This paper discusses the use of nonparametric preprocessing, specifically matching, to reduce model dependence in parametric causal inference. The authors argue that while parametric models are commonly used in political science research, the results presented are often based on a small number of model specifications, leading to potential bias and model dependence. Matching methods, which make fewer assumptions, can help address this issue by preprocessing data to make the treatment group more similar to the control group. This reduces the dependence on modeling choices and improves the accuracy of causal inferences. The paper explains how matching can be used to preprocess data before applying parametric methods, allowing researchers to use their familiar parametric techniques more effectively. By making the treatment variable independent of the background covariates, matching reduces the impact of model specification on the results. This approach has three key advantages: it is easy to use, it reduces model dependence, and it minimizes bias compared to raw data analysis. The authors also discuss the challenges of causal inference in observational data, where model dependence is a significant issue. They highlight the importance of avoiding selection bias and controlling for confounding variables. They propose a unified approach that combines nonparametric preprocessing with parametric analysis to improve causal inference. The paper defines causal effects and discusses the differences between fixed and random causal effects. It also addresses the issue of nonbinary and multiple treatments, emphasizing the importance of careful modeling and the need for assumptions about the data. The authors conclude that matching can be a valuable tool for reducing model dependence in parametric causal inference, especially when combined with parametric analysis. They argue that matching should be used as a preprocessing step to improve the accuracy and reliability of causal estimates. The paper also highlights the importance of understanding the assumptions underlying causal inference and the need for careful modeling to avoid bias.This paper discusses the use of nonparametric preprocessing, specifically matching, to reduce model dependence in parametric causal inference. The authors argue that while parametric models are commonly used in political science research, the results presented are often based on a small number of model specifications, leading to potential bias and model dependence. Matching methods, which make fewer assumptions, can help address this issue by preprocessing data to make the treatment group more similar to the control group. This reduces the dependence on modeling choices and improves the accuracy of causal inferences. The paper explains how matching can be used to preprocess data before applying parametric methods, allowing researchers to use their familiar parametric techniques more effectively. By making the treatment variable independent of the background covariates, matching reduces the impact of model specification on the results. This approach has three key advantages: it is easy to use, it reduces model dependence, and it minimizes bias compared to raw data analysis. The authors also discuss the challenges of causal inference in observational data, where model dependence is a significant issue. They highlight the importance of avoiding selection bias and controlling for confounding variables. They propose a unified approach that combines nonparametric preprocessing with parametric analysis to improve causal inference. The paper defines causal effects and discusses the differences between fixed and random causal effects. It also addresses the issue of nonbinary and multiple treatments, emphasizing the importance of careful modeling and the need for assumptions about the data. The authors conclude that matching can be a valuable tool for reducing model dependence in parametric causal inference, especially when combined with parametric analysis. They argue that matching should be used as a preprocessing step to improve the accuracy and reliability of causal estimates. The paper also highlights the importance of understanding the assumptions underlying causal inference and the need for careful modeling to avoid bias.