Using bayesstats summary on pooled multiple chain data.

[Ben A. Dwamena]

Dear Statalisters,
Is there a way to summarize amalgamated data from multiple chains with bayesstats summary that is undocumented?. I generated composite data using three chains as described in code below. On multiple attempts, bayesstats summary summarized third set of estimates only.

cd I:\
use http://fmwww.bc.edu/repec/bocode/m/m...ample_data.dta, clear

set seed 1356
gen num1 = tn+fp
gen num2 = tp+fn
gen study=_n
gen parameter1 = tn
gen parameter2 = tp
reshape long num parameter, i(study) j(dtruth)
tabulate dtruth, generate(mu)

forvalues nchain=1/3 {
bayes, remargl mcmcsize(10000) burnin(20000) initrandom thin(2) nomodelsummary saving( "I:\sim`nchain'.dta", replace): meglm parameter mu1 mu2, ///
noconstant || study: mu1 mu2, noconstant ///
cov(uns) family(binomial num) link(logit)
estimates store sim`nchain'
local estid : di "sim`nchain'" ""
local estnames "`estnames' `estid'"
}

nois grubin {parameter:mu2} {parameter:mu1} {U:Sigma_1_1} {U:Sigma_2_1} ///
{U:Sigma_2_2}, estnames("`estnames'")

forvalues i=1/3 {
use "I:\sim`i'.dta", clear
gen chain=`i'-1
qui save "I:\sim`i'.dta", replace
}
use "I:\sim1.dta", clear
forvalues i=2/3 {
append using "I:\sim`i'.dta"
}
nois bayesstats summary (sen:invlogit({parameter:mu2})) ///
(spe:invlogit({parameter:mu1})) ///
(logdor{parameter:mu2}+{parameter:mu1})) ///
(lrp:invlogit({parameter:mu2})/(1-invlogit({parameter:mu1}))) ///
(lrn1-invlogit({parameter:mu2}))/invlogit({parameter:mu1})), hpd noleg

Thank you,
Ben