Figure 3

Population model of a situation where RTM does not occur. Population model where the baseline association between health and predictor is due to causes with enduring effects on health. The sum of the indirect and direct effects from these causes on follow-up health is equal to their direct effect on baseline health. The effect from the enduring causes on health implies that this latent variable does not contribute to rank-order instability, and hence not to RTM in health. The statistical method used should therefore not include an assumption of RTM. A more technical explanation of why change score analysis is appropriate in this situation is provided in Additional file 1. The circle symbolizes a latent factor not observed by the researcher. Squares are observed variables used in the analyses of the generated data. To simplify the figure, residual variances of observed variables are not drawn. This population model partly corresponds to the model in Judd & Kenny [24] (p. 118), where the allocation variable to control versus intervention group has stable effects on test scores. The two models differ because the current model assumes that baseline health affects follow-up health. Nevertheless, the current population model implies that the effect from these causes on follow-up health is equal to their effect on baseline health.