The easiest approach to this problem, I think, is simply to take the
logarithm (natural, base 10, binary, or other, as convenient) of the
variable in question: that is, for raw variable X, compute a new
variable W = log(X). Do whatever statistical analyses you need in the
metric of the logarithms; at the end, transform back into the metric of
the raw data by taking antilogarithms (aka e^W, 10^W, 2^W, or whatever).
If a geometric mean is in some sense "natural" for your data, it is not
unlikely that the logarithm of the variable is approximately normally
distributed (at least, to the extent that *any* of our variables of
interest are approximately normally distributed!).
Any further advice would probably require rather more description of the
particular problem(s) you're trying to deal with.
-- DFB.
On 25 Mar 2003, Dr. S. Shapiro wrote:
> I am familiar with calculation of an arithmetic mean
> +/- a standard deviation. However, for certain problems
> now I need to calculate a geometric mean rather than an
> arithmetic mean. Is the "standard" standard deviation
> still valid for use with the geometric mean, or does a
> different kind of standard deviation need to be used in
> conjunction with the geometric mean?
-----------------------------------------------------------------------
Donald F. Burrill [EMAIL PROTECTED]
56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
