Using only Categorical variables as predictors

[John Brazner]

Hi Bruce -

I'm struggling with some bird point count data that was collected in 3 different wetland types, 3 different ecoregions and in 2 different years.  My main question is whether there were significant differences in certain species abundances or overall richness or abundance (response variables) among wetland types, ecoregions and years or the interaction of those factors. I'm guessing you will say this is better suited to an ANOVA approach using GLMs or GLMMs  where the underlying distribution can be assumed to be negative binomial or Poisson, but is it inappropriate to use NPMR with only categorical variables to examine which models would provide the best fit and look at the relative importance of these different categorical factors? For example, I'd like to be able to say something like "for Species A, Wetland Type was the most important factor in a 2 predictor model that also included Ecoregion as a second significant predictor." Or is this all a waste of time, and I really need to figure out how to do this with a GLM or GLMM?

Thanks a lot - 

John

[Chris Bowman-Prideaux]

You may want to consider an MRPP or Permanova. I have used MRPP and it does a multiple comparison test. You could see if there were significant differences between years or ecotypes that way.

--
You received this message because you are subscribed to the Google Groups "HyperNiche and NPMR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hyperniche+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[John Brazner]

Thanks Chris. I’ve used MRPP to look for differences in the whole community of birds but neither MRPP nor perMANOVA allow me to look at differences in individual species. 

[Bruce McCune]

It actually is possible to use MRPP or perMANOVA with a single variable, and yes, it's possible to use HyperNiche with only categorical predictors. For a one-way design with a single response variable and one categorical predictor, these should start to converge in their "answers". They all use a randomization test to evaluate statistical significance. The choice of distance measure becomes (mostly) moot with a single variable, since we are evaluating differences in a 1-dimensional space.

In some cases people who are using MRPP or perMANOVA for evaluating community differences also use this for individual species in the same paper. The main advantage of this is that the consistency in the conceptual framework for the analysis. The same rationale could apply to HyperNiche -- if you are doing more complex models, say with both continuous and categorical predictors, or just multiple continuous predictors, and wanted to compare your results to models with categorical-only predictors, I'd say go for it.

Bruce

[John Brazner]

Thanks a lot Bruce. It doesn’t look like I can use perMANOVA unless my group sizes are equal (mine are not), so I guess MRPP would be the way to go. I’ll save the NPMR for the next paper where I will have quite a few continuous habitat variables along with a few categorical predictors.

 

John

[John Brazner]

One more thing Bruce - I'm hoping you can recommend the distance measure you think would be most robust for MRPP based on single species and how to correct the p-values for multiple comparisons among gropus. I can't use Sorenson's because of zeros in the matrix, so am thinking squared Euclidean might be the most robust. I'm also planning to divide 0.05 by the number of comparisons to provide a significance threshold. Does that seem reasonable? Thanks again - John

[John Brazner]

One more thing Bruce - I'm hoping you can recommend the distance measure you think would be most robust for MRPP based on single species and how to correct the p-values for multiple comparisons among gropus. I can't use Sorenson's because of zeros in the matrix, so am thinking squared Euclidean might be the most robust. I'm also planning to divide 0.05 by the number of comparisons to provide a significance threshold. Does that seem reasonable? Thanks again - John

On Friday, January 11, 2019 at 11:42:44 AM UTC-4, Bruce McCune wrote: