The paper discusses the issue of model dependence in parametric causal inference, where the choice of control variables, functional forms, and other modeling assumptions can significantly affect the estimates. To address this, the authors propose using matching methods as a nonparametric preprocessing step to reduce model dependence. Matching methods aim to make the treated group similar to the control group, thereby reducing the relationship between the treatment variable and control variables. This preprocessing step allows researchers to use familiar parametric analysis methods without being overly influenced by specific model choices. The authors provide a unified approach and easy-to-use software (MatchIt) to implement these methods. They also define causal effects, assumptions for causal inference, and the challenges in parametric analysis methods for both experimental and observational data. The paper emphasizes the importance of avoiding selection bias and the curse of dimensionality, which can lead to model dependence. By preprocessing data with matching, researchers can achieve more accurate and less model-dependent causal inferences.The paper discusses the issue of model dependence in parametric causal inference, where the choice of control variables, functional forms, and other modeling assumptions can significantly affect the estimates. To address this, the authors propose using matching methods as a nonparametric preprocessing step to reduce model dependence. Matching methods aim to make the treated group similar to the control group, thereby reducing the relationship between the treatment variable and control variables. This preprocessing step allows researchers to use familiar parametric analysis methods without being overly influenced by specific model choices. The authors provide a unified approach and easy-to-use software (MatchIt) to implement these methods. They also define causal effects, assumptions for causal inference, and the challenges in parametric analysis methods for both experimental and observational data. The paper emphasizes the importance of avoiding selection bias and the curse of dimensionality, which can lead to model dependence. By preprocessing data with matching, researchers can achieve more accurate and less model-dependent causal inferences.