[R] partial mantel tests "Ecodist"

[Sarah Goslee]

Hi David,

On Fri, May 8, 2009 at 10:27 AM,  <ddepew at sciborg.uwaterloo.ca> wrote:

> My questions are as follows;
>
> 1) can "raw" data be used to construct the dissimilarity matricies? or
> should they be standardized? different variables have different measurment
> scales, my inclination is to standardize, but I don't know if this will
> dampen relationships between variables.

I'd standardize, especially if you're using Euclidean distances. The Goslee
and Urban JSS paper on the ecodist package goes into more detail (as
do some of the references cited therein).

> 2) If env1 and another variable are correlated, is the appropriate test
>        var1 ~ env1 + env2 + space?,
> or
>  var1 ~ env1 + space and then var1 ~ env2 + space?

Test for what? The first one partials out both env2 and space from the
relationship of var1 ~ env1, a very different thing than the second
example.

> 3) interpretation... Does the value of "r" (i.e. + or -) imply spatial
> overlap (+) or spatial exclusion (-)?

A negative value for r is usually uninformative (unless you've used
particular data transformations or something otherwise unusual). The
Mantel test question is generally: do differences in X correspond to
differences in Y, so the test you want is whether r > 0. Again, see the
JSS paper discusses this further.

Sarah

-- 
Sarah Goslee
http://www.functionaldiversity.org