It is suggested for reducing over-fitting of models, the variables should not be auto-correlated.
When each variable was tested through SAM module, it gave individual values of Moran's I which ranged from 0.2-0.6.
The question is how to know that that the variable has spatial auto-correlation?
Does Moran's I value of 0.2 with P level of 0.001 indicate auto-correlation?
and what is the cutoff limit for the value of Moran's I to select less or non auto-correlated variables for model development?
Hi Prakash Pradhan jee,
if you are looking for Moran's I test in spatial statistics through ARCGIS then there is a separate method where and how to handle this issue. If you are handing community data then I recommend you have a look vif (variance inflation factor) in vegan package. If you are working with multiple regression I suggest you have a look at step function and AIC values in glmm package.
Goodluck.
Chitra
Prakash,
Are you concerned with spatial autocorrelation in the residuals of the model or are you concerned with multicollinearity between explanatory variables?
If you are concerned with spatial autocorrelation in the residuals of the model than I suggest you consider the p-values associated with each Moran's I value. It is possible to have a larger Moran's I value that is not significant compared to a smaller Moran's I value that is significant.
If you are concerned with overfitting your model and/or multicollinearity among the potential variables, then I suggest the VIF function in the vegan package in R as suggested by Chitra.
Good luck,
Tim
Please note that Moran's I quantifies spatial correlation in a different sense than does a variogram/cross variogram or an autocovariance function. Hence it depends on how you are going to model/characterize "spatial distributions". As an aside you might want to do more exploratory analysis of your data, e.g see the free software called "ggobi". It comes with an electronic textbook and various sample data sets
For a way to decide on explanatory variables see attached.
Article Where the bears roam in Majella National Park, Italy
After preliminary MaxEnt run, which variables should be removed first,
is it those which are showing 0% contribution to the models?
or those which are giving less than 0.8 AUC value when used in isolation (by examination of test jackknife data)?
Thanks
Prakash
- Badges
- Science topic
- Similar topics
- Geoscience
- Geoinformatics
- Geographic Information System