I've specified a model with these analysis specifications: TYPE = TWOLEVEL RANDOM; ESTIMATOR=BAYES; biter=(5000); thin=10;
And this save specification: SAVEDATA: STDDISTRIBUTION = 1.dat; SAVE = fs(200); FILE = 2.dat; BPARAMETER = 3.dat;
The model has 2 MCMC chains (default). I have some questions about the BPARAMETER output ("3.dat") file:
(1) Why does chain 1 have 5,200 iterations? Chain 2 has 5,000. Does it have something to do with the 200 imputation runs for the factor scores or is that random coincidence?
(2) The mean, median, and SD for the parameter's posterior distribution in the MPlus-generated histogram are similar to, but do not match, the mean, median, and SD for the same parameter in either chain in the 3.dat file. Can you explain why? It makes me realize that I don't fully understand what's in the Mplus-generated histogram, which appears to match what's in the output.
(3) I'd like to use 99% CIs for some of my hypothesis tests, so I need to learn how to use the BPARAMETER output. Given the (slight) mismatch with the program/plot output, I'd appreciate pointers on how I can use the BPARAMETER output to get CIs other than 95%.
1) Yes. The last 200 iterations are provided in case you need to obtain the posterior distribution of a quantity involving both the factors and the parameters.
2) The posterior distribution is constructed by combing the second half of the two chains: iterations 2501 to 5000 from both chains for a total of 5000 values.
3) Use output:cint to get the 99% confidence interval. Mplus computes 90%, 95% and 99% CI. For other CIs you can use the BPARAMETER file and use the correct iterations or you can save the distribution from the plot (right click).