They treat horrifying statistical conditions such as unobserved heterogeneity, selection bias, heteroscedasticity, and autocorrelation.

FORBES: Publication Bias (or, Why You Can't Trust Any of the Research You Read)