Factors affecting reproducibility of numerical classifications.

Results obtained with reference organisms which had been included in each of three different investigations were compared. Positions of the reference organisms in diagrams based on highest-link numerical groupings varied with the composition of the microbial samples studied, but significant correlations were found between the results of different investigations. Correlations between the overall groupings obtained in different studies were often better than the correlations between actual similarity values computed between specific pairs of organisms. Similarity values based on different sets of features showed significant correlations, regardless of the kinds of features involved or of the number of features (above a minimum of about 40). Correlations between values based on totally different features were nearly as good as those between values based on similar features which had been determined by different investigators.