Reconstructing data from statistics

October 5, 2010 5:10 PM

Statistics Filter: Reconstructing data from F values?

I'm trying to reconstruct some data from statistical analyses in a paper. I know I can't recreate the actual numbers, but I'm trying to get as much descriptive information as possible. If I have results from a 2x2 F-test, including F values for the omnibus test and for each of the factors, as well as means for each level of each factor, is it possible to calculate the standard deviations? Or if not, can I figure out any constraints on what the standard deviations might be? Any other descriptive information I can figure out (effect sizes, etc.)? Thanks!

posted by svenx to Science & Nature (3 answers total)

I haven't done this in a while so I can't tell you how off the top of my head, but I know that the book "Practical Meta-Analysis" by Lipsey and Wilson will tell you how to calculate an effect size (standardized mean difference) if you know the means, F, and degrees of freedom. Seems like you could decompose that to get just the standard deviations. Hopefully someone else will chime in who has the formulas handy (I stupidly lent out my copy of the book and never got it back).
posted by nixxon at 5:29 PM on October 5, 2010 [1 favorite]

Wouldn't it be better just to ask the paper's authors if they will share the data with you?

What possible reason could there be to create a dataset working backwards from results other than the obvious nefarious one?
posted by srboisvert at 5:07 AM on October 6, 2010

Response by poster: I'm actually not sure what the obvious nefarious reason is, but yeah, asking the authors will probably turn out to be the easiest approach.
posted by svenx at 5:35 AM on October 6, 2010