www.cmm.bristol.ac.uk/forum

Hi all,

I have been trying to run a multiple imputation analysis using the practical14 worksheets for over 2 days now. It's been one problem after another and a lot of these I have managed to resolve using some tips from the forum so thanks everyone. Unfortunately I am stuck with an issue it seems no one else has experienced. After running the MI in Realcome, I import the imputation on MLwiN, when I click start analysis, i get this message in equations instead of the multiple windows I should see:
-2*loglikelihood(IGLS Deviance) : not available(3132 of 4873 cases in use).


I am not sure why the cases are not available, there are no error messages. Did anyone else experience this or have a solution?

Thanks all

The "Cases in use" numbers reported in the equations window reflect the missing data in the currently loaded data. While the imputation analysis is run this is temporarily replaced with each of the imputed dataset from Realcom, however once the analysis is complete and the model results are returned this is restored again, hence why there are still reported missing values in the output.

The results presented in the equations window represent a combination of the results for a model run on each of the imputed datasets. Rather than attempted to calculate an average of the likelihoods for each of these models MLwiN instead just reports these as not available as you observed.

Hi Chris,

Thank you for your responses.

I think your observation is right but the issue is that the MI is not being run, nothing happens beyond this screen, no analysis being completed. When I click start analysis (under the Imputation tab), within 5 seconds the display on the equation screen turns blue and that 'not available' message appears just before the 'cases in use'. My understanding is that this should not be happening, I should actually be seeing a number of different screens popping up until the MI is complete (which I assume the equation screen will then turn Green). So I am not sure why this is happening and no one else seems to have experienced this issue which is quite frustrating.

When you click "Start Analysis" what should happen is that messages related to reading in the various data files will be displayed in the status bar at the bottom of the window. If these files are small or reading is complete before the status bar updates then you may not see these however. The estimates will not turn green, as in MLwiN this signifies that the parameters have converged to a single value, which isn't the case here. If you run the model with your original data file with missing values (which will turn green), and then run the imputation analysis do the parameter estimates change? Could you also confirm which version of MLwiN you are using, as there have been bugs related to the imputation functionality in some previous versions?

To me this is a complex function. Thanks for your detailed feedback.
slope game