The social identities of elected politicians – such as their gender and ethnicity – affect who in society is favored with state resources and whose preferences are represented in government (Chattopadhyay and Duflo Reference Chattopadhyay and Duflo2004; Pande Reference Pande2003). But politicians with particular social identities can only be elected if they appear as candidates on the ballot. The choice sets presented to voters are determined by the candidate selection institutions adopted by political parties. In new democracies, parties increasingly use primary elections to select their nominees for legislative office (Field and Siavelis Reference Field and Siavelis2008), which may have important consequences for the representation of different identity groups. Primaries increase the number and types of voters who decide which politicians will reach the general election. Characteristics of the primary electorate may affect politicians' decisions about whether to even compete for a party's nomination.
We investigate how expanding the size and demographic diversity of the primary electorate affects the types of candidates who choose to compete in primaries and the characteristics of the nominees who appear on the general election ballot. We focus on the short-run effects of a reform to the legislative primary electorate in the National Democratic Congress (NDC), one of Ghana's two major political parties. Previously, the NDC primary electorate in each parliamentary constituency comprised roughly 200 to 800 local party branch executives. Leading up to the 2016 elections, however, the NDC opened primary voting to all rank-and-file members, expanding the primary electorate by an order of magnitude to roughly 2,000–10,000 voters per constituency. This increased the proportion of primary voters who are women or identify with groups outside the party's core ethnic base.
This reform allows us to assess the effects of expanding the primary electorate on candidate entry and selection because the New Patriotic Party (NPP), Ghana's other major party, provides an unusually good set of primaries for comparison. The two parties have similarly sized, stable bases and have alternated in power. Crucially, they also have very similar organizational structures (Riedl Reference Riedl2014), including similar gender gaps and ethnic disparities in local leadership and the same primary rules before 2016. Moreover, before the 2012 elections the NPP considered the same electorate expansion as the NDC for the same reasons and may still adopt it. While the stability and organizational parallels between these parties are unusual, they allow us to make a tight comparison of the primary processes of two very similar parties in similar constituencies in the same electoral environment. Such a comparison would not be possible in other new democracies with more inchoate party systems or a dominant party.
More broadly, this case offers a rare window into the effects of franchise extensions on descriptive representation in the developing world, in terms of both gender and ethnicity. Because democratization is typically a bundle of reforms and universal suffrage was frequently established at independence, past attempts to study the effects of franchise extensions on the social identities of candidates in these settings have struggled to disentangle their effects from other contemporaneous changes (Josefsson Reference Josefsson2014). The effects of franchise extensions within primary elections can be more clearly isolated from those of other factors in our case.
We estimate the average effects of expanding the NDC's electorate on its pool of aspirants and the characteristics of its nominees (average treatment effect on the treated). The NDC and NPP are not exactly the same, and our setting does not provide a ‘natural’ experiment. Instead, our analysis attempts to generate reasonable counterfactuals to NDC parliamentary primaries by creating matched sets with weighted NPP primaries. We apply optimal full matching (Rosenbaum Reference Rosenbaum1991; Rosenbaum Reference Rosenbaum2010) to an original dataset of biographical information on all aspirants seeking nominations in these two major parties in 2016, as well as a similar dataset for Ghana's 2012 elections.
We find that expanding the electorate changed the gender, ethnicity and political experience profiles of the pool of aspiring candidates and eventual nominees. It increased the average number of female aspirants and the probability that the nominee is a woman. In addition, more aspirants from ethnic groups outside the NDC's core coalition sought and won nominations. Neither change came at the expense of the party's incumbent Members of Parliament (MPs), however, as they became more likely to be renominated. Instead, the electorate expansion reduced the probability of nominating wealthy political newcomers – outsiders who were previously able to buy their way to nominations, sometimes even over long-serving incumbents.
We propose that these effects resulted from two shifts. First, expanding the electorate reduced the feasibility of direct vote buying (Lizzeri and Persico Reference Lizzeri and Persico2004), the dominant mode of campaigning in small-electorate primaries. Secondly, having more women and ethnic groups outside the party's core base in the electorate increased the expected success of aspirants from these groups.
After the reform, prospective female aspirants with ambitions to seek office, but without the financial resources and connections to the male-dominated patronage networks needed to buy votes, could expect to be more successful. They may also have expected to perform better in an electorate with more female voters. Aspirants from ethnic groups outside the party's core coalition could expect similar changes to their viability in primaries with more co-ethnic voters. These shifts also reduced the viability of wealthy political outsiders who lacked clear appeal among the broader party membership by limiting their ability to buy nominations. However, incumbent MPs often had the resources and reputations of constituency service needed to win primaries with any size of electorate.
We do not analyze the reform's effects on who is eventually elected to Parliament. It is difficult to construct a reasonable counterfactual of who would have won the general election in 2016 had the NDC kept its previous primary system and potentially selected different nominees, but still competed against the same 2016 NPP nominees.Footnote 1 Descriptively, however, the 2016 general election results appear to be consistent with the NDC's reform opening a path to office for a new group of politicians. The NDC had the same number of female MPs (fourteen) in 2016 as in 2012, even though it won forty-two fewer seats overall (106 vs. 148) – a 40 per cent increase in the female share of the party's caucus in Parliament. Moreover, eight of the fourteen women elected in 2016 were new MPs who had not been nominated in 2012 under the smaller electorate.
This article makes several contributions. First, we extend the literature on how electoral and party institutions, including parties' internal rules, affect women's descriptive representation (Bjarnegård and Zetterberg Reference Bjarnegård and Zetterberg2019; Norris and Lovenduski Reference Norris and Lovenduski1995; Wangnerud Reference Wangnerud2009). Our finding that opening up primary voting aids female candidacy contrasts with qualitative evidence that the democratization of candidate selection reduces women's access to nominations (Hinojosa Reference Hinojosa2012).Footnote 2 It also overcomes methodological challenges in other studies of the effects of franchise extensions on the election of women that are less able to isolate changes to the electorate from other contemporaneous political reforms (Josefsson Reference Josefsson2014).
We show that expanding the primary electorate can improve women's incorporation even in the short run, without changes to deeper constraints such as women's latent ambition (Fox and Lawless Reference Fox and Lawless2014) or party elites’ gendered recruitment efforts (Sanbonmatsu Reference Sanbonmatsu2002). Our findings also suggest that democratizing candidate selection may be an especially useful tool to improve women's representation in single-member district electoral systems, where the zero-sum nature of the contest is a particularly serious obstacle to women's candidacy. In these contexts, better-studied remedies like quotas and reserved seats may produce larger overall effects on women's representation than expanding the primary electorate (Duflo Reference Duflo2005; Krook Reference Krook2009; Tripp and Kang Reference Tripp and Kang2008), but they are often also more difficult to implement without backlash from men (Clayton Reference Clayton2015).
Secondly, and more broadly, we extend the study of franchise extensions to contemporary new democracies and show that changing the social identities of the pool of politicians seeking elective office is a key channel through which these reforms can affect governance. Previous research has found that historical suffrage extensions in advanced democracies improved policy and substantive representation (Aidt and Dallal Reference Aidt and Dallal2008; Berlinski and Dewan Reference Berlinski and Dewan2011; Lott and Kenny Reference Lott and Kenny1999; Morgan-Collins and Teele Reference Morgan-Collins and Teele2017), but it is unclear whether these effects can be attributed to changes in the identities of the politicians elected (Berlinski, Dewan and van Coppenolle Reference Berlinski, Dewan and van Coppenolle2014).
Expanding the Electorate: An Open Empirical Question
Whether women and other politicians from marginalized social identity groups seek and win office may be affected by electoral institutions, such as voting rules (Tripp and Kang Reference Tripp and Kang2008) and quotas (Krook Reference Krook2009), and internal party institutions, such as candidate eligibility criteria (Bjarnegård and Zetterberg Reference Bjarnegård and Zetterberg2019). However, prior studies offer little theoretical guidance on how expanding the franchise may affect who runs and wins in primaries or general elections in new democracies.Footnote 3 Analytical results are not available for a general formal model of candidate entry that involves simultaneous entry by an undefinable pool of potential aspirants from multiple groups. Multiple equilibria are possible even with a simple voting model and a known number of candidates and identity groups (Dickson and Scheve Reference Dickson and Scheve2010). Different combinations of entrants can also generate different outcomes, depending on how aspirants split votes within segments of the electorate. Consistent with this ambiguity, research on historical suffrage extensions in today's advanced democracies finds mixed effects on the identities of elected politicians (Berlinski and Dewan Reference Berlinski and Dewan2011; Berlinski, Dewan and van Coppenolle Reference Berlinski, Dewan and van Coppenolle2014; Morgan-Collins and Teele Reference Morgan-Collins and Teele2017).
More generally, studies of primaries in advanced democracies offer little guidance because ideology, which is central to most existing models (for example, Gerber and Morton Reference Gerber and Morton1998; Serra Reference Serra2011), is not a key feature of electoral competition in many new democracies. Primary aspirants in these settings are differentiated less by policy than by their ability to deliver patronage to small groups of primary voters (Ichino and Nathan Reference Ichino and Nathan2012).
In new democracies, changes to the size and demographic composition of the electorate should be more likely to affect primaries through changes to these vote-buying exchanges. Increasing the size of an electorate can induce a shift away from clientelism because narrow particularistic appeals are less feasible with larger numbers of voters (Lizzeri and Persico Reference Lizzeri and Persico2004). As vote buying declines, potential aspirants who have ambition but lack the resources or access to the patronage networks needed to be effective in clientelist primaries may expect to become more viable. Moreover, where there is significant ethnic voting, changing the composition of the electorate may also improve the expected prospects of aspirants from previously under-represented ethnic groups by bringing more potential supporters into the electorate.
Reforms to Parliamentary Primaries in Ghana
Since 1992, Ghana has held quadrennial elections for president and parliament, which has 275 members from single-member constituencies. The NDC and NPP dominate highly competitive elections and have alternated in government, with the NDC in power leading into 2016. These well-institutionalized parties have faced strong ‘isomorphic competitive pressures’ to mirror each other's behavior and organizations; reforms adopted by one party are often quickly matched by the other (Riedl Reference Riedl2014). Consequently, the NDC and NPP are unusually similar in many respects, including appeals to clientelism, the size of their ethnic bases, and their grassroots organizations and candidate selection procedures up until 2016.
The parties have very similar platforms, and policy is not a major axis of competition (Bob-Milliar Reference Bob-Milliar2012b; Nathan Reference Nathan2019; Riedl Reference Riedl2014). Although the parties have rank-and-file members from every group and voting is not exclusively along ethnic lines, ethnicity is a strong determinant of partisanship. The NPP's base is the Akan; the NDC's core is a collection of smaller groups, most notably the Ewe, the Ga and many northern groups. These bases are stable and similarly sized, leaving the two parties roughly evenly matched, with less subnational electoral volatility compared to more inchoate party systems (Ferree Reference Ferree2010).
Although the parties differ in some respects such as their elite factional politics (Bob-Milliar Reference Bob-Milliar2012a; Osei Reference Osei2016), they have nearly identical organizations. The NDC has a committee of nine and the NPP a committee of five polling-station executives at nearly all of Ghana's approximately 29,000 polling stations. These branch executives are the core ‘foot soldiers’ in each party's campaign efforts and serve as clientelist brokers linking ordinary supporters to politicians and government officials (Bob-Milliar Reference Bob-Milliar2012b; Brierley and Nathan Reference Brierley and Nathan2020).
The branch executives also comprised the primary electorates for each party for the 2004 through 2012 elections. These electorates were disproportionately male and composed of each party's core ethnic groups. For the NPP, the primary electorate numbered 200–800 in each constituency, with five branch leaders from each polling station. Women made up just 26 per cent of this electorate. In addition, the data in Brierley and Nathan (Reference Brierley and Nathan2020) show that the proportion of Akans among NPP polling-station-level executives is over 14 percentage points higher on average than in the estimated population of their polling stations' catchment areas. The NDC primary electorate in 2012 was similar, with four branch leaders voting per polling station; the women's organizer was typically the only female voter even though women made up approximately 37 per cent of ordinary party members.Footnote 4 Ewes, Gas and Northerners are similarly over-represented in NDC branch leadership (Nathan Reference Nathan2019).
With electorates of just a few hundred people, the main mode of competition in parliamentary primaries has been vote buying (Ichino and Nathan Reference Ichino and Nathan2012). Aspirants with significant personal wealth dominate. In some cases, aspirants return to their hometowns to ‘buy’ nominations despite having little history of working for their party or living in the constituency. Branch-level party leaders often view primaries as their ‘cocoa season’ – a key opportunity to extract rents. Aspirants pursue primary voters with gifts of TVs, motorbikes, school fees for their children and the like. Many aspirants openly admit in interviews to paying primary voters and report spending upwards of USD 75,000 to secure nominations.Footnote 5 They go to great lengths to monitor and enforce vote-buying transactions with the small number of party branch leaders voting in these elections.Footnote 6
As an NPP aspirant in the 2011 primaries described his experience: ‘Everybody [the voters] was about “What's in it for me? What have you brought for me? Politicians, after this election, they're not going to care about us any more…this is our chance”’.Footnote 7 Nominations often go to the highest bidder. One NPP MP described this bluntly, lamenting the ‘“money-ocracy” that has eaten into the fiber of our politics’. He argued the primaries ‘elect money into office rather than electing people based on competence’.Footnote 8
In 2013, NDC leaders expanded their primary electorate for the 2016 election to include all ordinary party members in each constituency. This was an attempt to reduce the influence of vote buying and select more popular nominees. NDC leaders explicitly stated that they hoped vote buying would be less feasible with a larger electorate.Footnote 9 This decision was made before the outcome of the 2016 election could be anticipated with any confidence, and without knowledge of what nomination system the NPP would use.
In June 2015, the NPP held primaries for the 2016 elections under the same rules as for 2012. The NDC primaries followed months later with a significantly expanded electorate, with most primaries held between late November 2015 and early 2016. The 2015/2016 NDC primary electorate was an order of magnitude larger than the NPP's, with 2,000–10,000 ordinary members voting on each nomination. Rather than taking place at a single venue in each constituency as in each party's small-electorate primaries, the new NDC primaries involved voting at every polling station. The NDC assembled a new digital register of all its rank-and-file members to use as its voter roll.
This electorate expansion could have occurred in either party. In 2009, the NPP also carefully considered expanding its electorate, deciding between two proposals: marginally increasing the number of executives from each polling station that could vote or allowing all rank-and-file members to vote, similar to the reform later adopted by the NDC. The NPP's reform advocates made nearly identical arguments to those that NDC leaders would make in 2013, claiming that increasing the electorate would ‘lead to election of people who actually … work for the party’, would ‘ensure that the selected … candidates … represent the popular will’, and ‘reduce expenditure on internal party elections’.Footnote 10 A main goal was similarly to reduce vote buying.Footnote 11 But due to the concern of one of the party's aspiring presidential candidates that a larger electorate would disadvantage his campaign, the party ultimately settled on the first proposal.
Possible Impacts of the Reform
The NDC's reform likely had two effects, which in turn may have changed the gender, ethnicity, and political experience profiles of the NDC's aspirants and eventual nominees for the 2016 election. First, because the party's rank-and-file membership is more gender balanced and ethnically diverse than its branch leadership, the new electorate had a greater proportion of women and non-core group members than before.
Secondly, as NDC leaders hoped, the electorate expansion may have changed the dominant campaign strategy in primaries. When the electorate is very small, primary aspirants can target specific voters with personalized benefits and enforce exchanges through direct monitoring. But this becomes more costly as the number of voters shifts into the many thousands; previously common forms of individual-level monitoring become logistically infeasible.Footnote 12 Aspirants might switch to employing intermediaries to buy votes on their behalf, but this is less efficient than the direct exchanges that are possible with very small electorates (Stokes et al. Reference Stokes2013).
In interviews, many aspirants anticipated that expanding the electorate would have exactly these impacts on their vote buying. Among the aspirants in our survey (Appendix A2), which was conducted prior to the NDC's primaries, expanding the electorate was by far the most common suggestion for reducing vote buying in primaries.Footnote 13 For example, one interviewee predicted, ‘if we allow all registered members to vote, it will make it difficult for vote buying… how are you going to pay money to everybody?’Footnote 14 Another explained:
That problem can be reduced if we … expand our register. It will be very difficult to buy everybody. But if you leave it to five people [per polling station], hah! You'll see at the voting grounds that they are registering people for Hajj. ‘I'll take you to Mecca.’ And that could change somebody right there who had been supporting you. Or ‘take this key to a motorbike’… All these things happen… But if it's an expanded album [register], I don't think that such things can be done. Nobody has that much resources to do those things.Footnote 15
We expect these two changes to lead women to anticipate that they may perform better in primaries than before for two reasons. First, prospective female aspirants are likely generally disadvantaged relative to male aspirants in their access to economic opportunities needed to amass the private fortunes needed for extensive vote buying (Arriola and Johnson Reference Arriola and Johnson2014). They may also be less embedded in male-dominated party networks, with fewer pre-existing social connections to branch executives that would facilitate the monitoring and enforcement of vote buying in small-electorate primaries. Expanding the electorate likely reduced both of these disadvantages, leading women to expect to perform better.
Secondly, aspirants may expect voters to use gender as an important heuristic for shared preferences,Footnote 16 leading prospective women aspirants to expect to perform better in an electorate with a significantly greater proportion of women voters. Female potential aspirants may also expect that female voters will be less intrinsically biased against them than male voters, all else equal. We do not have direct evidence about these expectations in Ghana. In qualitative interviews, female politicians make clear that they expect women candidates to be held back by damaging sexist attacks from male voters, similar to what Clayton et al. (Reference Clayton2020) observe in Malawi.Footnote 17 Regardless of whether female voters truly prefer female candidates, having more female voters in the electorate should encourage more female aspirants to compete as long as these aspirants expect to receive more support from female voters. We therefore expect the electorate expansion to increase (a) the number of female aspirants and (b) consequently, the probability that the nominee is female.
Based on a similar logic, we expect the electorate expansion to increase (c) the number of aspirants from outside the NDC's core ethnic coalition and (d) consequently, the probability that the nominee is from outside its core ethnic coalition. To the extent that aspirants have an advantage in winning support from voters with whom they share ethnic identities, aspirants from non-core ethnic groups that are under-represented in local party leadership may expect to perform better in primaries with more co-ethnic voters. Anticipating this, more potential aspirants from these groups compete for the nomination. With more aspirants from the party's non-core groups competing, the probability that one is nominated may also increase.Footnote 18
We are agnostic on the reform's effects on (e) the number of aspirants from the NDC's core groups and (f) the total number of aspirants, as these would depend on possible co-ordination by potential aspirants.
However, we expect that reduced feasibility of vote buying in primaries will reduce (g) the probability that the eventual nominee will be a newcomer to politics who has the wealth for vote buying, but not the public reputation and political experience for more general campaigning. Indeed, this is what NDC leaders hoped to accomplish by expanding the electorate. But we also expect the expanded electorate to (h) not reduce the probability that the nominee will be the incumbent. Incumbents have extended opportunities in office to build a reputation for service among local party members through their control of constituency development funds and by providing personal assistance to party supporters (Lindberg Reference Lindberg2010).Footnote 19 Unlike wealthy newcomers, incumbents can fall back on campaigning for re-nomination based on these reputations even if vote buying is more difficult.
Methods
We define the treatment as allowing rank-and-file members in a constituency to vote in primaries, with the party-constituency as the unit of analysis. Our outcomes are the numbers and characteristics of aspirants and the characteristics of nominees, and our estimand is the average treatment effect on the treated. We employ optimal full matching (Rosenbaum Reference Rosenbaum1991; Rosenbaum Reference Rosenbaum2010) to create sets of treatment (NDC) and control (NPP) primaries that are comparable on confounding and prognostic variables.
Optimal full matching creates sets of variable numbers of treated and control units that minimize the overall sum of distances between treated units and their matched control units on the propensity score. Treated units are each matched with at least one control, with each control used only once (Hansen and Klopfer Reference Hansen and Klopfer2006). This is implemented with optmatch v0.9-13 in R v3.6.2.Footnote 20 Treatment effects are estimated by regressing each outcome on the treatment indicator with effect of the treatment on the treated (ETT) weighting, in which each set is weighted in proportion to the number of treated units in the set (Hansen Reference Hansen2004). The units within each set are further weighted so that the weighted average of the outcomes in each set is the difference in means between the treatment and control units in the set.
We produce separate matches for each outcome variable, matching exactly on the 2012 value of each outcome (the lagged dependent variable) in each party-constituency and on a propensity score calculated from the remaining matching variables.Footnote 21 This means, for example, that for the analysis of the effect on the number of female aspirants, all units in a matched set of 2016 NDC and NPP primaries will have had the same number of female aspirants in 2012, when the parties had similarly-sized primary electorates. Past outcomes can be particularly informative for future outcomes since the Ghanaian party system is very stable, with negligible party switching by aspirants unlike in some other African countries. We regularly observe the same aspirants competing against each other in successive primaries.
Each outcome uses a different set of matching variables for its propensity score, but always includes four common variables from Ichino and Nathan (Reference Ichino and Nathan2012). The first is the party's vote share in the previous presidential election, which measures the attractiveness of a nomination to potential aspirants. The second is the party's vote share in the previous parliamentary election. When combined with presidential vote share, this reflects the performance of the party's past parliamentary nominee relative to the party's underlying support, measuring the previous nominee's quality, which would affect the entry decisions of potential aspirants in the next election. The third is population density, a proxy for economic development and a larger pool of potential wealthy aspirants who can buy votes. The fourth is fractionalization among the ethnic groups in the core coalition of the party. Additional matching variables specific to each outcome are described with the results.
The key non-interference assumption would be violated if aspirants sort across parties or constituencies in response to primary rules. But this is unlikely. To be eligible for a primary, aspirants in both parties had to be card-carrying party members in a constituency for four years, joining long before the electorate was changed. Since NPP primaries preceded NDC primaries, the assumption could also be violated if NDC primary voters chose a different nominee based on the outcome of the NPP's primary. But this would only make sense in the small set of very competitive constituencies where differences in the quality of candidates would affect the general election outcome. Primary voters generally lack the necessary information about both aspirants and other voters' preferences to successfully co-ordinate on a particular aspirant; in our extensive interviews, we did not encounter discussion of voters conditioning their choices in this way.
Data
Official results for primaries are not collated or publicly released in Ghana. By combining information from multiple sources, however, we are able to construct a dataset of NDC and NPP aspirants for all primaries that were held by early 2016 (Ichino and Nathan Reference Ichino and Nathan2021). Our final dataset includes 1,532 parliamentary aspirants across 272 constituencies for the NDC and 271 constituencies for the NPP, each out of 275. We define an aspirant as any candidate who publicly stated that he or she would seek a nomination. This includes some who announced that they would do so, but subsequently dropped out of the primary or were disqualified.Footnote 22 We drop some constituencies where we do not feel confident that we have identified all the aspirants, leaving us with 1,494 aspirants across 219 constituencies for the NDC and 252 constituencies for the NPP.Footnote 23 Summary statistics are in the Appendix (pg. A2).
We began with official lists of aspirants on the ballot on primary day in each constituency, but lists were missing for many constituencies for both parties due to inconsistencies in record keeping and primaries delayed by legal disputes. We supplemented the official lists with media reports. From March 2015, when politicians in the NPP began announcing their candidacies, through January 2016, we saved every article mentioning primaries in either party from ten prominent news organizations listed in the Appendix (pg. A4). This yielded 1,950 articles, from which we coded names and biographical information, including professions and past government and party positions. This allowed us to identify many more aspirants not on the official lists. We used a similar procedure to compile data on the 1,285 aspirants from primaries held before the 2012 elections. To add additional biographical details, aspirant names were merged with data from the Parliament of Ghana website and official lists of government ministers dating to 2000. We also supplemented the biographical coding with an in-depth survey of 125 aspirants in the NPP primaries in November 2015. With these sources, the number of aspirants in the NDC ranges from 1 to 9 per constituency, with a median of 3 and mean of 3.5, and the number of aspirants in the NPP ranges from 1 to 8, with a median of 3 and a mean of 2.9.
We also coded each aspirant's ethnicity based on their name, which is easily connected to the main ethnic categories in Ghana. We employed a dictionary of 3,503 names of Ghanaian politicians, using the procedure outlined in the Appendix (pg. A4). We measured constituency demographic characteristics, including the ethnic composition of each parliamentary constituency, using geo-coded enumeration area-level 2010 census data.
Results
Gender
Figure 1 plots the average number of women aspirants and share of nominees who are female, by party, before (2012) and after (2016) the NDC's electorate expansion. It shows an overall increase in the number of NDC aspirants and a substantial rise in the share of NDC nominees who are female, but little change in the number of female NDC aspirants.
Women aspirants and nominees, by year and party
We turn to matching to adjust for potentially important underlying differences across party-constituencies. For the number of female aspirants in 2016, we match on a propensity score using the four common variables described earlier and the population share of Muslims in the constituency.Footnote 24 We exact match on the number of female aspirants in 2012. For whether the nominee is female, we account for a constituency's underlying receptivity to female leaders using the same matching variables as for the number of female aspirants, and again exact match on the lagged dependent variable (female nominee in 2012). For the total number of aspirants, we use only the four common variables and exact match on the total number of aspirants in 2012.
Table 1 presents the number of sets and treated units remaining after matching for each outcome. We assess balance following Hansen and Bowers (Reference Hansen and Bowers2008). We compare treatment and control units within each set with the difference in means for each of our covariates, and assess their combination with a χ2 test. Our matching procedure significantly improves balance (Table 1, rows 1–3).Footnote 25
Overall balance before and after optimal full matching
Note: the fourth and eighth columns show the number of treated units in our sample before and after matching. The last column shows the number of sets created. The number of treated units before and after matching differs when there are no control units available that have the same value of the 2012 outcome.
Effects are estimated by weighted least squares regression, with standard errors calculated as in Hansen (Reference Hansen2004). Our results are at the top of Figure 2. We find that the NDC's reform increased the average number of female aspirants by 0.17 (p < 0.01) and the average number of total aspirants by 0.31 (p = 0.06). These are substantively large differences. In 2012, NDC primaries averaged 0.29 female aspirants out of 2.65 total, or only 11 per cent of all aspirants. An estimated effect of 0.17 additional female aspirants in Figure 2 is 55 per cent of the estimated average effect on the overall number of aspirants (0.31), equivalent to a 42 per cent (4.6 percentage point) increase in the overall proportion of female contestants compared to 2012. There are few historical estimates to benchmark our estimated effect, but it is comparable to those of more intensive interventions designed to increase female candidacy. For example, Bhavnani (Reference Bhavnani2009) finds that reserving city council seats for women in India increased the proportion of candidates who were female in subsequent elections for those seats after reservations were withdrawn by 7.4 percentage points from a baseline of 4.4 per cent.
Estimated average effects on the number of aspirants and characteristics of nominees
Note: the light and dark bars indicate 95 per cent and 90 per cent confidence intervals, respectively.
Furthermore, expanding the electorate led to an 8-percentage-point increase in the probability that the NDC nominee was female (p = 0.001), a large effect compared with a baseline of 9 per cent. In the Appendix (pg. A14), we show that women's advancements are not concentrated in constituencies where the NDC is unlikely to win.
The gender literature suggests two alternative explanations. First, learning about the NDC's reform may have increased women's ambition to seek office. This clearly affected women's ambition for a specific office. But latent ambition is deeply ingrained from early socialization (Fox and Lawless Reference Fox and Lawless2014) and unlikely to change quickly. More importantly, in order to become eligible for the NDC primaries, potential aspirants needed to become active party members in their constituencies at least a year before the electorate expansion was announced. This leaves the pool of potential aspirants largely fixed in the short run. It is more plausible that the expansion encouraged the entry of already-ambitious women who had been developing political careers in their constituencies, but were still undecided about whether to seek this particular office this cycle.
Secondly, personal recruitment (Sanbonmatsu Reference Sanbonmatsu2002) by party leaders or concurrent changes to formal candidate eligibility criteria (Bjarnegård and Zetterberg Reference Bjarnegård and Zetterberg2019) may lead to changes in women's ambition in the short run (Fox and Lawless Reference Fox and Lawless2011; Karpowitz, Monson and Preece Reference Karpowitz, Monson and Preece2017). However, this is also unlikely to account for the results. Starting with the 2012 elections, the two parties adopted very similar efforts to recruit more women to run for parliament, including reducing registration fees by 50 per cent for female aspirants. Moreover, the NPP has been more aggressive in encouraging female aspirants, moving first to reduce fees for women and publicly discussing reserving nominations for women.
Ethnicity
Next, we turn to ethnicity. Figure 3 shows the average number of aspirants and share of nominees from the parties' core and non-core ethnic groups. It indicates a substantial increase in the number of NDC aspirants from outside its core base, but no overall change in the ethnic composition of its nominees.
Aspirants' and nominees' membership in their parties' core ethnic coalition, by party and year
As before, we employ optimal full matching to estimate the effects of the electorate expansion, with balance statistics in Table 1. For the number of aspirants from core and non-core ethnic groups, we exact match on the lagged dependent variable. We add the population share of the largest ethnic group in the constituency, segregation between ethnic groups associated with the NDC and the NPP, and segregation among the sub-groups within each party's national ethnic coalition to the four common matching variables for the propensity score.Footnote 26 Figure 2 shows that the electorate expansion increased the average number of aspirants from ethnic groups outside the party's core coalition by 0.43 (p < 0.01), while reducing the average number of core group aspirants by 0.21 (p = 0.09).
Whether the entry of new aspirants from non-core ethnic groups leads to more of these aspirants winning primaries depends on several factors, including the total number of aspirants, the ethnic demography of party members and the resources available to each aspirant. Although the latter two are unmeasured, we proxy for these factors using the constituency-level population share of the ethnic groups in the party's coalition, spatial segregation between core and non-core groups in the constituency, and the total number of aspirants competing in 2012.Footnote 27 Including these as matching variables, we find that expanding the electorate increased the probability that the NDC nominee is from an ethnic group outside of the party's core coalition by 22 percentage points (p < 0.001). To put this into context, this is larger than the impact of reserving seats for ethnic minority candidates in India (Bhavnani Reference Bhavnani2017).
As with gender, party recruitment is an unlikely alternative explanation for these results. Although recruitment efforts are unobserved, national leaders from both parties have made public efforts to expand their ethnic appeal; the NPP has sought in recent years to shed its label as an ‘Akan party’. If national party leaders favored aspirants from non-core ethnic groups, this is just as likely to have been done in the NPP as in the NDC, which is at odds with our results.Footnote 28
Another alternative explanation is that aspirants from non-core ethnic groups may have won more often after the expansion due to greater co-ordination failures among voters from the party's core groups, not because non-core aspirants became more viable. However, this is inconsistent with our finding that the electorate expansion appears to have decreased the entry of aspirants from the NDC's core ethnic groups (see Figure 2; p = 0.09). The core-group vote likely became more – not less – concentrated, and non-core aspirants could only have won more often if they gained support among primary voters.
Political Experience
Finally, we examine the political experience of nominees. Figure 4 indicates a large drop in the share of NDC nominees who are wealthy newcomers and a smaller increase in the share who are incumbents. We proxy for wealth earned outside government or party positions by coding whether the nominee has a high-level private sector business background (Pinkston Reference Pinkston2016). We code an aspirant as having private sector experience if there is biographical information that he or she has served in a managerial or executive position in any private business or worked in a well-compensated profession (for example, lawyer). Because experience in the private sector and a background in government or party leadership are not mutually exclusive, we code private sector background as 1 only if a nominee had not served as an MP or in other government or party positions. This identifies the wealthy political newcomers.Footnote 29
Political experience of nominees, by party and year
Figure 2 shows that the NDC's electorate expansion made its nominees 11 percentage points less likely (p < 0.001) to be someone with a private sector background, but no prior political experience. For this outcome, we use the four common matching variables and exact match on the lagged outcome variable. Table 1 again presents balance statistics.
While the NDC's expansion of the electorate reduced the success of wealthy political newcomers in primaries, as NDC leaders intended, it did not appear to harm the party's existing MPs. For our analysis of whether incumbents were renominated, we include the population share of the incumbent's ethnic group as a matching variable, since the incumbent may be stronger, regardless of electorate size, in constituencies where he or she is from a more numerous group. Restricting the analysis to party-constituencies with incumbent MPs, we find that the expansion of the electorate led to incumbent MPs being 17 percentage points more likely to be renominated (p < 0.01).
Robustness to Alternative Specifications
Our results are generally robust to alternative specifications. We address concerns about potentially poor covariate overlap by rematching using a caliper that restricts matches to party-constituencies that fall within progressively smaller ranges on the propensity score (Appendix pg. A19). We also restrict the ratio of treated to control units to 10:1 and 1:10 to limit the influence of a single control (treated) constituency that may be matched with many treated (control) units (pg. A17). Neither affects our substantive results, although our point estimates vary somewhat as data is discarded by these procedures. In addition, we show that our results do not depend on matching exactly on the lagged dependent variable (pg. A21).
We also compare our preferred optimal full matching approach to three alternative matching methods: nearest-neighbor propensity score matching, Mahalanobis distance matching and genetic matching. We show that improvements in covariate balance are greater with optimal full matching, indicating that our preferred approach should produce less biased estimates conditional on the choice of matching variables (pg. A22).
Alternative Explanations
We proposed above that these results may result from two changes to the nature of primaries: the shift to a more diverse primary electorate – in terms of both gender and ethnicity – and a reduction in the prevalence of vote buying. While our data do not allow us to directly observe either mechanism, we are able to rule out several alternative explanations that do not operate through changes to the composition of the primary electorate or the extent of vote buying.
First, the NDC may have had more aspirants simply because nominations in the incumbent president's party are perceived to be more valuable. This is unlikely since our analysis only compares nominations with similar demand by exact matching on the number of aspirants in 2012, when the NDC was also the incumbent party and both parties used the same rules. Both parties ran the same presidential candidates in 2016 as in 2012, and the platforms and core bases of each party were unchanged. The main difference leading up to 2016 was the growing popularity of the NPP among swing voters, amid a backlash against the NDC's performance (Bob-Milliar and Paller Reference Bob-Milliar and Paller2018). Any increased expectations of becoming an MP as a member of the president's party should have increased the number of aspirants in the NPP relative to the NDC. But we find the opposite.
Secondly, the NDC's electorate expansion may have created a less certain electoral environment, leading aspirants from under-represented groups who did not believe they would have more support from new voters to enter to take advantage of a more open race. However, the result that more female and non-core group aspirants won primaries suggests that these new entrants did not misjudge their increased viability. Moreover, we find similar results when we exclude aspirants with lower viability, proxied by those who declared their candidacy but dropped out before the primary (Appendix, pg. A9).
Thirdly, the fact that the NPP considered, but rejected, the full expansion of its own electorate in the run-up to the 2012 election might indicate that it is a less democratic party than the NDC and explain why a larger and more diverse group of aspirants competed in and won the NDC's primaries. However, it was easier for the NDC to expand its electorate because the party is less democratic than the NPP and doing so did not require building consensus among multiple factions. The NPP national leadership had been riven by factional disputes, particularly between the two main aspirants for the party's presidential nomination in 2008, 2012 and 2016. One of these factions blocked an identical proposal to expand the NPP's primary electorate to all party members before 2012. The NDC's leadership is more centralized and unified around a powerful general secretary (Osei Reference Osei2016), who has greater discretion to change the party rules. This greater factionalism within the NPP should bias against our finding that more aspirants competed in the NDC: interviews with NPP aspirants suggest that the party's factions sometimes put forward rival candidates in the primaries, which should increase the number of aspirants in our control group observations.
Beyond Ghana
Expanding the primary electorate increased the number of aspirants who are female or members of under-represented ethnic groups in one of Ghana's major parties. This reform also increased the probability that the party's nominees were female or members of these ethnic groups and led to more incumbent MPs being renominated. But it decreased the probability that nominees were wealthy individuals with little political experience. Although we cannot directly test mechanisms, these results are consistent with our interpretation that expanding the electorate made vote buying a less viable strategy and changed the ethnic and gender composition of the electorate, which may have encouraged potential aspirants from groups that became better represented in the electorate to compete for nominations.
Our findings show that changes to primary rules are another institutional means – beyond quotas and reserved seats – to make candidate selection more inclusive. Although expanding the primary electorate has smaller effects than directly reserving seats, its effects are similar in magnitude to the longer-run impacts of seat reservations even after they are withdrawn (Bhavnani Reference Bhavnani2009; Bhavnani Reference Bhavnani2017). It may be easier to expand the primary electorate because it does not restrict competition by men or members of the ethnic majority who may be excluded from reserved seats. By making it more difficult for political outsiders with little reputation among general election voters to buy their way to nominations, this reform also appears to improve candidate quality. This echoes other findings that institutional shifts to improve descriptive representation among candidates can have carry-over effects on candidate competence (Besley et al. Reference Besley2017).
Should we expect similar short-run effects in other settings? Primaries are increasingly common across the developing world (Field and Siavelis Reference Field and Siavelis2008), including in Africa (Choi Reference Choi2018; Warren Reference Warren2018), where parties in at least fifteen countries now elect their legislative nominees through primaries (Ichino and Nathan Reference Ichino, Nathan and Boatright2018). Political parties such as the Botswana Democratic Party have also recently expanded their primary electorates in a similar manner to the NDC (Warren Reference Warren2018).
While it would not be possible to employ our empirical strategy in countries with single-party dominant or more inchoate party systems, our general theoretical answer is yes, conditional on three scope conditions that affect our proposed mechanism for how expanding the electorate affects the calculations of potential aspirants. The first condition is that primaries are real elections: the outcomes are not decided in advance through the manipulation of party elites. The second is that there are gender gaps and ethnic disparities in elite primary electorates compared to the broader party membership. The third is that both inter- and intra-party competition are patronage based and non-ideological. While the first condition may not hold where national party leaders strongly pressure primary voters to back favored aspirants (Choi Reference Choi2018), the latter two conditions should apply widely, especially in Africa (Logan and Bratton Reference Logan and Bratton2006; Riedl Reference Riedl2014).
Where these conditions hold, key differences in party systems may only lead to greater effects than for Ghana. For example, in more inchoate party systems, expanding the electorate may generate higher levels of uncertainty that encourage more non-viable aspirants to compete in the primaries, adding to the effects on aspirant entry. Expanding the electorate in such settings may also increase the likelihood of co-ordination failures within a party's core ethnic groups, adding to the viability of the path to nominations for aspirants from non-core groups. Similarly, the effects on the number and characteristics of aspirants may be even larger where there is greater potential for party switching than in Ghana. If aspirants can easily switch to a political party that expands its electorate to include more female and ethnically non-aligned voters, we expect more potential aspirants from under-represented groups to do so. Future research examining how the strategies of male and non-minority aspirants change when faced with these new competitors and types of voters can also help refine our argument for new settings.
Supplementary material
Online appendices are available at https://doi.org/10.1017/S0007123421000028.
Data availability statement
The data, replication instructions, and the data's codebook can be found at https://doi.org/10.7910/DVN/FYOEWG.
Acknowledgments
We thank Leo Arriola, Rikhil Bhavnani, Rob Blair, Mai Hassan, Valentino Larcinese, John Marshall, Maayan Mor, Alison Post, Rachel Riedl, Carlos Schmidt-Padilla, Pedro Vicente, Anna Wilke, Stephane Wolton, the anonymous reviewers, participants at the 2016 American Political Science Association Annual Meeting, the 2016 Midwest Political Science Association Annual Meeting, the 2016 U-M Africa Social Research Initiative Conference at the University of Ghana-Legon, and the 2016 University of Washington Conference on Voting, Elections, and Electoral Systems, as well as seminar participants at Columbia University, the University of California-Berkeley, Emory University, London School of Economics, Nova School of Business and Economics, and University of Wisconsin for their comments and suggestions. We thank Alhassan Ibn Abdallah, Peter Carroll, Thomas O'Mealia, Lalitha Ramaswamy and Crystal Robertson for research assistance; Dr. Franklin Oduro, Adu Kakra Duayeden, and the staff of the Ghana Center for Democratic Development (CDD-Ghana) for organizing the survey research; and Dr. Philomena Nyarko, Rosalind Quartey, and the Ghana Statistical Service for the census data.
