Run Bayesmh for unstructured covariance structure for the random effects

[Jack Cheng]

Dear all

I am following Bayesmh's document to run example 25 that deals with unstructured covariance structure for the random effects. The code on P.95 that uses pig example is below

bayesmh weight i.id i.id#c.week, likelihood(normal({var_0})) noconstant prior({weight:i.id i.id#c.week}, mvnormal(2,{weight:_cons}, {weight:week}, {Sigma,m})) prior({weight:week _cons}, normal(0, 1e2)) prior({var_0}, igamma(0.001,0.001)) prior({Sigma,m}, iwishart(2,3,I(2))) block({var_0}, gibbs) block({Sigma,m}, gibbs) block({weight:_cons}) block({weight:week}) block({weight:i.id}, reffects) block({weight:i.id#c.week}, reffects) noshow({weight:i.id i.id#c.week}) mcmcsize(5000) dots

However, Stata reports error message "prior mvnormal must have dimension 95 r(503);".

Does any one know what wrong is with the code, or how to run Bayes for unstructured covariance structure for the random effects in a right way? Thanks.

Best,
Jack

[Jack Cheng]

Or if anyone run this code successfully? Thanks.

The full code is below

. webuse pig
(Longitudinal analysis of pig weights)

. set seed 14

bayesmh weight i.id i.id#c.week, likelihood(normal({var_0})) noconstant
> prior({weight:i.id i.id#c.week},
> mvnormal(2, {weight:_cons}, {weight:week}, {Sigma,m}))
> prior({weight:week _cons}, normal(0, 1e2))
> prior({var_0}, igamma(0.001,0.001))
> prior({Sigma,m}, iwishart(2,3,I(2)))
> block({var_0}, gibbs) block({Sigma,m}, gibbs)
> block({weight:_cons}) block({weight:week})
> block({weight:i.id}, reffects)
> block({weight:i.id#c.week}, reffects)
> noshow({weight:i.id i.id#c.week})
> mcmcsize(5000) dots

prior mvnormal must have dimension 95
r(503);

[Joseph Coveney]

Perhaps StataCorp inadvertently broke something in bayesmh or the mvnormal() prior since that section of the user's manual was first produced.

This seems to be a bug.

You should contact Stata Technical Support about this.

[Jack Cheng]

Thanks, Joseph. I will contact Stata as you suggest.

Happy Weekend!
Jack

[Nikolay Balov (StataCorp)]

Jack,

to run the code it is essential to keep all levels of the group variable id. Before bayesmh you must run

fvset base none id

Otherwise the mvnormal prior does not have the right dimensions.

Nikolay

[Jack Cheng]

Thanks, Nikolay. The code works! I noticed this commend in previous samples but ignored it in example 25.

Also thank Joseph for the suggestion. It is a know-how that maybe only Statacorp can answer.