Take the 2-minute tour ×
Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

The outcome $y$ in my dataset is count data. There are three possible exposure variables $e_1, e_2, e_3$ conceivable. These exposure variables are mutually exclusive, i.e. refer to different physical entities, they are different groups. However, the exposure variables are quite correlated.

I would like to estimate the influence of the groups on the outcome. That is, I would like make a statement like "The outcome consists of 10% $e_1$, 70% $e_2$ and 20% $e_3$."

What modelling approaches can I use to address this problem? So far, I have two ideas (the model type would be negative binomial or Poisson)

  1. Model averaging. Here I would fit three models $y \;\tilde\;\; \text{offset}(\log e_i)$ and combine these models using for instance the model.avg function in the MuMIn package in R. However, I am not familiar with model averaging and would like to know if this approach is sensible before I look into it.

  2. I could use the sum $e=e_1+e_2+e_3$ as exposure variable and add the fractions $f_i=e_i/e$ as additional parameters. That is, $y\;\tilde\;\; \text{offset}(\log e)+ f_1 + f_2$. Here I have the problem of interpreting the parameters to derive a statement on the groups' influence.

Any hint appreciated -- also on how to find out if the data does not help to make such a statement.

share|improve this question
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.