
30% attrition rate

50% attrition rate

70% attrition rate


Dependency

Estimated β (95% C.I.)

95% coverage (% Sig)

Estimated β (95% C.I.)

95% coverage (% Sig)

Estimated β (95% C.I.)

95% coverage (% Sig)


y

x


.00

.00

.30 (.24,.36)

95% (100)

.30 (.22,.38)

94% (100)

.30 (.20,.40)

95% (100)

.10

.00

.30 (.24,.36)

95% (100)

.30 (.22,.38)

94% (100)

.30 (.20,.40)

94% (100)

.20

.00

.30 (.24,.36)

94% (100)

.30 (.22,.38)

94% (100)

.30 (.20,.40)

94% (100)

.30

.00

.29 (.23,.35)

93% (100)

.29 (.21,.37)

93% (100)

.29 (.19,.39)

93% (100)

.40

.00

.29 (.23,.35)

88% (100)

.29 (.21,.37)

89% (100)

.28 (.18,.38)

89% (100)

.10

.10

.29 (.23,.35)

95% (100)

.29 (.21,.37)

94% (100)

.29 (.19,.39)

94% (100)

.20

.10

.29 (.21,.37)

91% (100)

.28 (.20,.36)

91% (100)

.28 (.18,.38)

90% (100)

.30

.10

.28 (.20,.36)

83% (100)

.27 (.19,.35)

82% (100)

.27 (.17,.37)

84% (100)

.40

.10

.25 (.18,.33)

73% (100)

.26 (.18,.34)

69% (100)

.25 (.15,.35)

70% (100)

.20

.20

.28 (.20,.36)

83% (100)

.29 (.21,.37)

80% (100)

.26 (.16,.36)

85% (100)

.30

.20

.26 (.18,.34)

69% (100)

.25 (.17,.33)

65% (100)

.24 (.14,.34)

68% (99)

.40

.20

.25 (.18,.33)

44% (100)

.23 (.15,.31)

38% (100)

.22 (.10,.34)

42% (97)

.30

.30

.24 (.16,.32)

43% (100)

.23 (.15,.31)

33% (100)

.21 (.09,.33)

41% (95)

.40

.30

.22 (.14,.30)

16% (100)

.20 (.12,.28)

10% (99)

.18 (.06,.30)

15% (86)

.40

.40

.19 (.11,.27)

3% (100)

.15 (.07,.23)

1% (92)

.13 (.01,.25)

2% (59)

 Population β is the effect of the baseline predictor on the later outcome in the population. Dependency = direct effect (β) of study variables on Liability of attrition. Y is dependency between attrition and Followup outcome (path c in Figure 1), x is dependency between attrition and Baseline predictor (path b in Figure 1). N in the original sample is 1000. 95% C.I. = 95% confidence interval. Parameter estimates and 95% C.I.s are average results over the 500 generated samples. 95% coverage = proportion of the 500 samples where the 95% confidence intervals contain the population value. % Sig = proportion of the 500 samples where the false null hypothesis of zero association between Baseline predictor and Followup outcome is rejected at the .05 level. The dependency between study variables and the continuous Liability of dropping out is reported in βs. The corresponding ORs of the associations between study variables and a dichotomized dropout variable would be about the following (depending on attrition rate): β =.00 equals OR =1; β =.10 ≈ OR = 1.22; β =.20 ≈ OR =1.49; β =.30 ≈ OR =1.82; β =.40 ≈ OR = 2.23.