Visibility of autocratization and election outcomes
Visibility of autocratization and election outcomes (DOI: https://doi.org/10.1017/S1755773925100076) was written by Erin Hern in 2025. It was published in European Political Science Review.
"[T]he current wave of autocratization is subtler and slower than previous ones, such as the post-WWII wave characterized by coups. Instead, this wave is marked by elected officials making institutional or legal changes that consolidate their power and undermine the ability of opposition parties to gain office..."
The authors argue that visibility of anti-democratic actions determine when voters reject autocratization. 'Visible' is considered to be blatant, in the public, and sometimes violent.
The first contribution of this paper is in an identification strategy.
Data is collected from the competitive elections (free and fair, with at least one genuine opposition party) with a clear incumbent or incumbent party from African states between 1991 and 2023: 98 elections in total covering 25 states.
An episode of autocratization is defined as a decline in V-Dem Electoral Democracy Index (EDI) scores of 0.05 or more without a substantive recovery. An election is considered to have occurred under autocratization when it coincides with the episode, or when it follows an episode perpetuated by the incumbent. By this definition, 17 out of 98 occurred under autocratization.
Acts of autocratization are extracted from Freedom House reports: "incumbent taking steps to entrench their power, undermine opposition parties’ ability to compete, or inhibit free expression of the media, civil society, or general public." The act must be clearly attributable to the government and must be unambiguously extralegal.
The authors classify these acts by target (i.e., media, opposition, government, civil society, or citizens) and visibility. Through this process, they arrive at a categorization in which there are 13 types of visible autocratizing acts. A country's visible autocratizing score is the count of such acts in a year, ranging 0-13. The intuition is that reports of mass arrests of protestors is less visible than a report of arresting a protestor and also a report of arresting a journalist. Furthermore, reports of arrest counts will be driven by population level and size of the domestic journalist industry, rather than visibility. This score is expected to underestimate visible autocratization.
An election's autocratizing score is the sum of scores for the 24 preceding months, ranging 0-26. The intuition is that 2 years is a reasonable expectation for voters' attention span. The estimated scores range from 0-9. The sample mean was used as the threshold for identifying visibly autocratizing incumbents.
The first analysis is of election-level data. Specifically, whether the autocratizing incumbents won re-election.
V-Dem measures of polarization, ranging 0-4, is used as an estimate of affective polarization. This is examined as an alternative predictor of re-election.
Average annual GDP growth in this sample was 3.8%, which is used as the threshold for identifying high growth, another alternative predictor for re-election.
Looking at proportions of incumbents who win re-election, visibly autocratizing incumbents are less likely to win. The only other statistically significant split is high growth (more likely to win).
Looking at mean re-election rates, visibly autocratizing incumbents are less likely to win. This is the only statistically significant split.
The second analysis is of Afrobarometer data. This led to a subsetting of elections; the election must have been after 2016, the incumbent must have been identified as visibly autocratizing, and the Afrobarometer must have collected data in the state for the year prior to the election.
The full set of elections after 2016 are:
Botswana 2019
Ghana 2020
Lesotho 2017
Liberia 2017
Malawi 2020
Niger 2020
South Africa 2019
Zambia 2016 and 2021
The criteria above subset this to Ghana, Malawi, Niger, and Zambia (in 2021).
Respondents are asked which party they intend to vote for; the dependent variable is coded as 1 if they select the incumbent party, and 0 otherwise. Refusals and don't knows are excluded from analysis.
Respondents are also asked about their perceptions of the current state of freedoms (media, speech, and elections). A perceived autocratization score is calculated differently for each state, dependent on the classes of autocratizing acts reported by Freedom House.
- For Ghana, looking at media freedom alone
- For Niger and Zambia, looking at media freedom and free speech
- For Malawi, looking at all three
The score is normalized to range 0-3 for all respondents. The sample mean is 1.03.
Perception of the economy (as compared to 12 months ago) is measured as a Likert scale (much worse to much better). The sample mean is 2.47.
Partisanship is measured in two ways. First, a self-identification question; this is recoded to 0 for indifferent or nonpartisan responses, 1 for opposition parties, and 2 for incumbent parties.
An animosity question asks "How much would you like having neighbors who support a different political party?". Responses are on a Likert scale (strongly dislike to strongly like). Notably this question is not asked in Niger. Also notably, this is not the same as affective polarization (the gap between in-group animosity and out-group animosity). The interaction of this measurement with self-identification is used as an alternate predictor.
The authors regress on the dependent incumbent vote variable. They use a multilevel model with respondents nested within states, and also introduce a random effect for states. They control for age, sex, urbanicity, education, ethnicity, and lived poverty index.
They find that partisanship, perceptions of autocratizing acts, and perceptions of the economy are significant predictors.
The partisanship and animosity interaction is added, but is not significant.
An interaction of partisanship and perceptions of autocratizing acts is added; the negative effect estimated for those perceptions is actually largely confined to nonpartisans.
Reading Notes
Don't really buy into the author's characterization of recent autocratization.
Their findings on the interaction of partisanship and perceptions of autocratizing acts gives a pointer to the real story. Copartisans who are convinced by such reports are going to stop identifying with the party, so everyone else has surely come to the conclusion that the acts are 'not that bad' and will report that freedoms are protected still.