Causal Inference and Counterfactual Reasoning

Causal Inference and Counterfactual Reasoning




Identifying causal effects is an integral part of scientific inquiry, spanning a wide range of questions such as understanding behavior in online systems, effect of social policies, or risk factors for diseases. As computer systems increasingly impact all walks of life, questions of cause-and-effect also become important for designing and evaluating future systems. For instance, how do algorithmic recommendations affect our purchasing decisions? How do they affect a student’s learning outcome or a doctor’s efficacy? These are hard questions and require thinking about the counterfactual: what would have happened in a world with a different system, policy or intervention? Without randomized experiments and causal reasoning, correlation-based methods can lead us astray.

At Microsoft Research, we focus on answering causal questions from data. We draw on a rich literature from statistics, machine learning, epidemiology and the social sciences to rethink core methods for causal inference. Some of the topics we work on, from core inference and discovery algorithms to applications of causal inference include:

  • Finding natural experiments from large-scale data
  • Scaling and adapting causal inference methods to work on non-traditional data sources, like text and images
  • Predicting life event outcomes
  • Evaluating effects in online systems
  • Harvesting randomness for distributed systems optimization
  • Prediction and explanation in social systems
  • Explaining and interpreting machine learning