chaincheck: Checking the chains on convergence
In hmi: Hierarchical Multiple Imputation

Description Usage Arguments References

Formally tests the Gibbs-sampling chains on convergence. After the burn in is discarded, the remaining iterations of each chain are tested following Geweke (1992). In this test, the arithmetic means and their standard errors of the first 10% and last 50% of the chain (from now on always after discarding the burn in) are compared. In case of a stationary distribution, both means have the same expected value. The difference between both arithmetic means is divided the standard error. This is the Z-score, the test statistic. Chains not passing the test will be plotted. Each plot will flag which (fixed effect or variance) parameter was tested; and what variable was to be imputed and the cycle and imputation run. To see the next plot, the user has to hit <Return> (or "Enter").

1	chaincheck(mids, alpha = 0.01, thin = 1, plot)

`mids`	A mids object generated by hmi (alternatively a list), having an element called "gibbs" with the chains of the Gibbs-sampler runs.
`alpha`	A numeric value between 0 and 1 for the desired significance level of the test on convergence.
`thin`	An integer to set the thinning interval range. If thin = 1, every iteration of the Gibbs-sampling chain will be kept. For highly autocorrelated chains, that are only examined by few iterations (say less than 1000), the `geweke.diag` might fail to detect convergence. In such cases it is essential to look at a chain free from autocorrelation. When setting `thin = NULL`, the function will use internally a thinning of `max(1, round((nitt-burnin)/1000))` to get approximately 1000 iterations to be tested.
`plot`	Logical. Shall the chains be plotted in a traceplot or not. If the number of iterations and cycles is large, click through all traceplots can be interminable.

J Geweke (1992): Evaluating the accuracy of sampling based approaches to calculating posterior moments. In Bayesian Statistics 4 (ed. JB Bernando, JO Berger, AP Dawid and Adrian FM Smith) (pp. 169-193). Clarendon Press, Oxford, UK.

hmi documentation built on Oct. 23, 2020, 7:31 p.m.