reversible jump mcmc

[Qing Zhao]

Hi everyone,

I'm looking into the rjmcmc model selection method implemented in Nimble. Here, I want to consider a model that includes a quadratic form of a covariates, and I want the quadratic form to be included only when the main effect is in. I have the following line to achieve that (the full code is attached in case anyone is interested): 

    y_pred[i] <- beta0 + z[1] * beta1 * x1[i] + z[2] * beta2 * x2[i] + z[2] * z[3] * beta3 * x2[i] * x2[i]

which seems to work fine. However, meanwhile I figured there is this requirement that the length of beta's has to be the same as the length of z's. Say I want the main effect and the quadratic form to be in or out at the same time, what should I do? I cannot have the following:

    y_pred[i] <- beta0 + z[1] * beta1 * x1[i] + z[2] * beta2 * x2[i] + z[2] * beta3 * x2[i] * x2[i]

Is there a way to overcome this?

Best,

Qing

[Qing Zhao]

Sorry I forgot the attachment.

[Qing Zhao]

Thank you Ben for sharing the information. It seems to be a cool way to deal with this issue.

Best,

Qing

On Wed, Jul 10, 2024 at 9:04 AM Benjamin R Goldstein <ben.go...@berkeley.edu> wrote:

Hi Qing--

I was working on a similar problem recently (RJMCMC when one indicator variable is associated with multiple variables downstream). I found the answer in this old thread, "Sampling categorical variables with RJMCMC":

https://groups.google.com/g/nimble-users/c/O4A9mxW_vhU?pli=1

The code provided by Daniel Eacker in their final email on the thread worked great for me. Thanks Daniel!

Hope this helps--

Ben

--
You received this message because you are subscribed to the Google Groups "nimble-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nimble-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/633a7068-7fdd-4187-9bd6-231e592dba91n%40googlegroups.com.

[Benjamin R Goldstein]

Hi Qing--

I was working on a similar problem recently (RJMCMC when one indicator variable is associated with multiple variables downstream). I found the answer in this old thread, "Sampling categorical variables with RJMCMC":

https://groups.google.com/g/nimble-users/c/O4A9mxW_vhU?pli=1

The code provided by Daniel Eacker in their final email on the thread worked great for me. Thanks Daniel!

Hope this helps--

Ben

On Wed, Jul 10, 2024 at 10:44 AM Qing Zhao <white...@gmail.com> wrote: