Mixed logistic ridge regression

Question

I would like to build a logistic regression model in which I will be looking for predictor variables having a significant effect on the breeding success of a raptor bird.

The predictors in the dataset are highly correlated, which led me to consider logistic ridge regression. Furthermore, I investigated different breeding grounds in which one or multiple birds have been breeding. Since this makes the data clustered, I would need to add the breeding ground as a random effect in the model.

Thus, I would need a 'mixed logistic ridge regression' approach if I am getting things right here. This paper suggests this approach too for another problem:

http://www.ncbi.nlm.nih.gov/pubmed/22049265

Are there any people aware of the availability of an R package or something related having implemented a mixed logistic ridge regression approach as the paper and myself just described? I did not succeed in finding one.

Thank you.

library("sos"); findFn("lasso mixed") and findFn("ridge mixed") are pretty good starting points ... — Ben Bolker, Oct 03 '14 at 22:43

score 2 · Answer 1 · answered Oct 30 '14 at 18:33

2

There is an R package named glmmLasso that performs variable selection and shrinkage for generalized linear mixed models using the L1 penalty (the Lasso):

http://cran.r-project.org/web/packages/glmmLasso/index.html

The Lasso isn't the same as ridge regression (which uses the L2 penalty) but will still perform shrinkage of coefficients for correlated variables.

answered Oct 30 '14 at 18:33

RobertF

6,084

In addition to the papers referenced in the glmmLasso documentation, there is this 2011 study which incorporates ridge regression with linear mixed effect models (alas not logistic regression): http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3202941/. Maybe the same approach could be used with logistic regresion? – RobertF Oct 31 '14 at 14:31
That's correct, Robert. I found the paper too. I also agree this same approach could be used with logistic regression. However, I am afraid that it has not yet been implemented in any R package. – KoenVdB Oct 31 '14 at 19:08

score 1 · Answer 2 · answered Oct 03 '14 at 20:48

1

This might do what you want:

http://cran.r-project.org/web/packages/ridge/index.html

From the description:

Linear and logistic ridge regression for small data sets and genome-wide SNP data

answered Oct 03 '14 at 20:48

Roland Seubert

111

1

This is not what I need as it cannot incorporate random effects in the model. – KoenVdB Oct 04 '14 at 06:29

Mixed logistic ridge regression

2 Answers2

Linked