Regression discontinuity methods and its application and implementation

Objective To provide a complete and systematic introduction to the principle and statistical analysis method of regression discontinuity design (RDD), and to provide a reference for its application in the field of public health.

Methods Based on the introduction of RDD principle, the statistical analysis of RDD by Stata software was demonstrated with an example of formaldehyde intervention in indoor air. The analytical method and result of polynomial regression (parametric method) and local linear regression (nonparametric method) were introduced and compared.

Results RDD is a quasi-experimental design without randomized grouping of the study subjects. Instead, a variable related to the intervention is selected, and the intervention and control groups are divided according to a cut-off value determined by the value of this variable. RDD was superior to other quasi-experimental designs in terms of internal validity but showed limited external validity. RDD based on observational information controls not only observable confounders but also unobservable confounders, rendering it a more ethical approach. In this example, polynomial regression and local linear regression analyses showed similar result.

Conclusion RDD has a broad application in real-world public health research. Our result also provide new ideas and method for scientifically evaluating intervention policies and establishing a basis for high-quality decision-making.