ELECTORAL INTEGRITY AND POLITICIANS’ RESPONSIVENESS

Electoral accountability models suggest that elections affect politicians’ performance through two distinct but reinforcing principal channels (Ashworth Reference Ashworth2012). First, elections help voters to screen candidates, selecting competent or public-spirited types who voters believe work harder in office (Besley Reference Besley2005; Fearon Reference Fearon, Przeworski, Stokes and Manin1999). Indeed, empirical evidence suggests voters prefer honest and high-quality politicians in diverse settings (Besley Reference Besley2005; Bratton Reference Bratton2013; Galasso and Nannicini Reference Galasso and Nannicini2011).

Second, competitive elections can incentivize officeholders to perform well, irrespective of candidates’ quality, because voters can retrospectively sanction poor performance (Ferejohn Reference Ferejohn1986). While politicians may be self-interested and rent-seeking, expectations of electoral discipline motivates reelection-seeking incumbents to put in optimal effort, choosing a (costly) level of effort to satisfy voters’ endogenously established welfare utility threshold (Mayhew Reference Mayhew1974).

Although the selection and sanctioning models of electoral accountability provide plausible explanations for an incumbent’s performance in office, both models typically assume that elections are run honestly—that the will of the voters is accurately reflected in the results. I argue that both channels of influence can be subverted by election-day fraud and that the extent to which politicians can rig elections influences their incentives to cater to citizens’ demands (Collier and Hoeffler Reference Collier and Hoeffler2015).

Concerning selection, other things being equal, election-day fraud may undermine citizens’ ability to elect politicians who share their interests simply because the candidate most voters cast their votes for is not declared as the winner. Because the “winner” may not share the preferences of voters, (s)he is unlikely to satisfy citizens’ needs. Regarding sanctioning, I argue that incumbents can either rig elections to remain in office or “earn” their reelection by working harder to meet voters’ expectations. Obviously, incumbents can win office using other means such as vote buying, access to more campaign funds, media coverage (incumbency advantage), or obscuring information about their performance. Nevertheless, because officeholders cannot rely on these methods, they often supplement these assets at their disposal with vote rigging. All else equal, I argue that when it is easy for incumbents to engage in election-day fraud, they can reduce the time, personal resources, and amount of effort they devote to address constituents’ needs, and instead pursue their private business activities to earn outside rents.

In keeping with electoral accountability models, my argument implies that the quality of elections may encourage political responsiveness through two theoretically distinct channels. First, if we observe that an incumbent who was elected in a high-quality election (at time t − 1) works harder in office (at time t), it is possible that voters succeeded in selecting a competent candidate who shares their preferences. We can examine this possibility by simply comparing the personal attributes or policy preferences of incumbents elected in cleaner elections to those selected through manipulated ones. Second, incumbents’ expectations of competing in fairer election (at time t + 1) that would strengthen voters’ ability to punish shirking or “select them out” may incentivize higher performance (at time t). Such prospective beliefs may derive from past constraints on rigging or exogenous changes in election administration that would limit fraud.

Also, the latter channel of influence, expectation of sanction, suggests that an incumbent’s incentive to serve the interests of citizens depends not only on being elected in high-quality elections but also on “expecting” to compete in another one (at time t + 1). Even when elected in credible polls, incumbents who believe they can rig future polls may simply shirk their duties.

The effects of fairer elections may vary by levels of electoral competition. In Uganda, Grossman and Michelitch (Reference Grossman and Michelitch2018) find that disseminating information about the performance of politicians broadly to citizens increased officeholders’ efforts, but only in competitive constituencies where viable alternative candidates were available or encouraged by the initiative to enter the race. Accordingly, it is possible that an attempt to improve the quality of elections to induce political responsiveness may be more effective in competitive compared with noncompetitive constituencies. However, if parties and candidates use election fraud and intimidation to maintain their dominance in noncompetitive areas (Magaloni Reference Magaloni2006; Simpser Reference Simpser2008), then improving the quality of elections may motivate incumbents to roll up their sleeves (similar to their counterparts in competitive areas).

Political responsiveness involves doing what citizens want or acting in their interests (Pitkin Reference Pitkin1967; Powell Reference Powell2005). To get reelected, and unable to rig, I argue that incumbents are likely to exert a higher effort to satisfy voters’ priorities. Research on African legislators suggests that citizens prioritize constituency service over parliamentary work (Lindberg Reference Lindberg2010). Therefore, officeholders elected in high-quality elections are likely to increase their supply of constituency service (H1). On the other hand, fairer elections may induce incumbents to reduce the effort they put into parliamentary work (H2), perhaps shifting their efforts to providing constituency service (Ashworth and Bueno de Mesquita Reference Ashworth and Bueno de Mesquita2006). Scholars consider most African parliaments as weak, relative to the executive branch (Barkan Reference Barkan2009). Thus, if fairer elections further encourage legislators to reduce their attendance in parliament and instead dedicate their time to investigating citizens’ local needs and to putting pressure on their local governments or seeking donor funds to provide them, that would be theoretically important to know.

Regarding constituency service, which involves satisfying the nonpolicy concerns of citizens (Fenno Reference Fenno1978), MPs can prioritize the provision of public or private goods to maximize their votes. There is no consensus on which type of good—public or private–is more important in determining vote choice in democracies in sub-Saharan Africa. A dominant narrative of African politics would lead us to believe that to win votes in fairer elections, MPs resort to providing benefits to individuals or groups (Wantchekon Reference Wantchekon2003). Thus, legislators elected in high-quality elections are likely to satisfy citizens’ demand for private goods (H1a). However, emerging research suggests that African voters grant their votes to politicians in exchange for local public goods and services (Ichino and Nathan Reference Ichino and Nathan2013; Lindberg Reference Lindberg2010). Accordingly, high-quality elections would induce MPs to provide more public goods (i.e., local infrastructure) (H1b) including schools, clinics, bridges, and roads that are easily attributable to their efforts and enable them to satisfy more voters (Harding Reference Harding2015). Politicians may also simply provide both public and private goods, targeting the entire population with the former while rewarding supporters with the latter (Diaz-Cayeros, Estévez, and Magaloni Reference Diaz-Cayeros, Estévez and Magaloni2016). I combine original data on legislator spending and parliamentary attendance to examine these theoretical possibilities.

STUDY CONTEXT: GHANA

Ghana is an ideal setting for this study because the level of electoral competitiveness and turnover means that politicians have real incentives to think about how they use their resources when seeking reelection. Similar to many other countries, Ghana adopted multiparty elections in November 1992. Its 2012 general elections, which elected the 2013–17 Parliament, were the sixth. Ghanaian legislators are elected to four-year terms from single-member districts using plurality rule; they are not subject to term limits. Currently, the Parliament is composed of 275 members.^{Footnote 4} During the 2013–16 Parliament, 148 MPs belonged to the ruling National Democratic Congress (NDC), 123 to the main opposition party, the New Patriotic Party (NPP), and one to the People’s National Convention. There were three independent MPs. The NPP and NDC have dominated Ghanaian electoral politics since 1996: the two parties have controlled more than 98% of the seats. However, parliamentary races are increasingly competitive. Between 1996 and 2012, the average vote margin declined by about 11 percentage points (pp).^{Footnote 5} Also, between 2000 and 2012, the average turnover rates for incumbents seeking reelection was 24%.^{Footnote 6}

Despite the increasing electoral competitiveness, Ghanaians provide poor ratings of their legislators. For example, in Afrobarometer Round 6 conducted in 36 African countries, about 63% of Ghanaians said that they disapprove of their MP’s performance compared with the continental average of about 45%. Similarly, 48% of Ghanaians reported that all or most of their MPs are involved in corruption, compared with the continental average of 34%. These poor ratings may be explained by high expectations of Ghanaians of their representatives after more than two decades of democratic elections. However, they may also be due to politicians’ ability to manipulate local elections.

Several studies indicate that the country’s elections are often characterized by some level of fraud and violence (Gyimah-Boadi Reference Gyimah-Boadi2007; Ichino and Schündeln Reference Ichino and Schündeln2012; Jockers, Kohnert, and Nugent Reference Jockers, Kohnert and Nugent2010; Straus and Taylor Reference Straus, Taylor and Bekoe2012). Scholars argue that the enormous benefits and patronage resources that officeholders receive encourage these practices (Gyimah-Boadi Reference Gyimah-Boadi2009; Ninsin Reference Ninsin2016). Election fraud (and violence) routinely goes unpunished. For example, following the 2012 general election, the main opposition party (NPP) filed a petition in the country’s Supreme Court pointing to several irregularities in the polls. Although the Supreme Court acknowledged some of the allegations in its verdict, no election official or party was indicted, and the case was dismissed. Thus, fraud and violence are viable options for officeholders who face stiff competition or seek to ward off strong competitors (in noncompetitive constituencies).^{Footnote 7}

To curb electoral fraud, civil society groups such as the Coalition of Domestic Election Observers (CODEO), with support from international donors, have monitored the country’s elections since 1996. CODEO has observed all of the country’s general and local government elections held since its formation in 2000. The group is composed of 34 independent civil society organizations including religious, professional, and student bodies. In 2012, CODEO deployed about 4,000 observers to polling stations around the country on the election day. Similar to other domestic election observation groups, CODEO’s aim is to promote election integrity and strengthen political accountability. During the December 2012 elections, Asunka et al. (Reference Asunka, Brierley, Golden, Kramon and Ofosu2019) leveraged CODEO’s observation mission to examine the causal effects of election observers on indicators of election-day fraud and violence. Because observers wear uniforms (i.e., official CODEO T-shirts and caps), they are easily identifiable to election officials, party operatives, and voters. Empirical work suggests that the presence of observers deter election-day fraud and violence at polling stations and within constituencies. Because election observation remains a popular approach to promoting election integrity, it is important to examine whether such efforts, by reducing opportunities for fraud, ultimately improve the quality of political representation in new democracies.

RESEARCH DESIGN

Varying Expectations About Future Election Quality

In comparing the performance of incumbents elected in low-AIO with those in high-AIO, I assume that officeholders’ expectations of intensity of monitoring in the next polls coincide with previous levels. I provide evidence in the mechanism section below to show that such an assumption is plausible. However, to test whether such beliefs influence performance, I complemented the initial experiment by dispatching letters to a random set of 30 of the initial 60 MPs (blocking on their initial AIO); MPs in the control group did not receive a letter. In keeping with recent research on monitoring corruption, the letter was designed to alter officeholders’ beliefs about the chances of election-day fraud detection (Callen et al. Reference Callen, Gibson, Jung and Long2016; Olken Reference Olken2007). In consultation with CODEO, the letters stated that evidence from academic research on the country’s 2012 elections shows that more observers in a constituency reduced the overall levels of fraud (i.e., suspicious turnout rates) and voter intimidation. The legislators were then told that, to corroborate these results, I was collaborating with CODEO to replicate the study because, if true, the findings hold promise for democratic consolidation in the country. Appendix B shows copies of these letters.

The letter then informed the MPs that as part of the study, CODEO plans to deploy observers to 80% of polling stations in some constituencies in the upcoming (2016) elections, and that theirs happens to be one of those. The letter was stated in probabilistic terms because the number of observers CODEO could eventually deploy depended on the availability of donor funds, which was not known at the time I circulated the letters. However, I sent out the letters in November 2015 to give incumbents enough time to respond to the treatment in meaningful ways.^{Footnote 10} Indeed, CDF programs take time to implement. For example, Harris and Posner (Reference Harris and Posner2019) find that in Kenya 56% of the projects implemented by MPs using their CDFs took a year, whereas about a quarter took two years. Accordingly, the probabilistic nature of the letter represents a compromise with CODEO and implies that the treatment may be weak. Nevertheless, it provides a useful first step to understand the effects and potential causal channels through which quality elections influence political responsiveness.^{Footnote 11}

I refer to the letter treatment as expected intensity of observation (EIO). I delivered letters to treated MPs in person and read the content of the letter to them.^{Footnote 12} As a reminder, another letter was sent to MPs’ mailboxes (followed by phone calls to confirm receipt) in April 2016. By sending letters to MPs who received intensive and less-intensive observation during the 2012 elections, during their last year in office, my experiments yield a 2 × 2 design with four types of incumbents (see Table 1).

TABLE 1.

Experimental Design

In 2016, the two treatments at times t − 1 (AIO (a)) and t (EIO (l)) generate four sets of MPs (Y _al) represented by the row and column cells of Table 1 as follows:

1. 1. Y ₁₁: MPs elected in high-AIO constituencies in 2012 who received a letter to expect a large number of observers in their constituency in the 2016 elections

2. 2. Y ₁₀: MPs elected in high-AIO constituencies in 2012 who did not receive a letter

3. 3. Y ₀₁: MPs elected in low-AIO constituencies in 2012 who received a letter to expect high IO in 2016

4. 4. Y ₀₀: MPs elected in low-AIO constituencies who did not receive a letter to expect high IO.

This set of potential outcomes allows us to examine causal mechanisms linking intensely monitored election and the responsiveness of officeholders. First, if receiving a letter to expect greater monitoring boosts performance (i.e., E[Y ₀₁] − E[Y ₀₀] ≠ 0, and E[Y ₁₁] − E[Y ₁₀] ≠ 0), it would be consistent with the “expectation of sanction” channel of influence. Second, a higher EIO effect among legislators elected in high-AIO compared with low-AIO would provide further evidence to indicate that being elected in fairer elections and expecting to compete in another one is what drives political responsiveness. By contrast, if the letter does not affect performance, it would provide support for a potential selection effect—that it was the initial selection of an officeholder in high-AIO that mattered.

However, the analyses generated by the two treatments are limited to the final year of each MP’s four-year term and is further constrained by the limited number of cases in each of the treatment conditions (as shown in parentheses). Moreover, if legislators respond to different demands of citizens at various points in the electoral cycle (e.g., providing public goods early in their terms in office and supplying private benefits during an election year), the dimension of incumbents’ efforts that the EIO may affect could differ from that of the AIO that was implemented before the MP took office (Michelitch and Utych Reference Michelitch and Utych2018). However, these analyses provide an essential complement to the primary analysis on whether or not improving the quality of the election at time t − 1 increases the responsiveness of incumbents to explore causal mechanisms. The research design also provides a model for future investigation.

Measuring Politicians’ Responsiveness

To obtain direct measures of politician effort on constituency service and parliamentary work, I use data on legislators’ spending of their CDFs and attendance in parliament, respectively.

CDF spending provides an appropriate measure of responsiveness with which to test my theoretical predictions about the influence of quality elections on different types of constituency services for two reasons. First, MPs have to exert a significant amount of effort to use their funds, and their spending directly benefits members of their constituency.^{Footnote 13} Ghana established its MPs’ CDF in 1992 as part of the country’s District Assembly Common Fund (DACF) [Article 252 of the 1992 constitution]. The DACF represents a proportion (at least 5%) of national revenues disbursed to the country’s 216 local governments for community development projects.^{Footnote 14}

The central government sets aside a portion of the DACF, which it then allocates equally among MPs as CDFs each year. The national Fund Administrator (FA) deposits an MP’s money into a bank account maintained by the local government that serves the legislator’s constituency. To use these funds, MPs need to submit their plans to the local government and satisfy both legal and bureaucratic requirements. For example, to construct a bridge or repair a road in a local community, an MP must submit at least three price quotations from different vendors (Section 43 of the Public Procurement Act 663, 2003). The mayor and the procurement committee of the local government will then approve payment for the winner of the bid. These processes take time and energy.^{Footnote 15} In the case of providing personal assistance such as paying the school fees or medical bills of individual constituents, MPs must write letters providing reasons for the requests and the lists of selected recipients. Because MPs can decide whether to use these funds, the rate of usage provides a reasonable measure of effort. In this regard, this study joins an emerging literature that uses politician spending of CDFs or other central government transfers in their electoral districts as a measure of responsiveness (e.g., Asunka Reference Asunka2017; Harris and Posner Reference Harris and Posner2019; Keefer and Khemani Reference Keefer and Khemani2009).

Second, when MPs decide to use their funds, they have discretion over the allocation. They can either decide to provide public goods or private benefits to their constituents. Analyzing how MPs allocate their funds provides a way to examine which types of citizen demands they prioritize. I consider the proportion of funds that each legislator spends on public goods and private benefits with the assumption that spending on the former is more responsive because it aligns with what majority of voters want.

Between 2014 and 2016, each Ghanaian MP was allocated GHC1,264,987 (≈$316,246).^{Footnote 16} Unspent funds are rolled over to the next year. I gained access to and digitized 36 months of spending records for each of 60 MPs—totaling more than 9,400 ledger records that had been submitted to the FA.

I then constructed a database that contains the totals of how MPs allocated their funds among five principal expense categories: personal assistance to constituents (e.g., school fees, medical bills, business support, and house renovation); local public goods (e.g., construction or repairs of local roads, construction or rehabilitation of schools and clinics, streetlights, and bridges); monitoring of constituency projects and office expenses; transfers toward local government projects and activities (e.g., funds for national independence day or farmers’ day celebration); and donations to support local groups to undertake projects or activities (e.g., traditional authorities, religious groups, and youth associations). A last category of expenditure, which I code as unclear, includes expenses for which the purpose or beneficiary was not clear from the ledger.

Online Appendix I details my coding rules (Table I.1) and gives examples of the expenses sheets (Appendix Figures I.1 and I.2), as well as the summary statistics of these data (Table I.2). In the main analysis, I focus on the total amount of CDF spent and the part allocated to public and private goods to test my hypotheses. Although the other sets of expenses provide further insights on how legislators spend their CDFs in Ghana, these allocations only emerged from my detailed coding and did not form part of my initial expectations. Therefore, while Table 2 shows the summaries of the amounts incumbents spent on these items, Appendix Section E.3 provides detailed discussions of these expenses and results regarding the effects of fairer elections.

TABLE 2.

Average CDF Spending Across Six Expenditure Categories by the Intensity of Election Observation

Notes:

1. Table shows the average amount of CDF funds spent by Members of Parliament (MPs) in the sample between 2014 and 2016 by treatment conditions. Standard deviations are reported in parentheses. Amounts are in Ghana Cedis (GHC) ($1 ≊ 4 in August 2014 according to http://freecurrencyrates.com/en/exchange-rate-history/USD-GHS/2014/yahoo).

2. Source: Author’s coding of original expenditure sheets collected from Ghana’s District Assemblies’ Common Fund Administration.

Also, although I did not physically verify reported projects or surveyed individuals reported to have received personal assistance, MPs do not control these data; the local government that supervises the corresponding legislator report the spending and is directly held accountable for missing funds or mishandling of reports by the country’s auditor general (Williams Reference Williams2017). Thus, these data are reliable for my analysis.^{Footnote 17}

EXPLORING THE CAUSAL MECHANISM BETWEEN ELECTION QUALITY AND RESPONSIVENESS

What might explain the causal relationship between high-quality elections, generated by intense election monitoring, and improved political responsiveness regarding constituency service? I argued that election observation may strengthen the electoral connection by empowering citizens to either select quality candidates who in turn work hard in office, or by shaping incumbents’ expectations that voters might sanction their poor performance. To distinguish between the two channels, I first draw on features of the initial randomization of AIOs, electoral outcomes, and data collected through closed-ended interviews with 47 of the 60 MPs in the study. Second, I report results from randomizing letters to 30 (of 60) MPs in the initial sample that informed them to expect intense monitoring in the next election.

My analyses suggest that the selection of better candidates through intensely monitored election may only play a minimal role in explaining the behavior of incumbents in this case for two main reasons. First, the intensity of election observation was not announced ahead of the 2012 elections when the initial experiment was implemented. Thus, the treatment could not have influenced the candidate pool in the first stage of selection (Besley Reference Besley2005). Appendix Table A.2 shows that an equal number of candidates contested across constituencies in each treatment condition. Second, although we show that the presence of observers reduced the level of fraud and violence at polling stations, I do not believe the effects were sufficient to influence who won the elections (second stage). Although the treatment reduced the vote margins at the constituency level, especially in competitive districts, this reduction was not statistically significant. Also, the treatment did not produce legislators who were qualitatively different, on average, across multiple characteristics such as education, age, party affiliation, or term in office as I show in Appendix Table F.1. While voters’ choices may have been influenced by other candidate features that I do not capture here, based on the empirical evidence presented, the treatment is unlikely to have had a major influence on responsiveness through the selection of “quality” politicians.

Turning to the sanction mechanism, I provide tentative evidence to suggest that it provides a more plausible explanation for incumbents’ behavior in this study. I argue that, for election observation at time t − 1, to affect incumbents’ performance during their terms in office (in time t), at least two conditions must hold. First, incumbents must be aware of the intensity of election observation in their constituencies in the prior election (at time t − 1) and believe that monitoring was effective at reducing electoral fraud. Second, incumbents must believe that the intensity of election observation in their constituencies will be repeated during their reelection race (at time t + 1), thus reducing their ability to rig their reelection balloting.

To test the first condition, I conducted closed-ended interviews with MPs to determine whether they were aware of the intensity of observation in their constituencies. I asked them if they saw observers at polling stations they visited during the 2012 polls. I find a positive association between a higher AIO and MPs reporting that they saw observers. A higher concentration of observers in a constituency increased the probability that an MP reported that he or she personally saw an observer at polling stations they visited by about 17 pp (41.67% in low-AIO compared with 58.82% in high-AIO districts) (p = 0.31).^{Footnote 29} Moreover, MPs elected in intensely monitored elections reported that a higher proportion of polling stations (28%) was monitored in their constituencies, on average, compared with those who had fewer monitors (who reported that only 13% of stations were monitored), which represents a 15 pp increase.^{Footnote 30} These estimates are similar to the concentration of monitors, on average, as shown in Appendix Table F.4. Although these results are not statistically significant at conventional levels, they provide suggestive evidence that incumbents noticed the significant presence of observers in their constituencies during the 2012 election. I argue that this awareness, coupled with the reduction in fraud and violence induced by observers or the significant challenge in executing fraud pose by intense monitoring, signaled to MPs that they could not rely on future rigging.

Testing the second condition, that MPs’ past experiences influence their beliefs about the future, is more challenging. It is not clear that incumbents’ experiences with observers in their constituencies at time t—1 will automatically shape their beliefs about the intensity of observation in time t + 1. Although we can safely assume that MPs would expect some future monitoring in their constituencies because CODEO is credibly committed to observing each election, we cannot be certain about the intensity of observation that MPs would expect. Furthermore, we cannot be sure that these expectations map onto the treatment assignment in the 2012 elections. While the above results on the effect of AIO on politician performance imply incumbents were behaving as if they expected intense election observation, a randomized treatment to manipulate beliefs about future monitoring would allow making causal claims about the expectation of sanctioning mechanism.

To test the causal effect of expected intense monitoring on responsiveness, I analyze the outcome of my follow-up experiment that sent letters to half of the MPs in the sample a year before the 2016 elections.^{Footnote 31} Columns (1) and (2) of Table 4 display the average proportion of CDF spent in 2016 by incumbents according to whether they received a letter to expect more observers, respectively, and by the initially assigned intensity of monitoring. Column (3) shows the conditional effects of the letters by the prior AIO, whereas Column (4) reports the weighted average treatment effect. To assess potential interaction effects, I conduct a difference-in-difference (D-I-D) analysis comparing the impact of the letters among legislators who were elected in low intensely versus high intensely monitored elections. Column (5) reports the results. Panels A, B, and C report the results for the total use of CDF (utilization), public goods, and private benefits, respectively. In all cells, standard errors of the estimates are reported in parentheses.

TABLE 4.

Effect of Expectation of Intense Election Monitoring Conditional on Prior Intensity of Observation

Notes: Table shows the proportion of legislator spending in each experimental cell. It also shows the effect of expectation of intense observation on spending conditional on prior intensity of election monitoring in MPs’ constituencies (blocking variable), and the corresponding weighted ATEs. Randomization inference (accounting for the blocks) is used to generate the standard errors and the one-tailed test of the sharp null of no effect for each MP for the estimated weighted ATEs. 90% confidence intervals are estimated using boostrapping. I conducted 10,000 simulations.

Consistent with expectations, the results presented in Panel A show that receiving a letter increased the proportion of CDF spent in 2016 by 4.9 (s.e. 8) pp (one-tailed test, p = 0.265), on average. The effect of the letter was slightly higher [1.2 pp (12.4) increase] among incumbents elected in low-AIO [5.9 pp (s.e. 8)] compared with those in high-AIO [4.6 pp (s.e. 9.4)] constituencies, but not precisely estimated. Nonetheless, these conditional effects are large (although not statistically significant), representing about a 23 and 10% increase in low- and high-AIO, respectively.

Disaggregating these results into public goods (Panel B) and private goods (Panel C), the latter seems to drive the average increase in CDF spending. Specifically, while, on average, MPs who received a letter did not substantially change their expenditure on public goods, they increased their spending on private benefits. Specifically, the letter treatment increased spending on private benefit by 3.6 (s.e. 3.2) pp (one-tailed test, p = 0.14). For MPs who were elected in low-AIO districts, sending them a letter boosted their spending on individual benefits by 4.9 pp, a 58% increase. Those in high-intensity constituencies spent 3.2 pp more, a 30% increase in their expenditure on private benefits, compared with if they had not received the letter. These weighted average treatment effects masked the potential interaction between the AIO and EIO treatments.

A D-I-D analysis suggests that compared with legislators elected in low-AIO constituencies, the letter caused a further increase in spending on public goods by 4.1 (s.e. 7.8) pp and a decrease in the proportion of funds allocated to private benefits by 1.6 (s.e. 6.6) pp. These results suggest an interaction effect between the high-AIO and the letter treatments to increase spending on public goods further. Specifically, consistent with my argument, these results indicate that legislators who were elected in fairer elections and expected to contest their reelection in another one were more responsive to the priorities of their constituents, which provide tentative causal evidence the threat of sanctioning through fairer election increase democratic responsiveness. However, I acknowledge that a design with greater statistical power is required to confirm these results.

Nevertheless, the fact that the EIO treatment led to an increase in spending on private benefits raises concerns about the potential for fairer elections to promote clientelistic exchanges. However, the timing of the letter treatment may explain this finding (i.e., an electoral cycle effect Michelitch and Utych Reference Michelitch and Utych2018). The letters were sent one year before the elections. Scholarly work on the accountability pressures that legislators face in clientelistic polities such as Ghana suggests that voters demand more private benefits during election years (Lindberg Reference Lindberg2010; Michelitch and Utych Reference Michelitch and Utych2018). Indeed, Appendix Figure E.6 shows that MPs spending on private transfers doubled during the election year (from 6% in 2014, on average, to 12% in 2015 and 2016, a 100% rise), whereas that on public goods decreased (12% in 2014, 32% in 2015 and 24% in 2016, a 25% decrease between 2015 and 2016). Accordingly, the treatment may have further incentivized incumbents to respond to these demands.

I do not find support for two alternative explanations for the effect of the initial treatment on political responsiveness. The first indicates that the AIO treatment may have strengthened citizens’ pressure on the incumbent to supply constituency service. Using Afrobarometer data of respondents sampled from my study constituencies, I show in Appendix Table F.5 that the treatment had no influence on the number of times constituents report to have contacted their MPs, attended community meetings, joined a group to raised an issue, request government action, contacted local government officials, or their beliefs that it the duty of voters to ensure MPs do their work once elected.^{Footnote 32} Second, the AIO treatment did not affect the number of candidates competing in the next election in 2016, which would indicate incumbents’ fear of the entry of new challengers encouraged by cleaner elections (Besley Reference Besley2005; Grossman and Michelitch Reference Grossman and Michelitch2018) (see Appendix Table F.6). Ghana’s strong two-party system ensures that similar number of candidates usually run in each constituency.^{Footnote 33}

SUPPLEMENTARY MATERIAL

To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055419000479.

Replication materials can be found on Dataverse at: https://doi.org/10.7910/DVN/W8HUKY.