Research
Published
How to distinguish human error from election fraud: Evidence from the 2019 Malawi election
with Ryan Jablonski, British Journal of Political Science, Volume 55 2025
Abstract
Voters and politicians often blame tallying irregularities on fraud, undermining perceptions of democratic and electoral credibility. Yet such irregularities also result from capacity failures and human error. We introduce several methods to assess competing causes of tallying irregularities leveraging the quasi-random administration of polling stations. Using these methods, we revisit the case of the 2019 Malawian presidential election which was famously cancelled by the High Court due to widespread tallying irregularities and accusations of fraud. Contrary to the dominant consensus, we do not find evidence that irregularities were motivated by fraud or that they benefited the incumbent. Instead, we show that irregularities increased in proportion to the complexity of filling in result-sheets, suggesting a dominant role for human error. In addition to reinterpreting a historically important election, we also make the case that policy efforts to improve electoral credibility could productively be reallocated towards electoral administration rather than anti-fraud measures.
Please find the full article here
with Brigitte Seim and Ryan Jablonski, Journal of Development Economics, September 2020
Abstract
Does foreign aid shift public spending? Many worry that aid will be “fungible” in the sense that governments reallocate public funds in response to aid. If so, this could undermine development, increase the poorest's dependency on donors, and free resources for patronage. Yet, there is little agreement about the scale or consequences of such effects. We conducted an experiment with 460 elected politicians in Malawi. We provided information about foreign aid projects in local schools to these politicians. Afterwards, politicians made real decisions about which schools to target with development goods. Politicians who received the aid information treatment were 18% less likely to target schools with existing aid. These effects increase to 22–29% when the information was plausibly novel. We find little evidence that aid information heightens targeting of political supporters or family members, or dampens support to the neediest. Instead the evidence indicates politicians allocate the development goods in line with equity concerns.
Please find the full article here.
Work in progress
What drives public trust in elections? Experimental evidence from Malawi
with Alex Yeandle
Abstract
International donors have invested heavily in strenghtening electoral administration in low-income democracies, aiming to reduce irregularities and build trust. However, we know little about whether these interventions actually improve public perceptions. Using a conjoint choice experiment in Malawi, randomising organisational features of polling stations and their potential for political bias, we examine the determinants of public trust in a low-income setting. Voters are more trusting of stations with well-trained polling staff, independent monitors, security personnel, and transparency measures - effects driven by sanctioning the absence of these basic requirements. Respondents also prioritise procedurally fair measures over those that exclusively benefit their own party or ethnic group, challenging assumptions about the dominance of partisanship and ethnicity in African elections. We contribute to the literature on election administration and public opinion in low-income settings, while highlighting ways in which resource-constrained election bodies can build and maintain public support.
Please find a draft paper here.
with Alex Yeandle
Abstract
Research on projection shows that people overestimate the prevalence of their own views among others, and this can significantly shape political behaviour. But existing studies focus on wealthy information-rich democracies, rather than lower-income, uncertain settings where such perceptions are an important, high-stakes part of everyday political life. Misjudging others risks constraining voters' ability to coordinate, which can undermine provision of public goods, confidence in election outcomes, or efforts to overthrow dominant parties. Using original survey data from Malawi, in a pre-registered research design, we show that individuals in a low-income setting systematically overestimate support for their own party and the prevalence of their own ethnic group, while politically engaged individuals overestimate participation by those around them. These findings add new microfoundational insights to the study of politics in low-income states, while highlighting several avenues for future work going forward.
Foreign aid is neither a curse nor a blessing: Explaining the effect of foreign aid on voting behaviour and accountability
with Ryan Jablonski and Brigitte Seim
Abstract
How does foreign aid affect elections? To reconcile mixed evidence, we re-conceptualize the effect of aid on elections as a retrospective voting problem and show that the electoral effects of foreign aid are heterogeneous and depend on the distribution of aid and voter beliefs. To test our argument, we conducted a survey among 2,331 citizens around a sample of 180 schools in Malawi before and after the delivery of a foreign NGO project. Additionally, we conducted a SMS-based information experiment which varied voter knowledge about the aid allocation process. In line with expectations, voters who benefit from aid are more likely to anticipate voting for incumbents. But overall aid was a mixed blessing for incumbents: when citizens learn about aid, but fail to benefit, we document a sizable backlash against incumbents. Citizens were less likely to be satisfied with or vote for incumbents.
Natural Disasters and Electoral Politics in Low-Income States
with Alex Yeandle
Abstract
In wealthy democracies, the political impact of natural disasters depends largely on how governments respond to them. But climate shocks are most severe in low-income states, where politicians have limited capacity to provide post-disaster relief. In such settings, we argue that the electoral impact of natural disasters depends more on longterm partisan geography than short-term policy response. Governments skew infrastructure investments towards their own strongholds, limiting economic damage, while opposition parties are routinely prevented access to victims. We present evidence from Malawi, where Cyclone Idai caused mass flooding shortly before national elections. Using a matched difference-in-difference design, with geocoded polling station returns and high-resolution flood maps, we show that ruling party vote share falls in flooded communities, but only where historic incumbent support was weak. Consistent with the argument, household surveys and interviews indicate these areas were less financially resilient and more accessible to opposition campaigns.
Temporal updating of retrospective election integrity evaluations
with Alex Yeandle
Abstract
Survey-based retrospective evaluations of election integrity are widely used in comparative politics. These measures are typically interpreted as stable assessments of how voters think elections were conducted. We show, however, that retrospective integrity evaluations are temporally updated, that is, reconstructed at the time of the survey using contemporaneous political attitudes. Using repeated Afrobarometer surveys that ask identical questions about the same national election across multiple survey rounds, we analyse retrospective evaluations in 59 elections measured at different points in time. We show that respondents' assessments of a given election shift substantially across survey rounds and closely track changes in incumbent support, despite no change in the eleciton being evaluated. These patterns are robust across alternative measures and specifications, and are illustrated in an auxilliary case study. The findings demonstrate a systematic limitation of retrospective survey measures and caution against treating them as time-invariant indicators of past electoral quality.