Comparing change in longitudinal studies: adjusting for initial value.

Adjustment for initial value may yield misleading results when comparing change over time in two populations. This paper presents a methodology for correcting for the effects of regression to the mean when change over time is regressed on standardized initial level. The results are used to examine the bias introduced by failing to correct for the effects of regression to the mean in the two sample problem. We also examine the implications of adjusting for initial value in certain observational studies. When group assignment is known to be related to initial value, we argue that this adjustment procedure is inappropriate. The results are illustrated using data from a longitudinal study of lung function decline.