ExcelBanter - View Single Post

wrote...
....
Good point. I thought of that, too. But I try to
not make ass-u-me-tions about algorithms that I have
no control over, and especially about undocumented
features. For example, what if RAND() later can
utilize a hardware RNG, if it is installed?

Truly unknown algorithms are one thing. However, there's enough
documentation of Excel's PRNGs (both the one in Excel 2003 and the one
in previous versions) to establish that they *are* periodic. Undue
skepticism about documented functionality isn't wisdom, it's paranoia.
Whether paranoia is warranted when using Microsoft software with
Microsoft documentation is debatable.

Given the need for simulating sampling without replacement, would there
ever be hardware RNGs without a library routine to produce samples
without replacement?

Don't get wrong: I must admit that your assertion
is quite likely right in the case of generating only
15 RAND() values. I was probably being a little
anal. But I thought the point was important to make
since some people will apply these ideas to very
different circumstances.

The period is over 1 million (per Microsoft's summary of the specs,
it's only about 1E6, but the details of those specs imply the period is
orders of magnitude greater). Myself, I can't see even thoroughly
misguided Excel users trying to generate *samples* on the order of a
million values.

Anal, yes. Important point, no.