Do weights work in `stan_glm()`?

Summary:

I'm trying to use the weights argument to stan_glm() and stan_glmer() but it does not appear to affect the fit.

The same data in lm() or glm() works as desired.

Reproducible Steps:

Load the attached data weights_test.txt:

df <- read_csv("weights_test.txt")
head(df)

Then fit the model to all the data:

> stan_glm(y ~ x, data=df, iter=500)
            Median MAD_SD
(Intercept) 95.7   14.0  
x           38.5   20.0  

Now create a weighted dataset:

df_weighted <- df %>%
  group_by(x,y) %>%
  summarize(n=n())
head(df_weighted)

Then fit the weighted data:

>stan_glm(y ~ x, data=df_weighted, weights=n, iter=500)
            Median MAD_SD
(Intercept) 779.2  167.0 
x           -72.7  228.0 

This is very similar to what I get with df_weighted but not specifying any weights. On the other hand using lm() or glm(), the weighted fit is the same as the original fit.

RStanARM Version:

2.19.2

R Version:

3.6.1

Operating System:

Debian Linux (Stretch)

Reproduce the supplied comparison between stan_glm() and lm()/glm() using the grouped data and weights=n, then trace how weights are handled in stan_glm() and stan_glmer(). Done means weighted grouped data produces the same fit as the ungrouped observations, with regression tests covering the behavior.