Understanding Instrumental Variables

<p>A/B tests are the golden standard of causal inference because they allow us to make valid causal statements under minimal assumptions, thanks to&nbsp;<strong>randomization</strong>. In fact, by randomly assigning a&nbsp;<strong>treatment</strong>&nbsp;(a drug, ad, product, &hellip;), we are able to compare the&nbsp;<strong>outcome</strong>&nbsp;of interest (a disease, firm revenue, customer satisfaction, &hellip;) across&nbsp;<strong>subjects</strong>&nbsp;(patients, users, customers, &hellip;) and attribute the average difference in outcomes to the causal effect of the treatment.</p> <p>However, in many settings, it is&nbsp;<strong>not possible to randomize</strong>&nbsp;the treatment, for either ethical, legal, or practical reasons. One common online setting is on-demand features, such as subscriptions or premium memberships. Other settings include features for which we cannot discriminate customers, like insurance contracts, or features that are so deeply hard-coded that an experiment might not be worth the effort. Can we still do valid causal inference in those settings?</p> <p><a href="https://towardsdatascience.com/understanding-instrumental-variables-0ce5d3d6ba20"><strong>Read More</strong></a></p>