. Author manuscript; available in PMC: 2016 May 10.

Published in final edited form as: Stat Med. 2015 Jan 23;34(10):1621–1633. doi: 10.1002/sim.6414

Table 6.

Missing At Random (MAR)^a

Data Model	Interaction	Statistics^b	Hypothesis Test	Power^c	Type I Error^c
Full (CS)	Moderate	ζ _t	Last Time Point	0.193(0.328)	0.060(0.047)
		ζ _ANCOVA		0.285(0.327)	0.064(0.044)
		ζ _FUN		0.310(0.343)	0.049(0.049)
		ζ _MSL		0.989(0.987)	0.054(0.046)
		ζ _MBP	Overall Effect	0.968(0.965)	0.054(0.044)
		ζ _MSA	Overall Effect	0.995(0.995)	0.054(0.046)

Main Effects (CS)	None	ζ _t	Last Time Point	0.176(0.325)	0.060(0.047)
		ζ _ANCOVA		0.265(0.331)	0.064(0.044)
		ζ _FUN		0.303(0.338)	0.049(0.049)
		ζ _MSL		0.689(0.710)	0.054(0.046)
		ζ _MBP	Overall Effect	0.463(0.486)	0.054(0.044)
		ζ _MSA	Overall Effect	0.689(0.710)	0.054(0.046)

results based on 1000 simulations with 100 subjects at each treatment arm

model selection via BIC with 200 bootstrap resampling

values inside parentheses are based on no missing scenario

ζ_t: t-test for last time point effect

ζ_ANCOVA: test last time point effect via ANCOVA model

ζ_FUN: test last time point effect via full mixed effect model with UN covariance

ζ_MSL: model selection test statistics for testing last time point effect

ζ_MBP: mixed effects model based likelihood ratio test for overall effects

ζ_MSA: model selection statistics for testing overall effects