We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
On Measuring Fluctuating Environments: Predictability, Constancy, and Contingency.
- Authors
Stearns, Stephen C.
- Abstract
By simulating 640,000 time-series either 30 or 300 observations long, I evaluated the precision and power of significance tests for four measures of nonmetric time-series: Colwell's C and M and Goodman and Kruskal's @lR°C and @tR@?C. Significance tests for all four measures are powerful but do not fit published g2 and t values. An Appendix gives critical levels for the test statistics of all four measures for tables from 2 x 2 to 5 x 5 and for time-series with lengths 30 and 300. To detect the significant periodicities in a time-series with C and M, one should calculate the measures for many different time classes (contingency will peak when the number of time classes matches a natural periodicity), and sample at least twice as frequently as the period of the highest frequency fluctuation that one wants to detect (more frequently if asymmetric fluctuations are suspected). Colwell's measures were not as sensitive as Fourier transforms in detecting significant periodicities in metric test data, but their overall performance on metric data was good enough to justify their use on nominal and ordinal data. A method is given for estimating from pilot data the sampling effort required to detect a significant fluctuation.
- Subjects
TIME series analysis; MEASUREMENT; CONSTANCY; FOURIER transforms; CONTINGENCY theory (Management); POSSIBILITY
- Publication
Ecology, 1981, Vol 62, Issue 1, p185
- ISSN
0012-9658
- Publication type
Article
- DOI
10.2307/1936681