A multivariate method for determining optimal subsample size in the analysis of macroinvertebrate samples

Abstract

The depauperate macroinvertebrate communities of two disturbed urban streams were used to demonstrate a technique for determining optimal subsample size for the multivariate representation of community structure. Although multivariate analyses have commonly been applied to subsampled data, the effect of subsampling on multivariate patterns has not previously been investigated in detail. The minimum requirement for applying the method is eight fully processed community samples distributed across two treatments, with n = 4 (being a representative subset of a larger study to which subsampling will be applied). The data from each sample are subsampled repeatedly with the aid of a simulation program. Variation in community structure between treatments is tested by analysis of similarity for the complete data and for multiple sets of subsampled data. The optimal subsampling strategy is defined as the minimum effort required to achieve a median R-statistic no less than the value derived for the complete data. Subsampling to a fixed proportion is a less efficient strategy than subsampling to a fixed number of individuals. For the urban stream communities, subsampling to 300 individuals was adequate for four out of six comparisons and was adequate for all comparisons when a minimum proportion of 10% was applied. Several non-abundant taxa were found to be important contributors to differences between treatments. Subsamples need to be large enough to adequately estimate the abundance of such taxa. Further investigation is required to assess the relationship between optimal subsample size for multivariate analysis and the shape of cumulative species-abundance curves.