The article by Gary King and Richard Nielsen argues that propensity score matching (PSM), a widely used method in causal inference, often fails to achieve its intended goal of reducing imbalance between treated and control groups. Instead, PSM can increase imbalance, model dependence, and bias. The authors show that PSM approximates a completely randomized experiment, which is less efficient than fully blocked randomized experiments used by other matching methods. This makes PSM blind to important information in observational data, leading to worse performance in reducing imbalance and model dependence. The authors also demonstrate that when data are already balanced, PSM can actually increase imbalance, as it approximates random matching. This phenomenon is referred to as the "PSM paradox." The article highlights that while PSM has other useful applications, such as regression adjustment and inverse weighting, it is not the best method for reducing model dependence and bias in causal inference. The authors recommend using other matching methods that are more effective at reducing imbalance and model dependence. They also discuss the limitations of PSM theory and the dangers of random matching, which can increase imbalance and bias. The article concludes that while PSM has its place in causal inference, it is not the optimal method for reducing model dependence and bias, and other matching methods should be preferred when possible.The article by Gary King and Richard Nielsen argues that propensity score matching (PSM), a widely used method in causal inference, often fails to achieve its intended goal of reducing imbalance between treated and control groups. Instead, PSM can increase imbalance, model dependence, and bias. The authors show that PSM approximates a completely randomized experiment, which is less efficient than fully blocked randomized experiments used by other matching methods. This makes PSM blind to important information in observational data, leading to worse performance in reducing imbalance and model dependence. The authors also demonstrate that when data are already balanced, PSM can actually increase imbalance, as it approximates random matching. This phenomenon is referred to as the "PSM paradox." The article highlights that while PSM has other useful applications, such as regression adjustment and inverse weighting, it is not the best method for reducing model dependence and bias in causal inference. The authors recommend using other matching methods that are more effective at reducing imbalance and model dependence. They also discuss the limitations of PSM theory and the dangers of random matching, which can increase imbalance and bias. The article concludes that while PSM has its place in causal inference, it is not the optimal method for reducing model dependence and bias, and other matching methods should be preferred when possible.