THE QUALITY OF 
GOVERNMENT INSTITUTE
 
 
Cl   imate Shocks, Gender, 
 
an   d Incumbency 
 
Pu   nishment 
 
    
 20 
 
 
 
 
 
 25 
 
 
 
 
 Working paper series 2025:6  
  
 Emily Elia 
 
 Aksel Sundström 
 Stephen Dawson 
 
 
 
  
 
 
 
 
Climate Shocks, Gender, and 
Incumbency Punishment 
Emily Elia   
Aksel Sundström   
Stephen Dawson   
 
 
 
 
 
 
 
WORKING PAPER SERIES 2025:6  
QoG THE QUALITY OF GOVERNMENT INSTITUTE 
Department of Political Science 
University of Gothenburg 
Box 711, SE 405 30 GÖTEBORG 
September 2025 
ISSN 1653-8919 
© 2025 by Emily Elia, Aksel Sundström and Stephen Dawson. All rights reserved. 
 
Climate Shocks, Gender, and Incumbency Punishment 
Emily Elia   
Aksel Sundström   
Stephen Dawson   
QoG Working Paper Series 2025:6 
September 2025 
ISSN 1653-8919 
 
 
Abstract 
Abstract  
Do voters punish women incumbents more harshly than men during environmental disasters? Crises can 
exacerbate gender biases that lead voters to prefer masculine leadership, and these biases may make voters 
evaluate female incumbents’ performances during climate-related disasters more harshly. We test this 
argument in the context of South Africa, focusing on droughts – a salient crisis that informs voters about 
incumbents’ competence through their response. We complement a vignette experiment on incumbent 
gender and drought response with real-world geo-located data on drought onset, disaster declarations, and 
electoral results. We find consistent evidence that women incumbents receive greater electoral benefits than 
men when they provide drought relief, yet they receive a greater punishment when they fail to secure relief. 
Exploration of mechanisms highlights perceptions of male leaders’ involvement in stealing drought relief 
funds. These findings have implications for women’s representation, especially considering climate change 
where droughts are becoming more common and severe. 
 
 
 
 
 
Emily Elia   Aksel Sundström   
The Quality of Government Institute The Quality of Government Institute 
Department of Political Science Department of Political Science 
University of Gothenburg University of Gothenburg 
emily.elia@gu.se  aksel.sundstrom@pol.gu.se  
 
Stephen Dawson   
The Quality of Government Institute 
Department of Political Science 
University of Gothenburg 
stephen.dawson@gu.se  
 
 
 
 
 
 
 
 
 
Acknowledgement 
 
The authors are thankful for comments from seminars organized at the 2025 APSA annual meeting, the 
2025 Political Economy of Climate and the Environment (PECE) conference, the 2025 Politics & Gender 
Conference, the 2025 EPSA annual meeting, the 2025 Quality of Government (QoG) Institute 
conference, the 2025 GU the Gender and Politics workshop, and the 2025 Centre for Environmental 
Political Studies (CEPS) workshop. 
 
Research assistance by Paul Bederke, Zackarias Dahlberg and Herbert Ntuli is greatly acknowledged. 
 
This work was financially supported by Sundström’s ERC Starting Grant [project no. 101117889] and 
grants from The Swedish Research Council [grants no. 2020-03155, 2019-03218 and 2016-02119]. 
 
 
 
 
 
 
 
 
Introduction
Elected officials cannot control the weather, but an environmental shock like a major storm or
drought can teach voters about an incumbent’s competence. How a constituency fares during a
major weather event reveals information to voters about their government officials’ preparedness
in times of crisis. Environmental shocks can expose an incumbent’s ability to respond to citizens
in need and their willingness to mitigate further damages. Thus, though voters know a harmful
weather event is not necessarily an incumbent’s fault, they can still evaluate performance based
on how the incumbent responds to environmental shocks. Indeed, studies show that voters will
electorally punish officeholders for poorly handling a weather crisis (Arceneaux and Stein, 2006;
Blankenship et al., 2021; Carlin, Love, and Zechmeister, 2014; Cole, Healy, and Werker, 2012;
Eriksson, 2016; Gomez and Wilson, 2008).
A rapidly growing literature on the electoral responses to natural disasters studies the con-
ditions under which voters punish or reward incumbents in the wake of extreme weather events
(e.g. Heersink et al., 2022; Demirdogen and Olhan, 2025; Masiero and Santarossa, 2021; Visconti,
2022). However, this vein of research has not yet explored if voters enact accountability for an
environmental crisis against men and women incumbents in a similar fashion or if gender biases
lead to harsher punishment for women incumbents. Other work on electoral accountability in
general finds that women can face harsher scrutiny than men for their performance in office due
to gender biases that work against women officeholders (Bauer, 2015; Costa, 2021; Jones, 2014;
Kaslovsky and Rogowski, 2022; Murray, 2010), which suggests that women could also face harsher
voter backlash during environmental crises. Furthermore, in times of political and security crises,
voters often prefer male leadership (Holman, Merolla, and Zechmeister, 2016; Itzkovitch-Malka,
2024), which could spell trouble for women incumbents when an environmental shock hits. Taken
together, these different literatures suggest that when an environmental disaster occurs, female
incumbents may face disproportionate punishment from voters.
Environmental crises can create negative impacts that are uniquely felt by women, as women
tend to suffer from climate-related disasters at greater levels than men. These crises can negatively
impact women across many outcomes, from education to health to domestic violence (Arora-
Jonsson, 2011; Burke, Gong, and Jones, 2015; Corno, Hildebrandt, and Voena, 2020; Denton,
2002; Epstein et al., 2020; Moreno and Shaw, 2018; Neumayer and Plümper, 2007). Whether
women politicians also suffer the political impacts of an environmental crisis in a disproportionate
way remains understudied, though. Environmental crises can shift the gendered power dynamics in
government and create different political opportunities for women to enter politics (see Brulé 2023),
1
but what happens to the women already occupying office when an environmental crisis hits? Biased
voter assessments of female leaders could be one of the reasons for why women politicians remain
underrepresented (Hessami and Khasanboev, 2024). Our study contributes to this discussion by
providing novel insights into how assessments of incumbents in the wake of extreme weathers are
gendered.
In this paper, we ask if the political consequences of an environmental crisis are greater for
elected women. We investigate whether women officeholders face harsher electoral accountability
from voters for droughts and drought responses than men officeholders. By connecting theoretical
reasoning from different research veins and utilizing complementary types of originally collected
empirical data, this article provides several key insights through two interlinked studies.
We first use a preregistered vignette survey experiment1 fielded in South Africa that alters
information about droughts and incumbent ward councilors, local level representatives of South
African municipalities. 2 We ask respondents to evaluate a ward councilor who is either a man
or woman, and we manipulate whether a drought has occurred and how the incumbent councilor
has responded to the drought. We find that voters react similarly to female and male incumbents
when they are informed about a drought but not informed about the incumbent response; when
information about incumbent response is provided, though, interesting gender differences emerge.
Contrary to our expectations, the female incumbent who secures relief funds receives a greater
electoral reward for this performance than the male incumbent. However, when incumbents fail to
secure relief, the female incumbent is no longer preferred to the male incumbent. When analyzing
respondents who passed our manipulation check on incumbent gender, we find that, in line with
our expectations, women are evaluated more negatively and face greater electoral punishment than
men for failing to secure relief. Thus, voters’ reactions to women’s performance are more extreme
in both directions (punishment and reward) than they are to men’s performance. Our results
suggest that South African voters may view women as more agentic than men during a drought,
rewarding them more for good performance but also punishing them more for poor performance.
Our exploration of mechanisms, systematically analyzing open-ended responses in this survey and
also drawing on interviews from fieldwork, highlights that perceptions of male leaders’ involvement
in the theft of drought relief aid are integral to how voters form their views of incumbents.
We then investigate the external validity of these results by studying the real-world effects of
droughts on the electoral performance of men and women candidates in South African local elec-
1Anonymized OSF Preregistration: https://osf.io/6vrtf/?view_only=48b7c28ce372473fbc6354bc29ce0811
2This experiment received ethical approval from the authors’ institution’s ethical review agency (official name
currently redacted for anonymized peer review). More information about our ethical obligations can be found in
Appendix C.
2
tions between 2004-2024. We combine data from the Standardized Precipitation-Evapotranspiration
Index (SPEI) aggregated to electoral wards with municipality-level declarations of droughts to test
the effect of the onset of physical droughts and drought declarations close to elections on the elec-
toral fortunes of men and women incumbent ward councilors. A new dataset on local drought
emergency declarations amassed for this purpose and a total sample of all candidates standing for
elections in South Africa for the past two decades, that was verified by election authorities, enables
a precise analysis. Results highlight important similarities with experimental analyses: women
incumbents are electorally rewarded to a greater extent than men when droughts are declared in
their municipalities. Taken together, these results provide strong evidence for the gendered nature
of electoral accountability in the context of climate shocks and disaster relief in the South African
context. Voters respond to incumbent performance during weather-related crises differently de-
pending on incumbent gender.
This article makes at least three contributions. First, we bring insights to the literature that
examines the extent to which voters are attentive to incumbent’s actions after disasters and whether
they reward relief efforts (e.g. Gasper and Reeves, 2011; Healy and Malhotra, 2009), as this is the
first work to theoretically and empirically address how such processes have gendered elements.
Second, our study engages with scholarship on how different crises affect the standing of women in
politics (O’Brien and Piscopo, 2023; Blanton, Blanton, and Peksen, 2019) by documenting the local
impact that droughts and drought declarations may have on women’s electoral fortunes. Third,
by providing a multi-method inquiry on this topic, we speak to a growing literature that examines
linkages between gender and climate change such as environmental concern (Bush and Clayton,
2023; Benegal and Holman, 2025), women’s involvement in environmental policies (Clayton et al.,
2025), and theories on how disasters affect women’s capacities to run for office (Brulé, 2023), but
not yet how such climate disasters affect women incumbents. The focus on climate shocks in this
article reflects the growing relevance of these events in politics, as such events are projected to
become more common and severe, particularly in the Global South, due to climate change. We
show that as these extreme weather events increase in frequency and intensity, they can shape the
political trajectories of men and women in distinct ways.
Electoral Accountability, Environmental Crises, and Politi-
cian Gender
Voters sometimes attribute responsibility for natural disasters to incumbents, punishing them
electorally after the disaster occurs (Abney and Hill, 1966; Arceneaux and Stein, 2006; Blankenship
3
et al., 2021; Carlin, Love, and Zechmeister, 2014; Eriksson, 2016; Gomez and Wilson, 2008), or,
in cases where the incumbent’s reactions to the disaster were viewed favorably, rewarding them
for their good response (Lazarev et al., 2014; Bechtel and Hainmueller, 2011; Gasper and Reeves,
2011; Fukumoto and Kikuta, 2024). However, to our knowledge, scholars have not yet examined
this accountability for environmental shocks through a gendered lens despite evidence from other
research that voters evaluate politicians differently based on their gender.
Electoral accountability can be biased against women politicians, placing them under harsher
scrutiny than their male counterparts. Women politicians may be held to harsher standards by
voters due to voters’ tendency to evaluate male and female candidates through the filter of gender
stereotypes (Bauer, 2015; Dolan, 2010; Huddy and Terkildsen, 1993a,b; Lawless et al., 2010; San-
bonmatsu, 2002). When evaluating candidates, voters tend to associate “ideal” politician traits
with traditionally masculine traits, such as being assertive and strong, which leads male candidates
to better align with voters’ image of the ideal politician. This image is especially relevant during
times of crisis when traditionally masculine leaders are seen as more desirable by voters. Yet, voters
have also been found to dislike women candidates who take on too many masculine traits because
these women have “betrayed” their expected gender role (Cassese and Holman, 2018; Dolan, 2010).
As such, women leaders in the wake of disasters may face a “double bind:” women might be seen
as not conforming to traditional gender norms if they appear too assertive during a crisis, but
they may also be seen as violating the stereotypically ideal crisis-time leader if they respond in a
way that is seen as too caring and soft. Women are often evaluated as less fit for office than men
when voters rely on gender stereotypes to inform their candidate evaluation, and these gendered
stereotypes can even make women less likely to run for office than men in the first place (Fox and
Lawless, 2004; Lawless et al., 2010).
Once in office, women continue to face harsher scrutiny than men. One factor is media framing
and media visibility. Press could pay closer attention to women officeholders than men officeholders
and may portray women leaders’ responses more unfavorably (Bauer and Tremblay, 2011; Mur-
ray, 2010), increasing the chances that women’s behavior in office will be further scrutinized by
constituents. The role of visibility in news coverage has been pinpointed as especially important
in explaining why some incumbents are able to gain votes in the wake of disasters (Masiero and
Santarossa, 2021).
Another factor is how voters assess incumbents. Women in office are evaluated more meticu-
lously than men based on policy performance (Jones, 2014), congruence with constituents (Kaslovsky
and Rogowski, 2022), and legislator-constituent communication (Costa, 2021). Gendered differ-
ences in accountability have also been examined through the lens of corruption, and studies have
4
found that women officeholders face harsher consequences for corrupt behavior than men (Esarey
and Schwindt-Bayer, 2018; Reyes-Housholder, 2020). Voters are more likely to assess women pres-
idents’ performance by how well they handled the issue of corruption due to the common expecta-
tion that women should be less corrupt than men due to “fairer sex” stereotypes and/or women’s
outsider status in politics (Bélanger and Meguid, 2008). Similarly to gendered accountability for
corruption, women also face harsher consequences for sex scandals when voters hold hostile sexist
views (Barnes, Beaulieu, and Saxton, 2020). When women officeholders behave poorly in office,
voters react more negatively than they do to poorly behaved men.
However, not all work on gender and accountability finds that women face unequal punishment.
Sevi, Crabtree, and Blais (2024) find no evidence of women being punished more than men for
breaking campaign promises, and De Geus et al. (2021) find that voters judge the performance of
female and male politicians similarly. Other work shows that women do not face harsher account-
ability than men across the board, but they can face harsher accountability from certain sectors of
voters, particularly women voters (Eggers, Vivyan, and Wagner, 2018). Additionally, Eagly et al.
(2020) show that many gender stereotypes related to performance have eroded over time; while men
used to be perceived as more competent than women, stereotypes about competence today reflect
that men and women are often viewed equally, though men are still perceived as more agentic than
women. These studies paint a picture of mixed findings: electoral accountability is biased against
women in some instances but not all. However, it is critical to note that the bulk of this research
has been conducted in the context of advanced industrial Western democracies, with most of the
studies testing this relationship, as well as interrogating gender stereotypes, in the U.S. context. In
countries with less consolidated democratic institutions, weaker norms of electoral accountability,
higher levels of corruption, and a stronger presence of gendered stereotypes, different findings may
emerge. Where women’s representation is lower, gender biases may persist more strongly among
voters as well.
In sum, there is some evidence that electoral accountability is biased against women, particu-
larly where bad behavior is concerned. In the context of an environmental shock such as a drought,
women and men in office may face different reactions from voters, especially if officeholders fail
to properly respond to the issues caused by the drought. Though a politician cannot control the
weather, they can influence a constituency’s preparedness for environmental crises as well as the
constituency’s response to a drought. Voters may therefore attribute blame for the drought to
incumbents, holding them to account in subsequent elections for their (in)actions.
5
Punishing Incumbents for Droughts
An environmental shock can be a type of crisis that harms the economy, infrastructure, and life,
especially an event like a drought which can intensely deplete a community of resources and persist
over long periods of time. Particularly in contexts like Sub-Saharan Africa were drought conditions
can be extreme in an already-harsh climate, people experience loss of income due to agricultural
damage, diminished hygiene practices due to limited water, and scarcity of food and water, all
of which can plunge a community into crisis. In extreme cases, drought crises can even result in
the loss of human life. In times of crisis, many people wish for strong and assertive leadership,
which can activate gender stereotypes that favor male politicians over female politicians. Although
droughts are not the same kind of crisis as a terrorist attack or security threat, which existing
research on crisis and politician gender often focuses on, we argue that the hardship created
by a drought will trigger gender biases in voters that lead them to favor stereotypically strong
leadership in the same way that other common crises do. For example, Holman, Merolla, and
Zechmeister (2016) show that male politicians are favored in the US when the threat of terrorism
is salient. Similarly, Itzkovitch-Malka (2024) shows a preference for electing men during financial,
security, and education crises in Israel. Focusing on another type of crises—mortality from Covid-
19 in Germany— Hessami and Khasanboev (2024) document that female local incumbents had
a lower reelection probability than men, although local councilors in this setting had no role in
combating the pandemic.3 When an environmental crisis like drought strikes, voters’ preferences
for stereotypically strong and therefore masculine leadership may lead them to evaluate male
incumbents more favorably than women incumbents who do not fulfill the ideal image of a crisis-
time leader.
Voters may also believe that a politician who is well-connected and experienced is an asset
during crises because they will be more likely to successfully secure assistance for their constituency
as they “know the ropes” of politics better than political newcomers. This quality can be especially
valuable during a weather event like a drought when afflicted constituencies desperately need relief
resources to be acquired and distributed by their local government. Local governments are often
reliant on higher levels of government to provide resources for disaster relief, meaning that local
politicians need to know who to connect with and how in order to deliver relief to their constituents.
In a context like South Africa where women are underrepresented in politics, especially at the
subnational level, female officeholders may be perceived as political outsiders compared to their
male counterparts. This outsider perception could lead voters to believe that women lack the
3For related work showing that women are evaluated more harshly on the labor market and experience a “gender
punishment gap” for misconduct whereby women are punished more than men, see Egan, Matvos, and Seru (2022).
6
connections necessary to successfully secure drought assistance. The crisis of a drought leads
voters to prefer not only “strongman” leaders but also well-connected leaders who can obtain
relief resources quickly. Due to men’s historical and present-day domination of politics, voters
may believe that male politicians are best situated to get things done during difficult times. Men’s
greater embeddedness in government further enforces the belief among voters that men can perform
better during a drought crisis than women.
Taken together with the aforementioned findings that women incumbents can face harsher
accountability from voters in general and that voters will hold incumbents to account for their
responses to environmental disasters, we theorize that voters will react more negatively to female
incumbents than male incumbents during a drought. In drought times, women will face harsher
electoral accountability than men for the same drought responses as voters evaluate women’s
performance in office with greater scrutiny and look to stereotypically masculine leadership as
“ideal” leadership during crisis.
We expect that voters in a drought will have a greater affinity for male incumbents than female
incumbents in general. Tough times brought on by a drought will lead voters to prefer traditional,
strong, well-connected political leadership, most often embodied by male politicians. These gender
biases will lead voters to react more harshly to women incumbent’s political performance in drought
times, leading to greater electoral punishment for women than for men. When voters have no
information about incumbent response to the drought, we expect that gender stereotypes will
lead voters to assume that a man will be able to handle the crisis better than a woman. Male
incumbents will be more likely to be reelected than female incumbents in drought times, and they
will be perceived as more competent.
We also expect that voters’ preference for male leaders during a drought will result in a harsher
punishment for women than men when the incumbent fails to secure drought relief for their con-
stituency. In other words, voter support for and favorability of a woman incumbent will drop more
intensely than voter support for and favorability of a man incumbent when the incumbent does a
poor job responding to a drought. These expectations lead to our first two hypotheses:
H1: Female incumbents will be more negatively evaluated than male incumbents during a
drought when voters have no information about incumbent response
H2: Female incumbents will be more negatively evaluated than male incumbents during a
drought when the drought response fails
If voters punish incumbents differently for a poor drought response based on gender, will
they reward incumbents differently for a favorable drought response as well? While past research
provides convincing evidence that women often face harsher accountability from voters, it is less
7
clear if this finding also translates into a bigger electoral reward for women when they perform
well in office, particularly in times of crisis. On one hand, if voters’ gender biases lead them to
perceive male leadership as better fit for office, then a female leader who performs well in office
may experience a larger increase in voter support than a male counterpart because she handled
the crisis at hand in a manner that was “better than expected” by voters. On the other hand, it
may be possible that while a poor crisis response activates gender stereotypes against women, a
satisfactory crisis response neutralizes stereotypes, leading voters to react more uniformly to male
and female incumbents when they perform well. Voters may therefore reward men and women
in equal measure for a favorable crisis response. We hypothesize that when incumbents have a
successful response to a drought, meaning they were able to lessen the negative impacts of the
drought and bring some relief to constituents, voters will evaluate all incumbents positively in a
nongendered way.
H3: Female incumbents and male incumbents will be evaluated similarly during a drought
when the drought response succeeds
The Case
We focus our studies on South African local politics, a context that is well suited for research on
both droughts and women’s representation. Wards are the smallest district level in South Africa,
and they elect councilors via single-member district plurality who represent their ward at the mu-
nicipal level within the municipal council (People’s Assembly, 2024).4 Women are underrepresented
in councilor seats, with 3262 out of 4467 total councilor seats occupied by men (73.02%), giving
women a presence of just above a quarter of these seats (Electoral Commission of South Africa,
2024).5 However, there is variation in the amount of women’s representation across provinces;
at the highest level, Northern Cape province has 37.84% of its councilor seats held by women,
and KwaZulu-Natal province at the lowest level has 14.11% of its councilor seats held by women
(Electoral Commission of South Africa, 2024; Statista, 2024). Thus, while South African voters do
see women in the councilor role, women are far from achieving parity at the ward level. In fact,
women’s representation has recently decreased at the local level in South Africa (IDEA, 2024).
This low level of women’s representation may indicate that voter biases against women in office
4Municipal elections take place every five years, using a mixed-member system: half the councilors are elected via
proportional representation (party lists), and the other half through a first-past-the-post vote in individual wards.
Voters typically cast two ballots—one for a party and one for a ward candidate.
5The 1998 Municipal Structures Act encourages, but does not mandate, gender parity on party lists, recommend-
ing a 50% share of women and alternating placement (a “zebra” system). This soft quota excludes ward candidates
and is unevenly applied across parties, with the African National Congress (ANC) generally more supportive than
the Democratic Alliance (DA).
8
are present in the South African context, making it an appropriate case to test our hypotheses.
Public opinion data shed light on the challenges South African women face in politics. The
latest Afrobarometer reports that 58.8% of survey respondents believe that a woman who runs
for office in their community is somewhat likely (27.2%) or very likely (31.5%) to be criticized
or harassed by others in the community, and 50.6% of respondents believe that a woman who
runs for office is somewhat likely (20.4%) or very likely (30.2%) to face problems with her family
(Afrobarometer, 2024). South Africans perceive unique challenges for women officeholders, and
these perceptions reinforce the belief that politics is still very much a “man’s game” in South
Africa. These biases against women in politics may influence voters’ likelihood of enacting electoral
accountability against female incumbents in an unfair way.
South Africa is also a good fit for this study because of its experience with environmental
crises, particularly droughts. Recently, South Africa experienced an intense drought due to the
impact of the El Niño episodes in 2023 and 2024. Environmental projections suggest that the
country will witness an increase in mean annual temperatures, making future drought spells more
frequent and harmful (World Bank, 2021). South Africa is no stranger to drought conditions.
However, the record of South African authorities in responding to droughts has in general been
questioned. Authorities have responded with policies described as centered around relief schemes,
focusing more on immediate needs rather than building resilience to drought (Bahta, Jordaan, and
Muyambo, 2016; Meza et al., 2021). The water sector is plagued by corruption and relief funds
are known to often be siphoned off by local officials (Daniels, 2023; Corruption Watch, 2020).
In our observational analysis, we focus on the response of authorities to droughts through
the process of municipal declarations of drought emergencies. These emergency declarations allow
leaders to access relief funds. The formal drought response is governed by the Disaster Management
Act (Republic of South Africa, 2002), which outlines that local leaders, such as municipal councils
or mayors, are tasked with reaching out to the national authorities to obtain funds for their
locality when facing calamity (National Disaster Management Centre, 2019). Such declarations
are necessary for local leaders to access relief grants from the National Treasury.6 Importantly for
our study design, a declaration of drought emergency is described as a last resort for communities in
dire need; funds can be difficult to obtain due to burdensome administrative processes (World Bank,
2021, p 9). While any incumbent can initiate the process, there are several steps from the local
decision of attempting to declare a disaster to having a declaration granted by national authorities.
For example, actors at higher levels must “recommend or not recommend the request for a local
6We thank Mr. Jurgens Dyssel, Chief Director at the South African National Disaster Management Centre, for
his informative insights in these processes communicated in 2024.
9
disaster declaration.”7 Therefore, we consider a drought declaration to be a good indication of the
response of a local incumbent to droughts and whether this response is considered successful by
residents or not. Moreover, we assume that actual declarations are something that residents are
made aware of. Background interviews with residents in the Limpopo province (fieldwork in 2022)
who had recently experienced droughts corroborate this premise.8 In those interviews, respondents
voiced in various ways that residents facing droughts ask their locally elected leaders to approach
actors in central governments to make emergency funds available. People in drought conditions are
likely to be aware of the outcome of such requests, too. These announcements are made through
a government Gazette (Toxopeüs, 2025) and communicated to residents through news outlets.9
We also focus on droughts because they represent environmental crises that endure over time,
which means that there is a larger window of time for incumbents to respond to the crisis than
during acute events like a hurricane. Voters may therefore have higher expectations about how an
incumbent politician can help alleviate drought situations since droughts are not one-off disasters
that occur succinctly and then cease. In sum, we believe that evaluating elected officials’ drought
responses is an externally valid task for experimental survey respondents in the South African
context to complete. Additionally, the South African context provides rich observational data on
drought conditions, drought declarations, and women’s representation via electoral data, allowing
us to test our hypotheses with various types of data.
Study 1: Vignette experiment
We first test our hypotheses with a survey experiment, which allows us to identify causal relation-
ships between droughts, incumbent gender, and electoral accountability. We fielded our survey on
a sample of 5,00010 voting-age South Africans through the international survey firm YouGov. Our
sample is nationally representative by age, sex, province of residence, and socioeconomic status.
The survey was administered online, which, like all online surveys, comes with costs concern-
ing which populations can be best reached since internet access and internet-enabled devices are
needed. We sought to ameliorate this potential imbalance in our sample as best as possible by col-
7See https://www.gardenroute.gov.za/2018/05/31/declaring-a-local-disaster/.
8This refers to fieldwork that was carried out to inform us in formulating the research question guiding this article.
We draw on interviews with rural residents from drought-affected communities in Limpopo province, conducted in
spring 2022 when no drought was ongoing. Limpopo, located in the country’s northeast, is an ANC stronghold with
documented corruption in the water sector. A team of local author-trained enumerators conducted 50 face-to-face
interviews (with gender balance) in the local language. The villages were selected based on their potential eligibility
for drought relief. The background study received IRB approval.
9To illustrate, when a drought declaration was made in the Limpopo province in 2015, this was communicated
through media, see e.g. https://www.citizen.co.za/review-online/news-headlines/2015/11/05/r3-million-relief-for-
farmers-affected-by-drought/.
10We determined our sample size through a power analysis for eight groups with alpha set to 0.05, power set to
0.95, and effect size set to 0.3, a small-to-moderate effect according to common Cohen guidelines.
10
lecting nationally representative responses across province of residence and socioeconomic status so
that our sample is not disproportionately urban and/or wealthy. In terms of possible social desir-
ability bias among respondents when evaluating hypothetical politicians of different genders, there
may be an advantage to having respondents complete the survey online. When online, answers
are not affected by the gender-of-interviewer effect that has been shown to induce bias in answers
on gendered items among African respondents (Sundström and Stockemer, 2022). Nevertheless,
some participants may still refrain from passing hard judgments on female politicians in writing,
too. However, if anything, social desirability bias would lead to an underestimation of the degree
of punishment that women receive by voters. Still, we believe that the content of open–ended
answers in our survey where some respondents did not shy away from making negative comments
about women leaders suggests that people could answer without mannered responses.
The survey begins with standard demographic questions and political opinion questions like
partisan identity. Then, all survey respondents are randomly assigned (between subjects) to one of
eight conditions for the experimental portion. These conditions alter drought information across
four possibilities: a control condition with no drought, a drought treatment, a drought plus suc-
cessful response treatment, and a drought plus failed response treatment. The conditions also vary
by whether respondents read about a male or female incumbent councilor. These eight categories
are represented in Table 1.
The “No Drought” condition allows us to capture any baseline gender bias in respondents’ vote
choice that exist sans drought. Respondents may have a preference for men or women politicians
in non-crisis times that changes in the presence of a crisis; alternatively, if they have no gender
preference in non-crisis times but a gender preference when a crisis hits, then this change would
demonstrate the transformative impact a crisis like drought has on voter preferences. The “No
Drought” condition enables us to test this possibility.
The “Drought, No Response Information” condition then allows us to tease out if solely the
occurrence of a drought without any information about incumbent response causes a systematic
change in how respondents react to incumbents. The remaining two conditions, “Drought, Suc-
cessful Response” and “Drought, Failed Response,” provide respondents with information about
the incumbent’s behavior during the drought. We choose to have separate successful and unsuc-
cessful response conditions because electoral accountability is about both punishment (removing
one’s vote) for bad behavior and reward (giving one’s vote) for good behavior. It may be the case
that electoral punishment and electoral reward operate distinctly from one another.
Table 2 presents the different vignette texts. In this study, we are interested in the difference
between men and women incumbents without taking an intersectional approach that also considers
11
Table 1: Experimental Design
No Drought Drought, No Drought, Drought,
Response In- Successful Re- Failed Re-
formation sponse sponse
Incumbent Control Drought Drought + Suc- Drought + Fail-
Man cess ure
Incumbent Woman Woman + Woman + Woman +
Woman Drought Drought + Suc- Drought + Fail-
cess ure
race and ethnicity. Therefore, to avoid potentially confounding our gender treatment with race
and ethnicity cues that could significantly influence respondents’ evaluations, we do not state a
gendered name to signal the incumbent’s sex and instead tell respondents blatantly if the incumbent
is a man or woman.
The vignette texts reflect several design choices that we made to achieve as much internal and
external validity as possible. In the No Drought and Drought conditions, we had to present re-
spondents with information about the councilor that went beyond gender so that they had some
legitimate substance to use when evaluating the incumbent. However, we also wanted to avoid
introducing more variation or information about political performance into our vignettes that
could complicate causal inference when evaluating the treatment effects of the drought response
vignettes. We opted to describe the electoral process for and duties of a ward councilor in these
vignettes, using language directly from the South African People’s Assembly website about coun-
cilors’ responsibilities.11 Though this text is still vague when considering how a respondent may
evaluate the incumbent, we believe having respondents read this nonspecific information about
councilors is preferable to having respondents read information about some other aspect of per-
formance separate from a drought; we also believe this design choice is certainly preferable to
including no information about the councilor beyond their gender.
The vignettes deliberately do not specify the political party that the incumbent belongs to.
While there is a risk that respondents may have preconceptions about which party is more/less
likely to run with women candidates (and thus assign a party to the vignette), observational
data collected in conjunction in Study 2 alleviates this concern. Between 2006-2021, 34% of ward
candidates from the ANC were women, compared to 33% of ward candidates of the main challenger
DA. While the ANC is generally regarded as more concerned with the descriptive representation
of women – including a voluntary party quota at the local level – the overall discrepancy between
these parties does not apply at the ward level.
11https://www.pa.org.za/blog/what-role-ward-councilors
12
Table 2: Experimental Vignettes
No Drought:
Imagine you live in another ward within a municipality that is similar to yours.
The ward is represented by a male/female councilor.
He/She was elected by residents of the municipality in the last election. As a councilor, his/her
duties include monitoring the performance of the municipality and representing his/her ward
in the council. Councilors serve for 4 years and then can be reelected.
Drought:
Imagine you live in another ward within a municipality that is similar to yours.
The municipality has been experiencing a drought for weeks. The drought conditions could
bring harm to life, agriculture, and livestock in the municipality due to water shortages. These
conditions could result in lower incomes and limited access to clean water for people in the
municipality.
The ward is represented by a male/female councilor.
He/She was elected by residents of the municipality in the last election. As a councilor, his/her
duties include monitoring the performance of the municipality and representing his/her ward
in the council. Councilors serve for 4 years and then can be reelected.
Successful Drought Response:
Imagine you live in another ward within a municipality that is similar to yours.
The municipality has been experiencing a drought for weeks. The drought conditions could
bring harm to life, agriculture, and livestock in the municipality due to water shortages. These
conditions could result in lower incomes and limited access to clean water for people in the
municipality.
The ward is represented by a male/female councilor.
He/She successfully secured relief funds for the drought. Because he/she secured these funds,
residents received food parcels and water trucks were brought to the municipality, and so the
negative impacts of the drought were lessened.
Failed Drought Response:
Imagine you live in another ward within a municipality that is similar to yours.
The municipality has been experiencing a drought for weeks. The drought conditions could
bring harm to life, agriculture, and livestock in the municipality due to water shortages. These
conditions could result in lower incomes and limited access to clean water for people in the
municipality.
The ward is represented by a male/female councilor.
He/She failed to secure relief funds for the drought. Because he/she did not secure these funds,
residents did not receive food parcels and water trucks were not brought to the municipality,
and so the negative impacts of the drought were not lessened.
In the drought vignettes, we opted to describe the potential dangers of a drought in order to
ensure that all respondents understood why droughts are harmful events. Respondents enter the
survey with their own varied real-world knowledge of and experiences with droughts. We include
information about agricultural damage, detrimental income impacts, and a lack of clean water
access to showcase how droughts can harm all members of the population, not just those with
agricultural ties. In the Successful Drought Response and Failed Drought Response conditions, we
describe two relief efforts that are common in the South African context: supplying water trucks
13
and providing food parcels to constituents.
After reading the vignette, respondents answer questions about their evaluation of the incum-
bent councilor across different dimensions: likelihood of vote, competency, trust, favorability, and
perceived corruptibility. While likelihood of vote most closely captures our main outcome of inter-
est, whether voters punish women incumbents more than men, the additional outcome questions
allow us to also explore other forms of incumbent evaluation that may vary across incumbent
gender.
Respondents are first asked how likely they would be to vote for the councilor in the next elec-
tions and respond using a 5-point likelihood scale. Respondents are then asked about competence,
trustworthiness, and favorability; the order of these three questions is randomized across respon-
dents. Respondents indicate how competent they think the councilor is, how much they trust the
councilor to help people in the municipality, and how favorably they view the councilor. All three
questions use similar 4-point scales for responses. Last, respondents use a 5-point likelihood scale
to indicate how likely they think it is that this councilor will steal public funds while in office.
Respondents then answer an open response question that pipes in their response to the in-
cumbent’s perceived competence: “You previously answered that you think the councilor is [com-
petent/somewhat competent/somewhat incompetent/incompetent]. Why did you evaluate the
councilor in this way?” We include an open response question as a means to elicit evidence of
possible mechanisms for why voters may perceive the different incumbents as more or less compe-
tent. For this same reason, we also ask a second open response question only of respondents who
received one of the drought response vignettes: “In your own words, please write a few sentences
about why you think that the incumbent councilor [succeeded/failed] in securing relief funds for
the drought.”
Respondents then answer manipulation check questions to measure whether they properly re-
ceived the intended treatment. All respondents answer whether the councilor was a man or woman,
and respondents in the drought conditions also answer a question about the incumbent’s drought
response being a success, a failure, or not described. The full survey can be found in Appendix C .
Experimental results
To test our hypotheses, we present differences in means that show the effect of our eight condi-
tions on the outcome measures which capture if respondents evaluate the incumbent negatively
or positively. We show findings for respondents who passed a manipulation check for incumbent
14
gender, which comprises 81.11% of our sample.12. Treatment effects for these respondents reflect a
more precise test of our hypotheses, akin to the average treatment effect among the treated (ATT),
since we can verify that respondents properly received the treatment of incumbent gender. While
we largely focus on the vote likelihood outcome in order to analyze electoral accountability during
drought times, we also present findings for competence and likelihood of stealing funds in the main
text and trustworthiness and favorability in Appendix B, Figures A2 and A3.
Figure 1 presents differences in the mean likelihood of vote across our eight conditions with
95% confidence intervals.13 Mean likelihood of vote is measured on a 5-point scale where 1 equals
unlikely to vote for the incumbent and 5 equals likely to vote for the incumbent in the next election.
The figure presents means for the male incumbent conditions in circles and the female incumbent
conditions in triangles. The x-axis represents the vignette that each condition received: no drought
information, information that there is a drought without any information about the incumbent’s
response, information that there is a drought and the incumbent successfully obtained drought
relief, and information that there is a drought and the incumbent failed to obtained drought relief.
Figure 1: Mean Vote Likelihood by Experimental Condition
Mean Outcome by Condition, 95% CIs
Passed Gender Manipulation Check
5
4
3 Male IncumbentFemale Incumbent
2
1
t t t t f f f f
gh gh gh gh eli
e
eli
e lie ee eli
Dr
ou rou rou rou RD D D s s 
R o R  e o
 R
No No Y Y
e N N
Before considering incumbent gender, our findings first provide important insight into electoral
accountability during drought times. The occurrence of a drought reduces the likelihood that any
12Per our preregistration plan, we tested our hypotheses using our full sample and our sample of individuals who
passed the manipulation check. Findings for the full sample can be found in Appendix B
13Tukey HSD tests and t-tests can be found in Appendix B, Figures A14-A13 and Tables A6-A13.
15
Vote Likelihood
incumbent gets reelected when compared to non-drought times. If an incumbent can successfully
secure drought relief, though, they are largely rewarded by voters. If an incumbent fails to secure
drought relief, the electoral punishment is large: on our scale, the likelihood of vote drops from
somewhat likely to somewhat unlikely. These differences show that voters do respond to drought
conditions in their vote intentions. Electoral accountability for drought relief is highly evident in
our sample for both male and female incumbents, highlighting voters’ ability to react to drought
information and related incumbent performance.
Turning to how respondents evaluate incumbents during droughts in a gendered way, several
important differences. First, our findings reflect a baseline preference for women incumbents in
non-drought conditions: respondents have a significantly higher vote likelihood for the incumbent
in the female condition. This finding reflects other experimental work on elite gender that shows a
common preference for women candidates in hypothetical survey settings (Schwarz and Coppock,
2022). When respondents have information about a drought but do not know about how the
incumbent responded to the drought, though, they express nearly equal likelihood of voting for
male and female incumbents, which contradicts our first hypothesis that men should be evaluated
more positively by voters in drought times. For both female and male incumbents, the occurrence
of a drought reduces their electoral support.
When considering Conditions 4 through 8 where respondents read about incumbent responses
to the drought, we find mixed support for our expectations. When an incumbent successfully
provides relief funds for the drought, respondents reward a female incumbent for this behavior
significantly more than a male incumbent. We expected that men and women would be evaluated
similarly in the successful response conditions because their ability to deliver relief successfully
would neutralize gender stereotypes that work against female incumbents’ favor (H3), but instead
women are rewarded significantly more for this success. The female incumbent condition has a
mean vote likelihood of 4.34 while the male incumbent has a mean vote likelihood of 3.89. When
a female incumbent can provide drought relief to constituents, she enjoys a greater electoral boost
for that performance than a male counterpart.
However, our findings show that respondents are also more punishing of female incumbents.
In the relief failure condition, we find support for our second hypothesis: women incumbents
are punished more for failing to provide drought relief than male incumbents. Vote likelihood
drops lower for a woman who fails to provide relief than it does for a man, and this difference
in means achieves statistically significance (p=0.01). Women face harsher electoral accountability
for their performance. Given that women were also rewarded more than men for obtaining relief,
this finding demonstrates that women who perform poorly during a drought ”fall harder´´ than
16
men when compared to their vote likelihood during a good performance as well as during drought
times without any performance information. The female incumbent experiences a drop in mean
vote likelihood of 2.67 (4.34 to 1.67), while the male incumbent experiences a drop in mean vote
likelihood of 1.93 (3.89 to 1.96). For women, failing to provide drought relief generates a larger
punishment than it does for men.
Perceived competence of the incumbent paints a similar picture. Figure 2 shows that women
are perceived as more competent than men in non-drought times and especially when they provide
drought relief. In drought times without performance information, the mean estimate for the
female incumbent’s competence is slightly higher than the male incumbent’s, which is akin to
the competence outcome in the full sample, but the difference in these means does not achieve
statistical significance at the 0.05 level. When considering incumbents who did not get relief,
women incumbents again experience a much greater drop in perceived competence. Respondents
perceive the failing female incumbent in to be less competent than the failing male incumbent,
which reflect the results for vote likelihood in Figure 1. Results for perceived trustworthiness and
favorability in Appendix B are similar: the female incumbent received a more positive evaluation
when she acquired relief successfully, but she also received a more negative evaluation than the
male incumbent when she failed to secure relief funds for the drought.
Figure 2: Mean Incumbent Competence by Experimental Condition
Mean Outcome by Condition, 95% CIs
Passed Gender Manipulation Check
3.5
3
2.5 Male Incumbent
Female Incumbent
2
1.5
ht t t t f f f f
ug ug
h
ug
h gh lie lie lieu e li
e
ro ro ro ro s R s R
e Re Re  
 D  D D D e e No No
No No Y Y
Respondents evaluate the female incumbent more harshly in contexts of poor performance even
17
Competence
though they also evaluate the female incumbent more positively in contexts of good performance.
While this conclusion supports our second hypothesis that women should be punished more than
men for failing to respond to a drought, women are not preferred less than men during drought
times without any incumbent response information, and women are rewarded more than men for
good drought performance which contradicts our third hypothesis. Overall, the findings in this
sample tell a story where the female incumbent is viewed as more agentic than the male incumbent:
she is rewarded more for her success but also punished more for her failure. Interestingly, this
finding runs counter to the findings in De Geus et al. (2021), which show that voters treat male
incumbents as more agentic in the U.S. and Australia, thereby punishing female incumbents less for
poor performance but also rewarding them less for favorable performance. In contexts of drought
in South Africa, it is women politicians who present as more agentic to voters.
Overall, our findings reveal that female incumbents facing droughts in South Africa do not face
consistently harsher electoral accountability for their performance in office than male incumbents.
In fact, voters may even electorally reward women more than men in drought times, particularly
when they can successfully deliver drought relief. When women fail to deliver drought relief,
though, they are evaluated more negatively than failing men. These findings show how a female
incumbent who performs poorly during a drought may face a greater level of voter backlash for
her failures than a male incumbent. When she performs well, though, she may receive a higher
level of voter support than a male incumbent who performs equally well, suggesting that voters in
South Africa react to female incumbents’ performances more intensely (in positive and negative
ways) than they react to male incumbents’ performances.
Appendix B presents several exploratory analyses outlined in our preregistration plan that test
for heterogeneous treatment effects across respondent sex (A11), respondents’ real-life experiences
with droughts (A9), respondents’ urban vs. rural living conditions (A10), and respondent parti-
sanship (A12). Respondent sex could theoretically impact how voters evaluate male versus female
incumbents due to different stereotypes and expectations held by men and women. Respondent ex-
perience with droughts and urban versus rural living conditions could also impact how respondents
evaluate political performance during drought times. Drought conditions in a city, for example,
could feel less pressing than drought conditions in rural communities. Finally, partisanship could
influence how respondents evaluate incumbents in our design given that the ANC has a voluntary
gender quota, though it is important to note that the percentage of female candidates at the lo-
cal level is similar between the ANC and DA. Nevertheless, it is possible voters associate female
incumbents with the ANC more than with the DA. Overall, exploratory analyses reveal no or
small heterogeneous treatment effects across different subgroups. For example, the difference in
18
punishment of female vs. male incumbents who fail to acquire drought relief is slightly greater for
respondents who report they have not lived through a drought, for urban-dwelling respondents,
and for male respondents. DA partisans are overall more punishing of incumbents who fail to
get relief than ANC partisans, which may suggest that DA respondents assume the incumbents
they read about belong to the ANC due to their political dominance. Largely, though, subgroup
differences are small and follow the direction of the findings presented in the main analyses.
Mechanism: gendered corruption perceptions
To better understand why the female incumbent’s performance illicit stronger reactions from voters
during a drought, we explore other survey items that may shed light on voters’ gendered perceptions
of incumbents. One potential explanation, particularly in the context of drought relief in South
Africa, could be voters’ expectations of the incumbent’s corruptibility or honesty, a politician
quality that is often perceived by voters in gendered ways. Numerous studies show that voters
often believe women to be less corrupt than their male counterparts due to gender stereotypes
that paint women as more honest and risk-averse than men and due to women’s political outsider
status whereby they cannot access the networks necessary to behave corruptly like more embedded
men (Barnes and Beaulieu, 2014; Barnes, Beaulieu, and Saxton, 2018, 2020; Esarey and Schwindt-
Bayer, 2018; Bauhr and Charron, 2021). Voters in South Africa may express different expectations
about a male incumbent’s likelihood of misappropriating funds compared to a female incumbents’
likelihood of doing so, and these expectations may lead voters to evaluate an incumbent’s good
and poor performance differently based on if the incumbent is a man or woman.
Our final outcome question allows us to explore this possibility. We asked respondents to state
their believed likelihood that the incumbent would steal public funds. Figure 3 shows differences
in means across conditions for the likelihood of stealing outcome. Respondents used a 5-point
likelihood scale to answer, “How likely do you think it is that this councilor will steal public funds
while in office?” Five represents likely to steal, and one represents unlikely to steal, so higher means
indicate a higher perceived likelihood of corrupt behavior.
19
Figure 3: Mean Likelihood of Incumbent Stealing Funds by Experimental Condition
Mean Outcome by Condition, 95% CIs
Passed Gender Manipulation Check
4.5
4
3.5 Male IncumbentFemale Incumbent
3
2.5
gh
t ht ht ht
u ug ug ug eli
ef ief ief fel el li
e
o o er r ro ro  R  R  R  R
o D  D D D
s s o
o Ye Ye N N
o
N N
In the no drought, drought, and no relief conditions, the male and female incumbent are
perceived as equally likely to steal funds during their tenure. Taking the no drought and drought
conditions into account especially, respondents do not seem to have a baseline expectation that
women are less corrupt. In the relief conditions where the incumbent did secure relief funds for
the drought, the male incumbent is perceived as significantly more likely to steal funds. In this
condition, the likelihood of stealing is lowest because respondents are told that the incumbent put
the funds to use properly in order to provide water trucks and food parcels to the ward. However,
the significant difference between the male and female incumbent shows that, despite providing
drought relief, respondents have slightly higher suspicion of the male incumbent than the female
incumbent. Appendix B Figure A8 shows that, for the full sample, respondents believe men and
women incumbents to be equally likely to steal funds in all conditions. Overall, this outcome does
not strongly suggest that respondents believe that all female incumbents are less corrupt than male
incumbents, but the higher likelihood of stealing for the male incumbent in the successful relief
condition may indicate that respondents believe men are more likely to “skim a bit off the top” of
a pot of funds for themselves.14 Interestingly, respondents in the failed relief condition have the
14This generally mirrors insights from the background interviews in our fieldwork in the Limpopo province in 2022,
where interviewees explained in detail how they view male local leaders as more likely to steal drought relief aid.
For example, one interviewee noted the following: ”Unlike male leaders who will serve themselves or use drought
relief monies to feed their families and friends, women leaders will deliver services. Women will put in a lot of effort
to support the neighborhood in difficult times, offer aid, and develop drought-resistance plans.”
20
Likelihood of Stealing Funds
highest belief that the incumbent is likely to steal funds regardless of gender, which may suggest
that respondents assume the failure to acquire relief funds and distribute drought aid to the people
is a result of corrupt behavior.
Open response questions in our survey that ask respondents to explain why they evaluated the
incumbent’s competency as they did further demonstrate the role corruption perceptions played
in respondents’ reactions to the vignettes. Respondents in the relief conditions were also asked to
explain why they believe the incumbent failed or succeeded in acquiring drought relief. Many re-
spondents remark that women councilors are less corrupt than men. Several exemplary responses15
showcase this mindset:
• “[Be]cause sometimes women are less corrupt than men.”
• “Women are less likely to be involved in corruption [than] males.”
• “In my personal experience in life, I have come to know that women are more compassionate
and empathetic when compared to men. This lessens the possibility of the councilor being a
corrupt individual, and actually a person with decent moral values and a vision for a brighter
future.”
• “All the candidates are usually incompetent and her male counterparts would probably be
worse than her so I’d be hoping that the fact that she’s female would make her better or less
of a thief.”
• “She’s a woman, there are less incidents where females are involved in corruption[;] that’s
just my opinion and what I have been made aware of.”
• “Mostly the council[ors] in my community are male. They haven’t done any good in the com-
munity, they just use the government funds for their benefits so I believe a female council[or]
could do better.”
• “She is a woman and most women are fair enough.”
Respondents evaluate female councilors as more competent because they assume they are less
likely to be corrupt than male councilors. Some respondents also provide this logic to explain why
the female councilor succeeded at getting relief funds, implying that male councilors would be more
likely to steal funds and therefore fail to provide relief.
It is important to note that some respondents assigned to a female councilor condition remark
that they think men make better political leaders than women do; however, these responses focus
15Bracketed text represents edits made to the text by the authors to correct respondents’ typos for clarity.
21
on men having more leadership skills and opportunities than women, not on men being less corrupt
or more honest than women. In these answers, respondents state plainly that they believe men
are better leaders and women are not suited for politics, reflecting traditionally sexist views about
the role of women in society. It is also important to note that many respondents state that they
believe all politicians are corrupt. The belief that women are less corrupt than men is not universal.
However, our open responses reflect that many respondents do associate women with more honest,
fair politics than men.16
Study 2: Observational analysis
In an attempt to attest to the external validity and generalizability of the experimental evidence,
we also analyze observed incumbency performance in South African local elections between 2006-
2021. To emulate the experimental approach, we combine data on ward-level candidate gender,
incumbency status, real-world drought conditions, and official municipality-level drought-related
emergency declarations – which we conceive as as a proxy for the ability to get aid.
First, we calculate incumbency measures for all ward candidates between 2006-2021. These
data, from candidate registers provided by the South African Electoral Commission, contain infor-
mation on candidate gender and the political party to which they belong. To emulate the exper-
imental vignettes as much as possible, our primary incumbency stipulation relates to whether a
candidate won the same ward seat in the previous election.17 We identify 3827 such cases between
2006-2021 (24% of whom are women). As we are interested in electoral performance among ward
incumbents, our analyses are limited to these individuals.
Data on drought declarations were collected by compiling all government and provincial gazettes
related to disaster declarations between 2002-2024.18 Specifically, gazettes were initially screened
for keywords ”disaster”, ”disaster declaration act”, ”declare” and ”declaration” and subsequently
compiled into a database of disaster declaration gazettes. Further information about the disaster
declarations were then recorded including the type of disaster and the regional extent of the
declaration. As we are primarily interested in the ability of local incumbents to get aid for their
constituencies via disaster declarations, we focus our analysis on municipal declarations – the most
local-level of declaration possible – as these relate more to localized performance and accountability.
16Additionally, some open responses state that women are more competent than men due to reasons other than
corruption such as women being more nurturing, more community-oriented, and more likely to work hard than men.
For example, one respondent wrote, “Women are sympathetic unlike men[,] they think for the future.” Multiple
explanations center around women being committed to doing a better job than men, i.e. “Women like to do their
work perfect.”
17In additional tests reported in Appendix Table A2, we use alternative operationalizations of incumbency that
relate to municipal councilors and mayors, rather than the ward incumbent.
18We would like to thank Paul Bederke for his work in collecting and compiling data on disaster declarations.
22
However, we consider provincial declarations as a control variable (where aid may be accessed by
local councils though without a clear incumbency element regarding municipal politicians). The
result of this process is a measure which indicates whether, for each municipality, a drought was
declared or not in the 12-months prior to an election.
To measure geological drought conditions, we rely primarily on the Standardized Precipitation-
Evapotranspiration Index (SPEI): a global dataset of 0.5 degree geo-grids that approximates
drought conditions (as a combination of precipitation levels and temperature) on a monthly time-
series.19 The standardization of the SPEI means that values range from -3 to 3, with negative
values meaning dryer conditions and positive values meaning wetter conditions. We merge SPEI
grids with a time-variant measure of the primary crop grown in a given location (Jägermeyr et al.,
2021) as well as the months of that crop’s growing season in that same location (IFPRI, 2025).
These indicators are then aggregated to the electoral ward level. As a result, we are able to con-
struct a measure of whether drought conditions have occurred in the growing season of the primary
regional crop in the year prior to each election between 2006-2021. Specifically, we consider drought
onset as whether there have been two consecutive growing season months in a given location with
SPEI levels of below -1.5.20 The geographic and temporal distribution of the drought conditions
and declarations measures are illustrated in Figure 4.In total, this method records 6655 instances
of electoral wards experiencing a drought in the year preceding an election across four elections.
Of these, 272 (4.1%) resulted in a formal declaration of drought. The rate of disaster declaration
is higher for women than for men, however, as 5.4% of female incumbents in drought-affected ares
experienced drought declarations, compared to 3.6% of male incumbents.
We combine these data with two measures of electoral performance. First, we consider a simple
measure of whether the incumbent won re-election or not. This is a common outcome used in
incumbency effect studies. The mean re-election rate for incumbents in our sample is 0.78 (see
Appendix A.1. Secondly, we consider the change in vote share relative to the previous election.
This measure more precisely estimates the reward/punishment of an incumbent, even if they win
re-election. This measure is particularly useful in less competitive electoral contexts where one
party is dominant, as is the case in many regions in South Africa.21
19https://spei.csic.es/
20We also consider variations of this measure along both dimensions. Specifically, in Appendix A.2 we vary SPEI
levels (drought severity) between -1 and -1.5, and the number of months (drought duration) between 2 and 3.
21Descriptive statistics for all variables used in the observational analysis, broken down by gender, can be found
in Appendix A.1.
23
Figure 4: Drought conditions and official municipal (and provincial) declarations of drought by election
Note: This plot illustrates the spatial distribution of drought conditions (top row) and drought declarations (bottom
row) in the 12 months prior to South African local elections, 2006-2021. Boundaries represented are at the local
ward level. See main text for precise explanations of measurements.
Observational results
We estimate a series of linear regression models using two-way fixed effects to test the effect
of candidate sex and drought declarations on incumbent electoral performance. The results are
presented in Table 3. The dependent variable in models 1-3 is whether the incumbent won re-
election or not. The dependent variable in models 4-6 is the change in vote share relative to the
previous election: negative values indicate worse performance. The first set of models control for
the competitiveness of the ward, measured as the difference (in percentage points) between the
incumbent and the second-placed candidate in the previous election. The second set of models
include a control for the incumbent’s vote share in the previous election. All models control for
whether a drought was declared at the provincial level in the same electoral term and whether a
disaster of any other kind was declared in that time period. In models 2-3 and 5-6 the sample is
limited to incumbents in wards that actually experienced drought conditions in the electoral term,
measured as a dummy variable of whether two consecutive months accrued where precipitation
levels dropped below -1.5 in the SPEI Index during the growing season of the most common crop
in a given ward. All models include municipality and party-election fixed effects.
The results, presented in Table 3, suggest that female incumbents are no more or less likely to
win re-election than their male counterparts, on average (Models 1–3). However, they do experience
a greater electoral penalty in terms of vote share: across Models 4–6, female incumbents receive
24
Table 3: Observational results
DV: Incumbent win DV: ∆ vote share
(1) (2) (3) (4) (5) (6)
Female candidate 0.003 -0.013 -0.021 -0.011* -0.027*** -0.031***
(0.014) (0.022) (0.023) (0.006) (0.009) (0.009)
Drought declaration -0.135** 0.015
(0.055) (0.038)
Female cand. x Drought dec. 0.172*** 0.071**
(0.051) (0.030)
Ward competition 0.320*** 0.300*** 0.306***
(0.032) (0.047) (0.048)
Vote share (lagged) -0.448*** -0.459*** -0.454***
(0.028) (0.040) (0.041)
Provincial drought dec. -0.055 -0.029*
(0.038) (0.018)
Other disaster dec. -0.007 -0.021
(0.028) (0.016)
Municipality FE Yes Yes Yes Yes Yes Yes
Party-year FE Yes Yes Yes Yes Yes Yes
Drought conditions No Yes Yes No Yes Yes
Num.Obs. 3496 1279 1279 3812 1377 1377
R2 0.382 0.469 0.472 0.623 0.674 0.676
R2 Within 0.058 0.058 0.063 0.151 0.164 0.171
Notes: The table reports liner probability model (models 1-3) and ordinary least squares (models 4-6)
estimates. The dependent variable in models 1-3 is whether and incumbent won re-election or not
(0-1). In models 4-6, the dependent variable is the incumbents vote share - their vote share in the
previous election. * p ă 0.1, ** p ă 0.05, *** p ă 0.01
approximately 2 to 3 percentage points less in vote share compared to male incumbents, all else
equal. This suggests that women may contest less competitive districts where a reduction in vote
share may not necessarily translate to electoral turnover.
When a drought is declared within 12 months of an election, Model 3 suggests that the proba-
bility of re-election for male incumbents relative to females (the direct effect of drought declaration)
drops sharply by around 13.5 percentage points. In contrast, female incumbents facing a drought
are rewarded relative to men: they are 17.2 percentage points more likely to win (Model 3) and
enjoy a 7.1 percentage point boost in vote share (Model 6). These interaction effects demonstrate
an effective reversal of the gendered incumbency punishment observed under normal conditions.
By adding the coefficient of the interaction term to the coefficient of the direct drought declara-
tion term, we can calculate the total effect of drought declarations for female incumbents as a 3.7
25
percentage point increase in the likelihood of winning and approximately a 5.6 percentage point
gain in vote share. In a similar vein to the experimental evidence presented above, drought dec-
larations appear to mitigate and even reverse the typical electoral punishment imposed on female
incumbents.
Conclusion
Despite numerous studies exploring the impact of weather events and natural disasters on electoral
accountability and incumbent evaluation, the role of incumbent gender during such events remains
understudied. Borrowing from literature on the gendered nature of electoral accountability and
women’s representation in times of crisis, we hypothesized that during a drought—a weather
event that can severely harm affected populations—female incumbents will face harsher electoral
accountability and more negative evaluations than male incumbents.
Experimental and observational analyses provide evidence from South Africa that, contrary
to our expectations, female incumbents do not always face harsher voter evaluations than male
incumbents in drought times. Our survey experiment shows that when a female incumbent can
provide drought relief to her constituents, she is evaluated more positively than a male incum-
bent who does the same. However, when female incumbents fail to provide drought relief-and in
situations where official droughts are not declared-they are punished for this failure more harshly
than male incumbents. This greater punishment results in a lower vote likelihood for female in-
cumbents during poor performance. Since women also have a higher vote likelihood during good
performance, this harsher evaluation during poor performance means that women who fail to get
relief ”fall harder” in the eyes of voters than men. In other words, the difference in vote likelihood
between a woman who succeeds and a woman who fails to get relief is larger than the difference in
vote likelihood between a man who succeeds and a man who fails to get relief. While women enjoy
greater electoral support than men when drought relief is acquired, they also suffer from a greater
electoral punishment than men do when relief cannot be delivered. Voters react to incumbent
women’s performance in more extreme ways in both directions, reward and punishment.
With our observational analysis, we show that our experimental findings translate into real-
world electoral behavior. We approximate our experimental design with novel data on drought
declarations, our best available proxy for drought relief resources described in our experiment;
data on objective drought conditions via geo-coded precipitation levels and crop growth measures;
and electoral data for ward councilors in South Africa. We show that droughts are damaging to
incumbents’ reelection prospects, especially for women. However, if female incumbents can issue
26
a drought declaration, the necessary process to receive relief resources from central government
authorities, they significantly boost their probability of winning reelection and of increasing their
vote share. Importantly, this boost occurs for female incumbents specifically; a drought declaration
is not universally beneficial for male and female incumbents’ electoral performance. These results
mirror our experiment, adding robustness to our findings that women are rewarded more than
men for good performance in drought times. Importantly, these results also show that the voter
intentions measured in a hypothetical experimental setting do indeed travel into the real world.
Our findings suggest that women officeholders do not necessarily face consistent gendered ob-
stacles to maintaining power in the face of a weather crisis like a drought; in various drought
scenarios, women are actually preferred by voters. Thus, in drought times, women incumbents
may be able to hold onto office at a higher rate than men incumbents. This ability, though, comes
with a strong caveat that incumbents must perform well during a drought. This finding has the
potential for especially problematic outcomes for women incumbents if acts such as acquiring relief
aid for droughts are impacted by incumbent gender. For example, if women incumbents are sys-
tematically less likely to successfully secure relief funds for a drought than their male counterparts
due to hindrances such as institutional barriers to success or other political gatekeepers’ gender
biases against women, and women are punished more stringently than men for failing to get re-
lief, then women may lose their seats at a greater rate than men. Future research on women’s
representation and voter accountability in times of drought could examine the possibility that an
incumbent’s ability to deliver disaster aid may be hindered by gendered marginalization and bi-
ases within the institutions and political channels needed to effectively mitigate weather crises. A
further implication from the findings pertains to our understanding of relief preparedness. Women
incumbents should have higher incentives than men to maintain a level of crisis preparedness and
to be equipped to respond to such crises given the higher penalty they face for failing to address
the impact of such events.
Future research can also expand this study into other contexts to test the generalizability of
these findings. A context like South Africa has relatively low levels of women’s representation at
the local level and high rates of corruption, especially around relief funds, and these factors could
influence the gender biases that voters bring with them to the polls. We expect that our findings
may replicate in similar contexts, but voters may evaluate women incumbents differently in contexts
where women are more present in office. Additionally, what constitutes a good performance in the
face of an environmental disaster may differ where corruption levels are lower and voters have
higher expectations of government institutions and officials to deliver necessary aid. Our study
also focuses on one level of office, ward councilor. While ward councilors have the responsibility to
27
aid their municipalities in drought times, political actors at other levels of office undoubtedly play
a role in drought relief as well. Our results may travel across office levels, but it is also possible
that respondents are more (or less) punishing and rewarding towards higher levels of office, and
these differences could take on a gendered nature, too.
Along with limitations of generalizability, our study also has other important limitations that
may restrict how accurately our experimental findings translate to real-world observations of elec-
toral accountability in the face of droughts. We use drought declarations as a proxy for securing
drought aid in our observational analysis because this data is the best measure available to us to
approximate relief in the South African context, where data on actual monetary resources awarded
for drought purposes is not consistently available at the municipal level. Though the similar find-
ings across our studies give us confidence that these declarations area good proxy for relief aid, it
is possible that an observational analysis that has a precise measure of drought relief, such as the
value of food parcels distributed to a community, could produce different results.
Last, it is important to note that our vignettes intentionally exclude other political factors
like partisanship, race, and ethnicity of the incumbent in order to focus on the effect of gender as
cleanly as possible, but real-world candidates undoubtedly have these factors and more working for
or against them when voters evaluate their performances. Effects of gender may therefore be dulled
when voters also consider party affiliation or race, or gender effects may only be present when in
combination when other incumbent traits. Our analysis of observational data provides evidence
that incumbent gender does matter even outside of the controlled, hypothetical environment of
a survey experiment, and we show that electoral outcomes do indeed differ for men and women
officeholders in real-world drought scenarios. However, gender is just one aspect of incumbent
identity that may impact electoral accountability during environmental crises. Additional studies
that seek to alter more incumbent traits as well as more weather event scenarios and relief speci-
ficities could add valuable insight into how voters’ behaviors and preferences may change during
crises, particularly weather events and natural disasters.
28
References
Abney, F. Glenn, and Larry B. Hill. 1966. “Natural Disasters as a Political Variable: The Effect
of a Hurricane on an Urban Election.” American Political Science Review 60(4): 974–981.
Afrobarometer. 2024. http://www.afrobarometer.org “Afrobarometer.”.
Arceneaux, Kevin, and Robert M. Stein. 2006. “Who Is Held Responsible When Disaster Strikes?
the Attribution of Responsibility for a Natural Disaster in an Urban Election.” Journal of Urban
Affairs 28(1): 43–53.
Arora-Jonsson, Seema. 2011. “Virtue and vulnerability: Discourses on women, gender and climate
change.” Global Environmental Change 21(2): 744–751.
Bahta, Y.T., A. Jordaan, and F. Muyambo. 2016. “Communal farmers’ perception of drought in
South Africa: Policy implication for drought risk reduction.” International Journal of Disaster
Risk Reduction 20: 39–50.
Barnes, Tiffany D., and Emily Beaulieu. 2014. “Gender Stereotypes and Corruption: How Candi-
dates Affect Perceptions of Election Fraud.” Politics & Gender 10(3): 365–391.
Barnes, Tiffany D., Emily Beaulieu, and Gregory W. Saxton. 2018. “Restoring trust in the police:
Why female officers reduce suspicions of corruption.” Governance 31(1): 143–161.
Barnes, Tiffany D., Emily Beaulieu, and Gregory W. Saxton. 2020. “Sex and corruption: how
sexism shapes voters’ responses to scandal.” Politics, Groups, and Identities 8(1): 103–121.
Bauer, Gretchen, and Manon Tremblay, eds. 2011. Women in Executive Power. 0 ed. Routledge.
Bauer, Nichole M. 2015. “Emotional, Sensitive, and Unfit for Office? Gender Stereotype Activation
and Support Female Candidates.” Political Psychology 36(6): 691–708.
Bauhr, Monika, and Nicholas Charron. 2021. “Will Women Executives Reduce Corruption?
Marginalization and Network Inclusion.” Comparative Political Studies 54(7): 1292–1322.
Bechtel, Michael M., and Jens Hainmueller. 2011. “How Lasting Is Voter Gratitude? An Analysis of
the Short- and Long-Term Electoral Returns to Beneficial Policy.” American Journal of Political
Science 55(4): 852–868.
Benegal, Salil, and Mirya R. Holman. 2025. “Hostile sexism and overconfidence shape the public’s
views of climate change.” Climatic Change 178(8): 153.
29
Blankenship, Brian, Ryan Kennedy, Johannes Urpelainen, and Joonseok Yang. 2021. “Barking Up
the Wrong Tree: How Political Alignment Shapes Electoral Backlash from Natural Disasters.”
Comparative Political Studies 54(7): 1163–1196.
Blanton, Robert, Shannon Blanton, and Dursun Peksen. 2019. “The Gendered Consequences of
Financial Crises: A Cross-National Analysis.” Politics & Gender 15(4): 941–970.
Brulé, Rachel. 2023. “Climate Shocks and Gendered Political Transformation: How Crises Alter
Women’s Political Representation.” Politics & Gender 19(3): 928–934.
Burke, Marshall, Erick Gong, and Kelly Jones. 2015. “Income Shocks and HIV in Africa.” The
Economic Journal 125(585): 1157–1189.
Bush, Sarah Sunn, and Amanda Clayton. 2023. “Facing Change: Gender and Climate Change
Attitudes Worldwide.” American Political Science Review 117(2): 591–608.
Bélanger, Éric, and Bonnie M. Meguid. 2008. “Issue salience, issue ownership, and issue-based
vote choice.” Electoral Studies 27(3): 477–491.
Carlin, Ryan E., Gregory J. Love, and Elizabeth J. Zechmeister. 2014. “Natural Disaster and
Democratic Legitimacy: The Public Opinion Consequences of Chile’s 2010 Earthquake and
Tsunami.” Political Research Quarterly 67(1): 3–15.
Cassese, Erin C., and Mirya R. Holman. 2018. “Party and Gender Stereotypes in Campaign
Attacks.” Political Behavior 40(3): 785–807.
Clayton, Amanda, Boniface Dulani, Katrina Kosec, and Amanda Lea Robinson. 2025. “Represen-
tation increases women’s influence in climate deliberations: Evidence from community-managed
forests in Malawi.” American Journal of Political Science , ajps.12994.
Cole, Shawn, Andrew Healy, and Eric Werker. 2012. “Do voters demand responsive governments?
Evidence from Indian disaster relief.” Journal of Development Economics 97(2): 167–181.
Corno, Lucia, Nicole Hildebrandt, and Alessandra Voena. 2020. “Age of Marriage, Weather Shocks,
and the Direction of Marriage Payments.” Econometrica 88(3): 879–915.
Corruption Watch. 2020. https://www.corruptionwatch.org.za/
weak-public-governance-enables-water-sector-corruption/ “Weak public governance
enables water sector corruption.”.
Costa, Mia. 2021. “He Said, She Said: The Gender Double Bind in Legislator–Constituent Com-
munication.” Politics & Gender 17(4): 528–551.
30
Daniels. 2023. “Clamping Down on Corruption: The Battle in South Africa’s Water Sector.” Cape
Town Today .
De Geus, Roosmarijn A., John R. McAndrews, Peter John Loewen, and Aaron Martin. 2021.
“Do Voters Judge the Performance of Female and Male Politicians Differently? Experimental
Evidence from the United States and Australia.” Political Research Quarterly 74(2): 302–316.
Demirdogen, Alper, and Emine Olhan. 2025. “Fifty-four thousand deaths, zero electoral impact.”
Public Choice .
Denton, Fatma. 2002. “Climate change vulnerability, impacts, and adaptation: Why does gender
matter?” Gender & Development 10(2): 10–20.
Dolan, Kathleen. 2010. “The Impact of Gender Stereotyped Evaluations on Support for Women
Candidates.” Political Behavior 32(1): 69–88.
Eagly, Alice H., Christa Nater, David I. Miller, Michèle Kaufmann, and Sabine Sczesny. 2020.
“Gender stereotypes have changed: A cross-temporal meta-analysis of U.S. public opinion polls
from 1946 to 2018.” American Psychologist 75(3): 301–315.
Egan, Mark, Gregor Matvos, and Amit Seru. 2022. “When Harry Fired Sally: The Double Standard
in Punishing Misconduct.” Journal of Political Economy 130(5): 1184–1248.
Eggers, Andrew C., Nick Vivyan, and Markus Wagner. 2018. “Corruption, Accountability, and
Gender: Do Female Politicians Face Higher Standards in Public Life?” The Journal of Politics
80(1): 321–326.
Electoral Commission of South Africa. 2024. https://www.elections.org.za/pw/StatsData/
List-Of-Current-Ward-Councillors “Ward councillors statistics - Electoral Commission of
South Africa.”.
Epstein, Adrienne, Eran Bendavid, Denis Nash, Edwin D. Charlebois, and Sheri D. Weiser. 2020.
“Drought and intimate partner violence towards women in 19 countries in sub-Saharan Africa
during 2011-2018: A population-based study.” PLOS Medicine 17(3): e1003064.
Eriksson, Lina M. 2016. “Winds of Change: Voter Blame and Storm Gudrun in the 2006 Swedish
Parliamentary Election.” Electoral Studies 41: 129–142.
Esarey, Justin, and Leslie A. Schwindt-Bayer. 2018. “Women’s Representation, Accountability and
Corruption in Democracies.” British Journal of Political Science 48(3): 659–690.
31
Fox, Richard L., and Jennifer L. Lawless. 2004. “Entering the Arena? Gender and the Decision to
Run for Office.” American Journal of Political Science 48(2): 264–280.
Fukumoto, Kentaro, and Kyosuke Kikuta. 2024. “After a Storm Come Votes: Identifying the
Effects of Disaster Relief on Electoral Outcomes.” Political Behavior 46(4): 2357–2377.
Gasper, John T., and Andrew Reeves. 2011. “Make It Rain? Retrospection and the Attentive
Electorate in the Context of Natural Disasters: MAKE IT RAIN.” American Journal of Political
Science 55(2): 340–355.
Gomez, Brad, and Matthew Wilson. 2008. “Political Sophistication and Attributions of Blame in
the Wake of Hurricane Katrina.” Publius 38(4): 633–650.
Healy, Andrew, and Neil Malhotra. 2009. “Myopic Voters and Natural Disaster Policy.” American
Political Science Review 103(3): 387–406.
Heersink, Boris, Jeffery A. Jenkins, Michael P. Olson, and Brenton D. Peterson. 2022. “Nat-
ural Disasters, ‘Partisan Retrospection,’ and U.S. Presidential Elections.” Political Behavior
44(3): 1225–1246.
Hessami, Zohal, and Temurbek Khasanboev. 2024. “Gender Bias in the Reelection of Politicians
(When a Crisis Strikes).” IZA Discussion Papers, .
Holman, Mirya R., Jennifer L. Merolla, and Elizabeth J. Zechmeister. 2016. “Terrorist Threat,
Male Stereotypes, and Candidate Evaluations.” Political Research Quarterly 69(1): 134–147.
Huddy, Leonie, and Nayda Terkildsen. 1993a. “The Consequences of Gender Stereotypes for
Women Candidates at Different Levels and Types of Office.” Political Research Quarterly
46(3): 503.
Huddy, Leonie, and Nayda Terkildsen. 1993b. “Gender Stereotypes and the Perception of Male
and Female Candidates.” American Journal of Political Science 37(1): 119.
IDEA. 2024. https://www.idea.int/publications/catalogue/
womens-political-participation-africa-barometer-2024. “Women’s Political Par-
ticipation: Africa Barometer 2024.”.
IFPRI, International Food Policy Research Institute. 2025. https://dataverse.harvard.edu/
dataset.xhtml?persistentId=doi:10.7910/DVN/SWPENT “Global Spatially-Disaggregated
Crop Production Statistics Data for 2020 Version 2.0.”.
32
Itzkovitch-Malka, Reut. 2024. “How do gender stereotypes about leadership positions change in
the face of a crisis?” Research & Politics 11(4): 20531680241293602.
Jones, Philip Edward. 2014. “Does the Descriptive Representation of Gender Influence Account-
ability for Substantive Representation?” Politics & Gender 10(2): 175–199.
Jägermeyr, J., C. Müller, A. C. Ruane, J. Elliott, J. Balkovič, O. Castillo, B. Faye, I. Foster, C.
Folberth, J. A. Franke, K. Fuchs, J. R. Guarin, J. Heinke, G. Hoogenboom, T. Iizumi, A. K.
Jain, D. Kelly, N. Khabarov, S. Lange, T.-S. Lin, W. Liu, O. Mialyk, S. Minoli, El J. Moyer, M.
Okada, M. Phillips, C. Porter, S. S. Rabin, C. Scheer, J. M. Schneider, J. F. Schyns, R. Skalský,
A. Smerald, T. Stella, H. Stephens, H. Webber, F. Zabel, and C. Rosenzweig. 2021. “Climate
impacts on global agriculture emerge earlier in new generation of climate and crop models.”
Nature Food 2: 873–885.
Kaslovsky, Jaclyn, and Jon C. Rogowski. 2022. “Under the Microscope: Gender and Accountability
in the US Congress.” American Political Science Review 116(2): 516–532.
Lawless, Jennifer L., Jennifer L. Lawless, Richard Logan Fox, and Jennifer L. Lawless. 2010. It still
takes a candidate: why women don’t run for office. Rev. ed ed. New York: Cambridge University
Press.
Lazarev, Egor, Anton Sobolev, Irena Soboleva, and Boris Sokolov. 2014. “TRIAL BY FIRE:
A Natural Disaster’s Impact on Support for the Authorities in Rural Russia.” World Politics
66(4): 641–668.
Masiero, Giuliano, and Michael Santarossa. 2021. “Natural disasters and electoral outcomes.”
European Journal of Political Economy 67: 101983.
Meza, Isabel, Ehsan Eyshi Rezaei, Stefan Siebert, Gohar Ghazaryan, Hamideh Nouri, Olena
Dubovyk, Helena Gerdener, Claudia Herbert, Jürgen Kusche, Eklavyya Popat, Jakob Rhyner,
Andries Jordaan, Yvonne Walz, and Michael Hagenlocher. 2021. “Drought risk for agricultural
systems in South Africa: Drivers, spatial patterns, and implications for drought risk manage-
ment.” Science of The Total Environment 799: 149505.
Moreno, Jenny, and Duncan Shaw. 2018. “Women’s empowerment following disaster: a longitudinal
study of social change.” Natural Hazards 92(1): 205–224.
Murray, Rainbow. 2010. Cracking the Highest Glass Ceiling. Bloomsbury Publishing.
33
National Disaster Management Centre. 2019. https://www.gov.za/sites/default/files/gcis_
document/202003/guideline-classification-disaster-and-declaration-state-disaster.
pdf “Guideline - Classification of a Disaster and the Declaration of a State of Disaster.”.
Neumayer, Eric, and Thomas Plümper. 2007. “The Gendered Nature of Natural Disasters: The
Impact of Catastrophic Events on the Gender Gap in Life Expectancy, 1981–2002.” Annals of
the Association of American Geographers 97(3): 551–566.
O’Brien, Diana Z., and Jennifer M. Piscopo. 2023. “Gender and Political Representation in Times
of Crisis.” Politics & Gender 19(3): 891–899.
People’s Assembly. 2024. https://pa.org.za/blog/what-role-ward-councillors “What is the
role of Ward Councillors?”.
Republic of South Africa. 2002. “Disaster Management Act.”.
Reyes-Housholder, Catherine. 2020. “A Theory of Gender’s Role on Presidential Approval Ratings
in Corrupt Times.” Political Research Quarterly 73(3): 540–555.
Sanbonmatsu, Kira. 2002. “Gender Stereotypes and Vote Choice.” American Journal of Political
Science 46(1): 20.
Schwarz, Susanne, and Alexander Coppock. 2022. “What Have We Learned about Gender from
Candidate Choice Experiments? A Meta-Analysis of Sixty-Seven Factorial Survey Experiments.”
The Journal of Politics 84(2): 655–668.
Sevi, Semra, Charles Crabtree, and André Blais. 2024. “Do Voters Punish Women Politicians
More?” American Politics Research 52(6): 701–708.
Statista. 2024. https://www.statista.com/statistics/1498939/
share-of-female-ward-councilors-in-south-africa-by-province/ “South Africa:
share of female ward councilors 2024.”.
Sundström, Aksel, and Daniel Stockemer. 2022. “Measuring Support for Women’s Political Lead-
ership.” Public Opinion Quarterly 86(3): 668–696.
Toxopeüs, M. 2025. https://hsf.org.za/publications/hsf-briefs/drought “Waiting on wa-
ter - Drought management and its protracted timelines: An explainer.”.
Visconti, Giancarlo. 2022. “After the Flood: Disasters, Ideological Voting and Electoral Choices
in Chile.” Political Behavior 44(4): 1985–2004.
34
World Bank. 2021. “Southern Africa Drought Resilience Initiative (SADRI): South Africa Drought
Profile.”.
35
Climate Shocks, Gender, and Incumbency Punishment
Supplementary Material
A Observational Analysis
A.1 Descriptive Statistics
Table A1: Descriptive Statistics: All incumbents
Label length mean sd min max
Elected 3827 0.78 0.41 0.00 1.00
Candidate vote share 3814 0.72 0.16 0.28 0.99
Relative vote share 3812 -0.14 0.22 -0.95 0.49
Female candidate 3827 0.24 0.43 0.00 1.00
Drought declaration 3827 0.04 0.20 0.00 1.00
Ward competition 3816 0.53 0.28 0.00 0.98
Vote share (lagged) 3826 0.63 0.28 0.00 0.99
Provincial drought dec. 3827 0.36 0.48 0.00 1.00
Other disaster dec. 3827 0.44 0.50 0.00 1.00
Table A2: Descriptive Statistics: Women
Label length mean sd min max
Elected 915 0.81 0.39 0.00 1.00
Candidate vote share 909 0.73 0.15 0.30 0.98
Relative vote share 909 -0.14 0.22 -0.95 0.44
Female candidate 915 1.00 0.00 1.00 1.00
Drought declaration 915 0.07 0.25 0.00 1.00
Ward competition 909 0.55 0.27 0.01 0.97
Vote share (lagged) 915 0.65 0.27 0.00 0.98
Provincial drought dec. 915 0.39 0.49 0.00 1.00
Other disaster dec. 915 0.47 0.50 0.00 1.00
A1
Table A3: Descriptive Statistics: Men
Label length mean sd min max
Elected 2912 0.78 0.42 0.00 1.00
Candidate vote share 2905 0.72 0.16 0.28 0.99
Relative vote share 2903 -0.14 0.22 -0.94 0.49
Female candidate 2912 0.00 0.00 0.00 0.00
Drought declaration 2912 0.03 0.18 0.00 1.00
Ward competition 2907 0.52 0.28 0.00 0.98
Vote share (lagged) 2911 0.63 0.28 0.00 0.99
Provincial drought dec. 2912 0.34 0.48 0.00 1.00
Other disaster dec. 2912 0.44 0.50 0.00 1.00
A.1.1 Group Means
Figure A1: Mean ∆ vote pct. by drought condition and gender.
A2
A.2 Robustness Tests
Table A4: Alternate drought thresholds
DV: Incumbent win DV: ∆ vote share
(1) (2) (3) (4) (5) (6)
Female candidate -0.021 -0.019 -0.037 -0.031*** -0.019** -0.030*
(0.023) (0.019) (0.041) (0.009) (0.009) (0.018)
Drought declaration -0.135** -0.067 0.000 0.015 0.022 0.007
(0.055) (0.051) (0.150) (0.038) (0.036) (0.107)
Female x Drought dec. 0.172*** 0.147* 0.142** 0.071** 0.045 0.080
(0.051) (0.079) (0.064) (0.030) (0.051) (0.106)
Ward competition 0.306*** 0.234*** 0.272***
(0.048) (0.043) (0.066)
Vote share (lagged) -0.454*** -0.466*** -0.417***
(0.041) (0.044) (0.059)
Prov. drought dec. -0.055 0.018 0.066 -0.029* -0.024 0.012
(0.038) (0.034) (0.101) (0.018) (0.019) (0.055)
Other disaster dec. -0.007 -0.053* 0.023 -0.021 -0.026 0.042
(0.028) (0.031) (0.059) (0.016) (0.017) (0.033)
Municipality FE Yes Yes Yes Yes Yes Yes
Party-year FE Yes Yes Yes Yes Yes Yes
Drought condition -1.5, 2 mo. -1, 3 mo. -1.5, 3 mo. -1.5, 2 mo. -1, 3 mo. -1.5, 3 mo.
Num.Obs. 1279 1503 444 1377 1635 464
R2 0.472 0.513 0.614 0.676 0.671 0.682
R2 Within 0.063 0.040 0.065 0.171 0.166 0.149
Models 1 & 4 are replications of models 1 & 4 in Table 3. Replications of these models are presented
with alternative drought condition thresholds. Specifically, models 2 & 5 limit the sample to ward
where the SPEI index reaches -1 for three consecutive months. Models 3 & 6 limit the sample to wards
with a SPEI index of -1.5 for three consecutive months. * p ă 0.1, ** p ă 0.05, *** p ă 0.01
A3
Table A5: Alternate incumbent measure: councillors
Incumbents: Ward Incumbents: Councillor
(1) (2)
Female candidate -0.021 -0.031**
(0.023) (0.015)
Drought declaration -0.135** -0.073***
(0.055) (0.024)
Female cand. x Drought dec. 0.172*** 0.118*
(0.051) (0.065)
Ward competition 0.306*** -0.044
(0.048) (0.029)
Prov. drought dec. -0.055 0.029
(0.038) (0.025)
Other disaster dec. -0.007 -0.011
(0.028) (0.022)
Municipality FE Yes Yes
Party-year FE Yes Yes
Drought conditions Yes Yes
Num.Obs. 1279 4019
R2 0.472 0.577
R2 Within 0.063 0.003
Replication of model 3 in Table 3. Model 2 operationalizes incumbency as a
municipality councillor rather than the returning ward incumbent. * p ă 0.1,
** p ă 0.05, *** p ă 0.01
A4
B Additional Analyses, Survey Experiment
Figure A2: Mean Trustworthiness by Experimental Condition
Respondents passed incumbent gender manipulation check.
Mean Outcome by Condition, 95% CIs
Passed Gender Manipulation Check
3.5
3
2.5 Male IncumbentFemale Incumbent
2
1.5
gh
t
gh
t
gh
t
gh
t ef f
u li li
e lie
f iefl
ro rou rou rou  R
e e e e
D  
R  R  R
o o D D D e
s es No No
N N Y Y
Figure A2 shows mean trustworthiness of the incumbent by experimental condition for respondents
who passed the gender manipulation check. When a female incumbent (triangle) secures relief, she
is seen as significantly more trustworthy than a male incumbent (circle). Women are also more
trustworthy in the No Drought condition and Drought condition, though this difference is smaller.
When the incumbent fails to get relief, the male and female both seen as similarly untrustworthy.
A5
Trustworthiness
Figure A3: Mean Favorability by Experimental Condition
Respondents passed incumbent gender manipulation check.
Mean Outcome by Condition, 95% CIs
Passed Gender Manipulation Check
3.5
3
2.5 Male IncumbentFemale Incumbent
2
1.5
ht ht t t f f f f
ug ug ug
h
ug
h lie lie lie lie
ro ro ro o R
e e e e
 D  D D D
r
es
 
es
 R o R o R
No No Y Y
N N
Figure A3 shows mean favorability of the incumbent by experimental condition for respondents
who passed the gender manipulation check. When a female incumbent (triangle) secures relief, she
is evaluated as significantly more favorable than a male incumbent (circle). Women are also more
favorable in the No Drought condition and Drought conditions. When the incumbent fails to get
relief, though, the female incumbent is evaluated more unfavorable than her male counterpart.
A6
Favorability
Figure A4: Mean Vote Likelihood by Experimental Condition
Full sample
Mean Outcome by Condition, 95% CIs
4
3.5
Male Incumbent
3 Female Incumbent
2.5
2
ht ht ht ht f f f fg g g g lie lie lie lie
rou rou ou ou R
e Re Re Re
o D o D D
r Dr s s o  Ye Ye N N
o
N N
Figure A4 shows mean vote likelihood of the incumbent by experimental condition for the full
sample, meaning respondents who did not pass the manipulation check are included. When a
female incumbent (triangle) secures relief, likelihood of voting for her is significantly higher than
the likelihood of voting for the male incumbent who also secures relief (circle). Like our findings in
the main text, women receive a larger reward for their good performance. The female incumbent
also has a higher vote likelihood than the male incumbent in the No Drought condition and Drought
condition. When the incumbent fails to get relief, the male and female incumbent have similar vote
likelihoods. Unlike our sample who passed the manipulation check, the full sample does not punish
women more for a poor performance, but she does experience a greater drop in vote likelihood than
her male counterparty given her higher vote likelihood in the successful relief condition.
A7
Vote Likelihood
Figure A5: Mean Competence by Experimental Condition
Full sample
Mean Outcome by Condition, 95% CIs
3.5
3
2.5 Male IncumbentFemale Incumbent
2
1.5
gh
t ht htg h
t ef ef ef ef
rou rou rou
g g
ou Re
li li li li
r s s R
e
 R
e e
 D  D D D e e No No
 R
No No Y Y
Figure A5 shows mean competence of the incumbent by experimental condition for the full
sample, meaning respondents who did not pass the manipulation check are included. When a
female incumbent (triangle) secures relief, respondents evaluate her as significantly more competent
than the male incumbent who also secures relief (circle). Women are also more competent in the
No Drought condition and Drought condition. When the incumbent fails to get relief, the male
and female incumbent have similar competence ratings.
A8
Competence
Figure A6: Mean Favorability by Experimental Condition
Full sample
Mean Outcome by Condition, 95% CIs
3.5
3
2.5 Male IncumbentFemale Incumbent
2
1.5
ht ht ht ht ief f f fug ug ug ug el eli
e
eli
e e
o o o o R e
li
Dr Dr Dr Dr s s 
R
  e e No
 R o R
No No Y Y
N
Figure A6 shows mean favorability of the incumbent by experimental condition for the full
sample, meaning respondents who did not pass the manipulation check are included. When a
female incumbent (triangle) secures relief, respondents evaluate her as significantly more favorable
than the male incumbent who also secures relief (circle). Women also have higher favorability in
the No Drought condition and Drought condition. When the incumbent fails to get relief, the male
and female incumbent both lose favorability, and the female incumbent is rated as less favorable
than the male incumbent, though this difference falls shy of statistical significance.
A9
Favorability
Figure A7: Mean Trustworthiness by Experimental Condition
Full sample
Mean Outcome by Condition, 95% CIs
3
2.5
Male Incumbent
Female Incumbent
2
1.5
ht ht ht ht ief ief iefug ug ug ug l l l li
ef
ro ro ro ro s R
e Re e e
 D  D D D e es
 
No
 R o R
No No Y Y
N
Figure A7 shows mean trustworthiness of the incumbent by experimental condition for the full
sample, meaning respondents who did not pass the manipulation check are included. When a female
incumbent (triangle) secures relief, respondents evaluate her as significantly more trustworthy than
the male incumbent who also secures relief (circle). Women are also more trustworthy in the No
Drought condition and Drought condition. When the incumbent fails to get relief, the male and
female incumbent have statistically similar trustworthiness ratings.
A10
Trustworthiness
Figure A8: Mean Likelihood of Stealing by Experimental Condition
Full sample
Mean Outcome by Condition, 95% CIs
4
3.8
3.6
Male Incumbent
Female Incumbent
3.4
3.2
3
ht ht htg g g gh
t
lie
f f f
lie lie lie
f
e e e e
Dr
ou rou rou rou s R s RD D D o 
R o R  e
No No Y Y
e N N
Figure A8 shows mean likelihood of the incumbent stealing funds by experimental condition
for the full sample, meaning respondents who did not pass the manipulation check are included.
Higher values on the y-axis reflect a higher belief among respondents that the incumbent they
read about would be likely to steal public funds while in office. These findings are similar to the
findings reported in the main text for the subsample that passed the manipulation check. Women
and men share similar likelihoods of stealing across all conditions except for when drought relief
is secured. When drought relief is secured, respondents view the male incumbent as significantly
more likely to steal than the female incumbent.
A11
Likelihood of Stealing Funds
Figure A9: Mean Vote Likelihood by Respondent Experience with Droughts
Respondents passed incumbent gender manipulation check.
Mean Outcome by Condition, 95% CIs
By If Respondent Has Experienced a Drought
5
4
3 No ExperienceDrought Experience
2
1
an an gh
t t
gh elie
f lief liefe e elie
f
M omW ro
u rouD D R R R R
n, n, an
, n, o N No
 
Ma a M m
a
m o an
, n, 
Wo W M om
a
W
Respondents passed gender manipulation check.
Figure A9 shows mean vote likelihood for the incumbent across experimental conditions, rep-
resented on the x-axis, and by respondents’ self-reported experience living through a drought. A
binary measure of drought experience was created from a survey question that asked respondents
if they have lived through a drought before, yes or no. Real-life drought experience may influence
how respondents react to our vignettes. The plot shows that individuals with no drought expe-
rience (circle) are slightly more punishing of female incumbents, but not more rewarding, than
individuals who say they have lived through a drought (triangle).
A12
Vote Likelihood
Figure A10: Mean Vote Likelihood by Respondent Urbanity
Respondents passed incumbent gender manipulation check.
Mean Outcome by Condition, 95% CIs
Urban vs. Nonurban Respondents
5
4
3 Nonurban RespondentUrban Respondent
2
1
an an ht ht ef ef ef efM om rou
g ug li li li li
W ro , R
e Re e e
D D n n, o 
R o R
n, n, aa a M om
a N N
M m an
, n, 
W
Wo M om
a
W
Respondents passed gender manipulation check.
Figure A10 shows mean vote likelihood for the incumbent across experimental conditions, rep-
resented on the x-axis, and by respondents’ self-reported urban living conditions. A binary measure
of respondents’ urbanity was created from a survey question that asked respondents how urban
their area of living is, with individuals who report living in an ”urban” area coded as 1 and all
other individuals coded as 0. Urban environments may influence how respondents react to drought,
especially when compared to more rural individuals. The plot shows that individuals with urban
residencies (triangle) are slightly more punishing of female incumbents, but not more rewarding,
than individuals who live in nonurban areas (circles), though both subgroups exhibit gender dif-
ferences in punishing and rewarding incumbents for drought responses. Individuals in nonurban
areas are also more likely to vote for a female incumbent in nondrought times.
A13
Vote Likelihood
Figure A11: Mean Vote Likelihood by Respondent Sex
Respondents passed incumbent gender manipulation check.
Mean Outcome by Condition, 95% CIs
By Respondent Sex
5
4
3 Male RespondentFemale Respondent
2
1
an an ht ht lief lief lief liefM g g
Wo
m
Dr
ou rouD , R
e
, R
e e
o R o R
e
an
, n, an ana M om n,
 N n, 
N
M m a a
Wo W M omW
Respondents passed gender manipulation check.
Figure A11 shows mean vote likelihood for the incumbent across experimental conditions, rep-
resented on the x-axis, and by respondents’ sex, male (circle) or female (circle). Respondent sex
may influence which gender stereotypes people hold, and it also may influence how voters respond
to female versus male incumbents. The plot shows that men are slightly more punishing of female
incumbents, but not more rewarding, than women respondents. Both men and women in our sam-
ple exhibit gender differences in punishing and rewarding incumbents for drought responses, and
men and women respondents have a higher vote likelihood for female incumbents in the nondrought
condition.
A14
Vote Likelihood
Figure A12: Mean Vote Likelihood by Partisanship
Respondents passed incumbent gender manipulation check.
Mean Outcome by Condition, 95% CIs
ANC Partisan vs. DA Partisan
5
4
3 DA PartisanANC Partisan
2
1
an n t t f f f fM ma ug
h h e e e e
ug eli eli eli eli
Wo Dr
o
Dr
o R
n, n, 
R  R  R
, , a a No No
Ma
n
ma
n M om an
, n, 
o W aW M Wo
m
Respondents passed gender manipulation check.
Figure A12 shows mean vote likelihood for the incumbent across experimental conditions, rep-
resented on the x-axis, and by respondents’ partisanship, DA (circle) or ANC (triangle), the two
parties that currently dominate South African politics, with the ANC in an especially dominant
political position at both the national and subnational level. Partisanship was measured by asking
respondents if they identified with a political party; respondents were presented with a list of the
major parties in South Africa and could select one party,“other party,” or none. Though parti-
sanship is not included in our vignette, it is possible that respondents fill in partisanship mentally
when reading the scenario, especially when concerning incumbents and their gender. The ANC
has greater women’s representation than the DA and uses party quotas, which may mean that
respondents are more likely to assume a female incumbent belongs to the ANC. However, it is
important to note that differences in women’s representation between the ANC and the DA are
most prominent at the national level; at the local level, women’s representation is low in both
parties. These plots show that ANC partisans and DA partisans evaluate male and female incum-
bents similarly in the no drought and drought conditions, and they are also both more rewarding
of women than of men in the successful relief condition. In the failed relief condition, DA partisans
are overall more punishing of the incumbent than the ANC. Gender differences in punishment are
more pronounced with ANC partisans than DA partisans, too. ANC partisans exhibit a greater
punishment for the female incumbent who fails than for the male incumbent, akin to our main
findings.
A15
Vote Likelihood
Table A6: T-tests for Vote Likelihood, Male vs. Female Incumbent in No Drought Condition, Passed
Gender Manipulation Check Sample
Group N Mean Std. Err. Std. Dev. 95% CI Lower 95% CI Upper
Male, Drought 562 3.316726 0.0563641 1.336198 3.206016 3.427436
Female, Drought 472 3.722458 0.0552508 1.200356 3.613889 3.831026
Combined 1034 3.501934 0.0401577 1.291305 3.423134 3.580734
Difference -0.4057316 0.0796655 -0.5620564 -0.2494069
t = -5.0929
Pt(T <t) = 0.0000 Pr(—T— >—t—) = 0.0000 Pr(T >t) =1.0000
Table A7: T-tests for Vote Likelihood, Male vs. Female Incumbent in Drought Condition, Passed Gender
Manipulation Check Sample
Group N Mean Std. Err. Std. Dev. 95% CI Lower 95% CI Upper
Male, Drought 576 2.5625 0.0593277 1.423865 2.445975 2.679025
Female, Drought 462 2.584416 0.0691428 1.486167 2.448542 2.72029
Combined 1038 2.572254 0.045045 1.451259 2.483865 2.660644
Difference -0.0219156 0.0906794 -0.1998518 .1560207
t = -0.2417
Pt(T <t) = 0.4045 Pr(—T— >—t—) = 0.8091 Pr(T >t) =0.5955
Table A8: T-tests for Vote Likelihood, Male vs. Female Incumbent in Successful Relief Condition, Passed
Gender Manipulation Check Sample
Group N Mean Std. Err. Std. Dev. 95% CI Lower 95% CI Upper
Male, Yes Relief 545 3.893578 0.0549686 1.283254 3.785601 4.001555
Female, Yes Relief 432 4.347222 0.0483862 1.005688 4.25212 4.442324
Combined 977 4.094166 0.0380608 1.189666 4.019475 4.168856
Difference -0.4536442 0.0752862 -0.6013858 -0.3059026
t = -6.0256
Pt(T <t) = 0.0000 Pr(—T— >—t—) = 0.0000 Pr(T >t) =1.0000
A16
Table A9: T-tests for Vote Likelihood, Male vs. Female Incumbent in Failed Relief Condition, Passed
Gender Manipulation Check Sample
Group N Mean Std. Err. Std. Dev. 95% CI Lower 95% CI Upper
Male, No Relief 571 1.961471 0.0593684 1.418643 1.844864 2.078079
Female, No Relief 450 1.671111 0.0549155 1.164933 1.563188 1.779034
Combined 1021 1.833497 0.0413157 1.320166 1.752423 1.91457
Difference 0.29036 0.0827604 0.1279597 0.4527603
t = 3.5084
Pt(T <t) = 0.9998 Pr(—T— >—t—) = 0.0005 Pr(T >t) =0.0002
Table A10: T-tests for Vote Likelihood, Male vs. Female Incumbent in No Drought Condition, Full Sample
Group N Mean Std. Err. Std. Dev. 95% CI Lower 95% CI Upper
Male, No Drought 638 3.319749 0.0529344 1.337053 3.215802 3.423696
Female, No Drought 640 3.651562 0.0498751 1.261751 3.553624 3.749501
Combined 1278 3.485915 0.0366425 1.309936 3.414029 3.557802
Difference -0.3318133 0.0727229 -.4744829 -.1891437
t = -4.5627
Pt(T <t) = 0.0000 Pr(—T— >—t—) = 0.0000 Pr(T >t) =1.000
Table A11: T-tests for Vote Likelihood, Male vs. Female Incumbent in Drought Condition, Full Sample
Group N Mean Std. Err. Std. Dev. 95% CI Lower 95% CI Upper
Male, Drought 636 2.630503 0.0574333 1.448414 2.517721 2.743285
Female, Drought 628 2.83121 0.060978 1.528104 2.711464 2.950956
Combined 1264 2.730222 0.041947 1.491333 2.647928 2.812515
Difference -0.200707 0.0837386 -.3649892 -.0364249
t = -2.3968
Pt(T <t) = 0.0083 Pr(—T— >—t—) = 0.0167 Pr(T >t) =0.9917
A17
Table A12: T-tests for Vote Likelihood, Male vs. Female Incumbent in Successful Relief Condition, Full
Sample
Group N Mean Std. Err. Std. Dev. 95% CI Lower 95% CI Upper
Male, Yes Relief 604 3.900662 0.0516081 1.268341 3.799309 4.002016
Female, Yes Relief 611 4.10802 0.0483383 1.194846 4.01309 4.202949
Combined 1215 4.004938 0.0354531 1.235783 3.935382 4.074494
Difference -0.2073574 0.0706863 -0.3460383 -0.0686765
t = -2.9335
Pt(T <t) = 0.0017 Pr(—T— >—t—) = 0.0034 Pr(T >t) =0.9983
Table A13: T-tests for Vote Likelihood, Male vs. Female Incumbent in Failed Relief Condition, Full Sample
Group N Mean Std. Err. Std. Dev. 95% CI Lower 95% CI Upper
Male, No Relief 622 2.07074 0.058787 1.466144 1.955294 2.186185
Female, No Relief 639 2.098592 0.0568326 1.436639 1.98699 2.210193
Combined 1261 2.084853 0.0408543 1.450758 2.004703 2.165003
Difference -.027852 0.0817446 -0.1882227 0.1325187
t = -0.3407
Pt(T <t) = 0.3667 Pr(—T— >—t—) = 0.7334 Pr(T >t) =0.6333
A18
Figure A13: Mean Comparisons of Vote Likelihood Outcome, Tukey HSD Test, Passed Gender Manipula-
tion Check
Group 1 = Male, No Drought; Group 2 = Female, No Drought; Group 3 = Male, Drought; Group 4 = Female,
Drought; Group 5 = Male, Yes Relief; Group 6 = Female, Yes Relief; Group 7 = Male, No Relief; Group 8 =
Female, No Relief.
A19
Figure A14: Mean Comparisons of Vote Likelihood Outcome, Tukey HSD Test, Full Sample
Group 1 = Male, No Drought; Group 2 = Female, No Drought; Group 3 = Male, Drought; Group 4 = Female,
Drought; Group 5 = Male, Yes Relief; Group 6 = Female, Yes Relief; Group 7 = Male, No Relief; Group 8 =
Female, No Relief.
C Survey Information and Instrument
Ethical Obligations
Our research directly engages human participants in the research process and adheres to APSA’s
Principles and Guidance for Human Subject Research. This survey experiment received ethical
approval from the authors’ institution’s ethical review agency (official name currently redacted
for anonymized peer review). Respondents’ informed and voluntary consent was obtained at the
start of the survey through a consent form, and respondents could exit the survey at any time
even if they initially consented. Respondents also read a debriefing message at the very end of
the survey that reiterates the protection of their data and who to contact should they have any
questions. Our survey was conducted with the international survey firm YouGov. The YouGov
team reviewed our instrument before fielding to ensure that no questions or vignettes contained
information deemed inappropriate to field in the South African context. Respondents were told
A20
that their responses are anonymous, and the authors have received no identifying information in
the survey data about the respondents. Respondents were not deceived at any point during the
survey. Respondents were told explicitly that the vignette was a hypothetical situation. No real-
world names of individuals or political organizations were used in the vignettes. As determined
by our ethics review and the YouGov team’s pre-field assessment, no question or vignette in the
survey posed harm to respondents. Respondents were compensated for their time by YouGov.
YouGov determines compensation based on survey length and their knowledge of typical rates
in the operating country, and YouGov handles all compensation processes. Respondents were
compensated approximately 3.70USD.
The authors have no potential or perceived conflicts of interest arising from this research.
The authors received grants to fund this research. Specific grant information is currently
redacted here for the anonymized peer review process.
Survey Instrument
YouGov Demographics
[Age]
How old are you?
[Text box for answer entry]
[Sex]
Are you. . . ?
1. Male
2. Female
[Province]
In which of the following regions do you live?
1. Eastern Cape
2. Free State
3. Gauteng
4. KwaZulu-Natal
5. Limpopo
6. Mpumalanga
7. North West
8. Northern Cape
9. Western Cape
A21
[Education]
What is your highest level of education?
If you are currently in full-time education please put your highest qualification to date.
1. I did not complete any formal education
2. Kindergarten
3. Preschool – Grade 0
4. Primary school
5. High school
6. Post-secondary, non-tertiary education (generally vocational/professional qualification of 1–2
years, e.g. college, trade school)
7. Short-cycle tertiary education (vocational education and training, studying towards a non-
academic degree, e.g. nursing/teaching diploma)
8. Bachelors or equivalent level degree
9. Masters or equivalent level degree
10. Doctoral or equivalent level degree
[Income]
Gross HOUSEHOLD income is the combined income of all those earners in a household from all
sources, including wages, salaries, or rents and before tax deductions. What is your gross household
income?
1. Under R2,000 per month
2. R2,000 – R3,999 per month
3. R4,000 – R5,999 per month
4. R6,000 – R7,999 per month
5. R8,000 – R9,999 per month
6. R10,000 – R11,999 per month
7. R12,000 – R13,999 per month
8. R14,000 – R15,999 per month
9. R16,000 – R17,999 per month
10. R18,000 – R19,999 per month
11. R20,000 – R24,999 per month
12. R25,000 – R29,999 per month
A22
13. R30,000 – R34,999 per month
14. R35,000 – R39,999 per month
15. R40,000 – R59,999 per month
16. R60,000 – R79,999 per month
17. R80,000 – R99,999 per month
18. R100,000 – R149,999 per month
19. R150,000 or more per month
20. Prefer not to answer
21. Don’t know
[Country of Birth]
Please select from the list the country/region where you were born. If you would prefer not to say,
please select that option at the bottom.
[Drop down list of world countries]
[Intro]
This project aims to understand citizens’ attitudes towards political leaders. The project is led by
the University of Gothenburg, Sweden. The application has been approved by the Swedish Ethics
Review Authority.
All participation is voluntary. Participation consists of filling out a survey with your answers.
Your contact details will be handled by the research company and will not be made public. When
presenting results, only aggregated trends will be presented, never individual participant responses.
[Consent]
Your participation is voluntary, and you can choose to withdraw at any time. If you choose not to
participate or wish to withdraw, you do not need to state why.
Do you want to participate in this study?
1. Yes
2. No
#Screenout if consent=2
[Political]
Which of the following political parties, if any, are you the most supportive of?
1. ANC
2. EFF
3. DA
A23
955. Other
977. Don’t know / prefer not to say
[Ethnicity]
What is your race and/or ethnicity?
1. Black
2. White
3. Coloured
4. Indian
977. Don’t know / prefer not to say
[Rural]
How would you describe the place where you live?
1. City
2. Suburb
3. Town
4. Rural area
977. Don’t know / prefer not to say
[Living]
In general, how would you describe your own present living conditions?
1. Very bad
2. Fairly bad
3. Neither good nor bad
4. Fairly good
5. Very good
977. Don’t know / prefer not to say
[Corruption Perception]
If you had to guess, what percentage of the politicians in your country do you think are involved in
corruption? Use this scale below to answer, where 0% means you think no politicians are involved
in corruption and 100% means you think all politicians are involved in corruption.
[Sliding scale here from 0% to 100%]
A24
[Political Interest]
How much interest would you say you have in politics?
1. None
2. A little
3. Some
4. A lot
[Hypothetical Vignettes]
We will now present you with a hypothetical situation and ask you some questions. Please read
the text carefully.
# Split sample into 8 groups. Randomly assign 1 vignette to each respondent. Each vignette
should have equal probability of assignment.
Condition 1
Imagine you live in another ward within a municipality that is similar to yours.
The ward is represented by a male councilor.
He was elected by residents of the municipality in the last election. As a councilor, his duties
include monitoring the performance of the municipality and representing his ward in the council.
Councilors serve for 4 years and then can be reelected.
Condition 2
Imagine you live in another ward within a municipality that is similar to yours.
The ward is represented by a female councilor.
She was elected by residents of the municipality in the last election. As a councilor, her duties
include monitoring the performance of the municipality and representing her ward in the council.
Councilors serve for 4 years and then can be reelected.
Condition 3
Imagine you live in another ward within a municipality that is similar to yours.
The municipality has been experiencing a drought for weeks. The drought conditions could
bring harm to life, agriculture, and livestock in the municipality due to water shortages. These
conditions could result in lower incomes and limited access to clean water for people in the munic-
ipality.
The ward is represented by a male councilor.
He was elected by residents of the municipality in the last election. As a councilor, his duties
include monitoring the performance of the municipality and representing his ward in the council.
Councilors serve for 4 years and then can be reelected.
A25
Condition 4
Imagine you live in another ward within a municipality that is similar to yours.
The municipality has been experiencing a drought for weeks. The drought conditions could
bring harm to life, agriculture, and livestock in the municipality due to water shortages. These
conditions could result in lower incomes and limited access to clean water for people in the munic-
ipality.
The ward is represented by a female councilor.
She was elected by residents of the municipality in the last election. As a councilor, her duties
include monitoring the performance of the municipality and representing her ward in the council.
Councilors serve for 4 years and then can be reelected.
Condition 5
Imagine you live in another ward within a municipality that is similar to yours.
The municipality has been experiencing a drought for weeks. The drought conditions could
bring harm to life, agriculture, and livestock in the municipality due to water shortages. These
conditions could result in lower incomes and limited access to clean water for people in the munic-
ipality.
The ward is represented by a male councilor.
He successfully secured relief funds for the drought. Because he secured these funds, residents
received food parcels and water trucks were brought to the municipality, and so the negative
impacts of the drought were lessened.
Condition 6
Imagine you live in another ward within a municipality that is similar to yours.
The municipality has been experiencing a drought for weeks. The drought conditions could
bring harm to life, agriculture, and livestock in the municipality due to water shortages. These
conditions could result in lower incomes and limited access to clean water for people in the munic-
ipality.
The ward is represented by a female councilor.
She successfully secured relief funds for the drought. Because she secured these funds, residents
received food parcels and water trucks were brought to the municipality, and so the negative
impacts of the drought were lessened.
Condition 7
Imagine you live in another ward within a municipality that is similar to yours.
The municipality has been experiencing a drought for weeks. The drought conditions could
bring harm to life, agriculture, and livestock in the municipality due to water shortages. These
conditions could result in lower incomes and limited access to clean water for people in the munic-
ipality.
The ward is represented by a male councilor.
He failed to secure relief funds for the drought. Because he did not secure these funds, residents
did not receive food parcels and water trucks were not brought to the municipality, and so the
negative impacts of the drought were not lessened.
A26
Condition 8
Imagine you live in another ward within a municipality that is similar to yours.
The municipality has been experiencing a drought for weeks. The drought conditions could
bring harm to life, agriculture, and livestock in the municipality due to water shortages. These
conditions could result in lower incomes and limited access to clean water for people in the munic-
ipality.
The ward is represented by a female councilor.
She failed to secure relief funds for the drought. Because she did not secure these funds,
residents did not receive food parcels and water
[Vote]
Imagine you were allowed to vote in elections for the ward councilor in this municipality during the
next election cycle. How likely would you be to vote for this councilor in the next ward election?
1. Unlikely
2. Somewhat unlikely
3. Neither likely nor unlikely
4. Somewhat likely
5. Likely
# Randomize order of appearance for the following 3 questions only: Competence, Trust, and
Favorability
[Competence]
How competent do you think this councilor is?
1. Incompetent
2. Somewhat incompetent
3. Somewhat competent
4. Competent
[Trust]
How much do you trust the councilor to help people in the municipality?
1. Do not trust
2. Trust a little
3. Trust somewhat
4. Trust a lot
A27
[Favorability]
How favorably do you view the councilor?
1. Unfavorably
2. Somewhat unfavorably
3. Somewhat favorably
4. Favorably
[Stealing]
How likely do you think it is that this councilor will steal public funds while in office?
1. Unlikely
2. Somewhat unlikely
3. Neither likely nor unlikely
4. Somewhat likely
5. Likely
[Explanation]
You previously answered that you think the councilor is [# Pipe in respondent’s answer from
[Competence] question]. Why did you evaluate the councilor in this way?
[Open response box here]
[Success]
Only for individuals who received Condition 5 or Condition 6 (Group 5 or 6).
In your own words, please write a few sentences about why you think that the incumbent
councilor succeeded in securing relief funds for the drought.
[Text box here for response]
[Failure]
Only for individuals who received Condition 7 or Condition 8 (Group 7 or 8).
In your own words, please write a few sentences about why you think that the incumbent
councilor failed in securing relief funds for the drought.
[Text box here for response]
[Gender Check]
Was the councilor you read about a man or a woman?
1. Man
2. Woman
A28
[Response Check]
Only for individuals who received Condition 3, 4, 5, 6, 7, or 8 (Groups 3–8).
How did the councilor respond to the drought?
1. The councilor successfully secured relief funds
2. The councilor failed to secure relief funds
3. I did not read about how the councilor responded to the drought
[Attention]
Select the number 17 from the list of numbers below:
1. 6
2. 19
3. 25
4. 17
5. 89
[Women]
Please indicate your agreement with the following statements:
Women Jobs When jobs are scarce, men should have more rights to a job than a woman.
Women Leaders On average, men are better political leaders than women.
Women Honest On average, women are more honest than men.
1. Strongly disagree
2. Disagree
3. Neither disagree nor agree
4. Agree
5. Strongly agree
# Randomize order of appearance for the following 2 questions: Drought Experience, Drought
Worry
[Drought Experience]
Have you ever experienced a drought?
1. Yes
2. No
3. I don’t know
A29
[Drought Worry]
In general, how worried are you about future droughts in South Africa?
1. Very worried
2. Somewhat worried
3. A little worried
4. Not worried at all
[Outro]
The project will collect information about you, in the form of survey responses. They will be
aggregated for the analysis of statistical trends.
Survey responses will be saved in electronic files with a number only. The researchers involved
will study responses in an anonymized file. This will be stored through a solution established for
sensitive information. By working according to state-of-the-art with privacy, we ensure that data
is not disseminated outside the project, in accordance with [redacted for anonymized peer review].
According to this regulation, you have the right to access the data about you that is processed
in the project free of charge, and to have any errors corrected if necessary. You can also request
that data about you be deleted and that the processing of your personal data be restricted.
Responsible is [redacted for anonymized peer review].
A30