Figure 2

Population model of a situation where RTM occurs. Population model where the baseline association between health and predictor is entirely due to causes with transient effects on health. When the transient effect in the current figure comes from factors with variance > 0, it contributes to variance in change and thus to rank-order instability and RTM in health. These transient causes also affect the predictor, and RTM in health will therefore be correlated with the predictor. Thus, persons with different scores on the baseline predictor tend to regress toward the same mean on the health variable. To avoid attributing RTM in health to the predictor, the statistical method used should include an assumption of RTM. A more technical explanation of why MR is appropriate in this situation is provided in Additional file 1. The circle symbolizes a latent factor that is not observed by the researcher. Squares are observed variables used in the analyses. To simplify the figure, residual variances of observed variables are not drawn. This model corresponds to the model in Judd & Kenny [24] (p.111) where the allocation variable to control versus intervention group has time-limited effects on test scores.