You are here:
SAMPLING EFFORT AFFECTS MULTIVARIATE COMPARISONS OF STREAM COMMUNITIES
Citation:
Cao, Y., D P. Larsen, R M. Hughes, P. L. Angermeier, AND T. M. Patton. SAMPLING EFFORT AFFECTS MULTIVARIATE COMPARISONS OF STREAM COMMUNITIES. JOURNAL OF THE NORTH AMERICAN BENTHOLOGICAL SOCIETY. North American Benthological Society, Lawrence, KS, 21(4):701-714, (2002).
Description:
The estimation of ecological trends and patterns is often dependent on the size of individual samples from each site (sample size) or spatial scale in general. Multivariate analysis is widely used for determining patterns of community structure, inferring species-environment relationships and assessing human impacts on ecosystem conditions. The degree to which sample size affects the outcome of multivariate analysis is therefore a concern. Because calculation of similarity among samples is the enters of multivariate analysis, we examined the effect of sample size on two similarity indices, Jaccard Coefficient (JC) and Bray-Curtis Index (BC), using one benthic and two fish datasets. With increasing sample size, similarity between any two samples in the same dataset always increases, but most rapidly when they are from the same site, followed by from the same region, and then from different regions. Consequently, the degree of site/group separation or classification strength (CS), which depends on relative differences in similarity, increases with sample size. We illustrate this sample-size effect using Non-metric Multidimensional Scaling (NMDS). Sample size also affects NMDS stress level and the lengths of axes. CS and NMDS solutions entirely stabilized only when species accumulation curves leveled at all sites compared. As a completely stable solution is often infeasible, we suggest using the mean within-site similarity across sites to indicate the stability of multivariate comparisons. Coefficient of variation in within-site JC across sites can be used to measure the comparability among samples from different sites. Our study shows that the effect of sample size on multivariate comparisons is significant, scale-dependent, and varies with similarity index.