Hi Georgina, research indicates that after MI, we need to perform the analyses on each imputed data set using standard procedures and then, the results from these analyses are combined into a single inference (Rubin, 1987; Rubin & Schenker; 1986; and Little & Rubin, 1987). I do not use SPSS but if you are also able to use SAS, I could be provide some references. I hope this helps.

HI,

you did multiple imputation, the issue here is how much or how many imputation you did, and what was the reason for that? is it missing data, if you, how much the percentage, and what approach you used for imputation.

if the answer will be small amount of imputaiton, then it is safe to go ahead in your data analyses.

hope this help

Pat: The imputation gives me 5 datasets. Do I run the regression on each one? How do you pool the results of 5 regressions?

Muayyad: I had quite a bit of missing data which wasnt MCAR so needed to do MI to amend this. 

Hi Georgina, I am glad that you have the 5 imputed data sets (I was afraid that after having aggregated the datasets, you did not keep the 5 imputed). if your research requires that you run a regression analysis on each imputed dataset, then, yes, run your regression analysis. Depending on the statistical software that you are using, it should be an option for running all 5 analyses at once. For example, in SAS, one can include an statement for this using a "by" statement (e.g., by _imputation_;). In SAS, the "by" statement obtain separate analyses of observations in groups defined by the BY variable. Having all your 5 analyzes in one output should facilitate the analyzes because you can run an analysis calling, say, BY, CLASS, MODELEFFECTS, etc. I hope this helps.

I also think that after performing multiple imputation, you get, for example, in your case, 5 imputed data sets (I am assuming that you need to do multiple imputation due to missingness in your data), then you need to perform statistical analysis which was decided before based on your research question for 5 imputed data set separately. Then we can combine the result from these analysis.

Thats great! thanks you guys.

Just another question - again this is spss related so kind of regretting using this programme here as it really just seems counter intuitive with complex analysis like this- anyway-

I have generated a complex analysis plan because my data (ELSA)  is clustered and stratified. I have the MI value datasets but when I go to do descriptive stats through the complex analysis tab, there is no means of running this simultaneously with the imputed data. Is this a fault of spss 20 or is there another way to do it?

At this stage I am contemplating just weighting the data by one of the key variables and explicating that was all I could do based on the spss programme limitations-

Dear Georgina, I completely understand your feelings but I invite you to think that the fun is about to start! As I mentioned, I am a SAS user but I do have an SPSS guide to implement MI that I hope it is of help. In SAS, after MI, one "identifier" is created to distinguish each imputed set, the _imputation_, which is labeled 1, 2, and so on. Thus, we use this variable to do the analyses using a BY command (e.g., by _imputation_). The attached document explains that SPSS creates such variable: IMPUTATION_, which allows if wanted, to split the file by the IMPUTATION_ variable and do the pooling procedures. I hope this attached document helps. Do not get discouraged. Like I said, the fun is just about to start :). I have found the research gate community to be very nice and helpful and if you need further help, just write. Somebody will always try to help, me included.Take good care and keep up the great work. Best, Patricia

Thanks so much P. You're a diamond