| Ground water quality records have been collected with regularity for a relatively short time. Because of legal requirements for monitoring and liability constraints, information regarding changes in chemical constituents must be acquired very quickly. Large variations in data measurements can obscure small changes.;The study examined generated data sets with characteristics determined from a survey of the literature to be common in ground water quality. The characteristics were (1) seasonality, an annual cycle, (2) trend, both linear and step trends, (3) error distribution, normal or log normal, and (4) serial correlation, limited here to AR(1).;Results showed that the presence of seasonality reduces the power of trend detection, and increasing magnitudes of seasonality increase the reduction in power. Removal of levels equal to that when no seasonality was originally present.;Linear trends can be detected by Student's t-test on the slope estimate of a least squares regression at similar levels for both normal and log normal error distributions. Step trends are difficult to detect when the increase is near the level of process variance. Student's t-test was superior to the distribution-free Mann Whitney for samples with log normal errors and less than five observations, but the superiority was reversed for larger samples.;The present study was undertaken to examine the interaction of statistical characteristics that are often observed in ground water. Knowledge of the interaction will enable a more informed choice of statistical tests so that changes may be detected more accurately.;Serial correlation is difficult to detect and to estimate. It adversely affects the power and significance level of statistics to detect trend and seasonality. Initial high frequency sampling can improve estimates of the level of serial correlation. |
