Lord’s paradox
an effect in which the relationship between a continuous outcome variable and a categorical independent variable (e.g., treatment or control) is reversed when an additional covariate is introduced to the analysis. For example, suppose that a researcher is studying change in knowledge about a topic following an intervention and obtains results indicating that those who received the treatment had a better outcome (i.e., had learned more) than those in the control group. If, however, the researcher decides to use an analysis of covariance to adjust the treatment effect according to general educational level, the pattern of findings could be reversed: If participants receiving the treatment had a substantially higher level of education than those in the control group, it might now appear that they did not gain from the intervention. [Frederick M. Lord]