Host range testing methods
There is no perfect number to choose for how many replicates of a test to undertake. Statistical power analyses exist that can more accurately predict the number of replicates that need to be undertaken to generate data with certain probabilities of type I or type II errors. This calculation is based on the variance from the mean obtained in an a priori trial run or the mortality obtained in tests a posteriori (Hoffmeister 2005). It is possible that variance in biological parameters such as survival and development may differ according to the suitability of the host, and if so, different levels of replication may be required for different test plants.
It is important to avoid pseudo-replication. This can come about in host specificity testing if a shortage of test species results in long-lived individuals being utilised repeatedly in different tests. Pseudo-replication can also occur if the statistical testing is undertaken upon individual insects when in fact the data were gathered from insects caged together as a group (Hoffmeister et al. 2006). Hence, the samples are not independent. As a general rule of thumb true replication is achieved at the 'cage' level within which each individual test was run. Hoffmeister et al. (2006) goes as far as to say that it is essential that insects for the tests on non-target species do not come from one rearing container while control individuals come from another. Also it is essential that non-target hosts are not always tested in the same container or field cage or on the same plant, while target hosts are tested in another cage or on another plant. Equally important is that the location of experimental units (cages) within a chamber or laboratory, need to be randomised to avoid confounding effects of differences in temperature and light conditions etc. Similarly, tests with control and test insects should be carried out in parallel and not at different times. Randomisation of testing order or random assignment to plants or test cages assures that pseudo-replication can be avoided (Hoffmeister 2005). When writing applications to the EPA, it is unwise to expect the reviewers to assume that experiments have been carried out according to 'best practice', rather it is preferable to mention in detail that such steps were taken in the methods section of the host range testing reports.
Hoffmeister T.S. (2005). From design to analysis: effective statistical approaches for host range testing Pp. 672-682 In: Second International Symposium on Biological Control of Arthropods, Davos, Switzerland, 12-16 September, 2005, M.S. Hoddle (Ed.) United States Department of Agriculture, Forest Service, Washington.
Hoffmeister T.S., Babendreier D. and Wajnberg E. (2006). Statistical tools to improve the quality of experiments and data analysis for assessing non-target effects. Pp. 222-240 In: Environmental impact of invertebrates for biological control of arthropods: methods and risk assessment, F. Bigler, D. Babendreier and U. Kuhlmann (Ed.) CABI Publishing, Wallingford, Oxford.
Test designs for arthropod biological control