Judea Pearl, Professor
University of California, Los Angeles
Computer Science Department
“The Scientific Approach to Causal Inference”
The enourmous progress of causal analysis in the past few decades can safely be attributed to a paradigmatic shift and two fundamental principles. The shift involves a commitment to represent reality in terms of data-generating processes, rather than distributions. The principles are: (1) How counterfactuals are to be computed from a data-generating process, and (2) How dependencies in the observed data can be inferred from the structure of the process.
This talk will describe how these two principles led to remarkably powerful techniques of answering practical, yet non-trivial questions in epidemiological research. These include: policy evaluation, confounding control, transportability, sample selection bias, heterogeneity, mediation, causes-of-effects and missing data.
Background material can be viewed here:
J. Pearl, Causality (Cambridge University Press, 2000)
Causes of Effects