[amelia] Questions on Amelia (fwd)

---------- Forwarded message ---------- Date: Mon, 23 May 2005 14:27:36 -0400 From: "Marc P Berenson (berenson(a)Princeton.EDU)" <berenson(a)Princeton.EDU> To: king(a)harvard.edu, tercer(a)polisci.ucla.edu, scheve(a)umich.edu Subject: Questions on Amelia Dear Professors King, Honaker, and Scheve, I have been attempting an imputation with the Windows version of Amelia, and have a few questions. I would be very grateful for any assistance any one of you might be able to provide. 1. First, I seem to be having a problem similar to that others have had in the past with an AMvarnm error. Upon running the program with _AMidvar set to "1" such that Amelia will not consider the first variable, which is the weight, in imputing missing values, I receive the following message: _AMvarnm does not have enough variable names to match each variable in the dataset. Add variable names, or set _AMvarnm=. for default variable names (default= var1 var2 ...). As I am running the Windows version, I can't set _AMvarnm myself. However, when I did not specify any variables under the _AMidvar setting, the program does run. (For reference, I also set _AMnds to 10 and specify some nominal and ordinal variables, all of which run OK if _AMidvar is not specified.) 2. In running the imputation, is it best to pare down the data set to just the variables needed to run the model so that the program imputes faster and smoothly or should other variables be included to serve a descriptive function as Amelia imputes? 3. I also had some difficulty in discerning what value I should place the prior at for my two separate data sets. One data set is of at least 25 variables, 988 observations and has missing values for up to a few hundred observations for each variable (0 to 326 with most in the 100s). The second data set is of at least 27 variables, 2024 observations and has missing values from dozens to hundreds (up to 659). When I ran these data sets through at the default setting, I did not get any message suggesting that I use a prior, but given the high level of missingness and some correlation among the variables, perhaps I should. Many Thanks, Marc -- Marc P. Berenson Ph.D. Candidate Politics Department Princeton University 130 Corwin Hall Princeton, NJ 08544 E-mail: berenson(a)princeton.edu - Amelia mailing list served by Harvard-MIT Data Center [Un]Subscribe/View Archive: http://lists.hmdc.harvard.edu/?info=amelia