Priced but not paid: The electoral penalty for undemocratic behaviour in candidate-choice experiments, and the geography of a missing literature

Abstract

Purpose. Experimental political science has converged on a reassuring finding: when citizens are shown a politician who would gerrymander a district, defy a court, or purge the civil service, they withdraw support. This article asks how large that penalty actually is once the estimates are collected and placed on a common scale, and where in the world the design that produces them has been run.

Methods. A systematic search of candidate-choice and candidate-evaluation experiments that randomly assign a democracy-violating attribute to a politician and measure the effect on electoral support identified twenty-six studies, of which nineteen qualify, across thirteen countries. Seven independent samples report an estimate on the probability scale together with a usable measure of precision. These were harmonised into percentage-point changes in the probability of selection and pooled using a DerSimonian–Laird random-effects estimator, with Knapp–Hartung intervals, leave-one-out diagnostics and a prediction interval.

Findings. The pooled penalty is 12.0 percentage points (95% CI 9.2 to 14.9; Knapp–Hartung CI 7.7 to 16.3). Every individual estimate is negative and every confidence interval excludes zero. Heterogeneity is substantial in magnitude but not in sign (tau = 3.0 points; I² = 80.7 per cent), and the 95 per cent prediction interval, from 3.8 to 20.3 points, excludes zero. The same literature, however, reports that the penalty falls to roughly three points and loses statistical significance once the violating candidate is a co-partisan whose policy positions the respondent shares. Three of the seven pooled samples come from the United States. One qualifying study has been fielded in sub-Saharan Africa, and it measures electoral violence rather than institutional transgression. Where Global South cases are included in multi-country conjoints, the country-specific coefficient is generally not reported.

Implications. The literature has established with considerable precision what citizens will do when democratic principle is the only thing at stake, and almost nothing about what they will do when it is not. The evidence base sits in consolidated democracies at a moment when forty-four countries, twelve of them in sub-Saharan Africa, are recorded as autocratizing. The article specifies what an African replication would have to test, and why the existing design would have to change to test it.