[R] Loglinear models for missing data

fbielejec fbielejec at gmail.com
Mon Dec 6 20:43:07 CET 2010


I have the data in the following form:

   Location Variable Value Week
1   Africa   Africa    21 4 weeks
2     Asia   Africa     0 4 weeks
3   Canada   Africa    17 4 weeks
4    China   Africa    29 4 weeks
5   Europe   Africa    NA 4 weeks
6    Japan   Africa    68 4 weeks

It is a (melted) three-way count (Value is counts) table where for
example first row has the following meaning - when the Variable had its
maximal count, the Value for Location was 21, 4 weeks prior (covariate

The data has some missing Values, which I would like to impute.

What I have so far is a logit model predicting NA's in the Value, to
try to spot good predictors for missing entries. With those I hope to
come up with a loglinear (poisson) GLM and try to impute the NA's.
However coming up with a decent non-saturated model is difficult given
the data. 

Could You point me towards sth that could capture the nature of the
problem? Are lag models a good lead here?

while(!succeed) { try(); }

More information about the R-help mailing list