havior of these measures was calculated for small samples for all tables from 2 x 2 to 5 x 5; the resulting tables are presented and may be used in significance tests. Scientists who decide to use Colwell's measures should sample at least twice during the period of the most rapid fluctuation of interest. To accommodate asymmetrical fluctuations, they should sample frequently enough to include the regularly encountered minima and maxima and calculate contingency over many time scales to allow the data to reveal their natural periodicities. If there is some question concerning detection of a significant periodicity by increasing sample size, then pilot calculations should be run on samples of 5-6 cycles of the longest periodicity desired. If M exceeds 0.2, more data should be gathered to double the sample, which will almost certainly result in detection of a significant periodicity. If not, larger samples will probably not change the conclusions reached. Colwell's measures performed well on metric data where they could be compared with Fourier transforms. Fourier transforms remain the method of choice on metric data, but that Colwell's measures performed well and were consistent with Fourier measures on metric data gives us confidence in using them on nominal and ordinal data, where Fourier analysis is inappropriate. I urge their use on nominal and ordinal data and on smaller data sets than those traditionally exposed to time-series analysis.