On Sat, 12 Apr 1997, Douglas Dame <dougdame@HPE.UFL.EDU> wrote:

>Thanks to John Whittington for posting code and some stats on run-times that >show this solution to the posed problem is eminently affordable in terms of >run-time and machine cycles. It'd be hard to complain about a first-cut >solution with 8 minutes of run time, and especially not the improved version >at under 3 minutes. > >However, some situations you'd want to simulate (even in SAS) are much more >complex and time-consuming to run through for each iteration, and 10,000 >iterations based on my experience is an awful lot.

Douglas, I agree with you entirely - I have very rarely had to use simulations with more than 1,000 iterations - but in this case I was merely providing the code to do what Pamela had asked. Needless to say, I developed the code with just 10 iterations, then added the extra three zeros for final testing and 'posting'.

It obvioulsy does depend upon precisely what one is trying to do. If, for example, one's interest was in extreme tails of distributions, then 10K iterations, or even more, might sometimes be necessary.

I obvioulsy didn't know exactly what Pamela was trying to achieve. When one does know 'what one is doing', a good approach is to look at the incremental changes 'in whatever' as one runs the simulation and then stop the iterations when the incremental effect of each iteration falls below some particular threshold. In that fashion, one can sometimes find that surprisingly few iterations are quite adequate for the purpose in hand - so it may be necessary to build in a minimum just to avoid all the 'complaints' from folk who say you haven't used enough!!

Regards

John

----------------------------------------------------------- Dr John Whittington, Voice: +44 1296 730225 Mediscience Services Fax: +44 1296 738893 Twyford Manor, Twyford, E-mail: johnw@mag-net.co.uk Buckingham MK18 4EL, UK CompuServe: 100517,3677 -----------------------------------------------------------