Literature: autocratization and vote choice

Autocratization refers to the decline in the quality of democratic practice or institutions in a country due to an incumbent’s actions to limit competition or consolidate power.Footnote ² A rapidly growing body of literature seeks to measure citizens’ willingness to punish autocratizing incumbents electorally. Most of these studies conclude that voters prioritize partisanship, policy preferences, and/or economic performance over preservation of democracy (Albertus and Grossman, Reference Albertus and Grossman2021; Cohen et al., Reference Cohen2023; Gidengil et al., Reference Gidengil2022; Graham and Svolik, Reference Graham and Svolik2020; Jacob, Reference Jacob2025; Krishnarajan, Reference Krishnarajan2023; Lewandowsky and Jankowski, Reference Lewandowsky and Jankowski2023; Mazepus and Toshkov, Reference Mazepus and Toshkov2022; Orhan, Reference Orhan2022; Saikkonen & Christensen, Reference Saikkonen and Serup Christensen2023; Şaşmaz et al., Reference Şaşmaz2022; Simonovits et al., Reference Simonovits2022; Singer, Reference Singer2023). Much of this work relies on conjoint experiments requiring citizens to assess candidates with varying partisanship, policy platforms, and willingness to violate democratic principles. Selection of candidates aligning with respondents’ partisanship or policy preferences is taken as evidence of willingness to trade in democracy. These works suggest voters evaluate autocratizing incumbents no differently than other types of incumbents: effectively, voters prioritize their partisanship, policy preferences, and perceptions of economic performance, rendering autocratizing actions moot.

These studies provide a valuable starting point for understanding citizen responses to autocratization, but also have important limitations. Existing work has yet to systematically evaluate the extent to which respondents perceive the autocratization to which they are supposed to be responding. For example, Singer (Reference Singer2023) shows that citizens across Latin America rate the quality of their democracy more highly when sharing partisanship with the president, regardless of actual democracy levels. This analysis accounts for democracy quality country-by-country, but not for varying degrees of subtlety in democratic decline, or citizens’ perception of the varied attacks on democratic institutions. Other observational studies demonstrate that ruling party partisans hold more anti-democratic attitudes and tolerate more democratic erosion, but measure attitudinal responses to hypothetical situations rather than behavioral responses to incumbent actions (Bartels and Kramon, Reference Bartels and Kramon2020; Mazepus and Toshkov, Reference Mazepus and Toshkov2022). Graham and Svolik (Reference Graham and Svolik2020) supplement their experiment with a real-world analysis, measuring US congressional candidate Greg Gianforte’s vote loss after assaulting a journalist the night before the 2017 election. It is unclear, however, whether voters saw this incident as evidence of systematic anti-democratic behavior, or merely as pugilistic sensibilities and a short temper. While either should disqualify a candidate, they hold different meanings for the prospects of democratic survival.

Survey experiments in this literature, while causally precise, often measure citizens’ responses to subtle autocratizing behavior. For example, a survey experiment in Turkey measured support for “executive aggrandizement” through questions about whether the president should be able to make laws through decrees, make the budget, or have broad or symbolic powers (Şaşmaz et al., Reference Şaşmaz2022). An experiment in Germany measured illiberalism through candidate support for the government’s direct judicial appointments to the constitutional court, increased government influence on public broadcasting, and the ability of the government to pass bills without parliament (Lewandowsky and Jankowski, Reference Lewandowsky and Jankowski2023). By design, the authors note, these positions “do not reflect a too obvious anti-democratic attitude” (ibid, p. 44). However, these types of actions may not be legible to citizens as “anti-democratic.” Variation in other experiments provides some insight into the relative effects of subtle versus blatant acts. Saikkonen and Christensen (Reference Saikkonen and Serup Christensen2023) use a conjoint experiment to estimate the effects of “democratic norm transgressions” on support for hypothetical candidates. Subtle transgressions like failing to condemn violence against opposition officials or equivocation on whether judicial decisions must be respected moderately reduce support, while the most obvious transgression—inciting violence against opposition officials—elicits double the marginal effect.

Other works highlight the perceptual biases that may reduce recognition of democratic violations. Krishnarajan (Reference Krishnarajan2023, p. 474) argues that voters “do not give up democracy in a calculating way to gain politically,” but rationalize that the anti-democratic behavior they witness is, in fact, democratic. He demonstrates this with relatively subtle hypothetical democratic transgressions by U.S. Senators: barring some newspapers from attending a press meeting, promising quid pro quo to an interest group, making false statements in a campaign ad. Similarly, Grossman et al. (Reference Grossman2022) show that people with majoritarian conceptions of democracy think that actions taken by a democratically elected leader are, by definition, democratic through an experiment about an outgoing governor attempting to make norm-violating judicial appointments. Wunsch, Jacob, and Derksen (Reference Jacob2025) illustrate that people with illiberal understandings of democracy are more likely to support candidates with illiberal attitudes on judicial appointments and the role of the media. Affective polarization—when partisans hold deeply positive emotions about their partisan in-group and deeply negative ones about the out-group (Iyengar et al., Reference Iyengar2012)—may amplify perceptual biases. High affective polarization can raise the stakes of elections, and partisan animosity in particular can increase the social and political costs of crossing party lines (Somer & McCoy, Reference Somer and McCoy2019), thus heightening the prospects for perceptual bias (Krishnarajan, Reference Krishnarajan2023).

These studies underscore the importance of distinguishing between autocratizing behavior that is subtle—less visible to the population, less legible as anti-democratic, more subject to rationalizing or other perceptual biases—and behavior that is unquestionably undemocratic. While voters may be unaware of or able to rationalize away the undemocratic nature of more subtle encroachments on democratic institutions, they should be less able to do so with highly visible autocratic acts. There is some evidence thereof from studies that focused on more blatant anti-democratic actions. For example, McKie and Carlson (Reference McKie and Carlson2022) show that term limit contravention attempts—a highly visible and legible anti-democratic act—increase mass support for term limits in Africa. Frederiksen (Reference Frederiksen2024, Appendix C) shows that voters do sanction undemocratic candidates, even when they share partisanship and policy preferences, but notes when disaggregating the undemocratic treatments that some of the results are sensitive to the “strength of the experimental treatments.” Lastly, Carey et al. (Reference Carey2022) show that, in the United States, voters will penalize potential candidates who have expressed support for ignoring court decisions and promoting politically motivated investigations of opposition party members, regardless of whether they share partisanship with the candidate. However, support for candidates who endorse restrictive voter ID laws is highly partisan. The authors note that this attribute was the most controversial, but it is also the most ambiguous. Partisan arguments for voter ID laws typically focus on their importance for strengthening democracy, making it the most likely of their treatments to be prone to partisan rationalizing.

Most work on voter responses to autocratization focuses on a small set of liberal democracies, so it is important to consider how including electoral autocracies, such as those in Africa, might influence voter perception and response to autocratic acts. Existing work suggests that violations of democratic rules may be more severe when the country was less democratic to begin with, but that voters still perceive and respond to these violations. For example, McKie and Carlson (Reference McKie and Carlson2022) demonstrate that support for presidential term limits increases after term limit contravention attempts in Africa, and Emordi et al.(Reference Emordi2024) and Hern (Reference Hern2024) demonstrate that anti-democratic incumbent behaviors were important for generating electoral turnover in Senegal and Zambia, respectively.

When voters support autocratizing incumbents or candidates, then, they may not always be trading democracy for policy preferences or partisanship. Voters may sometimes be unaware of autocratizing behavior, or they may be engaged in partisan rationalizing. Across competitive regime types, these possibilities should be less likely when an incumbent’s actions are visibly, unequivocally anti-democratic.

Argument and hypotheses

While most people prefer democracy to other forms of government, they may not readily recognize an incumbent’s autocratizing behavior. Contemporary autocratization often takes place in subtle ways, such as obscure changes to legislative voting rules, re-districting, or judicial stacking. The average citizen is unlikely to understand the implications of these changes, particularly if they occur slowly. Even if these changes receive media coverage, the average person may be unmotivated to pay attention or comprehend how, cumulatively, they chip away at an opposition party’s ability to win elections or influence governance. Other forms of autocratization are far more visible and legible. Arresting opposition candidates on dubious grounds, shuttering media outlets critical of the government, or detaining private citizens for critical posts on social media are much more blatant tactics for consolidating power. I conceptualize “visible” autocratizing actions as blatant (i.e., hard to rationalize as “democratic”) and public (occurring within public view), and sometimes physically violent. Section five presents a full, detailed description and validation of the concept.

I argue that, for autocratization to influence an incumbent’s electoral prospects, it must be visible. Citizens are more likely to be aware of and interpret autocratizing actions as threats to democracy if they are visible. Citizens consider such actions in their evaluations of incumbents’ performance—typically to the detriment of the incumbent. The following hypotheses are coupled, wherein the “a” hypotheses focus on election outcomes, while the “b” hypotheses focus on voter motivations.

The existing literature suggests two alternative hypotheses. First, affective polarization may temper the impact of visible autocratization on vote choice and election outcomes. As noted above, partisan perceptual bias may lead voters to see autocratizing behavior as either justified or not autocratic at all. While I hypothesize that more visible forms of autocratization are less prone to such perceptual bias, partisanship may have a stronger influence under conditions of affective polarization (Simonovits et al. Reference Simonovits2022). Polarization may lead voters to feel they cannot support an out-party, even if they disapprove of autocratizing actions carried out by their in-party (Aarslew Reference Aarslew2023). Affective polarization primarily operates to weaken accountability linkages between voters and incumbents on the individual level. However, when a country as a whole experiences high levels of affective polarization, this context may insulate an autocratizing incumbent from electoral ouster by reducing co-partisan defection in response to bad behavior (McCoy & Somer, Reference McCoy and Somer2019). In countries with high levels of affective polarization, then, visibility of autocratization may be less important because too many voters are locked into their partisan “teams.”

Another alternative hypothesis is that citizens’ evaluations of an incumbent’s autocratization may be tempered by good performance in other areas. Specifically, if the economy is performing well, or an individual’s personal economic situation has improved, they may be willing to overlook autocratizing behavior—particularly if they believe the incumbent can best handle the economy (Şaşmaz et al. Reference Şaşmaz2022). This alternative follows from the idea that voters do trade democracy for performance in a more calculated way, making visibility of autocratization irrelevant.

The scope condition for evaluating these hypotheses is competitive elections. If the incumbent’s autocratization has gone so far as to make it impossible for an opposition party to win, then the preferences of citizens are irrelevant. The following section therefore details the universe of cases I use for the empirical application of concepts in this study.

Universe of Cases and Case Selection

I examine incumbent performance in the universe of competitive African elections since 1990: 98 elections across 25 countries from 1991 to 2023. Elections meeting the following criteria were included in this dataset: first, there had to be a clearly identifiable incumbent or incumbent party. This criterion eliminates inaugural elections after transitional regimes or those in which the incumbent was removed through a coup or other extraconstitutional method. Where the incumbent individual did not stand because of term limits, death in office, or retirement, I used the incumbent party instead. Second, the election had to be reasonably free and fair, meaning that there was no evidence of systematic irregularities identified by independent election observers and the outcome was not flagged as suspicious by international election monitors. This criterion does not preclude other forms of tilting the electoral playing field, like the incumbent using state resources to campaign. Rather, it restricts the elections to those in which the vote tally was probably unmanipulated and most citizens were able to vote if they chose to. Third, there must be a viable opposition party, which I define as an opposition party that drew at least 20% of the vote. This is a low bar for opposition party viability and includes elections in which opposition parties had little chance of winning. Functionally, these criteria include elections in competitive autocracies as well as electoral and liberal democracies. A list of the elections included is in the online Appendix.

To identify which elections were carried out under conditions of autocratization, I rely on V-Dem’s Electoral Democracy Index (EDI) (Coppedge et al Reference Coppedge2024). Ranging from 0-1, this index measures the extent to which a country adheres to the minimal procedural components necessary for democracy, including free and fair elections, freedom of speech, media and association, and the ability to elect the country’s chief executive. Using the Episodes of Regime Transformation dataset (Maerz et al., Reference Maerz2024), I identify episodes of autocratization as occurring when there is a decline of 0.05 or more without substantial recovery. I matched these episodes to the list of elections, considering an election as occurring under autocratization if it (a) took place during an autocratization episode after the decline of 0.05 or more, or (b) took place in the aftermath of an autocratization episode perpetrated by the incumbent since the previous election.Footnote ³ Using this definition, 17 elections in the universe of 98 cases took place in the context of autocratization. Section A2 in the online Appendix presents a lengthier discussion of the case selection criteria, use of alternative thresholds and measures, and a sensitivity analysis. The set of elections under autocratization varies slightly across measures but the results presented below are robust to the varied case selection methods. EDI scores do not identify the source of the decline. My selection criteria eliminate cases where coups or conflicts disrupt normal electoral processes but include cases where decline may be the result of gradual decay resulting from corruption or dysfunction rather than incumbent-driven autocratization. I contend that incumbents are most likely to lose re-election in such contexts where the autocratization is incumbent-driven and visible.

Concept development: visibility of autocratization

Concept measurement

Measuring “visibility of autocratization” is challenging because it requires identifying autocratic acts across countries with different levels of news coverage. It is further impossible to know whether something happened that was not reported. There are two possible approaches. The first is to pursue country-specific coding using original news media from each country. While generating a more precise measure, this approach would also be non-comparable and introduce systematic bias in that countries with less (or less available) media coverage would necessarily have lower scores. Using international media introduces a similar challenge, as some countries are much more heavily covered than others. The other approach (employed here) is a blunt instrument that trades precision for comparability and likely underestimates visibility. This approach uses a single set of coding rules and data from a single source, Freedom House (FH) country reports, to evaluate visibility of autocratization country to country. While these reports do not mention every act of autocratization within a country, they are likely to mention “newsworthy” incidents that fall into the “visible” category. Importantly, this approach is likely to systematically underestimate visibility of autocratization, biasing this study against a finding. The primary advantage of FH reports as opposed to those from other organizations is that they are available for all countries, are structured to be directly comparable, and cover relevant political developments for countries with and without autocratization.

To operationalize visibility of autocratization, I used FH reports to identify incidents of an 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. I divided these instances by “target” and whether the action was “visible” or “subtle.” The coding of each type of autocratizing act covered in this dataset is displayed in Table 1. For these acts to be counted, they must be likely attributable to the government. For example, a journalist being killed by insurgents would not count, while a journalist being killed by an agent of the state after publishing a critical report about the president would. Similarly, the acts must be outside the normal workings of the law. For example, an activist being arrested for destruction of property after setting a car on fire during a protest would not count, while an activist being arrested for a peaceful protest would. Where there were ambiguous cases—incidents that were not clearly (or likely) attributable to the government, or where it was unclear whether the incident occurred outside the normal workings of the law, I did not count them.

Table 1.

Types of autocratization by target and visibility

To generate a country-level score, I count the number of visible “types” of autocratization that FH recorded per year. There are 13 visible types of autocratization; annual country scores have a possible range of 0 to 13 (i.e., arrest of protestors is only counted once per year, even if there are multiple incidents in a year). The score measures the number of types of autocratization, not the frequency of such incidents. The reasoning for this coding choice is both theoretical and empirical: theoretically, a government performing a visible action even once should draw attention and indicates its willingness to do so. Empirically, an attempt to count all instances of each action (like “arrest of protestors”) is likely to generate a biased score, in that countries with a stronger independent media presence are more likely to catalog and report each incident. Furthermore, FH often indicates that the incidents it records happen “regularly” and provides a few high-profile examples rather than enumerating each one, making it impossible to generate a systematic total count of each incident. The resulting count is therefore very conservative and underestimates the “true” visibility value. Importantly, these visibility scores do not capture the full scope of autocratization undertaken by the incumbent, but operate as a lower bound on the portion of autocratization that is visible. Again, this underestimate biases the study against finding a result. An additional challenge in producing this score is that FH changed the structure of their reports in 2016, so that reports after 2015 are more detailed than earlier ones. Visibility scores are thus systematically lower before 2016. The analyses below account for this issue in varied ways (described in each section).

A country’s visibility score is the count of how many forms of visible autocratization the incumbent undertook annually during the 24 months prior to the election. This period reflects the likely attention span of a population, which will discount events that happened further in the past. The resulting visibility indicator has a theoretical range of 0–26; scores in the sample range from 0–9. As noted above, using the FH reports as source results in a likely underestimate of autocratizing acts in each country. Counting types rather than frequency further minimizes the difference in measurement between countries. Because the measure is an underestimate that minimizes difference, it is a highly conservative indicator that biases an analysis using it against finding an effect.

Due to the novelty of this measure, I use three forms of validation to evaluate whether it appropriately reflects the concept. Content validation, drawn from qualitative analysis of original interviews carried out in Zambia, supports the notion that people are indeed more likely to recognize as autocratic behaviors the ones I categorized as visible. Convergent validation confirms that the combined count of visible and subtle forms of autocratization from Freedom House reports correlates closely with other measures of democratic decline, while discriminant validation shows that visibility of autocratization (alone) is weakly correlated with, but distinct from, the severity of autocratization. These validations are available in full in the Appendix.

Empirical application: visibility of autocratization and election outcomes in Africa

Does visible autocratization reduce incumbents’ likelihood of re-election in Africa? In the universe of reasonably free African elections with an incumbent, official election results indicate the incumbent won 64% of the time. When individual incumbents stood for re-election, they were re-elected 69% of the time, whereas incumbent parties fielding new candidates won only 50% of the time.

To assess the three hypotheses related to election outcomes, I examine the data in two ways: first, I compare the means of the metric of interest for incumbents who were re-elected compared to those who were not, employing t-tests to determine whether the two populations are statistically distinct. Next, I look at re-election rates for incumbents in countries that were above and below the metric’s mean, again using t-tests to determine if re-election rates are statistically different. Unfortunately, the limited sample size (98 elections total, but only 17 under autocratization) precludes regression analysis. Figures 1 and 2 preview the findings for this section. In short, visibly autocratizing incumbents have lower re-election rates compared to subtle autocratizers as well as the full sample of incumbents (Figure 1), and visibility scores are much higher for incumbents who were voted out compared to those who were re-elected (Figure 2). It was the only metric that was consistently significantly related to incumbent re-election prospects at p < 0.1 (marked by *). High polarization appears to be bad for autocratizing incumbents’ re-election prospects, but the difference does not reach statistical significance. Good economic performance gives a slight but statistically insignificant boost to all incumbents.

Figure 1.

Re-election rates by population.

Figure 2.

Mean metrics in autocratizing sub-sample by re-election status.

Autocratizing incumbents won re-election at the same rate as incumbents in the broader sample (65% and 64%, respectively). However, differences emerge once taking visibility of autocratization into account. Using the operationalization described in section 5, the mean visibility score for the sub-sample of elections under autocratization was 3. The 11 autocratizing incumbents who were re-elected had an average visibility score of 1.8, while the six who were voted out had an average score of 5.2 (p = 0.046). Incumbents with visibility scores less than the sub-sample mean of 3 were re-elected 80% of the time, while those who had scores of 3 or greater were only re-elected 43% of the time (p = 0.08). As noted above, the quality of the reports used to generate the visibility scores improved in 2016. Restricting the sample to the eight elections that occurred after 2016 reveals the same pattern: the four re-elected incumbents had an average visibility score of 1.25, while the four who were voted out had an average score of 7.25 (p = 0.008). Visibly autocratizing incumbents were more likely to be voted out than their subtler counterparts as well as the broader sample of incumbents, supporting H1a.

There are obvious limits to this analysis given the sample size and the challenges around operationalizing visibility of autocratization. One specific concern is the possibility of a time trend: visibility scores for post-2016 elections tend to be higher because of improved reporting. If incumbency advantage also decreased over time, then the apparent relationship between visibility and incumbent re-election could be spurious. To address this possibility, I examine whether there is a time trend in incumbent advantage in the sample. Indeed, incumbents have become less likely to win re-election over time. The incumbent re-election rate in the first half of the data (49 elections spanning 1991–2010) is 69%, while in the second half (49 elections from 2011–2022) it drops to 59%. Post-2016, it drops to 54%, only slightly higher than the 50% re-election rate in the autocratizing sub-sample (N = 8). However, the re-election rate for autocratizing incumbents post-2016 with a visibility score of 3 or greater is only 20%: only one of the five incumbents in this category was re-elected. So, while incumbents’ re-election prospects have worsened over time, they remain uniquely bad for incumbents enacting visible autocratization.

Another possibility is that because visibility of autocratization is correlated (albeit weakly) with severity of autocratization, severity rather than visibility drives re-election prospects. I check this in two ways: by examining whether re-election is related to the size of the decline in V-Dem’s EDI, and whether the total autocratization score (count of both visible and subtle acts) is related to re-election. In short, while visibility correlates weakly with these measures of severity, neither total autocratization nor EDI decline are significantly associated with incumbent re-election.Footnote ⁴ Visibility, not severity, correlates with lower re-election rates.

The first alternative hypothesis (2a) posited that autocratizing incumbents would be more likely to retain office when affective polarization is high. There is no evidence to support this hypothesis in this sample. I use V-Dem’s measure of polarization the year prior to the election,Footnote ⁵ which ranges from zero (supporters of opposing political parties are friendly to one another) to four (supporters of opposing political parties are generally hostile to one another). The mean polarization score for the sample is 2.07. Contrary to the hypothesis, polarization is slightly higher in elections where the incumbent was voted out (2.2) compared to those in which the incumbent was re-elected (2), but this difference is not statistically significant (p = 0.11). When polarization is below average, incumbents are re-elected 69% of the time, versus 57% when it is above average (p = 0.21). Autocratization may condition the relationship between polarization and incumbent re-election, but not in the hypothesized direction. Polarization is slightly higher in autocratizing countries (mean of 2.3). The mean of polarization when incumbents were voted out was 2.6, and 2.1 when incumbents were re-elected (p = 0.18), though this difference is not significant. When elections in the autocratizing sub-sample took place with lower-than-average polarization, the incumbent was re-elected 85% of the time, but when polarization was higher than average, they won only 50% of the time (p = 0.15). Polarization does not insulate autocratizing incumbents in this sample. If anything, it hurts their chances at re-election.

This null result may be a result of the existing literature on polarization focusing primarily on countries in Europe and North America with highly institutionalized party systems, where partisan ties are comparatively deeper and more stable than in many African countries, which have had volatile party systems during the time under study (Dalton & Weldon, Reference Dalton and Weldon2007; Weghorst & Bernhard, Reference Weghorst and Bernhard2014). With more volatile party systems and malleable partisan ties, measures of affective polarization in such contexts may capture more fleeting sentiments than in countries with highly institutionalized, stable party systems. Party institutionalization may be an important scope condition for the relationship between affective polarization and re-election of autocratizing incumbents.

The second alternative hypothesis (3a) posited that incumbent re-election would be more likely when the economy was performing well. I operationalize economic performance by the growth rate in GDP per capita (PPP, to account for inflation) the year prior to the election. While this variable is an imperfect measure of economic performance, economic expansion and contraction are closely correlated with other indicators of economic wellbeing, particularly accounting for inflation. In the sample, average GDP growth per capita (PPP) was 3.8%. Economic growth seems to facilitate incumbent re-election in the broader sample. Average economic growth was 4.4% when incumbents were re-elected, and 2.8% when they were voted out (p = 0.09). When economic growth was worse than average incumbents won re-election 57% of the time, whereas when growth was better than average, they won 70% of the time (p = 0.10). In the autocratizing sub-sample, GDP per capita growth is lower, at 2%, and its relationship to incumbency dissipates.Footnote ⁶ Economic growth when these incumbents were voted out was 1.9%, compared to 2% when they were re-elected. Incumbents get a modest boost when their growth is greater than average, winning re-election 67% of the time (compared to 60% when growth is below average), but this difference is not statistically significant.

Economic growth is more sluggish in countries undergoing autocratization. Given that incumbent re-election in the broader sample is more likely with higher economic growth, it is plausible that incumbents become more likely to carry out autocratization when the economy is worse, attempting to circumvent the electoral consequences thereof. Whatever the causal direction of this relationship, once autocratization has begun, economic growth does not protect autocratizing incumbents (contrary to expectations).

Of the three factors analyzed here (visible autocratization, polarization, and economic performance), visible autocratization has the largest and most statistically robust relationship to incumbent re-election prospects. Contrary to expectations, affective polarization may negatively influence an autocratizing incumbent’s re-election prospects, though the difference is not statistically significant. Economic performance offers a boost to incumbents, but only if they have not undertaken autocratization.

This analysis provides a useful starting point for examining potential correlates of incumbent re-election under autocratization, but is limited in important ways. First, it cannot offer insights into the microdynamics that would drive these outcomes: namely, voter perception of autocratization, partisanship, experience of economic decline, and the relationships thereof to vote choice. Second, due to the small sample size, the ability to estimate precise relationships is limited. The following section turns to a large-N analysis of individual voter behavior in four countries to evaluate correlates of prospective vote choice.

Visibility of autocratization and vote choice: individual-level data

The data above suggest that incumbents are less likely to be re-elected when they carry out visible autocratization. The theoretical mechanism driving this outcome is that visible autocratization is legible to citizens and influences their vote choice. Alternatively, partisanship (especially under high affective polarization) or economic performance may be more important for vote choice.

I analyze the correlates of prospective vote choice in countries with visibly autocratizing incumbents using Afrobarometer data. I included countries that (a) were experiencing visible autocratization, (b) had an election after 2016, and (c) hosted an Afrobarometer survey within a year prior to the election. There are four cases that meet these criteria: Ghana (2020), Malawi (Reference VonDoepp2020), Niger (2020), and Zambia (2021). Each of these countries was sampled in Round 8 of the Afrobarometer less than a year before their elections. I restricted the sample to post-2016 elections to ensure that the coding for visibility of autocratization was comparable across elections (see discussion of challenges with earlier data in section 5). The analysis that follows employs multilevel mixed effects regression, nesting individuals within countries with random effects for country.

The dependent variable comes from a question that asks respondents what party they plan to vote for in the upcoming election. This binary variable is coded as 1 if the respondent selects the incumbent party, 0 if they select an opposition party, with a mean of 0.5. Those who state they do not know or they plan not to vote are dropped from the analysis.Footnote ⁷

The primary explanatory variable is “perceive autocracy,” which captures the extent to which respondents report perceiving the forms of visible autocratization taking place within their country. This variable is coded differently for each country depending on its form of visible autocratization, combining up to three indicators capturing perceptions about media restriction, speech restriction, and election tampering. In Ghana, where the only form of visible autocratization is infringement on free media, this variable only includes responses to a question about media freedom. In Niger and Zambia, it includes questions about both media and free speech restrictions, while Malawi includes all three (media, speech, and election tampering). The measure is standardized so that it ranges from 0–3 for all respondents, with higher scores reflecting stronger perceptions of autocratization. The sample mean is 1.03. If this approach accurately captures the forms of autocratization within each country, then we would expect scores for each of the constituent variables to be higher in the countries where that form of autocratization is documented. This is the case, as is detailed in the Appendix, alongside further description of the nature of visible autocratization in each country, alternative specifications of autocratization, and country-by-country regression analysis.Footnote ⁸

Independent variables include economic assessment, partisanship, and partisan animosity. Economic assessment is an ordinal variable that captures how respondents think the economy is doing compared to 12 months earlier, ranging from one (“much worse”) to five (“much better”) with a mean of 2.47. Partisanship is a three-category variable derived from a question asking whether respondents “feel close” to any political party. It is coded as 0 if respondents are nonpartisan (50.2%), 1 if respondents report feeling close to an opposition party (24.6%), and 2 if respondents report feeling close to the ruling party (25%). Animosity is an ordinal variable derived from the question “How much would you like having neighbors who support a different political party?”, where the responses range from one (strongly dislike) to five (strongly like”), with a mean of 3.84. This question was not asked in Niger. Notably, this indicator only captures half of the concept of affective polarization, which is the gap between in-group affinity and out-group animosity. Unfortunately, as the Afrobarometer does not include feelings thermometers for political parties, this variable is the closest proxy available. Control variables include age, sex, rural residence, education, ethnicity, and the lived poverty index. Full variable coding and complete results tables are in the Appendix.

The results, displayed in Table 2, show support for H1b (visibility) and H3b (economic assessment), but not H2b (animosity). Column I displays the coefficients for perceived autocratization, economic assessment, and partisanship in predicting incumbent vote. Unsurprisingly, the strongest predictor of incumbent vote in the sample was partisanship, with incumbent co-partisans much more likely to vote for the incumbent. Even after controlling for partisanship, however, perceived autocratization is negatively associated with prospective incumbent vote, offering evidence in support of H1b. The magnitude of this relationship is large: moving from the minimum (0) to the maximum (3) of perceived autocracy decreases the likelihood of prospective incumbent vote by 12 points, from 0.54 to 0.42. Positive assessment of the economy increases the likelihood of prospective incumbent vote, supporting H3b. Moving from the worst economic outlook to the best also increases the likelihood of prospective incumbent vote by 12 points, from 0.44 to 0.56.

Table 2.

Perceptions of autocratization and prospective incumbent vote

Note: Coefficients and standard errors (parenthesized) from multilevel mixed effects regression, with individuals nested within countries and random effects for country. Statistical significance denoted by ** if p < 0.01 and * if p < 0.05.

To evaluate whether partisan animosity amplifies co-partisan’s prospective vote choice for the incumbent, I interacted the measure of partisan animosity with partisanship (column II). The null results indicate that animosity does not moderate the relationship between partisanship and vote choice in this sample.

To unpack these results further, I consider the mechanisms underlying the relationship, including who is likely to report perceiving autocratization, and whether it may have a heterogenous relationship to vote choice. The mechanism is that people are exposed to information, which diminishes their perception of democratic quality in their country and ultimately their vote choice. If true, one might expect people who are more educated, politically engaged, and have more media exposure are more likely to perceive autocratization. If they connect their perception to democratic decline, there should be a negative association between their perception and assessment of the level of democracy in their country. In addition to being more likely to perceive autocratization, one might also expect the more educated, politically engaged, media consumers to be more likely to translate their perception to their vote choice. One might also expect this relationship to differ according to partisanship due to perceptual biases.

To evaluate these possibilities, I perform a series of additional analyses to examine the correlates of perceiving autocratization and its heterogeneous relationship to vote choice. First, I use the same modeling strategy as above to evaluate how education, political engagement (proxied by frequency of discussing politics)Footnote ⁹ , media exposure (additive index of frequency of accessing news across five sources), partisanship, and assessment of the country’s level of democracy (“to what extent is your country a democracy?”) relate to perceptions of autocratization. The results (Appendix A7) mostly accord with these expectations: people who are more educated, with more media exposure, and opposition partisans have higher levels of perceived autocratization, though political engagement is unrelated. Additionally, those who perceive autocratization also rate their country as less democratic. While not definitive, these associations are in line with what one would expect if people learn about anti-democratic incumbent behavior and connect that behavior to the quality of democracy.

Next, I evaluate whether the relationship between perceived autocratization and vote choice varies along the same dimensions. I re-run the regressions from the main analysis predicting vote choice with interactions between perceived autocratization and education, political engagement, media exposure, and partisanship. Education, political engagement, and media exposure do not moderate the relationship between perceived autocratization and vote choice, but partisanship does. Specifically, the impact of perceived autocratization on vote choice is almost entirely confined to nonpartisans. Figure 3 shows the linear predictions of expressing intent to vote for the incumbent by partisanship and perceived autocratization, ceteris paribus. As the figure illustrates, perceived autocratization does not influence vote choice for partisans, but the marginal impact of moving from 0 to 3 in perceiving autocratization decreases the likelihood that a nonpartisan votes for the incumbent from 66% to 29%. Given that nonpartisans make up 50% of this sample, that translates to a massive shift in vote share.

Figure 3.

Partisanship, perceived autocratization, and vote intent.

These results lend credence to the microdynamics underlying the national-level results, suggesting that the comparatively poor prospects of incumbents engaged in visible autocratization are linked to voters’ perceptions of their autocratizing actions. Partisan animosity was unrelated to re-election prospects at the national level, and the individual-level results suggest why: partisans are locked into their voting positions regardless of animosity, while nonpartisans are the malleable bloc. The individual- and national-level results on economic performance, however, are inconsistent. The individual results indicate that perceptions of the economy were just as substantively important as perceptions of autocratization for vote choice. This divergence suggests that national-level economic indicators do not necessarily reflect individual experiences of the economy.

Endogeneity: which incumbents are more visibly autocratic?

As there are reputational costs to visible autocratization (Cheeseman & Klaas, Reference Cheeseman and Klaas2018), it is plausible that leaders only choose to engage in visible forms of autocratization when they are already in a precarious position. In that case, their lower re-election rates reflect precarity rather than voters’ reactions to visible autocratization. Addressing this possibility, I examine the precarity of autocratizing presidents in the nine elections that occurred from 2016 onward. I rely on the existing literature to conceptualize precarity, drawing from works that examine when leaders resort to high-risk strategies for manipulating elections. The literature suggests that precarity is a function of a president’s ability to control key government offices alongside external factors related to performance (Cheeseman & Klaas, Reference Cheeseman and Klaas2018; Hafner-Burton et al., Reference Hafner-Burton2014; van Ham & Lindberg, Reference van Ham and Lindberg2015). I include seven factors that may intensify a president’s sense of precarity: autonomy of the election management body (EMB), judicial autonomy, lack of legislative majority, history of electoral turnovers (binary)Footnote ¹⁰ , poor recent economic performance, defections of high-level politicians to opposition parties, and opinion polling showing declining presidential support. These factors both indicate incumbent precarity and would augur poor election prospects.

I collect information on these factors in the period prior to these nine elections as mini-cases and use qualitative comparative analysis to evaluate whether visible autocratization follows precarity. Due to space constraints, the case detail is in the Appendix (A9). In short: total precarity, evaluated as a combination of all the aforementioned factors, is unrelated to visibility of autocratization across these nine cases. However, countries with visible autocratization share several factors.

Due to the small number of cases, this analysis examines the way characteristics cluster, rather than formal tests of statistical difference. Table 3 shows the raw scores for each election, shaded to indicate that it “counted” as an element of precarity. The table is organized by descending order of visibility to facilitate comparison across high- and low-visibility elections. First, precarious cases of high EMB and judicial autonomy (around 3 or 0.8 or higher, respectively, as measured by V-Dem) are clustered in the low-visibility elections, Lesotho 2017 being the exception for both. However, precarious cases where the incumbent lacked a legislative majority cluster in the high-visibility elections (excepting Zambia 2016). History of electoral turnover is equally distributed across high- and low-visibility elections, as are high-level political defections and cases of incumbent unpopularity in polls (though there was much uncertainty due to lack of polling data in high-visibility cases).

Table 3.

Elements of precarity prior to election

This comparison suggests that incumbents tended to engage in high-visibility tactics when they lacked a legislative majority but also had less autonomous EMBs and judiciaries. Additionally, it is notable that there was more uncertainty around public opinion prior to these elections, and that they tended to occur in countries with lower democracy scores at the time of the election (as measured by V-Dem’s EDI). This combination suggests that the lower autonomy of the EMB and judiciary created a more permissive environment for more visibly autocratic behaviors, while the lack of a legislative majority prevented autocratization through legislation, which tends to be more subtle.

Importantly for concerns of endogeneity, starting with a lower level of democracy and weaker EMB and judicial autonomy does not suggest that these incumbents were at higher risk of losing re-election. While they tended to lack legislative majorities because of the large numbers of parties competing in these systems, in most of these countries a legislative majority was not necessary for rule due to the separation of the legislative and executive branches. Rather, it appears that incumbents in these countries miscalculated how much malfeasance they could get away with—particularly in the context of uncertainty about their popularity. This analysis further suggests some important scope conditions for these concepts, as visible autocratization is particularly pronounced in weaker electoral democracies with limited institutional checks and balances.