Skip to main content

Table 2 Estimated associations between baseline predictor and change in health from appropriate use of change score analyses (i.e. in situations where RTM was assumed not to occur), as shown in Figure 3

From: Bias in the study of prediction of change: a Monte Carlo simulation study of the effects of selective attrition and inappropriate modeling of regression toward the mean

  50% attrition rate   70% attrition rate  
Dependency (bdrop) bpred(SE) Coverage of 95% C.I. Diff bpred(SE) Coverage of 95% C.I. Diff
Pred. V1 V2       
0 0 0 .10 (.04) 96% 0.00 .10 (.05) 96% 0.00
.3 .3 0 .13 (.04) 86% 0.08 .13 (.05) 91% 0.11
.3 .3 .1 .12 (.04) 90% 0.04 .13 (.05) 92% 0.06
.3 .3 .3 .10 (.04) 94% -0.02 .10 (.06) 93% -0.03
  1. Dependency is the magnitude (bdrop) of the regression of liability of dropping out on each of the three study variables. Pred = baseline predictor. V1 = the main variable at baseline (baseline health), V2 = the main variable at follow-up (follow-up health). SE = standard error. bpred = regression coefficient from predictor to change in health. Coverage of 95% C.I. = the percentage of the 500 samples with an estimated bpred with a 95% confidence interval containing the true population value. Diff = estimated change score (follow-up health minus baseline health). bpred, SE, and Diff are average results over the 500 generated samples. N in the original samples was 1000. The first line shows results when attrition was completely random.
  2. The true population value was bpred = .10.