How to deal with a high F-ratio using 'rstanarm'?

In the (1 | g) case, it is the standard deviation of the intercept across levels of g.