Revisiting the Marshmallow Test: On the Interpretation of Replication Results

In a recent paper, Watts, Duncan and Quan (2018; henceforth WDQ) replicate and extend the famous marshmallow studies. Their paper seemingly challenges the original evidence. We acknowledge WDQs contribution, but raise two concerns regarding WDQ’s findings and their interpretation. First, differences in measures between the two studies preclude a direct comparison of results. Most importantly, Monte Carlo Simulations indicate that the correlations reported in WDQ are biased towards zero due to censoring in the length of feasible waiting times. Second, a reduction in correlation size and significance in the presence of covariates which are not entirely exogenous or predetermined is difficult to interpret. We conclude that WDQ’s findings cannot serve to refute the evidence reported by Mischel and Shoda and suggest that assessing the importance of delay of gratification for later life outcomes ultimately requires more refined models of the skill accumulation process as well as targeted intervention studies.