In contrast, there are situations in which it is teachers or schools that elect to implement a given treatment, but there is no selection bias that relates to the children. For example, a researcher might want to compare the achievement gains of children in classes using cooperative learning, or schools using comprehensive reform models, to the gains made by demographically similar control groups starting at the same pretest levels. In such cases, random assignment of willing teachers or schools is still far preferable to matching, as matching leaves open the possibility that volunteer teachers or staffs are better than non-volunteers. However, the likely bias is much less than in the case of student self-selection. Aggregate pretest scores in an entire school, for example, should indicate how effective the current staff has been up to the present, so controlling for pretests in matched studies of existing schools or classes would control out much of the potential impact of having more willing teachers. For external validity, it is crucial to note that the findings of a well-matched experiment comparing volunteers to non-volunteers apply only to schools or teachers who volunteer, but the potential for bias is moderate (after controlling for pretests and demographic factors).