Ecologists use statistics for multiple purposes, but the amount of variance explained is a common denominator in assessing validity and importance of findings. Various authors have specifically pointed to the amount of variance explained as the true gauge of whether a branch of science is really onto something. Although that seems like an objective way to compare the performance or validity of different areas of science, multiple sources contribute to variance in data, and some of these sources are separate from the underlying scientific explanatory process. Measurement error and underlying noise both reduce the reliability of measurements, thereby reducing the maximum achievable R2. This is a particular problem as researchers are tending to examine patterns at small temporal and spatial scales, and as statistical pooling is used less and less. The R2 achieved in a study of a response variable that is measured reliably and aggregated over space and time is likely to be much greater than that of a study examining variables that are measured with substantial error in small research units (e.g., small plots, individuals, etc.)
Results/Conclusions
Disattenuation is the statistical process of measuring and accounting for reliability, allowing the researcher to estimate the fraction of explicable variance that is explained by the hypothesis or model. It is commonly used in some fields (psychometrics especially) but appears to have been largely ignored by ecologists. In some cases, reliability is restricted theoretically by the sampling process (e.g., binomial), so that theory can be used to gauge and estimate reliability. Other sources of noise contaminating measurements further complicate the process, but intentional study design or randomization procedures can provide empirical estimates of reliability. Here, I apply these procedures in real-world case studies of space use of foraging mice, and local abundance of foraging waterfowl, and show how disattenuation can be accomplished so as to enable "apples-to-apples" comparisons of explanatory and predictive power.