Strengthening statistical usage in marine ecology: Linear regression

• A non-technical overview of linear regression
• Emphasis on correct procedure and validation
• Examples from marine ecological contexts
• Most serious/common errors summarized

Linear regression is a frequently-used statistical technique in marine ecology, either to model simple relationships or as a component of more complex models. The apparent simplicity of this technique often obscures its far more complex underpinnings, upon which its validity, and ultimate ecological interpretations, wholly depend. We present a non-technical review of the foundations of linear regression and its application in marine ecology, with emphasis on correct model specification, the different concepts of linearity, the issues surrounding data transformation, the assumptions which must be respected, and validation of the regression model. The necessity of reporting the results of regression diagnostics is stressed; contrary to widespread practice in marine ecology, R2 and p-values alone do not provide sufficient evidence to form conclusions.