Position paper: Common mistakes and solutions for a better use of correlation- and regression-based approaches in environmental sciences

While empirical modelling remains a popular practice in environmental sciences, an alarming number of misuses of correlation- and regression-based techniques are encountered in recent research, although these techniques are described in courses and textbooks. This position paper reviews the most common issues, and provides theoretical background for understanding the interests and limitations of these methods, based on their underlying assumptions. We call for a reconsideration of misleading practices, including: the application of linear regression to data points that do not display a linear pattern, the failure to pinpoint influential points, the inappropriate extrapolation of empirical relationships, the overrated search for “statistical significance”, the pooling of data belonging to different populations, and, most importantly, calculations without data visualization. We urge reviewers to be vigilant on these aspects. We also recall the existence of alternative approaches to overcome the highlighted shortcomings, and thus contribute to a more accurate interpretation of the results. • Misapplications of bivariate analysis are frequently observed in environmental science papers. • Common practices actually rely on relatively strict assumptions to be accurate. • Their incautious use can lead to the fallacious identification of associations between variables. • In each situation, alternative approaches are proposed to overcome the encountered limitations. • Visual evidence should be given more weight versus automatic statistical procedures.