Hi there,

I think it might be useful to use the cor(new_data, use = "pairwise") function to get correlations based on pairwise complete data. Then you might find more highly correlated variables that should be removed before imputation.

Cheers,

Matt

~~~~~~~~~~~

Matthew Blackwell

Assistant Professor of Government

Harvard University

url: http://www.mattblackwell.org

On Sat, May 7, 2016 at 7:46 PM, Gülce Kale <gulce.kale@gmail.com> wrote:

Hi everyone,

I have a dataset with 20 features and 102 observation. When I use Amelia I got this error error: inv_sympd(): matrix appears to be singular. The resulting variance matrix was not invertible. Please check your data for highly collinear variables. Then I used spearman correlation to get rid of highly collinear features:

cor_spear <- rcorr(as.matrix(new_data), type="spearman")
cor<-cor_spear$r

I found the elements that have value >= 0.9 in my cor matrix. I removed corresponding features from dataset. There were some elements with value NA in cor matrix, I don't know what they mean but I removed them as well. Then, I only got 12 features which are correlated less than 0.9. However I still get this error:
The resulting variance matrix was not invertible. Please check your data for highly collinear variables.
How can I correct this error? 
Thank you!
--
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@lists.gking.harvard.edu

To unsubscribe from this list or get other information:

https://lists.gking.harvard.edu/mailman/listinfo/amelia