When does regression discontinuity design work? Evidence from random election outcomes

Abstract

We use elections data in which a large number of ties in vote counts between candidates are resolved via a lottery to study the personal incumbency advantage. We benchmark non‐experimental regression discontinuity design (RDD) estimates against the estimate produced by this experiment that takes place exactly at the cutoff. The experimental estimate suggests that there is no personal incumbency advantage. In contrast, conventional local polynomial RDD estimates suggest a moderate and statistically significant effect. Bias‐corrected RDD estimates that apply robust inference are, however, in line with the experimental estimate. Therefore, state‐of‐the‐art implementation of RDD can meet the replication standard in the context of close elections.

Relative age effects in political selection

“…We also verify the robustness to alternative range of bandwidths (columns (1) and 2in Tables 2 and 3). We conduct the RDD both with (somewhat outdated, see Hyytinen et al 2017, Calonico et al 2014and 2016a conventional local linear approach (columns (1) and 2of Tables 2 and 3 Tables and 3, referred to as the CCT-correction). We also report results where we fix the main and bias bandwidths to be the same following the advice in Calonico et al (2016a) and Hyytinen et al…”

“…Similar analysis for the local elections gives no reasons to doubt the validity and robustness of the zero effects in that sample (not reported). Hyytinen et al (2017) argue that the placebo cutoff tests are particularly useful for specification testing (as opposed to validity testing). Therefore, to test the appropriateness of applied specifications, in Figures 3a-3d, we artificially move the cut-off to placebo locations by the amount of days denoted in the x-axis, and report the coefficient and the associated 95% confidence interval estimated for each placebo location.…”

Rank effects in political promotions

“…Our implementation closely follows Hyytinen et al (2017), who evaluate different implementations of regression discontinuity designs in a close election setting by comparing RDD results with the results from actual randomizations that happen when two (or more) candidates tie for the last seat in Finnish local elections. More specifically, we use local linear and quadratic polynomials estimated separately on both sides of the cutoff.…”

The urban–rural gap in healthcare infrastructure: does government ideology matter?

“…Joshi (2015) and Béland and Oloomi (2017) use a similar approach to estimate the causal effect of government ideology on healthcare expenditure in US states. Hyytinen, Meriläinen, Saarimaa, Toivanen, and Tukiainen (2018) show that RD may well be equivalent to a randomized experiment in the context of close election outcomes when using certain RD estimation techniques developed by Calonico, Cattaneo, and Titiunik (2014) and Calonico, Cattaneo, and Farrell (2018). We follow this suggestion and estimate local polynomial RD using the optimal polynomial and bandwidth procedure and robust RD standard errors using the instructions of those authors.…”

“…Panel A in Table 1 shows the baseline RD estimates for our eight measures of hospital infrastructure and spatial inequalities in hospital infrastructure using three different ways of computing standard errors. 13 The third row refers to robust standard errors, which are shown to be the preferred specification (Hyytinen et al, 2018). When we use the growth rate of the four measures for hospital infrastructure, the left-wing government RD estimate does not turn out to be statistically significant (columns (1) to (4)).…”