0

I am currently running some generalized least squares (gls) models with a spatial correlation structure, using the R package nlme; they look like the following one:

bodymeanee2 <-gls(BS_mean_SCALED ~ LAT2_SCALED + LAT_SQUARED2_SCALED + SPRICHNESS_LOG_SCALED +
 SPRECORDS_LOG_SCALED + SST_mean_SCALED + NPP_mean_SCALED + OCEAN, correlation = 
corExp(form=~LAT_SQUARED2_SCALED+LON_SQUARED2_SCALED, nugget=TRUE), data=latitudecsv)

I am trying to extract the relative importance of each explanatory variable (predictor) by model avering, as the sum of the Akaike weights of the submodels in which the variable appeared, from the set of submodels including all combinations of explanatory variables. I generate the complete set of submodels with all the possible combinations of explanatory variables using the function dredge of the package MuMIn:

ms4.1<-dredge(bodymeanee2)

I would like to know how to extract the relative importance of each explanatory variable once the function dredge finishes. I would also like to know if it's normal that the function 'dredge' tooks hours and hours in finishing the complete set of submodels. Is there another possibility in less time?

Thank you in advance

Diego Cepeda
  • 35
  • 1
  • 8
  • that would be on your ANOVA output. Look at p values – Jorge Lopez Feb 22 '20 at 19:07
  • that should tell you how significant they are – Jorge Lopez Feb 22 '20 at 19:08
  • Thank you for your response. I am not looking for the degree of significance, which I already got in the ANOVA output as you suggested. Which I want to know is the relative importance of a predictor, because a variable may be statistically significant but still explain very little of the response variable in a model. – Diego Cepeda Feb 22 '20 at 19:12
  • Are you looking for the [effect size](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3444174/)? – dario Feb 22 '20 at 19:25
  • Yes, that's it. In other linear models that I performed, without spatial correlation with the function lm, I did this with the coefficients of determination (r squared) and the function 'dredge' that I mentioned. However, I think you cannot have r squared for gls, correct me if I'm wrong – Diego Cepeda Feb 22 '20 at 19:31

0 Answers0