From abstract (article is paywalled)

>Although modest bivariate associations were detected with educational attainment (r = .17) and body mass index (r = −.17), almost all regression-adjusted coefficients were nonsignificant. No clear pattern of moderation was detected between delay of gratification and either socioeconomic status or sex. Results indicate that Marshmallow Test performance does not reliably predict adult outcomes.

I guess the question is whether the covariates that were adjusted for in the regression are true confounders and not, say, something caused by ability to delay gratification.