SPSS and geometric mean and standard deviation

[dagn...@hotmail.com]

Dear Group,

I have (with help from the group, thankyou) learned how to calculate
the geometric mean and standard deviation (for GSD: take SD of log of
variable and then calculate antilog of that value) - 95% of values lie
between GBAR/(gsd2) and GBAR*(gsd2) where GBAR=geometric mean.

Questions:

1) Can I get SPSS to calculate the GSD? I can't find any evidence
of this command under the Means or any other command.

2) My dependent variable's log has a much better distribution than
the original variable. How does this relate to the reporting
and use of the geometric mean?

Thankyou in advance!
Beth

[Bruce Weaver]

Do exactly what you said above: compute the SD of the log-transformed
scores, and exponentiate it. Use something like this (untested) syntax:

compute const = 1.
compute logx = ln(x).
aggregate
/break = const
/meanlx = mean(logx).
/sdlogx = sd(logx).
compute gmean = exp(meanlx).
compute gsd = exp(sdlogx).
exe.

--
Bruce Weaver
bwe...@lakeheadu.ca
www.angelfire.com/wv/bwhomedir

[dagn...@hotmail.com]

I greatly appreciate the help and syntax.

How then would you calculate the range of the variable for which you
have calculated the geometric mean? Would you take the range of the
logarithm and exponientiate the endpoints (if this is correct, any
assistance with the syntax would also be appreciated)?

Thankyou,
Beth

[Bruce Weaver]

When I read your first post, I googled on "geometric standard
deviation", and got several hits. But when I google on "geometric
range", I get a bunch of hits for "geometric range seraching", which
appears to be something other than what you're describing. So it's not
clear to me whether this is something anyone uses in the way you propose.

Nevertheless, if you want to compute the min and max of your
log-transformed data, and exponentiate those points, that is easy enough
to do. Add two lines to the AGGREGATE command in my earlier post to get
the min and max values of the logged data, then exponentiate those values.

compute const = 1.
compute logx = ln(x).
aggregate
/break = const

/meanlogx = mean(logx)
/sdlogx = sd(logx)
/minlogx = min(logx)
/maxlogx = max(logx)
.
compute gmean = exp(meanlogx).
compute gsd = exp(sdlogx).
compute gmin = exp(minlogx).
compute gmax = exp(maxlogx).
exe.

By the way, I noticed that I had some unwanted command terminators in
the middle of the AGGREGATE command in my earlier post. There should
have been only one period at the end of the command, as shown above.

[dagn...@hotmail.com]

Dear Bruce and Group,

Yes, I found the same hits but these are not what I am looking for.
Thankyou for the new syntax. I have one final (I hope!) question. I
calculated geometric means and got the following means and SD (SD in
brackets):

X mean of log X geometric mean of
X

3.61 (5.39) .27 (1.78) 1.31
(5.87)

Do these numbers sound correct? I am not sure what to expect as I have
not worked with geometric means before.

Thankyou in advance and for all the help to date,
Beth

[Bruce Weaver]

The geometric mean and SD are correct, assuming the mean and SD of the
logged data are correct.