29 Appendix 2, Regulation (EU) 2017/735

In forcesince 14 Feb 2017

In this design, a minimum of 5 males and 5 females are tested at each concentration level resulting in a design using a minimum of 40 animals (20 males and 20 females, plus relevant positive controls.)

The design, which is one of the simpler factorial designs, is equivalent to a two-way analysis of variance with sex and concentration level as the main effects. The data can be analysed using many standard statistical software packages such as SPSS, SAS, STATA, Genstat as well as using R.

The analysis partitions the variability in the dataset into that between the sexes, between the concentrations and that related to the interaction between the sexes and the concentrations. Each of the terms is tested against an estimate of the variability between the replicate animals within the groups of animals of the same sex given the same concentration. Full details of the underlying methodology are available in many standard statistical textbooks (see references) and in the 'help' facilities provided with statistical packages.

The analysis proceeds by inspecting the sex x concentration interaction term in the ANOVA table [Statisticians who take a modelling approach such as using General Linear Models (GLMs) may approach the analysis in a different but comparable way but will not necessarily derive the traditional ANOVA table, which dates back to algorithmic approaches to calculating the statistics developed in a pre-computer age.]. In the absence of a significant interaction term the combined values across sexes or across concentration levels provide valid statistical tests between the levels based upon the pooled within group variability term of the ANOVA.

The analysis continues by partitioning the estimate of the between concentrations variability into contrasts which provide for a test for linear and quadratic contrasts of the responses across the concentration levels. When there is a significant sex x concentration interaction this term can also be partitioned into linear x sex and quadratic x sex interaction contrasts. These terms provide tests of whether the concentration responses are parallel for the two sexes or whether there is a differential response between the two sexes.

The estimate of the pooled within group variability can be used to provide pair-wise tests of the difference between means. These comparisons could be made between the means for the two sexes and between the means for the different concentration level such as for comparisons with the negative control levels. In those cases where there is a significant interaction comparisons can be made between the means of different concentrations within a sex or between the means of the sexes at the same concentration.

There are many statistical textbooks which discuss the theory, design, methodology, analysis and interpretation of factorial designs ranging from the simplest two factor analyses to the more complex forms used in Design of Experiment methodology. The following is a non-exhaustive list. Some books provide worked examples of comparable designs, in some cases with code for running the analyses using various software packages.

(1)

Box, G.E.P, Hunter, W.G. and Hunter, J.S. (1978). Statistics for Experimenters. An Introduction to Design, Data Analysis, and Model Building. New York: John Wiley & Sons.

(2)

Box G.E.P. & Draper, N.R. (1987) Empirical model-building and response surfaces. John Wiley & Sons Inc.

(3)

Doncaster, C.P. & Davey, A.J.H. (2007) Analysis of Variance and Covariance: How to choose and Construct Models for the Life Sciences. Cambridge University Press.

(4)

Mead, R. (1990) The Design of Experiments. Statistical principles for practical application. Cambridge University Press.

(5)

Montgomery D.C. (1997) Design and Analysis of Experiments. John Wiley & Sons Inc.

(6)

Winer, B.J. (1971) Statistical Principles in Experimental Design. McGraw Hill.

(7)

Wu, C.F.J & Hamada, M.S. (2009) Experiments: Planning, Analysis and Optimization. John Wiley & Sons Inc.