Sampling from predictive distribution when outcome has missingness

[Andrew Lawson]

Following on from Pedro's question: does this imply that if we have NAs in outcomes which are declared as data then they will not be automatically sampled unless we specify after the config stage : 

config$addSampler(allowData=TRUE)?

thanks

Andrew

[Chris Paciorek]

If the data nodes are scalar nodes, e.g.,

```

for(i in 1:n)

 y[i] ~ dnorm(mu, sd = sigma)

```

and you specify `y` as `data` with `nimbleModel` with some NAs in there, then nimble will flag the nodes corresponding to elements with NAs as *not* being data and will sample them using a predictive sampler.

The tricky part in terms of Pedro's question was having a multivariate node with a mix of NAs and non-NAs.

Ultimately, if you print the MCMC config object or run the $printSamplers method, you should be able to see what NIMBLE is doing by default.

-chris

-chris

--
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 visit https://groups.google.com/d/msgid/nimble-users/5fe2a52b-2806-4af6-b7e2-077e5f121f59n%40googlegroups.com.

[Lawson, Andrew B.]

Hi Chris

Thanks for the clarification. I did run an old parametric survival model with missing times and they were sampled correctly as missing. 

Thanks again

best

Andrew

---

From: 'Chris Paciorek' via nimble-users <nimble...@googlegroups.com>
Sent: Friday, July 25, 2025 6:19 PM
To: Andrew Lawson <abla...@gmail.com>
Cc: nimble-users <nimble...@googlegroups.com>
Subject: Re: Sampling from predictive distribution when outcome has missingness

 

If the data nodes are scalar nodes, e. g. , ``` for(i in 1: n) y[i] ~ dnorm(mu, sd = sigma) ``` and you specify `y` as `data` with `nimbleModel` with some NAs in there, then nimble will flag the nodes corresponding to elements with NAs as *not*

To view this discussion visit https://groups.google.com/d/msgid/nimble-users/CA%2B7Ts_rTZKJ1As2rHq0wnGv%2BfRhdJV9jJUKxHbOppWDEq%3DFu1A%40mail.gmail.com.