One of the issues with problems of reproducibility in the social sciences/psychology is that early studies usually choose WEIRD (white/educated from a rich/indu

One of the issues with problems of reproducibility in the social sciences/psychology is that early studies usually choose WEIRD (white/educated from a rich/industrialised/democractic country) subjects, which are often uni students who are very different from the rest of the world. One article: https://slate.com/technology/2013/05/weird-psychology-social-science-researchers-rely-too-much-on-western-college-students.html https://slate.com/technology/2013/05/weird-psychology-social... Could you not interpret this, in the widest sense, as overfitting interpretations to a 'weird' dataset, with the results that do not generalize, even though the stats (in this case, likely t-test) say everything is fine? In which case overfitting isn't a ML-only problem?