Appropriate Discriptive Statistics: Log or Original Data

[Patricia Ali]

Dear All,

I have log transformed my variables and want to provide descriptive stats. Would you advise that I provide descriptive stat based on original (non transformed data) or best on log data. Does it matter which one I use. I feel that original data would be better explained.

Thanks for the advise in advance

[Nick Cox]

These questions can't be answered well without knowing what you're writing and who will be reading (judging, assessing) your report, paper, dissertation, thesis or book. Conversely, it's often true that non-technical readers won't care (much) about technical details, while it is the more technical readers who will care.

But even when variables are better off log transformed, I usually expect to see for the original scale

mean and SD
minimum and maximum
median and quartiles

and I like to see geometric mean and geometric SD too.

[Patricia Ali]

thanks Nick

[David Benson]

I prefer to see the original scale in summary statistics.

In the note to the table, you could say something like:

"For ease of interpretation, summary statistics are actual values for variables x, y, and z, not their logarithmic transformations. Correlations and the regression analysis utilize the natural logs."

[Patricia Ali]

Brilliant idea Ben! Thanks