The Goldilocks Regression Discontinuity Design: A Dual-Threshold Framework for Evaluating Interventions Outside an Optimal Range

Regression discontinuity designs (RDDs) are widely used to estimate causal effects when treatment assignment is determined by a threshold on a continuous running variable. Many interventions, however, target individuals at both ends of a continuum, with treatment assigned when values fall either below a lower threshold or above an upper threshold. This paper introduces the Goldilocks Regression Discontinuity Design (GRDD), a dual-threshold framework for evaluating such interventions. GRDD estimates two locally identified treatment effects, one at each threshold, while allowing both effects to be interpreted within a common intervention framework. Separate local, simultaneous, and spline-based estimation strategies are presented. A 10-condition Monte Carlo simulation evaluates estimator performance across differences in sample size, functional form, heteroskedasticity, running-variable density, and treatment-effect magnitude. Results demonstrate an important tradeoff between precision and robustness. Simultaneous estimation provides substantial precision gains when the common functional form is correctly specified but can produce substantial bias and poor confidence interval coverage under misspecification. Separate local estimation is more consistently robust across conditions, while spline-based estimation was generally robust under nonlinear relationships. A semi-synthetic application using data from the National Health and Nutrition Examination Survey (NHANES) illustrates implementation of the framework at lower and upper body mass index thresholds. GRDD extends the application of RDD to dual-threshold interventions while preserving local identification at each cutoff and providing a unified framework for comparing intervention effects across distinct regions of need.