ANOVA using geometric mean

[mccombsj4]

Hello,

I am trying to replicate the analysis done in this paper, Table 2 and Table 3.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5332180/pdf/EHP27.pdf

They did an ANOVA of geometric means. I can generate geometric means using proc surveymeans and I'm familiar with proc anova, but I have not done an ANOVA of geometric means. Can anyone point me in the right direction.

Thanks you,

James

[PGStats · Accepted solution]

Usually it means doing an ANOVA on log-transformed dependent variables. Exp(mean(log(x))) = geomean(x). Thus, if you exponentiate the estimated means on the log scale, you get the geometric means.

[mccombsj4]

Do you know the sas code to do this?

[pau13rown]

this blog post from dr deng is likely helpful: http://onbiostatistics.blogspot.com/2014/03/analysis-of-2x2x2-crossover-design-data.html

also, see the appendix at this link to understand how it is related to percent change: https://www.cscu.cornell.edu/news/statnews/stnews83.pdf