Skip to main content
. 2016 Aug 30;35(29):5495–5511. doi: 10.1002/sim.7090

Table 3.

Properties of estimates for μ and τ 2 from the simulation study with K = 5 studies, using 1000 simulations for each τ 2 value. In each case, IS denotes values using Bayesian methods by importance sampling. ‘Z length’ denotes the average length of the nominal 95% interval for μ using the standard normal quantile, and ‘Z coverage’ denotes the proportion of nominal 95% intervals that cover the true value, using the standard normal quantile.

Properties of estimates for μ
Empirical mean Empirical std dev Z length Z coverage
IS MCMC IS MCMC IS MCMC IS MCMC
τ 2 = 0  − 0.001  − 0.001 0.106 0.106 0.771 0.771 0.9993 1
τ 2 = 0.029 0.003 0.003 0.143 0.144 0.800 0.805 0.995 0.995
τ 2 = 0.069 0.001 0.001 0.170 0.169 0.840 0.842 0.985 0.985
τ 2 = 0.206 0.005 0.005 0.260 0.260 0.971 0.984 0.996 0.996
τ 2 = 1.302 0.013 0.012 0.556 0.554 1.654 1.839 0.991 0.992
Properties of estimates for τ2
Empirical mean Empirical std dev
IS MCMC IS MCMC
τ 2 = 0 0.064 0.064 0.018 0.018
τ 2 = 0.029 0.073 0.073 0.027 0.027
τ 2 = 0.069 0.084 0.084 0.040 0.040
τ 2 = 0.206 0.141 0.142 0.112 0.111
τ 2 = 1.302 0.767 0.787 0.646 0.682

MCMC, Markov chain Monte Carlo.