Click It or Ticket at the Border: A Spatial Regression Discontinuity Analysis of Primary Seatbelt Enforcement Laws

Abstract

This paper investigates whether spatial regression discontinuity designs (RDD) at U.S. state borders can credibly identify the effects of seatbelt enforcement policy—and documents why they cannot. We examine geographic discontinuities where enforcement type changes from primary (police can stop drivers solely for non-use) to secondary (citation only if stopped for another violation). Using geocoded fatal crash data from FARS (2001–2019), we compare 289,916 crashes within 100 kilometers of enforcement borders. Our analysis yields a null point estimate (0.7 pp, 95% CI: $-$0.14 to 1.47 pp), but diagnostic tests reveal fundamental violations of RDD assumptions: McCrary density tests reject continuity ($p < 0.001$), placebo cutoff tests find significant "effects" away from true borders, and covariate balance fails for key variables. Crucially, our running variable (distance to nearest opposite-type state polygon) does not consistently correspond to actual treatment-changing border segments—a design flaw we document as a methodological warning for future spatial RDD applications. This paper serves as a cautionary case study: pooled multi-border spatial RDDs require careful construction of border-segment-specific running variables and conditioning, which we did not implement.