Pooled R^2 with Multiple Imputation?

[stodo53]

Hi all,

I am trying to figure out how to calculate edf to include in my proc mianalyze statement. The textbook I am following says to use:  dfcomplex=#PSUs-#Strata. However, although I am using complex survey data, I am only using weights (no clusters or strata). I have also used multiple imputation for my analysis. 

For exploratory purposes, I changed edf to multiple different values however, it still produced the same output anyways (even without changing it). See the attached output. Essentially, I am trying to compute a pooled F stat with num/denom df in order to calculate the pooled R^2. To do this I am following this textbook: "Multiple Imputation in Practice with Examples using Iveware" Section 3.6.3-3.6.5. 

Once I computed the table below I used this formula to get my R^2 value: rsq = dfnum*F/(dfnum*F + dfdenom). However, the computed, "pooled" R^2 doesn't make sense when comparing it to the 20 imputed datasets. The imputed datasets all have R^2 values around 0.5. Moreover, even in my complete case analysis, the R^2 value is also around 0.5 (I don't have that much missing data). So, it doesn't really make sense to me that this R^2 value is now 0.1? Any help would be very appreciated.