The confidence interval from lme4() is
based on evaluating the model likelihood at different values of the variance component and does
not assume symmetry, which is why it is more conservative at the high end and less conservative
at the low end compared to the approximate confidence interval based on the standard error.