Error message

[Ken]

I have been using HyperNiche to explore relationships between 9-10
habitat variables and freshwater fish communities as well as a focal
species.

When I use the randomization test, I often get an error message that
states that there is no variance in one or more of my variables and
that HN is trying to divide by zero.

My initial reaction was that my infile was in error, but that is not
the case. And as with most ecological data, there is rather large
variance in my data.

Any ideas on what this is all about and ways to deal with this?

Thanks-

[Bruce McCune]

Check the bootstrap tab under Evaluate Selected Models. Do you have
the bootstrap resampling turned on? Depending on your data, this can
create variables with no variance. (Explanation: When you use
bootstrapping you are taking repeated random samples of your data.
This is sampling with replacement, using the same sample size as the
actual data. Say you have binary data for one of your predictors,
N=100, and there are 5 ones and 95 zeros. Quite often a bootstrap
sample will come up with all zeros for that variable.)

A randomization test by itself will not normally create empty
variables, because it is just shuffling within the response variable,
not resampling it.

One exception to this is if you have one or more categorical
predictors and a response variable with discrete values (e.g. binary
0/1 data). In that case it is possible to shuffle the data such that
within categories all response values are equal. This problem would
be most severe with binary responses and categorical variables with a
small number of sample units per categories. If this is the case you
might rethink how you assign those categories.

If neither of these suggestions applies to your data, I suggest
sending your data to MjM Software technical support (Help | Technical Support).

-Bruce McCune