Patterns of ordination and classification instability resulting from changes in input data order

Random rearrangement of entry order in three data sets often changed ordination and classification results based on Reciprocal Averaging. Results varied with the data set and method used. Eliminating infrequently occurring species largely reduced, but did not always eliminate, the variability. Overall, results appeared related to data set complexity, the type of data or transformation, and the analysis method used. Detrended Correspondence Analysis had the greatest variability of the ordination methods tested. Results from quantitative data were usually more variable than presence/absence data. Variation in cluster analysis was related to the number of tie values in the similarity matrix. Detailed tests using randomization of entry order of individual data sets with each of the programs to be used are needed to individually assess the effects on the results.