# 6. Randomisation tests

TREEMAP offers a range of randomisation tests (some more experimental
than others) designed to help evaluate the relationship between
host and parasite phylogenies.

## 6.1 Is there evidence for cospeciation?

Given that the "best" reconstruction requires *n*
cospeciation events, how do we know if this is meaningful or not?
One approach is to ask what are the chances of getting *n*
cospeciations by chance alone. TREEMAP allows you to randomise
either the host or the parasite tree, or both simultaneously.
For each randomised tree (or pair of trees), TREEMAP uses its
heuristic algorithm to compute the maximum number of cospeciation
events between the two trees.

## 6.2 Generating random trees

You can choose to randomise either the host tree, the parasite
tree, or both simultaneously. Once you have chosen which option
you desire, TREEMAP then displays a dialog box asking you how
many random trees you want to generate, under which model they
are generated, and how to seed the random number generator.

If you choose **User value** then TREEMAP uses the value you
supply, otherwise TREEMAP takes a seed from your computer's system
clock. If you supply a seed it must be in the range 1 to 2,147,483,646.
If you use the same seed each time then the same set of random
trees will be generated.

The result of the randomisation test is a histogram of the frequency
of the maximum number of cospeciation events between the pairs
of trees generated. This is displayed in the Histogram window.

## 6.3 Comparing branch lengths

If your trees have branch lengths (e.g., amounts of sequence divergence)
then you can do one or more simple randomisation tests to assess
the significance of the correlation between branch lengths in
the host and parasite trees.

Choose the **Randomise | Branch lengths** command. TREEMAP
displays a dialog box similar to the one for random trees:

**Additive trees**

If you have additive trees, TREEMAP will randomly reassign the
observed branch lengths in the parasite tree, then compute the
correlation coefficient between the randomised branch lengths
and the observed host branch lengths.

**Ultrametric trees (molecular clock)**

If you have ultrametric host and parasite trees, you can either
use a simple Yule model to generate random coalescent times for
the parasites, or randomise the

intervals between the observed coalescence times to generate new
coalescence times (this will be described in more detail elsewhere).

**Results**

As with the random trees, the Histogram Window displays the results
of the randomisation test.