I am currently doing a project on environmental determinants of malaria vector distributions. I'm using remote sensing data for environmental variables linked via GIS. I have run univariate (binomial) models of my extracted environmental variables against vector presence/absence. Each of the factors I have extracted from over four scales: at the points of the vector sampling (within household compounds), the mean over a 100m radius, 500m, and 1km.

For some variables, more than one spatial scale was significant in the univariate analyses. I want to compare these univariate models to find out which is the most influential spatial scale for that variable, and use that in my multivariate.

The AIC values I have seem to disagree with the Log likelihood values and LRT test results (if I'm doing it right that is...), and so I'm not sure how to go forward. I thought that I would need to select the model with the lowest AIC and lowest LRT test P-value I have the AIC for each model, and the Log likelihood. I hope it is correct that I have tried to compare them by taking turns to compare each univariate to a maximal model (one with all scales) with an LRT test. Here is an example.

enter image description here

And also an example of what I get when i run lrtest() instead:

enter image description here

Before doing these things, I have the odds ratios etc. for the models elsewhere. Should I select the ones with the lowest AIC, or lowest P values in the above tests?

Thanks so much!



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.