Re: [amelia] Imputation too slow , and numerical options for tscs data?

Hi Suveyda, I've had a similar experience with using TSCS data, it can take a very long time, depending on the circumstances. Regarding the ridge prior, it's good to start with a value of 0.5 and then move your way up. For example, the code would read: empri = .05*nrow(data). When you say 40, do you mean you are using a 0.4 ridge prior? If you're using 40*nrow(data) in your code this may be a source of error. The time to complete the imputation varies according to a number of things. The first being the system you are running on and the number of datasets you are imputing (the m). Given the degree of missingness, the number of observations and the number of variables you describe, I'm afraid it will take some time. Generally , TSCS data takes much longer due to the number of interactions. For example, using a much smaller TSCS dataset with a 0.5 ridge prior took me approximately 4 hours. Using this as a heuristic, I think your problem will take much longer. Exactly how much longer, though, also depends on your code and what you're asking Amelia to do. Might I first suggest having a look at the Amelia manual (particularly sections 4.5-4.8) here: http://r.iq.harvard.edu/docs/amelia/amelia.pdf. I think this will guide you through most of your queries. Also, if you post your code to the list it would be easier to respond with a better answer about specific things - the key one being whether or not the syntax is correct. Hope this helps. Cheers, kyle From: amelia-bounces(a)lists.gking.harvard.edu [mailto:amelia-bounces@lists.gking.harvard.edu] On Behalf Of Karakaya, Suveyda Sent: Monday, 18 June 2012 7:38 AM To: amelia(a)lists.gking.harvard.edu Subject: [amelia] Imputation too slow , and numerical options for tscs data? Dear Prof King, I am new to Amelia. I try to impute a data set with 8400 obs and 32 vars in total, 14 vars have missing obs, some a lot like around 4 000. Since I have many missing obs, I set the ridge prior to 40, is that correct? I also chose time spline with k=1 and interaction with cross section, is that also correct option for a tscs data? My second question is: how long on average does it take to complete imputation? I am waiting for 12 hours and do not see any progress at all yet, the computer is running but there is no imputation yet. Is it because I put ridge option to 40 or because of many logical bounds and log transformations that I set for my vars? Thank you very much Suveyda Karakaya PhD Candidate The University of Tennessee