Hi everybody,
Graphpad prism is able to calculate geometric means and geometric standard deviation.
- Is there any requirement for the data to have a normal distribution in order to calculate the geometric mean of the data ?
- Is it possible to calculate the geometric standard error of the mean instead of the geometric standard deviation? Or is this question completely mathematically dumb?
Thanks for your answer to a very statistic naive person...
Juliette
- the requirement is that the data should have an approximate log-normal distribution.
geom. mean and geom. sd are calculated from the log-transformed values: use the transformed values to calculate the mean "log-mean" and the sd ("log-sd"). The geometric versions are obtained by un-doing the log-transformation on these statistics, i.e. geom.mean = exp(log-mean) and geom.sd = exp(log-sd).
- you can do this: calculate the "log-sem" (as usual, only using the log-transformed values), then exponentiate the result.
One can do a lot, but it might be difficult to interpret these numbers.
If you are really interested in etimating means, better use a statistical model with a distribution that makes sense with your data and estimate the mean (not something different, like geometric means).
I agree with Jochen's reply entirely but you may want to Google search some of the terms and procedures prior to use. Best wishes to all, David Booth
See these links:
https://blogs.sas.com/content/iml/2019/10/02/geometric-mean-deviation-cv-sas.html
https://www.jstor.org/stable/2235723
https://stats.stackexchange.com/questions/155823/geometric-mean-standard-error?noredirect=1&lq=1
for logtransfomed data calculate Geometric mean and SD then Standard error of GM can be interpreted.
Ravishankar Suryanarayana : Did I get this correctly: you mean to calculate the geometric mean of the log-transformed data?
In any case, could you explain what the interpretation is?
instead of SD, the geometric coefficient of variation can be used to describe the variability. even geometric sd can be computed.Then as usual SE computed from GSD
Hi Juliette Bitard ,
as practical example on what Jochen Wilhelm said, please have a look to the paper in the link, especially Appendix B:
https://www.efsa.europa.eu/en/efsajournal/pub/3662
No, there is no requirement for the data to have a normal distribution in order to calculate the geometric mean. The geometric mean is a measure of central tendency that is suitable for log-normally distributed data, where the logarithm of the data values follows a normal distribution.
It is possible to calculate the geometric standard error of the mean (GSEM) instead of the geometric standard deviation (GSD). The GSEM is a measure of the precision of the estimate of the geometric mean, and it takes into account both the variability of the data and the sample size. The GSEM can be calculated as:
GSEM = exp(SD of ln(x))/sqrt(n)
where SD of ln(x) is the standard deviation of the logarithm of the data values and n is the sample size.
However, some statistical software may not provide an option to directly calculate the GSEM. In this case, you can estimate the GSEM using the formula above or by taking the ratio of the GSD to the square root of the sample size.
- Badges
- Science topic
- Similar topics
- Biological Science
- Anatomy
- Tissue