Journal of Public Economics 200 (2021) 104455 Contents lists available at ScienceDirect Journal of Public Economics journal homepage: www.elsevier.com/locate/jpube First among equals: The first place effect and political promotion in multi-member plurality elections Dean Dulay a, Laurence Go b,⇑ a Singapore Management University, Singapore b Universitat Autònoma de Barcelona, Spain a r t i c l e i n f o a b s t r a c t Article history: We study the impact of rank-based decision-making in a multi-member plurality electoral system by Received 23 February 2021 examining the decisions of Philippine legislative councilors to run for and win higher office. By focusing Revised 18 May 2021 on multi-member plurality elections, we identify the effect of rank amongst politicians that hold the Accepted 19 May 2021 same office and received a similar number of votes. To identify the causal effect of rank, we conduct a Available online 12 June 2021 close-elections RD at the village, municipality, and province levels. Our main result is the first place effect: incumbent first placers are 5–9% (1–4%) more likely to run (win) in future elections than incumbent sec- JEL Codes: ond placers. The first place effect is unique among rank effects: subsequent rank comparisons yield sub- O10 D72 stantially weaker or insignificant results. Further evidence suggests that a variety of potential mechanisms—party alignment, strategic voting, differential levels of media exposure or the better perfor- Keywords: mance of first placers—do not seem to explain our results. These results improve our understanding of the Rank effect variety of ways rank effects interact with electoral systems. Multi-member plurality Ó 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http:// Political promotion creativecommons.org/licenses/by/4.0/). Regression discontinuity design The Philippines 1. Introduction toral systems. We therefore know little about how rank interacts with political promotion in alternative electoral schemes. Recent scholarship in both economics and political science has This paper examines the effect of rank on political promotion in established the prevalence of rank-based decision-making on a multi-member plurality electoral systems. How rank affects polit- variety of political outcomes, including reelection, the propensity ical promotion in such a system remains an open question, and to contest elections, political appointments and government for- there are reasons to believe that differences may exist. Most obvi- mation (Anagol and Fujiwara, 2016; Fujiwara and Sanz, 2020). Per- ous are the institutional differences between proportional repre- haps the most established empirical relationship in this field is the sentation and different types of plurality electoral systems. effect of electoral rank—based on a candidate’s relative number of Research has pointed to the many differences in both voter and votes in a given election—on political promotion, for example how politician behavior between these two broad systems, such as rank determines who is promoted within a political party or who when voters engage in strategic or tactical voting, the level of runs for higher office in future elections. The relationship between politician’s effort, whether candidates seek minority votes, and rank and political promotion has been documented in diverse con- when candidates cultivate the personal vote (Blais et al., 2001; texts: Swedish and Brazilian councilors (Folke et al., 2016), Norwe- Carey and Shugart, 1995; Cox and Shugart, 1996; Myerson, gian councilors and mayors (Cirone et al., 2021), Finnish councilors 1993). Differences over both the existence of rank effects and the (Merillainen and Tukiainen, 2018), and Slovakian MPs (Crisp et al., mechanisms by which it operates naturally come into question 2013). In spite of their relative geographical diversity, these studies as well. Furthermore, the extant research has identified political focus on the dynamics of rank in proportional representation elec- parties as the primary decision-makers that take rank information and decide on matters of political promotion. Yet political parties are notoriously weak in several countries that hold plurality elec- tions (Hicken, 2009). Moreover, by virtue of its construction, plu- ⇑ Corresponding author at: Universitat Autònoma de Barcelona, Edifici E1, rality systems incentivize candidates to seek the personal vote. Campus de la UAB, 08193 Bellaterra (Cerdanyola del Vallès), Barcelona, Spain The underlying factors that lead to the personal vote—candidate E-mail address: laurence.go@uab.cat (L. Go). https://doi.org/10.1016/j.jpubeco.2021.104455 0047-2727/Ó 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 qualities, qualifications, and record—may thus be mechanisms that offered by our focus on multi-member plurality system is twofold. mediate the effect of rank on political promotion. Finally, the link First, we examine rank effects outside the context of proportional between rank and political promotion is further complicated in representation. This allows us to examine rank dynamics beyond multi-member plurality systems—which ranks matter more than the electoral system that is most often studied. Second, multi- others?. member plurality elections allow us to examine rank effects in a We thus need a setting that employs a multi-member plurality context where both candidates win and have similar formal polit- electoral system and can clarify the underlying conceptual differ- ical power while their ranks may differ. This allows us to build on ences outlined above. Legislative elections in the Philippines pro- work, such as Anagol and Fujiwara (2016) that examine rank vide this ideal context. Legislators are elected through multi- dynamics in single-member plurality.1 Rank effects still exist in member plurality, where rank varies among legislators (depending multi-member plurality systems, yet the particular forms in which on who gets more votes), while their institutional responsibilities they manifest themselves—the first place effect—and the mecha- remain similar. Furthermore, the weakness of party alignment in nisms by which they come about—not through parties—differ from the Philippines opens up opportunities to explore alternative existing work. While this paper does not ultimately pin down a mechanisms. We use three tiers of legislatures—village, municipal- mechanism, we discuss some possibilities and provide some sugges- ity, and provincial councils. We examine whether the rank of tive evidence against other plausible mechanisms. These results are incumbent councilors—placing first, second, and so on in a given the first step towards a deeper understanding of the impacts of rank- election—increases their probability of running for and winning based decision-making. higher office in a future election. Examining three different tiers of government shows not only the validity of our results, but also that they are robust throughout a country’s political system. Estab- 2. Institutional background lishing the rank effect can be challenging, as comparing first and second placers suffers from selection issues: higher ranked candi- 2.1. Local Governments in the Philippines dates may have more resources, better innate ability or larger net- works. To address these endogeneity concerns, we employ a This paper examines Philippine elections at the province, regression discontinuity design on close elections between first municipality, and village levels. We focus on how rank in multi- and second placers. Because of the political structure of local coun- member legislatures affects the probability of running for and win- cils, first and second placers are both elected to the same office— ning higher office. In particular, we examine whether village coun- providing us with the ideal setting to isolate the impact of rank cilors run for village chair, municipal councilors run for municipal on future political outcomes. vice mayor, and provincial councilors run for provincial vice gover- Our results provide evidence for a novel empirical association nor. A summary of the relevant information on these positions is between rank and political promotion in multi-member plurality given in Table 1. systems. We call this the first place effect: the effect of ranking first We begin by describing villages, the smallest unit of local gov- (versus second) on the probability of running for and winning ernance in the Philippines. There are 42,046 villages throughout higher office (our measure of political promotion). First place coun- the Philippines, with an average population of 2,400 inhabitants. cilors in a close election are 5–9% (1–4%) more likely to run for Each village is headed by the punong barangay (village chair), (win) higher political office in the subsequent election—a result charged with general executive and legislative functions such as that holds in all levels of local government. First place village coun- drafting and implementing ordinances and resolutions, enacting cilors are more likely to run for village chair, first place municipal the budget, as well as delivering basic goods and services (LGC councilors are more likely to run for vice mayor, and first place Book 3 Title 3 Chapters 3 and 4). All villages have one legislative provincial councilors are more likely to run for vice governor. district and the sangguniang barangay (village council), presided These findings suggest that rank has a significant impact on future by the village chair, is composed of 7 regular council members. electoral success, increasing the probability of first placers running The village council is tasked with creating ordinances and resolu- for and winning higher office by at least 30% more than that of the tions for their village constituency. Village elections take place second placer. The first place effect is robust to different specifica- every three years. All village positions face a three-term limit tions and bandwidth values, and satisfies covariate balance. Fur- regardless of the length of term. thermore, we show that the first place effect is unique, the Each village has exactly one village chair and 7 elected coun- empirical link between rank and promotions becomes significantly cilors. The position of village chair is elected through single- weaker once we compare second place to third place candidates member plurality voting, whereas the village councilors are elected (around 1–2%), and becomes more insignificant for subsequent through multi-member plurality voting. For the village council comparisons. Third, we provide suggestive evidence that party elections, each voter may vote for as many as 7 councilors, and organization (and hence party-based decision-making) does not the 7 councilors with the highest number of votes obtained com- serve as the mechanism that links rank and political promotion. prise the village council. Village candidates are not affiliated with We provide a test for party alignment between the councilor and any political party, so voters can vote for individual candidates.2 the politician occupying higher office or between the first and sec- Note that the multi-member plurality council elections will be the ond place councilors, and find that rank effects are not stronger for unit of analysis when we consider rank effects. councilors from the same party. In addition, we explore whether A collection of villages forms a municipality. There are 1,634 first placers receive more campaign funds from their party than municipalities in the country with an average population of second placers, and find no significant differences. These results roughly 61,800 inhabitants. The executive branch of the munic- suggest that candidates’ decisions to run are not subservient to ipality is headed by a mayor and vice mayor, whose roles and party decisions, a result that differs from previous papers. Finally, responsibilities include enacting the budget, providing public we provide evidence suggesting that alternative mechanisms— 1 such as voter coordination and media exposure—do not mediate Our work is also closely related to Pons and Tricaud, 2020 who show that in multi-election races, higher marginal rank in the first round of the election makes a our results. candidate more likely to continue to vie for political office. Our results add to the existing literature on the link between 2 The Omnibus Election Code stipulates that ‘‘barangay (Village) elections shall be rank and political promotion in particular, and the effect of rank non-partisan” and ‘‘no person who files a certificate of candidacy shall represent or allow on political selection in general. The primary analytical leverage himself to be represented as a candidate of any political party or any other organization’’. 2 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Table 1 Political System in Philippine Local Government. Level of Executive Executive Legislative Legislative Number of Constitutional Government Position/s Voting Position Voting Council Roles of System System Seats Council Members Village Chair Single-member Councilor Multi-member 7 Same across plurality plurality ranks Municipality Mayor and Single-member Councilor Multi-member 8 Same across or City Vice Mayor plurality plurality ranks Province Governor and Single-member Board Multi-member 10 Same across Vice Governor plurality Member plurality ranks Note: In the remainder of the paper, we use the terms provincial board members and provincial councilors interchangeably. In addition, we use municipalities to refer to both cities and municipalities to simplify the exposition. goods, and implementing ordinances and resolutions (LGC Book The executive positions are often thought of as the most 3 Title 3 Chapter 3 Articles 1 and 2). The municipality’s legisla- desirable local office, owing to their significant control of the ture is headed by the municipal council. The municipal council is budget and thus over the choice of public works projects. It is composed of 8 councilors, also elected via plurality-at-large vot- thus common for legislators—the village, municipal, and provin- ing, and with each councilor holding a constitutionally equiva- cial councilors—to aspire for their jurisdictions’ executive offices. lent position.3 Municipal boards are in charge of passing and That is, village councilors aspire to be village chairs, municipal approving ordinances and resolutions (LGC Book 3 Title 3 Chapter councilors seek the mayorship and vice mayorship, and provin- 3 Article 3). Mayors, vice mayors and municipal council members cial councilors hope to become governors and vice governors. have three year terms with a term limit of three consecutive This claim is supported by qualitative scholars of the Philippines. terms. For example, Sidel (1999) argues that the position of mayor is a Voting at the municipal level goes as follows: Municipal mayors major prize, the benefits associated with it being ‘‘the awarding and vice mayors are elected via single-member plurality voting, of building permits, the passage of municipal zoning ordinances, whereas the municipal council members are elected through the use of government-owned land, the allocation of public works, multi-member plurality voting, with voters voting for as many the approval of reclamation projects. . .[and] the awarding of petty councilors as the number of municipal council seats. Council votes monopoly franchises and concessions”. It is also worth noting that are cast for individual candidates and not for a set of party a councilor ‘‘jumping the line” is uncommon—only 0.02% of vil- candidates. lage councilors run for municipal mayor or vice mayor and only Several municipalities in turn comprise a province. Provinces 0.01% of municipal councilors run for provincial governor or vice are the most aggregated local government in the Philippines. governor. There are 81 provinces in the country with an average popula- In summary, the Philippine political system is an appealing con- tion of 1.25 million people. The executive branch of the province text for our study and clearly provides a structure that lends itself is led by a governor and a vice governor. Among their responsi- to testable hypotheses. First, councilors are of varying rank while bilities include enacting the budget, appointing provincial offi- having similar institutional powers, thus controlling for institu- cials, and implementing ordinances and resolutions (LGC Book tional differences that may confound our results. Second, the 3 Title 4 Chapter 3 Articles 1 and 2). The legislative branch of appeal of local executive positions makes them the obvious next the province consists of the provincial council. Its responsibilities step for councilors. The choice of outcome variable—running for include the approval and passing of new ordinances and resolu- and winning executive office—is thus straightforward. Third, the tions as well as licenses (LGC Book 3 Title 4 Chapter 3 Article 3). personalistic nature of Philippine politics allows for the testing of The council is composed of 10 members, elected via a plurality- alternative mechanisms beyond selection by political parties. at-large voting procedure and with constitutionally equivalent functions and responsibilities.4 Governors, vice governors and provincial council members have three year terms with a three 3. Data consecutive term limit. Voting at the province level follows similar processes as in the The dependent variables of interest are whether a candidate municipal and village levels. The governor and vice governor are runs in, and wins, a subsequent election for a higher position. elected through single-member plurality voting, whereas the Table 1 shows the executive positions that the corresponding provincial councilors are elected via multi-member plurality vot- legislative councilors run for. The elections data do not contain ing, again with as many votes per voter as council seats. The per- numeric politician identifiers, so individuals are determined sonalistic character of voting in local elections belies the unstable using a name matching technique that matches both first and nature of political parties in the Philippines (Hicken, 2009). Candi- last names of candidates within the same province. This is done dates consistently shift from one party to another across electoral for two reasons. First, we match at the province level because it years, and sometimes even during a single election season. is very uncommon and costly for candidates to run in provinces different from their own. Second, we do this to prevent false matches—the larger the constituency considered, the higher 3 While municipalities have 8 councilors regardless of size, cities may have from 10 probability of having similar names. In Appendix A, we consider to 36 councilors. As before, the same voting rules (i.e. multi-member plurality voting) other more stringent matching techniques where we match, in apply. addition to first and last names, the suffix, the middle initial 4 Although most provinces have 10 council members, the number of council seats or the middle name. Results are robust to these stricter match- varies according to their size: ranging from 6 for the smallest provinces to 16 for the largest provinces. But even these cases follow the same voting procedures as do ing rules. conventional councils: voters may vote for as many councilors as there are council The main independent variable is a candidate’s rank, which is seats in their respective districts. based on the total votes obtained in the candidate’s respective leg- 3 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 islative district. We exclude from our analysis cases where multi- Candidacyiptþk ¼ a þ bFirst it þ f ðMV it Þ þ Firstit  f ðMV it Þ þ it ple candidates tie for the same rank.5 Electioniptþk ¼ a þ bFirst it þ f ðMV it Þ þ Firstit  f ðMV it Þ þ it Our elections data contain the name, constituency, votes 8i s:t: MV it 2 ½h; h obtained and party affiliation (except for village level) of all candi- ð1Þ dates. Our data spans province and municipal elections from 1988 to 2016. Provincial and municipal elections are held every three where Candidacyiptþk (Electioniptþk ) is 1 if politician i runs for years since 1992. The provincial dataset contains 3,274 observa- (wins8) higher office p in election year t þ k9, Firstit is 1 if politi- tions, while the municipal dataset contains 29,294 observations. cian i is ranked first (0 if ranked second) in election year t; MV it Village elections are also held every three years, although they is the margin of victory between the first and second placers have been postponed several times in the past. Given that earlier and f ðÞ are linear, quadratic and cubic functions of the running village elections are done manually, election records are only avail- variable. The optimal bandwidth value h is given by the band- able from 2002, 2007, 2010 and 2013. However, because only the width selection procedure used: Calonico et al. (2014) and winning candidates are listed in the 2002 and 2007 election Imbens and Kalyanaraman (2012). Robust standard errors are clus- returns, we drop 2002 from our regressions as we are unable to tered at the level of the constituency. accurately construct the variable for the probability of running in the subsequent election. The final village dataset contains 5. Empirical results 160,192 observations.6 Auxiliary datasets are discussed in Appendix B. Fig. 1 shows the existence of first place effects in all three levels of local government in the Philippines. The x-axis variable, ‘vote share difference between first and second’, is equal to either the first 4. Identification strategy placer’s vote share minus the second placer’s vote share for the first placer (which is positive and to the right of the cutoff), or Establishing rank effects—the impact of rank on the probabil- the second placer’s vote share minus the first placer’s vote share ity of running for or winning higher office—is the central contri- for the second placer (which is negative and to the left of the cut- bution of this paper. In particular, we provide evidence for the off).10 The y-axis is either the probability of running for (blue circles first place effect, where we show that, holding other factors con- and curves) or winning (red triangles and curves) higher office in the stant, councilors who rank first are more likely to run for and next election. Graphical analysis provides suggestive evidence of a win higher office than councilors who rank second. However, discontinuous and large increase in the probability of running for comparing first and second placers and their political outcomes and winning higher office for first placers compared to second presents an identification challenge and suffers from selection placers. concerns. Candidates who finish first may garner the most votes Table 2 presents estimated coefficients from Eq. 1. The depen- due to social networks (Cruz et al., 2017), resources (Gerber, dent variables are either the probability of running for 1998), innate ability (Mondak, 1995) and other factors (Lee, (Candidacy; t þ k) or winning (Election; t þ k) higher office—vice 2008; Ansolabehere et al., 2001). In this case, the ideal experi- governor for the province, vice mayor for the municipality and vil- ment is to take two otherwise similar individuals, randomly lage chair for the village. Consistent with regression discontinuity assign them the label of being first and second and observe their designs, we show robustness of results by varying our choice of eventual political decisions. the control function f ðÞ and the bandwidth h. Columns 2 and 3 To mimic a randomized design, we employ a two-pronged use the full sample and include quadratic and cubic functions of identification strategy that exploits a unique institutional feature the running variable. For specifications using optimal bandwidth of the Philippine local government and utilizes the standard procedures, Columns 5 and 7 estimate local linear regressions with close elections regression discontinuity approach. First, we only the Calonico et al. (2014) and Imbens and Kalyanaraman (2012) consider elections for positions with at least two winners to selection methods. Columns 4 and 6 provide the optimal band- guarantee that rank, not other extraneous factors, is driving width value for the respective methods. To make sense of coeffi- our results. Second, following the rank effects literature, we cient magnitudes, Column 1 shows the second place mean, compare candidates who barely ranked first with those who which is the estimated value of the dependent variable for the barely ranked second. This identification strategy allows us to close second placer.11 Finally, the number of observations is skirt around selection issues and compare two individuals with reported in brackets. ex ante similar characteristics.7 Table 2 corroborates the graphical evidence: simply being Studying all three levels of local government in the Philippines, labeled first versus second significantly increases the probability we estimate the following regression at the provincial, municipal of running and winning for all levels of government. In other and village levels. In each scenario, politician i refers to a provin- words, keeping other factors constant, current rank matters signif- cial, municipal or village councilor while position p refers to the icantly in increasing the future likelihood of running for and win- office of the provincial vice governor, municipal vice mayor or vil- ning a higher position. lage chair, respectively. 5 Including elections with tied candidates in the analysis does not change our results. The proportion of elections for which candidates tied for first is negligible: 8 0.05% for provincial elections, 0.3% for municipal elections and 2% for village elections. The Electioniptþk variable is defined as winning not conditional on running. Hence, 6 Specifically, the provincial dataset includes 9 election cycles for each of the 81 the dummy variable is equal to one when politician i wins, and zero if politician i loses provinces, with 6,469 candidates running, of which 3,274 are first and second placers. or does not run for higher office p. 9 The municipal dataset contains 9 election cycles for each of the 1,634 municipalities, As a general rule, k = 3, although there are years when village elections are held with 119,670 candidates running, of which 29,294 are first and second placers. The after 5 years. 10 village dataset covers 2 election cycles for each of the 42,046 villages, with 573,908 Because of our multi-member plurality setup, we define vote share as the number candidates, of which 160,192 are first and second placers. of votes obtained by the first or second place candidate, divided by the total number 7 Owing to this design, bunching at the cutoff is not possible. Specifically, for every of votes garnered by all candidates. 11 first place candidate, we have a counterpart second placer, which implies that we We follow the definition from Anagol and Fujiwara (2016), who use a linear cannot have imbalance in this empirical setup. In Fig. C1, we show that observations specification and the Imbens and Kalyanaraman (2012) bandwidth to compute for are symmetric around the cutoff. this mean. 4 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Fig. 1. First Place Effect Note: Horizontal axis shows the vote share difference between the first and second placers (i.e. a positive vote margin implies ranking first and a negative vote margin implies ranking second). Each circle (triangle) corresponds to the unconditional mean of running (winning) within 0.005 or 0.001 intervals of the vote margin. Solid curves are smoothed local polynomials on either side of the discontinuity. Table 2 First Place Effect. Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) Province Candidacy, t þ k 0.056 0.0845 0.0868 0.046 0.0925 0.013 0.0472 [N = 3,274] (0.0148) (0.0178) [N = 2,614] (0.0180) [N = 1,356] (0.0338) Election, t þ k 0.021 0.0356 0.0422 0.039 0.0398 0.012 0.0265 [N = 3,274] (0.0115) (0.0133) [N = 2,458] (0.0148) [N = 1,260] (0.0228) Municipality Candidacy, t þ k 0.122 0.0805 0.0749 0.014 0.0643 0.006 0.0546 [N = 29,294] (0.00618) (0.00656) [N = 23,258] (0.00812) [N = 14,972] (0.0110) Election, t þ k 0.057 0.0383 0.0287 0.012 0.0175 0.006 0.0113 [N = 29,294] (0.00465) (0.00520) [N = 21,998] (0.00651) [N = 14,384] (0.00875) Village Candidacy, t þ k 0.091 0.0674 0.0636 0.025 0.0614 0.008 0.0562 [N = 160,192] (0.00228) (0.00245) [N = 134,472] (0.00302) [N = 70,704] (0.00485) Election, t þ k 0.035 0.0320 0.0289 0.026 0.0278 0.008 0.0236 [N = 160,192] (0.00167) (0.00179) [N = 135,248] (0.00213) [N = 71,446] (0.00342) Note: Robust standard errors clustered at the constituency level are in parentheses. Dependent variable is the probability of running for or winning higher office: vice governor for province, vice mayor for municipality and village chair for village regressions. The unit of observation is a candidate. Columns 2–3 use the full sample with quadratic and cubic control functions. Column 5 uses the Calonico et al. (2014) selection method while column 7 uses the Imbens and Kalyanaraman (2012) selection method. Columns 4 and 6 provide the optimal value for each procedure. The number of observations for the full sample and bandwidth methods is reported in brackets. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. First place effects on running for the provincial level range from 6. Robustness checks 5–9% (from a second placer’s baseline mean of 6%) while those on winning range from 3–4% (from a base of 2%). Municipal level coef- 6.1. Covariate balance ficients are slightly smaller than the provincial estimates: coeffi- cients on running span from 5–8% (from a base of 12%) while As an indirect test of our identification assumption, we check coefficients on winning go from 1–4% (from a base of 6%). For the whether first and second placers differ on predetermined variables. village level, the coefficients are also slightly smaller than those The idea behind this placebo test is to investigate whether preex- for the municipal level: estimates for running wander around 6– isting differences in variables may be explaining the differences 7% (from a base of 9%) and estimates for winning vary from 2–3% in post-treatment outcomes. Graphical evidence shows that (from a base of 4%). covariates are smooth around the cutoff point (see Fig. C2). More The following are key observations from our empirical results. formally, Table C1 presents results for regressions with the follow- First, the estimated coefficients are stable, statistically significant ing as the dependent variables: probability of running and win- and generally robust to using different control functions and band- ning, vote share and membership in the party of the politician in width selection methods. This holds for all levels of local govern- higher office in the previous election, as well as past candidacies ment. Second, the coefficients are substantively meaningful. For (i.e. number of times the candidate ran in the past). Reassuringly, example, the provincial estimates show almost a doubling of a can- all variables are balanced and support the validity of our empirical didate’s propensity to run simply by being labeled first, with strategy. municipal and village level estimates showing a 45–74% increase in this probability. These results provide evidence for the individual 6.2. Robustness to bandwidths effect of rank while controlling for other confounding factors. This significant and robust finding adds to and complements the exist- Fig. D1 shows how our coefficient estimates on candidacy and ing body of results in the rank effects literature, showing the impact election vary with the chosen bandwidth value. For a wide range of rank on political promotion in multi-member plurality systems. of values, our estimates are relatively stable and remain statisti- 5 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 cally significant for all levels of government. As expected, results ning for and winning a higher position.14 In this subsection, we pre- on running are more robust to bandwidth values than the results sent results for rank effects on: (i) promotion to the highest position on winning. While our earlier tables presented estimates using within a constituency, and (ii) re-election to the same position. only the optimal bandwidths, this exercise proves that for many Instead of considering the immediate position above the coun- other bandwidth values, our results continue to hold. cilor, we now test whether the same rank effects are present when provincial and municipal councilors run for the highest executive 6.3. Identifying politicians position: the provincial governor or the municipal mayor.15 In Table F1, we show that at the provincial level, councilors do not seek Consistent with literature in political economy (Querubin, the office of the governor, and there is no difference in running for or 2016), we identify politicians across time by matching names. A winning as governor across ranks. While we see statistically signifi- potential issue is that our results may be affected by name mis- cant rank effects for municipal councilors running for mayor, the matches. For example, without the suffix name, we may consider magnitudes are much smaller (around 1–2%) than those for vice both father/uncle and son as one, potentially causing overesti- mayor (5–8%). This result implies that while the first place effect mates to the first place effect if we think that males are more likely leads to political promotion, a councilor’s prospects is still limited to rank first.12 To address this and other possible sources of mis- by the hierarchy of political offices—they cannot make too big of a match, we perform two checks. First, we use more stringent match- jump from municipal councilor to mayor or from provincial coun- ing techniques, where in addition to matching first and last names, cilor to governor. we also match the suffix name, the middle initial or the middle Next, we explore the case for re-election, when councilors run name. Tables A2, A3 and A4 show that our results are robust to these for the same position in the succeeding election. Table F2 shows matching techniques. This implies that while mismatches may occur, that rank effects are negative when considering the same position: they do not significantly impact the results that we find. Second, we first placers are less likely than second placers to run for and win identify the most common last names and drop observations with the same position in the following election. Combining this with those last names. The idea behind the test is that mismatches occur our main finding, this suggests that political promotion is a path more frequently in individuals with less unique last names. For reserved for first placers, leaving second placers only with pro- example, for politicians with the popular surname ‘Dela Cruz’, it is spects for political re-election. possible that different individuals are mistakenly considered one and the same. Dropping individuals with common surnames from the sample, we find that our results are robust to this exercise (see Table A5). 8. Discussion of potential mechanisms 7. Other results This section provides a brief discussion of the potential mecha- nisms driving the rank effect. While we do not pin down the pri- 7.1. Other rank effects mary mechanism in this paper, a clear statement of the underlying mechanisms at play will be beneficial for future work. In this subsection, we examine the effects of other ranks on First, one potential mechanism is that rank effects are a product political promotion. Exploiting the multi-member nature of coun- of intra-party decision-making (Folke et al., 2016; Merillainen and cils in the Philippines allows us to uncover the rank effect not just Tukiainen, 2018; Cirone et al., 2021). Parties use rank as a way to for the first placer but also for lower ranked candidates. determine who is popular, and hence who to promote within the For example, in the seven-member village council, we can study party. At the conceptual level, there is little evidence that parties rank effects comparing second and third placers, up until sixth and in the Philippines exert top-down organizational control. Rather, seventh placers. In Fig. 2, we show rank effects between nth and party decision-making is the result of a bottom-up process that aggregates the preferences of powerful political families operating n þ 1th placers for n > 1.13 Effects see a sharp drop after n ¼ 1: that within the party (Hicken, 2009). We also do not see empirical sup- is, first place effects for running and winning are significantly greater port for such a hypothesis. We test for potential coordination by than other rank effects. Other rank effects remain generally statisti- political parties by including a control for party alignment and its cally significant, although the effect significantly diminishes as one interaction with the rank variable. We define party alignment in goes down the rank. The size of the constituency seems to be nega- two ways: (i) whether the provincial (municipal) councilor belongs tively associated with the presence of other rank effects: for the to the same party as the vice governor (vice mayor), and (ii) provincial level, only the first place effect exists; for the municipal whether the first place councilor belongs to the same party as level, all but four rank effects are present; for the village level, all the second place councilor. Table 3 shows that this mechanism is rank effects are positive and significant. We therefore conclude that seemingly not the driving force behind the rank effect since the while some ranks matter, the first place effect is 3–6 times larger interaction of rank and party alignment is not statistically signifi- than the second place effect, and an order of magnitude greater than cant in most specifications. Even with the inclusion of the party the rest. variable, the rank effect remains robust and significant. We also explore whether parties are more likely to financially support first 7.2. Other positions placers than second placers. To test this, we collect a novel dataset on campaign contributions and expenditures, where we use data Are there rank effects for running and winning for other posi- on contributions from political parties as our main dependent vari- tions? So far, we have centered our discussion on rank effects on able. We find that first placers do not receive more donations from political promotion, where ending up in a higher rank leads to run- their party in the subsequent elections than second placers (see Table G3). While this evidence does not rule out all the ways by 12 A suffix name is normally used for male children or parents who share the same which parties may influence rank effects, we consider this prelim- first and last names—Sr. for Senior, Jr. for Junior, II, III and so on. 13 Alternatively, we present the corresponding RD figures in Appendix E. 14 15 We base our definition of the ‘higher position’ on the constitutional rule of For the village, there is no higher position than the village chair, which subsumes succession. For example, when municipal vice mayors vacate their position, the the responsibilities of the local chief executive and the head of the local village highest ranking councilor replaces the vice mayor. council. 6 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Fig. 2. Other Rank Results Note: Dependent variable is the probability of running for or winning higher office. The horizontal axis shows the comparison between consecutive ranks. The blue circle is the coefficient estimate, with the thick blue bar showing the 90% confidence interval and the thin blue bar showing the 95% confidence interval. inary evidence that implies that rank effect may be explained by coordination (broadly defined), such that the higher the turnout the mechanisms beyond party alignment. greater the level of coordination (Arias et al. (2019); Rolfe (2012), An alternative mechanism is voter coordination. Voters use Rolfe (2012); Cruz et al. (2020), Cruz et al. (2020)). Table G1 shows rank as a signal to coordinate on popular candidates (Anagol and that the first place effect does not differ heterogeneously with Fujiwara, 2016).16 Candidates know this, and hence choose to run. respect to voter turnout. The coefficients on voter turnout and its Such a mechanism would support a growing literature that links interaction with rank are not statistically significant and relatively the spread of information provision with the ability to coordinate small (in absolute value).17 In our setting, voter coordination does over preferred candidates (Adida et al., 2020). We test this indirectly not seem to explain the existence of the rank effect. As a caveat, it by comparing the rank effect in areas where voter coordination may have greater impacts on politicians’ decision-making. Following the extant literature, our control variable voter turnout proxies for voter 17 Because the voter turnout variable is only available for half of our sample, our province-level results suffer from smaller sample sizes and present null rank effects. 16 Anagol and Fujiwara (2016) suggest rational voters strategically coordinating to Even if the inclusion of voter turnout may have weakened the rank effect, that the vote for the runner-up as a mechanism to explain why second place candidates are turnout variable itself is not significant seems to suggest a minimal, if any, role of substantially more likely to run in and win first-past-the-post elections. voter coordination in driving our results. 7 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Table 3 Party Alignment. Full Sample Optimal Bandwidth Quadratic Cubic Linear Linear (1) (2) (3) (4) Board Member & Vice Governor in the Same Party First 0.0625 0.0697 0.0785 0.0323 (0.0205) (0.0243) (0.0259) (0.0445) Same Party 0.0237 0.0169 0.0149 0.0961 (0.0164) (0.0198) (0.0206) (0.0337) First  Same Party 0.0460 0.0407 0.0363 0.178 (0.0304) (0.0388) (0.0412) (0.0595) Second Place Mean .042 .042 .042 .042 Bandwidth 1 1 CCT=.046 IK=.013 N 3,130 3,130 2,484 1,278 Board Members (First and Second Placers) in the Same Party First 0.0931 0.0887 0.0942 0.0159 (0.0291) (0.0339) (0.0332) (0.0518) Same Party 0.0176 0.0230 0.0221 0.0264 (0.0198) (0.0227) (0.0236) (0.0394) First  Same Party 0.0127 0.0029 0.0025 0.0578 (0.0378) (0.0434) (0.0424) (0.0560) Second Place Mean .056 .056 .056 .056 Bandwidth 1 1 CCT=.046 IK=.013 N 3,274 3,274 2,614 1,356 Full Sample Optimal Bandwidth Quadratic Cubic Linear Linear (1) (2) (3) (4) Councilor & Vice Mayor in the Same Party First 0.0871 0.0817 0.0750 0.0610 (0.0092) (0.0096) (0.0126) (0.0169) Same Party 0.0313 0.0347 0.0258 0.0351 (0.0074) (0.0083) (0.0104) (0.0145) First  Same Party 0.0129 0.0166 0.0204 0.0119 (0.0122) (0.0133) (0.0167) (0.0227) Second Place Mean 0.122 0.122 0.122 0.122 Bandwidth 1 1 CCT = 0.014 IK = 0.006 N 29,290 29,290 23,254 14,970 Councilors (First and Second Placers) in the Same Party First 0.0688 0.0693 0.0592 0.0587 (0.0090) (0.0102) (0.0125) (0.0167) Same Party 0.0290 0.0244 0.0233 0.0146 (0.0075) (0.0082) (0.0106) (0.0145) First  Same Party 0.0200 0.0135 0.0094 0.0077 (0.0121) (0.0134) (0.0165) (0.0224) Second Place Mean 0.122 0.122 0.122 0.122 Bandwidth 1 1 CCT = 0.014 IK = 0.006 N 29,294 29,294 23,258 14,972 Note: Robust standard errors clustered at the constituency level are in parentheses. Dependent variable is the probability of running for vice governor (vice mayor). The party variable is an indicator for whether the provincial (municipal) councilor is in the same party as the incumbent vice governor (vice mayor). The unit of observation is a candidate. Columns 1–2 use the full sample with quadratic and cubic control functions. Column 3 uses the Calonico et al. (2014) selection method while column 4 uses the Imbens and Kalyanaraman (2012) selection method. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. is worth noting that other alternative models of voter coordination parties making selections based on rank—does not seem to apply have not been explored. in this case. Finally, suggestive evidence shows that other potential We also test two more potential mechanisms: greater media mechanisms such as voter coordination, media exposure, access to exposure for first placers and first placers’ better on-the-job perfor- resources, political appointments and performance do not explain mance. We show that both mechanisms do not seem to be driving the result. our main results. To save space we move our discussion of these This paper’s main contribution is to establish the empirical rela- mechanisms to Appendix G. tionship between rank and political promotion in a multi-member plurality electoral system. The main result, the first place effect, is 9. Conclusion robust across different levels of government, highly significant, and is substantive. The paper also opens up avenues to consider novel This paper has provided evidence for the link between rank and non-party based mechanisms to explain the first place effect. political promotion in multi-member plurality legislative elections This paper will hopefully lead to further work that will tease out in the Philippines. We document the first place effect: barely rank- the relevant mechanisms behind rank effects in multi-member ing first in legislative councils leads to a significant increase in the plurality elections. Informal norms within legislative councils or probability of running for higher office. The first place effect is politician beliefs are among the many plausible mechanisms that unique; the rank effect for second versus third as well as subse- may explain this result. Overall, given the richness of political sys- quent comparisons gives us much smaller or insignificant magni- tems the world over, this paper is a step forward in our under- tudes. The standard mechanism in the extant literature—political standing of how rank affects politics. 8 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Funding: This work was supported by the Institute for Humane follows naturally from definition given the availability of data. For Studies and the Mack Institute for Innovation Management at the example, Folke et al. (2016) and Eggers and Hainmueller (2009) Wharton School, University of Pennsylvania. use birth registers and historical biographies to identify family connections. However, in the absence of high quality data, we Declaration of Competing Interest exploit naming conventions to recognize the same politician through time (as is common in the literature in developing The authors declare that they have no known competing finan- countries). cial interests or personal relationships that could have appeared In the Philippines, the naming tradition is as follows: to influence the work reported in this paper. given_name middle_name last_name suffix_name Jose Garcia dela Cruz Jr. Acknowledgments The given name may consist of one or more names. Although the middle name is used as in the United States, the main differ- We are deeply indebted to Santosh Anagol, Allen Hicken, Timur ence is that it must be the mother’s maiden name. The middle Kuran, Edmund Malesky, Livia Schubiger, Shing-Yi Wang, Erik Wib- name is not considered a surname, and the mother’s maiden name bels and Maisy Wong for their guidance and advising on this pro- comes before the father’s surname. In the Philippines, only the last ject. We would also like to thank Anna Cororaton, Nicolas Fliess, name, which comes from the father’s last name, is considered an Marco Morucci, Ngoc Phan and Mateo Villamizar for insightful individual’s surname. When males get married, no change occurs comments, and Teresita Cadiben from the Commission of Elections in their names. For females, however, the usual practice is to take for sharing institutional knowledge on election-related issues. All on the spouse’s last name. In addition to the usual components, errors are our own. males may also have a suffix name if their given and last names match their father’s. Suffixes can be Jr. (or Sr. for fathers), II, III, Appendix A. Name matching and so on. How do we identify politicians across time and their family A.1. Matching politicians across time members given our data? For some existing studies, identification In the elections data, names of candidates are provided, Table A1 although some components may be incomplete or missing. The Distribution of Top 10 Last Names. most common source of name discrepancies originates from the Last Name Frequency Percent shortening of the middle name to just the middle initial (in some years, the same candidate maybe listed as having a middle name Dela Cruz 7,472 0.50 Garcia 5,388 0.36 ‘G’ or ‘Garcia’). Another potential source is providing a suffix name Reyes 5,276 0.35 in some years but not others. Ramos 5,076 0.34 In this section, we test whether our results are robust to using Mendoza 4,642 0.31 different methods for name matching. Our baseline approach uses Bautista 3,822 0.25 the first and last names to match individuals across time. First, we Flores 3,761 0.25 Gonzales 3,407 0.23 show that including the suffix name, middle initial or the middle Santos 3,406 0.23 name does not change the results. In fact, all estimates remain sta- Fernandez 3,362 0.22 tistically significant and robust. Tables A2, A3 and A4 present the Total 45,612 3.04 provincial, municipal and village results using the different meth- ods. Second, we drop observations whose last name belongs to the Table A2 First Place Effect (Province), by Name Matching Method. Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) First Name + Last Name + Suffix Name Candidacy, t þ k 0.054 0.0826 0.0838 0.047 0.0904 0.017 0.0497 [N = 3,274] (0.0147) (0.0176) [N = 2,638] (0.0176) [N = 1,294] (0.0351) Election, t þ k 0.021 0.0346 0.0413 0.038 0.0403 0.012 0.0253 [N = 3,274] (0.0115) (0.0133) [N = 2,452] (0.0149) [N = 1,248] (0.0231) First Name + Last Name + Middle Initial Candidacy, t þ k 0.047 0.0901 0.0921 0.048 0.101 0.014 0.0637 [N = 3,274] (0.0139) (0.0167) [N = 2,654] (0.0164) [N = 1,404] (0.0316) Election, t þ k 0.019 0.0341 0.0402 0.039 0.0377 0.012 0.0164 [N = 3,274] (0.0108) (0.0124) [N = 2,472] (0.0137) [N = 1,280] (0.0213) First Name + Last Name + Middle Name Candidacy, t þ k 0.036 0.0612 0.0616 0.042 0.0642 0.015 0.0353 [N = 3,274] (0.0126) (0.0149) [N = 2,532] (0.0154) [N = 1,442] (0.0275) Election, t þ k 0.015 0.0227 0.0269 0.043 0.0249 0.016 0.0178 [N = 3,274] (0.00918) (0.0106) [N = 2,560] (0.0116) [N = 1,510] (0.0151) Note: Robust standard errors clustered at the province level are in parentheses. Dependent variable is the probability of running for or winning as the vice governor. The unit of observation is a candidate. Columns 2–3 use the full sample with quadratic and cubic control functions. Column 5 uses the Calonico et al. (2014) selection method while column 7 uses the Imbens and Kalyanaraman (2012) selection method. Columns 4 and 6 provide the optimal value for each procedure. The number of observations for the full sample and bandwidth methods is reported in brackets. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. 9 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Table A3 First Place Effect (Municipality), by Name Matching Method. Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) First Name + Last Name + Suffix Name Candidacy, t þ k 0.12 0.0782 0.0724 0.013 0.0613 0.006 0.0522 [N = 29,294] (0.00613) (0.00649) [N = 23,034] (0.00810) [N = 14,716] (0.0110) Election, t þ k 0.056 0.0374 0.0282 0.012 0.0185 0.006 0.0124 [N = 29,294] (0.00459) (0.00511) [N = 21,602] (0.00647) [N = 14,100] (0.00870) First Name + Last Name + Middle Initial Candidacy, t þ k 0.118 0.0828 0.0770 0.012 0.0615 0.006 0.0509 [N = 29,294] (0.00599) (0.00646) [N = 22,092] (0.00832) [N = 13,998] (0.0113) Election, t þ k 0.055 0.0413 0.0319 0.011 0.0184 0.005 0.0131 [N = 29,294] (0.00453) (0.00518) [N = 21,088] (0.00667) [N = 13,380] (0.00901) First Name + Last Name + Middle Name Candidacy, t þ k 0.076 0.0504 0.0473 0.013 0.0369 0.006 0.0250 [N = 29,294] (0.00482) (0.00515) [N = 22,720] (0.00656) [N = 13,938] (0.00904) Election, t þ k 0.037 0.0250 0.0192 0.012 0.0107 0.005 0.00974 [N = 29,294] (0.00363) (0.00402) [N = 22,244] (0.00522) [N = 12,824] (0.00740) Note: Robust standard errors clustered at the municipal level are in parentheses. Dependent variable is the probability of running for or winning as the vice mayor. The unit of observation is a candidate. Columns 2–3 use the full sample with quadratic and cubic control functions. Column 5 uses the Calonico et al. (2014) selection method while column 7 uses the Imbens and Kalyanaraman (2012) selection method. Columns 4 and 6 provide the optimal value for each procedure. The number of observations for the full sample and bandwidth methods is reported in brackets. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. Table A4 First Place Effect (Village), by Name Matching Method. Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) First Name + Last Name + Suffix Name Candidacy, t þ k 0.089 0.0654 0.0618 0.024 0.0593 0.009 0.0568 [N = 160,192] (0.00226) (0.00242) [N = 132,850] (0.00302) [N = 75,424] (0.00460) Election, t þ k 0.035 0.0310 0.0280 0.025 0.0267 0.008 0.0227 [N = 160,192] (0.00165) (0.00177) [N = 133,738] (0.00213) [N = 73,450] (0.00331) First Name + Last Name + Middle Initial Candidacy, t þ k 0.090 0.0678 0.0639 0.025 0.0613 0.009 0.0596 [N = 160,192] (0.00227) (0.00244) [N = 133,480] (0.00302) [N = 77,298] (0.00455) Election, t þ k 0.035 0.0326 0.0293 0.025 0.0276 0.007 0.0231 [N = 160,192] (0.00166) (0.00178) [N = 133,446] (0.00215) [N = 66,680] (0.00356) First Name + Last Name + Middle Name Candidacy, t þ k 0.084 0.0646 0.0617 0.026 0.0590 0.010 0.0586 [N = 160,192] (0.00221) (0.00234) [N = 136,020] (0.00289) [N = 81,380] (0.00428) Election, t þ k 0.033 0.0304 0.0279 0.024 0.0248 0.009 0.0228 [N = 160,192] (0.00160) (0.00168) [N = 132,080] (0.00209) [N = 77,046] (0.00312) Note: Robust standard errors clustered at the village level are in parentheses. Dependent variable is the probability of running for or winning as the village chair. The unit of observation is a candidate. Columns 2–3 use the full sample with quadratic and cubic control functions. Column 5 uses the Calonico et al. (2014) selection method while column 7 uses the Imbens and Kalyanaraman (2012) selection method. Columns 4 and 6 provide the optimal value for each procedure. The number of observations for the full sample and bandwidth methods is reported in brackets. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. Table A5 First Place Effect (Excluding Common Last Names). Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) Province Candidacy, t þ k 0.57 0.0875 0.0897 0.055 0.0950 0.013 0.0511 [N = 3,159] (0.0157) (0.0187) [N = 2,670] (0.0177) [N = 1,300] (0.0348) Election, t þ k 0.021 0.0368 0.0438 0.040 0.0418 0.014 0.0329 [N = 3,159] (0.0119) (0.0138) [N = 2,416] (0.0149) [N = 1,398] (0.0197) Municipality Candidacy, t þ k 0.123 0.0813 0.0748 0.013 0.0656 0.007 0.0626 10 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Table A5 (continued) Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) [N = 28,515] (0.00625) (0.00695) [N = 22,450] (0.00828) [N = 15,784] (0.0107) Election, t þ k 0.056 0.0389 0.0266 0.012 0.0183 0.006 0.0138 [N = 28,515] (0.00470) (0.00515) [N = 21,382] (0.00663) [N = 14,040] (0.00887) Village Candidacy, t þ k 0.091 0.0670 0.0629 0.026 0.0609 0.007 0.0548 [N = 155,365] (0.00232) (0.00248) [N = 131,848] (0.00303) [N = 62,920] (0.00522) Election, t þ k 0.035 0.0321 0.0287 0.027 0.0280 0.007 0.0229 [N = 155,365] (0.00170) (0.00181) [N = 132,331] (0.00214) [N = 64,386] (0.00363) Note: Robust standard errors clustered at the constituency level are in parentheses. Dependent variable is the probability of running for or winning higher office: vice governor for province, vice mayor for municipality and village chair for village regressions. The unit of observation is a candidate. Columns 2–3 use the full sample with quadratic and cubic control functions. Column 5 uses the Calonico et al. (2014) selection method while column 7 uses the Imbens and Kalyanaraman (2012) selection method. Columns 4 and 6 provide the optimal value for each procedure. The number of observations for the full sample and bandwidth methods is reported in brackets. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. top 10 most common last names in the Philippines (see Table A1). dataset includes thousands of ordinances from 1988–2019, and This addresses the issue of mismatches due to common names. contains the ordinance title, date of enactment and the names of Comparing Tables 2 and A5, we find that results remain the sponsors or authors. unchanged. B.4. Standing committees Appendix B. Auxiliary datasets To control for traditions and ‘unwritten rules’ within the council, B.1. Voter information we again collected a novel dataset of city council positions for each councilor. It is common practice in newly elected councils to deter- Voter turnout data can be used to understand the mechanism of mine the standing committees for the following three-year term. voter coordination. We calculate voter turnout in each province The dataset includes standing committee positions for 25 cities, and city from 2004 to 2013 by dividing the number of registered from 1988 to 2019. The dataset lists all committees and the respec- to actual voters. Unfortunately, the data is not available for the vil- tive position of each councilor (i.e. chairperson, vice chairperson or lage level. member). For some cities which have special positions such as the majority floor leader, minority floor leader, etc., the dataset also notes which councilors are appointed or voted to those positions. B.2. Campaign contributions and expenditures To control for campaign spending, we obtained the statements B.5. Online news mentions of all provincial board members in 2010 and 2013 who ran again in 2013 and 2016, respectively. The dataset includes the percent- An emerging source for news in the Philippines are websites or age of campaign expenditurers originating from personal funds, online platforms of newspapers and broadsheets. To measure party contributions and contributions from private companies or media exposure, we scanned the websites of top national and individuals. regional newspapers and created an algorithm to count the num- ber of times a particular candidate’s name was mentioned. We did this for 29 news sites, scanning all news articles from 2008 B.3. Ordinances to 2019. As objective measures for performance and productivity in the council, we collected a novel dataset of city ordinances from 31 Appendix C. Balance cities in the Philippines. Conditional on choosing the city, we obtained all city ordinances for years which data is available. The Fig. C1. Number of Observations Balance Note: Horizontal axis shows the vote share bins, which are within 0.005 or 0.001 intervals of the vote margin. The blue bars correspond to the number of observations in each bin. Because of the empirical design, the distribution of observations is symmetric around the cutoff. 11 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Fig. C2. Covariate Balance Note: Horizontal axis shows the vote margin of the first placer. Dependent variables are the probability of running for or winning higher office, vote shares, party affiliation in the previous election, and past candidacies. Each circle or triangle corresponds to the unconditional mean within 0.005 or 0.001 intervals of the vote margin. Solid curves are smoothed local polynomials on either side of the discontinuity. 12 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Table C1 Placebo Test and Covariate Balance. Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Value Linear Value Linear (1) (2) (3) (4) (5) Province Candidacy, t  k 0.014 0.030 0.00556 0.023 0.0121 [N = 2,344] (0.00825) [N = 2,032] (0.00916) Election, t  k 0.002 0.042 0.00252 0.017 0.000270 [N = 2,694] (0.00321) [N = 1,702] (0.00403) Vote Share, t  k 0.219 0.042 0.0125 0.030 0.00597 [N = 1,928] (0.0162) [N = 1,663] (0.0185) Party Affiliation, t  k 0.525 0.043 0.0215 0.016 0.0725 [N = 860] (0.0565) [N = 494] (0.0864) Past Candidacies 2.396 0.042 0.148 0.031 0.123 [N = 2,708] (0.114) [N = 2,404] (0.130) Municipality Candidacy, t  k 0.035 0.011 0.00145 0.006 0.00479 [N = 22,102] (0.00447) [N = 16,060] (0.00554) Election, t  k 0.007 0.014 0.000155 0.004 0.00228 [N = 24,512] (0.00180) [N = 11,950] (0.00304) Vote Share, t  k 0.132 0.013 0.00622 0.007 0.00321 [N = 14,978] (0.00526) [N = 10,592] (0.00669) Party Affiliation, t  k 0.546 0.011 0.0125 0.004 0.0442 [N = 4,735] (0.0242) [N = 2,464] (0.0369) Past Candidacies 1.41 0.014 0.0499 0.006 0.0421 [N = 24,508] (0.0314) [N = 15,212] (0.0444) Village Candidacy, t  k 0.045 0.019 0.00120 0.007 0.000834 [N = 72,016] (0.00263) [N = 43,240] (0.00385) Election, t  k 0.009 0.015 0.00100 0.004 0.00175 [N = 65,406] (0.00138) [N = 26,924] (0.00258) Vote Share, t  k 0.101 0.013 0.00172 0.004 0.00350 [N = 39,961] (0.00190) [N = 17,791] (0.00319) Past Candidacies 1.073 0.019 0.0169 0.009 0.0206 [N = 71,084] (0.0107) [N = 49,988] (0.0142) Note: Robust standard errors clustered at the constituency level are in parentheses. Dependent variables are the probability of running for or winning higher office, vote shares and party affiliation in the previous election and number of past candidacies. The unit of observation is a candidate. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. 13 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Appendix D. Robustness to bandwidths Fig. D1. Robustness to Bandwidths Note: Dependent variable is the probability of running for or winning higher office. The horizontal axis shows the range of bandwidth values until h ¼ 0:4. The blue circle is the coefficient estimate, with the thick blue bar showing the 90% confidence interval and the thin blue bar showing the 95% confidence interval. 14 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Appendix E. Other rank effects (see Fig. E1). Fig. E1. Other Rank Effects Note: Horizontal axis shows the vote share difference between the nth and n þ 1th placers (i.e. a positive vote margin implies ranking nth and a negative vote margin implies ranking n þ 1th ). Each circle (triangle) corresponds to the unconditional mean of running (winning) within 0.005 or 0.001 intervals of the vote margin. Solid curves are smoothed local polynomials on either side of the discontinuity. 15 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Fig. 6 (continued) 16 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Appendix F. Rank effects for other positions Appendix G. Potential mechanisms Table F1 First Place Effect (for the Highest Position). Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) Province Candidacy, t þ k 0.013 0.00572 0.00499 0.036 0.000002 0.011 0.0195 [N = 3,274] (0.00513) (0.00558) [N = 2,380] (0.00696) [N = 1,198] (0.0104) Election, t þ k 0 0.00251 0.00231 0.027 0.00610 0.012 0.00196 [N = 3,274] (0.00318) (0.00368) [N = 2,098] (0.00384) [N = 1,246] (0.00142) Municipality Candidacy, t þ k 0.024 0.0182 0.0143 0.016 0.0146 0.006 0.0145 [N = 29,294] (0.00301) (0.00338) [N = 24,810] (0.00393) [N = 15,000] (0.00546) Election, t þ k 0.007 0.00510 0.00325 0.016 0.00253 0.01 0.00066 [N = 29,294] (0.00165) (0.00178) [N = 24,788] (0.00216) [N = 20,192] (0.00250) Note: Robust standard errors clustered at the constituency level are in parentheses. Dependent variable is the probability of running for or winning the highest office in the constituency: provincial governor for province, and municipal mayor for municipality regressions. The unit of observation is a candidate. The number of observations for the full sample and bandwidth methods is reported in brackets. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. Table F2 First Place Effect (for the Same Position). Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) Province Candidacy, t þ k 0.53 0.0543 0.0691 0.035 0.0372 0.017 0.0674 [N = 3,274] (0.0257) (0.0308) [N = 2,352] (0.0329) [N = 1,588] (0.0455) Election, t þ k 0.425 0.0297 0.0519 0.039 0.0298 0.021 0.0747 [N = 3,274] (0.0251) (0.0291) [N = 2,474] (0.0318) [N = 1,792] (0.0408) Municipality Candidacy, t þ k 0.605 0.0687 0.0679 0.015 0.0556 0.005 0.0535 [N = 29,294] (0.00777) (0.00850) [N = 23,838] (0.0106) [N = 12,366] (0.0161) Election, t þ k 0.505 0.0448 0.0505 0.019 0.0494 0.007 0.0401 [N = 29,294] (0.00768) (0.00827) [N = 25,896] (0.00984) [N = 16,246] (0.0138) Village Candidacy, t þ k 0.625 0.0552 0.0507 0.026 0.0452 0.011 0.0501 [N = 160,192] (0.00326) (0.00354) [N = 135,228] (0.00434) [N = 91,976] (0.00592) Election, t þ k 0.487 0.0261 0.0290 0.023 0.0339 0.007 0.0451 [N = 160,192] (0.00332) (0.00361) [N = 131,142] (0.00458) [N = 67,226] (0.00752) Note: Robust standard errors clustered at the constituency level are in parentheses. Dependent variable is the probability of running for or winning the same office again: provincial councilor for province, municipal councilor for municipality and village councilor for village regressions. The unit of observation is a candidate. The number of observations for the full sample and bandwidth methods is reported in brackets. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. Table G1 Voter Coordination. Full Sample Optimal Bandwidth Quadratic Cubic Linear Linear (1) (2) (3) (4) Province First 0.0418 0.0349 0.0395 0.0456 (0.0285) (0.0311) (0.0317) (0.0528) Voter Turnout 0.0123 0.00438 0.00201 0.0233 (0.0219) (0.0294) (0.0298) (0.0538) First  Voter Turnout 0.0309 0.0443 0.0314 0.0883 (0.0408 (0.0464) (0.0454) (0.0833) Second Place Mean .042 .042 .042 .042 Bandwidth 1 1 CCT=.046 IK=.013 (continued on next page) 17 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Table G1 (continued) Full Sample Optimal Bandwidth Quadratic Cubic Linear Linear (1) (2) (3) (4) N 1,604 1,604 1,284 662 Municipality First 0.0546 0.0361 0.0338 0.0251 (0.0146) (0.0180) (0.0173) (0.0233) Voter Turnout 0.0190 0.0315 0.0277 0.0322 (0.0131) (0.0156) (0.0154) (0.0204) First  Voter Turnout 0.00101 0.0180 0.0255 0.0120 (0.0215) (0.0245) (0.0266) (0.0315) Second Place Mean 0.121 0.121 0.121 0.121 Bandwidth 1 1 CCT = 0.014 IK = 0.006 N 12,830 12,830 10,742 6,984 Note: Robust standard errors clustered at the constituency level are in parentheses. Dependent variable is the probability of running for vice governor (vice mayor). The voter turnout variable is an indicator for above median voter turnout defined as the share of actual to registered voters. The unit of observation is a candidate. Columns 1–2 use the full sample with quadratic and cubic control functions. Column 3 uses the Calonico et al. (2014) selection method while column 4 uses the Imbens and Kalyanaraman (2012) selection method. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. Table G2 Media Exposure. Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) Province First 1.74 0.507 0.303 0.029 0.530 0.024 1.216 [N = 1,246] (0.881) (1.019) [N = 796] (0.947) [N = 720] (0.833) Municipality First 1.04 0.0419 0.151 0.006 0.378 0.005 0.241 [N = 8,536] (0.232) (0.218) [N = 4,646] (0.238) [N = 4,042] (0.314) Village First 0.162 0.0081 0.0059 0.018 0.0017 0.006 0.0110 [N = 159,941] (0.0135) (0.0146) [N = 138,199] (0.0189) [N = 75,221] (0.0312) Note: Robust standard errors clustered at the constituency level are in parentheses. Dependent variable is the number of news articles where the candidate was mentioned. The unit of observation is a candidate. The number of observations for the full sample and bandwidth methods is reported in brackets. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. Table G3 Access to Resources. Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) Dependent Variable: Campaign Contributions by Party First 11.46 25.37 4.470 0.032 15.26 0.031 15.26 [N = 373] (17.37) (12.48) [N = 241] (10.30) [N = 241] (10.30) Dependent Variable: Total Campaign Contributions First 22.07 27.96 16.34 0.031 18.95 0.053 8.475 [N = 373] (16.77) (16.15) [N = 237] (12.08) [N = 300] (12.79) Dependent Variable: Personal Campaign Expenditure First 155.97 5.233 5.477 0.032 47.80 0.036 22.24 [N = 373] (22.98) (25.01) [N = 243] (27.35) [N = 251] (43.15) Note: Robust standard errors clustered at the province level are in parentheses. Dependent variable is the amount received by candidates from their party, from all sources, and the amount spent originating from personal sources. Amounts are expressed in thousands of Philippines pesos. The unit of observation is a candidate. The number of observations for the full sample and bandwidth methods is reported in brackets. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. A mechanism closely related to both party alignment and voter nal news outlets in the Philippines from 2008 to 2019 and use the coordination is media exposure—first placers might enjoy more number of times a councilor’s name appears in the articles as our extensive media mileage, rallying parties or voters to converge main outcome variable. Table G2 shows that first and second plac- towards the first placer because of this boost in popularity. We test ers do not differ in the number of times they are mentioned in local this by collating all online articles from 29 top national and regio- media during their terms. In addition, the low number of name 18 D. Dulay and L. Go Journal of Public Economics 200 (2021) 104455 Table G4 Council Positions & Performance. Full Sample Bandwidth: CCT Bandwidth: IK Second Place Optimal Optimal Mean Quadratic Cubic Value Linear Value Linear (1) (2) (3) (4) (5) (6) (7) Dependent Variable: Number of Committee Chairmanships First 2.25 0.0271 0.0757 0.010 0.0289 0.007 0.240 [N = 121] (0.163) (0.184) [N = 83] (0.194) [N = 73] (0.229) Dependent Variable: Number of Ordinances Sponsored First 7.31 0.353 1.728 0.009 1.740 0.005 1.475 [N = 308] (1.185) (1.364) [N = 202] (1.259) [N = 156] (1.422) Note: Robust standard errors clustered at the municipality level are in parentheses. Dependent variable is the number of committee chairmanships held or the number of ordinances sponsored. The unit of observation is a candidate. The number of observations for the full sample and bandwidth methods is reported in brackets. Second place mean is the estimated value of the dependent variable for the second placer using the Imbens and Kalyanaraman (2012) method. mentions regardless of rank shows that councilors do not seem to Arias, E., Balan, P., Larreguy, H., Marshall, J., Querubin, P., 2019. Information Provision, Voter Coordination, and Electoral Accountability: Evidence from be the focus of media attention. Mexican Social Networks. American Political Science Review 113 (2), 475–498. We also explore a politician-level mechanism which measures Blais, A., Nadeau, R., Gidengil, E., Nevitte, N., 2001. Measuring Strategic Voting in the councilor’s performance in office. Councilors who end up Multiparty Plurality Elections. Electoral Studies 20 (3), 343–352. achieving a higher rank may update their beliefs and desires for Calonico, S., Cattaneo, M., Titiunik, R., 2014. Robust Nonparametric Confidence Intervals for Regression-Discontinuity Designs. Econometrica 82 (6), 2295– political promotion. As a result, they perform better and exert 2326. more effort towards running for higher office. To test this, we col- Carey, J., Shugart, M., 1995. Incentives to Cultivate A Personal Vote: A Rank Ordering lated ordinances from various municipalities and linked them to of Electoral Formulas. Electoral Studies 14 (4), 417–439. Cirone, A., Cox, G., Fiva, J., 2021. Seniority-Based Nominations and Political Careers. each councilor. In Table G4, we find that first placers do not spon- American Political Science Review 115 (1), 234–251. sor more ordinances than second placers, thus providing sugges- Cox, G., Shugart, M., 1996. Strategic Voting Under Proportional Representation. The tive evidence that effort and performance are not a mechanism Journal of Law, Economics, and Organization 12 (2), 299–324. Crisp, B., Ollivella, S., Malecki, M., Sher, M., 2013. Vote-Earning Strategies in Flexible driving the first place effect.18 List Systems: Seats at the Price of Unity. Electoral Studies 32 (4), 658–669. While we have argued that councilors regardless of rank have Cruz, C., Labonne, J., Querubin, P., 2017. Politician Family Networks and Electoral constitutionally similar roles, differences may arise between what Outcomes: Evidence from the Philippines. American Economic Review 107 (10), 3006–3037. is legally recognized (de jure) and what is practiced in reality (de Cruz, C., Labonne, J., Querubin, P., 2020. Social Network Structures and the Politics of facto). First, a potential hypothesis is that first placers have Public Goods Provision: Evidence from the Philippines. American Political amassed more wealth or have better access to resources (regard- Science Review 114 (2), 486–501. Eggers, A.C., Hainmueller, J., 2009. MPs for Sale? Returns to Office in Postwar British less of source) due to their position. While this variable is hard Politics. American Political Science Review 103 (4), 1–21. to capture, we use data from campaign filings on personal or other Folke, O., Persson, T., Rickne, J., 2016. The Primary Effect: Preference Votes and sources of campaign funds to measure wealth and access to Political Promotions. American Political Science Review 110 (3), 559–578. resources. In Table G3, we find that total contributions and per- Fujiwara, T., Sanz, C., 2020. Rank Effects in Bargaining: Evidence from Government Formation. The Review of Economic Studies 87 (3), 1261–1295. sonal campaign expenditures in the subsequent election for first Gerber, A., 1998. Estimating the Effect of Campaign Spending on Senate Election and second placers are not significantly different. Second, an Outcomes Using Instrumental Variables. American Political Science Review 92 important concern is that informal norms surrounding political (2), 401–411. Hicken, A., 2009. Building Party Systems in Developing Democracies. Cambridge appointments may dictate that better positions be reserved for University Press. the higher ranked candidate. To test this, we collect data on com- Imbens, G., Kalyanaraman, K., 2012. Optimal Bandwidth Choice for the Regression mittee chairmanships in different municipal councils and show in Discontinuity Estimator. Rev. Econ. Stud. 79 (3), 933–959. Lee, D., 2008. Randomized Experiments from Non-random Selection in U.S. House Table G4 that first placers are not more likely to become commit- Elections. Journal of Econometrics 142, 675–697. tee chairs. Merillainen, J., Tukiainen, J., 2018. Rank Effects in Political Promotions. Public Choice 177, 87–109. Mondak, J., 1995. Competence, Integrity, and the Electoral Success of Congressional Incumbents. Journal of Politics 57 (4), 1043–1069. Myerson, R.B., 1993. Incentives to Cultivate Favored Minorities Under Alternative References Electoral Systems. American Political Science Review, 856–869. Querubin, P., 2016. Family and Politics: Dynastic Persistence in the Philippines. Adida, C., Gottlieb, J., Kramon, E., McClendon, G., 2020. When Does Information Quarterly Journal of Political Science 11 (2), 151–181. Influence Voters? The Joint Importance of Salience and Coordination. Rolfe, M., 2012. Voter Turnout: A Social Theory of Political Participation. Cambridge Comparative Political Studies 53 (6), 851–891. University Press. Anagol, S., Fujiwara, T., 2016. The Runner-Up Effect. Journal of Political Economy Sidel, J., 1999. Capital, Coercion, and Crime: Bossism in the Philippines. Stanford 124 (4), 927–991. University Press, Stanford, CA. Ansolabehere, S., Snyder Jr., J., Stewart III, C., 2001. Candidate Positioning in U.S. Pons, V. and C. Tricaud (2020). The large effects of a small win: How past rankings House Elections. American Journal of Political Science, 136–159. shape the behavior of voters and candidates. 18 To shed more light on this and the following mechanisms, we hand-collected these novel datasets. Given the difficulty in data availability, the tests conducted using these datasets have smaller samples and the resulting estimates suffer from larger standard errors. 19