Possible bug while passing resampled data directly to bayesglm

[bakaburg1]

Hello,
I'm observing a strange phenomenon while using bayesglm(). If I create a resampled data.frame directly in the function call, it will create estimates centered around the null, while if I first create the resampled data.frame and then pass it seems to produce correct estimates:

> bayesglm(I(Species == 'versicolor') ~ I(Sepal.Width <= 2.9), binomial(), data = iris %>% sample_frac(1, replace = T)) %>% tidy(exp = T)
# A tibble: 2 x 5
  term                      estimate std.error statistic  p.value
  <chr>                        <dbl>     <dbl>     <dbl>    <dbl>
1 (Intercept)                  0.458     0.215    -3.63  0.000278
2 I(Sepal.Width <= 2.9)TRUE    1.29      0.355     0.716 0.474   
> local({
+     DF <- iris %>% sample_frac(1, replace = T)
+     bayesglm(I(Species == 'versicolor') ~ I(Sepal.Width <= 2.9), binomial(), data = DF) %>% tidy(exp = T)
+ })
# A tibble: 2 x 5
  term                      estimate std.error statistic      p.value
  <chr>                        <dbl>     <dbl>     <dbl>        <dbl>
1 (Intercept)                  0.212     0.278     -5.57 0.0000000256
2 I(Sepal.Width <= 2.9)TRUE    7.79      0.379      5.42 0.0000000593

The non-resampled estimates are:

bayesglm(I(Species == 'versicolor') ~ I(Sepal.Width <= 2.9), binomial(), data = iris) %>% tidy(exp = T)
# A tibble: 2 x 5
  term                      estimate std.error statistic      p.value
  <chr>                        <dbl>     <dbl>     <dbl>        <dbl>
1 (Intercept)                  0.214     0.270     -5.71 0.0000000114
2 I(Sepal.Width <= 2.9)TRUE    6.72      0.377      5.05 0.000000442 

and are similar to the version in which the data.frame is passed after creation.

The problem doesn't seem to appear if using simple glm().