The inaccuracy of sample-based confidence intervals to estimate a priori ones