Bootstrapping to investigate the effect of number of macroinvertebrate samples on confidence limits of the mean

Abstract

A resampling technique (bootstrapping) was used to evaluate the effects of increasing sample number on the confidence limits of mean density and the mean number of taxa (Families) in addition to the performance of four different sampling devices used to collect macroinvertebrates in three streams of contrasting anthropogenic impact. The four sampling devices were a Hess sampler, a modified Hess sampler equipped with an internal water pump, a Surber sampler, and a two-pole kick-net. In general, confidence limits decreased predictably as the number of samples increased, most notably with the addition of a third sample. The density of organisms captured with the Surber sampler was significantly lower than that of the Hess sampler but greater than that of the kick-net sampler. No significant difference in the number of Families was found among sampling devices. Based upon a minimal 10% reduction in confidence limits compared to the previous sample, a minimum of four replicates was suggested when using the Surber sampler. In general, other methods required a greater number of replicates that the Surber to adequately represent density, but this did not hold when applied to the number of taxa. Despite the inconvenience of having to bend down to use the Surber sampler, this study found no evidence to reject its apparent popularity as a sampling method for macroinvertebrates.