Intragovernmental Grant Distribution and Party Alignment Bias under Democratic and Authoritarian Governments: The Case of Poland

Abstract

This article studies the party alignment bias in Poland. Relying on quasi-experimental regression discontinuity design and 2006–2019 data for over 2,400 municipalities, we aim to establish whether the impact of political alignment on the distribution of grants from the central to the local governments has increased under the rules of populist, authoritarian government. We report robust evidence that provides an affirmative answer to this question.

Keywords

party alignment bias local public finances regression discontinuity design intergovernmental transfers fiscal federalism

Introduction

Distributive politics has attracted considerable attention in the literature. One of its noteworthy aspects is the “party alignment bias”—the proposition that grant distribution from higher to lower levels of government is affected by party links between local and central governments and that lower level constituencies that are politically aligned with the central (or in general, higher level of) government are rewarded with more grants. The literature provides persuasive evidence on the links between party alignment and grant distribution.¹

Party alignment bias is harmful—it flaws the democratic process,² undermines the possible efficiency gains stemming from decentralisation,³ exacerbates the political business cycle,⁴ and may lead to increased corruption and lower quality of local government officials.⁵

In this article, we aim to revisit the question of whether the distribution of grants from the central to the local governments in Poland is affected by the political alignment of the local authorities with the central government. More importantly, we aim to verify the hypothesis that the scale of party alignment bias has increased under the rules of an authoritarian, populist government.

Poland is an interesting case study allowing us to dig deeper into the phenomenon of party alignment bias. The country, which was a forerunner of democratic transition in Central and Eastern Europe (CEE), has recently witnessed one of the biggest drops in democracy score⁶ in the region under the rules of an authoritarian, populist party Law and Justice (Prawo i Sprawieliwość [PiS]) that won the 2015 and the consecutive 2019 parliamentary elections.⁷ This allows investigating whether, under the rules of an authoritarian, populist government,⁸ the party alignment bias has changed.

Authoritarian governments aim to undermine plurality and democracy; consequently, they may also aim to strengthen the central authority and limit local autonomy—a phenomenon currently observed in Hungary and Poland.⁹ One of the instruments for reinforcing the power of the ruling party and for capturing the local governments is to manipulate the probability of re-election of the local, aligned politician; discretionary transfers that are biased in favour of aligned municipalities may be one of the methods. While—as the evidence clearly shows—alignment bias is observed in liberal democratic countries,¹⁰ we hypothesise that since authoritarian governments systematically dismantle a democratic system of checks and balances, such policies might be used even more eagerly by the authoritarian party. Our findings support this claim—they show that in Poland the party alignment bias is indeed stronger during the period when the populist, authoritarian government led by the PiS Party has been in power.

As far as we know, this is one of the few studies that focus on party alignment bias in CEE countries that tries to shed more light on distributive politics under the rules of the authoritarian party, using a quasi-experimental approach.¹¹ Among other studies that discuss the impact of party alignment on the distribution of public funds in CEE economies are those by Banaszewska and Bischoff,¹² who study the impact of party alignment between the regional and local governments on the distribution of European Union (EU) funds in Poland; Ionita,¹³ who studies the impact of political ties on the allocation of funds to the local governments in Romania; Flis and Swianiewicz¹⁴ and Sześciło et al.,¹⁵ who elaborate on the distribution of a special Central Government Fund for Local Investment in Poland in 2020 and show that aligned municipalities were the main beneficiaries. These studies, while very insightful, do not use the regression discontinuity design (RDD) approach, which offers many benefits for the empirical studies of phenomena, where it might be difficult to extract exogenous changes, including the party alignment bias.¹⁶

While our results need to be treated with reservation, as the data are still limited, we consider them important, especially for CEE countries. In the region, the democratic institutions are relatively young and weak, and authoritarian, populist governments are on the rise, threatening the very existence of liberal democracy, judicial independence, civil society, and free media.¹⁷ From a policy perspective, our results show that a rules-based, non-discretionary system of local government financing might be especially preferred in countries where non-democratic parties are on the rise. From the research perspective, more evidence for other countries regarding the party alignment bias under populist, authoritarian governments is needed.

The structure of this article is as follows. Section “Literature review” discusses the empirical literature on the effects of party ties on intergovernmental transfers, concentrating on studies that rely on the RDD, and states the research questions of this article. This is followed by a description of the institutional background of the Polish municipalities. The identification strategy, data, and results are presented in section “Data, Methodology, and Results,” and the last section concludes. In our study, we use the rich political and economic data on over 2,400 Polish municipalities and four consecutive elections at both the central and local level, between 2006 and 2019. We rely on quasi-experimental RDD to find the causal link between the alignment of the municipal mayor and the per capita amount of discretionary grants that the municipality receives from the central government.

Literature Review

There are a number of papers that theoretically discuss and empirically test the proposition that intragovernmental grants are allocated not only according to objective criteria but also in line with various political considerations and ties. An important issue studied by this literature is whether the political alignment between the central and local governments influences the allocation of grants. From a theoretical perspective, the political alignment bias has been explained, among others, in Brollo and Nannicini,¹⁸ Curto-Grau et al.,¹⁹ Bracco et al.,²⁰ and Arulampalam et al.²¹ These works show that transfers from the central government, by increasing the supply of public goods at the local level, may also increase the chance of re-election of the local politician currently in power. Under the assumption that the voters are not fully informed, the increased supply of (local) public goods serves as an indication of the high quality of the local politician.²² Since the central government—for different reasons—cares about electing the politically aligned incumbent,²³ the grants from the central level will be channelled to the aligned municipalities. Therefore, these models show that there might be a discretionary, upward jump in the size of grants transferred from the central level, in municipalities where the mayor belongs to the ruling party compared with municipalities where the non-aligned candidate has won the election. The only reason for this discretionary increase is political alignment.

From a normative perspective, the party alignment bias is a harmful policy. Borcan,²⁴ relying on data for Romania and RDD, rigorously demonstrates that local political alignment reinforced by increased transfers provides the ruling party a political capital in local governments, “which politicians can exploit at the expense of democratic values and institutions.”²⁵ Allocation of grants along political, instead of equity or efficiency, criteria is likely to hurt local development and long-run growth.²⁶ Finally, Brollo et al.²⁷ argue that windfall, non-tax revenues may increase corruption and rent-seeking as policymakers with larger budgets face a lower probability of punishment for corruption from imperfectly informed voters. They also show that higher rents increase the share of low-quality political candidates. Empirical evidence for Brasil reported in Brollo et al.²⁸ supports theoretical propositions. Hence, as noted by Bracco et al.,²⁹ there is “a perverse trade-off between the level of discretion in the distribution of intergovernmental grants and the disciplining and selection role of elections.”

Another strand of literature aims at empirically verifying the existence and scale of party alignment bias. There are a significant amount of papers that test the party alignment bias,³⁰ but only several, relatively recent studies test the alignment bias relying on the quasi-experimental method of RDD that compares municipalities, where the aligned candidate has (barely) won the elections to municipalities where the aligned candidate has (barely) lost. RDD, which we describe in greater detail in section “Data, Methodology, and Results,” allows for minimising the danger that the results are driven by endogeneity and by municipal characteristics that jointly determine the election results and the size of transfers.³¹ Among these studies is the work by Migueis,³² who confirms the bias for municipalities in Portugal; Bracco et al.³³ and Curto-Grau et al.,³⁴ who show evidence of the bias relying on data for Italy; or Brollo and Nannicini,³⁵ who report evidence supporting the existence of alignment bias for Brazil. Baskaran and Hessami,³⁶ studying intergovernmental grants in one of the German states, do not find a discontinuous jump in grants in favour of the aligned municipalities, but they do report that the alignment effect depends on the extent of overall local support—that is, while the simple alignment effect is not statistically significant, aligned municipalities receive higher transfers when the overall share of aligned municipalities exceeds a given threshold.

In this article, we are especially interested in examining the possible change in party alignment bias in Poland, after the year 2015 when the authoritarian, populist government came to power, as, to the best of our knowledge, party alignment bias has not been explicitly studied during the change from democratic to authoritarian governments. Yet, the hypothesis that authoritarian, populist governments are likely to increase the size of alignment bias may be justified by conclusions stemming from the existing studies. As Bracco et al.³⁷ show, the optimal size of grants transferred to aligned municipalities depends on their marginal benefits and marginal costs, as perceived by the central government. In the model presented by Bracco et al.,³⁸ the marginal benefit of transfers depends, among others, on the positive effect of transfers on the re-election probability of the aligned mayor, while the marginal cost is assumed constant and equal to 1. However, as party alignment bias is inefficient from the social point of view, voters may dislike it, which may decrease the chance of re-election of the central government. Hence, the marginal cost of grants to aligned municipalities may be conditional on the probability that voters will be informed about the alignment bias and its inefficiencies. To the extent that modern populist, autocratic governments enjoy undue electoral advantage elections—due to biased media coverage, restrictive campaign regulations, unequal suffrage, and so on³⁹—the probability of voters learning about the bias falls under autocratic government. Therefore, it might be argued that under authoritarian governments, the marginal cost of politically biased grants decreases, leading to higher grants to aligned municipalities.

From a different perspective, Bellodi et al.⁴⁰ and Sasso and Morelli⁴¹ show that populist governments are characterised by lower quality of bureaucracy and lower policy effectiveness. They discuss a vicious cycle of poor-quality bureaucrats and politicians maximising their short-term benefits. Hence, the proposition of a stronger party alignment bias under an autocratic, populist government might be also justified along the lines proposed by Bellodi et al.⁴² and Sasso and Morelli.⁴³ The populist government replaces high-quality experts with low-quality bureaucrats who optimally choose to implement the loyal policy instead of the efficient one. Hence, also grants from the central to the local governments are more likely to be distributed not according to efficiency criteria, but along the lines of party loyalty.

In this article, we aim to answer two research questions. First, we want to assess whether discretionary grants from the central to the local governments in Poland are higher for aligned municipalities, that is, municipalities where the mayor belongs to the ruling party. Second, we aim to check whether the party alignment bias—if it exists—is uniform across time or whether it has changed after the populist, authoritarian government came to power in the end of 2015.

Local Public Finances and Elections in Poland

Local Public Finances

In Poland, the local governments play a significant fiscal role: overall, their budgets encompass around 30 percent of total general government revenues and expenditures. The main unit of the local government is the municipality—of the 2,478 Polish municipalities, 303 are officially classified as urban, 1,559 as rural, and 616 as mixed urban-rural character.⁴⁴ The municipal responsibilities include the financing and supply of pre-school and primary education, communal services (water supply and distribution, sewage disposal, waste collection), the maintenance of local roads and street lighting, and in case of cities, also local public transportation. Other own municipal competencies include the supply of communal housing, spatial planning, selected social services, maintenance of voluntary fire brigades, local libraries, public parks, and graveyards.⁴⁵

Municipal revenues come from a number of sources. Own revenues (defined as all taxes, charges, and other payments where the municipality has some degree of autonomy⁴⁶) constitute approximately 30 percent of overall local revenues. They encompass real estate tax, agricultural tax, forest tax, tax on vehicles, and so on.

The remaining sources of revenue that finance local budgets are the following: shared taxation which includes personal income taxation (PIT) and corporate income taxation (CIT), which constitute around 20 percent of total municipal revenues, and various grants received from the central government and higher levels of the regional government that represent, on average, 48 percent of real per capita municipal revenues.

The grants transferred to municipalities from the central government (and from higher levels of the local governments) encompass the general purpose grants and conditional grants. The former have mostly a non-discretionary character and are distributed according to legally codified formulas based on predefined indicators⁴⁷; they are a major source of funding for all local governments in Poland. These grants amount to approximately 27 percent of total municipal income.

The remaining grants are conditional grants. These grants are distributed to local governments for several purposes. A significant part of these grants is transferred to finance expenses stemming from activities that were imposed on municipalities by the central government and that derive from certain legal agreements or social programmes that are administered by the local governments, but are fully financed by a grant from the central government.

However, these grants have also a fully discretional component that intends to support municipalities in implementing their own tasks. The decision regarding the distribution of these grants is made at the level of the central government. This discretional part of the grants transferred from the central to the local governments are the grants that we focus on.

More exactly, this study focuses on discretionary grants transferred from the central government to municipalities with the purpose of co-financing the municipal own, current, and investment spending. These grants, depending on the year, amount to an average 3 to 5 percent of the total municipal revenue (see Table 1). However, for 25 percent of municipalities with the largest share of discretionary grants in total revenue, these grants may be higher than 6 percent of the total revenue (see Table 2).

Table 1

The Share of Different Sources of Municipal Revenue in Total Municipal Revenue, for the Years 2006–2019.

	2006 (%)	2008 (%)	2010 (%)	2012 (%)	2014 (%)	2016 (%)	2018 (%)	2019 (%)
Own revenues	32	34	33	32	34	30	28	27
Shared taxation	21	24	20	21	21	21	23	23
General purpose, formula-based grants	28	27	27	27	25	23	21	21
Other grants, out of which	15	14	14	12	13	22	23	24
Discretionary grants	3.4	3.6	3.9	3.1	4.1	3.5	3.0	2.4

Source: Own calculations, based on Statistics Poland (www.stat.gov.pl).

Table 2

The Share of Discretionary Grants in Total Municipal Revenue: The Median Share and the Share in the First Quartile and the Third Quartile Where Municipalities Are Arranged According to the Size of the Share of Discretionary Grants in Their Total Revenue.

	2006 (%)	2008 (%)	2010 (%)	2012 (%)	2014 (%)	2016 (%)	2018 (%)	2019 (%)
First quartile	2.5	2.9	2.7	2.3	3.6	3	2.4	2
Median	3.7	4.2	4.2	3.5	4.8	4	3.4	2.7
Third quartile	5.4	6.2	6.5	5.	6.5	5.4	5	3.6

Source: Own calculations, based on Statistics Poland (www.stat.gov.pl).

These grants may finance infrastructure investments, sports facilities, scholarships, school allowances, and so on. Since they are discretionary, they are easy to manipulate. Furthermore, such grants may be used to finance highly visible projects, that is, they are an ideal target for local politicians willing to attract voters. The precise definition of these grants is provided in Table A1 in the appendix.

Central and Local Elections in Poland

In this article, we assess the municipal party ties with the central government, based on the party alignment of the mayor. Since 2002, municipal mayors are elected in direct, local elections held at the same time every four years; before 2002, mayors were elected by the local councils. Direct, local elections were held in autumn of 2002, 2006, 2010, 2014, and 2018. Since 2002, mayors may only be recalled by the citizens in the local referenda. Therefore, mayors are simultaneously politicians with strong electoral legitimacy and public officers heading the local administration.⁴⁸

Mayors may not only belong to national parties, but may also be nonpartisan and supported by local committees that are autonomous from national parties. Among political parties in Poland, only those represented in parliament count in local politics; the significance of other parties is marginal.⁴⁹

During the twenty-first century, national elections in Poland have been held in 2001, 2005, and 2007, and after that time, elections were held every four years: in 2011, 2015, and 2019. Since the parliamentary elections of 2005, two large parties—the right-wing PiS and centrist Civic Platform (Platforma Obywatelska [PO])—have dominated the national political scene and have formed successive governments.

The 2005 parliamentary election resulted in the victory of PiS, which formed a coalition government with League of Polish Families (Liga Polskich Rodzin [LPR]) and Self-Defence (Samoobrona). After the earlier parliamentary elections in 2007, the centrist PO ruled until 2015, in coalition with Polish People’s Party (Polskie Stronnictwo Ludowe [PSL])—an agrarian, Christian-democratic party. In autumn 2015, PiS returned to power after eight years in opposition.⁵⁰ PiS came to power with 38 percent of the popular vote, but secured an absolute majority of 235 of the 460 parliamentary seats. They formed a right-wing government, with strong ties to the Catholic Church. The 2019 parliamentary election in Poland resulted in continuity of rule of the incumbent PiS party. The government that was been formed was a coalition with parties: United Poland (Solidarna Polska) and Gowin’s Agreement (Porozumienie Gowina). As we have already stressed, the scholarly literature describes the PiS government (since 2015) as authoritarian and populist, comparing it to the Orban government in Hungary.⁵¹ After winning the 2015 elections, the PiS government has started to dismantle judicial independence and other democratic institutions.⁵² The policy implemented by the PiS government has caused the biggest drop in democracy score in the region over the years 2016–2021.⁵³ Before 2016, all ruling parties, including PiS that ruled in 2006–2007, were democratic, and the Polish democracy scores were high and stable.

In the next part, we aim to empirically verify whether and how the party alignment bias has changed during the period when these processes were unfolding.

Data, Methodology, and Results

Data Description and Methodology

Our empirical analysis is based on Polish fiscal and electoral data from 2006 to 2019 at municipal level, for over 2,400 local units. As already indicated, local elections were held in October or November 2006, 2010, 2014, and 2018, while parliamentary elections during this period were held in October 2007, 2011, 2015, and 2019. Since the elections in autumn of 2015, the country has been ruled by the authoritarian, populist party PiS.

The dependent variable in all regressions is the log of annual per capita discretionary grants that were transferred from the central government to municipalities to co-finance their own, local tasks. As explained in section 2 above, the purpose of these grants is to co-finance selected tasks that the municipalities are responsible for. These grants are fully flexible and discretional, hence perfectly suited to examine the plausible party alignment bias. Our goal is to verify the hypothesis of the party alignment bias in the distribution of these grants and, more importantly, to verify whether the bias has changed under the rules of the autocratic government.

However, identifying the alignment bias on grants is difficult due to possible endogeneity.⁵⁴ To alleviate this problem, we employ a sharp RDD. Sharp RDD is a quasi-experimental method that allows to control for endogeneity and omitted variable bias in estimating the impact of a well-defined discontinuity or the “treatment effect.” RDD compares a certain “outcome” variable among individuals or entities that have been subject to a “treatment” relative to units that did not receive the “treatment.” To execute this approach, a “running” or “forcing” variable that assigns “treatment” according to a well-specified cut-off value is needed. The assumption is that observations that are close to both sides of this cut-off are very similar and may be treated as if they were randomly assigned to the treatment. A crucial assumption is that there are no discontinuities in confounding variables at the cut-off.⁵⁵

Intuitively, the important advantage of using the RDD in studying the party alignment bias is that it allows comparing the size of per capita grants focusing on municipalities where the aligned politician has barely won, as opposed to municipalities where she has barely lost. It is assumed that municipalities in which the election results were very close are homogeneous, and any difference in the size of per capita grant can be attributed only to the mayor’s alignment and not to other characteristics. Therefore, the RDD allows to address the problem of endogeneity and eliminate the bias that might be due to preferences or other factors, which under alternative empirical approaches are difficult to control for. It is often called a “quasi-experimental” method, as the aim is to compare very similar units, where some units have (almost randomly) “received the treatment” (in our case, the aligned politician became a mayor) and others did not (in our case, the aligned politician has lost). Hence, it is assumed that any effect on the variable of interest (in our case, per capita grants) in only due to the “treatment” (aligned mayor) and not due to other differences between units.

Our identification strategy is to focus on local elections, where the partisan-aligned politician has won the elections and became a mayor or was the first runner-up. Our forcing variable is the vote margin (VM)—the difference between the share of votes received by the aligned politician and her competitor. A positive VM implies that the mayor belongs to the ruling party, and a negative VM means that the aligned politician has lost and her unaligned opponent is the mayor. Based on the VM, we construct the “treatment”—a dummy variable (Alignment) equal to 1 if VM >0 and equal to 0 if VM <0. Note that we have dropped all elections where neither the mayor nor the runner-up were members of the ruling party.

More exactly, we apply the RDD to estimate the following equation:

\begin{array}{l} S_{i, t + j} = β_{0} + β_{1} A l i g n m e n t_{i, t} + β_{2} p {(V M)}_{i, t} + β_{3} A l i g n m e n t_{i, t} \cdot (V M_{i t}) \\ + \sum_{z = 1}^{n} β_{z} C_{z, i, t + j} + e_{i, t} \end{array}

(1)

where:

S_i, _{t + j} is the log of per capita discretionary grants transferred to municipality i from the central government in year t + j, where j is from 1 to 4 and t is the election year: 2006, 2010, 2014, or 2018. Since elections are usually held in November of year t, we focus on outcome variable in year t + 1, t + 2, t + 3, and t + 4.

VM_i, _t is the VM, that is, the difference between votes cast on the aligned political and her competitor in elections in municipality i, in year t, where t is the year 2006, 2010, 2014, or 2018, and where the aligned politician has won the elections or was the first runner-up.

Alignment is a dummy variable that is equal to 1 if the aligned politician has (barely) won the elections in year t and became a mayor and is equal to 0 if the aligned politician has (barely) lost the elections in year t and the mayor is unaligned:

F_{i t} = 1 if M V_{i t} > 0

(2)

F_{i t} = 0 if M V_{i t} < 0

(3)

The function p is a linear or polynomial function of the forcing variable VM. The forcing variable is allowed to have a different slope to the left and right of the cut-off value; hence, we include the interaction of Alignment and VM.

Finally, in most estimations, we include the following:

A range “z” of covariates for each municipality i in years t + j, where j is between 1 and 4 (C_z, _{i, t + j}). Among covariates, we include the size of municipalities, measured by population (size), the age structure of the population (age), measured by the dependency ratio, and the level of municipal development. We do not have data on gross domestic product (GDP) per capita on municipal level, but we proxy them by the amount of local, per capita PIT revenues of the local government (taxes). Naturally, this proxy is far from perfect and needs to be treated with reservation; in Poland, farmers do not pay the PIT. Nevertheless, this measure is often used in empirical studies to model the approximate level of development of municipalities.⁵⁶

All variables are listed in Table A1 in the appendix.

Although in RDD including covariates is not necessary, researchers often augment the analysis by including them to increase efficiency.⁵⁷ As already indicated, in most of our estimations, we also control for other municipal characteristics, although we also show results without these covariates.

In all regressions, the most important coefficient is β₁ of the Alignment dummy, as it illustrates the “jump” or the percentage change in per capita grants for aligned municipalities, around the zero VM.

The RDD can be implemented using parametric or local, non-parametric analyses—to make sure our results are robust, we implement both. A key consideration when implementing the local, non-parametric approach is the choice of window or bandwidth around the discontinuity. There is a trade-off between statistical power and bias of the estimated coefficients. In the selection of the bandwidths, we rely on Calonico et al.⁵⁸ and use a data-driven, mean squared error (MSE)-optimal bandwidth, common on both sides of the cut-off. The MSE-optimal bandwidth is an upgraded version of the often-used Imbens and Kalyanaraman⁵⁹ (IK) and Calonico et al.⁶⁰ (CCT) bandwidths.

The political data were provided by the National Electoral Commission (Państwowa Komisja Wyborcza), and data on municipality characteristics have been taken from the Local Database of Central Statistical Office (Bank Danych Lokalnych GUS).

Regression Results

Before presenting the results, we need to make sure that the RDD is valid. A necessary condition for the validity of the RDD is that the running variable—in our case, the VM—has not been manipulated and sorted at the cut-off. In particular, it could be possible that the ruling party might tilt the elections in favour of its candidate.⁶¹ The local elections in Poland are usually closely observed by various party and independent organisations such as Watchdog Poland, and there has been so far no evidence that local elections were subject to fraud or manipulation.

The McCrary plot, shown in the appendix (Graph A1), does not show any signs of election manipulation in favour of the ruling party.

Another relevant test is whether the pre-treatment variables exhibit any discontinuity at the threshold. Such discontinuities are very problematic, as their presence makes it unclear whether the jump in transfers is explained by the change in alignment or by some other underlying variable. We find no evidence of such discontinuity. In Table A2 in the appendix, we present the relevant calculations, where we apply the RDD to model municipal characteristics, which we also use as covariates in our main analysis of party bias (i.e., municipalities’ size, age structure, and per capita PIT revenue). Each column in Table A2 corresponds to a particular dependent variable and a particular time period. Table A2 presents the results using the parametric approach; however, we also estimated these regressions using the local RDD with triangular kernel, and the results were very similar.⁶² Therefore, it seems that the RDD is valid and we may proceed to regression analysis.

Before we proceed to discussing the results, it has to be noted that in 2019 a new governmental, outside-the-budget fund was established, and one of its tasks was to co-finance the local roads, which has affected the amount and structure of the discretionary grant that we focus on. To make sure our results are robust, we calculated all regressions, excluding the year 2019, and the results were very similar.⁶³

Table 3 shows the results of the RDD where the left-hand-side variable is the log of annual, per capita discretionary grants transferred to the municipality from the central government, for the whole time period under study—the years 2006–2019. Following the literature, we just present the coefficient that illustrates the “jump” for the aligned municipalities. Columns 1–2 of Table 3 show the results for the parametric regression, encompassing the whole sample, using a first-order (Column 1) and second-order polynomial (Column 2) of the VM VT. Columns 3–5 contain results estimated by means of the local regression model, with optimal bandwidths; Columns 3 and 5 show the results when triangular kernel was applied, and Column 4, as well as Columns 1 and 2, contains the results with uniform kernel.⁶⁴ In most regressions, we include covariates: municipalities’ size, age structure, and per capita PIT revenue. Regression with no additional covariates is reported in Column 3. All regressions rely on robust standard errors, clustered at the municipality level.

Table 3

Regression Results for the Whole Period of 2006–2019.

Variables	(1)	(2)	(3)	(4)	(5)
Variables	Parametric regression	Parametric regression	Local, linear regression	Local, linear regression	Local, linear regression
Alignment	0.117***	0.117***	0.0571	0.110	0.032
	(0.035)	(0.035)	(0.052)	(0.067)	(0.077)
Covariates: size, taxes, age, time dummies	Yes	Yes	No	Yes	Yes
Kernel	Uniform	Uniform	Triangular	Uniform	Triangular
Observations	5,870	5,870	4,383	2,203	2,006

Note: Own calculations. This table reports the results of sharp RDD, where the dependent variable is the log of annual, per capita grants for own tasks. Aligned municipalities are defined as those where the mayor belongs to the party that forms the ruling collation at the central (state) level. Estimates in Columns 1 and 2 are from parametric RDD, with linear (Column 1) and second-order polynomial (Column 2). Estimates in Columns 3–5 are derived from a local linear regression. In the selection of bandwidths, we rely on Calonico et al.⁶⁵ and use a data-driven, mean squared error (MSE)-optimal bandwidth, common on both sides of the cut-off. Results shown in Columns 1–2 and 4–5 include covariates. Standard errors are reported in parentheses, are clustered at the level of municipalities, and are robust to heteroscedasticity; stars denote significance. Dependent variable: log of per capita discretionary grants on own tasks, from the central level to municipalities. RDD = regression discontinuity design.

p < .1. **p < .05. ***p < .01.

Parametric regressions (Columns 1 and 2) covering the whole sample show a statistically significant result of alignment on transfers from the central to the local governments, for both a linear and second-order polynomial impact of the VM. However, when we run a local regression, restricting the sample by means of optimal bandwidths, to close-margin elections only, the alignment effect turns out not to be significant (Columns 3, 4, and 5).

Hence, we conclude that for the whole time period 2006–2019, we do not detect a robust, statistically significant evidence of a party alignment bias in Poland. This is somewhat surprising, as most of the existing empirical evidence confirms the existence of the bias; the only paper we know that failed to find a significant effect, relying on a general RDD, was by Baskaran and Hessami⁶⁶ who studied the case of Hesse in Germany.

Graph 1 contains the illustration of the alignment effect in the period 2006–2018.

Graph 1

The RD plot: log of per capita grants on own tasks to non-aligned and aligned municipalities, 2006–2019.

Next, we turn to analysing two subsamples: years 2006–2015 and 2016–2019 to investigate whether the party alignment bias is different under democratic governments that ruled till 2015 and the PiS government that came to power at the end of 2015.

During the first period under study (2006–2015), there were three local elections: in 2006, 2010, and 2014; during the second period, there was just one local election: in 2018. This of course implies that the presented results need to be interpreted with much caution, as data are limited and surely more research is needed to further confirm or refute them. This is an interesting research avenue for the future.

Table 4 shows the results estimated for the first period 2006–2015, and Table 5 shows the results estimated for the second, “authoritarian” period, 2016–2019.

Table 4

Regression Results for the Non-Authoritarian Period of 2006–2015.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
Variables	Parametric regression	Local, linear regression	Local, linear regression	Local, linear regression	Local, linear regression	Local, linear regression
Alignment	0.104** (0.0455)	0.005* (0.068)	−0.0279(0.103)	0.037(0.102)	−0.092(0.112)	−0.172 (0.158)
Covariates: size, taxes, age, time dummies	Yes	No	Yes	Yes	No	Yes
Kernel	Uniform	Triangular	Triangular	Uniform	Triangular	Triangular
Observations	4,239	3,008	1,382	1,195	882	499

Note: Own calculations. This table reports the results of sharp RDD, where the dependent variable is the log of annual, per capita grants for own tasks. Aligned municipalities are defined as those where the mayor belongs to the party that forms the ruling collation at the central (state) level. The time period is when the democratic governments were ruling. Estimates in Column 1 are from parametric RDD. Estimates in Columns 2–6 are derived from local linear regressions. Columns 5 and 6 show calculations in the subsample restricted to years t and t−1, where t is the year of local elections. In the selection of the bandwidths, we rely on Calonico et al.⁶⁷ and use a data-driven, mean squared error (MSE)-optimal bandwidth, common on both sides of the cut-off. Standard errors are reported in parentheses, robust to heteroscedasticity; stars denote significance. Dependent variable: log of per capita discretionary grants from the central level to municipalities. RDD = regression discontinuity design.

p < .1. **p < .05. ***p < .01.

Table 5

Regression Results for the Authoritarian Period of 2016–2019.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Variables	Parametric regression	Local, linear regression	Local, linear regression	Local, linear regression	Local, linear regression	Local, linear regression	Local, linear regression
Alignment	0.152***	0.138**	0.1368*	0.163***	0.200**	0.0285	0.060
	(0.0419)	(0.064)	(0.077)	(0.063)	(0.102)	(0.0993)	(.082)
Alignment × Election Year						0.150** (0.0695)	0.280*** (0.1058)
Covariates: size, taxes, age, time dummies	Yes	No	Yes	Yes	Yes	Yes	Yes
Kernel	Uniform	Triangular	Uniform	Triangular	Uniform	Uniform	Uniform
Observations	1,631	833	511	876	238	467	465

Note: Own calculations. This table reports the results of sharp RDD, where the dependent variable is the log of annual, per capita grants for own tasks. Aligned municipalities are defined as those where the mayor belongs to the party that forms the ruling collation at the central (state) level. Estimates in Columns 1 are from linear, parametric RDD. Estimates in Columns 2–7 are derived from local linear regressions. Column 5 shows calculations in the subsample restricted to 2017–2018, where 2018 is the local election year. In the selection of the bandwidths, we rely on Calonico et al.⁶⁸ and use a data-driven, mean squared error (MSE)-optimal bandwidth, common on both sides of the cut-off. Standard errors are reported in parentheses, are robust to heteroscedasticity, and clustered at the level of municipalities; stars denote significance. Dependent variable: log of per capita discretionary grants from the central level to municipalities. RDD = regression discontinuity design.

p < .1. **p < .05. ***p < .01.

As Table 4 reveals, during the period 2006–2015, there is no evidence of a statistically significant “jump” in grants for municipalities, where the aligned politician has barely won the elections, that is, no evidence of party alignment bias. Table 4, Column 1 contains the results of the parametric regression using all observations in the sample, and the remaining columns show the results of local regressions, with observations restricted by optimal bandwidths. All estimations rely on robust standard errors, clustered at municipality.

While alignment effect is significant in parametric regression (yet smaller in size, compared with the period 2006–2019), it loses significance in local regressions, with covariates. We also check for the alignment effect, when the sample is restricted to local election year and one year prior to elections (hence 2006; then 2009–2010 and 2013–2014). The alignment effect can be particularly strong before elections, as the authorities try to increase the chance of the aligned mayor of winning the elections. The results are presented in Column 6. Again, we do not find evidence of statistically significant party alignment effect in election years.

The results for the sub-period 2016–2019—the time when PiS has been in power—are reported in Table 5. The results indicate a statistically significant and robust aliment effect (we show many different results to make sure that the statistically significant alignment effect is robust). Columns 1 to 4 show a statistically significant alignment effect in both parametric and local regressions; the local regressions are estimated with either uniform or triangular kernel to make sure that the results do not change under different kernels. The size of the alignment effect is substantial—the coefficient suggests that aligned municipalities, where the aligned mayor has barely won, received approximately 13–15 percent more grants compared with non-aligned municipalities, where the aligned politician has barely lost.

In Column 5, we report results when the sample is restricted pre-election and election year (2017 and 2018). The alignment effect is significant and the value of the coefficient is higher, compared with the results reported for the sample covering the whole period 2016–2019.

We also include an interaction term between the Alignment and local election year. In including an interaction between the Alignment and another covariate, we follow Baskaran and Hessami.⁶⁹ The interaction term between the election year and Alignment is positive and significant, suggesting that the party alignment bias is especially visible during local election years.

To further check the robustness of the results, we follow Bellodi et al.⁷⁰ and address the possibility that higher per capita grants are confounded by incumbency effects. The likelihood for aligned candidates to be incumbents might be different compared with other candidates, and the differences in grants might be the product of different experiences. Therefore, we exclude from the sample all incumbent mayors and present the results in Column 6. The exclusion of incumbent mayors does not significantly change the results.

Summarising, the results presented in Tables 4 and 5 show that indeed in Poland the party bias became more pronounced after 2015 that corresponds with the rules of the authoritarian, populist PiS government. The party alignment bias seems to be particularly strong during election years, indicating that the grants mostly aim at increasing the chance of winning the next elections for the aligned mayor.

The graphical illustration of these effects is shown in Graphs 2 and 3. Graph 2 shows results for the period 2006–2015 and Graph 3, for the period 2016–2019. Graph 3 shows a clear jump in per capita grants at the cut-off (i.e., in aligned municipalities) after year 2015, while Graph 2 shows no evidence of the jump before 2016.

Graph 2

The RD plot: log of per capita grants to non-aligned and aligned municipalities, before the year 2006–2015.

Graph 3

The RD plot: log of per capita grants on own tasks to non-aligned and aligned municipalities, 2016–2019.

Hence, both the regression and graphical analyses confirm that party alignment bias in Poland grew stronger over time, and the time during which the bias is clearly visible in data corresponds to the period when the authoritarian, populist government took power. To make sure that the difference between two different time-periods is robust, we also implement RDD subgroup analysis. The procedure relies on propensity score weighting implemented to increase the similarity of the subgroups.⁷¹ The subgroup analysis confirms the difference in the alignment effect between the sub-periods.

In our case, the two subgroups are defined as pre-2015 and post-2015. The results of the analysis confirm the different alignment effects in the two time-periods. In each case, the bandwidth has been chosen, like in other estimations, to minimise the MSE of the local polynomial RD estimator. The covariates used were: the size of municipality, the log of per capita personal tax income, and the percentage of the adult population. The results are presented in Table A3 in the appendix.

Conclusion

This article studied the party alignment bias in Poland. Relying on RDD and 2006–2019 data for over 2,400 municipalities, we have revisited the question of whether the distribution of grants from the central to the local governments is affected by political alignment of the local authorities with the central government. We found that over the whole time period under study, the party alignment bias is not statistically significant. However, when we divided the sample into two subsamples—2006–2015 and 2016–2019, we found robust evidence that the party alignment bias was not uniform across these time-periods, but has increased after 2015, which corresponds to the period when populist, authoritarian government of PiS took power. In the years before 2016, party alignment bias turned out not to be robustly different from zero; while in the years 2016–2019, it is positive and robust. The bias is particularly strong during election years, indicating that it serves to increase the chance of winning elections by the aligned incumbent mayor.

Although these results need to be interpreted with great caution, they suggest that in Poland the party alignment bias has increased in the period when an undemocratic, populist government took power. Naturally, these results are based on limited data, and hence, much more research is needed before any definite conclusions are formulated. However, we strongly believe that these results deserve attention. First and foremost, the causal link between populist, nondemocratic governments and party alignment bias is plausible, but much more evidence, for different countries and time-periods, is needed. Hence, our results show a new, important research avenue. From the policy perspective, our results suggest that as party alignment bias exists and under different governments may be more or less pronounced, rules-based as opposed to discretionary grants to local governments seem to be the preferred option that can increase efficiency of grants.

Footnotes

Appendix

Table A3

Subgroup Analysis: The Two Subgroups Are Pre-2015 and Post-2015.

Subgroup	Alignment
1 (before year 2015)	−0.077 (0.107)
1 (after year 2015)	0.155** (0.076)
Difference	0.232* (0.132)

Note: The results of sharp regression discontinuity design subgroup analysis. The procedure relies on propensity score weighting implemented to increase the similarity of the subgroups.⁷³ The two subgroups are defined as observations before 2015 and after 2015.

ORCID iDs

Joanna Siwińska-Gorzelak

Grażyna Bukowska

Notes

Joanna Siwińska-Gorzelak is a professor in the Faculty of Economic Studies at the University of Warsaw. She specialises in local public finance, fiscal policy, growth theory, and political economy.

Grażyna Bukowska is a senior lecturer in the Faculty of Economic Studies at the University of Warsaw. She specialises in local public finance, political economy, and EU policies.