Combining convex regression with the regression discontinuity design: Effectiveness of e-scooter providers during the Covid-19 lockdown
提出一种将凸回归与断点回归结合的一阶段估计方法，用于分析挪威解封后电动滑板车运营商的表现，发现运营商在有效性和对政策响应上存在异质性。
• Contributes to the emerging literature on endogeneity and causal inference in efficiency analysis. • Proposes 1-stage estimation combining regression discontinuity and convex regression. • Adoption of novel bandwidth selection and robust inferences for heterogenous treatment effects. • Pioneers in analysis of effectiveness of e-scooter operators. • Operators are heterogenous wrt. effectiveness and responsiveness to lifting of Covid-regulations. The efficiency literature has only recently begun to address endogeneity and causal inference in frontier estimation. Most previous studies combine efficiency analysis with causal inferences using a two-stage estimation strategy, hence not properly addressing correlation between efficiency and treatment. This paper proposes a one-stage approach that combines a semi-nonparametric estimator subject to shape-constraints with the quasi-experimental regression discontinuity design. The proposed method enables joint estimation of the impacts of contextual factors and a non-parametric production frontier. Building on recent methodological developments within the literature on regression discontinuity design, the current paper develops strategies for estimation, bandwidth selection and inference for both average and heterogenous treatment effects. The novel methodology is applied for analyzing the reopening of Norway after lockdown using the Regression Discontinuity in Time design, considering both average and firm-specific treatment effects. Empirical analysis of the novel e-scooter markets in the cities Oslo and Drammen indicates that while the average performance of e-scooter providers was unaffected by the reopening of Norway during the Covid-19 pandemic, operators are heterogenous with regards to both effectiveness and responsiveness. While the empirical analyses provide mixed evidence that lockdown policies affect transport operator performance, they offer important insights into the benefits of accommodating heterogenous policy impacts in assessment of diverse and volatile industries.