Working Paper: NBER ID: w20405
Authors: Andrew Gelman; Guido Imbens
Abstract: It is common in regression discontinuity analysis to control for high order (third, fourth, or higher) polynomials of the forcing variable. We argue that estimators for causal effects based on such methods can be misleading, and we recommend researchers do not use them, and instead use estimators based on local linear or quadratic polynomials or other smooth functions.
Keywords: identification; policy analysis; polynomial regression; regression discontinuity; uncertainty
JEL Codes: C01; C1
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| High-order polynomial regressions (C29) | Misleading estimators for causal effects (C51) |
| High-order polynomial regressions (C29) | Sensitivity to polynomial order (C69) |
| High-order polynomial regressions (C29) | Potential biases (D91) |
| High-order polynomial regressions (C29) | Confidence intervals that are too narrow (C46) |
| High-order polynomial regressions (C29) | Excessive influence from extreme values (C46) |
| Local linear or quadratic polynomials (C29) | More reliable estimates (C51) |