Systems Archive 1

[ Read Responses | Return to the Index ]
[ Previous | Previous in Thread | Next in Thread | Next ]

Re: Far Too Few

Posted by John Auston on Tuesday, 20 January 1998, at 7:39 a.m., in response to Re: Far Too Few, posted by Lounge Lizard on Monday, 19 January 1998, at 9:30 p.m.

My point is that it is always better to run many more
rounds than 25 million. Granted, depending on what you
are trying to show, this may be overkill. For example,
25 million rounds are plenty to show the % freq of running
or true counts.

But, when I was trying to contrast Hi-Lo Lite indices versus
actual indices (or perhaps it was Zen Lite, I forget), the
correct 'winner' did not show up until near 1 billion rounds.
Meaning, less than that could have 'shown' that the rounded
indices were 'better' than the actual indices, which would
have been 'wrong'.

The problem with posts that show a lot of different data,
some of which settle down within 25 million trials, some
not, is that some folks may tend to draw conclusions that the
researcher did not intend.

So, better to run more trials, just to be safe - IMO.
Personally, I would never run less than 100 million, and
often run 200-400 million. SBA makes this easy.

Especially in posts where there is the hint of comparing
one system to another, or one set of indices to another,
more is simply better. To be sure, a lot of past
research is based on samples in the few millions, but they
did not have access to the programs and computers we have

Just my thoughts.




The Systems Archive 1 is maintained with WebBBS 2.24.060398.