Moore et al ^{Reference Moore, Carter and van Goozen1} found a ‘novel and robust’ relationship between confectionery consumption during childhood and conviction for violence in adulthood. However, there are serious methodological concerns, which make the overall findings questionable. As the authors recognise, the number of violent people in their cohort is very low. The lack of descriptive information in the paper, contrary to recommendations on the reporting of observational studies, ^{Reference Vandenbroucke, von Elm, Altman, Gotzsche, Mulrow and Pocock2} forces the reader to an exercise of reconstruction. What emerges is that only about 33 participants were violent (0.47% of 6942) and of these, only 23 (69% of the 33 violent individuals) had eaten confectionery excessively. With such numbers, it is highly discouraged in the biostatistical literature to model the probability of being violent using as many parameters (8) as the authors did, since the fit is essentially driven by the number of cases and not by the entire sample size. ^{Reference Harrell, Lee and Mark3} The deficiencies of this approach are well known and numerous, affecting all aspects of the modelling process, from variable selection to effect size estimation, ^{Reference Harrell, Lee, Califf, Pryor and Rosati4} and are not, generally, accommodated by the adoption of rare-events logistical models, which only provide a fix for bias in estimating regression parameters. With such few cases, no interactions have been considered, even though some may be very intuitive (e.g. confectionery consumption and child-oriented parenting). With no serious attempt at considering interactions in the model, the risk of finding spurious associations is well documented (Simpson's paradox). ^{Reference Agresti5} Unfortunately, no details are provided in the paper concerning distribution of the other seven factors included in the multivariable model (gender, late birth, etc.) between violent and non-violent people, so that it is almost impossible to understand how low the cell frequency is in some such combinations. With these considerations in mind, the conclusions suggesting a relationship between confectionery and violence seem an over-interpretation of the fitted model.