py-why/econml

EconML is a Python package from Microsoft Research that helps analysts and researchers answer a specific kind of question: if we change something, what effect does that change have, and does that effect differ for different groups of people? This is the problem of causal inference, and it sits at the heart of decisions like whether a drug works better for certain patients, whether a pricing change grows or shrinks revenue, or whether a policy program actually improves the outcomes it targets. The core challenge EconML addresses is that most real-world data comes from observation rather than controlled experiments. You cannot always run a randomized trial, so you have to use statistical methods to estimate what would have happened under different conditions. EconML combines ideas from economics and statistics with modern machine learning to do this more accurately and at larger scale than older methods allowed. It implements several research-backed techniques, including double machine learning, which was developed to isolate causal signals from noisy datasets with many variables. The package is designed around a common structure: you specify a treatment variable (the thing that was changed or that you want to understand), an outcome variable (what you are measuring), and a set of background features about each observation. EconML then estimates not just an average effect but how that effect varies across different subgroups, which is what the word "heterogeneous" refers to in its full name. It also produces confidence intervals so you can see how certain the estimates are. EconML is installable via pip and works alongside the standard Python data science stack. The project includes Jupyter notebooks with worked examples, and the documentation site covers the main estimation methods, policy learning tools, and guidance on selecting the right approach for a given problem. It is actively maintained and has been receiving regular releases since 2019. The full README is longer than what was shown.