Nonmetric clustering analysis

Since cluster analysis methods are of limited nature, and MVA methods are not really data-driven, we must conclude that these methods as well as the metric MDS methods are not qualified as maximally data-driven data-mining aids. Therefore, our only hope is to promote the methods broadly called nonmetric MDS (nMDS) as a versatile general reduction methodology for relational data. 

Developed for the analysis of ecological data (Matthews and Hearne, 1991), nonmetric clustering and association analysis (NCAA) is a multivariate derivative of artificial intelligence research. NCAA has a fundamentally different approach to discovering patterns in data sets. 

Matthews, G.B., R.A. Matthews, and W.G. Landis. 1995. Nonmetric clustering and association analysis Implications for the evaluation of multispecies toxicity tests and field monitoring. Environmental Toxicology and Risk Assessment, Vol. 3, ASTM 1218. J.S. Hughes, G.R. Biddinger, and E. Mones, Eds. American Society for Testing and Materials, Philadelphia, PA, pp. 79-93. 

The following examples demonstrate the usefulness of multivariate methods in the evaluation of field ecological data and laboratory multispecies toxicity tests. In each of the examples, several multivariate techniques were used — generally Euclidean and cosine distances (Figure 11.29), principal components, and nonmetric clustering and association analysis. 

In both studies, nonmetric clustering outperformed the metric tests, although both principal components analysis and correspondence analysis yielded some additional insight into large-scaled patterns, which was not provided by the nonmetric clustering results. However, nonmetric clustering provided information without the use of inappropriate assumptions, data transformations, or other dataset manipulations that usually accompany the use of multivariate metric statistics. The success of these studies and techniques led to the examination of community dynamics in a series of two multispecies toxicity tests. 

As a first test of the use of multivariate analysis in the interpretation of multispecies toxicity tests, the dataset used to analyze the CR microcosm experiment was presented in a blind fashion for analysis. Neither the purpose nor the experimental setup was provided for the analysis. Nonmetric clustering was used to rank variables in terms of contribution and to set clusters. 

The second major application of nonmetric clustering to the analysis of SAM data has been the investigation of the impact of the complex Jet-A (Landis et al. 1993b). The major modification to the SAM protocol was the means of toxicant delivery. Test material was added on day 7 by stirring each microcosm, removing 450 ml from each container, and then adding appropriate amounts of the water soluble fraction (WSF) of Jet-A to produce concentrations of 0,1,5, and 15% WSF. After toxicant addition the final volume was adjusted to 3 1. 

---