Design of experiments is a cornerstone of the practice of statistics, with applications in virtually all areas of research. The goal is to design an experiment in order to confirm or reject a hypothesis. Data scientists are faced with the need to conduct continual experiments, particularly regarding user interface and product marketing. This chapter reviews traditional experimental design and discusses some common challenges in data science. It also covers some oft-cited concepts in statistical inference and explains their meaning and relevance (or lack of relevance) to data science.
Whenever you see references to statistical significance, t-tests, or p-values, it is typically in the context of the classical statistical inference “pipeline” (see Figure 3-1). This process starts with a hypothesis (“drug A is better than the existing standard drug,” “price A is more profitable than the existing price B”). An experiment (it might be an A/B test) is designed to test the hypothesis—designed in such a way that, hopefully, will deliver conclusive results. The data is collected and analyzed, and then a conclusion is drawn. The term inference reflects the intention to apply the experiment results, which involve a limited set of data, to a larger process or population.