Sample size estimation for savanna monitoring protocol development

Abstract

When designing data collection protocols for a new research project, it is important to have a large enough sample size to detect a desired effect, but not so large to be wasting time collecting more data than needed. Power analysis methods can be used to estimate this sample size. In this report, power analyses used to estimate sample sizes needed for a savanna monitoring study, for which the U.S. Fish and Wildlife Service are developing protocols, are described. Power analyses were run to estimate the sample sizes needed to detect a specified difference (that is, effect size) between means from two savanna areas or between yearly means for a savanna area. Sample sizes were estimated for nine different vegetation metrics that will be measured in savanna areas. Analyses were run for each metric using a range of means and variances, effect sizes, and correlation among repeated measures. Sample size estimates varied among vegetation metrics. Within each vegetation metric, estimated sample sizes varied with means, variances, effect size, and correlation. Many of the sample size estimates were too large to be feasible when sampling; therefore, the tables of estimated sample sizes may be first used as a guide to determine an adequate and feasible sample size that will detect differences in some vegetation metrics. Then, using this sample size, the tables can be used to estimate the effect sizes for each vegetation metric that may be detectable for a given mean, variance, and correlation.