Optimising the capacity of field trials to detect the effect of genetically modified maize on non-target organisms through longitudinal sampling

To assess risks of cultivation of genetically modified crops (GMCs) on non-target arthropods (NTAs), field tests are necessary to verify laboratory results and in situations where exposure pathways are very complex and cannot be reproduced in the laboratory. A central concern in the design of field trials for this purpose is whether the tests are capable of detecting differences in the abundance or activity of NTAs in a treated crop in comparison with a non-treated comparator plot. The detection capacity of a trial depends on the abundance and variability of the taxon, the values assumed for type I (alpha) and II (beta) errors, and the characteristics of the trial and statistical design. To determine the optimal trial layout and statistical analysis, 20 field trials carried out in Spain from 2000 to 2009 to assess risks of GMCs on NTAs were examined with alpha and beta set at 0.05 and 0.20, respectively. In this article we aim to determine the optimal number of sampling dates during a season, or longitudinal samples, in the design of field trials for assessing effects of GM maize on NTAs, and the ones that contribute most to achieving detectable treatment effects (d(c)) less than 50% of the mean of the control. Detection capacities are a function of the number of individual samples taken during the season but a high number of samples is rarely justified because gains of repeated sampling can be relatively low. These gains depend primarily on field tests relative experimental variability in individual samplings (i.e. experimental variability relative to the mean of the control in each sampling date) which in turn depends on the sampling method (visual counts, pitfall traps or yellow sticky traps) and the density (or abundance) of the taxon in question. Taxa showing more density (or abundance) have less relative experimental variability. The smaller the experimental variability, the lower the profit of increasing the number of sampling dates. Sticky traps have a good effect detection capacity and need very few sampling dates, whereas visual counts and pitfall traps have a poorer effect detection capacity and need more individual samples to achieve d(c) values lower than 50%. In maize field trials, it is recommended to concentrate sampling efforts in certain growth stages; the optimal ones for achieving an acceptable detection capacity are variable but, in general, samples in the first half of the season render better detection capacity than samples in the second half.