Skip to Content

Causal AI & LLM synergies: Enterprise decision making needs more than LLMs

Read more

Causal Effect Estimation

One of the great challenges in causal inference is to estimate the magnitude of the effect that a treatment has on the outcomes. This can be quite challenging, especially in situations in which the data wasn’t generated in a randomized control trial. Despite the usual nomenclature, causal effect estimation isn’t limited to medicine – every single decision made in a business setting is a treatmentwhich hopefully will lead to a positive outcome: from giving discounts to clients to rolling out a new ad campaign. In all of these settings we want to know not only whether the treatment worked, but also how well it worked and which cohort of the population was more susceptible to the treatment. All of these are questions addressed by causal effect estimation.This presentation is a general overview of all concepts related to causal effect estimation, including:

  • Randomized control trials (RCTs) and observational versus interventional studies
  • Important types of effect estimation measures, such as:
  • Average treatment effects (ATE)
  • Conditional average treatment effects (CATE)
  • Heterogeneous treatment effects (HTE)
  • Individual treatment effects (ITE)

Also discussed in this presentation are several estimation methods, such as the back-door criterion, front-door criterion, instrumental variables and propensity score matching.