Learning causal models from correlations: learning the science of control from observations

Most people think they perfectly understand causality and, if asked, probably would say that scientific explanation is causal. Psychologically this is understandable, because we leave in a world full of instances of causality. From cars to planes, from home appliances to gardening equipment we are surrounded by artefacts that can be controlled.

But reality is much more complicated; people with decision making responsibility should learn to distinguish sharply what can be observed from what can be controlled. In other words, they should separate processes that can be only observed from causal processes. Causation means that we can intervene and control the course of events. Basic science such as dynamics or electromagnetism is observational not causal. And basic scientific explanation is deductive not causal.

However, there are systems that are designed to be causal. Most human artefacts are designed to be causal artefacts. However, there are systems such as economies, social systems, firms and their markets whose behaviour is not known with precision. We know approximately some observational laws, for instance we know correlations, but decision making needs to understand causation.

A number of tools for learning causation from observational correlations have been discovered. These tools are very useful for decision making in business. I am offering an executive level course on how to deploy causal models. Go to my site www.sergiofocardi.net to learn more on causal models and, if of interest, to participate in my course.