Model ecosystem replicability challenged by the "soft" reality of a hard bottom mesocosm

The concept of model ecosystem replicability is challenged by a case study of a land-based brackish-water mesocosm. Coefficients of variation (CVs) are calculated for a total of 50 population and community variables between unpolluted control systems, during five separate test years. These CVs are used for calculations of minimum detectable differences (MDDs) for a given number of treatments and replicates at specific levels of statistical significance and power, when analysis of variance (ANOVA) techniques are used. The CVs are also used for calculation of the minimum number of replicates (MNRs) theoretically needed to confirm that a specific deviation from the control is statistically significant. While variables with CVs lower than 20-30% have been regarded both as normal and manageable in mesocosm studies (in economic and practical terms), the mean CV values calculated in this study are generally higher. Exactly half of the studied variables have CVs higher than 60%. Therefore, there will be obvious difficulties in finding statistically significant deviations from the controls even if the test systems are rigorously replicated, the investigated effect sizes are large and the postulated acceptable magnitudes of statistical error are high. (C) 1998 Elsevier Science B.V.