The power of noise and the art of prediction

Data analysis usually aims to identify a particular signal, such as an intervention effect. Conventional analyses often assume a specific data generation process, which implies a theoretical model that best fits the data. Machine learning techniques do not make such an assumption. In fact, they encourage multiple models to compete on the same data. Applying logistic regression and machine learning algorithms to real and simulated datasets with different features of noise and signal, we demonstrate that no single model dominates others under all circumstances. By showing when different models shine or struggle, we argue that it is important to conduct predictive analyses using cross-validation for better evidence that informs decision making.