Michael, You asked:
-----Original Message----- From: Listar To: impute digest users Sent: 4/22/01 2:00 AM Subject: impute Digest V2 #16 1) Given the limited range of response options (0 - 2) and the skewed nature of the data, is the use of MCMC estimation under the assumption of normality not appropriate? Response: For the most part, I'll leave this one to the more learned members of this list, except to comment that what you said in the first part of your message and the citations you put forward are consistent with my understanding of the situation: As long as you employ an analysis method post-imputation that adequately addresses the non-normality issue, the Gaussian-based imputation approaches should offer fairly robust performance. Note that the the SAS MI procedure does offer you the option of using transformations if you decide you want to go that route. 2) Assuming that I can proceed with the imputation using a normal model, should I impute within subscale, or should I impute using the full instrument? Response: I would think it would make most sense to impute at the finest grain level possible in order to make use of the maximal amount of information offered by the correlations among the items. I'll be curious to see what the other responses are to this question for I've occasionally wondered this myself when I've used MI. 3) If I should impute within subscale, how do I deal with the items that are assigned to more than one subscale? Response: Hmmm...that would be another argument in favor of imputing at the item level. 4) After imputing the missing data and I have several complete-data data sets, how should I combine the parameter estimates in the IRT analysis? I will have three parameters per item to estimate. How do I determine the between-imputation variance, the within-imputation variance, and the total variance? How do I determine the relative efficiency? Response: I'm not familiar with IRT, but if the procedure generates parameter estimates and standard error estimates for parameter estimates, you can use the MIANALYZE procedure in SAS to combine the parameter estimates and standard errors appropriately. Any advice or words of wisdom will be greatly appreciated. Thanks in advance, Michael Bohlig Response: I'm afraid what I've said here probably wasn't of much help. I'm looking forward to what others on the list have to say in response to your questions. With best wishes, Tor Neilands Center for AIDS Prevention Studies UC San Francisco [email protected]
