{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# Chapter 28: Matching, Propensity Scores, Weighting, and Doubly Robust Estimation\n",
        "**Economic question:** How can observational studies improve comparability when treatment is not randomized?\n",
        "\nPropensity methods rebalance observed covariates; they do not eliminate bias from unobserved confounding.\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "from sklearn.linear_model import LogisticRegression\n",
        "rng=np.random.default_rng(28); n=1500; x=rng.normal(size=(n,3)); ps=1/(1+np.exp(-(x[:,0]-.5*x[:,1]))); d=rng.binomial(1,ps); y=2*d+x[:,0]+rng.normal(size=n)\n",
        "logit=LogisticRegression().fit(x,d); phat=logit.predict_proba(x)[:,1]\n",
        "w=d/phat+(1-d)/(1-phat)\n",
        "print('IPW ATE',np.average(d*y/phat)-np.average((1-d)*y/(1-phat)))\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Interpretation checklist\n",
        "- State the unit of observation and units of every variable.\n",
        "- Separate association, prediction, and causation.\n",
        "- Report magnitude and uncertainty.\n",
        "- Identify the most important threat to validity.\n"
      ]
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "name": "python",
      "version": "3.13"
    },
    "ceteris_lab": {
      "course_slug": "fundamentals-python-econometrics",
      "course_title": "Fundamentals of Python for Financial Econometrics",
      "chapter": 28,
      "source_origin": "supplied"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 5
}