You are here:
BOOTSTRAPPING AND MONTE CARLO METHODS OF POWER ANALYSIS USED TO ESTABLISH CONDITION CATEGORIES FOR BIOTIC INDICES
Citation:
Blocksom, K A., F A. Fulk, D J. Klemm, AND S M. Cormier. BOOTSTRAPPING AND MONTE CARLO METHODS OF POWER ANALYSIS USED TO ESTABLISH CONDITION CATEGORIES FOR BIOTIC INDICES. Presented at Society of Environmental Toxicology and Chemistry, Nashville, TN, November 12-16, 2000.
Impact/Purpose:
The goal of this research is to develop methods and indicators that are useful for evaluating the condition of aquatic communities, for assessing the restoration of aquatic communities in response to mitigation and best management practices, and for determining the exposure of aquatic communities to different classes of stressors (i.e., pesticides, sedimentation, habitat alteration).
Description:
Biotic indices have been used ot assess biological condition by dividing index scores into condition categories. Historically the number of categories has been based on professional judgement. Alternatively, statistical methods such as power analysis can be used to determine the number of categories distinguishable by a biotic index. Lacking the replication necessary to conduct traditional power analysis we used bootstrapping and Monte Carlo methods to conduct power analysis on two different biotic indices to establish condition categories. For the Stream Benthos Integrity Index (SBII) developed for wadeable streams in the Mid-Atlantic highlands we used several pairs of within year revisits to generate sets of bootstrap replicates. Using these samples we tested for differences between each possible pair of sites using t-tests. Repeating this process 500 times we determined the effect size (mean difference in scores) at which the power (proportion of tests for which the null hypothesis was rejected) was equal to 0.80. For the Lake Macroinvertebrate Integrity Index (LMII) developed for New Jersey lakes and reservoirs, we had replicate data for only 2 of 58 lakes. Thus we estimated sample variance from these replicates and used the maximum of the two variance values in our analysis. To generate samples for t-tests we used Monte Carlo methods to generate normal random variates based on a mean equal to the actual LMII score and variance equal to the maximum variance mentioned above. In each case the effect size was used to determine statistically the number of distinct condition categories for the index by dividing the maximum index score by the effect size.