1

A few months back I created a spatial logistic regression model of terrestrial skunk rabies in Arizona/Colorado/New Mexico/Utah. A link to the project can be found here (https://docs.wixstatic.com/ugd/0a0a58_4a5a522e0ee944e7a186e8d1f106010d.pdf)

I'm attempting to run the model for Colorado only. Aside from only modeling for Colorado the changes to the model include:

  • A higher sampling ratio. 3:1 controls/cases
  • More sample points 578 positive specimens
  • More years of sampling. 2008 - 2017
  • More years of independent variables averaged. (e.g. monthly surface temperature averaged 2008-2017 vs 2013-2017)
  • reduction in the # of rabies positive species (there are fewer species in CO positive for rabies than in the four states)
  • changing from the range of species to the modeled habitat of a species based on USGS Gap View

The results for Colorado are very similar to the Colorado results in the Four Corners model. The same non-autocovariate variables seem to be significant, However the estimates as well as predicted risk are infinitesimally smaller:

enter image description here

enter image description here

I also tried standardizing my variables given the wide variety of scales. The results were the same as using non-standardized variables.

What is going on? My only hypotheses are that the slight changes in the variable definitions altered the results.

Also that the heterogeneity of the landscape in four states maybe teases out more differences than just in one state?

E.g. You don't find rabid skunks in the barren desert of Utah compared to prairie of CO, but you can find rabid skunks in the prairie steppe of Colorado and the scrubland of Western Slope.

I used Spatial Analyst for some map algebra and cell statistics functions and multi-values to points. The first go around I used ArcGIS/QGIS/R. This go-around just ArcGIS and R. – BenW 33 mins ago

  • You have a tag for ArcGIS Spatial Analyst but is that the GIS software that you are using to do this? – PolyGeo Dec 19 '18 at 22:47
  • Yep I used Spatial Analyst for some map algebra and cell statistics functions and multi-values to points. The first go around I used ArcGIS/QGIS/R. This go-around just ArcGIS and R. – BenW Dec 19 '18 at 23:16
  • 1
    Looks like mess up with units in your model. – FelixIP Dec 20 '18 at 7:09
  • 2
    Is this a GLM with no consideration for spatial autocorrelation? And what are you comparing it to and which estimates are infinitesimally smaller? Do you mean "infinitesimally small" (ie zero) or "smaller" (ie not really different)? – Spacedman Dec 20 '18 at 9:27
  • 1
    The intercept terms are concerning, for two reasons. In the four-state model, the p-value reported for the intercept is obviously incorrect. If that's not a copy-paste error, you should be deeply concerned about the software. In the CO-only model the standard error of the intercept is huge, which implies the model's prediction limits for any particular value would essentially be from 0 to 1. It's very hard to determine what's going on from just this output, but one avenue of research that should be high on your list would be to check collinearity diagnostics. – whuber Dec 20 '18 at 22:24

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.