The lmer
results are suspect because the correlations are 1 or NaN.
The stan_lmer
ones might be okay but you get a very small intercept (I think) due to standardizing your variables, which isn’t recommended. Just do it with unstandardized variables, and if you have to interpret things in terms of standard deviations. Do posterior_predict
once with the actual data and once with a one standard deviation increase for all observations on the variable of interest. The distribution of the difference between those two is what you want.