Photo credit: Carol Sahley/USAID

How should political parties choose candidates?


Published 13.03.24
Photo credit:
Carol Sahley/USAID

An experiment in Sierra Leone varied how much say voters, compared to party officials, had in selecting candidates for Parliamentary elections, and sheds light on which systems of leadership selection are most effective

Editor's Note: Check out the companion podcast for this article which delves into how this research happened - listen here.

The competence and integrity of political leaders drive government performance. How we choose our leaders, and the characteristics we value in the selection process, thus become critical inputs for good governance. These are not new ideas: more than 200 years ago, James Madison argued for the importance of carefully designed leadership selection processes in the Federalist Papers (Hamilton et al. 1788). Yet we have scant evidence on how different selection processes influence the types of people who become politicians, as the contemporary political economy literature has largely neglected the issue (Besley 2005), at least until recently (Dal Bó and Finan 2018).

One critical component of leadership selection is how political parties choose candidates. In most democracies around the world, parties rely on nominations or appointments by party officials. In the United States, by contrast, parties use direct vote primaries. As these are two quite distinct approaches, a natural question is which system works better: are party elites or ordinary voters better positioned to pick candidates?

The answer is not obvious. Primary skeptics worry that voters are poorly informed about politics, so letting voters choose might bolster representation, but it will come with a cost of selecting candidates of dubious quality. Advocates counter that party elites might have different preferences than ordinary citizens and may exploit their positions of power to “manipulate and profit from the nomination process” (Hirano and Snyder 2019, page 15). Adjudicating these rival claims empirically is challenging, as parties rarely alter how they select candidates for anything but purely strategic, and hence endogenous, reasons.

An experiment in candidate selection

To make progress, we partnered with the two major political parties in Sierra Leone on a novel field experiment that varied how much say voters, as compared to party officials, had in selecting candidates for the 2018 Parliamentary elections (Casey, Kamara and Meriggi 2021). In control races, the parties implemented “politics as usual,” and chose candidates via recommendations from party officials at various levels, with no direct participation by voters. For a randomly selected subset of treatment races, the parties implemented a new selection method with two components: i) a party convention where potential candidates (called “aspirants”) presented their qualifications to party officials and local residents, and engaged in policy debates that were broadcast over radio; followed by ii) opinion polling, representative of all registered voters in the constituency, that asked voters which aspirant they would like to become the candidate, which we aggregated and shared with party officials via one page reports. While parties were free to ignore the reports, if they followed them everywhere, the intervention would approximate a US-style primary.

We evaluate treatment effects on two key outcomes: representation, defined as the candidate chosen to run in the general election is the aspirant who ranks first among voters, and the quality of selected candidates.

Research design

The Political Parties Regulation Commission (PPRC), which has the constitutional mandate to regulate political parties, invited all registered parties to participate in the initiative and associated research. National party leaders decided whether to participate, and both major parties—the All People’s Congress (APC) and the Sierra Leone People’s Party (SLPP)—opted in. For budgetary reasons, the PPRC asked each party to select 46 races from the 132 Parliamentary constituencies nationwide where they would be willing to experiment with the new selection mechanism. The research team randomly assigned half of each party’s list to treatment and control groups (see Figure 1).

Figure 1: Random assignments by party

Random Assignments by Party

Field enumeration teams collected rich data from voters, aspirants and party leaders before and after the conventions and radio broadcasts, in both treatment and control races (see Figure 2). Data from the second voter surveys (labelled V2) determine which aspirant is most preferred by local constituents. For treatment races, this information goes into the voter reports shared with parties. In controls, it shows who would have won a primary, if one had occurred.

Figure 2: Implementation timeline

Implementation Timeline

The problem for representation in the status quo

Analysis of polling data in control races reveals a sizeable distortion away from voter preferences: under “politics as usual,” voters get their first choice candidate only 38% of the time. Consider what this distortion implies in party strongholds, where the APC has long-standing ties with ethnic groups in the north of the country, while the SLPP has ties to groups in the south (Kandeh 1992). In these areas, whomever the party chooses is near guaranteed to win the general election, so if voters do not get their favorite candidate, they will not get their preferred representative in Parliament.

What might explain this status quo distortion: is it due to a conflict in preferences between voters and party leaders, or something else? When we examine the characteristics that predict how popular aspirants are among voters and party officials, we find that they broadly agree on what makes for a good candidate. The strongest predictor for both groups is an individual’s record of providing local public goods and economic development projects. To the extent that past provision predicts future performance in office, this makes intuitive sense.

Information constraints seem more likely to explain why party officials frequently fail to select the most popular candidates. To see this, we compare party officials’ answers to two questions: i) “if the choice to award the symbol was up to you today, who would be your first choice?” and ii) “if the registered voters in this constituency voted directly today for the symbol, who do you think would get the most votes?” While 90% of party officials indicated that voters shared their first choice, this is incorrect: only about half were in fact a match with the voter polling data.

Impacts of the more democratic primary selection method

The primary intervention alleviates information constraints in two ways: the conventions and broadcasts provide rich information about aspirant qualifications; and the voter reports elicit, aggregate, and deliver information about citizen preferences to party officials. Our experiment evaluates whether this combination improves representation and selection on quality.

Figure 3 reveals large gains for representation: in treatment races, the frequency with which parties select the candidate most preferred by voters increases to 61%, an improvement of 23 percentage points. Back-of-the-envelope calculations suggest that parties thereby chose a different candidate in 11 races, and because about half of sampled races are in strongholds, this effectively changed the identity of 6 elected MPs.

Figure 3: Positive treatment effect on representation

Positive treatment effect on representation

To test whether this increase in representation comes at the expense of selection on quality, we compare the characteristics of candidates ultimately sent to the general election across treatment and control races. Experimental estimates suggest that the more democratic method selected candidates with stronger public goods records, meaning that aspirants who had provided more development projects in the past were more likely to be chosen to advance to the general election. Figure 3 shows this result for a bundle of related measures rolled up into an index, where zero is the mean score for aspirants in control races. The average score increases to 0.46 standard deviation units for candidates selected in treatment races, which is a large and highly significant improvement. This is important as it reflects increases in exactly the characteristics that were found above to be important to voters and party leaders.

Figure 4: Positive treatment effect for selection on quality

Positive treatment effect for selection on quality

Note: this figure presents treatment effect estimates for selection on quality, where the error bars correspond to the 95% confidence interval as estimated via ordinary least squares regression that includes the 23 party-region randomization strata and robust standard errors; the development spending index compiles multiple measures of a candidate’s provision of local public goods over the previous three years, which is normalized with respect to all aspirants in control group races and expressed in standard deviation units.

Concluding thoughts

Elections are large public investments, for both domestic governments and their donor partners. The efficacy of these investments in delivering representative and competent elected politicians depends critically on how candidates are selected. If party officials select candidates with little input from voters, citizens may be perfectly enfranchised on paper—entitled to participate in free and fair general elections—but wholly irrelevant in practice, at least for partisan strongholds.

Political parties in Sierra Leone demonstrate that there are practical ways to improve candidate selection. Their openness and responsiveness to the new mechanism attests to their willingness to experiment at the frontier of democratic practice. Our headline results show how giving voters more say can improve representation without compromising candidate quality.

Regarding generalisability, note that we compiled voter reports from a representative sample of voters, whom we visited at their residences. This sidesteps some contentious issues in American primaries, like the relationship between selective turn out and the choice of ideologically extreme candidates. Also, the set of aspirants under consideration was largely fixed, whereas in other environments a switch to primaries might alter who was interested in running in the first place. We leave the consequences of these distinct design choices to future research.


Besley, Timothy (2005), "Political Selection", Journal of Economic Perspectives, 19(3): 43-60.

Casey, Katherine, Abou Bakarr Kamara, and Niccolò F. Meriggi (2021), "An Experiment in Candidate Selection", American Economic Review, 111(5): 1575–1612.

Dal Bó, Ernesto and Frederico Finan (2018), "Progress and Perspectives in the Study of Political Selection", Annual Review of Economics, 10: 541-575.

Hamilton, Alexander, et al., editors (2008), The Federalist Papers, Oxford University Press, October 1787-May 1788.

Hirano, Shigeo and James M. Snyder Jr. (2019), Primary Elections in the United States, Cambridge University Press.

Kandeh, Jimmy D. (1992), "Politicization of Ethnic Identities in Sierra Leone", African Studies Review, 35: 81-99.