The effects of rounding on likelihood procedures