stan-dev
models with constant outcomes
Summary:
A rstanarm model will fail if any of the variables (including the outcome) are constant. But I'm not sure what we really want to do in this case.
Description:
Because of check_constant_vars
Reproducible Steps:
http://stackoverflow.com/questions/41354745/rstanarm-updating-priors-with-binary-data-without-any-events
RStanARM Version:
2.14.1
R Version:
3.3
Operating System:
Debian
With improper priors, it would be an improper posterior. With proper priors, you could learn your priors were wrong.
On Fri, Dec 30, 2016 at 6:30 PM, Jonah Gabry [email protected] wrote:
Yeah I'm not sure what we want do to either
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/stan-dev/rstanarm/issues/149#issuecomment-269834782, or mute the thread https://github.com/notifications/unsubscribe-auth/ADOrqpFDKsbeswmTEe9SFYj9M3LcluBaks5rNZQTgaJpZM4LYblI .
The bernoulli.stan file would fail anyway because N is declared with a lower bound of 1 but that is somewhat minor compared to the question of whether we want to estimate the model.
yeah
On Fri, Dec 30, 2016 at 6:46 PM, Jonah Gabry [email protected] wrote:
This might be something to hold off on settling until the next release.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/stan-dev/rstanarm/issues/149#issuecomment-269835791, or mute the thread https://github.com/notifications/unsubscribe-auth/ADOrqpdVZ7EbgheSqqIWAhOyeo7HZledks5rNZfPgaJpZM4LYblI .
I think the following should work, but fails due to constant y error
d_bin <- data.frame(N = c(674, 674), y = c(39,39), grp2 = c(0,1))
fit_bin <- stan_glm(y/N ~ grp2, family = binomial(), data = d_bin,
weights = N, refresh=0)
