Causal inference is a field of study which aims to elucidate true directed relationships between variables in order to understand cause and effect. We can argue that causality is innate since a toddler would be quick to learn the effect of holding down the power button of a computer running. However, many cause-to-effect relationships can be masked or disturbed by more complex mechanisms. Examples of concrete cause-to-effect questions include: Does advertising increase product sales? How does education affect income? And do dietary carotenoids reduce the risk of lung cancer? Unfortunately, extracting causal evidence from collected data to assess cause-to-effect relationships is often very challenging and usually requires careful thought investigations. While data from controlled experiments are key to answer causal questions, these experiments are often imperfect, costly, unethical , or simply infeasible.
In many cases, available data come from observational studies which do not readily address the causal questions of interest. Classical statistical analysis can identify strong associations between a modifiable exposure and an outcome from observational data, but the identified relationships may simply be a result of confounding factors or reverse causality problems, common in observational studies, instead of causal effects. Answering causal questions requires tools that enable understanding how the distribution of response variables change under changing conditions or interventions. Analysis approaches to causality draw from counterfactual concepts, structural equations and graphical models. Valuable methods identified to explore the paths from cause to effect are random assignment, regression, instrumental variables, regression discontinuity, and differences in differences. Many applications and extensions of these methods have been proposed along with novel methodologies to identify and estimate causality.