Randomization and Monte Carlo Testing
The use of randomizing techniques for the analysis of data is now a widely accepted method. For some data sets, there simply is no acceptable parametric or nonparametric statistical test to apply to the data. Only randomizing techniques can be used. The greatest difficulty is that no standard program or routine can handle all requests. BEAST 2005 includes several randomization techniques for sequence analysis. Very often, a custom computer program must be written to implement the randomization technique. Windward Technology (WWT) will write the program for you and then give you both the compiled and the source code for your own use.
You must very careful define your null or alternative hypotheses and diagnostic statistics. If you want to test whether some set of data meets a theoretical model, you must carefully define the model. WWT will then interview you to carefully define your needs and craft a Monte Carlo test for your study. If, however, you need to see whether two or more sets of observations could have come from the same undefined population, a Randomization test will be crafted. Both techniques sample from the theoretical model or your data 10,000 to 100,000 times using exactly the techniques and sample sizes that were used in the study. This establishes the statistical boundaries for whatever diagnostic statistics were desired and the probability that your data could be the result of sampling error or random events.