Abstract:
|
When a dominated rejection algorithm is used to generate random variates of a continuous distribution on a bounded interval of real values, the extent to which calculated uniform variates, and the density function evaluation at those uniform variates, introduce evaluation error into the acceptance or rejection decision could possibly interfere with the intended statistical inference. In fact, if the actual distribution being produced significantly differs from the intended distribution, then whatever statistical inference uses this data is invalid. The analytical methods found in this paper provide for objectively quantifying the extent to which spurious candidates (variates that should be rejected but are actually accepted due to calculation error) and missed candidates (variates that should be accepted but are actually rejected due to calculation error) may either be eliminated, minimized, or tolerated.
|