. Author manuscript; available in PMC: 2018 Jan 5.

Published in final edited form as: J Am Stat Assoc. 2017 Jan 5;111(516):1454–1465. doi: 10.1080/01621459.2016.1167693

Table 4.

Posterior mean and 95% credible interval for the quantities of interest under each missingness assumption. The posterior probability row gives the probability that the exercise treatment is superior: higher cessation rate, lower weight change, lower correlation.

		Missing data assumption

Quantity of interest	Treatment	Z_it MAR W_it MAR	Z_it MNAR W_it MNAR	Z_it MNAR W_it MAR	Q_it = 0 if missing W_it MAR
	Wellness	0.53 (0.40, 0.65)	0.50 (0.37, 0.63)	0.50 (0.37, 0.63)	0.30 (0.21, 0.39)
P (Q_iT = 1)	Exercise	0.47 (0.35, 0.59)	0.44 (0.32, 0.56)	0.43 (0.32, 0.56)	0.28 (0.20, 0.37)
	Post. prob.	0.25	0.24	0.23	0.42
	Wellness	3.0% (2.3, 3.8)	2.9% (2.2, 3.6)	3.0% (2.2, 3.7)	2.6% (1.8, 3.3)
E(W_iT)	Exercise	3.0% (2.3, 3.7)	2.8% (2.0, 3.4)	3.0% (2.3, 3.6)	2.7% (2.1, 3.4)
	Post. prob.	0.51	0.61	0.51	0.38
	Wellness	0.18 (− 0.08, 0.42)	0.18 (− 0.08, 0.42)	0.18 (− 0.07, 0.42)	0.21 (0.02, 0.39)
corr(Q_iT, W_iT)	Exercise	0.13 (− 0.14, 0.36)	0.13 (− 0.13, 0.36)	0.13 (− 0.14, 0.36)	0.17 (− 0.04, 0.35)
	Post. prob.	0.61	0.59	0.61	0.62