Hi Seyda, 

Please see this article, which has a section on missing categorical data. http://www.annualreviews.org/doi/abs/10.1146/annurev.psych.58.110405.085530

I would suggest to impute itemwise. Otherwise youbare loosing information you already have when you define a sumscore as missing even might just one item is missing. O would check MICE for R, i think it offers a lot of flexible possibilities to imputr different kinds of data.

Best

Thanks for valuable suggestions.

Mr Heinrich, could you please explain what you mean by saying itemwise impute? I am able to impute categorical data so far. However, the problem is when I do some descriptive statistics, system-missing values have emerged in large numbers (34) and I don't understand why. Do you know the answer or how to handle with these system-missing values?

Thanks

Sorry, i was not precise enough. I would suggest to impute tue item scores, and not the sum scores. The sum scores can be formed after the imputation process. 

It is difficult to say why you have system missings, even they should be imputed. Have you checked the SPSS output? maybe there are some hints why this occurs. Maybe these cases have no data or only data that is NOT part of your imputation model, so that imputation can't be performed, but I#m not sure.

Best

Dear Seyda,

I understand that you would like to impute values for a group membership variable.

The imputation technique can be increasingly complex with the number of categories. For example, you could do imputation on a few binary variables but imputation of nominal multicategorical variables should probably be kept to a minimum. 

The MICE package in R offers different imputation methods for binary (logreg) and nominal (polyreg and LDA). You can perform different imputation methods for different types variables simultaneously in R, while it is impossible or more imprecise with the missing data module of SPSS.

I hope that helps. Can you describe your variables?

Fantastic suggestions. Thanks all

suppose R is better in terms of handling the missing data, as well as other statistics such as SEM.

Will go through it,

Thanks again

Seyda Eruyar

You can use the imputation command instead of missing values analysis command. The command will generate a new set of data with number of desired imputations. Further analysis will then be carried out using the imputed datasets.