Skip to content

Distance for logistic models

2 messages · Daniel Patón Domínguez, Sarah Goslee

#
Dear all:

I am preparing a study using Non-Metric Multidimensional Scaling with some acoustic parameters. The final fitting indicate a R2 of 0,99 using canberra distance. However, the tendency of fitting between ordination distances and observed dissimilarities show a logistic form. I had tested 12 mathematical distances-simmilarity-disimilarity analysis but probably additional distances not include in dist or ecodist libraries can be used. Any suggestion?

Regards
8 days later
#
Hi Daniel,

Choosing a dissimilarity based on what gives you the best fit is
suspect, just like any other "try everything and take the lowest
p-value" approach.

Better, look at the properties of the various dissimilarities and
choose the one that best fits your data and your hypothesis.

Also note that NMDS fits the rank order distance of the dissimilarities.

Sarah

On Wed, Jan 10, 2018 at 10:07 AM, Daniel Pat?n Dom?nguez
<d.paton.d at gmail.com> wrote: