electoral vulnerability and size of local governments ... · electoral vulnerability and size of...
TRANSCRIPT
Electoral vulnerability and size of local governments:
Evidence from voting on municipal mergers
Ari Hyytinena Tuukka Saarimaab Janne Tukiainenc
aJyväskylä School of Business and Economics, University of Jyväskylä. P.O. Box 35, 40014 University
of Jyväskylä, Finland. Email:[email protected]. bGovernment Institute for Economic Research VATT. P.O. Box 1279 (Arkadiankatu 7), FI-00101
Helsinki, Finland. Email: [email protected]. cCorresponding author. Government Institute for Economic Research VATT. P.O. Box 1279
(Arkadiankatu 7), FI-00101 Helsinki, Finland. Email:[email protected]. Tel: +358295519 451.
Helsinki Center of Economic Research. Arkadiankatu 7, 00100 Helsinki, Finland.
Abstract
We analyze how anticipated changes in the electoral vulnerability of municipal
councilors affect their voting behavior over municipal mergers. The electoral
vulnerability changes due to a merger because it changes the composition of political
competitors and the number of available seats in the next election. We use this variation
for identification and find that the smaller the increase in the electoral vulnerability of a
councilor, the more likely he is to vote for the merger. The documented effect is not
driven by the behavioral response of the voters, or by party-line considerations. The
councilors’ desire to avoid personal electoral competition may lead to sub-optimally
small municipalities from the local citizens’ point of view.
Key words: Electoral vulnerability, local politics, municipal mergers
JEL classification numbers: H11, H77, D72
1. Introduction
The efficient provision of local public services depends on the number and size of local
jurisdictions, such as municipalities and school districts. Boundary reforms and regional
amalgamations are a means to change them, but the required adjustments are often
conflict-prone and difficult to achieve politically. It is thus natural to ask, do politicians
want to reset the boundaries of their local jurisdictions, if given a chance to cast a vote
in favor of such a change? If not, why?
The economics literature on the endogenous formation of political jurisdictions
concentrates on the determinants of jurisdiction size (e.g. Miceli 1993, Alesina and
Spolaore 1997, Ellingsen 1998, Bolton and Roland 1997, Casella 2001). The core
theoretical result from this literature is that the optimal size is determined by a trade-off:
The existence of economies of scale and inter-jurisdictional spillovers favors large
jurisdictions while regional heterogeneity in preferences over local public goods favors
small jurisdictions.1 This trade-off appears to be real, as people seem to be willing to
forego the scale benefits in order to avoid heterogeneity in their local jurisdictions (e.g.,
Brasington 2003, Alesina et al. 2004 and Gordon and Knight 2009, Hanes et al. 2012,
Saarimaa and Tukiainen 2014).
A clear gap in this literature is that both empirical and theoretical work abstracts
away from possible political agency and electoral concerns in the formation of local
jurisdictions and resetting of their boundaries. In this paper, we put this question on the
center stage. We do so by asking how local politicians’ (municipal councilors) re-
1 A large number of local jurisdictions can be optimal also due to yardstick competition, if there are informational asymmetries between politicians and voters (Besley and Case 1995), or due to tax competition, if the public sector is a revenue maximizing leviathan (Brennan and Buchanan 1980, Edwards and Keen 1996). Perroni and Scharf (2001) show that tax competition can lead to larger jurisdictions in equilibrium.
2
election concerns affect, independently of voter preferences and their party-line
pressures, whether they vote for or against a municipal merger and to what extent their
concerns of (increased) electoral vulnerability are reflected in the resulting municipal
structure.
A municipal merger may affect a councilor’s electoral vulnerability in future
elections in a number of ways. First, a merger is a policy decision that affects the
service-tax bundle provided by the municipality. Voters may punish or reward a
councilor for these changes in policy. Second, a merger changes the boundaries of
electoral districts, and thus, has a direct effect on electoral competition and electoral
vulnerability of a councilor. The way this latter channel works depends on the
particularities of the election system.
This paper’s empirical context is a multi-party, open-list proportional
representation (PR) system where, after having been elected but with an eye to future
elections, local politicians decide whether or not to vote for a municipal merger. There
are three potential levels of political decision-makers in such a system: First, a
municipal council may vote according to its true electoral incentives, representing the
interests of the pre-merger municipality, and thus the preferences of the councilors’
current constituency. The second level is political parties, at the national and also pre-
and post-merger municipality levels. In our context, the parties and the political
coalitions to which they belong can both gain and lose local political power due to the
proposed merger. These party-line considerations may largely determine whether the
councilors vote for or against the merger. Third, it may be the self-interest of the
councilors that drive their voting decision. For example, a councilor may want to
3
prevent the merger from taking place if it is expected to increase his electoral
vulnerability in the future elections.
Taken together, the foregoing suggests that there is a myriad of interdependent
electoral, party and individual effects which determine whether incumbent local
politicians want to vote for a change in the boundary of their local jurisdictions. We aim
at quantifying one of these effects, the behavioral response of the incumbent politicians
to the (anticipated) electoral consequences of a change in the boundary. We estimate
this councilor-level response by using data from a recent wave of municipal mergers in
Finland and by focusing on the within municipality and within party-line variation in
the data.
The voting decision is directly and predictably linked to the councilors’ electoral
vulnerability because if the merger goes through, it changes not only the composition of
voters, but also the set of political competitors and the (relative) size of the municipal
council. We show how these changes can be captured empirically and explore whether
they affect the voting decisions of local politicians. In this regard, we follow the recent
work by Blais et al. (2011) and Fiva and Folke (2014), who study electoral reforms and
estimate their mechanical (i.e., how the reforms change the way votes are transformed
into seats for a given vote distribution) and behavioral (i.e., how voters, parties or other
political actors react to the reforms) effects on ex post electoral outcomes.2 We ask,
instead, do the merger-induced mechanical effects have a feedback effect on the
councilors’ voting decisions.
Our key empirical finding is that a councilor is less likely to vote for a merger if
his electoral vulnerability increases in the merger state relative to the status quo of no
2 This taxonomy dates back to Duverger (1954), but he used the term ‘psychological’ instead of ‘behavioral’.
4
merger. We find, in particular, that the anticipated changes in the composition of voters
and political competitors are important in explaining this councilor-level behavioral
response. This result is robust to holding electoral (municipality-level) and party-line
(party-municipality -level) merger incentives constant. The source of the councilor-level
desire for maintaining the status quo is thus neither the councils’ (uniform) electoral
incentives at large, nor party-line considerations. Our findings imply that when
boundary reforms are delegated to the local level, local politicians’ desire to avoid
political competition may lead to too little consolidation, and thus to jurisdictions being
too small from the point of view of the economy at large. We conclude that the self-
interest of local politicians is, at least on the margin, a source of resistance to
consolidation and may lead to sub-optimally small municipalities.
The remainder of this paper is as follows: In the next section, we provide a brief
literature review. We describe the institutional framework and our data in Section 3. In
Section 4, we discuss the various mechanical and behavioral effects that a merger may
have on different political actors. We present our empirical approach in Section 5 and
report our results in Section 6. Section 7 concludes.
2. Literature Review
Besides the literature on the endogenous formation of political jurisdictions (as
discussed in the introduction), our paper builds on four other branches of the prior
literature in economics and political science.
The first branch of papers has evaluated the effects of mergers both from pre- and
post-merger perspective. The pre-merger effects are related to common pool problems
after a merger is decided, but before it comes to effect. For example, Hinnerich (2009),
5
Jordahl and Liang (2010) and Saarimaa and Tukiainen (2013a) find that merging
municipalities free-ride on their merger partners by increasing debt prior to the merger
taking place. Another group of papers has analyzed the effects of mergers on post-
merger outcomes. For example, Reingewetz (2012) and Blom-Hansen et al. (2014) find
that mergers decrease local government expenditures, while Lassen and Serritzlew
(2011) and Saarimaa and Tukiainen (2013b) document that mergers have an effect on
political efficacy and voting behavior.
The second branch of the literature deals with the disciplinary effects of elections
and the way re-election concerns affect politicians’ policy choices. Much of the earlier
empirical literature builds on term limits, which make incumbents ineligible to run for
re-election after a certain number of terms in office. The typical finding is that term
limits have clear effects on implemented policies and politicians’ effort (e.g. Besley and
Case 1995, Dal Po and Rossi 2011, Ferraz and Finan 2011). This suggests that elections
discipline politicians.
The third related branch of the literature builds on Duverger’s (1954) insights and
focuses on the estimation of mechanical and behavioral effects of electoral rules and
reforms. Lijphart (1990) studies cross-country variation in electoral systems and
outcomes from 1945 to 1985 and argues that the mechanical effects of the electoral
system variables are large. Benoit (2001) finds that both district magnitude and the
electoral formula influence the number of parties in Hungarian local elections. Cox et
al. (1999 and 2000) demonstrate, in turn, that electoral rules affect the number and
characteristics of intraparty factions in Japan. Blais et al. (2011) use variation in the
electoral systems in nine elections in Switzerland and four in Japan to distinguish
between the mechanical and psychological effects of the electoral systems on the
6
effective number of parties. They find evidence for behavioral responses by parties and
voters, but the mechanical effects appear to dominate in many cases. Fiva and Folke
(2014) study a nationwide change in the seat allocation method in Norway, which
mechanically raised the proportionality of the seat allocation. They find evidence both
for mechanical and behavioral effects, but the relative importance of the two appears to
depend on the outcome. Fiva and Folke also found evidence for a strategic response by
the incumbent politicians, who appear to reduce the size of the local council in order to
mitigate the effects of the reform-induced increase in proportionality.
The fourth branch of the literature studies the political and economic
consequences of redistricting (including gerrymandering) and annexation in a
majoritarian election context. The papers in this branch have explored both the effects
of such boundary reforms on electoral outcomes (e.g., Gelman and King 1990, 1994)
and on politician behavior, especially in the US (e.g., Glazer and Robbins 1985 and
Levuax-Sharpe 2001, Boatright 2004). This US evidence shows that the effects of
boundary reforms and redistricting depend in a subtle fashion on by whom and how the
district boundaries are redrawn (e.g., Gelman and King 1994, Carson and Crespin 2004)
as well as on how easily the politicians can anticipate and prepare for the political
consequences of redistricting (Boatright 2004). The tradeoffs are real and severe, as
redistricting may for example lead to a loss of the incumbency advantage (Desposato
and Petrocik 2003).3
3 Coate and Knight (2007) consider socially optimal redistricting and the role of electoral competitiveness therein. Districting and resetting of the local government boundaries have also been found to shape policy outcomes (Ansolabehere et al. 2002, Besley and Preston 2007).
7
We contribute to these four branches of the prior literature by estimating the
feedback effect of a reform-induced mechanical change in the electoral vulnerability on
the ex ante behavior of local politicians in the context of a PR system.
3. Institutional background and data
Our analysis uses data from Finnish municipalities and concentrates on the behavior of
municipal councilors who were elected in 2004 for a four year term and who voted for
municipal mergers that eventually took (or, if turned down, did not take) place between
2007 and 2009. The next elections were held in October 2008 and they took into
account the mergers that would then take place subsequently at the start of 2009.
3.1 Local decision making and municipal mergers
In Finland, public goods and services are provided by two tiers of government where
municipalities constitute the local level. The Finnish public sector is highly
decentralized and municipalities are responsible for providing more services than in
most other countries, including social and health care services and primary education.
Municipalities are therefore of considerable importance to the whole economy, with the
GDP share of municipality spending being roughly 18%.
During the past decade smaller municipalities have found it increasingly difficult
to provide the large scale of services that they are responsible for. These difficulties are
due to many things, such as population aging and internal migration from rural to urban
8
areas. As a result of this, there has been a constant pressure to reduce the number of
municipalities in recent years.4
Each municipality has a council which is responsible for all major strategic and
financial decisions.5 Councils are elected every four years using open list elections,
which apply the D’Hondt method. Parties select the candidates, but voters determine
their order within the lists.6 Each municipality has only one electoral district. Council
size is a step function of the municipality’s population and is determined by law as
follows: 13, 15 or 17 for municipal population 2000 or less, 21 for 2,001–4,000; 27 for
4,001–8,000; 35 for 8,001–15,000; 43 for 15,001–30,000; 51 for 30,001–60,000; 59 for
60,001–120,000; 67 for 120,001–250,000; 75 for 250,001–400,000 and 85 for over
400,000.
Being a member of a municipal council is a part-time job, with meetings taking
place monthly. There are limited pecuniary rewards from having a seat in the council.7
Of course, holding a council seat can also generate other rewards, such as prestige,
better chances of getting elected in the national parliamentary elections, improved non-
4 In 2005, the central government initiated a plan that aimed at strengthening the operating environment of the municipalities. The plan aimed at reforming municipal revenue structure and making the provision of municipal services more efficient. In 2007, a provisional law was enacted, stating that mergers between municipalities are the main tool for achieving sounder municipalities. However, the merger decisions were left to be made voluntarily by the municipality councils. 5 The council also chooses the municipal board, which has a preparatory role. The composition of the board is based on party shares in the council, i.e., each party in the council get seats in the municipal board according to their share of council seats. 6 In the open list D’Hondt method, each voter casts a single vote to a single candidate. Parties gain seats based on the sum of votes that their candidates get. Within the parties, the seats are allocated by ranking the candidates based on their individual votes. There are eight parties in the Finnish parliament, which also dominate the municipal politics. Some local single-issue groups exist as well. The parliament and municipal councils are dominated by the three biggest parties with a combined overall share of votes of around 60 percent in both 2004 and 2008 elections. 7 The reward consists mainly of meeting fees, which vary roughly from 50 Euros to more than 300 Euros per meeting. There are also separate fees for subcommittee meetings, such as the subcommittee of education or health care. Council and municipal board chairmen also get an annual fee on the top of the basic meeting fees. All the fees increase with municipality size.
9
political labor market prospects and the ability to move policy closer to their preferred
point (e.g. Besley and Coate 1997, Lundqvist 2011, Kotakorpi et al. 2013).
Mergers between municipalities are voluntary and the municipality councils are
allowed to decide which potential mergers they consider. A typical merger process is as
follows: After an initial feasibility study, the municipal boards make a proposal of the
merger to the municipal councils. This proposal is voted on by the councils. In about
half the cases, the potential merger includes more than two municipalities. The merger
votes are in most cases conducted simultaneously among the municipalities
contemplating a merger.8 If the proposed merger gains a majority in all the participating
councils, the merger goes through. If not, it is cancelled and the municipalities continue
as they were.
If the merger goes through, the next elections are organized so that the entire new
municipality constitutes a single electoral district (i.e., there are no regional quotas, or
equivalent, for the pre-merger municipalities). We are thus analyzing redistricting
across many at-large elections. This environment differs from e.g. the U.S. congress,
where voting is by electoral districts and where redistricting refers to the process of
redrawing their boundaries.
3.2 Data sources
We have collected data on how each individual councilor voted in the merger votes.
These data were collected separately from each municipality and were often available
online. We have linked these voting data to the data on municipal elections held in
8 The timing is not entirely simultaneous, because the lengths of council meetings differ. Moreover, it seems that in some rare cases, the voting was sequential on purpose.
10
2004. These elections data were provided by the Ministry of Justice.9 We focus on
councilors representing the eight largest parties. Our final data consist of 3,804
individual councilors coming from 135 municipalities and 59 (potential) municipal
mergers. Out of the 135 municipalities in our data, 99 eventually underwent a merger.10
Besides data on the election outcomes, such as the number of votes the councilors
received in the 2004 elections, we have data on the councilors’ age, gender, and whether
they were elected in the 2000 elections. In addition to this, we have augmented the data
with variables describing different municipal characteristics, such as population and
mean income, provided by the Statistics Finland. Moreover, we used geographic
information system techniques and Statistics Finland’s Grid Database to calculate the
mean distance of municipal population to the center of the municipalities that
contemplate a merger. We report descriptive statistics for these variables in Table A1 in
Appendix A.
4. Mechanical and behavioral effects in municipal mergers
Municipal amalgamations are boundary reforms that change the size and political
landscape of local jurisdictions. If an amalgamation goes through, it reduces the size of
the municipal councils relative to the size of the municipalities, changes the
composition of voters, and mixes the set of political competitors from the merging
municipalities. It thus shares features of electoral reforms and redistricting (and also of
9 The data are managed on behalf of the Ministry of Justice by a commercial operator (Tieto Oyj). 10 Between 2004 and 2008 elections, 130 municipalities decided to merge. We were unable to collect individual councilors’ voting behavior from 31 merging municipalities. In a few cases, the same municipality was involved in two separate merger votes during the time period. These were, however, separate merger processes. For example, the city of Rauma underwent one merger in 2007 and another in 2009.
11
annexation) and is likely to have an effect on electoral outcomes. Building on this and
on Duverger (1954), Blais et al. (2011) and Fiva and Folke (2014), we predict that the
resetting of municipal boundaries have the following mechanical and behavioral effects.
Mechanical effects: The resetting of municipal boundaries results in two
mechanical effects, which reflect how the reform changes the way votes are transformed
into seats (when the political actors do not change their behavior). First, if a municipal
amalgamation takes place, it increases the size of the local jurisdiction. The resulting
change in the council size increases the vulnerability of marginal seats, because council
size is, by law, an increasing but concave function in the population of the
municipalities. From now on, we call this Council size mechanism. Second, the resetting
of municipal boundaries mixes the set of candidates competing against each other for
the seats in the post-reform council. Henceforth, we call this Competition mechanism.
We call the sum of the two Total mechanical change.
To see why both Council size and Competition mechanisms are at work, consider
the (open-list) D’Hondt method, where voters cast votes to individual candidates and
where the available seats are allocated to multiple parties based on the total vote count
of the parties’ candidates. Furthermore, in the D’Hondt system, a candidate’s election
outcome depends not only on his personal and his party’s number of votes, but also on
how the votes are distributed to the other candidates and parties. This means that a
candidate can be a close competitor to multiple candidates both from his own party and
also from the other parties.
In such a seat allocation method, the two mechanical effects affect both parties
and individual councilors. The reason for this is that when at work, Council size
mechanism mechanically removes one or more of the marginal seats. Besides having an
12
adverse impact on a single marginal candidate, this reduction may disproportionately hit
those parties that have multiple marginal candidates on their lists.
Competition mechanism in turn reflects how merging two or more given pre-
merger vote distributions of the merging municipalities affects the vote shares of
candidates and parties, and thus the election outcomes. These resulting changes affect
competition both between and within parties. First, the number of seats that party A gets
can change either when party A’s own vote share changes, or when party B’s vote share
increases and party C’s decreases, even if A’s vote share remains the same. These party
level effects may affect individual candidates in different ways. Second, the election
outcome for a candidate depends on his within party rank. A merger may profoundly
affect the within party rankings, especially for the candidates that come from relatively
small municipalities. This means that changes in the vote distribution and composition
of the party lists can change the election outcome in a complex way both for the
candidates and for their parties.
It is worth pointing out that Competition mechanism is not explicitly driven by an
electoral rule, but Council size mechanism is. Council size mechanism is thus a policy
variable, determined at the national level. Indeed, in the first elections after the merger,
some municipalities were allowed temporarily to use a larger council size than the
council size step function would indicate. This suggests that introducing a less concave
council size schedule might be an instrument to induce more mergers and that Council
size mechanism may be more salient to the councils, parties and councilors. If it is, it
could even be a source of conformity in the voting behavior. These observations provide
a motivation for us to try to distinguish between the effects of the two mechanisms.
13
Ex post voter, party and councilor behavioral effects: The resetting of
municipal boundaries may lead to an ex post behavioral effect by voters, parties and
councilors who in the subsequent election decide how to adjust and react to the merger.
For example, there can be an ex post behavioral effect by the councilors, as the resetting
of municipal boundaries may change their willingness to rerun for a seat as well as their
campaign efforts in the post-merger election, as compared to the state of affairs when
the boundaries remain intact. One reason for this is that resetting the municipal
boundaries may cut some voters loose from their old representative in a heterogeneous
way (as e.g. the work by Desposato and Petrocik 2003 suggests). Moreover, parties,
and/or voters may adjust strategically (as e.g. the work by Boatright 2004, Fiva and
Folke 2014 and Saarimaa and Tukiainen 2013b suggests). Voters and parties may, for
example, adjust to the new political landscape by e.g. reducing turnout or entry.
Ex ante voter, party and councilor behavioral effects: Voters, parties and
councilors can also respond to the proposed amalgamations ex ante, before they are
decided upon. Voters may, for example, have been forward-looking and voted with the
subsequent consolidation in mind in the municipal elections that preceded the merger
votes. Parties, representing the supply side of politics, can both gain and lose their
political power due to the proposed amalgamations.11 Because parties are forward-
looking, too, this can lead to various party-line considerations both at the local and at
the national level. Such considerations may even determine how party-loyal councilors
vote in the merger votes.
11 See e.g. Howitt and Wintrobe (1995), who consider the possibility of inefficient political inaction in a majoritarian, two-party system. A main driver of the desire for the status quo is the desire of the political parties to avoid electoral competition. Such party-considerations may well characterize a multi-party, PR system as well.
14
Last, but not least, there is room for an ex ante behavioral effect by individual
councilors, which mirrors their proactive behavioral response to the anticipated political
consequences of the change in the municipal boundaries. This refers to the possibility
that the self-interest of the individual councilors drives their voting decision. For
example, a councilor may want to prevent the merger from taking place if it is expected
to have an adverse effect on his electoral vulnerability. We focus on quantifying this ex
ante behavioral effect of the councilors. This requires that we can determine how the
mechanical effect changes the electoral vulnerability of the municipal councilors and
how this (anticipated) change in the vulnerability then feeds back to the councilors’
voting behavior when they vote for the proposed merger.
5. Econometric approach
5.1 Econometric specification
To analyze how anticipated changes in the electoral vulnerability of individual
municipal councilors affect their voting behavior in the merger votes, net of ex ante and
ex post party-line considerations and the councilors’ electoral incentives at the
municipal or merger level, we consider the following econometric model
(1) 1 0( ) ,iikm km iki mp pv uμ δ= −+ +
where vikm equals one if councilor i representing party k from municipality m votes in
favor of the merger (and is zero otherwise); μkm denotes a set of fixed effects; (p1i − p0i)
refers to Total mechanical change and is thus our measure for the anticipated effect of
the merger on electoral vulnerability that is due to Council size and Competition
15
mechanisms; and uikm is an error term. For some specifications, we augment (1) to
include a vector of control variables, xikm.
We are interested in parameter δ, which captures the effect of Total mechanical
change in electoral vulnerability on councilors’ voting behavior. We focus on the
mechanical change for two reasons. First, it is possible to measure the anticipated
mechanical effects, whereas it is hard to capture all of the above listed behavioral
responses empirically. When measuring how the electoral vulnerability changes (i.e.,
when calculating Total mechanical change; see below) due to the amalgamation, we use
the vote distribution from the 2004 election, which does not mirror any ex post
reactions.12 This means that we can abstract away from the myriad ex post behavioral
reactions and adjustments by the different actors. Second, we focus on the mechanical
change also because it is harder for the incumbent politicians to predict how voters (or
other actors) react in the subsequent 2008 election than to anticipate the likely effects of
the mechanical change. This is important, because the councilors are unlike to react to
something that cannot be reasonably anticipated.
Indeed, anecdotal evidence suggests both that re-election concerns are a source of
concern among local politicians who contemplate mergers and that the mechanical
change corresponds to public perceptions about how a merger would affect re-election
prospects. The anecdotal evidence to which we refer here comes from i) what a number
of councilors have explicitly shared with us in confidential discussions and from ii)
what can be inferred from public discussion, both in national and local media. In
particular, at the time when mergers were contemplated some newspapers calculated
12 The party-lists and vote distribution of the pre-merger election does not contain information on ex post voter or party reactions, whereas those of the subsequent (post-merger) election do. The latter information was not available to the councilors when the mergers were voted upon.
16
what the post-merger councils would look like given the vote distribution from the pre-
merger elections. This is the type of exercise we do, except that we explicitly allow for
uncertainty in the election outcomes when measuring electoral vulnerability (see
below).13
In what follows, we first discuss in detail how we use fixed effects to control for
the heterogeneous preferences of voters and politicians and for how the different
political actors in the PR system influence the merger votes. We then explain how we
measure the councilor level electoral vulnerability.
5.2 Heterogenous policy preferences and party-line effects
We include the fixed effects, μkm, in various combinations and at various levels in order
to control for the possibility that voters and politicians have heterogeneous preferences
and, in particular, to account for how the different political actors in the PR system
influence the merger votes.
First, we use merger-level and party fixed effects. The former allow for a shared
view by the councils of the municipalities about the desirability of a merger.14 These
merger-level fixed effects account, for example, for the resemblance of policy
preferences in the involved municipalities. It is likely that such resemblance increases
the probability of a merger getting uniform support among the councilors in all the
councils who vote about a particular merger. The merger-level fixed effects also allow
for the possibility that the expected gains, such as the anticipated economies of scale in
the provision of local public goods, are so great (or so miniscule) for some of the
13 We document this more carefully in Appendix B. 14 The merger-level fixed effects refer to the local jurisdictions that would result from the municipal mergers if they go through.
17
mergers that the councils of the merging municipalities have a shared view about the
merger and vote therefore uniformly. We include party fixed effects to allow for a
national party-line vote. Most of the parties that are represented in the municipal
councils are also represented in the national politics. The party fixed effects allow for
the possibility that all council members of a political party vote in the same way,
irrespectively of where they live. A party may have a nation-wide policy against or for
municipal mergers, depending on whether the party’s leaders think that the mergers
benefit it politically or not.
Second, instead of allowing for merger and party fixed effects separately, we use
party-merger fixed effects. These fixed effects allow each party to have a merger-
specific effect on the councilors’ voting outcome. This means that merger-level party
politics can be a decisive factor for the local consolidation of municipalities. For
example, it is possible that the voting behavior of all councilors from party k depends on
the anticipated political strength of party k in the proposed merger.
Third, we can go deeper and allow for municipality fixed effects (in place of the
merger fixed effects). A municipal council may vote according to its true electoral
incentives, representing the interests of the pre-merger municipality, and thus the
preferences of the councilors’ current constituency. We allow for this by using the
municipality fixed effects, as they account for example for the fear that the merger will
result in closing down of local services (e.g. elementary schools) in the municipality
contemplating the merger.15 When we have both the municipality and (national) party
15 It is important to control for this fear, because it may be correlated with changes in electoral vulnerability. It is, if the councilors of the smaller municipalities are more likely to lose their political power as a result of the merger. See e.g. Knight (2008) on the link between regional representation in a legislative body and the geographic distribution of centralized spending.
18
fixed effects in the model, we estimate the councilor-level responses only using the
within municipality and within party-line variation in the data.
Fourth, instead of allowing for municipality and party fixed effects separately, we
can use party-municipality fixed effects. These disaggregated fixed effects allow
municipal-level party politics to be a decisive factor for the voting behavior of the
councilors. For example, it is possible that all councilors from party k in municipality m
think that their political strength decreases if the proposed merger takes place. If party-
line considerations at the municipal level determine how councilors vote, party-
municipality fixed effects explain (most of) the variation in the voting data.
5.3 Measuring electoral vulnerability
As Folke (2014) has stressed, measuring closeness of elections − and thus electoral
vulnerability − is not straightforward in PR election systems. We nevertheless need such
a measure, because in equation (1), (p1i − p0i) refers to Total mechanical change in
electoral vulnerability due to amalgamation. In alternative specifications, we also want
to decompose Total mechanical change into Council size and Competition mechanisms.
To be more precise, we define Total mechanical change in electoral vulnerability
for councilor i as (p1i − p0i), where p0i and p1i are proxies for the anticipated security of
the seat of the councilor in the next (2008) elections in the no-merger state (p0i) and the
merger state (p1i), respectively. We calculate these proxies as if all the voters voted or
abstained as they did in the 2004 (pre-merger) elections, but mimic uncertainty related
to the election outcomes using a bootstrap procedure, as explained below.
19
Bootstrap procedure
We construct p0i and p1i by using a bootstrap elections procedure introduced by
Kotakorpi et al. (2013), which allows us to create a set of counterfactual re-election
prospects.16 The procedure uses information on the identity of the candidates in the
2004 election, the vote distribution from the 2004 election and the number of available
seats in the different merger states based on the council size rule. Because we focus on
the councilors’ reaction to the anticipated mechanical change, we do not use
information about the post-merger 2008 elections in the bootstrap procedure. This
means, in particular, that the councilors’ ex post decisions to run for a council seat in the
subsequent 2008 election are not used.
To calculate Total mechanical change, we implement the bootstrap procedure
twice, once for two different scenarios. First, we consider a scenario in which all
mergers are assumed not to go through. For this scenario, we use the 2004
municipalities as the constituencies, irrespectively of whether the merger actually took
place or not. For the second scenario, we proceed as if all the mergers took place. The
hypothetical post-merger constituencies are constructed by allowing both the set of
candidates and voters as well as the number of available council seats to mirror the
properties of the post-merger entity.
For both scenarios, the bootstrap procedure is run in three steps:
Step 1: For each bootstrap round s, we sample votes with replacement for each
candidate from the vote distribution of the 2004 municipal election. The sampling
probability of a vote for a candidate is the share of the votes that he or she received in
16 The use of simulated elections to generate counterfactual outcomes is commonplace (e.g. Odendahl and Freier 2012, Chen and Rodden 2013). Kotakorpi et al. (2013) use the procedure to identify close winners and losers in a proportional election system for an RDD analysis. This corresponds to p0i in our analysis.
20
the 2004 election, but, crucially, the share is different in the scenario in which none of
the mergers is assumed to take place as compared to the scenario in which all of them
are assumed to take place. During each bootstrap round, we draw as many votes as were
given in the real 2004 elections to produce a distribution of votes over the candidates.
The resulting set of votes varies across the bootstrap repetitions within each of the two
scenarios due to randomness, but also differs systematically between the two scenarios,
because the sampling probabilities are different.
Step 2: We use the sampled votes to calculate a hypothetical election outcome
using the actual election rules for each bootstrap round. We do this differently for the
two scenarios: For the first scenario, we use the 2004 municipalities as the
constituencies to determine whether a candidate gets elected in a given bootstrap round.
We do these calculations for the second scenario by assuming that all the mergers took
place.
Step 3: We repeat the bootstrap elections of Step 1 and 2 many times (S = 10,000)
and count the share of times a particular candidate is elected both in the scenario in
which none of the mergers takes place and in the scenario in which all of them take
place. The former gives us p0i and the latter p1i. We then obtain a measure for Total
mechanical change in electoral vulnerability as (p1i − p0i).17
Total mechanical change in electoral vulnerability mirrors, as mentioned above,
both Council size mechanism (reduction in the relative number of seats) and
Competition mechanism (change in the set of candidates and voters). We can make use
of these sources of variation to further dissect the mechanical change into its sub-
components. To identify the latter of these two sub-components, we repeat the above
17 Further details of the procedure are explained in Appendix B.
21
bootstrap procedure as if all the mergers had taken place, but with the new council size
artificially set equal to the sum of the pre-merger council seats (instead of its actual
legal size).18 We call the resulting variable so that ( − ) gives us a measure for
Competition mechanism. This change in electoral vulnerability arises, because the set of
candidates and the distribution of votes over parties and candidates are different in the
event the merger takes place, as compared it not taking place, even if the election
outcomes are calculated in both cases using the vote distribution from the 2004
elections. Finally, subtracting Competition mechanism from Total mechanical change
gives the change in electoral vulnerability that is due to the change in the number of
council seats. This gives us Council size mechanism. It is equal to
( ) ( ) ( )1 1 1 0 1 0i i i i i ip p p p p p− = − − − . We provide further information of the bootstrap
procedure and a concrete example in Appendix B.
Descriptive statistics of changes in electoral vulnerability
The histograms for the electoral vulnerability variables are presented in the three panels
of Figure 1. Notice that this figure includes only the elected councilors that voted upon
the mergers, but not the candidates that were not elected in the 2004 elections.
The histogram of p0i is displayed in Panel A. It shows that the probability
distribution has a lot of mass on the right. This means that it is very likely that many of
the existing councilors would be re-elected in the 2008 election, if it was organized so
that the contemplated merger did not take place and voters behaved like in the 2004
election. The reason for why the probability of re-election is less than one for some of
18 This means that each candidate “competes” in these bootstrap elections in the new, merged constituency with all the candidates from the merging municipalities, but with the twist that the overall number of available seats is not reduced as the law would require.
22
the councilors is related to them being marginal (i.e., lucky) and subject to electoral
competition.
The histograms of and p1i are displayed in Panels B and C, respectively. They
show, in turn, that holding other things constant (but for the merger outcomes), the
contemplated mergers have a large mechanical effect on the election outcomes. The
mass on the left of these histograms means that it is likely that a number of the existing
councilors would not be re-elected in the 2008 election if it was organized so that the
contemplated merger took place and voters behaved like in the 2004 election. A closer
look at the data showed that these potential drop-outs are typically councilors from the
smaller municipalities that are contemplating a merger with a larger municipality.
[Figure 1 about here]
In Figure 2, we display the sub-components of Total mechanical change. First, as
the panel titled B–A shows, Competition mechanism decreases the electoral
vulnerability of some candidates, but hurts the most. The fact that electoral vulnerability
decreases for some is mainly due to the larger council size used in the bootstrap
elections for .19
Second, Council size mechanism is shown in the panel titled C–B. This
mechanism hurts most of the candidates, but, as expected, benefits no one. Third, Total
mechanical change is displayed in the panel titled C–A. It shows that the merger
increases the electoral vulnerability of most of the candidates, but actually benefits
19 The prior literature on redistricting suggests (e.g., Gelman and King 1994, Desposato and Petrocik 2003, Carson and Crespin 2004) that in a majoritarian election system, changes in the population of voters and/or in the mix of political competitors can either increase or decrease the electoral vulnerability of a given councilor.
23
some rare candidates. Those who appear to benefit were typically marginal in the 2004
election in a municipality that then subsequently contemplated a merger with a much
smaller municipality. We provide more details and intuition in Appendix B.
Overall, there is a lot of variation in our bootstrapped measures for electoral
vulnerability and in Total mechanical change over the candidates. The measures seem
to work as expected (e.g., the council size mechanism is non-positive for everybody)
and mirror what they were constructed for.
[Figure 2 about here]
Table 1 reports descriptive statistics for the electoral vulnerability measures
unconditionally and conditional on the councilors’ voting behavior and merger
outcomes. The upper part of the table shows that Council size and Competition
mechanisms as well as Total mechanical change are on average more negative in the
group of councilors who voted against the merger than they are in the group of
councilors who voted for the merger. This means that the councilors who voted in favor
of the mergers experience a smaller increase in their electoral vulnerability. This is
mostly due to a difference in p1i (and ) between the two groups. The lower part of the
table shows that there are similar differences if the numbers are conditioned on the
merger eventually taking or not taking place.
Table 1 also shows that the mean of p0i does not vary a lot between those
councilors who voted for the merger and who did not vote for it, or between those who
come from the merging municipalities and who come from the municipalities that did
not eventually merge. Since this measure of electoral vulnerability is based on the pre-
24
merger 2004 vote distribution and constituencies, it can be seen as a measure of the
level of political competition in the municipalities at the time the mergers were
contemplated. This suggests that a large part of the cross-sectional variation in the
mechanical change in electoral vulnerability is driven by the variation induced by the
mergers, and not by variation in the electoral vulnerability in the pre-merger
municipalities.
[Table 1 about here]
6. Empirical results
6.1 Main results
We start from Table 2, which presents the results for model (1). The reported results are
for a set of models without fixed effects (Panel A), with separate merger and (national)
party fixed effects (Panel B) and with party-merger fixed effects (Panel C). The
standard errors are clustered spatially, using the constituencies that would result from
the mergers (if they go through) as the clustering unit. In the first column of each panel,
there are no additional control variables. As we move to the right across the columns,
the models have progressively more controls, xikm.
The group of individual controls includes gender, age, age squared, and an
incumbency dummy (= 1 for those who were elected also in the 2000 elections; = 0
otherwise).20 The group of municipality controls include population, per capita mean
20 We also include a dummy for the rare cases where a vice-councilor voted in the merger-vote because the actual councilor was absent. We assume for the purposes of this paper in these cases that the vice-councilor is a perfect and obedient substitute for the councilor. We therefore use the vote of the vice councilor on the L.H.S. but the characteristics of the absent councilor on the R.H.S.
25
income, mean population distance from the municipal center, unemployment,
dependency ratio, per capita taxes, per capita grants and per capita expenditures.
Finally, in the rightmost specifications, we control directly for the vote shares of each
councilor at the municipal and the merger level. These vote shares refer to the
individual councilor’s vote share in the old municipality and to the corresponding
(hypothetical) share in the contemplated merger, as calculated using the 2004 vote data.
Holding the vote shares of a candidate constant means that the only source of variation
in the electoral vulnerability is how the votes are distributed over the other candidates.21
The estimated effect of Total mechanical change in electoral vulnerability on
councilors’ voting behavior, , is statistically significant and positive in all panels and
columns of Table 2. In particular, as the first columns of each panel show, the effect is
robust to adding the merger and party fixed effects separately (Panel B) and to adding
party-merger fixed effects (Panel C). These findings mean that the smaller the increase
in a councilor's electoral vulnerability, the more likely the councilor is to vote for
merger.
[Table 2 about here]
Table 3 repeats the regression analyses of Table 2, but uses separate municipality
and party fixed effects (Panel A) and party-municipality fixed effects (Panel B). The
point estimates are smaller, but still uniformly positive. They are statistically significant
21 Of course the vote shares are strongly correlated with our measures of electoral vulnerability. The main reason for including them is to control for voters’ preferences for a particular candidate. We include the vote shares in one of the specifications to err on the side of having an overly conservative set of controls, but acknowledge that doing so may in fact remove much useful identifying variation.
26
at 10 percent level in the first two columns of both panels, but are insignificant in the
third column, which include the vote shares as controls. These findings show that
neither the councils nor the parties vote uniformly. The decrease in the estimated effect
is however not surprising, because including the municipality and party fixed effects in
the model means that the effect of the mechanical change in electoral vulnerability is
identified from within-municipality and within-party variation only.22
[Table 3 about here]
Tables 4 and 5 present the results from models in which (p1i − p0i) is replaced by
terms that reflect Council size mechanism, ( − ), and Competition mechanism, ( − ). In Table 4, we present the results from the various models that are equivalent
to those of Table 2. Its three panels show that the effects of Council size and
Competition mechanisms are positive and statistically significant. The former is also
larger than the latter. Thus, it seems that temporarily allowing a larger council size in
the merged municipalities may be associated with the greater likelihood of the mergers
taking place. These findings are robust across the columns and, as Panels B and C show,
to the inclusion of separate merger and party fixed effects and to having party-merger
specific fixed effects in the model.
[Table 4 about here]
22 This means that a lot of potentially useful identifying variation is “closed down”. Viewed from this perspective, these results should be interpreted as the (conservative) lower bounds of the effect of interest. See Appendix A for more details on the variation in the data within the fixed effect groups.
27
In Table 5, we present the results from models that mirror those of Table 3. It
shows that when the municipality and party and party-municipality fixed effects are
added, the effect of Competition mechanism is positive and statistically significant.
However, in these models the effect of Council size mechanism is small and also
statistically insignificant.
[Table 5 about here]
To make sense of these results, we consider the different actors that are present in
a PR system. The overall picture that emerges from the results presented in Tables 2
through 5 is that adding more disaggregated party fixed effects into the models do not
change our baseline finding about the importance of councilors’ re-election concerns.
This is evident, for example, by comparing the results in Panels B and C in Tables 2 and
4 and also by comparing Panels A and B in Tables 3 and 5. This comparison also shows
two other things. First, parties do matter, because the explanatory power of the models
is greater, the more disaggregated party fixed effects we use. However, it seems that
their effect is largely orthogonal to that of individual electoral vulnerability. Second,
having municipality level fixed effects in the model makes for a bigger change. This
suggests that heterogeneous preferences across municipalities and local democracy
considerations are important drivers of merger decisions. Nonetheless, we find that
holding the electoral (municipality level) and party-line (party-municipality level)
incentives constant, a councilor is the more likely to vote for a merger the less his
electoral vulnerability increases in the merger state relative to the status quo of no
merger. However, a limitation of this analysis is that it does not reveal the concrete
28
reasons for why the councilors’ care about re-election prospects. It is likely that the
reasons include private gains of holding the office, such as pecuniary gains, prestige,
future labor market prospects and the ability to move policy closer to their own
preferred point in the post-merger council.
What about the discrepancy in the effects of Council size and Competition
mechanisms? Perhaps the most intuitive explanation for it is the use of fixed effects.
One might, for example, conjecture that the municipality fixed effects wash away most
of the useful variation in Council size mechanism. However, a closer look at the
variation in the data reveals that this is not the case: There is more unexplained variation
in Council size mechanism than in Competition mechanism after allowing for the most
disaggregated fixed effects (party-municipality).23
To provide further intuition on the drivers of these findings, we split our sample
according to whether the party members voted unanimously (or not) within the
municipalities. The idea is to compare the subsample of the data where it seems that
council or party discipline may have been binding to the subsample where due to split
voting we know that such discipline has not been the sole determinant of voting
behavior.24
The results for these two samples are presented in Table 6. In these models, we
use the party-merger fixed effects. It is evident from Panel A of Table 6 that Council
size mechanism has a positive and statistically significant effect in the sample where the
23 This can be seen from Table A2 in Appendix A, which illustrates how the amount of variation in the outcome and the different mechanisms of interest react to including the various levels of fixed effects. 24 The unanimous sample includes all councilors from the municipalities that voted unanimously and also the councilors from the parties that were unanimous even when the whole municipality was not. In most of these cases, the whole council was unanimous. Here we concentrate on the three dominant parties that have a large number of councilors so that it makes more sense to think about party discipline.
29
party members voted unanimously within the municipalities. In this sample,
Competition mechanism has no effect. The opposite is true for the sample where there
was split voting (Panel B). These results suggest that once we can rule out strict party
discipline, individual councilors seem to be concerned only about the anticipated change
in the composition of voters and political competitors (which is what Competition
mechanism measures). On the other hand, Council size mechanism seems to drive the
decision of whether the entire council or all the party members in a council vote for or
against the merger. These findings are consistent with the view that a policy of
temporarily allowing a larger council size in the merged municipalities may have
increased the likelihood of the mergers taking place.
[Table 6 about here]
6.2 Robustness tests
We have probed the robustness of our results in a number of ways. In the interest of
brevity, we report them in Appendix C. It suffices to note here that our baseline results
are robust to, e.g., using i) the subsample of data that includes only the councilors from
the three traditionally dominant (largest) parties; and ii) the subsample of data where
there is split voting within municipalities or party-municipalities and where we use the
corresponding fixed effects.
30
6.3 Policy significance
In-sample analysis
How do the estimated effects translate into changes in the likelihood of mergers? This
question cannot be answered solely on the basis of the point estimates, because for a
merger to go through, it needs a majority in each municipal council contemplating the
merger. To assess the policy significance of the estimates calls therefore for an
evaluation of how much the likelihood of a merger increases if the effect of mergers on
electoral vulnerability is neutralized.
To do this evaluation, we use the estimated models as follows: First, we simulate
merger outcomes by setting Total mechanical effect to zero (Counterfactual). We then
compare the rate of occurrence of mergers thus obtained to the simulated rate of
occurrence when the effect is set at its non-zero estimated value (Actual).25
We report the results of the simulations in Table 7. The reported numbers are the
rate of occurrence of mergers in the simulations for three different sets of
municipalities: First, the set of municipalities that underwent the merger; second, the set
of contemplated mergers that did not take place; and third, all the municipalities that
voted for a merger. The columns of the table report results from four different estimated
models: From a model with separate merger and party fixed effects (Column 1), with
25 The results we report in this section are based on the following simulation: First, we draw a random shock for a councilor from a uniform distribution (on the unit interval) and compare it to the fitted value generated by either the actual estimated model or the counterfactual model (with the effects set to zero). If the draw is smaller (larger) than the fitted value of the given councilor, he is assumed to vote for (against) the merger in the simulation. When we draw such a shock for all councilors, we can calculate whether a certain merger gains the required majority in all the participating municipalities or not. Second, we repeat this 1,000 times and take note of each merger occurrence.
31
party-merger fixed effects (Column 2), with separate municipality and party fixed
effects (Column 3), and with party-municipality fixed effects (Column 4).26
Table 7 illustrates two things. First, it shows that the estimated models predict a
much higher rate of occurrence of mergers in those municipalities that actually merged,
as compared to those who did not merge, which indicates a reasonable model fit.
Second, having the fixed effects in the model improves the model predictions,
especially in the group of municipalities that did not merge. Third and most importantly,
the table shows, depending on the level of fixed effects, that the increase in the rate of
occurrence of mergers increases at least by 2.0 (0.6894–0.6691) and at most by 7.7
(0.7806–0.7032) percentage points when we turn off the effect of electoral vulnerability.
These effects are in relative terms larger in the group of municipalities that did not
merge than in the group that merged. Since there are 59 contemplated mergers in our
data, these numbers mean that from one to four mergers did not take place due to the
desire of the councilors to personally avoid electoral competition. Because the average
number of municipalities in a merger is 3.4, this means, in turn, that there are from 3 to
13 municipalities in the group of municipalities that did not merge that would have
merged, had the councilors not wanted to avoid electoral competition.
[Table 7 about here]
26 The merger fixed effects results are based on model (7) of Table 3, the party-merger fixed effects results are based on model (11) of Table 3, the municipal fixed effects results are based on model (2) of Table 4 and the party-municipality fixed effects results are based on model (5) of Table 4.
32
Out-of-sample analysis and external validity
We are inclined to think that our results are representative of how political agency and
electoral concerns have affected the resetting of municipal boundaries in Finland. There
are three arguments for this view: First, our sample is rather comprehensive, as it
includes almost a third of Finnish municipalities. Second, our baseline models include
the municipal level fixed effects, which control for sample selection, at least to an
extent.27 Third, to check the representativeness of the simulation results reported in
Table 7, which apply to the group of municipalities that voted for a merger, we
considered the possibility that the anticipated increase in electoral vulnerability
prevented some municipalities from voting on a merger, and thus reduced the likelihood
of such mergers taking place. This analysis shows (see Appendix D) that the anticipated
increase in electoral vulnerability may have prevented some mergers from taking place
also outside our sample (i.e., among a stratified random sample of potential mergers that
were never voted on). Accounting for them would only strengthen our main
conclusions.
Whether our results generalize to other countries depends on the institutional
differences in electoral systems and in the way nations reset the boundaries of local
governments. For example, in Belgium, Canada, Denmark, Israel and Sweden mergers
were implemented, or at least strictly overseen, by the central government, whereas in
27 To see why, it is useful to recall that sample selection due to unobservables can be formulated as an omitted variable problem (Heckman 1979). This problem can be corrected for by introducing the inverse Mills’ ratio as an additional explanatory variable. Since the decision to vote for a merger is decided at the municipal level, selection into our sample on the basis of unobserved municipal level characteristics can be controlled for by using the municipality fixed effects. Note, moreover, that if sample selection is related to observable explanatory variables, it does not induce a bias in the standard OLS estimation. It could be, for example, that the municipalities that decided to vote for a merger are those where the adverse changes in electoral vulnerability are smaller than in the municipalities that did not decide to have a vote. This kind of selection is not a source of concern to us.
33
Finland, Germany, Japan and the Netherlands, the recent merger decisions have been
taken at the local level, by local politicians. Our findings ought to generalize to
institutionally similar settings, because many, if not most, redesigns of electoral
boundaries change both the composition of voters and the set of political competitors,
like they do in our case. Moreover, council size step functions are used in many
countries (e.g. in Brazil, France, Germany, Italy, Norway and Sweden). While the
relation of population to council size varies across countries, the differences in the
council size rules appear to be surprisingly small, even for countries of very different
sizes, such as Finland and Germany.28
7. Conclusions
The size and number of local governments is a key policy decision from the point of
view of the efficient provision of local public goods and services. We ask whether the
concerns of local politicians (municipal councilors) about their electoral vulnerability,
independently of voter preferences and party-line concerns, affect whether they vote for
or against a municipal merger and to what extent this is reflected in the resulting
municipal structure. We analyze this question by estimating the feedback effect of an
anticipated (mechanical) change in electoral vulnerability on the ex ante behavior of
politicians in a merger vote that determines whether the boundaries of their
constituencies change or not. The novel feature of our paper is that it considers political
agency and re-election concerns to be a driver of the merger decisions.
28 E.g., a Finnish municipality with population of 10000 has 35 seats in its council, whereas a German one has 24. The ratio of the difference in council sizes to the threshold is 0.0011. This means that for each 1000 inhabitants, there is one councilor more in Finland than in Germany. With population 150000, a Finnish municipality has 67 seats and a German one 50. This means that in larger cities, there is one councilor more in Finland than in Germany for each 10000 inhabitants.
34
Our key finding is that holding the electoral (municipality-level) and party-line
(party-municipality -level) incentives constant, a councilor is more likely to vote for a
merger if his electoral vulnerability does not increase in the merger state relative to the
status quo of no merger. Moreover, we find that the expected change in the composition
of voters and political competitors is important in explaining this behavioral response.
Our findings imply that incumbent politicians vote for policies that allow them to
escape political competition. Because councilors react proactively to the expected
changes in political competition, the design of the boundaries of the Finnish local
governments appears to be subject to strategic gerrymandering-type considerations (see
e.g. Gul and Pesendorfer 2010). This means that local politicians’ self-interest may lead
to sub-optimally small municipalities from the local citizens’ point of view. This makes
endogenous merging of (local) jurisdictions inefficient for a reason that the prior
literature has not so far explicitly considered: The foregone mergers are the price that
the society at large pays because the councilors care about the private gains that holding
a public office generate and because they want to avoid electoral competition.
35
Online Appendices (Intended for online publication only)
Appendix A: Data
Control variables
Table A1 presents the descriptive statistics of the control variables in our regression
models.
[Table A1 about here]
Variation in the data
A possible issue with our analysis it that it is unclear what the proper decision-making
level in a PR system is. It could e.g. be that councilors’ parties have all the agenda
setting power, and thus, the individual councilors simply vote according to the party
line. However, our data shows clearly that this is not case. Out of the 3,804 councilors
in our data, 2,134 councilors come from municipalities where there is no variation in
vote within the councilors’ home municipality. If these were the only data, it would be
impossible to analyze empirically the determinants of the voting decisions using
councilor-level data. However, of the remaining 1,670 councilors, for which the voting
varies within their home municipalities, there is councilor-level variation also within the
parties in the voting behavior (in 1,057 cases). Moreover, these deviations from the
party-municipality line are mostly not about only one councilor deviating.
In Table A2, we report the R2 measures from regression models where we regress
our voting outcome and our key explanatory variables on different sets of fixed effects
used in the models reported in the paper. The table shows, for example, that the party-
36
municipality fixed effects explain 63 percent of the variation in the voting outcome.
More disaggregated fixed effects explain systematically more of the variation in both
the voting outcome and the electoral vulnerability measures. The overall picture that
emerges is that we still have a considerable amount of variation left even with the most
disaggregated fixed effects.
[Table A2 about here]
Finally, in Figure A1, we report histograms of the shares of votes in favor of the
merger, separately for each of the various groups of political actors that also correspond
to the different, disaggregated levels of fixed effects that we use in the empirical
analysis. The figure shows that there is a non-negligible amount of within group
variation even within the party-municipality groups in the outcome variable: Not all
data is concentrated either at zero or at one. Bars between zero and one represent
deviations from the respective group line. For example, when there are deviations from
the party-municipality line, typically more than one councilor deviates (see bottom-right
histogram in Figure A1).
[Figure A1 about here]
Appendix B: Details on the bootstrap procedure
This Appendix provides additional details on and a concrete example of the bootstrap
procedure that we use to generate our empirical measures for the electoral vulnerability.
37
Basic idea
As we explain in the main text, our procedure builds on that of Kotakorpi et al. (2013),
from which the reader can find additional details. In our application, the procedure uses
information on the identity of the candidates in the 2004 election, the vote distribution
from the 2004 election and the number of available seats in the different merger states
based on the council size rule. An underlying assumption of the procedure is that the
distribution of voter preferences at the margin of abstaining and participating is identical
for those who turned out to those who did not, because allocating each new vote draw in
the bootstrap is based on the vote shares in the real elections.
The aim of the bootstrap procedure is to construct a smooth (continuous) measure
for electoral vulnerability in the pre-merger and post-merger elections that is capable of
mirroring the complexities of a multi-party PR system and that varies between the
councilors who cast a vote in the merger votes. In the Finnish open-list local election
system, each voter gives a single vote to a single candidate. This implies that unlike in
the closed-list elections, a vote distribution over individual candidates is available in the
Finnish system. It is therefore sensible to measure electoral vulnerability at the level of
candidates as opposed to the level of parties. The purpose of the bootstrap procedure is
to mimic uncertainty naturally present in the election outcomes of individual candidates:
Some marginal councilors could lose their seat due to a change in the vote distribution
and party-lists, whereas for the non-marginal councilors, such changes would have to be
much larger to have an impact.
38
Hypothetical example
To explain how the bootstrap procedure works in practice, we focus on a concrete but
hypothetical example. To this end, we consider candidates that come from two
imaginary municipalities, a small municipality A and a large municipality B, which
merge with each other.
Table B1 reports results from the pre-merger elections for these municipalities and
the electoral vulnerability calculations based on these election results. From left to right,
we report municipality IDs, party IDs, candidate IDs, the number of votes that each
candidate received in the pre-merger election, elections status (1= elected; 0 = not
elected), vote shares both in the old municipality and the merger, and our bootstrap
variables, 0ip , 1ip and 1ip . These are used to calculate the change in the electoral
vulnerability measures, which are reported in last three columns: Total mechanical
effect ( 1ip – 0ip ), Competition mechanism ( 1ip – 0ip ) and Council size mechanism ( 1ip –
1ip ).
As the table shows, municipality A is assumed to have 6 candidates and
municipality B 18. We assume that the council size in municipality A is 3 and in
municipality B it is 10. We also assume that the official council size in the post-merger
elections is 10. To be able to show that our measure also captures differences in political
preferences across municipalities, we assume there are 3 active parties in municipality
A but 4 active parties in municipality B.
[Table B1 about here]
39
To calculate Total mechanical change, we implement the bootstrap procedure
twice, once for two different scenarios. First, we consider a scenario in which the
merger is assumed not to go through. For this scenario, we use the pre-merger
municipalities as the constituencies. For the second scenario, we proceed as if the
merger took place. The hypothetical post-merger constituency is constructed by
allowing both the set of candidates and voters as well as the number of available council
seats to mirror the properties of the post-merger entity.
Consider now a single bootstrap round: In Step 1, we sample votes with
replacement for each candidate from the pre-merger vote distribution (column 4). The
sampling probability of a vote for a candidate is the share of the votes that he/she
received in his own municipality (sampling probabilities for 0ip ) or in the entire merger
(sampling probabilities for 1ip and 1ip ). Thus, for example, sampling probabilities are
4/52 and 4/498 for the candidate on the first row (Candidate 1 in Municipality A). We
record the outcome of each draw, which results in one of the candidates getting one
vote. We repeat this vote sampling 52 (for 0ip ) or 498 (for 1ip and 1ip ) times to produce
a set of votes for each candidate in the municipality A.
In Step 2, we use the sampled vote distribution to calculate a hypothetical
election outcome using the actual election rules. For the first scenario, we use the old
municipalities as the constituencies to determine whether a candidate gets elected (for
0ip ). For this calculation, we use council size of 3 for municipality A and 10 for
municipality B. We then repeat the calculations for the second scenario, i.e., as if the
merger took place. For this calculation, we use council size of 10 for 1ip and 13 for 1ip .
The only difference between p1i and is the different council size used in their
calculation.
40
In Step 3, we repeat the bootstrap elections of Steps 1 and 2 10,000 times and
count the share of times a particular candidate is elected the three scenarios. This gives
us p0i, p1i and 1ip .
We are now ready to make a number of observations about the bootstrap
procedure and the nature of the measures that it generates:
• Monotonicity: Within each party list, p0, p1 and are increasing in votes, as
they should be.
• Nature of electoral vulnerability: Looking at the column for p0i in Table B1
shows that all those councilors for whom p0i is clearly below one, are lucky and
subject to electoral competition. Their seats are vulnerable to start with, because
they are more likely to lose their seats to non-elected candidates if the pre-
merger elections were re-organized (and all the voters voted in the same way as
they actually did in the pre-merger elections) in the old municipality and if we
allow a degree of randomness in the voting outcomes. In particular, there are no
certain candidates (p0 exactly 1) in the smaller municipality A, but there are
some in the larger municipality B.
• Sources of electoral vulnerability: The two sources of electoral vulnerability are
between parties and within party competition. To consider the former first,
notice that party 3 in municipality A gains on average 1.22 (0.003 + 0.542 +
0.678) seats in the bootstrap procedure. This means that in 22% of our bootstrap
rounds they manage to steal a seat from either party 1 or 2, and about equally
often from each. Party 3 in municipality A provides an example of within party
competition: Candidates 5 and 6 compete against each other when party 3 gains
only one seat.
41
• Total mechanical effect ( 1ip – 0ip ): The total effect is typically negative, but can
also be positive for some candidates (as we explain in detail below). A
comparison of the column for p0i with that for p1i shows that should the merger
take place, the candidates from municipality A have to compete against the
politicians from municipality B. The values of p1i show that even the strongest
candidates from municipality A are subjected to intense competition since the
maximum value of p1i is 0.751. Therefore, all the candidates from municipality
A face uncertainty over their re-election if the merger goes through. However,
Total mechanical effect is nevertheless quite small for candidates 1 and 4, who
had little chance to get elected even in the old municipality and for candidate 3,
who due to his large vote share in the old municipality seems to be able to rise
high enough in the within-party ranking also in the merged municipality.
• Competition mechanism ( 1ip – 0ip ): As the second rightmost column of the table
shows, Competition mechanism is mostly negative for candidates from
municipality A but large and positive for some candidates in larger municipality
B. The candidates who gain are typically those who are marginal in the old, pre-
merger municipality, face little within party threat from the members of own
party in municipality A, and who benefit from the hypothetical increase in the
number of total seats available (from 10 to 13 when we compare p0i to ) that
this measure captures. For example, candidates 2 and 3 from municipality B are
marginal in the old election because they often compete with each other for the
second seat of their party. They face little threat from party 1 candidates from
municipality A. To see this, note that top party 1 candidate in municipality A has
11 votes whereas candidates 2 and 3 from municipality B have 20 and 22 votes.
42
As one of the new three seats would be allocated to party 1 (if the council size
was not reduced), Competition mechanism is positive for candidates 2 and 3
from municipality B. The same applies to candidate 13 from municipality B.
• Council size mechanism ( 1ip – 1ip ): As the last column of the table shows,
Council size mechanism is always non-positive, as it should. This column shows
how, besides having an adverse impact on a single marginal candidate, this
mechanism may disproportionately hit some parties that have multiple marginal
candidates on their lists. Moreover, we can see that Council size mechanism has
a smaller impact in the smaller municipality, but it is important for some
candidates in the larger municipality. The column also shows how the negative
Council size mechanism nearly offsets the positive Competition mechanism for
some candidates of municipality B (see e.g. candidates 2 and 3).
In Table B2, we show an example of a single real municipality to show that the
patterns of Table B1 are present also in the real data. In order not to explode the size of
the table, we show only the candidates from a single small municipality, which in our
data merges with one much larger partner.
[Table B2 about here]
Technical detail: Block size in Step 2
An important detail of step 2 in the bootstrap procedure is the size of each draw (i.e., the
number of votes sampled per draw). In the case of small municipalities, the size of each
draw is one vote. However, in larger municipalities, re-sampling will not introduce
variation in the election outcomes across the bootstrap repetitions if the draw size is
43
small relative to the total number of votes given in the election. The reason for this is
the following: When the number of total votes gets larger, the differences in the amount
of votes between the marginal candidates increases on average. In other words, the
probability that a single voter is pivotal decreases as the size of the municipality
increases. This means that the larger the municipality is, the less likely it is for us to find
candidates for whom the election outcome would vary over the bootstrap repetitions.
This would lead to a discrete and coarse measure of electoral vulnerability.
To avoid this, we follow Kotakorpi et al. (2013) and sample votes in blocks. This
re-introduces variation in the case of larger municipalities. For each municipality in
which more than 2000 votes are given, we take only 2000 draws in Step 2. If a
candidate is drawn to get a vote, we weight the vote so that the total number of votes in
each bootstrap election round matches with the amount of votes given in the actual
election. For example, in a municipality where 7000 votes were given, 3.5 votes are
allocated to the candidate in each draw, if he/she is drawn. The number 2000 is ad hoc,
but the distributions of the electoral vulnerability measures are fairly stable over a
reasonable range of the size of draws. We regard the use of block size as a technical
device that helps us to get a smooth (continuous) measure for electoral vulnerability.
However, it is not entirely without a theoretical backing. The approach could be
motivated, for example, by the group voting theory (e.g., Coate and Conlin 2004).
Anecdotal evidence on re-election concerns and municipal mergers
It is difficult to find examples of councilors publicly stating that they are concerned
about their personal re-election prospects after a merger. However, in public discussions
councilors often raise concerns over regional political representation after a merger. Of
44
course from these statements it is impossible to distinguish between self-interest and
actual concerns over regional representation. See for example:
• “Municipal politicians oppose a municipal merger due to its impact on local
democracy” (in Finnish: “Kuntapoliitikot vastustavat kuntaliitosta - perusteluna
lähidemokratia”), an article on the website of Finnish Broadcasting Company,
YLE (11.4.2014), discussing the reasons why local politicians oppose proposed
mergers.29
In addition, a number of local and national newspapers have calculated election
outcomes using pre-merger vote distributions and actual election rules when a particular
merger has been contemplated. These calculations correspond to our measure of p1, but
without the re-sampling procedure. See for example:
• “Municipal reform would increase the political power of the Centre Party” (in
Finnish: “Kuntauudistus nostaisi keskustan kannatusta keskuskaupungeissa”), an
article in Helsingin Sanomat newspaper (9.2.2012), discussing the relative
power of parties in different proposed mergers based on the vote distribution of
pre-merger elections.30
• “Social Democrats would stay in power in the Jyväskylä municipality” (in
Finnish: “Demareiden valta-asema säilyisi uudessa Jyväskylässä”), an article in
Keskisuomalainen newspaper (21.1.2014), discussing the composition of the
new municipality council (based on the vote distribution of pre-merger
elections) if a merger between 9 municipalities would take place.31
29 http://yle.fi/uutiset/kuntapoliitikot_vastustavat_kuntaliitosta_-_perusteluna_lahidemokratia/5672050 30 http://www.hs.fi/kotimaa/Kuntauudistus+nostaisi+keskustan+kannatusta+keskuskaupungeissa/a1305555315424 31 Not available online.
45
Appendix C: Robustness checks
In this Appendix, we report a series of sensitivity tests that we have implemented to
probe the robustness of our results. We start by repeating the analyses of Table 3 and 5
of the main text, using the subsample of data that includes only the councilors from the
three traditionally dominant (largest) parties. The models always include the party-
municipality fixed effects. These estimates are reported in Panel A and B of Table C1.
The results echo our baseline findings, but are somewhat stronger in terms of the size of
the effects and their statistical significance.32 There are several potential explanations
for this stronger ex ante behavioral effects. For example, the larger party organizations
may be better able to inform certain councilors on the potential adverse effects of the
mergers on the electoral vulnerability of their fellow party members. Moreover, the
politicians for whom political careers (and thus re-election) matter more may self-select
into the larger parties.
[Table C1 about here]
We also confirm in Table C2 that our results are robust to omitting all the
observations which have no variation in the outcome within the fixed effect groups that
are used in the estimation. This exercise is useful in reconfirming that we have enough
variation in the voting decisions for identification even after including the most
disaggregated fixed effects.
32 The same is true also for the no fixed effects and the merger fixed effects models both in the total effect and decomposition regressions in all specifications. We do not report these for the sake of brevity.
46
[Table C2 about here]
Appendix D: Effect of increased electoral vulnerability out-of-sample
The simulation results reported in Table 7 of the main text apply to the group of
municipalities that voted for a merger. This is a selected group of municipalities. It is
therefore of interest to consider the possibility that the anticipated increase in electoral
vulnerability prevented some municipalities from voting on a merger, and thus reduced
the likelihood of such mergers taking place.
To shed some light on this issue, we draw a stratified random sample from the
universe of all potential mergers. To limit the set potential mergers to a feasible size, we
do not allow them to cross county borders, they have to share a common border and
they cannot include more than six partners.33 The sample is stratified based on the
number of merger partners so that the drawn sample matches our data in this dimension.
Since we have councilor level election data for all municipalities, we can calculate the
Total mechanical change in electoral vulnerability due to these potential mergers also
for the councilors in this stratified random sample. We find that Total mechanical
change is on average -0.418 in the stratified random sample of potential mergers. This
effect is only slightly larger in absolute terms than in the sample that voted on the
mergers (-0.407, see Table 1 in the main text).
We can also repeat the merger simulations of Table 7 for the stratified random
sample of potential mergers. To this end, we generate out-of-sample predictions of
merger votes for individual councilors. A limitation of this procedure is, however, that
we do not have estimates for the merger or municipal fixed effects for the municipalities
33 See Saarimaa and Tukiainen (2014) for the details of creating all the potential mergers and of the sampling procedure.
47
that are in the stratified sample. Therefore, we follow a pragmatic approach and conduct
the merger simulation using the model (3) in Table 2, but in addition include a set of
merger level control variables (see Table A1 in Appendix A).34 To account for the lack
of fixed effects, i.e. to allow the voted mergers to be unobservably different from those
potential mergers that were never actually voted upon, we calibrate the estimated
models by introducing a shock to all potential mergers. The idea of these shocks is to
reduce the likelihood of the merger going through in the simulated merger votes so that
the vote outcomes would better reflect the lower empirical rate of occurrence among the
potential mergers that were never voted upon. We repeat the simulations over a range of
such shocks.
We find that the anticipated increases in the electoral vulnerability reduce the rate
of merger occurrence from 3 to 15 percentage points, as the calibrated shock varies
between [0, -0.5]. These numbers suggest that the anticipated increase in electoral
vulnerability may have prevented some mergers from taking place also in the stratified
random sample of potential mergers. The effect appears to be non-negligible for a
number of potential merger constellations, except for those that are the most unlikely to
succeed (i.e., when the calibrated shock is very negative).
In sum, anticipated increases in electoral vulnerability make municipal mergers
less likely in two ways. First, as we showed in the main text, conditional on a potential
merger being voted, it reduces the likelihood of the merger going through by 2–7
percentage points. Second, the anticipated increase in electoral vulnerability seems to
have prevented some mergers from taking place also among the potential mergers that
were never voted upon.
34 In this model, the effect of Total mechanical change on voting for a merger is 0.209 and it has a standard error of 0.057.
48
Acknowledgements
We thank Tim Besley, Micael Castanheira, Torun Dewan, Olle Folke, Ronny Freier,
Phil Haile, Kaisa Kotakorpi, Gilat Levy, Gerard Padro i Miquel, Torsten Persson, Tanja
Saxell, Frode Steen, Marko Terviö, Otto Toivanen, Jouko Verho, the editor Brian
Knight and anonymous referees for helpful comments. We also thank seminar
participants at EPCS 2011 at Rennes, EPSA 2011 at Dublin, HECER, IEB, Jyväskylä,
Oulu, STICERD at LSE and VATT for useful discussions. Oskari Harjunen and Niklas
Jahnsson provided excellent research assistance. Tuukka Saarimaa is grateful to the
Finnish Cultural Foundation for funding and SERC at the London School of Economics
and Political Science for hospitality. Janne Tukiainen is grateful to Emil Aaltonen
Foundation, Jenny and Antti Wihuri Foundation and Yrjö Jahnsson Foundation for
financial support and STICERD at the London School of Economics and Political
Science for hospitality.
References
Alesina, A., Spolaore, E., 1997. On the Number and Size of Nations. Quarterly Journal
of Economics 112, 1027–1056.
Alesina, A., Baqir, R. Hoxby, C., 2004. Political Jurisdictions in Heterogeneous
Communities. Journal of Political Economy 112, 348–396.
Ansolabehere S., Snyder J. M., Gerber A., 2002. Equal Votes, Equal Money: Court-
Ordered Redistricting and the Public Spending in the American States. American
Political Science Review 96, 767–777.
Benoit, K., 2001. District Magnitude, Electoral Formula, and the Number of Parties.
European Journal of Political Research 39, 203–224.
49
Besley, T., Coate, S., 1997. An Economic Model of Representative Democracy.
Quarterly Journal of Economics 112, 85–114.
Blais, A., Lachat, R., Hino, A., Doray-Demers, P., 2011. The Mechanical and
Psychological Effects of Electoral Systems: A Quasi-experimental Study.
Comparative Political Studies 44, 1599–1621.
Blom-Hansen, J., Houlberg, K., Serritzlew, S., 2014. Size, Democracy, and the
Economic Costs of Running the Political System. Forthcoming in American Journal
of Political Science.
Boatright, R. G., 2004. Static Ambition: Legislators’ Preparations for Redistricting.
State Politics and Policy Quarterly 4, 436–454.
Bolton, P., Roland, R., 1997. The Breakup of Nations: A Political Economy Analysis.
Quarterly Journal of Economics 112, 1057–1090.
Brasington, D., 2003. Snobbery, Racism, or Mutual Distaste: What Promotes and
Hinders Cooperation in Local Public-Good Provision? Review of Economics and
Statistics 85,874–883.
Brennan, G., Buchanan, J.M., 1980. The Power to Tax: Analytical Foundations of a
Fiscal Constitution. Cambridge: Cambridge University Press.
Carson, J. L., Crespin, M. H., 2004. The Effect of State Redistricting Methods on
Electoral Competition in United States House of Representatives Races. State
Politics and Policy Quarterly 4, 455–469.
Casella, A., 2001. The Role of Market Size in the Formation of Jurisdictions. Review of
Economic Studies 68, 83–108.
Chen, J., Rodden, J., 2013. Unintentional Gerrymandering: Political Geography and
Electoral Bias in Legislatures. Quarterly Journal of Political Science 8, 239–269.
50
Coate, S., Conlin, M., 2004. A Group Rule: Utilitarian Approach to Voter Turnout:
Theory and Evidence. American Economic Review 94, 1476– 1504.
Coate, S., Knight, B., 2007. Socially Optimal Districting: A Theoretical and Empirical
Exploration. Quarterly Journal of Economics 122, 1409–1471.
Cox, G. W., Rosenbluth, F. M., Thies, M. F., 1999. Electoral Reform and the Fate of
Factions: The Case of Japan's Liberal Democratic Party. British Journal of Political
Science 29, 33–56.
Cox, G. W., Rosenbluth, F. M., Thies, M. F., 2000. Electoral Rules, Career Ambitions,
and Party Structure: Comparing Factions in Japan's Upper and Lower Houses.
American Journal of Political Science 44, 115–122.
Dal Bo, E., Rossi, M. A., 2011. Term Length and the Effort of Politicians. Review of
Economic Studies 78, 1237–1263.
Desposato, S. W., Petrocik, J. R., 2003. The Variable Incumbency Advantage: New
Voters, Redistricting, and the Personal Vote. American Journal of Political Science
47, 18–22.
Duverger, M., 1954: Political Parties. London: Methuen.
Edwards, J., Keen, M., 1996. Tax Competition and Leviathan. European Economic
Review 40, 113-134.
Ellingsen, T., 1998. Externalities vs Internalities: A Model of Political Integration.
Journal of Public Economics 68, 251–268.
Ferraz C., Finan, F., 2011. Electoral Accountability and Corruption: Evidence from the
Audits of Local Governments. American Economic Review 101, 1274–1311.
Fiva, J., Folke, O., 2014. Mechanical and Psychological Effects of Electoral Reform.
Forthcoming British Journal of Political Science.
51
Folke, O., 2014. Shades of Brown and Green: Party effects in Proportional Election
Systems. Forthcoming Journal of the European Economic Association.
Freier, R., Odendahl C., 2012. Do Parties Matter? Estimating the Effect of Political
Power in Multi-party Systems. DIW Berlin Discussion Papers 1205.
Gelman A., King, G., 1990. Estimating the Electoral Consequences of Legislative
Redistricting. Journal of the American Statistical Association 85, 274–282.
Gelman A., King, G., 1994. A Unified Method of Evaluating Electoral Systems and
Redistricting Plans. American Journal of Political Science 38, 514–554.
Glazer, A., Robbins, M., 1985. Congressional Responsiveness to Constituency Change.
American Journal of Political Science 29, 259–273.
Gordon, N., Knight, B., 2009. A Spatial Merger Estimator with an Application to
School District Consolidation. Journal of Public Economics 93, 752–765.
Gul, F., Pesendorfer, W., 2010. Strategic Redistricting. American Economic Review
100, 1616–1641.
Hanes, N., M. Wickström, E. Wångmar, 2012. Municipal Preferences for State-imposed
Amalgamations: An Empirical Study Based on the Swedish Municipal Reform of
1952. Urban Studies 49, 2733–2750.
Heckman, J., 1979. Sample Selection Bias as a Specification Error. Econometrica 47,
153–61.
Hinnerich, B. T., 2009. Do Merging Local Governments Free Ride on Their
Counterparts when Facing Boundary Reform? Journal of Public Economics 93, 721–
728.
Howitt, P., Wintrobe, R., 1995. The Political Economy of Inaction. Journal of Public
Economics 56, 329–353.
52
Jordahl, H., Liang, C.-Y., 2010. Merged Municipalities, Higher Debt: On Free-riding
and the Common Pool Problem in Politics. Public Choice 143, 157–172.
Knight, B., 2008. Legislative Representation, Bargaining Power and the Distribution of
Federal Funds: Evidence from the US Congress. Economic Journal 118, 1785–1803.
Kotakorpi, K., Poutvaara, P., Terviö, M., 2013. Returns to Office in National and Local
Politics. CESifo Working Paper No. 4542.
Lassen, D. D., Serritzlew. S., 2011. Jurisdiction Size and Local Democracy: Evidence
on Internal Efficacy from Large-scale Municipal Reform. American Political Science
Review 105, 238–258.
Leveaux-Sharpe, C., 2001. Congressional Responsiveness to Redistricting Induced
Constituency Change: An Extension to the 1990s. Legislative Studies Quarterly 26,
275–286.
Lijphart, A., 1991. The Political Consequences of Electoral Laws, 1945-85. American
Political Science Review 84, 481–496.
Miceli, T. J., 1993. The Decision to Regionalize in the Provision of Education: An
Application of the Tiebout Model. Journal of Urban Economics 33, 344–360.
Osborne, M. J., Slivinski, A., 1996. A Model of Political Competition with Citizen-
Candidates. Quarterly Journal of Economics 111, 65–96.
Perroni, C., Scharf, K., 2001. Tiebout with Politics: Capital Tax Competition and
Constitutional Choices. Review of Economic Studies 68, 133–154.
Reingewertz, Y., 2012. Do Municipal Amalgamations Work? Evidence from
Municipalities in Israel. Journal of Urban Economics 72, 240–251.
Saarimaa, T., Tukiainen, J., 2013a. Common Pool Problems in Voluntary Municipal
Mergers. VATT Working Papers 53.
53
Saarimaa, T., Tukiainen, J., 2013b. Local Representation and Strategic Voting:
Evidence from Electoral Boundary Reforms. IEB Working Papers 2013/032.
Saarimaa, T., Tukiainen, J., 2014. I Don’t Care to Belong to Any Club That Will Have
Me as a Member: Empirical Analysis of Municipal Mergers. Political Science
Research and Methods. 2, 97–117.
54
Figure1. Histograms of , and p1i.
Notes: Panel A reports p0i, which refers to the share of times a particular candidate is elected in the bootstrap scenario in which none of the mergers takes place. Panel B reports , which refers to the share of times a particular candidate is elected in the bootstrap scenario in which all of the mergers takes place, but assuming that the council size was not reduced as the law would require. Panel C reports p1i, which refers to the share of times a particular candidate is elected in the bootstrap scenario in which all of the mergers takes place, but assuming that the council size is reduced as the law required.
05
1015
Den
sity
0 .2 .4 .6 .8 1Bootstrap election (merger = 0)
Panel A
05
1015
Den
sity
0 .2 .4 .6 .8 1Bootstrap election (merger = 1, old council size)
Panel B
05
1015
Den
sity
0 .2 .4 .6 .8 1Bootstrap election (merger = 1, new council size)
Panel C
55
Figure 2. Histograms of the mechanical changes.
Notes: Panel titled B-A reports Competition mechanism, –p0i. Panel titled C-B reports Council size mechanism, p1i– . Panel titled C-A reports Total mechanical change, p1i – .
01
23
45
Den
sity
-1 -.5 0 .5 1Competition mechanism
B - A
01
23
45
Den
sity
-.8 -.6 -.4 -.2 0Council size mechanism
C - B
01
23
45
Den
sity
-1 -.5 0 .5Total mechanical change
C - A
56
Table 1. Descriptive statistics of the measures of electoral vulnerability.
Notes: p0i refers to the share of times a particular candidate is elected in the bootstrap scenario in which none of the mergers takes place. refers to the share of times a particular candidate is elected in the bootstrap scenario in which all of the mergers takes place, but assuming that the council size was not reduced as the law would require. p1i refers to the share of times a particular candidate is elected in the bootstrap scenario in which all of the mergers takes place, but assuming that the council size is reduced as the law required.
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
Number of councilors
p 0 0.841 0.185 0.852 0.179 0.838 0.186
0.670 0.303 0.582 0.289 0.690 0.303
p 1 0.433 0.358 0.309 0.309 0.462 0.362
Competition mechanism -0.171 0.303 -0.270 0.268 -0.148 0.307
Council size mechanism -0.236 0.160 -0.272 0.150 -0.228 0.161
Total mechanical change -0.407 0.342 -0.543 0.289 -0.376 0.346
Mean Std. Dev. Mean Std. Dev.
Number of councilors
p 0 0.853 0.178 0.836 0.187
0.631 0.304 0.685 0.302p 1 0.357 0.341 0.464 0.360
Competition mechanism -0.222 0.301 -0.151 0.302
Council size mechanism -0.274 0.154 -0.221 0.160
Total mechanical change -0.496 0.330 -0.373 0.340
All
3,804
Vote = 0 Vote = 1
720 3,084
1,077
Merger = 0 Merger = 1
2,727
1p
1p
57
Table 2. Total mechanical change (no, merger and party-merger fixed effects).
Notes: The results are from linear probability models where the dependent variable is whether the councilor voted in favor of the merger. Sample size in each regression is 3,804. Standard errors are robust to clustering at the merger level and are reported in parentheses. ***, ** and * indicate statistical significance at 1, 5 and 10 percent level, respectively.
(1) (2) (3) (4)
Total mechanical change 0.219*** 0.220*** 0.153** 0.152***
(0.057) (0.057) (0.058) (0.055)
R2
0.04 0.04 0.09 0.09
(5) (6) (7) (8)
Total mechanical change 0.165*** 0.165*** 0.146*** 0.143***
(0.049) (0.049) (0.053) (0.051)
R2
0.26 0.26 0.33 0.34
(9) (10) (11) (12)
Total mechanical change 0.173*** 0.173*** 0.158*** 0.164***
(0.049) (0.050) (0.051) (0.052)R
20.39 0.39 0.45 0.45
Individual controls No Yes Yes Yes
Municipality controls No No Yes Yes
Vote shares No No No Yes
Panel A: No fixed effects
Panel B: Separate merger and party fixed effects
Panel C: Party-merger fixed effects
58
Table 3. Total mechanical change (municipality and party-municipality fixed effects).
Notes: The results are from linear probability models where the dependent variable is whether the councilor voted in favor of the merger. Sample size in each regression is 3,804. Standard errors are robust to clustering at the merger level and are reported in parentheses. ***, ** and * indicate statistical significance at 1, 5 and 10 percent level, respectively.
(1) (2) (3)
Total mechanical change 0.074* 0.072* 0.062
(0.040) (0.040) (0.046)
R2
0.49 0.49 0.49
(4) (5) (6)
Total mechanical change 0.072* 0.071* 0.070
(0.041) (0.041) (0.045)
R2
0.66 0.66 0.64
Individual controls No Yes Yes
Vote shares No No Yes
Panel B: Party-municipality fixed effects
Panel A: Separate municipality and party fixed effects
59
Table 4. Decomposition results (no, merger and party-merger fixed effects).
Notes: The results are from linear probability models where the dependent variable is whether the councilor voted in favor of the merger. Sample size in each regression is 3,804. Standard errors are robust to clustering at the merger level and are reported in parentheses. ***, ** and * indicate statistical significance at 1, 5 and 10 percent level, respectively.
(1) (2) (3) (4)
Council size mechanism 0.270** 0.275** 0.270*** 0.264**
(0.104) (0.105) (0.100) (0.118)
Competition mechanism 0.205*** 0.205*** 0.111* 0.132**
(0.062) (0.061) (0.065) (0.058)
R2
0.04 0.04 0.09 0.09
(5) (6) (7) (8)
Council size mechanism 0.251*** 0.257*** 0.198*** 0.196**
(0.068) (0.070) (0.058) (0.076)
Competition mechanism 0.149*** 0.149*** 0.122* 0.128**
(0.054) (0.053) (0.067) (0.059)
R2
0.26 0.26 0.33 0.34
(9) (10) (11) (12)
Council size mechanism 0.279*** 0.283*** 0.223*** 0.258***
(0.076) (0.078) (0.062) (0.085)
Competition mechanism 0.153*** 0.152*** 0.126* 0.141**
(0.054) (0.054) (0.066) (0.059)
R2
0.39 0.39 0.45 0.45
Individual controls No Yes Yes Yes
Municipality controls No No Yes Yes
Vote shares No No No Yes
Panel A: No fixed effects
Panel B: Separate merger and party fixed effects
Panel C: Party-merger fixed effects
60
Table 5. Decomposition results (municipality and party-municipality fixed effects).
Notes: The results are from linear probability models where the dependent variable is whether the councilor voted in favor of the merger. Sample size in each regression is 3804. Standard errors are robust to clustering at the merger level and are reported in parentheses. ***, ** and * indicate statistical significance at 1, 5 and 10 percent level, respectively.
(1) (2) (3)
Council size mechanism 0.007 0.009 -0.052
(0.044) (0.044) (0.055)
Competition mechanism 0.134*** 0.128*** 0.148***
(0.046) (0.046) (0.054)
R2
0.49 0.49 0.49
(5) (6) (7)
Council size mechanism 0.034 0.036 0.013
(0.039) (0.038) (0.050)
Competition mechanism 0.109*** 0.106** 0.116**
(0.047) (0.047) (0.051)
R2
0.66 0.66 0.66
Individual controls No Yes Yes
Vote shares No No Yes
Panel A: Separate municipality and party fixed effects
Panel B: Party-municipality fixed effects
61
Table 6. Results for unanimous and split voting samples (party-merger fixed effects).
Notes: The results are from linear probability models where the dependent variable is whether the councilor voted in favor of the merger. The samples include councilors only from the three largest parties. Standard errors are robust to clustering at the merger level and are reported in parentheses. ***, ** and * indicate statistical significance at 1, 5 and 10 percent level, respectively.
(1) (2) (3) (4)
Council size mechanism 0.196** 0.197** 0.166*** 0.219***
(0.083) (0.085) (0.060) (0.072)
Competition mechanism 0.028 0.027 0.033 0.052
(0.049) (0.049) (0.074) (0.069)
R2
0.63 0.63 0.70 0.70
N 2,141 2,141 2,141 2,141
(5) (6) (7) (8)
Council size mechanism 0.222 0.235 0.130 0.033
(0.174) (0.170) (0.162) (0.193)
Competition mechanism 0.143* 0.138* 0.369*** 0.388***
(0.077) (0.075) (0.105) (0.114)
R2
0.17 0.18 0.20 0.20
N 967 967 967 967
Individual controls No Yes Yes Yes
Municipality controls No No Yes Yes
Vote shares No No No Yes
Panel A: Unanimous
Panel B: Split voting
62
Table 7. Counterfactual simulation results.
Notes: The table presents results from a merger vote simulation exercise. The numbers correspond to shares of realized mergers for each model specification and merger sub-sample. The simulations are based on 1,000 repetitions.
Model specification: Merger FE
Party-merger FE
Municipality FE
Party-municipality FE
(1) (2) (3) (4)
Merger = 1:
Actual 0.9449 0.9511 0.9369 0.9528
Counterfactual 0.9765 0.9856 0.9661 0.9755
Merger = 0:
Actual 0.1955 0.1682 0.0637 0.0569
Counterfactual 0.3565 0.3382 0.0869 0.0722
All:
Actual 0.7076 0.7032 0.6604 0.6691
Counterfactual 0.7802 0.7806 0.6877 0.6894
63
Table A1. Descriptive statistics.
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
Number of councilors
Councilor characteristics:
Age 48.3 11.0 48.0 11.2 48.4 10.9
Female 0.365 0.482 0.358 0.480 0.368 0.482
Two or more terms in council 0.600 0.490 0.617 0.486 0.594 0.491
Vote share in municipality 0.024 0.016 0.025 0.016 0.023 0.016
Vote share in merger 0.008 0.006 0.006 0.005 0.008 0.007
Municipal characteristics:
Population 9,656 15,241 7,896 10,333 10,686 17,450
Taxable income (€ per capita) 10,511 1,761 9,691 1,476 10,991 1,742
Mean population distance to centre (km) 5.49 10.32 7.26 16.71 4.45 1.99
Unemployment rate (%) 10.9 3.87 12.5 3.74 9.9 3.65
Dependency ratio 1.51 0.27 1.64 0.26 1.44 0.25
Municipal income tax rate (%) 19.0 0.68 19.2 0.6 18.8 0.72
Central government grants (€ per capita) 1,563 574 1,827 529 1,408 544
Total expenditures (€ per capita) 5,000 737 5,266 825 4,845 635
Merger characteristics:
Population 28,323 27,271 27,014 21,815 28,930 29,693
Taxable income (€ per capita) 11,171 1,583 10,328 1,288 11,562 1,567
Mean population distance to centre (km) 8.91 4.71 12.68 6.16 7.16 2.39
Unemployment rate (%) 11.2 3.72 12.8 3.47 10.4 3.63
Dependency ratio 1.52 0.23 1.64 0.19 1.47 0.23
Municipal income tax percent (%) 19.0 0.51 19.2 0.4 18.9 0.54
Central government grants (€ per capita) 1,431 501 1,688 440 1,312 487
Total expenditures (€ per capita) 5,043 583 5,326 722 4,913 460
Cooperation 0.5 0.5 0.4 0.5 0.6 0.5
Merger size 2.9 1.5 3.4 1.6 2.7 1.5
All Merger = 0 Merger = 1
3,804 1,077 2,727
64
Table A2. Variation in key variables in different fixed effect groups.
Note: The table reports R-squared measures from regression models where the variables are regressed on different fixed effect level dummy variables.
Variable Merger Party-merger Municipality Party-municipality
Vote decision (0/1) 0.24 0.37 0.48 0.66
Total mechanical change 0.22 0.31 0.81 0.84
Council size mechanism 0.35 0.41 0.43 0.52
Competition mechanism 0.11 0.24 0.83 0.86
65
Figure A1. Share of councilors voting in favor of the merger.
Notes: In each histogram, the unit of observation is a merger, party-merger, municipality or party-municipality -group. The magnitude of interest is the share of councilors in each group that voted in favor of the merger.
05
1015
Den
sity
0 .2 .4 .6 .8 1Yes vote share within merger
05
1015
Den
sity
0 .2 .4 .6 .8 1Yes vote share within party-merger
05
1015
Den
sity
0 .2 .4 .6 .8 1Yes vote share within municipality
05
1015
Den
sity
0 .2 .4 .6 .8 1Yes vote share within party-municipality
66
Mun
icip
ality
Part
yC
andi
date
N
umbe
r of
vot
esEl
ectio
n st
atus
Vot
e sh
are
in o
ld
mun
icip
ality
Vot
e sh
are
in
new
mun
icip
ality
p0
p1
Tot
al m
echa
nica
l ef
fect
Com
petit
ion
mec
hani
smC
ounc
il si
ze
mec
hani
sm
A1
14
00.
077
0.00
80.
035
0.00
00.
001
-0.0
35-0
.034
-0.0
01
A1
211
10.
212
0.02
20.
838
0.01
90.
107
-0.8
20-0
.731
-0.0
89
A2
316
10.
308
0.03
20.
904
0.75
10.
949
-0.1
520.
046
-0.1
98
A3
42
00.
038
0.00
40.
003
0.00
00.
000
-0.0
03-0
.003
0.00
0
A3
59
00.
173
0.01
80.
542
0.02
20.
060
-0.5
20-0
.483
-0.0
38
A3
610
10.
192
0.02
00.
678
0.03
70.
111
-0.6
41-0
.567
-0.0
74
B1
13
00.
007
0.00
60.
000
0.00
00.
000
0.00
00.
000
0.00
0
B1
220
00.
045
0.04
00.
449
0.49
20.
930
0.04
40.
482
-0.4
38
B1
322
10.
049
0.04
40.
658
0.68
70.
964
0.02
90.
306
-0.2
77
B1
456
10.
126
0.11
21
11
00
0
B2
56
00.
013
0.01
20.
065
0.01
00.
074
-0.0
550.
009
-0.0
64
B2
69
00.
020
0.01
80.
326
0.08
90.
360
-0.2
370.
034
-0.2
71
B2
711
10.
025
0.02
20.
620
0.22
80.
658
-0.3
920.
038
-0.4
30
B2
814
10.
031
0.02
80.
876
0.57
30.
881
-0.3
040.
005
-0.3
09
B2
936
10.
081
0.07
21
11
00
0
B2
1040
10.
090
0.08
01
11
00
0
B2
1180
10.
179
0.16
11
11
00
0
B3
1212
00.
027
0.02
40.
006
0.09
90.
271
0.09
30.
265
-0.1
72
B3
1314
00.
031
0.02
80.
017
0.19
60.
528
0.17
90.
511
-0.3
32
B3
1433
10.
074
0.06
60.
997
0.99
71.
000
0.00
10.
003
-0.0
02
B4
1510
00.
022
0.02
00.
002
0.00
10.
008
-0.0
010.
006
-0.0
07
B4
1617
00.
038
0.03
40.
089
0.06
40.
183
-0.0
260.
094
-0.1
20
B4
1726
10.
058
0.05
20.
900
0.75
70.
916
-0.1
430.
016
-0.1
59
B4
1837
10.
083
0.07
40.
995
0.97
90.
998
-0.0
160.
003
-0.0
19
Tab
le B
1. B
oots
trap
ele
ctio
ns e
xam
ple
with
two
fictio
nal m
unic
ipal
ities
.
67
Table B2. Illustration of p0, p1 and in an actual municipality.
Party IDCandidate
IDNumber of
votesElection status
p 0 p 1
1 7 14 1 0.7476 0.0001 0.0037
1 8 15 1 0.7986 0.0004 0.0039
1 10 28 1 0.9983 0.0106 0.0652
1 11 29 1 0.9988 0.014 0.0759
1 14 29 1 0.9989 0.0131 0.0777
1 15 32 1 0.9997 0.0221 0.1096
1 12 32 1 0.9999 0.0211 0.1103
1 9 47 1 1 0.1298 0.3667
1 13 81 1 1 0.6676 0.8715
2 18 7 0 0.0162 0.0001 0.0002
2 17 30 1 0.957 0.0318 0.1259
2 16 45 1 0.9972 0.1806 0.4002
3 23 9 0 0.0389 0.0001 0.0013
3 19 16 0 0.5762 0.001 0.0146
3 22 18 1 0.7634 0.0019 0.0199
3 21 19 1 0.8333 0.0026 0.0302
3 20 66 1 1 0.5376 0.7964
4 2 67 1 1 0.4858 0.7437
8 6 3 0 0 0 0
8 5 6 0 0 0 0
8 4 11 0 0.011 0 0
8 3 24 1 0.9364 0 0.0001
19 27 3 0 0.0001 0 0
19 28 4 0 0.0002 0 0
19 25 7 0 0.0141 0 0.0006
19 26 8 0 0.026 0 0.0007
19 29 14 0 0.2977 0.0015 0.0081
19 24 29 1 0.9905 0.0464 0.1218
1p
68
Table C1. Results for large parties (party-municipality fixed effects).
Notes: The results are from linear probability models where the dependent variable is whether the councilor voted in favor of the merger. Standard errors are robust to clustering at the merger level and are reported in parentheses. ***, ** and * indicate statistical significance at 1, 5 and 10 percent level, respectively.
(1) (2) (3)
Total mechanical change 0.098** 0.099** 0.092*
(0.044) (0.044) (0.049)
R2
0.63 0.63 0.63
N 3,108 3,108 3,108
(4) (5) (6)
Council size mechanism 0.049 0.054 0.024
(0.045) (0.045) (0.059)
Competition mechanism 0.142*** 0.138*** 0.143***
(0.046) (0.045) (0.049)
R2
0.50 0.50 0.50
N 3,108 3,108 3,108
Municipality-party fixed effects Yes Yes Yes
Individual controls No Yes Yes
Vote shares No No Yes
Panel A: Total mechanical effect for large parties
Panel B: Decomposition for large parties
69
Table C2. Results for split-vote samples.
Notes: The results are from linear probability models where the dependent variable is whether the councilor voted in favor of the merger. The models are estimated using subsamples where there is split voting within corresponding fixed effect groups. Standard errors are robust to clustering at the merger level and are reported in parentheses. ***, ** and * indicate statistical significance at 1, 5 and 10 percent level, respectively.
(1) (2) (3)
Council size mechanism -0.032 -0.027 -0.179
(0.102) (0.099) (0.118)
Competition mechanism 0.260*** 0.241*** 0.294***
(0.085) (0.084) (0.099)
R2
0.24 0.24 0.24
N 1,670 1,670 1,670
(5) (6) (7)
Council size mechanism 0.034 0.039 -0.053
(0.164) (0.160) (0.196)
Competition mechanism 0.296** 0.282** 0.350**
(0.124) (0.122) (0.140)
R2
0.23 0.23 0.23
N 1,057 1,057 1,057
Individual controls No Yes Yes
Vote shares No No Yes
Panel A: Separate municipality and party fixed effects
Panel B: Party-municipality fixed effects