Gary King, "Why Propensity Scores Should Not Be Used for Matching"
Why Cross Validated Stack Exchange recommends it
Cross Validated users link it when asking whether propensity score matching is worse than other matching methods, citing King's critique of it for causal inference.
Recommended by 1 people across 1 different months, 2020 to 2021.
What it is
A one-hour talk by Harvard political scientist Gary King, given at the International Methods Colloquium in 2015, arguing that propensity scores can increase imbalance and model dependence when used for matching. It accompanies his paper on the same topic and surveys alternative matching methods.
What people said
Recommends John Kazowski · 2021-08 · Cross Validated Stack Exchange · on “Matching with a continuous treatment”
I am trying to match a ~60k observation data set, where the treatment (annual days of sun exposure) is continuous. The data also has several confounders, some of which are continuous and some categorical. I was trying to use the MatchIt R package to get covariate balancing propensity scores (CBPS) but got an error message saying "The treatment must be a binary variable." Using the CBPS package directly did work and the cobalt package indicates that matching indeed yielded balance. But now I am unsure how to actually match the data. library(CBPS) cbps_out <- CBPS(days_of_sun ~ gender + race + age + state + history_of_cancer, method = "exact", data = df) I was looking everywhere online, but th
Cites Shade · 2020-10 · Cross Validated Stack Exchange · on “Propensity score matching vs non-parametric regression”
I am trying to understand the benefit of propensity matching over non-parametric regression for causal inference from non-experimental data. As background: the way I understand it, parametric regressions are generally a poor choice for causal inference when selection bias exists. One can try to create a model that takes into account the different baselines between treatment and control groups, but she/he will be be extremely vulnerable to model misspecification.* Matching solves this issue by making the two datasets look "alike." The exact meaning of alike depends on the matching algorithm used, but all the algorithms strive to eradicate model dependence, with the most common types being str
Cites Pik-Mai Hui · 2021-12 · Cross Validated Stack Exchange · on “Is propensity score matching worse than other matching methods?”
Gary King explained why he thought propensity score matching should not be used for matching. See Paper [LINK] and the Video Lecture [this video] . After all these years, have the academics/practitioners reached consensus on the following question: should propensity score matching be avoided in favor of other matching methods? Is there any counter-arguments given by other researchers to Gary King's explanation?
Cites Pik-Mai Hui · 2021-12 · Cross Validated Stack Exchange · on “Metric for unsupervised recommender-system competition?”
Out of the three options, let's first talk about option 2 and option 3 first, then option 1. Option 2: Panel expert judgement No. Never trust subject-matter experts on their judgement about anything quantitative. This applies to even ML experts on their topic-of-interest ML models' performance metrics. In fact, the more the experts know about the subject, the more biased they are about that subject. In addition, experts tend to be over-confident on their ability to avoid biases, which makes it even worse. That's why almost all companies insist on A/B test before fully deploying models. Listen to how Gary King explains a version of it: [this video] Option 3: Accept that it's unsupervised and look