No apologies necessary!  Thanks, Matt!  Good points.  I hadn't considered the possibility that the model might be nearly saturated (in addition to having some collinearity problems).  I chose 12 more variables to remove, and Amelia ran!  You made my day!

Wendy

> From: blackwel@fas.harvard.edu
> Date: Wed, 26 May 2010 14:44:03 -0400
> Subject: Re: [amelia] infinite or missing values in 'x'º
> To: wnuzzo@hotmail.com
> CC: amelia@lists.gking.harvard.edu
>
> Hi Wendy,
>
> Sorry for taking so long to get back to you. This feels like a
> numerical problem stemming from some variables being linear
> transformations of one another. You may want to track down the
> problematic variable by including variables one at a time (perhaps
> using a for loop).
>
> Another problem might be the number of parameters for the number of
> cases that you have. 69 variables implies roughly 2300 parameters and
> that is close to the number of observations you have.
>
> Cheers,
> matt.
>
> 2010/5/13 Wendy Nuzzo <wnuzzo@hotmail.com>:
> > Hello!
> >
> > I am having an issue running Amelia to impute missing dichotomous variables
> > (listed under "ords" below).  My data set has n=2628 and 69 variables with
> > missing data.  For about half of the variables, the missingness is less than
> > 20%.  For the other half of the variables, the missingness is between 30%
> > and 40%.  I have updated R & Amelia on my Mac:  R 2.11.0 and Amelia 1.2-17.
> >
> > This is the code I'm running:
> >
> > az.out<-amelia(x=dat, m=5,
> > p2s=2,ords=c("recid","gender","bin_race","pf2_1","pf2_7","pf3_1","pf3_4","pf3_6","pf4_5","pf4_16","pf5_1","pf5_10","r2_2","r2_3","r2_4","r2_5","r2_8","r3_2","r3_3","r4_1","r4_3","r4_9","r4_10","r4_12","r4_14","r4_15","r5_2","r5_3","r5_4","r5_6","r5_7","r5_8","r6_1","r6_2","r6_3","r6_4","ipf2_1","ipf2_7","ipf3_1","ipf3_4","ipf3_6","ipf4_5","ipf4_16","ipf5_1","ipf5_10","ir2_2","ir2_3","ir2_4","ir2_5","ir2_8","ir3_2","ir3_3","ir4_1","ir4_3","ir4_9","ir4_10","ir4_12","ir4_14","ir4_15","ir5_2","ir5_3","ir5_4","ir5_6","ir5_7","ir5_8","ir6_1","ir6_2","ir6_3","ir6_4"),
> > idvars=c("recidcount"))
> >
> > And this is the error it results in:
> >
> > amelia starting
> > beginning prep functions
> > Variables used:  recid age gender bin_race pf2_1 pf2_7 pf3_1 pf3_4 pf3_6
> > pf4_5 pf4_16 pf5_1 pf5_10 r2_2 r2_3 r2_4 r2_5 r2_8 r3_2 r3_3 r4_1 r4_3 r4_9
> > r4_10 r4_12 r4_14 r4_15 r5_2 r5_3 r5_4 r5_6 r5_7 r5_8 r6_1 r6_2 r6_3 r6_4
> > ipf2_1 ipf2_7 ipf3_1 ipf3_4 ipf3_6 ipf4_5 ipf4_16 ipf5_1 ipf5_10 ir2_2 ir2_3
> > ir2_4 ir2_5 ir2_8 ir3_2 ir3_3 ir4_1 ir4_3 ir4_9 ir4_10 ir4_12 ir4_14 ir4_15
> > ir5_2 ir5_3 ir5_4 ir5_6 ir5_7 ir5_8 ir6_1 ir6_2 ir6_3 ir6_4
> > running bootstrap
> > -- Imputation 1 --
> > setting up EM chain indicies
> >
> >  1Error in eigen(thetanew[2:nrow(thetanew), 2:ncol(thetanew)], only.values =
> > TRUE,  :
> >   infinite or missing values in 'x'
> >
> > I was concerned about collinearity, so I calculated the VIFs and removed the
> > variables with high VIFs.  Now all the variables have VIF < 10.   I've also
> > tried a ridge prior of up to 300 to account for the high missingness rate.
> > I've looked in the listserv archives and of course "googled" my error as
> > well.
> >
> > Any guidance on what might be causing this error would be appreciated.
> > Thank you!
> >
> > Wendy Nuzzo
> > soon-to-be M.S. Stat
> > Portland State University
> >
> > ________________________________
> > Hotmail has tools for the New Busy. Search, chat and e-mail from your inbox.
> > Learn more.

---

Hotmail is redefining busy with tools for the New Busy. Get more from your inbox. See how.