Visual Inference and Graphical Representation in Regression Discontinuity Designs.

The article investigates visual inference—the ability of readers to detect discontinuities in regression discontinuity (RD) design graphs—and how graphical representation choices affect this process. Using randomized experiments with nonexpert and expert participants viewing graphs generated from data calibrated on 11 published economics papers, the study finds that bin width and the inclusion of fit lines significantly influence visual inference accuracy, with small bins and no fit lines recommended for clearer interpretation. Comparing visual inference to econometric inference procedures (IK, CCT, and AK), visual inference achieves comparable or lower type I error rates but generally higher type II error rates, and the two approaches appear complementary when combined. The findings provide evidence-based guidance on RD graph construction and highlight the potential for integrating visual and formal statistical inference in empirical research.