A problem that occurs when the available data sample is truncated according to the value of some variable correlated with the dependent variable. For example, pupils in private schools are likely to come from wealthy families. Suppose post-school earnings are used as a measure of quality of a given school. If people with wealthy family backgrounds tend to have higher earnings for reasons other than their schooling, then ignoring this selection leads to the confounding of the effect of family background with the effect of training received at a particular private school. In such cases the ordinary least squares estimator is biased and inconsistent. See also truncated sample.