Generalization bias is “a cognitive tendency that operates automatically and frequently leads researchers to unintentionally generalize their results from particular samples to broader populations including when their evidence does not warrant it and when the generalization is avoidable.” The limited cognitive resources of brains require individuals to generalize beyond observations, as a heuristic. Over-generalization is not therefore necessarily a bias, but a cognitive necessity. However, whilst science is supposed to help avoid such over-generalizations, it actually does the opposite. One reason is that science is meant to produce explanations, and providing explanations is shown to facilitate broad generalizations, for example when drawing explanations about possible rare exceptions. In psychology this takes the form of the “law of small numbers” shown by Tversky and Kahneman. This over-generalization applies to extrapolating the samples’ population, but is augmented because samples are often not representative of other populations to which results are generalized (e.g. WEIRD samples) or all the study designs they are being generalized to. Socio economic conditions (e.g. cost of producing large and stratified samples, pressure to get grants, …) interact to increase the generalization bias. Irreproducibility follows due to inevitable over-generalizations when scientists: 1) do not consider or understand the full set of variables relevant to the phenomenon; 2) have erroneous preconceptions about the phenomenon. In either case, they overlooked factors that limited the generalizability of the finding.