# Combining multiple imputation results for hierarchical regression in SPSS

I'm running a hierarchical regression model in SPSS. I used multiple imputation to handle missing data (14 imputations) and then ran the regression. The regression is:

Step 1: 3 dummy coded predictors, 2 continuous predictors Step 2: 1 continuous predictor DV: Continuous outcome

SPSS generated individual output for my original data set plus each of the 14 imputations (15 sets of results total). If I need to report change in R2, the ANOVA results, and the significant coefficients, how do I combine these values from the imputed sets to get a pooled set of results? I've looked at Rubin's rules (http://sites.stat.psu.edu/~jls/mifaq.html#howto) but I'm not sure how to apply this to multiple regression. I would appreciate any advice, links, or examples on how to combine this output!

Thanks!

• I'm not aware of a pre-packaged way of combining MI output in SPSS, but I once followed the guidelines from the link your provided. It required me doing some coding (or you could hand calculate some things feasibly). You apply the rules the same for multiple regression as for any other regression: for each estimated quantity (a coefficient or p-value, for example) let Q represent that quantity of interest (Q_i ranges over 14 values in your case) and then run through the calculation steps in that link. – zkurtz Sep 14 '13 at 19:53
• SPSS computes the combined ("pooled") estimates and other relevant statistics for you. See Edit - Options - Multiple Imputation if the setting is to show pooled results. – ttnphns Sep 15 '13 at 7:57
• ttnphns - Thanks - after splitting the file I was able to get pooled coefficients. Still no luck with the ANOVA f stats or change in R2. I may just have to report those stats based on unimputed data. – user30295 Sep 16 '13 at 13:30