Small-sample theory for steady state confidence intervals