The choice of the environmental covariate affects the power to detect variation in reaction norm slopes

Many traits are phenotypically plastic, i.e. the same genotype expresses different phenotypes depending on the environment. Individuals and genotypes can vary in this response to the environment and this individual (IxE) and genetic variation in reaction norm slopes (GxE) can have important ecological or evolutionary consequences. Studies on variation in plasticity often fail to show IxE or GxE; this can indicate a genuine absence or simply a lack of power. There is, however, another factor that could potentially affect the power to detect IxE or GxE: the choice of the environmental variable included in the analysis. Identifying the genuine environmental driver of phenotypic plasticity will mostly be impossible and hence only a proxy is included in the analysis. However, if this proxy is too weakly correlated with the real driver of plasticity, this will bias IxE and GxE downwards, could lead to spurious negative results and invalid conclusions. As the mean phenotype in a given environment captures all environmental effects on the phenotype, using it as ‘environment’ in the analysis should increase the power to detect IxE or GxE. Using simulations I here showed that using weakly correlated proxies indeed biased estimates downwards but that when using the environment-specific means this was not the case. Using environment-specific means as a covariate has been commonly used in animal and plant breeding but rarely used outside these fields despite its potential usefulness as ‘yardstick’ to test whether IxE or GxE is present or absent in the phenotype of interest.