Premium
Can We Determine the Significance of Key‐Event Effects on a Recruitment Time Series?—A Power Study of Superposed Epoch Analysis
Author(s) -
Prager Michael H.,
Hoenig John M.
Publication year - 1992
Publication title -
transactions of the american fisheries society
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.696
H-Index - 86
eISSN - 1548-8659
pISSN - 0002-8487
DOI - 10.1577/1548-8659(1992)121<0123:cwdtso>2.3.co;2
Subject(s) - nonparametric statistics , statistics , parametric statistics , autocorrelation , statistic , statistical hypothesis testing , test statistic , epoch (astronomy) , mathematics , series (stratigraphy) , statistical power , econometrics , computer science , geology , paleontology , computer vision , stars
A persistent question in fishery research is whether extreme environmental events, such as climatic perturbations or discharges of toxic substances, influence recruitment. Superposed epoch analysis has been proposed as a statistical test to address such questions. In a superposed epoch analysis of recruitment, a test statistic is computed from the differences between recruitments in years with extreme environmental events (“key years”) and recruitments in immediately surrounding years; the significance of the test statistic can be determined either parametrically or nonparametrically. Here we examine the power of two parametric and four nonparametric test statistics to detect, in a variety of simulated data sets analyzed by the superposed epoch method, associations between key events and unusual values of recruitment. The statistical significance of the nonparametric test statistics is determined by randomization, the significance of the parametric statistics by consulting tabled distributions. Under the simulated conditions, we observed essentially no loss of statistical power when conducting the superposed epoch analysis with a randomization test when the parametric approach was also appropriate. However, in situations when parametric testing was not appropriate, the randomization test was often much more powerful than a parametric test. We also evaluated the statistical power of superposed epoch analyses conducted with test statistics in which recruitment data from each key year were compared in a paired fashion to data from the surrounding years. For data with strong trend or a high degree of autocorrelation, such paired test statistics outperformed the corresponding unpaired statistics; otherwise, the unpaired statistics tended to be more powerful. In testing simulated conditions patterned after the population of chub mackerel Scomber japonicus off southern California, we estimated the power of the proposed randomization test as approximately 0.35 to 0.50.