followed to age 32 with 96% retention. E-Risk - 2,232 British twins born 1994-95, with sibling fixed-effects models on 509 same-gender dizygotic pairs.

New Zealand birth cohort and a British twin cohort.

The paper that made self-control a policy target, and it contains the within-family analysis that makes it worth taking seriously. The Dunedin self-control composite pools nine measures - researcher-observer ratings at ages 3 and 5 plus parent, teacher and self-reports at 5, 7, 9 and 11 (alpha 0.86) - so it is partly observational and mostly rater report. The age-32 outcomes are largely objective records including clinical and biomarker health measures, income data and court conviction records searched in New Zealand and Australian police systems. The E-Risk sibling models use the same age-5 observational measure and compare dizygotic co-twins, who are no more alike than ordinary siblings.

Across the extremes, comparing the highest and lowest fifths of childhood self-control at age 32 - multiple health problems 11% versus 27%, polysubstance dependence 3% versus 10%, annual income under NZ$20,000 10% versus 32%, criminal conviction 13% versus 43%. Adjusted odds ratios were 1.186 (95% CI 1.038 to 1.355) for substance dependence and 1.714 (1.425 to 2.063) for criminal conviction. THE WITHIN-FAMILY RESULT - among 509 dizygotic twin pairs, the less self-controlled twin at age 5 smoked more by 12 (B = 0.07, p < .03), performed worse at school (B = -0.13, p < .001) and was more antisocial (B = 0.09, p = .007). Controlling sibling IQ differences the school-performance coefficient roughly HALVES to -0.07 (p = .01) while smoking and antisocial behaviour hold.

Medium overall, low for the sibling analysis specifically. The Dunedin gradient is observational and open to the archive's default explanation; the E-Risk co-twin comparison is the strongest within-family evidence in this literature and it survives, though the achievement coefficient halves under IQ control and the horizon is only age 5 to 12.

Mixed. The gradient replicates across cohorts; what does not is any demonstration that it is manipulable. The strongest within-family estimate here is short-horizon and shrinks under IQ control.