Electoral Rules, Political Competition and Fiscal Expenditures: Regression Discontinuity Evidence from Brazilian Municipalities

Abstract

We exploit a discontinuity in the rules of Brazilian mayoral elections to investigate whether political competition has a causal impact on fiscal policy choices. In municipalities with fewer than 200,000 voters, mayors are elected under a plurality voting system. In all other municipalities, a runoff election takes place between the top two candidates if neither achieves the majority of votes. Our results suggest that political competition induces more investment and less current expenditures, particularly personnel expenditures. The impact is larger when incumbents can run for re-election, suggesting incentives matter insofar as incumbents can themselves remain in office.

Acknowledgements

The views expressed in this paper are those of the authors and do not necessarily represent those of the IMF or IMF Policy or of the Brazilian Ministry of Finance. We would like to thank Helios Herrera, Leonardo Rezende, Filipe Campante, Alexandre Samy, Claudio Ferraz, and seminar and conference participants at FGV-SP, PUC-Rio, USP, LAMES 2008, NEUDC 2008, PEG-LACEA, IPEA-RJ, IBMEC-RJ, UFJF and FGV-RJ for useful comments. Edson Macedo, Mariano Lima, João Pedro Borges and Julia Folescu provided excellent research assistance. Datasets used in the paper are available from the authors upon request. Usual disclaimer applies.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. See, for example, Besley et al. (2010), Besley and Case (1995a), Chhibber (1995), Rodgers and Rodgers (2000), Besley and Case (2003) and Gordon and Huber (2007).

2. See, for example, Campante, Chor, and Do (2009).

3. See, for example, Duverger’s Law, which is formally proved in Palfrey (1989).

4. Note that the presence of a runoff would not necessarily rule out a right-leaning party victory in this example. Suppose there are four left-leaning parties each of which receives 15 per cent of the vote, and two right-leaning parties that receive 20 per cent of the vote. Then, a runoff would take place with the two right-leaning parties.

5. The presence of a runoff is more likely to affect the outcome of the election when voters choose their first choice of candidate (sincere voting) than under strategic voting. The political economy literature has interest in sincere versus strategic voting because of their different implications for modelling. Some empirical evidence is available, mainly using structural modelling strategies. See Degan and Merlo (2011) and Merlo (2006).

6. The economic literature has devoted large attention to the role of human capital on economic growth and welfare. See, for example, the AEA’s presidential address by T. W. Schultz (Schultz, 1961) on the foundations of that relationship.

7. RDD was introduced by Thistlewaite and Campbell (1960) but has been widely diffused in the empirical economics literature since the work of Van Der Klaauw (2002), Angrist and Lavy (1999), Hahn, Todd, and Van Der Klaauw (2001), among others. See also the recent work of Imbens and Lemieux (2008) for an extensive survey on RDD.

8. Municipality fixed effects regressions would be unfeasible given the small number of switcher municipalities they have in the data.

9. Starting with Laakso and Taagepera (1979), the political science literature has used the number of effective parties as their main measure of political competition. The number of effective candidates is the inverse of the sum of squared vote shares, that is, the inverse of the Herfindahl-Hirschman Index of vote shares multiplied by 10,000.

10. It is consensual among Brazilian economists (within the mainstream of the profession) that current expenditures are excessive while public investment is at a sub-optimal level. See, for example, Werneck (2008).

11. The last sharp discontinuity in the formula takes place at 156,216 habitants, which involves municipalities that are typically smaller than those with 100,000–300,000 voters. We present in the Supplementary Materials empirical evidence that these transfers are not relevant in our research design.

12. It is possible that, by reducing the probability of remaining in office, higher political competition worsens policies by shortening the incumbent’s horizon. Campante et al. (2009) show that stability can have non-monotonic effects on the quality of policies, which is empirically supported in their cross-country analysis. Cleary (2007), in contrast, assesses through OLS regressions the link between competition and output at the municipal level in Mexico and concludes there is no relationship. In fact, according to Padovano and Ricciuti (2009), there is no unambiguous evidence of this link at the national level.

13. Most of results in the literature linking political competition to policy choices are obtained by identification strategies that are not robust to confounding effects from unobservables. Besley et al. (2010) is an exception and uses an IV strategy. We believe that by using RDD we can make the causal link between political competition and policy choices even stronger.

14. On empirical evidence on visible expenditures, see for instance Akhmedov and Zhuravkaya (2004) and Veiga and Veiga (2007). Construction and infrastructure investments are typically manipulated along the electoral cycle. See, for example, Gonzalez (2002) and Drazen and Eslava (2010).

15. The electorate is composed of three groups. All citizens between 18 and 64 years are automatically registered, and voting is mandatory for these voters. Second, between 16 and 18, registration in optional, but voting is mandatory once registered. Finally, voting is optional for registered voters older than 64 years. Besides fines, sanctions for not voting include becoming ineligible for public sector jobs, passport issuance and, more importantly, government transfers.

16. In 1996, the first round took place on 3 October, and the second on 15 November. In 2000, it took place on 1 October and 29 October. In 2004, the dates were 3 October and 31 October.

17. The total number of municipalities in Brazil is currently 5570. This figure oscillates slightly because of new municipalities, which are normally created by dismembering from another municipality. No municipality around the discontinuity (between 100,000 and 300,000) was dismembered during our sample period, so our results are not affected.

18. Data is available after the electronic ballot was introduced by Law number 9.100, from 1995 onwards. The 1996 municipal elections were the first to have electronic ballots in the vast majority of races.

19. In our sample for 2000, 77 per cent of the incumbents ran for re-election. In 2004, out of those mayors who could run, 91 per cent ran for re-election.

20. There was a significant expansion of education in Brazil during the sample covered in our study. Primary school net enrolment has been about 95 per cent over the last decade. But secondary school net enrolment increased from about 65 per cent in the late 1990s to about 80 per cent in the most recent years.

21. We focus on school construction rather than hospital construction since municipal hospitals for that period were less than a quarter of all hospitals in the country. That makes the construction of a municipal hospital a rare event. That is in contrast with education: around two-thirds of the schools in that period were municipal schools.

22. Investment is a rubric that mostly measures expenditures with infrastructure expansion. School construction is part, but not all, of it. In that sense, these two fiscal variables, investment and school construction are functionally, but not perfectly related to each other.

23. Ministério da Fazenda is the Brazilian equivalent of the Ministry of Finance.

24. We could conjecture only two possible objections to this assumption, neither of which seems relevant. One is that, in anticipation to the possibility of a second round, incumbents would invest more to inaugurate public works between rounds. The runoff takes place approximately three weeks after the first round, so this channel seems quite farfetched, particularly since electoral law forbids inauguration for a period before elections. One could also argue that incentives for accepting lobbying money from contractors are higher in two-round elections because one must finance a longer campaign. However, since rounds are so close in time, the additional campaigning comes at relatively low cost above and beyond that of first-round campaigning. TV advertising is allocated in a centralised manner and is free of charge. Thus, little room is left to spend campaign money between rounds.

25. Starting with Laakso and Taagepera (1979), the literature has used the number of effective parties as their main measure of political competition in the political science literature.

26. The number of observations is lower for the change in municipal schools because data is not available for 1996.

27. It is conceivable to use opinion polls during the administration cycle. However, these polls are not conducted at enough mid-sized municipalities to implement any quantitative empirical procedure.

28. Another proxy for expected competition is past competition. If potential candidates decide to run in the very beginning of the electoral cycle, then arguably political competition in the previous race would be a better proxy. Newspapers start covering discussions within parties on potential candidates a year before elections, but we have no hard evidence. In any case, political competition shows significant persistence over time. Regressing either of our measures of political competition (see below) on her lag yields a very significant statistically estimated coefficient of around 0.4. Sixteen per cent of the variation in political competition is explained by past political competition. Thus, even if, contrary to anecdotal evidence, decisions were made early on, actual future political competition is not a bad proxy for expected political competition.

29. Under continuity assumptions on the conditional expectations of potential outcomes, which can be found in a general framework in Hahn et al. (2001), one could identify the local effect of changes in electoral rules on political competition and fiscal policy. The ratio of these two effects could be interpreted as the causal effect of political competition on fiscal policy on municipalities with 200,000 voters if we interpret political competition as a continuous treatment variable under a fuzzy regression discontinuity design.

30. Results are robust to selecting different subsamples. In fact, when we use narrower samples – for example, 150,000 through 250,000 – we get stronger results. In addition to estimating the models parametrically as described above, we also estimate them nonparametrically using local linear regressions as in Hahn et al. (2001) as robustness check. Our nonparametric estimates reveal that the results are robust to specification.

31. In the Supplementary Materials (Table A.3), we report summary statistics, as in Table 1, but for that subsample.

32. The municipalities do not bear the costs of the election, so there is no reason why crossing the 200,000 voter threshold should impact fiscal policy other than through political competition.

33. It would also be interesting to check how payroll expenditures on teachers behaves as a function of political competition. Unfortunately, only total payroll expenditures are available.

34. We avoid using the sample with all races for estimation since the first stage statistics are not all statistically significant, which could cause the problem of weak instruments.

35. A one-standard deviation change is far for a local impact. Thus, we compute the figures as exact percentage effects, not extrapolations of elasticities that are valid only locally. For the elasticities of investment, current expenditures and payroll expenditures, we use: . For the semi-elasticity of the percentage change in schools we use: .

36 The procedure consists of two parts. The first stage estimates the histogram in Figure 1. The second stage consists of estimating two local linear regressions, above and below the discontinuity point. The percentage of observations in each bin is treated as the dependent variable, and the midpoint of the bins as regressors. See McCrary (2008) for further details.

37. We computed the t-statistic based on the test proposed in McCrary (2008).