I'm just a little bit confused and hope someone might can help me:
I'm working on the statistics part of my thesis and performed some ANOVAs with Greenhouse Geisser as correction. Now the thing that suprises me is that SPSS gives me decimal numbers as degrees of freedom for the F value.
I.e. F(1.461 / 29.227)=....
In papers I've never seen decimal numbers as df's for F values. They're always whole numbers.. So am I doing sth wrong or is it just normal to round the values?
Would be glad for some help...
No, you're not doing any wrong. It's just an adjustment to the degrees of freedom when the underlying assumption of homogeneity associated with the covariance matrix is violated....e.g. for a One-Way Within Subjects Design (or Repeated Measures) you would have df(num) = epsilon*(a - 1) and df(den) = epsilon*(n - 1)(a - 1) and where the computation of epsilon is rather complicated.
Because epsilon is bounded between 1/(a - 1) and 1.0, it will create the decimals your observing.
Thank you dragan! So would you rather round the decimals and use whole numbers or keep the decimals for a publication? Do you know wheater there is a convention?
Thank you Dragan for your comment clarifying the Greenhouse Geisser ANOVA adjustment!!! It was super helpful
Just a quick question. How can you derive the "usual degrees of freedom" when a manuscript only presents the adjusted numerator degrees of freedom and the adjusted F test? For example, I'm reading a paper and trying to calculate the p-value, but I only see the following (N = 74): F = 4.01 (adjusted), df = 1.46, p = .03.
Other than that information, the manuscript has nothing such as epsilon (although I do see they had that info when producing the output as they used SPSS 19.0). I tried contacting the author to no avail.
Any thoughts on how to back-transform?
Advertise on Talk Stats