Multilevel model (mixed model)

A multilevel model works with data that sit in groups: children in classes, students in seminars, measurements within people. People in the same group are often more alike than people in different groups, so their values are not independent. The model gives every group its own level (random intercept) and thus separates the differences between groups from those within groups. The effect of a characteristic (fixed effect) is then estimated with the groups taken into account.

Example: A reading test of 720 children from 40 classes: 16% of the variation lies between classes. The characteristic "socioeconomic status" is related to the reading test (b = 2.61, p < .001), even with the class taken into account.

Caution: It needs several groups with several cases each; with fewer than 10 groups the estimates at the group level are very uncertain, from about 30 they are reliable. A multilevel model shows relationships, not causes.