Space–Time Correlation and Its Effects on Methods for Detecting Aquatic Ecological Change

Abstract

The analysis of variance (ANOVA) is commonly used to analyze observations collected from aquatic monitoring programs designed to detect ecological change. ANOVA assumes that the deviations of the observations from their true means (the errors) are uncorrelated in space and time. Aquatic monitoring data often violate this assumption. The results of Monte Carlo simulations using simulated data generated from both statistically and mechanistically based models show that the presence of either spatially or temporally correlated errors can significantly affect the outcome of ANOVA tests. In practice, spatial correlation is more likely to be a problem than is temporal correlation, given typical monitoring frequencies. The effects of spatial correlation can be minimized through judicious use of control station pairing in the monitoring design. However, when insufficient flexibility exists in the monitoring design, alternate models, such as multivariate time series analysis, or multivariate analysis of variance, must be used in place of ANOVA.