Explaining and Predicting Outcomes with Linear Regression using Python or R

The first two sessions of this module introduce the conceptual framework of the method using the simple case of a single predictor. Formulas and technicalities are kept to a minimum. The main focus is on interpretation of results and assessing model validity. Our reported results will include precision statements on expected outcomes and explanatory effects (hypothesis tests and confidence intervals). We further use the regression model to predict future results. For this, we provide prediction intervals and verify model performance on independent, left out datasets.

In session 3 and 4 we allow for more than one predictor leading to the multiple linear regression model. We focus on either explanation or prediction. How to come to a parsimonious but possibly flexible model starting from a large number of predictors will be discussed in detail. In these complex linear models, special attention will be given to interpreting individual predictor effects, as they critically depend on other terms in the model and underlying relations between predictors (confounding and effect modification).

In the last session a more elaborate data analysis is discussed. We touch on problems where linear regression is too restricted and replaced by related approaches such as generalized linear models and mixed models.

Different features will be illustrated with case examples from the instructors’ practical experience, and participants are encouraged to bring examples from their own work.

Hands-on exercises are worked out behind the PC in two parallel groups using Python or the R software.