1 Introduction

An expansive literature on fiscal federalism concerns itself with understanding how the political alignment between different levels of government within a country may affect the decisions taken by politicians and parties, as well as by voters at the ballot box (e.g., Baskaran & Hessami, 2017; Brollo & Nannicini, 2012). As part of this literature, numerous studies bring forth the idea of “coattails”—whereby a candidate’s electoral performance in one type of race can directly influence that of co-partisans in elections taking place at another governmental level (Golder, 2006). In this scholarship, much work has been done on so-called “top-down coattails”, addressing the question of whether the success of someone running for a high-level office can spill down to lower-level races (Meredith, 2013). Substantially less understood, however, is the complementary “bottom-up” process, wherein the performance of party members in elections for lower levels of government can trickle upwards, directly benefiting or hindering up-ticket co-partisans.

More specifically, the key question in the study of reverse coattails (a terminology coined by (Ames, 1994)) is whether the political control of subnational governments can directly benefit parties in national elections.Footnote 1 Following Ames’s work, scholars have corroborated the existence of reverse coattails in various countries and time-periods.Footnote 2 That said, empirical support for reverse coattails is not universal: Broockman (2009) does not find evidence in the US, a result echoed by Migueis (2013) in Portugal, and Feierherd (2019) in Brazil, with the latter study showing that mayors may actually hurt up-ticket co-partisans, particularly when voters are dissatisfied with their socio-economic conditions.

Given these conflicting findings, recent studies have turned towards disentangling the channels by which aligned mayors can help their parties (Avelino, Biderman, and Barone, 2017), and it is here that we make our contributions, which we summarize as follows: first, we find evidence for reverse coattails in a unique political setting—under a technocratic central government unexpectedly instituted in Romania after an exogenous tragic event. These findings allow us to assert that mayors are electorally valuable even in periods when party-affiliated central governments do not support them via politically discriminatory measures. Second, exploring the mechanism underlying our findings, we find evidence for prospective voting, with voters supporting their mayor’s party in national elections partially in anticipation of preferential treatment from the future political central government.

More specifically, as part of the effort to understand the channels underlying reverse coattails, a leading explanation—drawing from a literature on “pork-barrel” politics (Arulampalam, Dasgupta, Dhillon, and Dutta, 2009)—is that aligned mayors are able to stimulate the vote by using resources preferentially allocated to them by their co-partisans controlling the central government. As summarized by Feierherd (2019), “workhorse models of re-distributive politics typically assume that re-election seeking [central politicians] use fiscal transfers to reward co-partisans in the hope that stronger local incumbents will transfer votes” (p. 195).

Therefore, a core mechanism underlying reverse coattails is proposed to be a retrospective one: targeting resources to aligned constituencies increases the discretion of one’s co-partisans and “ties the hands” of rivals (Brollo & Nannicini, 2012), thus creating electoral rents in upcoming elections—an insight echoed by Ventura (2021), who argues that mayors use local resources to generate support, and finance these efforts via preferential access to central capital controlled by co-partisans. Henceforth, we label this retrospective mechanism “realized favoritism”.

The present paper refines the existing literature on reverse coattails in three ways. First, in light of the expansive scholarship highlighting local-central political alignment and the associated realized favoritism channel as a key driver of reverse coattails, we begin by asking the complementary question: is sub-national incumbency electorally rewarding in national races absent a contemporaneous party alignment linking local and central governments, and the associated retrospective realized favoritism channel?

We argue that quantifying the reverse coattails effect absent realized favoritism is important from a theoretical perspective, as such an analysis provides insights into the versatility of local incumbency as an electoral resource. To this point, strong positive effects would indicate that local incumbency can influence national power even in periods when a party does not control the central government and its associated administrative capacities—for instance, under a technocracy,Footnote 3 a form of government that has become increasingly widespread in Europe (Pastorella, 2016) as an alternative form of representation to party governance (Caramani, 2017).

More broadly, retrieving evidence for reverse coattails absent realized favoritism would bring into question the reverse coattails relationship being a straightforward retrospective exchange one, thus refining the conclusions of the aforementioned workhorse models of distributive politics (Brollo & Nannicini, 2012; Feierherd, 2019).

That said, partialling out the role of realized favoritism is a difficult statistical task—evidence for central favoritism is widespread,Footnote 4 and generally, when establishing an empirical association between local incumbency and national electoral performance, one of the counterfactual political affiliations considered is that of the party controlling the central government. If so, the estimated reverse coattails effect, even if causal, will be informative only of the bundled impact of alignment-enabled realized favoritism, alongside any other potential mechanisms driving reverse coattails. Such statistical difficulties may explain why breaking down the channels by which local incumbents stimulate the vote remains a challenge (Avelino, Biderman, and Barone, 2017).

We address these difficulties by studying reverse coattails in a unique political setting: under a technocratic central government, unexpectedly instituted after a national tragedy in Romania.

In the first part of our analysis, we document strong reverse coattails—of roughly eleven percentage points in our preferred specification—using a close-elections regression discontinuity [RD] design for causal inference. Additionally, we use the same methodology to rule out pre-elections preferential fiscal allocations by the central government, thus providing evidence against realized favoritism as a driver. Our results suggest that mayoral partisanship did not have a significant effect on the disbursement of central resources under the technocratic government.

These two results constitutes our two main contributions: first and most straightforwardly, the strong positive reverse coattails effect documented here adds another empirical piece of evidence in favor of the scholarship highlighting local incumbency as a valuable electoral resource. Additionally, our findings suggest that a contemporaneous party alignment between local and central governments is not necessary for mayoral partisanship to provide an electoral advantage in national elections, adding to the wider aforementioned literature exploring realized favoritism as a key driver of reverse coattails.

Of course, with these results in mind, an immediate question arises: if contemporaneous local-central alignment and realized favoritism do not explain the documented reverse coattails effect, then what does? This is the question that the final part of our analysis concerns itself with, and it is here that we make our third contribution to the literature.

In particular, to investigate the underlying mechanism driving the reverse coattails effect, we draw from the scholarship on prospective voting, wherein voters are argued to be forward-looking, using information on the likelihood of future events to form expectations and make decisions. That is, the prospective view paints elections as serving a mandate selection goal: bygones are bygones, and voters choose candidates based on their expectations of future behavior (Konrad & Sherif, 2019; Erikson, MacKuen, and Stimson, 2000),Footnote 5

We take our analysis one step further, and investigate whether a prospective channel may help explain the documented reverse coattails effect. In line with work on the signaling value of distributive politics (Bracco, Lockwood, Porcelli, and Redoano, 2015), our argument is that local voters are ceteris paribus more likely to support their mayor’s party not only as a result of preferentially-distributed resources prior to the national elections, but also in anticipation of future such preferential treatment, to take place after the national elections, conditional on their mayor’s party gaining power in the central government.Footnote 6

We argue that studying this mechanism, henceforth labeled “anticipated favoritism”, is worthwhile, both for better understanding the political effects of mayoral incumbency, but also as a novel empirical assessment of prospective voting, adding to the scholarship attempting this statistical task (Woon 2012; Feltovich and Giovannoni 2015).Footnote 7

In support of anticipated favoritism driving the documented reverse coattails effect, we provide two pieces of evidence. First and most straightforwardly, we show that central resources were distributed in a preferential manner to co-partisans after the national elections investigated here, when the technocrats stopped ruling and a political government ascended to power. Our results show that mayors politically-aligned with the new government received just under eleven percent additional funding from the center.

These findings suggest that a mayor’s party gaining central power significantly affects a constituency’s finances, something which forward-looking voters may account for at the ballot box. That said, solely based on this result, it is still difficult to ascertain whether the reverse coattails we document are indeed driven by voters anticipating future favoritism, rather than, say, a simple mobilization effort conducted by the mayor.

While directly corroborating the former mechanism is challenging, since surveys breaking down voters’ motivations were not systematically conducted, we provide an additional piece of evidence suggesting that anticipated favoritism plays a role in the last part of our analysis. Concretely, we argue that if voters are indeed forward-looking, then the documented reverse coattails should be lager in magnitude in constituencies where central resources are a meaningful component of local funding, because the electorate in these center-reliant constituencies generates greater expected utility by voting for those more likely to attract “pork” from the center, relative to voters in more self-reliant constituencies.

In a heterogeneity analysis, we provide evidence supporting this interpretation: we find that the documented reverse coattails are twice as large in center-reliant constituencies, operationalized as those constituencies where central funds as a percentage of total revenues are above the median. Our result show that mayoral incumbency led to a 14.7 percentage points increase in the score obtained by the mayor’s party in center-reliant constituencies, relative to 7.92 percentage points in the rest of the sample.

Summarizing our third contribution: we document strong reverse coattails, and provide novel suggestive evidence that anticipated favoritism plays a role in explaining the result, thus bringing empirical support for the prospective, mandate selection representation model of voting.

2 Main hypothesis, setting and data

Our primary objective is to study whether the reverse coattails effect may materialize absent a retrospective channel hinging on realized favoritism by the central government.

We first address the most pressing issue: in “normal” times—that is, when a country’s central government is controlled by political parties—it is difficult to assess whether voters are more likely to support their mayor’s party in national elections as a result of pre-elections policies favoring co-partisans, as opposed to other channels such as anticipated favoritism, which we investigate in the final part of our analysis. This difficulty of distinguishing the realized and anticipated favoritism channels mirrors the broader challenge of separating the prospective and retrospective views (Feltovich and Giovannoni 2015).

Therefore, an intuitive first step is to assess whether mayors are electorally valuable when no partisan link exists between local constituencies and the central government, thus muting realized favoritism. This constitutes our first hypothesis.

Hypothesis 1: A contemporaneous party alignment between local and central governments, and the associated realized favoritism channel, are not necessary for local incumbency to provide an electoral advantage in national elections.

Two remarks are useful. First, our goal is not to establish that central-local alignment is generally unnecessary for reverse coattails. Indeed, via the anticipated favoritism channel we investigate below, voters may support their mayor’s party precisely in anticipation of preferential disbursements driven by future (i.e., following the national elections) local-central alignment. Hence, we solely focus on ruling out the necessity of a contemporaneous (i.e., in-between the local and national elections) partisan link.

Second, we do not argue that contemporaneous alignment is insufficient. Indeed, our objective is not to rule out pre-elections co-partisan preferential treatment as an important driver of reverse coattails. Generally, retrospective voting may very well play an important role (Ventura 2021). Rather, our goal is to study the necessity of realize favoritism and subsequently provide a complementary narrative whereby forward-looking considerations are also a driver: then, studying reverse coattails in a setting where the realized favoritism channel is muted constitutes an important step. This analysis does not suggest that retrospective voting is irrelevant in other contexts.

2.1 Setting: an unexpected technocracy in Romania

We exploit a unique political setting to quantify reverse coattails absent realized favoritism. We illustrate the relevant timeline in Fig. 1.Footnote 8

Fig. 1
figure 1

Political Timeline and Central Government Incumbency. Note: The Colectiv Nightclub Fire led to the unexpected resignation of the Social Democratic cabinet, and to the institution of a technocratic government, which ruled throughout 2016. After winning the December national elections, the Social Democrats returned to power

Following a fire at the Bucharest “Colectiv” Nightclub in October 2015, protests erupted linking the event to corruption. A demand of the protesters was the resignation of the then central government, controlled by Romania’s Social Democratic [SD] Party.Footnote 9 As a result, the government resigned early in November.

Subsequently, Dacian Cioloş, former European Commissioner, became prime-minister, with the new government instituted mid-November. We note that no cabinet members belonged to a party at the time, making this the first fully-technocratic government in Romania’s history. Importantly, the technocrats governed throughout 2016, a year in which both local and national elections took place in June and December, respectively.

We assess Hypothesis 1 in this context. Concretely, we ask whether mayoral incumbency resulting from the June elections benefited the SDs in the December parliamentary race.

Given the central government’s technocratic nature and the unanticipated timing of the succession, our argument is that mayors did not receive preferential co-partisan treatment from the centre in the period leading up to the national elections, an argument we corroborate empirically below. Therefore, any reverse coattails documented in this setting would materialize absent contemporaneous local-central party alignment and the realized favoritism channel, in line with Hypothesis 1.

2.2 Romania’s administrative and political organization

Before describing our data and empirical strategy, we briefly present here the institutional setting, focusing on the responsibilities and discretion of Romanian mayors, as well as the main parties present in Romanian politics at the time.

Romania is divided into 41 counties, and over 3,000 territorial-administrative areas, or constituencies—the main unit of observation used in our analysis. Constituencies, further classified as municipalities, cities or communes, depending on their population size and urbanization level,Footnote 10 are administered by a mayor and local council, elected via first-past-the-post ballots, typically held in June every four years.

The mayor—the local incumbent—represents a constituency’s executive authority. Fiscally, mayors are responsible for managing the local budget, both in terms of accruing revenues and allocating funds across projects (education, infrastructure and so forth).Footnote 11 Revenues are generated either internally, mainly via income, capital and property taxes, or externally, primarily via intergovernmental grants received from the center.Footnote 12 We note that the extent to which constituencies rely on central funds is on average substantially higher in rural areas, where self-funding capacities are limited.

In the vast majority of cases, local fiscal initiatives are proposed by the mayor. Once brought forward, these projects are voted on by the local council and, subject to approval, implemented once again by the mayor. Thus, mayors enjoy great levels of discretion—an appealing contextual feature for our analysis, which focuses on the partisanship of mayors to evaluate the reverse coattails effect.Footnote 13

In terms of Romania’s political organization, while the country is a multi-party democracy, with several parties holding mayoral offices, the political scene at the time was dominated by two major ones: the aforementioned center-left Social Democrats and the center-right National Liberals [NLs]. In Fig. 2, we show a map of Romania’s constituencies depicting the partisan affiliation of the mayors winning the June 2016 local elections. Overall, the SDs secured 53.6% of offices, while the NLs secured 33.9%.Footnote 14

Fig. 2
figure 2

Party Affiliation of Mayors by Constituency. Note: We map the party affiliation of mayors elected in June 2016. Data source: Romania’s Central Electoral Bureau

2.3 Data, operationalization, and sample restrictions

We combine information retrieved from Romania’s Central Electoral Bureau for the June local elections and the December parliamentary elections. We have data on the the party of all mayoral candidates, turnout, and the number of votes obtained by all parties in both elections.

We construct the running variable SD MARGIN\(_{i}\) for each constituency i as the difference between the vote share obtained by the SDs \({V}_{i}^{SD}\) and that of their closest competitor \({V_{i}^{O}}\):

$$\begin{aligned} \text {SD MARGIN}_{i} ={\text {V}_{\text {i}}}^{\hbox {SD}} - {\text {V}_{\text {i}}}^{\hbox {O}} \end{aligned}$$

Since we want to measure reverse coattails, we only consider races where the SDs placed first or second. Thus, in constituencies where the SDs won (lost) the ballot, SD MARGIN\(_{i}\) is positive (negative). Hence, we define our treatment variable \(\textit{INC}_\textit{i}\) equal to one if the SDs won constituency i’s mayoral race (zero otherwise).

Our electoral dependent variable is the vote share (percent of turnout) the SDs obtained in the December parliamentary elections.

To operationalize central favoritism, we use fiscal data. We combine information provided by the Ministry of Regional Development with demographic data from the National Statistics Institute. The resulting dataset tracks per capita public revenues for each constituency between 2015 and 2018, broken down by funding sources - locally generated or received from the center. 2015 data are used to assess the RD design’s validity. We use 2016 data to rule out realized favoritism by the technocrats. Finally, we use 2017–2018 information to investigate anticipated favoritism following the SDs’ return to power.

Throughout, we focus on total income and central income, defined as the difference between total and locally-generated revenues, both measured in log per capita terms,Footnote 15 as our dependent fiscal variables. Moreover, we code a variable TRANSFERS\(_{i}\), recording revenues disbursed by the centre with the purpose of balancing the local budgets and financing several local public goods investments.Footnote 16

The data-set contains 3180 constituencies. We restrict the sample in four ways. First, as discussed, we remove constituencies where the SDs did not secure the first or second position in June 2016. We impose three additional restrictions which, despite being theoretically appealing, are not essential: our results are robust to dropping them. First, we remove constituencies where candidates ran unopposed in the local elections to diminish the influence of these outliers. Second, we remove constituencies where the SDs went up against ALDE, the smaller party alongside which the SDs governed both before 2015 fire and after the 2016 national elections.

Finally, we remove constituencies classified as municipalities, and focus our analysis on rural communes and smaller towns. Essentially, the reason for doing so boils down to our focus on the partisan affiliation of the mayor as our key explanatory variable. As detailed above, whilst in smaller towns and rural communes, mayors have a significant amount of discretion and power, the same does not hold for municipalities whose policies are often instead dependent on decisions made by the corresponding county council (Romania is divided into 41 counties and a county council is elected every four years in each of them). The county council is responsible for county-level financial planning, which largely depends on the economic activity of municipalities specifically—by definition the most populous and economically important of Romanian localities. Hence, since our focus is on local-central relationships, we chose to exclude municipalities and focus on the units of observation where mayoral power and discretion are present.

2339 constituencies remain in the final sample. Appendix Table A1 presents summary statistics.

3 Analysis

We use a sharp RD (Imbens and Lemieux 2008) to causally estimate reverse coattails.Footnote 17 We restrict our sample of constituencies such that the running variable SD MARGIN\(_{i}\) \(\in \) [\(-h +h\)], where h is a positive number chosen using the methodology from Calonico et al. (2014) [CCT], and estimate:

$$\begin{aligned} \text {Y}_{i}= \alpha + \gamma \text {SD MARGIN}_{i} + \beta _0 \text {INC}_{i} + \beta _1 \text {INC}_{i} * \text {SD MARGIN}_{i} + \rho X_i + \epsilon _{i} \end{aligned}$$
(1)

In equation 1, Y\(_{i}\) is an outcome: either the SD vote share obtained in the 2016 parliamentary elections, or different fiscal measures. X\(_{i}\) is a vector of controls,Footnote 18 which we employ in certain specifications to assess the robustness of our estimates. \(\beta _0\) is the coefficient of interest. Its estimate gives the local average treatment effect of mayoral incumbency at the zero win margin threshold. When estimating Eq. 1, we use local linear regressions [LLRs] within a subsample of municipalities restricted in accordance to the bandwidth mentioned above. The standard errors used are robust to heteroskedasticity.

Fig. 3
figure 3

Reverse Coattails in an Unexpected Technocracy (Main Result). Note: We illustrate the discontinuous jump in SD vote share in the 2016 parliamentary elections, taking place at the zero SD MARGIN threshold

We begin by studying whether local incumbency benefited the SDs in the parliamentary race. Figure 3 addresses this visually. We find significant differences in the vote share obtained by the SDs when comparing constituencies where their candidates narrowly won or lost the local election. We document a large jump precisely at the zero win margin threshold, a result that is difficult to explain by models which do not account for the direct effect of mayoral partisanship.

Table 1 Reverse coattails in an unexpected technocracy (main result)

Quantitatively, the effect is meaningful politically as shown in Table 1. According to our preferred specification in Panel A, column (2), mayoral incumbency causally led to a strongly statistically-significant increase in the parliamentary vote share obtained by the SDs of eleven percentage-points on average.

It is reassuring that this estimate is largely unaffected by bandwidth alterations, or by the inclusion of electoral, demographic or policy-related covariates. Politically, the magnitude of this estimate suggests that local incumbency had a meaningful electoral impact, given that the retrieved point estimate is equal to roughly one fourth of the overall score obtained by the SDs in the parliamentary ballot—45 percent.

Now, the goal of studying reverse coattails absent a contemporaneous local-central party alignment is to rule out realized favoritism as a driver. As discussed above, the 2016 technocratic government in Romania had theoretically limited incentives to implement preferential policies. Here, we use local-level fiscal information to corroborate this empirically. The results are reported in Table 2, which shows the estimated effects of mayoral SD incumbency on local 2016 fiscal outcomes.

Table 2 Lack of evidence for realized favoritism under the technocratic government

All in all, we find little to no evidence for politically-discriminatory fiscal allocations. We are unable to reject the null hypothesis of a zero incumbency impact for any of the variables considered. Looking at central income in our preferred specification (column 5), the retrieved coefficient is not significantly different from zero, suggesting that the economic impact of sub-national incumbency is also negligible. The coefficient estimates from Table 2 are also graphically illustrated in Fig. 4 below.

Overall, these findings are reassuring in that we do not find evidence for favoritism behavior by the central government.Footnote 19 Taken together with the results above, where we document meaningful positive reverse coattails, we argue that the findings presented here shed light on the underlying mechanism linking local incumbency with national party performance. Chiefly, in line with hypothesis H1, our results suggest that concurrent central favoritism practices are not necessary for strong reverse coattails to materialize, complementing the existing scholarship which has so far identified the alignment bias as an important mechanism explaining the electoral value of local office control.

Fig. 4
figure 4

Lack of Evidence for Realized Favoritism under the Technocratic Government. Note: We graphically illustrate the coefficient estimates from Table 2

3.1 Does prospective voting partially explain our results? Suggestive evidence on the role of anticipated favoritism

The above findings suggest that reverse coattails may arise absent a retrospective voting channel hinging on realized favoritism by the central government. This constitutes our paper’s key contribution, adding to the broader literature on intergovernmental relations and the emergence of reverse coattails. Of course, as discussed in the introduction, if realized favoritism does not explain the documented reverse coattails effect, then what does? In this section, we provide several pieces of evidence suggesting that a prospective channel driven by anticipated favoritism may play a role. Concretely, we investigate whether voters are more likely to support their mayor’s party in anticipation of future preferential treatment, to be carried out after the elections, conditional on their mayor’s party ascending to central power.

Therefore, we conduct three further analyses corroborating anticipated favoritism as a driver of reverse coattails.

We begin by asking whether post-elections political favoritism actually occurred in this setting. Again, the political timeline investigated here provides a convenient way to address this: recall from Fig. 1 that the SDs regained control of the central government in 2017. Therefore, if prospective voting partially explains reverse coattails, we expect to see preferential transfers from the SDs after their return to power. This constitutes our second hypothesis.

Hypothesis 2: Once in control of the central government, political parties preferentially distribute resources to co-partisan mayors.

We empirically assess this prediction by quantifying the causal effect of mayoral partisanship on 2017–2018 central transfers. We show our findings in Table 3, panel A, where we aggregate the observations for both years.

Table 3 Central resources allocations under the social democratic government

Overall, these results corroborate the hypothesis whereby the now political government practiced preferential resource allocations. Looking at central income, the coefficient estimate suggests that SD incumbents received just under twelve-percent additional funds from the center, relative to mayors belonging to other parties, with similar effects documented for all the dependent variables. We illustrate these results graphically in Appendix Fig. B2.

We next address the question of anticipated favoritism credibility (Fehrler et al. 2020; Artés et al. 2022): when voting in the national elections, is it sensible for voters to anticipate their mayor’s party preferentially distributing “pork” if victorious? After all, one might posit that the key incentive for co-partisan favoritism (securing a higher vote-share) disappears after the elections.

We argue that anticipated favoritism is credible since voters know that future national elections will follow the one they are currently voting in, and anticipate preferential treatment accordingly (Ventura 2021)—in line with the canonical prediction of the discretionary inter-governmental grants literature (Arulampalam et al. 2009), asserting that central incumbents have incentives to favor aligned local incumbents. This leads to our third hypothesis.

Hypothesis 3: The preferential distribution of resources is more notable in periods leading up to national elections.

Again, the setting here allows us to study this empirically, since two further national elections took place in 2019: a race for the European Parliament in May, and a presidential race in November. Therefore, if electoral concerns indeed drive co-partisan favoritism, we expect the effect of mayoral partisanship on fiscal allocations to be stronger in 2018, a pre-electoral year, relative to 2017.

As shown in Table 3, we find supportive evidence: for example, looking at central income, the coefficient estimate for SD incumbency is approximately three times as large in panel B, where we restrict our sample to 2018, relative to panel C, where only 2017 observations are included. Once more, these findings are also graphically illustrated in Appendix Figs. B3 and B4.

Finally, we argue that if indeed anticipated favoritism plays a role, then we should expect reverse coattails to be stronger in constituencies where central resources are a meaningful component of local funding. We posit that the electorate in these centre-reliant constituencies generates greater expected utility by voting for those more likely to attract “pork” from the centre, relative to voters from constituencies that are better able to cover their expenses using internally-generated funds, in line with prospective voting. This constitutes our final hypothesis.

Hypothesis 4: Reverse coattails are stronger in constituencies where resources received from the central government are an important component of local revenues.

To test this, we first operationalize centre-reliant constituencies as those with above-median 2015 central income as a fraction of total income,Footnote 20 and fit equation 1 separately for the above and below-median subsamples.

Fig. 5
figure 5

Reverse Coattails: Heterogeneity by Central Funds Reliance. Note: For the full sample, we replicate the analysis from Fig. 3. Constituencies in the high (low) central-reliance subsample are those for which 2015 per-capita central income as a fraction of total income is above (below) the median measured in the full sample. The dependent variable is the 2016 parliamentary SD vote share

As shown in Fig. 5 and Appendix Table B2, the results of this heterogeneity analysis corroborate Hypothesis 4. The effect of mayoral incumbency is twice as large in center-reliant constituencies, relative to their more self-sufficient counterparts.Footnote 21

With these results in mind, we conclude this section by emphasizing that we are not arguing here that anticipated favoritism (and prospective voting more generally) is the sole driver of reverse coattails. After all, mayors have other means at their disposal to stimulate the vote, whose effectiveness does not hinge on voters being forward-looking.Footnote 22 Rather, our objective in this section was to provide several pieces of empirical evidence corroborating anticipated favoritism as a mechanistic driver for the reverse coattails effect, more generally adding to a larger literature attempting to quantify prospective voting using observational data.

3.2 Empirical design validity check

We address two issues that could invalidate the consistency of our main estimates. First, we consider the possibility of strategic treatment selection at the forcing variable’s threshold (McCrary 2008). This problem might arise, for instance, if the SDs committed voter fraud. If so, comparing constituencies where the SDs barely won and lost would not lead to consistent estimates of the incumbency effect. Intuitively, if strategic selection took place, the SDs would presumably be inherently more predisposed to illicit tactics in affiliated constituencies, which would lead to an overestimate of the true reverse coattails.

Fig. 6
figure 6

SD MARGIN Histogram. Note: We present the histogram of the running variable SD MARGIN around its zero threshold. Each bin represents a one percentage point interval

Fig. 7
figure 7

McCrary (2008) Test. Note: We perform the formal McCrary (2008) strategic selection test, which involves a kernel estimation of the density of local races involving the Social Democrats, performed on each side of the threshold. Bin width as in McCrary (2008). We restrict the sample to races where the win margin lies between ± 90 percentage points to diminish the influence of outliers

However, we note that illicit practices were contextually unlikely, as the 2016 local races were the first elections in Romania where polling stations were equipped with video cameras to prevent fraud. Corroborating the implications of this appealing feature, we provide visual evidence showing that the SD MARGIN\(_{i}\) varies smoothly at the discontinuity threshold in Figs. 6 and 7. Figure 6 presents a conventional histogram. In Fig. 7, we follow McCrary (2008) to address non-random sorting formally. As shown visually, there is insufficient evidence to reject the null hypothesis wherein the distribution of the running variable is continuous at the threshold, thus diminishing sorting concerns.

Second, we test whether observable pre-2016 variables differ between SD affiliated and non-affiliated constituencies at the zero win margin threshold. Results are provided in Table 4, where we focus on electoral and fiscal/demographic variables in panels A and B, respectively. Although the naive OLS regressions reveal significant associations between local incumbency and almost all the variables considered, these differences become statistically indistinguishable from zero in the RD design.

Overall, we are unable to statistically reject the zero effect null hypothesis for any variable. In light of these findings, the RD method appears contextually valid. Therefore, we posit that the results discussed above can (cautiously) be interpreted as consistent estimates of the causal effects of local incumbency.

Table 4 Design Check: Mayoral Partisanship and Past Observable Variables

4 Related literature

Our paper primarily adds to a growing literature investigating reverse coattails. As previously discussed, early work in Brazil by Ames (1994) documents a positive association between the party of municipal incumbents and the vote shares of the corresponding presidential candidates which he attributes to campaigning and opposition intimidation actions. Although insightful, whether differences in national performance can be causally linked to the partisanship of local politicians in settings such as this is unclear, due to potential issues of omitted variables or reverse causality.

More recent work addresses this internal validity issue by applying RD strategies similar to our own. In Germany, Hainmueller and Kern (2008) show that incumbency in single-member districts led to positive spillovers in elections occurring in a second, proportional representation tier, while Ade and Freier (2013) only find local incumbency effects on council elections if they are organized concomitantly with the subsequent local race. In other studies, Brollo and Nannicini (2012), Magar (2012), Dey and Sen (2016), Bonilla-Meíja and Higuera-Mendieta (2017), Avelino et al. (2017), Bardhan et al. (2018), Borcan (2020), and Ventura (2021) corroborate the existence of reverse coattails in Brazil, Mexico, India, Romania and Colombia. Conversely, in the US, Broockman (2009) shows that, although congressmen themselves enjoy incumbency premiums, these benefits do not spill-over to the presidential vote shares obtained by the incumbents’ parties. Similarly, Migueis (2013) documents that municipal incumbency did not affect national voter shares in Portugal.

A point of commonality between these studies is that a political party controlled the central government in the investigated settings, thus enabling realized favoritism as a mechanistic driver. In fact, several of these studies either explicitly attribute the reverse coattails effect they document to central co-partisan preferential treatment, or caution that it is currently unclear whether strong reverse coattails would materialize absent an alignment between administrative layers. For instance, Dey and Sen (2016) conclude that one of the parties they analyze (the TMC in India) benefited from local incumbency precisely because it practiced favoritism in the allocation of funds under the National Rural Employment Guarantee Scheme. Similarly, Bonilla-Meíja and Higuera-Mendieta (2017) argue that “(...) coalitions are able to assign discretionary transfers to aligned mayors”, thus creating “electoral advantages for their candidates in national elections”, while Firpo et al. (2015) find that voters supported precisely those candidates who brought resources to their local constituencies. Finally, Ventura (2021) posits that having access to central resources controlled by co-partisans is a core channel explaining the reverse coattails effect he documents in Brazil.

Our paper makes a contribution by assessing the reverse coattails effect in a setting where the central government is technocratic. In addition to extending the external generalizability of the existing studies to this alternative form of governance, this analysis helps us better break down the theoretical mechanisms underlying the reverse coattails effect.

As a secondary contribution, our retrospective versus prospective voting analysis adds to the scholarship exploring the mechanisms underlying democratic representation, which asks whether voters are more responsive to politicians’ past actions (Key 1966; Fiorina 1981) or expectations on future policies (Downs 1957).

Empirically, evidence on retrospective voting comes from studies showing that voting is influenced by changes in socio-economic conditions (e.g., Powell and Whitten 1993; Hopkins and Pettingill 2018; Artés et al. 2022), as well as natural disasters (e.g., Ramos and Sanz 2020), crime (e.g., Arnold and Carnes 2012), and the quality of local public goods (e.g., Burnett and Kogan 2017). Conducting a time-series analysis linking US presidential approval with consumer surveys, Norpoth (1996) concludes that “the results unequivocally reject the prospective claim and confirm the retrospective one”.

The prospective view has also received empirical support. Several studies focused on consumer sentiment (MacKuen et al. 1992; Erikson et al. 2000) or an economy’s stages of development (Hsieh et al. 1998; Singer and Carlin 2013) to provide suggestive evidence. More recently, work has turned towards the evaluation of political promises directly. Here, Elinder et al. (2015), using a difference-in-differences strategy, show that voters in Sweden responded prospectively to a 1994 proposed cut in childcare transfers. Similarly, Alpino (2018) finds evidence from Italy suggesting that voters acted prospectively in response to a surprise promise to abolish the property tax.

Our findings add to this sub-strand specifically, and extend the empirical support for prospective voting to local politics, by providing evidence suggesting that prospective considerations may partially explain why mayors matter politically.

Here, our results first constitute an additional piece of empirical evidence supporting the existence of strong reverse coattails. More importantly, our findings also suggest that prospective considerations concerning future inter-governmental transfers may partially explain why reverse coattails arise, adding to an emerging scholarship on why such coattails effects arise mechanistically (Avelino et al. 2017; Ventura 2021).

5 Conclusion

We investigate the reverse coattails effect following an exogenous tragic event which unexpectedly severed the partisan alignment generally linking local and central governments. Using novel constituency-level data from Romania and an RD strategy, we find evidence suggesting that local incumbency causally generated meaningful vote share premiums in the 2016 parliamentary race.

We argue that our findings further our understanding of intergovernmental relationships by providing an electoral justification for why parties direct large amounts of resources towards strengthening their hold on local offices (Brollo and Nannicini 2012). Because of the unique setting in which the 2016 elections took place, our results suggest that local incumbency remains a valuable electoral resource even under a technocratic government (Pastorella 2016), an alternative form of governance where the scope of contemporaneous partisan favoritism-driven measures is reduced—meachanisms that had been identified as key drivers of reverse coattails in previous studies (Samuels 2002; Baskaran and Hessami 2017).

That said, our findings do not imply that central favoritism is not pertinent in a dynamic sense: taking advantage of the technocratic government’s short-lived nature, we bring evidence or an anticipated favoritism channel exacerbating the reverse coattails effect. We show that, after the national elections, the now political central government carried out electorally-driven co-partisan “pork” disbursements, and that the documented reverse coattails are strongest precisely in centre-reliant constituencies—where one would expect anticipating preferential central policies to matter most. Thus, our results suggest that one important aspect future work estimating reverse coattails might consider is the nature of the central government resulting after the national elections. For instance, the expected party composition of the next central government may dictate the extent to which future favoritism is feasible, and thus whether mayors are valuable, potentially explaining existing disagreements in the broader scholarship.

Besides this political dimension, appraising the consequences of local incumbency allows us to better understand how electoral incentives shape local distributive politics more broadly. For instance, Romania is a transition economy where local administrations are characterized by relatively weak self-funding capacities, and hence a strong dependence on central resources. Such resources are used for local level investments in education, infrastructure, social services and so forth. Thus, better understanding how intergovernmental favoritism-driven relationships shape the allocation of these resources can ultimately help us better explain large variations in local development.

Finally, prospective voting is an important feature of electoral models employing sequential rationality to explain the political process (Konrad and Sherif 2019). However, empirically studying whether voters are forward-looking is challenging (Woon 2012)—to make progress, scholars were required to use an array of creative methods, ranging from laboratory experiments (Feltovich and Giovannoni 2015), to exploiting institutional features such as term-limits (Alt et al. 2011) and exploring political promises (Elinder et al. 2015). Our study contributes to this effort. While acknowledging that the external validity of our findings may be justifiably scrutinized, we argue that our results constitute a novel piece of evidence for the prospective voting view. More broadly, our work highlights how studying local politics can help us address broader questions on voting behavior.