Skip to main content
. Author manuscript; available in PMC: 2009 Nov 30.
Published in final edited form as: Biometrics. 2008 Mar 19;64(4):1032–1042. doi: 10.1111/j.1541-0420.2008.01011.x

Table 3.

Results from simulation study under setting 1

Joint analysisa
 Separate analysisa
Type of
parameterb 		True
value		 Mean
estimate		 Sample
SE		 Mean
Asy. SE		 Mean
estimate		 Sample
SE		 Mean
Asy. SE
β12 0.036 0.036 0.0048 0.0016 0.032 0.0068 0.0065
β13 0.075 0.075 0.024 0.0031 0.069 0.028 0.028
β21 −0.015 −0.015 0.0045 0.0035 −0.012 0.0067 0.0068
β23 −0.13 −0.11 0.074 0.074 −0.096 0.067 0.063
γ12 1 1 0 0
γ13 −0.5 −0.7 1.26 0.93
γ21 1 1 0 0
γ23 5.0 5.29 2.01 1.95
var(ηi1) 0.16 0.14 0.043 0.062
cov(ηi1, ηi2) −0.05 −0.042 0.014 0.021
var(ηi2) 0.02 0.016 0.0053 0.0073
a

In joint analysis, the variance components and the loadings of the latent traits are estimated. In separate analysis, each type of transition is estimated separately assuming no association and without using the latent traits. The “Sample SE” denotes the standard deviation of the parameter estimates from 100 simulated data sets and the “Mean Asy. SE” denotes the average of the asymptotic standard errors obtained from the 100 simulated data sets.

b

We use k to index the state from which transition is made. We use l = 2 and 3 to denote the states of disability and death that can be transited to from state 1. We use l = 1 and 3 to denote the states of independence and death that can be transited to from state 2.