Re: [amelia] Problem with logs - supplemental

Dear Ms. Janz, Since you were kind enough to post the data so we could track down this issue, I thought I'd just mention a few more specifics about your issue in the context of your data, off-list since it's your own data. The Amelia error revealed itself because there are variables to be logged, "l_US_fdi_petrol" and "l_US_fdi_mining", which had no observations in common (every observation is missing one or the other or both). That's not a problem with the data necessarily, just a curious feature I assume. Additionally, I see that in working around this you have made additional copies of all the variables to be logged with all non-positive values changed to 1's. That is fine, if the negative values really are better measured as 1's, but not necessary. Amelia will shift variables to be logged to the right so they are positive in the event there are negative values for the imputation dataset, and then return the imputations on the original scale. That's slightly different than your fix; it's of course your choice to substantively decide which you think is appropriate for your data. Finally, the dataset you posted has both the variables to be logged, and the original unlogged variables in the imputation model. That's certainly identified, and a good choice if you don't know whether the effect is with the linear or logged term, but perhaps also an oversight if you believe from theory these need to be logged. Thanks again for your help in tracking this down and making your data so easily available. Let me know if anything just raises more questions, Best, James. -- James Honaker, Senior Research Scientist //// Institute for Quantitative Social Science, Harvard University ________________________________________ From: Honaker, James Sent: Friday, January 04, 2013 6:49 PM To: N. Janz; amelia(a)lists.gking.harvard.edu Cc: nicolejanz(a)gmail.com Subject: RE: [amelia] Problem with logs Dear Ms. Janz, There was an error in one of the checking routines we use in Amelia to see if there are any problems in the data that can be preempted. I fixed the error and it will be changed in the next Amelia release. In the meantime, you can avoid this issue by turning off the checking routines as an option in the amelia() call with the "incheck" argument. In your example, this would be: amelia.out <- amelia(n, m=5, ts="year", incheck=FALSE, cs="country",intercs=TRUE,polytime=1,logs=vlogs) Thanks for your help in finding this error and making your data available so we could resolve it. Best, James. -- James Honaker, Senior Research Scientist //// Institute for Quantitative Social Science, Harvard University ________________________________________ From: amelia-bounces(a)lists.gking.harvard.edu [amelia-bounces(a)lists.gking.harvard.edu] On Behalf Of N. Janz [nj248(a)cam.ac.uk] Sent: Thursday, January 03, 2013 4:31 PM To: amelia(a)lists.gking.harvard.edu Cc: nicolejanz(a)gmail.com Subject: [amelia] Problem with logs Dear all, Amelia works fine when I do NOT specify any variables that should be logged. As soon as I include logs, I get an error ("#Error in matrix(unlist(value, recursive = FALSE, use.names = FALSE), nrow = nr, : negative extents to matrix"). I checked that the variables to log can be logged (no negative/zero values). I'm grateful for any hints. Best, Nicole --Rcode-- #load MI package require(Amelia) set.seed(1000) #load data set load(url("http://schreiberin.de/nicole.Rdata")) n <- nicole dim(n) #3822 59 class(n) #"data.frame" # Declare variables to be logged vlogs <- c("GDP_const2000" , "GDP_curr" ,"population" , "lifeexp" ,"infmortality" ,"trade" , "l_US_fdi_total" , "l_US_fdi_petrol" , "l_US_fdi_total_manuf","l_US_fdi_food" , "l_US_fdi_chemical" , "l_US_fdi_prim_fab_metal" ,"l_US_fdi_machinery", "l_US_fdi_electrical" , "l_US_fdi_transport" ,"l_US_fdi_whole_trade", "l_US_fdi_depository" , "l_US_fdi_finance_except" ,"l_US_fdi_mining" , "l_WB_FDI_percentGDP" , "l_WB_FDI_curr" , "l_UN_FDI_flow" , "l_UN_FDI_flow_pgdp" , "l_UN_FDI_stock", "l_UN_FDI_stock_pgdp" ) amelia.out <- amelia(n, m=5, ts="year", cs="country",intercs=TRUE,polytime=1,logs=vlogs) #Error in matrix(unlist(value, recursive = FALSE, use.names = FALSE), nrow = nr, : negative extents to matrix --END Rcode-- -- Nicole Janz, PhD Cand. Lecturer at Social Sciences Research Methods Centre 2012/13 University of Cambridge Department of Politics and International Studies www.nicolejanz.de | nj248(a)cam.ac.uk | Mobile: +44 (0) 7905 70 1 69 4 Skype: nicole.janz -- Amelia mailing list served by HUIT [Un]Subscribe/View Archive: http://lists.gking.harvard.edu/?info=amelia More info about Amelia: http://gking.harvard.edu/amelia Amelia mailing list Amelia(a)lists.gking.harvard.edu To unsubscribe from this list or get other information: https://lists.gking.harvard.edu/mailman/listinfo/amelia