Persisting challenges in multiple models: a note on commonly unnoticed issues regarding collinearity and spatial structure of ecological data

There is a growing need to heed some caveats in numerical data analysis. In 2013, I set out some issues regarding multiple regression frameworks. Here, I used both hypothetical and real data sets collected in Brazil to discuss the implications of, and provide suggestions for, some statistical issues regarding to collinearity and spatial structure of ecological data. For example, a weak treatment of collinearities might lead to discarding important variables for the model, and this can be avoided by a correct approach to collinearities before the model selection. Moreover, studies have demonstrated that the spatial structure in both predictor and response variables is an important point to be addressed, rather than the presence of this structure only in the residuals. Aiming to facilitate the controlling of such bias, I provide two fully explained scripts for R language. Considering the seriousness of spatial structure, my opinion is that no article that presents confirmatory analysis should be considered for publication if their authors do not heed that caveat; facing this issue, I strongly suggest that one performs a variance partitioning scheme.