Multidimensional Contingency Table Analysis as a Tool for Biological Oceanography

Abstract

AbstractAnalyzing biological oceanographic data sets can be hampered by several practical problems, e.g., the large experimental errors generally associated with low concentrations, the non-linearity and even non-monotonicity of biological responses to environmental changes, and the interactions between causal variables. The generalized linear model can be used to analyze sets of non-monotonically interacting variables. When experimental errors are large, grouping values into discrete states helps to reduce the impact of such errors on statistical analyses. Log-linear models (multidimensional contingency table analysis) can be fitted to interacting qualitative variables and also ordered variables partitioned into discrete states. As an example, a data set from Baie des Chaleurs (Gulf of St. Lawrence, Canada) that includes physical, chemical, and biological variables is analyzed using conventional oceanographic and statistical approaches and also multidimensional contingency table analysis. The horizontal ...