DR (#15)

Author: matheusfacure

causal-inference-for-the-brave-and-true/12-Doubly-Robust-Estimation.ipynb

@@ -550,7 +550,8 @@
    "source": [
     "I hope I've convinced you about the power of doubly robust estimation. Its magic happens because in causal inference, there are two ways to remove bias from our causal estimates: you either model the treatment mechanism or the outcome mechanism. If either of these models are correct, you are good to go\n",
     "\n",
-    "However, in practice, it's very hard to model precisely either of those. What ends up happening is that neither the propensity score nor the outcome model are 100% correct. They are both wrong, but in different ways. Still, doubly robust estimation can combine those two wrong models to make them less wrong. It's like one model corrects the error of the other, providing an end result that is better than each one individually. \n",
+    "One caveat is that, in practice, it's very hard to model precisely either of those. More often, what ends up happening is that neither the propensity score nor the outcome model are 100% correct. They are both wrong, but in different ways. When this happens, it is not exactly settled [\\[1\\]](https://www.stat.cmu.edu/~ryantibs/journalclub/kang_2007.pdf) [\\[2\\]](https://arxiv.org/pdf/0804.2969.pdf) [\\[3\\]](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2798744/) if it's better to use a single model or doubly robust estimation. As for me, I still like using them because at least it gives me two possibilities of being correct. \n",
+    "\n",
     "\n",
     "## Keys Ideias\n",
     "\n",
@@ -598,7 +599,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.6.9"
+   "version": "3.8.5"
   }
  },
  "nbformat": 4,