Power analysis for beach surveys of marine debris

Abstract

In the US, a monitoring programme is being planned to evaluate national trends in beach litter. Power analysis was used to determine if the programme had a high probability of detecting a specified effect. We compared the use of a repeated measures model and a one-way analysis of variance model to investigate the power of detecting a 20% linear decrease in litter on beaches over a 5-year period, with power of 0.84 or more, a Type I error rate of 0.05, and quarterly sampling. We used the average coefficient of variation and, for the repeated measures model, average autocorrelations as estimates of model parameters. Common debris items typically had positive autocorrelations and use of the repeated measures model produced sample size estimates smaller than those from the analysis of variance model. Sample size estimates critically depend on reliable estimates of the mean, variance, and covariance of debris items of interest.