ESSAYS IN LOCAL PUBLIC FINANCE

By

Walter Thomas Melnik

A DISSERTATION

Michigan State University

in partial fulﬁllment of the requirements

Submitted to

for the degree of

Economics – Doctor of Philosophy

2018

ABSTRACT

ESSAYS IN LOCAL PUBLIC FINANCE

By

Walter Thomas Melnik

My dissertation studies how political factors, local labor demand shocks, and voting behavior aﬀect
state and local public goods provision. Following a brief introduction, I begin with “Legislative
Redistricting, Party Politics, and the Spatial Distribution of Transportation Expenditure.” In this
essay, I estimate how a state representative’s political party aﬀects road construction expenditure
in areas that she represents. An extensive literature asks how a legislator’s party aﬃliation af-
fects public expenditure in the area the legislator represents. Unfortunately, almost all studies
estimate this eﬀect using party changes through election outcomes, which could be correlated with
unobservable determinants of transportation expenditure. To overcome this issue, I identify my
estimates using changes in party aﬃliation engendered by the 2012 state legislative redistricting
in Ohio. In many cases, redistricting moved a geographic area into a district whose incumbent
representative belonged to the opposing political party. This created variation in partisan alignment
unrelated to election outcomes. From 2010-2017 the Republican party controlled the Ohio House
of Representatives, the Ohio Senate, and the governorship. Using variation due to redistricting for
identiﬁcation, I ﬁnd that areas moving from Republican to Democratic districts due to redistricting
received $3.5 million (0.19 standard deviations) less annual highway construction funding than
areas that remained in Republican districts. This funding decrease derives from a decline in the
number of large construction projects in these areas. The estimated eﬀects diﬀer substantially
when identiﬁed using variation through voting in non-redistricting years, perhaps due to selection
issues concerning the type of districts changing parties through election outcomes. In addition, the
expenditure change associated with a party change through election outcomes depends on whether
the incumbent lost an election or retired, further evincing selection issues associated with this
variation.

In my next essay, “Municipal Government Reaction to Mass Layoﬀs in Ohio,” I study how
municipal government ﬁnances respond to negative local employment shocks. Using data from
595 municipalities in Ohio, I estimate the change in municipal revenue after reported mass layoﬀs
and plant closings, as well as the municipality’s response: possible adjustments to tax rates,
expenditure, and borrowing.
I ﬁnd that income tax revenue plummets in the year after a mass
layoﬀ, driven by a large decline in income tax base. Municipalities do not raise income or property
tax rates to compensate for the income tax drop - rather, tax rates decline slightly. Property tax
revenue also declines, while revenue from service charges and fees and intergovernmental revenue
do not change signiﬁcantly. Thus, total revenue drops substantially for several years after a mass
layoﬀ.
In response, municipalities cut expenditure across several categories, including general
government, public safety, leisure and community environment, and capital outlay. Cities also
draw down their unreserved fund balance substantially, avoiding deeper cuts to expenditure by
depleting their accumulated funds.

In my last essay, “Ballot Order and Ballot Roll-oﬀ: Evidence from Ohio,” coauthored with
Mike Conlin and Paul Thompson, I study how an election item’s position on the ballot aﬀects the
probability that voters abstain from voting on that item (“roll-oﬀ”), and on the probability that
voters choose to vote yes conditional on casting a ballot. Local tax referenda in Ohio rotate ballot
position every year based on the level of local jurisdiction that placed the referendum on the ballot,
providing a source of exogenous variation to test these propositions. Previous research suggests
that voters are less likely to cast a vote for election items lower on the ballot, and more likely to
choose the status quo. These ﬁndings support the idea of choice fatigue, suggesting that facing more
decisions impairs voters’ decision making ability. Unlike previous papers, I am able to control for
demographic characteristics (age and party aﬃliation) of voters who see each referendum. I ﬁnd
that voters tend to cast more yes votes for items lower on the ballot. I also ﬁnd that older voters are
much less likely, and partisan voters much more likely, to abstain from ballot items, showing the
importance of controlling for these characteristics when estimating the eﬀect of ballot position on
roll-oﬀ.

ACKNOWLEDGEMENTS

I thank my committee - Leslie Papke, Ron Fisher, Cory Smidt, and especially my chair Mike Conlin
- for their counsel and support.
I thank the Lincoln Institute for supporting my research through a C. Lowell Harriss Dissertation
Fellowship.
I also thank my parents for their investment of time and eﬀort in my education.

iv

TABLE OF CONTENTS

LIST OF TABLES .
LIST OF FIGURES .
CHAPTER 1

.
.
.
.
INTRODUCTION .

.
.

.
.

.
.

.
.

.
.

.
.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

CHAPTER 2 LEGISLATIVE REDISTRICTING, PARTY POLITICS, AND THE SPA-

.
.
.

.
.
.

.
.
.

. .

Introduction .

.
2.1
.
.
2.2 Related Literature .
2.3 Road Project Data .
.
2.4 Districts and Redistricting .
2.5 Empirical Speciﬁcations

3
TIAL DISTRIBUTION OF TRANSPORTATION EXPENDITURE . . . . .
3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5.1 Trends in Outcome Variables . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5.2 Regression Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5.3 Controls . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Percentage changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5.4
2.5.5 Accounting for the Senate
. . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5.6 Number of Construction Projects . . . . . . . . . . . . . . . . . . . . . . . 27
2.6 Variation from Voting and Redistricting . . . . . . . . . . . . . . . . . . . . . . . 28
Incumbent Representatives and Redistricting Party Change . . . . . . . . . . . . . 32
2.7
2.8 Conclusion .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

. .

. .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

. .

3.2

3.1

3.3 Eﬀect of a Mass Layoﬀ on Tax Base . . . . . . . . . . . . . . . . . . . . . . . .

CHAPTER 3 MUNICIPAL GOVERNMENT REACTION TO MASS LAYOFFS IN OHIO 37
Introduction .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.1.1 Municipal Reaction to Mass Layoﬀs . . . . . . . . . . . . . . . . . . . . . 37
3.1.2 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Institutions and Data
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2.1 Ohio Municipalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2.2 WARN Act Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
. 45
Income Tax Base . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Property Tax Base . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
. .
3.4.1 Tax Rates .
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
.
.
3.4.2 Revenue .
.
3.4.3 Expenditure . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4.4 Cities’ Assets and Debt . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3.4 Municipal Response .

3.5 Conclusion .

3.3.1
3.3.2

. .

.
.
.
.

.

.

.

.

.

v

CHAPTER 4 BALLOT ORDER AND BALLOT ROLL-OFF: EVIDENCE FROM LO-
CAL REFERENDA IN OHIO (COAUTHORED WITH MIKE CONLIN
AND PAUL THOMPSON)

. . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 Data and Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
. . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Empirical Methodology and Results
4.4 Conclusion .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

Introduction .

. .

. .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

APPENDICES .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
APPENDIX A NOTE ON PROJECT LEVEL STATISTICS . . . . . . . . . . . . . . 71
APPENDIX B TABLES .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
.
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
APPENDIX C FIGURES .
. 100
.
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

BIBLIOGRAPHY .

.

.

.

.

.

.

vi

LIST OF TABLES

Table 2.1: Summary Statistics, by Party-Area of Intersection . . . . . . . . . . . . . . . . . 72

Table 2.2: Summary Statistics, by Party-District

. . . . . . . . . . . . . . . . . . . . . . . 73

Table 2.3: Representative Transitions and Relationship with Other Gov. Branches . . . . . . 74

Table 2.4: Eﬀect of Redistricting Party Change on Road Construction Spending, 2007-2018

75

Table 2.5: Eﬀect of Redistricting Party Change on Number of Projects, 2007-2018 . . . . . 76

Table 2.6: Eﬀect of Party and Alignment with Gov. on Construction Project Funding,

Compare Voting Changes to Redistricting Changes 2007-2018 . . . . . . . . . . 77

Table 2.7: Eﬀect of Redistricting Party Changes on Construction Project Funding, Con-

trolling for Areas that Changed Representative 2007-2018 . . . . . . . . . . . . 78

Table 2.8: Summary Statistics, Projects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

Table 3.1: Descriptive Statistics .

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

Table 3.2: The Eﬀect of Job Loss in a Mass Layoﬀ on Tax Base . . . . . . . . . . . . . . . 82

Table 3.3: The Eﬀect of Job Loss in a Mass Layoﬀ on Tax Rates . . . . . . . . . . . . . . . 83

Table 3.4: The Eﬀect of Job Loss in a Mass Layoﬀ on Revenue

. . . . . . . . . . . . . . . 84

Table 3.5: The Eﬀect of Job Loss in a Mass Layoﬀ on Expenditure

. . . . . . . . . . . . . 85

Table 3.6: The Eﬀect of Job Loss in a Mass Layoﬀ on Debt and Asset Balance . . . . . . . 86

Table 4.1: Summary Statistics .

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

Table 4.2: Share of Voters who did not Vote on Referendum . . . . . . . . . . . . . . . . . 90

Table 4.3: Share Yes for Township Referenda . . . . . . . . . . . . . . . . . . . . . . . . . 91

Table 4.4: Share Yes for County, School District, and Municipality Referenda . . . . . . . . 92

vii

LIST OF FIGURES

Figure 2.1: Partisan Control of Ohio Government . . . . . . . . . . . . . . . . . . . . . .

. 93

Figure 2.2: House Districts in Ohio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

Figure 2.3: Summed Road Construction Funding, 2007-2012 vs. 2013-2018 . . . . . . . . 95

Figure 2.4: Trends in Expenditure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. 96

Figure 3.1: Job Loss in Mass Layoﬀs as Percent of Population, Ohio Municipalities 2009 . . 97

Figure 3.2: Percentage Change in Income Tax Base and Employment after Mass Layoﬀ . . 98

Figure 4.1: Voter Characteristics by Election Date and Jurisdiction . . . . . . . . . . . . . 99

viii

CHAPTER 1

INTRODUCTION

State and local government activities comprise a substantial share of economic activity in the United
States, and provide goods demanded by almost all citizens. According to the Census Bureau, state
and local governments spent $3.4 trillion in 2015 [United States Census Bureau, 2017b], roughly
19% of United States Gross Domestic Product. State and local governments produce many public
services often considered essential, including public safety, roads, education, sanitation, and utility
services. State and local tax and expenditure decisions drive important choices by private citizens:
where to live, where to start a business, and where to educate one’s children.

Besides the impact of their ﬁscal decisions, state and local governments merit special attention
for three important reasons. First, despite the scope and inﬂuence of state and local ﬁscal decisions,
some issues, like the eﬀect of partisanship on expenditure, have received more study at the federal
level than at the state or local level. Second, state and local governments face special challenges that
matter less to national governments. For example, I argue that state and local governments are more
vulnerable to economic ﬂuctuations than national governments, while facing legal limitations on
their tax and borrowing decisions, especially at the local level. Finally, state and local governments
provide a broad array of institutions and populations. Certain economic phenomena may manifest
themselves diﬀerently within diﬀerent institutional frameworks - in addition, certain institutions
may provide a unique opportunity to test theories about economic behavior.

My dissertation studies how local labor demand shocks, the political process, and voting
behavior aﬀect state and local public good provision. Three papers comprise my dissertation.
Although all relate to state and local public ﬁnance, each approaches the topic from a diﬀerent
perspective. “Legislative Redistricting, Party Politics, and the Spatial Distribution of Transportation
Expenditure,” combines political economy and public ﬁnance by asking how the political process
aﬀects the allocation of public goods. “Municipal Government Reaction to Mass Layoﬀs in Ohio”
joins labor economics and public ﬁnance by asking how negative labor demand shocks aﬀect tax,

1

expenditure, and borrowing decisions by local governments. Finally, “Ballot Order and Ballot
Roll-Oﬀ: Evidence from Local Referenda in Ohio” combines behavioral economics and public
ﬁnance, studying the eﬀect of voting behavior on local tax decisions.

2

CHAPTER 2

LEGISLATIVE REDISTRICTING, PARTY POLITICS, AND THE SPATIAL

DISTRIBUTION OF TRANSPORTATION EXPENDITURE

2.1 Introduction

The condition of public roads serves as a critical benchmark of state government performance.
The average US state devotes ﬁve percent of its direct expenditure to highways. Although transporta-
tion lags behind education and public welfare in dollar amount, the importance of transportation
spending stems from its visibility and broad use. While only a fraction of the population receives
social assistance, almost everyone uses roads [Fisher, 2009, pp. 150-151]. Besides conferring
direct beneﬁts to citizens, a functional highway system fosters local development by providing a
necessary input for most industries. In the 2017 edition of Area Development magazine’s annual
site selection survey, corporate executives and managers regarded highway accessibility as the most
important of twenty-seven location draws. In fact, the survey has rated this the most or second most
important factor for ﬁve years in a row [Fisher, 2009, Gambale, 2016, 2017].

State governments bear primary responsibility for road construction in most of the United
States. In Ohio, the state spent $3.07 billion on highways and roads in 2014. This constituted the
majority of the $5.41 billion of state and local spending on Ohio roads and highways, including
the bulk of capital expenditure ($2.41 billion out of $3.19 billion).1 Because road construction and
maintenance disproportionately beneﬁt citizens living near the work site, the state government may
favor certain geographic areas through the provision of road construction projects. I test whether a
geographic area in Ohio receives more road project funding when its state representative belongs
to the Republican party, which has controlled the Ohio State Assembly and governorship for most
of the past two decades.

1These totals closely reﬂect national spending patterns. On aggregate, states undertook 59% of expenditure on
highways (57% in Ohio), including 73% of capital expenditure (76% in Ohio). [United States Census Bureau, 2017a]

3

I estimate how changes in political party, generated by the 2012 state legislative redistricting,2
aﬀected the distribution of transportation expenditure throughout Ohio. The redistricting process
divided the state into 265 areas of intersection between the ninety-nine old legislative districts and
the ninety-nine new legislative districts. I estimate the change in dollar value of state sponsored
construction projects before and after redistricting in these areas of intersection, depending on
whether an area shifted to a state representative of a diﬀerent party after redistricting.

Previous papers studying partisan alignment’s eﬀect on resource allocation [Berry et al., 2010,
Case, 2001, Solé-Ollé and Sorribas-Navarro, 2008] have focused on change through elections.
Change in a district’s party through voting depends on voters’ preferences and beliefs about the
party’s ability to provide public goods. In addition, party change through voting may be correlated
with unobserved demographic or economic shifts that determine road construction spending. For
these reasons, voting outcomes in a district likely depend on public good provision to the district,
biasing any estimates that depend on party variation through voting.

The redistricting process creates variation in political party that is independent of voting out-
comes. After redistricting, 56 areas of intersection moved to a representative of the other political
party - 34 moved from the Republican to the Democratic party, and 22 moved from the Democratic
to the Republican party. Using variation from redistricting, I ﬁnd that areas changing from a Re-
publican to a Democratic state representative experienced a large decrease in highway construction
expenditure, relative to areas represented by a Republican before and after redistricting. Areas
switching from a district with a Republican representative to a district with a Democratic represen-
tative received $3.5M less per year after redistricting than areas that remained in the Republican
party. In percentage terms, areas that remained in a Republican district before and after redistricting
received 66% more annual construction expenditure than areas that moved from a Republican to
a Democratic district. The governor belonged to the Republican Party and the Republicans held a
2Ohio, like every other state, redraws its state legislative districts every ten years. Redistricting ensures that each
district contains the same share of the state’s population as other districts. According to the United States Supreme
Court in the 1964 Reynolds v. Sims ruling, the equal protection clause of the United States Constitution guarantees
that all state legislative districts have “substantially equal” population. The Department of Justice and the federal court
system have interpreted “substantially equal” as allowing each district to vary by 10% from an equal share of the state’s
population, and Ohio’s State Code establishes the same standard. See Section 2.4 for more on the redistricting process.

4

majority in both chambers of the State Assembly during this time. Thus, this ﬁnding suggests a
preference in public good allocation toward areas represented by the governing party.

Despite the drop in expenditure, the total number of projects in areas moving from Republican to
Democratic districts did not change. Areas represented by a Republican before and a Democrat after
redistricting, however, received fewer large projects (projects involving new construction, widening,
intersection and interchange work, miscellaneous projects, and bridge repair) than areas remaining
with a Republican representative. From 2013-2016, areas that moved from a Republican district to
a Democratic district due to redistricting gained 76% fewer large construction projects than areas
that belonged to a Republican district before and after redistricting. These results suggest that, while
representatives’ political aﬃliation does not inﬂuence routine repairs and maintenance, partisan
politics impact allocation of large projects either through legislation, representatives lobbying the
Department of Transportation, or action by the Governor’s oﬃce. Meanwhile, areas that moved
from a Democratic to a Republican district during redistricting years did not receive an expenditure
increase. If it is easier to delay major projects than to undertake new ones, I would expect to see
a drop in large projects within areas moving to Democratic districts sooner than an increase in
large projects in areas moving to Republican districts. I also ﬁnd that districts with an incumbent
representative receive less funding than those with an incumbent. Areas that moved from a
Democratic to a Republican district lost their incumbent representative, and the associated decline
in funding may have oﬀset the gain from moving to a Republican representative.

I compare my estimates, using variation in representative’s party through the redistricting
process, to estimates based on variation in representative’s party through election outcomes. I ﬁnd
that the negative eﬀect of moving from a Republican to a Democratic district identiﬁed through
redistricting variation exceeds the eﬀect identiﬁed using variation through election outcomes.
Meanwhile, the estimates based on changes through elections suggest that areas moving from
Democratic to Republican districts would experience higher expenditure - this doesn’t happen after
changes due to redistricting. In addition, the estimates based on voting suggest that the change in
expenditure was much larger when an incumbent representative had been voted out of oﬃce than

5

when the incumbent had chosen not to run again. In years with a Democratic governor, areas where
a Democratic incumbent had been voted out received $6.4M less annual construction funding in the
next two years. If the incumbent Democratic representative had retired and a Republican replaced
her, however, expenditure increased slightly. The diﬀerent eﬀect for incumbents who lost rather
than retired suggest possible selection issues associated with variation through election outcomes.
This diﬀerence in estimated eﬀect may reﬂect reverse causality - in areas where expenditure had
begun to fall, voters blamed their incumbent representative. More generally, unobservable changes
in economic or demographic determinants of transportation expenditure are likely to be correlated
with residents’ political preferences that drive election outcomes.

2.2 Related Literature

Numerous empirical papers have estimated the impact of partisan alignment on expenditure dis-
tribution. Case [2001] distinguished two possible objectives: a ruling party seeking to maximize the
size of its majority will target swing districts, while a risk averse ruling party seeking to maximize
the probability of maintaining its majority will target its base. Studying poverty assistance expen-
ditures in Albania, she found that the majority Democratic Party directed transfers toward its own
base. Similarly, Castells and Solé-Ollé [2005] estimated a structural model that disentangled three
motives for distributing transportation expenditure across Spanish electoral districts: the marginal
eﬀect of funds on output, redistributive concerns, and expenditure’s eﬀect on the probability that
the majority party would win reelection. They found that funds were distributed more according
to their eﬀect on the ruling party’s probability of winning reelection than eﬃciency and equity
concerns. Solé-Ollé and Sorribas-Navarro [2008], presented a model where the party controlling
the national legislature prefers to distribute funds to lower level governments with the same partisan
aﬃliation. Testing the model’s predictions on Spanish municipalities from 1993-2003, they found
that aligned municipalities received 40% more grants than unaligned. Other papers have used
variation in the ruling party [Johansson, 2003] or a regression discontinuity design in legislator
vote share [Bracco et al., 2015] to evaluate the proposition that expenditure distribution across

6

districts depends on the party that represents those districts. Albouy [2013] ﬁnds that United States
senators and representatives belonging to the majority party attract more funds to own states, also
using a regression discontinuity design. Finally, some authors have asked whether federal house
districts [Berry et al., 2010] or states [Larcinese et al., 2006] aligned with the president receive
more federal expenditure.

The above literature relies on variation in partisan alignment through election outcomes. But
expected growth in expenditure could inﬂuence who runs in an election and who wins. Redis-
tricting provides changes in partisan alignment that do not depend on election outcomes. Voters
from Republican districts are regrouped with voters in Democratic districts, and vice-versa. In
many areas, the process changes the party of the legislator representing an area, due to shifting
district boundaries rather than election outcomes. By exploiting redistricting variation to identify
coeﬃcient estimates, I avoid selection issues associated with election variation, issues that could
lead to biased estimates.

A few papers within the political science literature have used redistricting as a source of ex-
ogenous variation. Ansolabehere, Snyder, and Stewart [2000] exploited this variation to estimate
the component of “personal voting” in an incumbent politician’s electoral advantage. After redis-
tricting, an incumbent only possesses a personal advantage among the voters in the intersection of
her new and old district. Thus, the authors measured personal voting advantage by comparing the
incumbent’s vote share between old and new counties in her district. Ansolabehere, Snyder, and
Stewart [2002] ﬁnd that, following a 1960s court ruling equalizing the population of state legislative
districts, more spending went toward counties in formerly underrepresented districts. Chen [2010]
uses a State Senate expansion in New York to test whether areas with more overlap in the upper
and lower house receive more earmark expenditure. Chen’s identiﬁcation strategy is similar to
my own - he considers the eﬀect of a zip code area moving from one district to another. He also
includes indicators for zip codes moving from Democratic to Republican districts or vice-versa,
although a much smaller share of areas changed party in his sample relative to mine. My paper
considers a diﬀerent outcome - both count and expenditure on transportation expenditure, rather

7

than aggregate earmark expenditure. My paper considers not only the eﬀect of representative’s
party alignment with the majority in the state government, but also the eﬀect of the senator and
representative belonging to the same party. My paper ﬁnds diﬀerent eﬀects from Chen [2010]
- he ﬁnds that areas moving from the minority to the majority party in the senate experience a
larger magnitude change than areas moving from the majority to the minority party, while I ﬁnd the
opposite. I also compare change in the number of total projects and the number of large projects
when an area changes parties. Perhaps most importantly, my paper compares estimates identiﬁed
from variation through voting to estimates identiﬁed from variation through redistricting, and ﬁnds
that they diﬀer.

I estimate that areas belonging to Republican districts before and after redistricting receive
a 66% increase in construction expenditure, compared to areas that move from Republican to
Democratic districts. This ﬁgure exceeds the estimates in previous papers. At the federal level
in the United States, Berry, Burder, and Howell [2010] estimate that House districts belonging
to the president’s party receive 4.5% more federal spending than other districts. Albouy [2013]
estimates that states with an entirely Republican congressional delegation would earn 31% more
transportation funding when the Republican party controlled the Senate, House and presidency -
states with an entirely Democratic delegation received only 17% more when Democrats controlled
the Senate, House, and presidency. Solé-Ollé and Sorribas-Navarro [2008] and Bracco, Lockwood,
Porcelli, and Redoano [2015] each estimated that lower level governments controlled by the same
party as the national government receive a roughly 40% increase in grants.

Estimates in my paper diﬀer from previous estimates for multiple reasons. First, I estimate the
eﬀect on expenditure when a district boundary change moves part of a district to a new representative
belonging to a diﬀerent party, while previous papers estimate the change in expenditure after
voters elect a representative from a diﬀerent party. Change in party through election outcomes
likely depends on changes in voters’ preferences, changes that may be correlated with shifts in
unobservable determinants of transportation expenditure. This fact implies that there are selection
issues associated with the type of districts that change party through elections. At the end of the

8

paper I estimate the change in transportation expenditure associated with a party change through
election variation. Using this variation, I estimate that areas moving from the Republican to the
Democratic party in years with a Republican governor only lose $1.2M - $2.2M in expenditure per
year, depending on the speciﬁcation.

Rather than studying expenditure distribution at the national level, I focus on state level al-
location. The spending item I consider is particularly amenable to this research question. Road
construction expenditure contrasts with spending on poverty assistance, earmarks, or education,
items that beneﬁt only a subset of the population. With these items it is diﬃcult to account for
demand heterogeneity, and demand for these items will likely correlate with partisan preferences.
Also, because state legislative districts are small and most people use roads, a road construction
project within a state legislative district serves a larger share of district population than other
expenditures. Thus road construction expenditure more eﬃciently targets the constituency that a
representative wishes to favor. In addition, it is diﬃcult to account for heterogeneity in demand for
items used by a fraction of the population. Need for education or poverty assistance will ﬂuctuate
widely depending on economic and demographic trends within a district. Because of the diﬀerent
environment, as well as the exogenous source of regressor variation, the estimated eﬀect in my
paper may diﬀer from the results in previous studies.

2.3 Road Project Data

I estimate the eﬀect of representative’s party aﬃliation on highway construction funding, using
all highway projects in Ohio that received direct state funding from 2007 to 2018.3 The dataset,
from the Ohio Department of Transportation’s “Transportation Information Management System,”
includes 9780 projects with 30,045 work locations and total estimated expenditure of $23.1 billion.

3Ohio Department of Transportation, Transportation Information Management System,

https://gis.dot.state.oh.us/tims. The dataset includes all projects that have secured funding from 2003 onward. There
are some projects in the ﬁle budgeted for years after 2018 (as late as 2040), but I exclude these. I have chosen to include
projects beginning in 2007 for two reasons. First, this centers the dataset symmetrically around January 2013, when
representatives from the new districts took oﬃce. Second, Governor Ted Strickland, a Democrat, took oﬃce in 2007.
The dataset begins with his tenure, followed by Republican John Kasich’s two terms in oﬃce. This allows me to easily
control for the political party of the governor.

9

The unit of observation is work location. Some work locations (such as bridge repair) occupy a
single point. Other work locations involve repairs extending over a stretch of road and are coded as
lines. Many projects include multiple work locations, often with some coded as points and others
as lines. The dataset includes geographic coordinates for each work location to six decimal places
- I can place each work location precisely within a house and senate district.4,5

The Ohio Department of Transportation (ODOT) maintains more than 50,000 lane miles of
roads and 15,000 bridges. ODOT is responsible for the state’s interstate highways, numbered federal
highways, and state routes. Outside of the state highway system, local governments maintain more
than 200,000 lane-miles of county, township, and municipal roads through a mix of state formula
grants and local taxes. On locally managed roads, the state provides funding or directs the
distribution of federal funds for some construction projects. Ohio’s transportation budget passes
every odd-numbered year, with the legislature allocating funding within line-item categories and
dictating general parameters for how these funds will be spent. Ultimately, through legislation, the
legislature determines the level of state funding by setting the state fuel tax rate and tolls. The
state splits its fuel tax, by formula, between the ODOT and local governments - the legislature
also decides what percentage will go to each. Outside of the biennial budget, the state legislature
exercises some control over the distribution of expenditure through speciﬁc programs that distribute
state and federal funds. The governor’s oﬃce exercises more direct control over these decisions.

The ODOT, a branch of the governor’s oﬃce whose director is a gubernatorial appointee,
plans and approves speciﬁc projects. Large new construction projects in excess of $12M6 require

4This means the location is precise to within a few feet.
5A project may span multiple house districts if it includes multiple work locations. Cost is recorded at the project
level, so I must allocate the cost among work locations for each project, and among districts for some work locations
that span multiple districts. If a project has multiple work locations then I divide the cost evenly among work locations.
A project may also span multiple districts if a work location coded as a line crosses district boundaries.
If one
work location spans multiple districts, I allocate its cost between those districts according to the length of segment in
each district. Consider a work location allocated expenditure est_cost across N areas of intersection, with each area
containing a roadwork segment of length lengthi. Each area i is allocated est_costi = est_cost ×
.
Summary statistics of some key characteristics at the project level are listed in Table 2.9 of the Appendix.

lengthi
n=1 lengthn
6From 1997 to 2012 the cut-oﬀ was $5M. In 2012 the legislature raised it in accordance with the ODOT’s estimation

N

of increased costs.

10

approval by the Transportation Review Advisory Council (TRAC), a committee whose members
include the Director of Transportation, six gubernatorial appointees, and two legislative appointees.
Beyond the ODOT’s regular budget, some funds administered by the ODOT may be allocated by
executive or legislative action on a biennial or a triennial basis. For example, the Bridge Partnership
Program, initiated in 2014 by executive action, distributed $120M of federal money for ﬁscal years
2015-2017. In May 2017, the legislature extended the plan, and promised additional funds, in a
separate bill from the biennial budget.

A representative’s party could inﬂuence expenditure toward her district in several ways.

If
the representative belongs to the governor’s party, then she may be able to lobby the ODOT for
funds more eﬀectively. A representative could also coordinate with members of her own party
in the House or Senate. For example, she could threaten to introduce a bill altering the ODOT’s
funding. A representative’s power within the state assembly may be limited, however, if she is
not a member of the majority party. Similarly, a representative belonging to the majority party
may possess more power to protect scheduled projects in her district from delay or cancellation.
Partisanship could also inﬂuence decisions by the TRAC. As Republican governors or Republican
legislators appointed almost all members of the TRAC during this period, the TRAC may have
favored Republican controlled areas. A Republican representative lobbying or negotiating with the
TRAC to direct projects toward her own district would likely have an advantage, compared to a
Democrat.

Table 2.1 presents summary statistics for a few variables at the area of intersection-by-year level.
Recall that “area of intersection” refers to one of the 265 geographic areas formed by the intersection
of old and new House districts. These areas form the unit of observation in the empirical analysis
that follows. The ﬁrst and second column divide area of intersection by political party, while the
third column totals all areas. For reference, I have summarized all variables at the house district
level in Table 2.2.7

The ﬁrst two rows describe the number of projects and estimated total expenditure8 on all
7Many more characteristics of projects are available in the TIMS database.
8This is the sum of estimated total cost of all projects in the area of intersection. Estimated total cost is set during

11

projects. On average, each area of intersection receives $6.8M per year in state construction
project expenditure on 32 construction projects.9 The third and fourth rows show number and
estimated total cost of large projects. Large projects belong to the six categories of road work
with the largest average project cost (new construction, intersection or interchange work, widening,
miscellaneous projects,10 and bridge repair - see Table 2.9). Roughly one-tenth of the projects in
each area of intersection are large projects, although these projects account for more than half of
total expenditure. The ﬁfth row displays estimated total cost for projects where the primary sponsor
is either the ODOT or some other state agency. Roughly one-ﬁfth of state construction project
expenditure goes toward projects sponsored by local governments.

Table 2.2 provides a better picture of how these variables diﬀer between Republican and
Democratic districts - districts all contain similar population, unlike the areas of intersection
described in Table 2.1. Total expenditure remains almost the same between Republican and
Democratic districts, but Republican districts receive more projects. This likely reﬂects Republican
districts’ location in more rural areas (see Figure 2.2 in the next section). Because rural areas
contain more road mileage but fewer bridges and intersections, Republican districts receive more
small projects and fewer large projects.

Localities also contribute their own funds to road maintenance. Three variables in Tables 2.1
and 2.2 measure contribution to road provision by local governments: public works expenditure by
townships and counties, transportation expenditure by municipalities,11 and average property tax
rate devoted to roads by all local governments.12 Local expenditure data come from summaries

the planning stage, before ﬁrms bid on the contract - it may diverge somewhat from actual expenditure, but it probably
more closely reﬂects policymakers’ expectation of costs when allocating projects.

9I assign projects to years based on the ﬁscal year when the project will be funded. The Ohio ﬁscal year runs from
July 1 of the preceding calender year through June 30 of the same calender year, so that ﬁscal year 2017 runs from July
1, 2016 through June 30, 2017. I assign projects to the calender year before the ﬁscal year when they are funded, so
that all projects funded in ﬁscal year 2017 are 2016 projects.

10Includes all projects that could not easily be classiﬁed. Includes section improvement, intelligent vehicle systems,

building demolition, and realignment or relocation.

11In the case where a local government’s boundaries straddle multiple districts, I assign expenditure to each district

proportional to the fraction of the local government’s population on each side of the boundary.

12The Ohio Department of Taxation publishes annual property tax millage rates levied by every local government
in the state. The population weighted average millage rate devoted to roads is constructed by summing all road millage
rates in the area, each weighted by the local government’s share of the area’s population.

12

of Annual Financial Reports13 published by the Ohio Auditor’s oﬃce. Unfortunately, only the
ﬁnancial reports for municipalities feature a speciﬁc item for transportation.14 For townships and
counties, the Annual Financial Reports summarized total expenditure on public works without
separating transportation expenditure.15
I derive local expenditure for each area of intersection
by summing expenditure by all local governments within the area.16 When a local government’s
jurisdiction overlaps two areas, I divide spending between the two areas according to the proportion
of the local jurisdiction’s population within each area. Municipalities spend more on transportation
in Democratic districts, while county and township governments spend more on public works in
Republican districts - in more rural areas, county and township governments play a larger role in
public service provision than in urban areas.

Ohio funds a large share of local road expenditure through gas tax revenue. The state splits
revenue from its gas tax levies between the ODOT and distributions to counties, townships, and
municipalities. The state distributes funds to municipalities according to the share of registered
vehicles inside each municipality, and to townships according to the share of registered vehicles
and road miles in the township.17 Changes in these variables over time, then, reﬂect changes in
13The municipality data from 2006-2012 were taken from municipal audits conducted by the Ohio auditor. I also
ﬁlled some data that were missing in the county Annual Financial Reports from audits conducted by the Ohio auditor.
All other data came from summarized annual ﬁnancial reports.

14Municipalities reported transportation expenditure diﬀerently in the Annual Financial Reports - some listed
expenditures on “Public Works” or “Public Services” without listing transportation expenditures separately. As
cities that reported zero transportation expenditure always reported zero utility expenditures and often reported zero
expenditure on public health, I gather that these items are often grouped together under “Public Works” or “Public
Services.” For municipalities with no spending on transportation listed in the Annual Financial Report, I impute
transportation spending as the sum of expenditure on public health, basic utilities, public works, public services, and
transportation, multiplied by the average ratio of transportation to these ﬁve items for all other municipalities in that
year.

15The local ﬁnancial data also includes a separate item for capital outlay. For many localities, this item includes
capital expenditure on roads. But according to these data, capital expenditure by municipalities, counties, and townships
throughout the state exceeds $1.7 billion per year, roughly the same amount as state road construction expenditure.
According to the census, however, local capital spending on roads equaled only one third of state capital spending,
implying that the majority of municipal capital spending did not go towards roads and highways. For this reason, I do
not include local capital expenditure as a covariate.

16The data on municipal transportation expenditure and public works expenditure, are not available for all local
governments in all years. In cases where only one consecutive year of data is missing, I interpolate the missing year as
the mean of the prior year and the year after. I do not include the municipalities and townships that are missing more
than one consecutive year of data.

17More precisely there are two funds for the distribution of gas tax: the Gasoline Excise Tax Fund (Fund 7060)
and State and Local Government Highway Fund (Fund 7068). For municipalities, distributions from both funds are

13

the number of registered vehicles in each area. Localities also fund expenditure through their own
tax eﬀort. Municipalities, counties, and townships may, with referendum approval from residents,
levy a special property tax rate for road maintenance. Tables 2.1 and 2.2 summarize the population
weighted average local property tax devoted to roads in each area of intersection and each district.
On average, local governments raised 0.75 mills of property tax for road repair each year. More
is raised in Republican than in Democratic districts - again, likely because Republican areas tend
to be located in more rural areas with township governments that collect more revenue through
property tax.

Employment data in the next two rows come from the LEHD branch of the Census Bureau.
These data are available at the census block level, and have been summed at the area of intersection
and district level. Republican districts have less employment and larger employed population -
many people reside in Republican areas but commute to Democratic areas for work. Population
is available only for 2010, from the Census Bureau. Note that Republican districts had slightly
higher population in 2010 than Democratic districts. As mentioned in the introduction, more areas
shifted from Republican to Democratic districts than vice versa, despite Republicans picking up
one more seat - the redistricting process needed to add population to Democratic districts. The
last two variables provide some measure of need for transportation expenditure. Suﬃciency rating
of bridges comes from the National Highway Administration’s bridge inventory. Bridges tend to
be in better repair in Republican districts. This may reﬂect a pattern of past political partisanship,
but may stem from other diﬀerences between Republican and Democrat controlled regions. The
last row shows daily vehicle miles traveled on roads in USDOT functional classes 1-6 - the ODOT
provides this data, as well.18

proportional to the number of registered vehicles in the municipality. Distributions from Fund 7068 are equal for all
townships. From Fund 7060, roughly half is distributed equally. For the other half, the township gets either a minimum
distribution of funds, or 70% of the share it would receive if the funds were allocated half according to share of road
miles and half according to share of registered vehicles. Funds for counties are split evenly. For townships, I include
only distributions from Fund 7060, which accounts for about 2

3 of the total distributed to townships.

18For a description of functional classes, please see Footnote 26 in Section 2.5.3.

14

2.4 Districts and Redistricting

The Ohio legislature (“General Assembly”) has an upper house (“Senate”) and a lower house
(“House of Representatives”). There are ninety-nine representatives and thirty-three senators in
Ohio. Each representative serves one of the ninety-nine state house districts. Each senator serves
one of the thirty-three senate districts, each constructed from three contiguous house districts.
Voters choose their representatives for two year terms every even-numbered year and the elected
representatives take oﬃce in January of every odd-numbered year. Senators serve four year terms,
with half of the senate elected every even-numbered year. In recent years the Republican party has
dominated state politics in Ohio: from 2007-2018 Republicans controlled the state Senate every
year, the state House of Representatives in all but two years, and the governorship in all but four
years (see Figure 2.1).

State legislative districts are redrawn every ten years, based on the population information
In Ohio, a political commission determines state legislative
provided by the decennial census.
districts. This commission is composed of the governor, state auditor, secretary of state, and one
member each selected by the House Majority and Minority Leaders.19 The commission draws
districts such that each district contains roughly 1/99th of the state’s population. By law no district
may vary more than 10% from this benchmark. Redistricting occurs the year after the census,
most recently 2011, meaning state representatives were elected from the new districts starting in
2012. Similarly, half of state senators were elected from the new districts in 2012, half in 2014.
This means that state representatives elected from the new districts took oﬃce in 2013, and state
senators elected from the new districts took oﬃce either in 2013 or in 2015.

Figure 2.2 displays the geographic intersection between pre-redistricting (“old”) and post-
redistricting (“new”) House districts, the unit of observation in my analysis.20 Redistricting created
19(Ohio Constitution XI § 1). The House and Senate Majority leaders each select three members for an advisory
committee, including one from the opposing party [State of Ohio, Ohio Revised Code §103.51]. Note that Ohio voters
amended the constitution to change these procedures in 2015, and yet another amendment began collecting signatures
in 2017. The 2015 amendment expanded the redistricting committee from ﬁve to seven (with two from the opposing
party). The circulating amendment would require new districts to be redrawn based on past political preferences and
restrict the breakup of counties.

20I derive each intersection between new and old districts by spatially joining 2003-2012 districts and 2013-2022

15

301 intersections between old and new house districts, and ninety-two intersections between old
and new senate districts. I restrict attention to areas with population over 100; there were 265 of
these for the House and seventy for the Senate.21 The boundaries of all but six House districts and
two Senate districts changed. In some districts that had experienced population growth or decline,
the commission simply added or removed a slice of land. As of the 2010 Census, for example, old
house District 2 had grown to a population of 170,573, 53% larger than 1
th of state population.
99
116,760 people remained in new District 67, a geographic subset of old District 2, while 54,813
people were moved to new District 68. Other districts were carved up more thoroughly: in thirteen
house districts no intersection of the old and new district contained more than half of the old
district’s population.

I estimate the eﬀect of party change on spending using transitions between parties during the
redistricting year: “R to R,” “R to D,” “D to R,” and “D to D.” Figure 2.2 plots this “transition
matrix” for the House of Representatives onto the map of Ohio. Generally, urban areas like Toledo,
Cleveland, and Cincinnati lean Democrat, while rural areas lean Republican. Appalachia and the
“Rust Belt” region around Lake Erie are Democratic strongholds. Areas that changed parties after
redistricting concentrated in these regions, around the borders between Democratic and Republican
districts near Columbus, Cincinnati, Toledo, and in the North and East of the state.

Table 2.3 documents how areas of intersection transitioned between representatives in all ﬁve
state House of Representatives elections from 2008 to 2016. The ﬁrst column displays the proportion
of representative transitions in each party transition for election years that did not immediately follow
redistricting: 2008, 2010, 2014, and 2016. The second column shows representative transitions for
the redistricting election year, 2012. The third column reports test statistics for the diﬀerence in
means between redistricting and non-redistricting years.

districts using ArcMap10.1 GIS software.

21The spatial merge process, rather than true change in the boundaries, may have produced the very small areas
of intersection. Most areas with population under 100 have zero population. There are eleven areas with population
between 100 and 1000. These may result from the commission “ﬁne tuning” the district’s size or composition. Such ﬁne
tuning could conceivably include the exchange of an area with zero population, such as transferring the headquarters of
a powerful corporate donor [Daley, 2016], although this seems unlikely at the state government level. More likely, the
areas of intersection with population in the hundreds reﬂect a legal requirement to respect local government boundaries
where possible.

16

Table 2.3’s ﬁrst panel shows how areas transitioned between parties.22 During the redistricting
year, many more areas changed from a Republican to a Democratic representative, or vice-versa,
than in other years.
In non-redistricting years, only 8% of areas changed party. But in the
redistricting year, 13% (34 areas) changed from a Republican to a Democratic representative, and
8% (22 areas) switched from Democratic to Republican.

The second panel shows the probability that an area reelects its old representative after an
election. This is much more common in non-redistricting years, occurring with a 0.39 + 0.31 =
0.70 probability, compared to only a 0.30 probability in the redistricting year. The second panel
also shows the proportion of representatives who replaced an incumbent from their own party.
Most new representatives belonged to the same party as their predecessors: 73% (0.12 + 0.10
1 − 0.70 ) in
non-redistricting elections, and 70% (0.28 + 0.21

1 − 0.30 ) in the redistricting election year.

The third panel tabulates the possible reasons why an area did not keep its incumbent.

In
typical election years, the incumbent had either been voted out or did not run for reelection. Voters
rarely removed an incumbent candidate from oﬃce - this occurred in only 4% of non-redistricting
area-by-year observations, and in only 1% of redistricting area-by-year observations.
In most
areas that changed representative in non-redistricting years, the area’s incumbent did not run for
reelection23- this occurred in 26% of area-by-year observations during non-redistricting election
years, out of the 30% of area-by-year observations where the representative changed. In contrast,
the incumbent chose not to run in only 17% of areas during the redistricting year, while 70% of
areas changed representative. Rows 13-14 show the share of areas whose last state representative
won reelection, but where the area had moved into a diﬀerent district than that representative -
this category includes more than half of all areas and the vast majority of areas that had changed
representative

52%

100%−30% = 74%.

Table 2.3 shows that far more areas changed representative and changed party during the 2012
election than in other election years. More importantly, almost all of these changes derived from
22Columns 1 and 2 each sum to one in the ﬁrst panel. Column 1 sums to 0.92 in the second panel, the proportion

of areas that changed party in non-redistricting elections. Column 2 sums to 0.79 for the same reason.

23This category includes the eight area-by-year observations where the incumbent lost in her own party’s primary.

In each of these cases, the successful primary challenger won the general election.

17

the redistricting process. Even among the 17% of areas that changed representative because the
incumbent retired or lost, most had moved to a diﬀerent district than the old incumbent and would
have changed representative anyway. But even this fact does not reveal fully how redistricting drove
the change in representatives. In total, there were twenty districts where at least one incumbent
did not win reelection in 2012.24 This number includes eight districts with two incumbents - 16
incumbents ended up in the same district as another sitting representative, and in seven of these
double incumbent districts, one of the incumbents did not run. Thus, there were only 14 areas
where the incumbent resided in the area, and was not reelected. Only four of these areas changed
party: These were the two areas where a Republican lost the general election, and two areas where
the sitting Democrat did not run and the Democrat who did was defeated.

Because a political commission composed primarily of Republicans drew the new House dis-
tricts, some suspect that gerrymandering aﬀected the new district boundaries [Daley, 2016]. A
party that gerrymanders attempts to maximize either the number of seats that it wins or the proba-
bility that it will win a majority [Owen and Grofman, 1988]. According to the theoretical literature
on gerrymandering [Friedman and Holden, 2008, Gilligan and Matsusaka, 1999, Owen and Grof-
man, 1988, Sherstyuk, 1998] the gerrymandering party does this by adjusting the proportion of
support for its own party within districts. The optimal solution to this problem involves conceding
a minority of districts with a vast majority of voters (ideally 100%) supporting the opposing party,
while crafting another set of districts with a “safe ” majority for the gerrymandering party. Thus,
areas that shift to a new district or to a representative of the opposing party may have systemati-
cally diﬀerent levels of partisan support than other areas. Because I use ﬁxed eﬀects estimation,
correlation between changes in party through redistricting may be correlated with the level of
partisan support, without biasing the estimates. My estimates remain consistent as long as changes
in partisan preferences over time remain mean independent of changes in representative’s party
24Two incumbents lost in a general election, one lost his primary, 17 incumbent representatives did not run for
reelection. Another representative ran for reelection in a diﬀerent district that did not overlap with her old district. Six
of the representatives who did not run were term limited, one held the oﬃce in place of a diﬀerent representative who
had joined the governor’s cabinet, and one resigned while facing corruption charges. Thus, only seven representatives
chose not to run voluntarily.

18

through redistricting, conditional on the various control variables. Changes in representative’s
party through election outcomes, on the other hand, are almost certainly correlated with changes
in partisan preferences over time.

2.5 Empirical Speciﬁcations

I estimate the diﬀerence in construction expenditure between areas represented by a majority
party Republican and areas represented by a minority party Democrat. Because variation in
representative’s party aﬃliation through election outcomes may be correlated with unobservables
determining road construction, I initially use only variation from redistricting for identiﬁcation.
The empirical speciﬁcations are based on the redistricting transitions shown in Figure 2.2. I test: (i)
whether areas that left the Republican party during the redistricting year (the “R to D” transitions)25
suﬀered a drop in funding, relative to areas that remained Republican (the “R to R” transitions) and
(ii) whether “D to R” areas received more funding after redistricting than “D to D” areas. First, I
consider aggregate construction funding before and after redistricting. At the area of intersection
between old and new districts, I sum the total cost of all projects from 2013-2018 (the year after
redistricting took eﬀect and the ﬁrst ﬁve years after) and compare this average to the total cost of
projects from 2007-2012. Figure 2.3 plots average expenditure before and after redistricting for
each category: R to R, R to D, D to R, and D to D.

The bar chart indicates that expenditure did not increase in areas that switched from the
Republican to the Democratic party, while generally increasing in all other areas. The average
annual increase is large: $21M over six years for Republican areas that remained Republican, and
slightly less for areas that had a Democratic representative in 2012. Only Republican areas that
moved to Democratic representatives during the redistricting year experienced no increase.

25From now on, when I refer to party transitions, or to R to D, D to R, etc., I refer only to transitions during the

redistricting year.

19

2.5.1 Trends in Outcome Variables

The estimation in Figure 2.3 and the regressions that follow involve a diﬀerence in diﬀerences
approach. According to this conception, I compare the R to D treated areas to a set of R to R
control areas, and I compare the D to R treated areas to a set of D to D control areas. To ensure the
validity of this approach, I must ensure that the expenditure change associated with a party change
does not reﬂect an upward expenditure trend during the pre-treatment period.

Figure 2.4 plots changes in expenditure for each party transition category, with R to R areas
compared with R to D areas and D to R with D to D. As seen in Figure 2.4, R to R areas received
a relatively constant amount of expenditure from 2007-2012, while R to D areas experienced a
slight uptrend.
In the two years following redistricting, the R to D areas fell to a much lower
level of expenditure than the R to R areas, although the time trends remain roughly parallel. From
2007-2012 D to R districts experienced a general decrease in expenditure and D to D a general
increase, although there was much ﬂuctuation around this trend - after redistricting, D to R districts
did not change appreciably, while expenditure in D to D districts generally trended downward.

2.5.2 Regression Analysis

Although Figure 2.3 depicts how district expenditures vary with party transitions caused by re-
districting, this simple diﬀerencing does not account for changes in time varying determinants of
transportation expenditure. Additionally, these diﬀerences compare post-treatment years with a
Republican governor and a Republican controlled legislature (2013-2018) to pre-treatment years
with a Democratic governor (2007-2010) and a Democratically controlled House of Representatives
(2009-2010). To address these issues, I ﬁrst estimate the following speciﬁcation:

expenditureit = β0 + (β1R_to_Di + β2D_to_Ri + β3D_to_Di) × (yeart > 2012)
(1)
+ (β4R_to_Di + β5D_to_Ri + β6D_to_Di) × (yeart < 2011) + δt + γi + i

where expenditureit is area of intersection i’s road expenditure in year t. The party change
indicators - R_to_Di, D_to_Ri, and D_to_Di - capture how each area’s representative transitioned

20

between parties during the redistricting year (2012). R_to_Di, for example, equals one for areas
that moved from a Republican district in 2012 to a Democratic district in 2013. The β1 coeﬃcient
represents the diﬀerence in expenditure for these areas after redistricting for areas that changed
from Republican to Democratic, relative to areas that remained Republican (R_to_Ri is the omitted
category).

Post-redistricting elections in November 2012 closely followed the change from Democratic
to Republican governor and from Democratic to Republican House in November 2010. I want to
identify the eﬀect of changing from a Republican to a Democratic representative, or vice-versa,
conditional on the Republican party’s control of the state government. Thus, I include an indicator
for years 2007-2010 interacted with the redistricting party change indicators. The coeﬃcient
on each of the post redistricting party-change variables ((R to D) × (year > 2012),(D to R) ×
(year > 2012),(D to D) × (year > 2012)) indicates the increase in funding between the two
years immediately preceding redistricting (2011-12) and the six years after. γi denotes area of
intersection ﬁxed eﬀects, while δt denotes year ﬁxed eﬀects.

The estimates from this speciﬁcation are provided in the ﬁrst column of Table 2.4. The coeﬃcient
for areas that changed from a Republican to a Democratic legislator - (R to D)×(year > 2012) - is
quite large in magnitude. Areas leaving the Republican party received roughly $3.5M less in annual
construction funding than those that remained. From Table 2.1, the mean annual estimated cost
of projects in each area is $6.8M, with a standard deviation of $17.8M. Thus, areas that switched
from a Republican to a Democratic representative saw their funding drop by 20% of a standard
deviation. The coeﬃcient estimate is statistically diﬀerent from zero at the 95% conﬁdence level.
This is consistent with partisanship having a large eﬀect on construction funding. However, areas
that moved from a Democratic to a Republican representative received $0.73M less funding than
those that retained a Democratic representative, though this estimate is not statistically signiﬁcant.
Thus, areas leaving the Republican party seem to suﬀer a decrease in funding, while areas moving
to a Republican representative experience no increase.

This asymmetry may reﬂect an immediate decrease in funding to R to D areas from canceled

21

or postponed projects, with no immediate increase in new projects directed toward D to R areas
- planned projects may be canceled or delayed quickly, while it takes some time to develop and
begin work on new construction. Figure 2.4 supports this explanation. R to D areas experience an
immediate drop in funding, relative to R to R areas. For D to R areas, however, there is no immediate
change compared to D to D areas - instead, expenditure in D to R areas gradually increases over
the six years following redistricting, while expenditure in D to D areas gradually increases. The
asymmetry may also reﬂect an advantage for areas that retain their incumbent representative after
redistricting. The loss of funding from a D to R area moving to a new representative could oﬀset
the gain from moving to a majority party representative.

2.5.3 Controls

Construction spending likely depends on population, economic activity, and road usage, which
vary within areas across years. Local preferences for spending on roads may also change over
time. If areas that changed party during redistricting experienced greater economic growth, greater
population growth, or more growth in residents’ preferences for roads than other areas, this could
lead to biased estimates. The next speciﬁcation includes covariates to more eﬀectively model
demand for road construction spending.

First, I control for local government spending on roads and highways. Local spending likely
responds to changing demand for roads in an area. To the extent that spending on local routes
complements or substitutes for state spending, local spending in an area may attract or crowd out state
funded construction. I include the the lagged population weighted average property tax rate devoted
to roads and highways, lagged municipal spending on transportation expenditure, and lagged county
and township spending on public works within in area. Because residents directly approve the
property tax rate devoted to roads in referendum elections, this variable in particular reﬂects local
demand by residents. The expenditure variables control for local government expenditure on roads
from funds other than property tax revenue and state gas tax levies, such as income tax revenue,
sales tax revenue, or fees.

22

I also include lagged distribution of state gas tax revenue to municipalities and townships within
each area. Distributions to municipalities and townships depend on the number of registered ve-
hicles within each area, so that these variables provide a measure of residents’ road usage. I also
include lag of employment, as well as lag of the number of employed people in each area. Em-
ployment provides a measure of business activity within each area. Areas experiencing economic
growth may demand more spending on roads. On the other hand, the state may attempt trans-
portation capital projects to stimulate areas in economic decline. Employed population provides a
measure of demand for roads in the area by residents, as well as a proxy for population.

To account for road condition, I include the lagged average suﬃciency rating for bridges
inspected within each area, from the Federal Highway Administration’s National Bridge Inventory.
Finally, I control for lagged usage of state roads, measured by vehicle miles traveled on non-local
roads.26 ODOT records daily vehicle miles traveled at the county level. Although this variable
should track roadway usage and need better than any of the other control variables, its measurement
at the county level makes it a less reliable predictor of demand for smaller areas in my sample.27,28
Column 2 of Table 2.4 reports estimates when these covariates are added to the party change
indicators in Column 1. The coeﬃcient on the (R to D) × (year > 2012) becomes slightly smaller
in magnitude than the coeﬃcient in Column 1 (2.6M instead of 3.5M), and becomes marginally
statistically insigniﬁcant (p-value = 0.113). Still, the coeﬃcient remains large and negative - the
26The US Department of Transportation divides roads into seven “functional classes”: Interstates, Other Freeways
or Expressways, Other Principal Arterial, Minor Arterial, Major Collector, Minor Collector, and Local. A road’s
functional class is based on its physical characteristics and connection to other roads. Vehicle miles traveled is reported
by functional class. Although functional class does not depend on who owns or maintains the roads, only 0.003% of
state roads were classiﬁed as Local, compared to 83.5% of local government roads. For counties that cross district
boundaries, I calculate vehicle miles traveled in an area as that area’s population weighted share of vehicles miles in
the county that contains it, the same as I calculated the local expenditure variables.

27I add lagged controls, rather than contemporaneous values, to avoid problems of reverse causality. For example,
state road construction funding may substitute or complement local eﬀort, and better roads within an area may
increase traﬃc or vehicle ownership among residents. Recall that the previous sample included projects where a
local government had been the primary sponsor as long as the state or federal government distributed some aid. The
speciﬁcation with controls excludes these projects to focus only on projects sponsored by the ODOT or (in rare cases)
some other state agency.

28Unfortunately, while the sample of transportation data run from 2007-2018, the data on local expenditure and
employment only run through 2015, meaning that I lose the last two post-redistricting years when I include the lagged
controls. I also drop ﬁfteen observations where suﬃciency rating was unavailable for several years.

23

estimate’s 26% drop in magnitude mostly reﬂects the exclusion of locally sponsored projects,
which accounted for 22% of estimated cost. While restricting the data to years 2007-2016 does not
appreciably change the coeﬃcient estimates, the decrease in sample size does slightly increase the
standard errors.29 Nonetheless, the similar coeﬃcients for this truncated sample suggest that the
eﬀect of representative party change occurs quickly, and that projects on state highways rather than
locally sponsored projects drive this eﬀect.

While the set of covariates is extensive, these may not adequately control for all trends in
unobservable factors determining road construction spending. It is possible that areas changing
from the Republican to the Democratic party, or vice versa, had been growing at diﬀerent rates than
areas remaining in the same party. Areas may have been more likely to switch party in districts that
were broken up more thoroughly during the redistricting process. The redistricting process changes
districts that grew or shrank in population, and population change is likely tied to demand for roads.
Additionally, there may be time trends in distribution of districts, perhaps due to dissipating eﬀects
of the Democratic control of the House and governorship at the beginning of the sample. To deal
with these issues, Column 3 of Table 2.4 adds a time trend for each party change category, as well
as the lagged controls described above.

The diﬀerence between areas that left the Republican party and areas that stayed in the Repub-
lican party increases in magnitude to $6.6M and remains statistically signiﬁcant. Interestingly, the
coeﬃcient for areas that switched from a Democratic to a Republican representative becomes less
negative than the coeﬃcient for areas that retained a Democratic representative (by about $3.5M),
although the diﬀerence remains statistically insigniﬁcant. These estimates suggest that when the
Republican party controls both levels of the Assembly and the governorship, areas with Republican
representatives reap the beneﬁt.

29For the equations with controls - reported in Columns 2, 4, and 5 in Table 2.4, and Equations 3-6 in Table 2.5 - I
estimated the regression without controls on a sample that excluded state sponsored projects from 2007-2016. I also
excluded the ﬁfteen areas with missing data on bridge suﬃciency rating. The coeﬃcient estimates were very similar
to the estimates with controls. Estimating Equation 2 on the reduced sample without controls resulted in a lower
coeﬃcient for (R to D)×(year > 2012) coeﬃcient than in the estimates with controls ($3.1M instead of $2.6M), and a
slightly larger diﬀerence between the (D to D)×(year > 2012) and (D to R)×(year > 2012) coeﬃcients. The standard
errors were generally slightly smaller in the speciﬁcations with controls. I also ran each regression on the full set of
years but only including state sponsored projects, again excluding controls.

24

2.5.4 Percentage changes

Given the substantial dispersion in the dependent variable (see the ﬁrst column of Table 2.1), it
is important to estimate the proportional change in construction projects caused by party change.
Estimating proportional changes ensures that a few outlier areas do not drive the results. Ideally,
the natural logarithm of construction projects would replace the level as dependent variable, but
the log dependent variable is not deﬁned for much of the sample. Although each area received
33 construction projects per year on average, 14% of area-by-year observations in the sample
have received zero construction projects. Instead, Poisson quasi-maximum likelihood estimation
(QMLE) with an exponential conditional mean function provides consistent estimates of the semi-
elasticities, as long as the conditional mean is correctly speciﬁed. I estimate a model with both area
of intersection and year ﬁxed eﬀects.30,31

Columns 4 and 5 of Table 2.4 present the results of this speciﬁcation. Other than a diﬀerent
functional form for the conditional mean, these speciﬁcations are identical to Equation 1 and
Equation 2 in the ﬁrst two columns. The coeﬃcients suggest that areas belonging to a Republican
district before and after redistricting receive 56% to 66% more highway funding than areas moving
from a Republican to a Democratic district. The coeﬃcients on the party change variables,
especially the (R to D) × (year > 2012) variable, change very little with the addition of controls,
although standard errors increase rather substantially.

2.5.5 Accounting for the Senate

In Ohio, each House of Representatives district lies entirely inside of a state Senate district. We have
seen that, although the Senate was redistricted along with the House, no area changed to a senator
30Poisson QMLE remains robust to distributional misspeciﬁcation - in particular, it does not require Var[y|X]
to equal E[y|X] for consistency. If the conditional variance depends linearly on the mean (Var[y|X] = σE[y|X]),
moreover, the estimator is eﬃcient. Uniquely among parametric non-linear models, the distribution of parameter
estimates does not depend upon the distribution of the ﬁxed-eﬀects, so that the ﬁxed eﬀects coeﬃcients need not
be estimated directly. This frees the estimator from the “incidental parameters problem” - the parameter estimates
converge to their true values as the number of areas, not the number of years, increases [Wooldridge, 2010, Chapter
19].

31The number of observations is slightly lower for the Poisson regressions, as areas where expenditure was zero in

all years are excluded.

25

of the opposite party after redistricting took eﬀect. This means that areas whose representative
changed parties either moved from having a senator and a representative of the same party to
diﬀerent parties, or vice versa. In bicameral legislatures, constituents may assign credit for funds
received partially to their representative and partially to their senator [Chen, 2010]. The governor
or other members of the legislature may be less willing to direct funds to an area if they know that
a member of the opposing party will receive credit for that funding. At the same time, it is likely
easier for a representative to direct funds toward her district in coordination with a senator32 with
the same party aﬃliation.

In the next set of regressions I include an indicator for areas with both a Republican state
representative and Republican senator, and an indicator for areas with a Democratic state represen-
tative and a Democratic senator, each interacted with a post-2012 indicator. This speciﬁcation tests
whether it is advantageous for an area to have a representative and senator belonging to the same
party. In addition, it tests whether the eﬀects of party change estimated in the previous speciﬁcations
really represent a majority party advantage. Perhaps most areas had the same party senator and
representative prior to redistricting, or perhaps areas that changed party had belonged generally to
senate districts controlled by the opposing party. In either case, the eﬀect in the previous regression
would be partly an eﬀect of senator and representative party alignment, rather than an aﬀect of the
representative’s party directly.

The last two columns of Table 2.4 contain the estimates from this speciﬁcation. Column 6
shows estimates from the speciﬁcation in Column 1 with the addition of the alignment controls,
while Column 7 includes controls. The point estimates suggest that, if an area switches to a
representative from the same party as the senator representing that area, the area receives increased
highway funding. But an F-test of the joint signiﬁcance of the coeﬃcients for representative
and senator belonging to the same party fails to reject the null that both coeﬃcients are zero
(p-value = 0.33 without controls, 0.56 with controls). Moving from the Republican district to a
Democratic district still causes an estimated decline in construction spending. The decline does

32Recall that in Ohio, each Senate district is composed of three contiguous House districts.

26

vary substantially, however, depending on whether the senator and representative for an area belong
to the same party. The estimates in Column 6 suggest that areas switching from a Republican to a
Democratic representative receive $5.7M ($3.4M + $2.3M) less in construction spending when the
senator is a Republican, and $1.1M ($3.4M - $2.3M) less when the senator is a Democrat. Adding
controls for demand while controlling for alignment between House party and Senate party causes
the standard errors to increase. Still, the estimates in Column 7 show that areas moving from a
Republican Senate and House district to a Democratic House district are estimated to lose $4.6M
($2.3M + $2.3M), with this estimate marginally diﬀerent from zero in a statistical sense (p = 0.08).

2.5.6 Number of Construction Projects

The prior speciﬁcations consider construction expenditure as a continuous variable distributed
across areas of the state. Total construction expenditure, however, is allocated to a certain number
of projects. An area may receive less funding because the state undertakes fewer projects in the
area. On the other hand, the state may undertake the same number of projects while reducing the
average funding devoted to each project. I now consider whether change in representative party
aﬃliation aﬀects number of projects and number of large projects. Large projects include new
construction, intersection/interchange, widening, other projects, and bridges and tunnels - the six
categories with largest average project cost (see Table 2.8 of the appendix).33 A state representative
may be more able to inﬂuence allocation of large projects than allocation of projects in general.
Lobbying or negotiating for a new construction project yields a higher payoﬀ than lobbying or
negotiating for maintenance work or landscaping. In addition, large projects take longer to develop
between planning and the beginning of construction.

I estimate the eﬀect of redistricting party change on number of projects and number of large
33Given the closeness in mean cost between the bridges and tunnels category and the maintenance category, it may
seem arbitrary to count one as a large project and not the other. Work on bridges and tunnels, however, involves work
at a speciﬁc point and only crosses district boundaries when there are multiple work areas within a single project. On
the other hand, maintenance often extends over a stretch of roadway and thus more often crosses district boundaries.
There are far more bridge and tunnel projects than maintenance projects (2186 vs. 1255 - see Table 2.8). But there are
fewer bridge and tunnel projects per year in each area of intersection, compared to maintenance (1.8 vs. 2.3).

27

projects using ﬁxed-eﬀects Poisson QMLE. The base speciﬁcation includes indicators for the (R
to D), (D to R), and (D to D) party change categories, each interacted with a pre-2011 and a
post-2012 indicator for each category, as well as area of intersection and year ﬁxed eﬀects. The
second speciﬁcation includes controls. The third speciﬁcation includes controls and a linear year
trend for each party change category (R to R, R to D, etc.). Each speciﬁcation is estimated twice,
with two dependent variables: total number of projects, and total number of large projects.

Table 2.5 indicates that when the dependent variable is number of projects, the estimates
show no diﬀerence between areas that remained within Republican districts and areas that moved
to Democratic districts. However, areas moving from Republican to Democratic districts do
experience a sharp decline in the number of large projects. Depending on the set of covariates,
areas in Republican districts before and after redistricting received from 43% to 107% more large
projects than areas that move to a Democratic district. Clearly, a decrease in the number of large
projects drives the decline in total construction expenditure for areas that moved from Republican
to Democratic districts. The results suggest that while partisan politics do not aﬀect mundane
maintenance decisions, majority party legislators and the governor’s oﬃce inﬂuence the location
of large projects.

Areas moving from the Democratic to the Republican party receive more projects in general,
but fewer large projects, than areas that belonged to Democratic districts before and after redis-
tricting. However, these diﬀerences diminish as controls are added. As in the previous section’s
speciﬁcations, any diﬀerence between areas that left Democratic districts and areas that remained
Republican is much smaller than the diﬀerence between areas that left Republican districts and
areas that remained.

2.6 Variation from Voting and Redistricting

Party changes through election outcomes may aﬀect distribution of transportation funding dif-
ferently than party changes driven by redistricting, for two reasons. First, there may be selection
issues associated with variation based on voting - election outcomes may reﬂect changes in unob-

28

servable determinants of transportation funding. Second, party changes through election outcomes
occur in more closely contested districts. Identifying party changes through redistricting avoids
these problems. In the next set of estimates, I provide descriptive evidence that selection issues
bias the estimated eﬀects when party changes result from election outcomes.

In Section 2.5, I estimated the eﬀect of having a Republican representative when Republicans
held majorities in the House and Senate under a Republican governor. Now, I compare the eﬀect of
having a Republican representative in years with a Democratic governor and years with a Repub-
lican governor. Each area of intersection by year observation falls within one of four categories:
Republican representative and Republican governor, Republican representative and Democratic
governor, Democratic representative and Republican governor, or Democratic representative and
Republican governor. I estimate diﬀerences in expenditure levels within each of these categories
using the following speciﬁcation with area and year ﬁxed eﬀects.

expenditureit = β0 + β1R_rep_R_govit + β2D_rep_R_govit + β3R_rep_D_govit

+ γi + δt + errorit

Coeﬃcients are identiﬁed using variation in representative’s party through redistricting and through
election outcomes. If expenditures are higher in areas represented by the party that controls the
House and Senate, with no advantage to areas represented by members of the governor’s party,
then I would expect β1 = β3 > 0 and β2 = 0. If expenditures are higher in areas represented
members of the governor’s party but not for areas represented by the party controlling the House
and Senate, then I would expect β2 = β3 < 0 and β1 = 0.

The estimated coeﬃcients, in Table 2.6, demonstrate an advantage for areas with a Republican
representative - these areas gain $1.2M more in funding during years with a Republican governor,
and $1.0M more in years with a Democratic governor, relative to areas with a Democratic repre-
sentative. After adding controls in the second column, the estimated diﬀerence between districts
with a Republican representative and a Democratic representative increases to $1.4M, while the

29

advantage for Republican districts during Democratic governor years rises to $1.3M. In neither
case, however, can I reject the null hypothesis that areas with Republican representatives attract
more transportation funding than areas with Democratic representatives - the p-value is 0.20 in
the regression without controls, and 0.22 in the regression with controls. There is no evidence
that areas with a representative belonging to the governor’s party receive more road construction
expenditure when using variation through both election outcomes and redistricting.34

The next two regressions separate the eﬀect of party changes due to election variation from
the eﬀect of redistricting party changes. Recall that the ﬁrst set of regressions in Table 2.4
compared spending during the six years after redistricting to spending in the two years before.
Here, I combine the redistricting party change indicators from those regressions with the party and
governor alignment indicators. 35

Column 3 displays the estimates when controls are not included. Areas that changed from a
Republican to a Democratic representative (or from a Democratic to a Republican representative)
during a regular election year lost (gained) $1.2M ($2.98M - $1.83M) annual construction expen-
diture during years with a Republican governor, and $1.0M in years with a Democratic governor.
Column 4 includes controls (and drops years 2016-2018). Including controls, I estimate that areas
moving from the Republican to the Democratic party (or the Democratic to the Republican party)
lose (gain) $1.7M ($5.35M - $3.64M) in years when there is a Republican governor, and $1.4M in
years when the governor is a Democrat. Column 3 shows that areas changing from a Republican
to a Democratic representative through redistricting lost $1.84M + $2.98M - $1.83M = $2.99M in
annual construction expenditure (p-value = 0.03). During the redistricting year, of course, a Re-
publican governor held oﬃce. Similarly, Column 4 shows that areas moving from the Republican
34These estimates suggest little advantage in belonging to the governor’s party. The variation in governor’s party,
however, derives from Democratic governor Ted Strickland’s term from 2007-2010, which followed sixteen years
of Republican governors.
It is possible that during his single term Governor Strickland did not possess the same
political clout as his predecessors or successors. From Section 2.3, for example, recall that the Transportation Review
Advisory Council approves the largest projects (those with value exceeding $12 million dollars). Although the TRAC
is largely appointed by the governor (six seats out of nine), members are appointed in overlapping terms of seven years
- Democratic Governor Taft appointed only two members to this committee during his term.
35The pre-2011 party change indicators, e.g. (R to D)×(year < 2011), are not included because I am controlling for

party and alignment with the governor in each year.

30

to the Democratic party due to redistricting lost $3.1M ($1.41M + $5.35M - $3.64M) in annual
funding (p-value = 0.06).

Thus, the change in expenditure associated with a party change through redistricting diﬀers
from the change in expenditure associated with a party change through elections by millions of
dollars. The estimates are somewhat imprecise - a joint test of the coeﬃcient on (R to D)×(year >
2012) and the diﬀerence between the coeﬃcients on (D to R)×(year > 2012) and (D to D)×(year >
2012) gives a p-value of 0.12 in the equation without controls and 0.16 when controls are included.
Still, the large coeﬃcients on the redistricting change indicators strongly suggest that estimates
identiﬁed through election results are biased.36

Outside of the redistricting year, areas changed parties in two ways. Either the incumbent
representative lost the election, or the incumbent representative chose not to run and her party’s
candidate failed to win the general election.37,38 The last set of regressions, in Columns 5 and
6, include an indicator for areas where the representative was in her ﬁrst term and the previous
36I considered the possibility that the diﬀerence in coeﬃcients reﬂects asymmetry between the eﬀect of changes
from R to D and D to R in non-redistricting years. If areas moving from R to D in non-redistricting years experience
a large decrease in funding while areas moving from D to R change very little, then the overall positive eﬀect of
Republican representation would be small. I estimated the following equation:

(β1R_to_D_redisti + β2D_to_R_redisti + β3D_to_D_redisti) × (yeart > 2012)

+ β4R_to_Dit + β5D_to_Rit + β6D_to_Dit + δt + γi + it

expenditureit = β0 +

Here, R_to_Dit indicates that area i changed from a Republican representative in the previous election cycle to a
Democratic representative in any year t from 2007-2018 - similarly for D_to_Rit and D_to_Dit. R_to_D_redisti
indicates party change during the redistricting year, the same as R_to_Di in the previous speciﬁcations. For the sake of
brevity I do not show the results here. The coeﬃcients on the R_to_D_redisti × (year > 2012), D_to_R_redisti ×
(year > 2012), and D_to_D_redisti × (year > 2012) are almost identical to the coeﬃcient estimates in Columns
3 and 4. Again, I test whether the coeﬃcients on R_to_D_redisti × (year > 2012) and the diﬀerence between the
coeﬃcients on D_to_R_redisti × (year > 2012) and D_to_D_redisti × (year > 2012) are jointly signiﬁcant.
In a regression without covariates, the joint test gives a p-value of 0.08, and the R_to_D_redisti × (year > 2012)
coeﬃcient is marginally signiﬁcant on its own, with a p-value of 0.06. With the addition of covariates (and the loss
of three years) the p-value for the joint test increases to 0.13. Again, the estimates yield suggestive evidence that the
coeﬃcients identiﬁed from party changes through voting diﬀer from the coeﬃcients identiﬁed from party changes
through redistricting.

37In 41 year-by-district observations (or 110 year-by-area of intersection observations) the representative left oﬃce
during the term (usually after an appointment to a preferable oﬃce, due to a scandal, or due to unexpected death).
There were only three year-by-district observations where the appointed replacement ran in the general election and
lost. In these regressions, I have considered the replacement representative as the incumbent.

38The third possibility - the incumbent losing the primary and her replacement losing the general election - never
occurred. Ten incumbents lost the primary election between 2007 and 2018, but each candidate who beat the incumbent
in the primary won the general election.

31

incumbent lost. I interact this variable with an indicator for the representative belonging to the
same party as the governor. This means that the ﬁrst three variables are identiﬁed from years when
a representative retired after her ﬁrst term and a member of the opposing party replaced her. In
years after an incumbent loses, her district suﬀers a large decrease in funding, particularly if the
incumbent had belonged to the governor’s party. Under a Republican governor, areas where an
incumbent Republican representative lost suﬀered $8.54M ($7.09M + $1.65M + $5.59M - $5.79M)
less road construction expenditure (p-value = 0.05). If, on the other hand, an incumbent from the
Republican governor’s party retired and a non-incumbent of the opposite party replaced her, her
district would lose only $1.45M ($1.65M + $5.59M - $5.79M) in expenditure.39 The diﬀerence
could reﬂect the eﬀect of expenditure trends on partisan preferences, instead of the other way
around. If an incumbent of the governor’s party suﬀers defeat, this likely means her constituents
are less satisﬁed with the governor’s party in that year than in previous years. Dissatisfaction
with the governing party may stem from poorer public service provision. The lower level of
public services provided likely causes the incumbent legislator’s defeat, not the other way around.
As an alternative explanation, districts that reject an incumbent representative have experienced
particularly stark changes in preferences, and these changes reﬂect stark changes in economic and
demographic determinants of road expenditure.

2.7 Incumbent Representatives and Redistricting Party Change

According to the results in Section 2.5 (using variation through redistricting), shifting from a
Republican district to a Democratic district creates a large drop in predicted construction funding.
But shifting from a Democratic to a Republican district generates no corresponding increase in
funding. As discussed at the end of Section 2.5.2, the asymmetry may reﬂect an incumbent
advantage - areas that change from a Democratic to a Republican district gain a party aﬃliation
boost, but lose their incumbent representative. The last two columns of Table 2.6 suggested a strong
39In years with a Democratic governor, areas where an incumbent Democrat lost would only receive $6.46M
($7.09M + $1.65M - $2.28M) less funding, according to the regression without controls. But areas where the
incumbent Democrat retired and a Republican replaced her would actually gain $0.63M.

32

incumbent advantage, based on variation through both elections and redistricting. In this section,
I use redistricting variation to test whether the party change disadvantage estimated in Section
2.5 partly reﬂects the advantage of retaining the same representative and belonging to the same
district post-redistricting. Each area that changed party from 2012-2013 necessarily switched to a
representative who had not represented that area in the past. Many areas switched representative but
did not switch party, however. Recall from Section 2.3 that 185 areas switched to a non-incumbent
representative from 2012-2013 - of these, only 56 changed party.

For several reasons, an incumbent representative whose district boundaries have changed may
reward the old area of her district relative to the new area. First, the incumbent representative
possesses greater knowledge of the old area’s needs. Second, it is likely quicker to postpone
old projects than to begin new projects. If an area that moves from a Democratic district for a
Republican district is less likely to suﬀer postponement of old projects and more likely to obtain
new projects, then in the short term it will gain no funding increase, because new projects take time
to develop.

Table 2.7 shows the results of reestimating the ﬁrst ﬁve columns of Table 2.4, adding two
variables: an indicator for areas where the representative was Republican before and after redis-
tricting, but the representative changed, and areas where the representative was a Democrat before
and after redistricting, but the representative changed. In Column 3, I also add trends for R to
R and D to D districts that retained the same representative, while Columns 4 and 5 use Poisson
QMLE to estimate the percentage changes in dependent variable associated with a change in party
or representative.

The results give strong indication that incumbent representatives attracted more funding that
new representatives. According to the speciﬁcation in Column 1, areas that had belonged to the
Republican party and retained the same representative attracted more funding than areas that moved
from one Republican representative to another (the omitted category). R to D areas experienced a
much smaller increase in funding when compared to R to R areas that changed representative than
when compared to all R to R areas. On the other hand, D to R areas experienced a post-redistricting

33

increase in funding relative to D to D areas that changed representative ranging from $1.6M in the
ﬁrst column to $8.1M in the third column with controls and time trends. Thus, after controlling for
whether areas moved to a new incumbent representative, the eﬀect of moving from a Democratic to
a Republican district is estimated to be large and positive, relative to the speciﬁcation that compared
D to R to all D to D.

According to the regressions based on redistricting variation in Section 2.5, an increase in
spending for areas that moved to Republican districts did not accompany the estimated drop in
spending for areas that left the Republican party. The results in Table 2.7 suggests that areas
moving to a new district experience a decrease in funding, regardless of party. After controlling for
this factor, R to D areas experience an increase in funding that equals or even exceeds the decrease
in funding for areas that move from the Republican to the Democratic party.

2.8 Conclusion

Policymakers allocate road construction expenditure across regions based not only on concerns
for eﬃciency or redistribution, but perhaps also according to partisan preferences.
I show that
areas with a state representative belonging to the majority Republican party draw substantially
more road construction funding than areas with a Democratic representative. I estimate an eﬀect
of representative’s party on spending comparable to or larger than estimates of party alignment on
nationwide spending distribution in other studies, such as in Spain [Solé-Ollé and Sorribas-Navarro,
2008] and Italy [Bracco et al., 2015]. My estimates, moreover, exceed estimates of Berry, Burder,
and Howell [2010] and Albouy [2013] regarding the eﬀect of congressional representation on federal
spending in the United States. I ﬁnd that areas moving from a Republican to a Democratic district
receive 66% less expenditure in years where the assembly and governorship were Republican, while
these studies found from 5% increase for districts aligned with the president to 31% increase for
states whose entire congressional delegation belonged to the majority party.

While previous studies used variation in partisan alignment through election outcomes, changes
in voter preferences are likely correlated with unobservable shifts in need or preferences for roads.

34

In addition, voters may choose a representative based on their expectation that she can direct public
expenditure toward their district. For these reasons, endogeneity of party changes through elections
likely leads to biased estimates. Unlike most previous papers, I avoid this endogeneity issue by
focusing on party changes generated by the redistricting process. Besides concerns about bias,
estimates identiﬁed using party changes through elections heavily weight districts where it is most
likely for a party change to occur. Thus estimates identiﬁed using voting variation tend to estimate
eﬀects for swing districts. It is unlikely that the majority party would target expenditure equally
between safe and swing districts [Case, 2001, Dixit and Londregan, 1998]. Because the variation
that I use derives from boundary changes rather than voting, the eﬀect that I identify more closely
represents the average over the entire distribution of voter preferences, instead of the average over
swing districts only.

The large eﬀect estimated in this paper probably also reﬂects the institutional setting considered.
At the federal level, each state has two senators who may belong to diﬀerent parties. Although
each federal representative is accountable to her own sub-state district, the prevalence of state-level
block grants provides an incentive for state representatives to work across party lines in bringing
funds back to the state. At the state level, however, one party represents each district in the House
of Representatives. Because state house districts are geographically small and the entire population
beneﬁts from transportation expenditure, each representative faces strong incentives to direct funds
toward her own district. I also provide evidence that areas with a representative and senator in the
same party receive substantially more transportation expenditure - this may indicate coordination
of eﬀorts, or a willingness of other Republican policymakers to aid districts where their own party
will receive all of the credit. Areas represented by the majority party reap substantial beneﬁts,
particularly if the same party controls the House and overlapping Senate district. Areas belonging
to the minority party suﬀer.

The empirical analysis shows that areas moving from a district represented by the governing
Republican party to a district represented by the minority Democratic party suﬀer a large decline in
construction project funding. The eﬀect reﬂects a diminished number of major construction projects

35

directed toward these areas. Although partisan concerns do not enter into mundane construction
decisions like maintenance and landscaping, the results imply that legislators in the majority party
are able to reward their constituents by directing large projects toward their own districts.

I also estimate the change in expenditure associated with a change in representative’s party
through election results in non-redistricting years. The estimates show that the manner in which
districts change through election results determines the change in expenditure. Areas where an
incumbent representative from the governor’s party is voted out of oﬃce suﬀer a much larger
expenditure decline than areas where party changes when the incumbent does not run. This is
consistent with a non-random selection of districts changing parties through election results. More
importantly, estimates of the change in expenditure associated with a party change through voting
diﬀer from estimates based on redistricting. This diﬀerence may derive from endogeneity of
representative party changes due to election results.
It may also reﬂect that districts changing
party through voting form a selected subsample of swing districts. The comparison conﬁrms the
importance of isolating an exogenous source of variation in political party to estimate these eﬀects.
This paper advances the literature by applying such a credible identiﬁcation strategy to this question.

36

CHAPTER 3

MUNICIPAL GOVERNMENT REACTION TO MASS LAYOFFS IN OHIO

3.1 Introduction

3.1.1 Municipal Reaction to Mass Layoﬀs

On October 17, 2008, General Motors closed its assembly plant in Moraine, Ohio, costing the city
1150 jobs. Between this plant closing, GM’s layoﬀ of 960 second-shift employees in September,
and layoﬀs at two smaller auto manufacturing plants, Moraine lost 2553 jobs in 2008. This number
equaled 41% of the city’s population. Beyond the hardship suﬀered by residents employed at the
plant and at businesses serving it, the closing depleted the city’s ability to provide public services.
Moraine’s 2% income tax raised $19.9 million of revenue in 2007 - in 2009, it raised only $12.4
million. Property tax revenue fell from $772,000 in 2007 to $474,000 in 2009. Despite the drop in
revenue, the city did not raise tax rates until 2013, when it tripled its property tax rate before hiking
its income tax rate in 2014. Both increases required approval in referendum elections, asking that
residents relinquish a larger share of their diminished incomes in tax. Moraine also cut expenses.
Total annual expenditure averaged $27.4 million from 2005-2008, but fell to $19.9 million annually
from 2009-2013. Capital expenditure and expenditure on leisure and community environment
(such as parks and recreation) dropped by more than one-half. The city also increased its debt over
this time period, from $5.3 million in 2009 to $11.8 million in 2013. At the same time, the city ran
down its available assets - government fund cash and investments halved from $24 million to $12
million.

Sudden declines in economic activity threaten local government ﬁnances more than national
or state government budgets. Due to greater volatility of taxable activity at the local level, local
governments must particularly concern themselves with revenue variability. In Ohio, a state hit

37

hard by the Great Recession, state income tax base fell 8.3% between the 2008 and 2009 tax years.1
But many municipalities imposing an income tax suﬀered a more severe decline. Income tax base
dropped by 43% in Leipsic, a village raising most of its revenue through a local income tax. Leipsic
was one of 20 municipalities where taxable income fell more than 25%. Meanwhile, income tax
base rose at least 10% in 23 municipalities. Ohio municipalities also raised funds through the
property tax - assessed property value fell more gradually than income tax base, but the decline
persisted for longer. Between 2008 and 2009, assessed real property value fell 1.9% across Ohio.
By 2013, assessed real property value had fallen 9.8% from its high in 2008, although the annual
decline never exceeded 4%. Relative to the income tax, the property tax ﬂuctuates less year by year.
Local governments’ responses to tax base declines, either by increasing taxes or by shifting
to other revenue sources, are limited by state laws restricting the type of tax imposed and the
tax rate. In Ohio, state law prohibits municipalities from imposing general sales taxes, although
municipalities may impose property and income taxes. Residents must approve any increase in the
property or income tax rate beyond 1% through a referendum election. These limitations protect
residents from unexpected tax increases, but also impair municipalities’ ability to adjust revenue
collection to changes in economic conditions. If a municipality cannot raise revenue to its previous
level, it must either cut expenditures, borrow, or deplete any saved funds it may have accumulated.
I consider the ﬁscal policy response of municipal governments in Ohio to an important shock:
shutdown, or contraction in employment, by local businesses. I identify these shocks using notices
of closings and mass layoﬀs given under the Worker Advisory and Retraining Notiﬁcation (WARN)
Act. From 1997-2013, 1664 closings and layoﬀs caused the loss of 321,077 jobs in the 595 Ohio
municipalities I consider. Mass layoﬀs decreased municipal revenue starting in the year of the
layoﬀ and continuing for several years afterward. The revenue drop was often unexpected - in most
cases the municipal government was not notiﬁed until a few months before a layoﬀ occurred. In
addition to estimating the size and duration of the revenue decline associated with a mass layoﬀ,
I investigate municipalities’ ﬁscal policy response. In doing so, I provide a positive analysis of

1Average municipal income tax base fell 9.2%.

38

municipal decision making when the trade-oﬀs associated with these decisions may be particularly
severe.

To discover the eﬀect of mass layoﬀs on ﬁscal outcomes, I model each outcome as a function of
reported job loss in mass layoﬀs, scaled to municipal population (“proportional job loss”). Using
an event study framework, I include lags of reported job loss in past layoﬀs to estimate a layoﬀ’s
dynamic eﬀect over several years.
I estimate that an announced mass layoﬀ with proportional
job loss equal to 1% of municipal population results in a 0.75% decline in income tax base from
the year before the layoﬀ to the year after. This means that a mass layoﬀ of average size, with
proportional job loss equal to 3.2% of the population, will decrease income tax base by 2.4%. In
subsequent years, the income tax base does not recover to the pre-layoﬀ level. I ﬁnd that assessed
property value does not decrease signiﬁcantly after a mass layoﬀ occurs. This is likely because
assessed property value reﬂects sale prices only after a triennial appraisal process. Because an
appraised increase in value is phased in proportionally over three years, it takes as long as six years
for assessed property value to reﬂect changes in sale price.

I ﬁnd that municipalities do not change their income tax rate signiﬁcantly in response to a mass
layoﬀ. In fact, income and property tax rates decline slightly. As the income tax base falls with no
responding increase in tax rates, I ﬁnd that revenue declines. Income tax revenue falls, remaining
below the pre-layoﬀ level four years after the layoﬀ. Due largely to the drop in income tax base,
total revenue decreases after a reported mass layoﬀ, and remains below the pre-layoﬀ level four
years following.

Total municipal expenditure declines following a mass layoﬀ, and also recovers slowly. Con-
sistent with previous literature, expenditure on general government, leisure and community envi-
ronment, and capital outlay all decrease. Somewhat unexpectedly, public safety expenditure also
declines sharply, perhaps due to a decrease in business demand for those services, or because expen-
ditures in this category are more easily adjusted. In addition to cutting expenditures, municipalities
shift their structure of liabilities and decrease outstanding fund balances. For cities, where data
on assets and liabilities are available, long term debt increases while short term debt decreases.

39

Unreserved fund balance, moreover, decreases substantially, and remains below the pre-layoﬀ level
at least four years after a mass layoﬀ occurs. Thus, cities avoid deeper cuts to present expenditure,
at the cost of a more precarious ﬁnancial position in the future.

3.1.2 Related Literature

A number of papers address local governments’ reaction to ﬁscal distress. Poterba [1994] studies
state governments’ response to unprojected decreases in revenue and increases in expenditure,
but focuses primarily on how budget institutions and the characteristics of elected oﬃcials aﬀect
adjustment to equilibrium. In an analysis of Danish municipalities similar to Poterba, Rattsø and
Tovmo [2002] ﬁnd that property tax rates respond inversely to unexpected increases or decreases
in revenue, while the income tax rate only responds to negative shocks. Skidmore and Scorsone
[2011] analyze the expenditure decisions of Michigan cities undergoing ﬁscal stress.2 Certain
categories of expenditure, speciﬁcally capital outlay and parks and recreation, proved vulnerable
to cuts.

Other papers address how local governments respond to speciﬁc causes of revenue decline.
Cromwell and Ihlanfeldt [2015] ﬁnd that local governments in Florida increased millage rates after
decreases in property values and state aid during the great recession. Dye and Andrew Reschovsky
[2008] discuss the response of property tax revenue per capita to a change in state aid to all levels
of local government. Among local governments in general, there was no statistically signiﬁcant
relationship between change in state aid and change in property tax revenue per capita.3 Chernick,
Langley, and Reschovsky [2011] ask how revenue and expenditure changed in large cities during
the Great Recession, using “constructed governments” that aggregated the tax and expenditure of
all local governments in each area. They ﬁnd that property tax revenue ﬂuctuated less than local
income or sales tax revenue during this period. Lutz, Molloy, and Shan [2011] investigate several
2They measure ﬁscal stress by an index of the diﬃculty of raising revenue minus an index of the cost of providing

a bundle of services.

reforms in some states.

3There was a large, statistically signiﬁcant negative relationship for school districts, likely driven by school ﬁnance

40

ways a decrease in housing prices could aﬀect local revenues, including a decrease in property tax,
sales tax, income tax, or conveyance taxes on property transfers. Testing these channels during
the great recession, they ﬁnd that only property tax revenue decreased. Lutz [2008] estimates the
elasticity of property tax revenue with respect to property values, using Census data for a panel of
MSAs. He ﬁnds the elasticity is highest after 3 years, at 0.4.

A few papers consider the ﬁscal impact of a boom or bust within a particular industry. Feler
and Senses [2017] estimate the response of local government expenditures to import competition,
although they aggregate all local governments at the commuting district level. Wu and Korman
[1987] regress changes in property value, property tax revenue, and non-property tax revenue
on changes in manufacturing employment in New York State counties. They ﬁnd that each
manufacturing job lost was associated with a decline in property values and property tax revenue,
but do not consider government response to the revenue decline. Finally, Newell and Raimi [2015]
investigate how discovery of new shale oil and gas resources aﬀected local government ﬁnances in
eight states. Although discovery of fossil fuel resources boosted property and sales tax base, it also
expanded the cost of government service provision (e.g. infrastructure expenditure).

My paper advances the literature in four ways. First, while others have studied the eﬀect of
an ailing local economy on municipal revenue, none have considered mass layoﬀ events directly.
Second, papers about a declining economy’s eﬀect on local governments have either ignored local
government response, or have considered the aggregate response by various levels of government
across states.4 I consider the response of one level of government, where each municipality is a
decision making unit responding to a layoﬀ event under the same institutional constraints. Third,
I consider a wider range of policy responses by the local government, including tax rate changes,
shifts to other revenue sources, expenditure cuts in the aggregate or within speciﬁc categories, and
debt ﬁnance. Finally, I consider the case of a government with two tax instruments, an income and
property tax. I estimate the eﬀect of a mass layoﬀ on each tax base, as well as the government’s
potential decision to shift between revenue bases.

4Skidmore and Scorsone [2011] is an exception, but the authors consider only expenditure response while I consider

changes in tax rates, shifts to non-tax revenue sources, borrowing, and change in fund balance.

41

3.2 Institutions and Data

3.2.1 Ohio Municipalities

Ohio’s 932 municipal governments provide basic government services, including ﬁre, EMS, and
police protection; roads, sewers, and other infrastructure; and amenities like parks, swimming
pools, and cemeteries.5 Municipalities raise revenue from various sources, including income
and property taxes. The income tax is imposed, at the same rate, on income earned within the
municipality and on residents’ income earned outside the municipality. Wages, rental income, and
net business proﬁts attributable to activity in the jurisdiction are taxed.6,7

A municipality that levies a property tax imposes the tax on both commercial and residential
property at the same statutory rate.8 Property is appraised every three years, with one-third of the
change in value applied in each of the three years following an appraisal - the assessed value equals
35% of appraised value.9 Assessed value changes more rapidly if the property is demolished or
sold for lower than its appraised value. Voted property taxes are a levy on value at the time the tax
is approved. An increase beyond that value lowers the eﬀective tax rate, rather than raising revenue.
This cap is applied separately to Class 1 (residential and agricultural) and Class 2 (commercial
and industrial) property. Each year, the county assessor adjusts property tax rates so the revenue
5Municipalities with population less than 5000 are referred to as “villages,” while those with population greater
than 5000 are referred to as “cities.” Of the 932 incorporated municipalities in Ohio, 251 are currently cities, while the
rest are villages.

6In many cases a resident may be granted a credit against income earned in other jurisdictions. Suppose that there
are two jurisdictions. In jurisdiction A the tax rate is 2%, in jurisdiction B, 4%. If each jurisdiction allows a 100%
credit on income tax paid in the jurisdiction where workers are employed, then residents of jurisdiction A commuting
to jurisdiction B will pay no income tax in jurisdiction A, while residents of jurisdiction B that commute to jurisdiction
A would pay 2% income tax in jurisdiction A and 2% in jurisdiction B. If only 50% credit were granted, then residents
of jurisdiction A commuting to jurisdiction B still only pay taxes to jurisdiction B, while residents of jurisdiction B
commuting to jurisdiction A will be “double taxed” paying 2% in jurisdiction A and 3% in jurisdiction B. Credits
allowed vary across municipalities.

7Other sources of revenue include fees and ﬁnes, payment for services by residents or other governments, and
intergovernmental revenue. Municipalities are allowed to impose a lodging tax and an admissions tax on events. Each
comprises a negligible fraction of revenue: in 2013 202 municipalities collected $59.8 million total through lodging
taxes, while 46 municipalities collected $26.3 million in admissions tax, $14.4 million by Cleveland.

8Although the municipality cannot choose diﬀerent rates, state property tax relief measures mean that the eﬀective

tax rate on residential property is typically lower.

9An exception applies to agricultural land, which may be evaluated according to its current use rather than its

appraised market value.

42

collected from each class of real property will not exceed the amount levied when the rate was
approved. If Class 2 value decreases while Class 1 value increases, or vice versa, revenue could
fall in a municipality while overall assessed property values increased. The revenue cap does not
apply to taxes within the 10 mill limit, or any higher limit approved by a charter.10

Three other local governments impose property taxes in Ohio:

local and vocational school
districts, counties, and townships.11 The total property tax burden in a municipality depends upon
the taxes imposed by these other three governments. For each municipality in the sample, I calculate
the sum of tax rates imposed by all overlapping jurisdictions. In cases where a municipality spans
several school districts or counties, the school district or county rate is calculated as the average
property tax rate of those jurisdictions weighted by the share of the municipality’s real property
lying in each overlapping jurisdiction.12

The data comprise a balanced panel of 595 municipalities in Ohio from 1997-2013, including
gross income tax revenue, real and tangible personal property value, property and income tax rates,
and all WARN notices issued in Ohio during this time. Revenue and expenditure by category are
available from 1999-2013, while employment is available from 2002-2013. Finally, for 181 cities,
data on outstanding debt, cash and investments, and unreserved fund balance were available from
2002-2013.13 I provide summary statistics for income tax base, property tax, population, and other
10Until 2009, property tax was levied on the tangible personal property of manufacturers. Tangible personal
property included equipment and machinery, inventories, and ﬁxtures and furnishings. The tax was imposed at the
same statutory rate as the tax on real property, although none of the discount factors applied to real property applied to
tangible personal property. Tangible personal property composed a small share of property value, in my sample 10%
of real property across all municipalities, higher in industrialized areas. From 2005-2009, Ohio phased out the tax on
tangible personal property, reducing the rate of appraised value to assessed value 6.25 percentage points per year from
the initial rate of 25%.

11Each municipal jurisdiction overlaps one or more county governments, and one or more school districts; in
suburban or rural areas villages may overlap with township governments. Because local school districts, counties,
townships, and municipalities all impose a property tax, the ten unvoted mills are typically allocated between these three
levels of government. The municipality may be allowed to impose three or four “inside” mills of property tax without
referendum approval, while the remainder of the inside millage is allocated to the school district, county, and township
governments. As seen in the second panel of Table 3.1, the property tax rate imposed by municipalities generally
accounts for a small proportion of the total property tax imposed by all local governments within the boundaries of the
municipality (7.11

54.5 = 13.0%).

12Although school districts may impose an income tax, I consider only the municipal income tax rate in determining
tax burden. The municipal income tax rate is usually larger than the school rate, and data on the school income tax rate
is not available in all years.

13Data on tax rates and tax base were collected by the Ohio Department of Taxation [Ohio Department of Taxation,

43

municipal variables in Table 3.1. The majority of municipalities raised revenue through an income
tax. All municipalities raised some revenue through the property tax (only one municipality-by-year
observations shows zero property tax revenue), but the income tax composed the largest source of
revenue during this period. The municipal income tax raised on average $58.9 million per year,
more than twice as much as the property tax at $24.3 million.14

3.2.2 WARN Act Summary

The 1988 Worker Adjustment and Retraining Notiﬁcation (WARN) Act requires employers of 100
workers or more to disclose site closures and mass layoﬀs 60 days in advance. The employer must
notify employees, labor unions, local elected oﬃcials, and the state rapid response department.15
There are four thresholds that trigger mandatory WARN notiﬁcation: shutdown of a employment
site with employment loss for more than 50 employees within 30 days; a mass layoﬀ causing
employment loss for more than 500 workers, again within 30 days; an employer with more than
150 workers laying oﬀ more than 33% of its workforce; or multiple sites satisfying any of these
thresholds jointly during a 90 day period.16

The Ohio Department of Job and Family Services provides data on plant closings from 1996-

“Tax Data Series”]. Employment data comes from the Census Bureau’s Longitudinal Employer-Housing Dynamics
project [United States Census Bureau, “LEHD Origin Destination Employment Statistics”]. Revenue, expenditures,
and debt were collected from both audit reports of Ohio municipalities by the Ohio Auditor’s Oﬃce, and from annual
ﬁnancial reports submitted to the Ohio Auditor’s Oﬃce by municipalities. Unfortunately, data on revenues and
expenditure were missing for some municipalities. I have restricted the sample to the 595 municipalities for whom
these data was available in every year from 1999-2013.

14Tax revenue variables and income tax base are tabulated in the year collected. Income tax collections reﬂect

economic activity in the previous year.

15Information in this section is from United States Department of Labor, “The Worker Adjustment and Retraining

Act: Revising the Act and Educational Materials Could Clarify Employer Responsibilities and Employee Rights.”

16To qualify as an employee under these thresholds, a person must work at least 20 hours per week for six months
out of the year. Employment loss includes those who lose at least half of their hours or those who are laid oﬀ for more
than six months. Employment loss does not include those who were transfered or refused a transfer within “reasonable
commuting distance.” An employer who fails to ﬁle a WARN notice must give workers back pay and beneﬁts as well
as up to $500 to the local government for each day of violation. This requirement can only be enforced in court by the
aggrieved party (who may have their legal fees reimbursed). This penalty and enforcement structure may not be severe
enough to ensure compliance. A 2003 Government Accounting Oﬃce Report found that only 33% of 2001 mass layoﬀ
events triggering the WARN threshold could be matched to a WARN notice; however, this study could only match 13%
of WARN notices to mass layoﬀs in BLS data [United States Government Accounting Oﬃce, 2003].

44

2015. These data indicate the date of the notice, the date of the plant closing or mass layoﬀ,17 the
number of workers aﬀected, and the name of the county, municipality, and ﬁrm. From 1997-2013,
978 municipality by year observations had at least one reported closing. Table 3.1 shows summary
statistics for plant closings.18 At least one WARN notice occurred in 9.7% of all municipality-year
observations. In municipalities where at least one mass layoﬀ occurred during the year, 325 jobs
were lost on average.

Because the unit of observation is the municipality, the key regressor will be job loss scaled to
municipal population, or “proportional job loss.” Rescaling job loss to be a fraction of municipal
population gives a better picture of the layoﬀ’s importance in the municipality where it occurs. In
municipalities where a mass layoﬀ occurred, job loss in mass layoﬀs averaged 3.2% of municipal
population. Figure 3.1 illustrates the distribution of this variable across incorporated municipalities
in 2009, the sample year when the largest number of layoﬀs occurred. Job loss scaled to population
was largest either in smaller municipalities located far from any major metropolis, or in smaller
municipalities on the outskirts of large urban areas.

3.3 Eﬀect of a Mass Layoﬀ on Tax Base

3.3.1

Income Tax Base

A mass layoﬀ may occur multiple times for each municipality, with each layoﬀ aﬀecting ﬁscal
outcomes for several years. Thus, I model each ﬁscal outcome (income and property tax base,
income and property tax rates, revenue and expenditure, outstanding debt and asset balances) as a
function of a series of proportional job loss lags. Proportional job loss is deﬁned as number of jobs
lost in a reported mass layoﬀ divided by municipal population. Failing to control for past layoﬀs
would decrease the precision of the estimates, because the eﬀect of each layoﬀ would be conﬂated
with the eﬀects of past layoﬀs. More importantly, because the probability of a layoﬀ occurring in
17I do not observe in the data whether these notiﬁcations apply to plant closings or mass layoﬀs where the site does

not shut down, so I will refer to them all as “mass layoﬀs.”

18In municipalities where more than one closing occurred during the same year, the multiple closings were counted

as a single closing with job loss equal to the sum of job loss in all closings.

45

a given year is positively related to the probability of a layoﬀ occurring in previous years, failing
to control for job loss in past layoﬀs would bias the estimates. Including four lags of the job loss
variable as well as the contemporaneous term also illustrates the eﬀect of a single mass layoﬀ over
time. The coeﬃcient on the fourth lag of job loss, for example, illustrates the lingering eﬀect of the
mass layoﬀ on income tax base four years after the layoﬀ occurs.

For municipality i in year t, I model the relationship between job loss and income tax base using

the following event study speciﬁcation:19

log(income_tax_base)it = (job_lossi,t+1, job_lossit, job_lossi,t−1, ..., job_lossi,t−4)γ

+ Xi,t−5β + ξi + δt + errorit

(1)

Since the time-constant component ξi is likely correlated with the probability of a mass layoﬀ
occurring, I estimate the model using ﬁxed eﬀects. I also include a set of year dummies, and a
vector of control variables, lagged by ﬁve years.20 In municipalities where a layoﬀ occurred, mean
income tax base may be lower throughout the sample time period than in years before the layoﬀ. In
other words, if a layoﬀ has a lasting eﬀect, this will drag down mean income tax base. This means
the dependent variable would be higher than its sample mean in the year before the layoﬀ, and lower
in years after the layoﬀ. Thus, I include the lead of the job loss variable. The coeﬃcient on the
lead of announced job loss shows how much higher the dependent variable is predicted to be the
year before a mass layoﬀ event of a certain size, relative to other sample years for that municipality.
Note that the dependent variable is logged while proportional job loss is expressed as a fraction.
Thus, the diﬀerence between the coeﬃcient on the lead term and the coeﬃcients on the lags show
the percentage change in the dependent variable associated with a mass layoﬀ with job loss equal
to 1% of population, from the year before the layoﬀ to however many years after.

19For more on event studies in economics, see MacKinlay [1997] and Sandler and Sandler [2013].
20

X includes ﬁve year lagged controls for log of population, log of real property value, log of residential property

value, income tax rate, and aggregate property tax rate for all levels of government overlapping the municipality.

46

The ﬁrst column of Table 3.2 reports estimates with income tax base as the dependent variable.21
In municipalities where a mass layoﬀ occurred, income tax base was higher the year before the
layoﬀ than in other sample years.
In the year that the layoﬀ began, a job loss equal to 1% of
municipal population causes income tax base to drop about 0.293 − 0.0805 = 0.21% from the
pre-layoﬀ level, although the diﬀerence is not statistically signiﬁcant (p-value = 0.16). Because the
income tax base variable reﬂects the year that income was collected, this means income dropped
the year before the layoﬀ took place. This could mean that plants began cutting some staﬀ prior
to the mass layoﬀ, reducing hours but not ﬁring employees, or that proﬁts began to fall in plants
that would shut down. During the year following the mass layoﬀ income tax base drops by 0.54%
(-0.459 - 0.0805) compared to the year of the layoﬀ, or 0.75% compared to the year before the
layoﬀ. This means a layoﬀ at the mean of the proportional job loss variable, 3.2% of the population,
results in a 2.4% drop in tax base from the year before the layoﬀ to the year after. Income tax base
remains below the pre-layoﬀ level - a mass layoﬀ with proportional job loss equal to 1% results in
0.46% lower income tax base four years after, with the diﬀerence statistically signiﬁcant at the 99%
conﬁdence level.

The loss of income tax revenue persists for a number of possible reasons. First, employees of the
ﬁrm who lived in the municipality may emigrate to another location for work. Second, employees
who commuted into the municipality may ﬁnd work in a diﬀerent location. Finally, employees
who ﬁnd another job in the municipality, or former resident employees who now commute outside
the municipality, may receive lower wages. Finally, in the case of a plant closing, the municipality
loses tax that the business paid on its net proﬁts, and enterprises serving the business or its workers
will suﬀer losses and reduce employment as well.22 These tend to amplify the eﬀects described
21The marginal tax rate does not change with income. For this reason, I can divide gross revenue by the income tax
rate to measure income tax base. The panel of income tax base, property values, and tax rates runs from 1997 to 2013.
Because WARN notice data is available from 1997 to 2014, ﬁve periods are lost due to the inclusion of lags and leads.
Only municipalities that had an income tax in all years (1997-2013) are included in the income tax and employment
regressions.

22Greenstone, Hornbeck, and Moretti [2010] structurally estimate the eﬀect of a large plant opening on total factor
productivity of plants in the same county, discovering that opening a new “million dollar plant” increases the total
factor productivity of incumbent plants by 12%, increasing labor demand and wages. In the converse scenario, when a
ﬁrm closes, labor demand and wages at nearby ﬁrms decline. Firms that marketed their goods to the ﬁrm conducting

47

above.

Figure 3.2 plots the coeﬃcients on the job loss variables, illustrating the layoﬀ’s eﬀect over
time. Figure 3.2 also plots the coeﬃcients from the second column of Table 3.2, when log of
employment is the dependent variable in Equation 1. Employment follows the same path as income
tax base, but the drop in employment is somewhat less severe initially. The drop in income tax
base may be somewhat more severe because it includes lost tax on proﬁts, as well as on lost wages
from laid oﬀ workers. The decline in income tax base likely also reﬂects skill mismatch for laid oﬀ
workers who quickly ﬁnd new jobs. If some laid oﬀ workers quickly accept a position that does not
require as much human capital, then they will receive low wages despite remaining employed.

3.3.2 Property Tax Base

The third and fourth columns of Table 3.2 display results of regressions with assessed real property
value as the dependent variable. The right hand side is identical to Equation (1), with one exception:
because I include ﬁxed eﬀects, I exclude controls for real property value and Class 1 property value
from the real property value equation.23 After a mass layoﬀ, real property value declines slightly.
For a layoﬀ with job loss equal to 1% of population, real property value falls 0.05% from the year
before the layoﬀ to the fourth year after, although the diﬀerence is not statistically signiﬁcant.

Several factors may prevent a decline in real property value from showing in the data. Perhaps
property owners anticipate the layoﬀ, so that a downtrend in property value precedes the layoﬀ. If
not, there will still be some lag between the layoﬀ and any decline in sale prices, as unemployed
workers may delay choosing to relocate, for example. Finally, there will be a lag between the layoﬀ
and the next reappraisal of property; depending upon the time between the layoﬀ and the triennial
reappraisal, it could take as long as six years for a decline in sale prices to fully reﬂect in appraised
value.
the layoﬀ or its employees experience lower proﬁts, and labor demand shifts to the left.

23The property value regressions, unlike the income tax regression, include all municipalities in the sample. Like
the income tax base regression, they include a control for the income tax rate, with a rate of zero for municipalities that
do not have an income tax.

48

3.4 Municipal Response

A municipal government may respond to a mass layoﬀ by altering its ﬁscal policy. It can attempt
to recover lost revenue by raising tax rates or by switching to other revenue sources. Failing this,
it must cut expenditure, undertake more debt ﬁnance, or deplete its saved resources. Expenditure
cuts may not aﬀect all items proportionally, depending upon the perceived necessity of the item
and changes in demand for the item. A municipality may issue debt or draw upon rainy days funds,
rather than cutting expenditure, if administrators perceive the eﬀects of a mass layoﬀ to be short
lived. If the layoﬀ’s eﬀects are expected to last, then the municipality is more likely to reduce its
issue of new debt, but may draw on accumulated funds or reﬁnance outstanding debt to adjust to a
new level of expenditures.

3.4.1 Tax Rates

Table 3.3 shows the result of the ﬁxed eﬀects event study speciﬁcation from Equation 1 with
income tax rate and property tax rate as dependent variables. Both income tax and property tax
rates decline. For income tax rate, the diﬀerence for the coeﬃcient in the period before the layoﬀ
and the coeﬃcient four years after is (0.127 − 0.0390) = 0.09. A mass layoﬀ of mean size with
proportional job loss of 3.2% lowers the income tax rate by 0.032 × 0.08 = 0.0029 percentage
points, 0.34% of a standard deviation. By a similar calculation, property tax rate (measured in
mills) falls by 0.4% of a standard deviation from the year before the layoﬀ to four years later.24
Neither diﬀerence proves statistically signiﬁcant. But clearly municipalities do not compensate for
a decline in revenue base by raising tax rates.

Tax rates fall after a mass layoﬀ either because the municipal government does not wish to
increase its tax rate, because voters refuse tax increases or renewals, or because the municipal
government anticipates that a tax referendum will be unpopular and does not attempt it.
In
some cases, a voted increase in property or income tax rate is approved for a ﬁxed duration; the

24 0.032 × (0.570−(−0.0333))

4.79

= 0.004

49

municipality must pass a renewal of an expiring rate, or the total rate decreases. It is unlikely that
residents would be more willing to approve a rate increase following a mass layoﬀ. Marginal utility
of income increases for voters whose income has fallen, and if a mass layoﬀ causes job loss for
commuters, this shifts more tax burden onto residents. The government’s decision to propose a
tax referendum depends on the marginal beneﬁt of revenue and its expectation that the referendum
will pass. For a tax base decrease to increase the probability that a referendum passes, the median
resident must have highly inelastic demand for public services.

3.4.2 Revenue

If the tax base falls with no compensating increase in tax rates, revenue will fall. Regressions in
Table 3.4 conﬁrm this, using diﬀerent categories of revenue as the dependent variable in Equation
1.25 Income tax revenue declines, with the magnitude of the decline and its persistence over time
very similar to the pattern for income tax base displayed in Table 3.2. Interestingly, property tax
revenue decreases as much as income tax revenue in the ﬁrst and second years after a reported
mass layoﬀ. As seen in Tables 3.2 and 3.3, property tax rate and real property values fell as
well, contributing to the decline. But the small declines in real property value and tax rate cannot
account for the large property tax revenue decline. Property tax revenue may fall without a decline
in property value or tax rates for at least two reasons. First, property tax delinquency may increase.26
An increase in volatility of either Class 1 or Class 2 property values following a mass layoﬀ could
cause overall property tax revenue to decline (c.f. Section 3.2.1). If Class 2 value falls while Class
1 value rises, or vice-versa, then revenue from Class 2 value falls while revenue from Class 1 value
is capped at the amount authorized by the existing levies and likely remains the same.27

25Although only the coeﬃcients on the proportional job loss variables are shown, these regressions include the

same controls as the regressions in Section 3.3.

26Statewide, net property tax delinquency equaled 0.7% of property tax revenue between 1997 and 2010. Property
tax delinquency was countercyclical, reaching 1.5% in the 2001 recession year and 1.6% in the 2008 recession year.
Unfortunately municipal data on property tax delinquency are not available.

27Besides the data on income tax revenue, all revenue and expenditure data are from audit reports and annual
ﬁnancial reports for Ohio municipalities from 1999-2013. These data reﬂect the year when the revenue was available;
thus, total revenue reﬂects income tax revenue from activity in the previous year.

50

The last three columns of Table 3.4 shows how a mass layoﬀ aﬀects revenue from fees, ﬁnes,
and service charges, intergovernmental revenue, and total revenue. Because there are many zeros
in the data, I alter Equation 1 and use ﬁxed eﬀects Poisson estimation instead of OLS with a logged
dependent variable. As with a logged dependent variable, each coeﬃcient shows the percentage
change in the dependent variable associated with a change in the right hand side variable.28 For
a mass layoﬀ with 1% proportional job loss, services, fees and ﬁnes revenue declines 0.27% from
the year before the layoﬀ to two years after. Still, the coeﬃcient on the forward job loss variable
is not statistically diﬀerent from the coeﬃcients on any lag. Intergovernmental revenue does not
change systematically following a mass layoﬀ.29 Column 5 shows that the declines in income tax
revenue, property tax revenue, and ﬁnes, fees and services revenue drive a substantial decline in
total revenue. Revenue declines 0.41% from the year before a mass layoﬀ with 1% proportional job
loss to the year after. Standard errors are relatively large, but a test that average revenue in the years
after a mass layoﬀ equals average revenue in the year of the layoﬀ and the year before marginally
rejects the null (p-value = 0.10).

3.4.3 Expenditure

The decline in revenue associated with a mass layoﬀ pressures municipalities to cut expenditure.
In addition, a mass layoﬀ may reduce the demand for some municipal services. Amenities such as
ﬁre protection, emergency medical services, police protection, well maintained roads and bridges,
28The Poisson ﬁxed eﬀects estimator is robust to distributional assumptions, conditional on a correctly speciﬁed
mean. Like ﬁxed-eﬀects OLS, it does not suﬀer from the incidental parameters problem [Wooldridge, 2010, Chapter 19].
Any municipality with all zero values of the dependent variable drops out of the estimation - thus three municipalities
with all-zero values for services, ﬁnes and fees revenue are dropped from Column 3 of Table 3.4, leading to slightly
fewer observations in this regression.

29Ohio municipalities participate in a revenue sharing program, where the state distributes unconditional aid to
municipalities based on population, property tax base, and whether the municipality had any income tax. In 2007,
distributions to municipalities totaled $416 million. In the same year, gas tax distributions to municipalities from the
Ohio Department of Transportation totaled $245 million, based on number of registered vehicles in each municipality.
$661 million of the $1.59 billion in intergovernmental revenue derived from factors that would either not change, or
that would shrink during a local economic downturn. Other sources of aid included partial or total payment for capital
projects distributed by the Department of Transportation or the Ohio Public Works Commission. As will be seen next,
a drop in expenditures accompanies the drop in revenue detailed here. Any aid comprised of a state match of local
funds likely decreases with this decrease in expenditure.

51

and access to water and power are highly demanded by most enterprises. Most mass layoﬀs in the
sample stem from an establishment closing, and the rest likely accompany an establishment scaling
back its operations, leading to a decline in the demand for services that it uses.

A municipal government will seek to cut services where the marginal beneﬁt from providing
the service is relatively elastic, or where the marginal beneﬁt from the service decreases with a
plant shutdown. Previous literature suggests that municipalities facing a decline in revenue or a
ﬁscal crisis will cut expenditure on general government operations, on parks and recreation, and on
capital outlay [Cromwell and Ihlanfeldt, 2015, Skidmore and Scorsone, 2011], which seem to have
a ﬂat marginal beneﬁt curve. The marginal beneﬁt curve for ﬁre, ambulance, and EMS services
may be steeper, but a plant closing may shift the demand for these services to the left.

Table 3.5 shows that total expenditure declines after a mass layoﬀ.30 For a layoﬀ with propor-
tional job loss of 1%, I estimate that total expenditure declines 0.26% from the year before the
layoﬀ to two years after, or 0.52% from the year of the layoﬀ to two years after. Most categories
of expenditure also fall after a mass layoﬀ. General government expenditure falls 0.38% from the
year before to two years after a mass layoﬀ with proportional job loss of 1%. For a layoﬀ this size
over this time frame, public safety expenditure falls by 0.24%, leisure and community environment
expenditure falls by 1.75%, and capital outlay falls by 1.02%. The drop in public safety expenditure
contrasts with previous literature estimating municipal response to decreases in revenue, and may
reﬂect a decrease in demand. Alternatively, municipalities facing a large, unexpected shock to
revenue simply need to cut expenditure quickly, and public safety expenditures are a large category
that may be easily reduced. The drop in infrastructure spending more likely reﬂects a decrease in
demand, since it begins the year of the layoﬀ, before the larger decline in revenue. In contrast,
expenditure on public health increases, reinforcing the ﬁndings of Kuhn, Lalive, and Zweimüller
[2009] and Browning and Heinesen [2012] that plant closings raise demand for public health
services.

30Controls are included, but their estimated coeﬃcients are suppressed in the output table.

52

3.4.4 Cities’ Assets and Debt

The large, persistent revenue decline associated with a mass layoﬀ brings about a decrease in
expenditure, but this decrease need not perfectly track the drop in revenue. Municipalities may
also borrow to fund certain expenditures, like infrastructure improvements and capital outlays.
A municipal government with declining tax base and no desire to raise taxes may choose to
fund necessary capital expenditures by borrowing. A municipality may not borrow to directly fund
operating expenditures. Still, municipalities may free up current revenue for operating expenditures
by funding capital outlays through debt rather than current revenues.

Because municipalities may only borrow for certain purposes, however, accumulated asset
balances may provide a more important source of emergency funds than debt issue. Unlike states,
municipalities may do not set aside a restricted “rainy day fund,” but often accumulate an unreserved
fund balance for the same purpose [Marlowe, 2005, 2013]. Unreserved fund balance refers to the
diﬀerence between spendable assets and short term liabilities that is not reserved by law for a
speciﬁc expenditure - essentially the municipality’s accumulated resources that it may designate for
any use. Thus, municipalities may use unreserved fund balance to avoid expenditure cuts during
periods of revenue decline.31

Summarized Annual Financial Reports from the Ohio Auditor’s Oﬃce provide data on assets
and liabilities for cities, but not villages, from 2002-2013.32 Beyond data availability issues, it
makes sense to analyze cities separately. Cities likely possess more sophisticated resources for
managing investments, and thus may ﬁnd it easier to manage accrued assets. Similarly, issuing debt
31If Ricardian equivalence holds, then a municipality will increase its debt (or reduce its fund balance) in two
circumstances. A municipality may borrow or draw down savings to intertemporally smooth expenditure if it expects
a decrease in revenue to be temporary, or to gradually adjust to a lower level of expenditure if it expects the decrease to
be permanent. If Ricardian equivalence does not hold, then present administrators and voters may approve temporarily
higher spending at the expense of future taxpayers.

32Unreserved fund balance includes designated fund balance and undesignated fund balance, although the distinction
is not particularly meaningful in practice [GASB, Governmental Accounting Standards Board, 2006]. Starting in 2013,
Summarized Annual Financial Statements reported restricted, committed, assigned, and unassigned fund balances
[Government Accounting Standards Board, 2009]. For 2013, I replace unreserved fund balance with assigned and
unassigned fund balances. Some cities did not report assets or debt every year.
If a city had not more than one
consecutive year of missing data, then I interpolated the variable’s value in that year as the linear combination of the
value in the year before and the year after.

53

involves substantial ﬁxed costs of legal fees, ratings fees, and underwriter compensation. Thus,
villages typically issue debt only in special circumstances, while the largest cities issue debt on
an annual basis. Summary statistics on cities’ outstanding debt, unreserved fund balance, and
cash and investments are presented in Table 3.1. Most cities had some form of long-term, general
obligation33 bonds issued, while a smaller number had outstanding notes or loans. Most cities also
had a positive unreserved fund balance. Cities often held substantial cash and investment reserves,
as well - indeed, cash and investments nearly equaled debt owed by cities, on average.

I estimate the eﬀect of mass layoﬀs on debt, unreserved fund balance, and cash and investments
using a ﬁxed eﬀects event study regression. The speciﬁcation is similar to Equation 1, but the
dependent variable is measured in per capita terms (divided by municipal population) and not
logged. While debt and cash and investments are either positive or zero, unreserved fund balance
is negative for many observations. Besides accounting for negative observations in unreserved
fund balance, the linear speciﬁcation allows for a direct comparison between changes in sources of
funds, while expressing the dependent variable in per capita terms keeps large observations from
dominating the estimation.

Table 3.6 shows the results of this estimation. For a mass layoﬀ with job loss equal to 1% of
population, general obligation bonds decrease in the second, third and fourth years following a mass
layoﬀ by about $11 per capita, relative to the year before. Notes and loans payable, however, increase
in the second, third, and fourth years following a mass layoﬀ, by roughly the same amount. The short
term debt undertaken by municipalities includes bond anticipation notes and tax anticipation notes,
preparing municipalities for an increase in expenditure. As municipalities decrease expenditure
(especially capital expenditure), this type of short term debt decreases. Despite increasing capital
expenditures, however, municipalities increase long-term outstanding debt. The increase may
reﬂect a choice of how to ﬁnance expenditures undertaken in the past. For example, Moraine Ohio
(discussed in the introduction) issued $5.8 million of new long term debt in 2010, two years after
33General obligation bonds are backed by the full credit of the municipality issuing them. Other types of long term
debt, such as revenue bonds and special assessment bonds are backed only by a speciﬁc source of revenue and are not
considered in what follows.

54

experiencing a series of devastating layoﬀs. This issue retired outstanding notes ﬁnancing past
expenditure [Moraine, 2010]. Moraine’s administration likely decided to ﬁnance a larger share of
these expenditures through debt, rather than tax, after the layoﬀ occurred.

Per capita unreserved fund balance decreases $47 in the year after a mass layoﬀ with 1%
proportional job loss, before recovering to a level roughly $32 per capita below its level at the
year before the layoﬀ. A mass layoﬀ of average size, with proportional job loss equal to 3.2% of
population, lowers unreserved fund balance by 12% of a standard deviation from the year before
the layoﬀ to the year after. The results suggest that cities, unwilling or unable to raise tax rates and
unwilling to cut operating expenditures, choose to instead deplete their accrued assets. Column 4
shows that most of the long term decline stems from municipalities drawing down their balance of
cash and investments - cash and investments per capita decline $24 per year from the year before
the layoﬀ to four years after. While municipalities may hold a fund balance to avoid expenditure
cuts in response to temporary shocks, the results in Table 3.2 suggest that the decline in income tax
base associated with a mass layoﬀ persists over time. At the same time, unreserved fund balance
does not recover to the pre-layoﬀ level after four years, leaving cities vulnerable to further ﬁnancial
stress.

3.5 Conclusion

The shock to municipal ﬁnances following a reported mass layoﬀ proves to be large and eco-
nomically signiﬁcant. Revenue declines substantially for several years, and municipal governments
are unwilling or unable to increase tax rates in response to this drop. Instead, municipalities re-
spond by cutting expenditures and increasing outstanding debt. Municipalities do not seem to fully
recover within four years of a layoﬀ’s occurrence. Income tax revenue, total revenue, and total
expenditure remain below their pre-layoﬀ level four years after the layoﬀ’s occurrence. Munici-
palities cut expenditure not only on general government, leisure and community environment, and
capital outlays, but even on public safety. Municipalities run up outstanding long-term debt, and to
an even greater degree dig into discretionary funds accumulated from past revenue.

55

These results bear at least two important implications for policy. First, by estimating the size
and duration of the shock to revenue, the impact on diﬀerent types of revenue and expenditure,
and how eﬀectively municipal ﬁscal policy ameliorates the shock, I provide information to state
and federal policymakers evaluating diﬀerent types of aid to economically distressed communities.
Second, as most Ohio municipalities impose an income tax, unexpected job losses will impede
revenue raising capabilities more in this state than in others. Economists and policy researchers
have weighed the advisability of relying on traditional property taxes against the advantages of
increasing revenue through sales or income taxes.34 Because the income tax base ﬂuctuates more
than the property tax base, high reliance on an income tax exposes local governments to greater
revenue variability. Municipalities’ ability to respond to an income tax decline casts light on the
wisdom of tapping this revenue source.

Although longer term public safety cuts and reduced assets seem potentially harmful, further
research is needed to better ascertain the exact welfare implications of these eﬀects. Do the
expenditure cuts associated with a mass layoﬀ correspond to lower quality of government service
provision, such as higher crime, or lower 911 and EMS response times? Are municipalities able to
maintain a sound ﬁscal position after the reduction in fund balance, or do they enter ﬁscal crisis?35
Knowing the eﬀect of a mass layoﬀ, and municipalities’ response, is a good start, but open questions
remain.

34Fisher [2009] summarizes the policy debate over property taxes. McGuire, Papke, and Reschovsky [2014]
compare the property tax to its alternatives in school ﬁnance, while Mikesell [2010] provides a descriptive overview
of local sales and income taxes. Both Sobel and Holcombe [1996] and Dye [2004] measure the procyclicality of the
state income tax, while Jesse Edgerton and Andrew F. Haughwout and Rae Rosen [2004] show a case where increased
income taxes lead to higher revenue variation at the local level. Chernick, Langley, and Reschovsky [2011] ﬁnd that
property tax displayed less cyclicality in large cities during the Great Recession than local sales or income taxes.

35Beckett-Camarata [2004] surveyed oﬃcials in Ohio local governments suﬀering from a state-deﬁned ﬁscal
emergency. 47% of respondents attributed the ﬁscal emergency to a plant closing, suggesting that many localities
suﬀer long-term ﬁscal hardship after these events.

56

CHAPTER 4

BALLOT ORDER AND BALLOT ROLL-OFF: EVIDENCE FROM LOCAL
REFERENDA IN OHIO (COAUTHORED WITH MIKE CONLIN AND PAUL

THOMPSON)

4.1 Introduction

Research in economics, marketing and political science has demonstrated that the size of the
choice set and the sequence of choices aﬀect an individual’s decisions, in environments ranging
from on-line grocery orders to car sales to elections. Numerous empirical papers provide evidence
of “choice fatigue,” testing whether individuals’ choices depend on how many decisions they have
made previously. A signiﬁcant number of these papers use election ballot order to test: (i) whether
individuals are less likely to vote on contests located further down the ballot, a phenomenon that
political scientists call “roll-oﬀ”; and (ii) whether a contest’s ballot location inﬂuences the choice of
those who do vote. Augenblick and Nicholson [2016], hereinafter referred to as A&N, provide the
most convincing strategy for identifying a causal relationship between ballot position and election
outcomes, using a set of 392 contests in San Diego County between 1992 and 2002.1 Because the
ballot location of a given contest varies across precincts in San Diego, A&N are able to convincingly
address contest heterogeneity. Due to a diﬀerent set of local contests on diﬀerent precinct ballots,
a contest’s position varies across ballots, which allows A&N to use within-contest variation for
identiﬁcation. In addition, because they observe precincts across multiple elections, they are able
1A number of other papers consider similar questions in terms of the relationship between ballot order or ballot
length on the one hand, and roll-oﬀ or election outcomes on the other, but we feel that no other identiﬁcation strategy
to determine the causal eﬀect is as convincing as A&N. Bowler, Donovan, and Happ [1992] regress both roll-oﬀ and
probability of voting no on ballot rank for a set of California ballot proposals. Although they ﬁnd signiﬁcant eﬀects in
both cases, they fail to account for many of the proposition characteristics that may be correlated with ballot length.
Selb [2008] considers the eﬀect of ballot length on voting patterns, using a set of Swiss referendum elections. Although
the author fails to ﬁnd a statistically signiﬁcant relationship between ballot length and no votes, he does estimate a
statistically signiﬁcant eﬀect of ballot length on the variance of no votes. Ho and Imai [2008] estimate the causal
relationship between the position in which a candidate’s name is listed on the ballot within a contest and the vote
share that the candidate receives, using a dataset of state oﬃce elections in California. They identify a statistically and
economically signiﬁcant eﬀect for minor party candidates and for primary elections. Ho and Imai [2006] also ﬁnd a
positive relationship between ballot position and votes received in the 2004 California recall election.

57

to control for time invariant precinct-level factors that inﬂuence voter behavior.

A&N ﬁnd evidence of roll-oﬀ and, perhaps more interesting, they ﬁnd that voters are more likely
to choose the status quo – including voting no on propositions and voting for the ﬁrst candidate
listed – for contests that appear lower on the ballot. While the results of this thorough analysis
are consistent with voter fatigue, A&N also mention two other possible explanations: time-varying
precinct characteristics such as voter composition being correlated with ballot location and the
probability of a voter “rolling oﬀ” being correlated with a voter’s preferences (especially in terms
of the status quo results). While A&N are unable to directly test for these alternative explanations,
they do ﬁnd that choice fatigue does not occur on contests toward the top of the ballot that have no
ballot position variation and that the results hold when ballot length is included as a covariate.

Using information on 2004-2012 Ohio tax referenda proposed by local administrative divisions
(townships, municipalities, school districts, and counties) and diﬀerent variation for identiﬁcation,
our paper complements the work of A&N. In terms of variation used for identiﬁcation, Ohio law
regulates order within the “questions and issues2 portion of the ballot as follows (Revised Code §
3505.06 (B) (1)). State questions and issues are at the top of this portion of the ballot, followed
by questions and issues involving local jurisdictions. The order of local questions and issues
rotates annually based on jurisdiction type. In 1997, county referenda were placed at the top of the
ballot, followed by municipal referenda, township referenda, and ﬁnally school and other district
referenda. The next year, and every year following, the top jurisdiction category moved to the
bottom of the ballot, while every other jurisdiction category moved up one rank. Thus, in 2010,
municipalities were placed ﬁrst, followed by townships, school and other districts, and counties. In
2011, however, townships were placed ﬁrst, and municipalities fell to the bottom. This rank rotation
provides a source of exogenous variation in ballot order, allowing us to test for voter fatigue. The
main disadvantage of this paper relative to A&N is that rank, our proxy for ballot location, does
2The ballot in Ohio is divided into three sequentially ordered sub-ballots (Revised Code §§ 3505.03 – 3505.06).
First, the “Oﬃce type ballot” presents choices of candidates chosen through the nominating process, including state and
federal candidates, as well as many candidates for local oﬃces. Second, the “Nonpartisan ballot” includes candidates
for state or local school board, candidates for judicial oﬃce, and candidates for municipalities and townships not
nominated through a primary. Finally, the questions and issues ballot contains proposals to be accepted or rejected
through the referendum process.

58

not vary for a given referendum. For this reason we are unable to use within-contest variation to
estimate voter fatigue. To adequately address referendum heterogeneity, we restrict our analysis to
tax referenda because Ohio law standardizes the proposal process for these referenda and speciﬁes
a particular “taxing authority” which may propose referenda for each administrative division. More
importantly, we are able to control for many characteristics of the referenda, including whether
the referendum renews an existing tax, the purpose of the tax revenue, the type of tax and the tax
rate. These characteristics are likely to aﬀect the salience, palatability, and welfare impacts of the
referenda.

The main advantage of this paper over A&N is that we have voter level information from which
we can construct measures of voter composition for each referendum: speciﬁcally, the fraction of
voters over 65 years of age, the fraction registered democrats and the fraction registered republicans.
These composition measures may be important. Numerous studies document how an individual’s
attention span, cognitive abilities and decision making vary with age and how voter preferences
vary with political party. Another diﬀerence is that while voting no on a proposition is a vote for
the status quo in A&N, the status quo for our referenda depends on whether a referendum renews
an existing tax. A yes vote is a vote for the status quo when the referendum involves a tax renewal
and a no vote is a vote for the status quo when the referendum does not involve a tax renewal.

The empirical results in this paper suggest that controlling for variation in voter composition
may be important when testing for choice fatigue. We ﬁnd that the fraction of individuals at the
ballot box choosing not to vote on a township referendum increases with the fraction of voters
under the age of 65 and with the fraction registered as either a democrat or republican. However,
we do not ﬁnd the eﬀect of ballot location on this decision to vary signiﬁcantly based on the age or
political aﬃliations of the voters. Thus, our results suggest that the evidence of roll oﬀ is not due
to variation in the age and party aﬃliation of voters. That said, we do ﬁnd that those referenda with
large fractions of voters under the age of 65 or registered as democrat or republican have a much
larger fraction of voters choose to not cast a referendum vote - irrespective of where the referendum
appears on the ballot. This is important when considering status quo bias because we also ﬁnd

59

that while the fraction voting yes on the referendum does not appreciably vary with political party
aﬃliation, it does vary signiﬁcantly based on age. Not only does the share voting yes vary with
the fraction of voters over 65 years of age, this relationship depends on whether the referendum
involves a tax renewal or a new tax. For tax renewal referenda, where a yes vote supports the status
quo, a larger fraction of voters over the age of 65 results in a larger fraction of yes votes compared
to a referendum involving a new tax. While we do not ﬁnd this diﬀerential relationship to vary
based on where the referendum appears on the ballot, it does suggest that if voter composition
measures are not controlled for, results that are consistent with status quo bias could actually be
due to unobserved voter heterogeneity.

4.2 Data and Summary Statistics

The Ohio Board of Elections provided referendum and voter characteristic information from
2004 through 2012.3 The referendum data consists of the outcomes of local tax referenda located
on general and primary election ballots and the ballot language from the Ohio Board of Elections
website.4 In terms of the diﬀerent types of jurisdictions, there are 3,504 township referenda, 688
county referenda, 2,136 municipality referenda and 2,362 school district referenda. In terms of
referendum outcomes, we have the number of yes votes and no votes. The ﬁrst column in Table 4.1
contains summary statistics for only township referenda and indicates that the average numbers of
yes and no votes are 734 and 490, respectively.5 This results in the share of yes votes averaging
3For every administrative division that may levy taxes, the Ohio Revised Code speciﬁes a “taxing authority”
authorized to propose tax referenda. For a county, the taxing authority is the county commissioners; for a municipality,
the legislative authority (e.g. city council); for a township, the township trustees; and for a school district, the district
board of education. Generally, in order to place a tax or bond referendum on the ballot, the taxing authority must ﬁrst
adopt a resolution describing the purpose of the tax issue, and describing various characteristics of the tax or bond
issue (such as the rate of taxation in the case of a tax levy, or the total debt issued in the case of a bond).

4Elections in Ohio are held on four uniform election dates during years without a presidential election, and on
three uniform election dates during presidential election years. Taxing authorities may propose ballot questions for
any of these election dates. The general election occurs in November. In years where no presidential election occurs,
primary elections take place in May, while special elections may be held in February or August. During presidential
election years, primary elections take place in March, superseding both the May primary and, in every year since 2000,
the February special election.[Ohio Secretary of State, 2013]

5We gathered information on over 13,000 referenda which were on an Ohio ballot between 2004 and 2012. We
dropped all referenda that did not involve a tax or had missing information (such as number of yes and no votes). We
also dropped referenda that took place during either February or August (i.e., special elections). We also dropped

60

0.62 and 85.8 percent of these referenda passing.

Along with vote counts on the referenda, the individual Ohio voter ﬁle6 identiﬁes the jurisdiction
of each registered voter and which elections the individual went to the ballot box to vote. By
identifying the jurisdiction of each registered voter and which elections the individual went to
the ballot box to vote, the Ohio Board of Elections individual voter ﬁle allowed us to credibly
identify the number of voters who did not vote on the township referendum. The percent of voters
who did not cast a vote on the referendum averaged 15.1 percent across all township referenda.
Unfortunately, due to missing voter speciﬁc jurisdiction designations, it was diﬃcult to credibly
identify the fraction of voters who voted on the other types of referenda.7 This is the reason
we distinguish between townships and the other jurisdiction types and provide summary statistics
for county, municipality and school district referenda in column 2 of Table 3.1.8 Compared to
township referenda, these jurisdictions’ referenda average seven times more votes (8,343 total
votes compared to 1,224), with a smaller share of voters supporting the referendum and a lower
probability of passing.

The individual voter ﬁle also speciﬁed birth year and political aﬃliation for each registered
voter. From this information, we constructed the fraction of voters over 65 years of age, the
fraction registered democrats and the fraction registered republicans. Because these measures are
fractions of the total voters identiﬁed from the jurisdiction and we have no reason to believe the
misclassiﬁcations of jurisdiction location in the ﬁles are correlated with an individual’s age or

all “special district” referenda because we were not able to identify the number of voters in the district and unable
to identify whether an election would have been ranked among the “school and other districts” or with a diﬀerent
jurisdiction. We dropped all referenda involving seven of the 83 Ohio districts because of concerns with the individual
voter ﬁles which are maintained by each county. For the remaining 76 counties, we individually went through each
referendum to ensure the referenda and voter information was reliable. For those referenda that caused concern, such
as those having more total referenda votes than voters at the ballot box, we dropped them. These were relatively few
referenda and the results of our empirical analysis do not change appreciably if we kept them. In the end, dropping
these referenda left us with a dataset containing 8,690 referenda.

6This ﬁle contains voting records for all registered voters in the state of Ohio. These ﬁles can be accessed at:

https://www6.sos.state.oh.us/ords/f?p=111:1

7For most counties, we are reasonably comfortable with the number of voters identiﬁed for county referenda. We
chose not to include these referenda because the number of county referenda is limited and, of the jurisdiction types,
controlling for heterogeneous referenda characteristics is most problematic for county referenda.

8We do not present separate estimates for county, municipalities and school district referenda because the estimates
are similar across these jurisdiction types but more precisely estimated when all three jurisdiction referenda are used.

61

party aﬃliation, we construct these demographic variables for all four types of jurisdictions. Table
3.1 indicates that approximately a quarter of all voters are over 65 years of age irrespective of the
jurisdiction type and that about twice as many individuals are registered republicans as registered
democrats. Interestingly, about half of all voters do not indicate a party aﬃliation.

Figure 4.1 depicts how these demographic characteristics of voters vary across elections for the
townships that have a referendum and for the other three jurisdictions that have a referendum. It is
clear that voter characteristics vary signiﬁcantly across elections with younger voters being much
more likely to vote in presidential and gubernatorial elections relative to other general elections and
with a higher proportion of party aﬃliated voters turning out for primary elections. Not only are
voter preferences likely to vary with these demographic characteristics but the decision to abstain
from voting on the referendum may also vary. In addition, how abstaining varies with ballot location
may also vary across age and party aﬃliation. The speciﬁcations in the next section address these
issues.

The ability to identify when the referendum appeared on the ballot (i.e., which election) allows
us to determine the rank ordering for the particular jurisdiction’s referenda in that election year.
This rank order proxies for the referendum’s exact ballot position and varies between one and
four (e.g., a one indicates that referenda for the particular jurisdiction type precedes referenda
for the other three jurisdiction types and a four indicates it succeeds these other jurisdiction
types).9 The average rank is close to 2.5 for all four types of jurisdictions because the number
of referenda proposed remains roughly constant across years. The exact wording on the ballot
allows us to identify not only the jurisdiction but also whether the referendum renews an existing
tax the purpose of the tax, the type of tax and the tax rate. The referenda propose to fund many
diﬀerent types of expenditures which vary across jurisdiction type. The most prevalent types for
9We could construct another measure of ballot location by estimating the average number of contests that would
appear on the ballot for each township referendum. The problem with this measure is that, unlike A&N, we can only
estimate an average number due to overlapping jurisdiction issues. In addition, we are concerned that the number of
diﬀerent types of referenda in a geographic area is correlated with unobserved voter preferences. We prefer using rank
for this reason, as well our concern that when an administrative unit places a tax referendum on the ballot may depend
on their expectation of whether the other overlapping administrative divisions have a tax referendum on the ballot. We
are less concerned that an administrative division’s decision to place a tax referendum on the ballot depends on the
division’s rank in that particular year.

62

township referenda are police/ﬁre/emergency medical services (ems), roads and cemeteries; for
county referenda, police/ﬁre/ems, community/senior center, health services and child services; for
municipality referenda, police/ﬁre/ems, roads, current expenditures and parks/recreation facilities;
and for school district referenda, capital projects, current expenditures, emergency expenditures and
permanent improvements. Table 3.1 indicates that slightly less than half of the referenda involve
a tax renewal (48.3% for townships and 41.3% for counties, municipalities and school districts)10
and identiﬁes the proportion of the referenda pertaining to the diﬀerent purposes. Table 3.1 also
provides descriptive statistics on the type of tax and tax rate. All of the township referenda involve
property taxes and their average millage rate is 1.52. Counties can raise revenue through a sales
or property tax while school districts and municipalities can raise revenue through an income or
property tax. Column 2 of Table 3.1 indicates that 83% of the referenda for these three jurisdiction
types involve a property tax while only 7% propose a sales tax and only 11% propose an income tax.
The average millage rate, sales tax rate and income tax rate are 3.351, 0.456 and 0.693, respectively.
Along with the rank ordering, the ability to identify which election the referendum appeared on
the ballot allows us to identify other aspects of the election that might be correlated with whether
an individual at the ballot box voted on the referendum and also how they voted. Table 3.1 indicates
that about a quarter of the referenda were placed on the November ballot in presidential election
years while 18.7% of township referenda and 15.4% of county/municipality/school district referenda
occurred in the November ballot in gubernatorial election years. The percentage diﬀerence between
presidential and gubernatorial elections is primarily attributable to the fact that there were three
presidential and only two gubernatorial elections during the time period analyzed and not due
to a signiﬁcant diﬀerence in the number of referenda appearing on a given election.11 However,
these percentages are signiﬁcantly greater than the percent of referenda that appear on the primary
elections for president and governor. Over 40% of the referenda appear on the ballot in odd number
10If an existing millage is set to expire, the taxing authority may request a “renewal levy” to extend a tax of the
same millage rate for a longer period of time. The taxing authority may submit the renewal levy at any election during
the last year in which the tax is collected, or at the general election of the previous year.

11Presidential elections took place in 2004, 2008 and 2012 while gubernatorial elections took place in 2006 and

2010.

63

years when there is neither a presidential nor gubernatorial election (general or primary). These
numbers indicate that while slightly more referenda appear during general elections in the even years,
when a president or governor is chosen, referenda are much more likely to appear in a November
general election than in a March or May primary election.12 Table 3.1 indicates that 84.3% of
township referenda appear on a November election while 71.2% of county/municipality/school
district referenda appear on a November election. Since voter composition diﬀers in presidential
and gubernatorial elections as well as across general and primary elections, these percentages seem
particularly worthy of note.

4.3 Empirical Methodology and Results

To credibly estimate how ballot location aﬀects whether and how people vote, contest hetero-
geneity must be accounted for. As discussed by A&N, much of the existing literature involving the
eﬀect of ballot location just correlates ballot length and aggregate votes which is problematic if the
types of contests vary, in terms of aspects like saliency and importance, with ballot location and
length. While some of this literature attempts to control for speciﬁc characteristics of the contest,
this is often diﬃcult due to the limited set of observables and the signiﬁcant diﬀerences across
contests. A major advantage of the A&N paper is that they are able to convincingly address contest
heterogeneity because the ballot location of a given contest varies across local jurisdictions in San
Diego due to other local contests on a jurisdiction’s ballot. Therefore, they are able to use within
contest variation to identify the eﬀect of ballot location. Since rank, our measure of ballot location,
does not vary across a referendum, we are unable to use within referendum variation to identify
the eﬀect of ballot location.
Instead of including referendum ﬁxed eﬀects, we address contest
heterogeneity by focusing on tax referenda and controlling for a rich set of referendum observables.
Along with contest heterogeneity, it is important to control for jurisdiction and election het-
erogeneity. Tastes/preferences for and importance of a particular contest are likely to vary across
voters in diﬀerent jurisdictions and this could inﬂuence not only the choice whether to vote but also
12Primary elections occur in May except when there is a November presidential election in which case they occur

in March.

64

how to vote on a contest. Similar to A&N, we address the time invariant jurisdiction characteristics
by including jurisdiction speciﬁc ﬁxed eﬀects. We also account for election speciﬁc heterogeneity
by controlling for election characteristics that do not vary across jurisdictions. These election
characteristics, such as whether a referendum is held during a presidential election, likely inﬂuence
the type of individuals who decide to go to the ballot box.13 Unlike A&N, we allow the eﬀect of
these election characteristics on voter composition to vary across heterogeneous jurisdictions.

In an eﬀort to address the contest, jurisdiction and election heterogeneity in Ohio tax referenda,

we estimate the following speciﬁcation for the 3,504 township referenda:

Sharere = αt + β1(Ranke) + β2(Renr) + βr Xr + β j X j + βe Xe + re

(1)

where Sharere is the share of voters who did not vote on referendum r in election e; Ranke
equals one through four; Renr equals one if referendum r involved a tax renewal; Xr is the set of
referendum characteristics (purpose of the tax, type of tax being proposed, and the tax rate); Xj is
the set of township/jurisdiction speciﬁc indicator variables; and Xe is the set of election speciﬁc
characteristics (number of state issues, presidential election, gubernatorial election, presidential
primary, gubernatorial primary, and election month).

Column 2 of Table 3.2 provides estimates from the above speciﬁcation while Column 1 provides
estimates when jurisdiction ﬁxed eﬀects are not included. In both, the positive coeﬃcient estimate
associated with rank, while not statistically signiﬁcant, is consistent with much of the literature
on roll-oﬀ and suggests that voters are less likely to vote on referenda that are further down the
ballot.14 The estimate of β2 is close to zero, indicating that the fraction of individuals at the ballot
13A&N do not need to include election characteristics that do not vary across jurisdictions because they use within
contest variation and a contest appears on the same election date for all jurisdictions. Therefore, including contest ﬁxed
eﬀects allow A&N to control for contest as well as election heterogeneity.

14While the robust standard errors result in the rank coeﬃcient not to be statistically signiﬁcant, the t-statistics are
greater than one and the coeﬃcients are statistically signiﬁcant at conventional levels if diﬀerent assumptions are made
in terms of the error terms (such as being uncorrelated across referenda). One potential reason these estimates are not
precisely estimated is that rank is a noisy measure of where the referendum occurs on the ballot.

65

box who choose not to vote does not depend on whether or not the referendum renews an existing
tax.

To the speciﬁcation denoted in Equation (1), we also add three election speciﬁc covariates that
characterize voter composition: fraction of voters over 65 years of age, fraction of voters who are
registered democrats, and fraction of voters who are registered republicans. Voters’ preferences for
a tax referendum are likely to vary with both age, whether they are registered with a party and, for
those registered, which party they are aﬃliated with. While the township ﬁxed eﬀects account for
election invariant diﬀerences in age and party aﬃliation across townships, they do not account for
the variation across elections in voter composition within a township. Figure 4.1 depicts signiﬁcant
variation in these voter characteristics across elections. The coeﬃcient estimates associated with
these voter composition measure are in Column 3 of Table 3.2. They indicate that the fraction not
casting a referendum vote decreases with the fraction of voters over 65 years of age and increases
with registered voters (both democrat and republican). The eﬀect of the fraction of voters over
65 is especially large with a one standard deviation increase (6 percentage points) resulting in a
decrease in the fraction voting on the township referendum of 4.2 (6 times 0.6953) percentage
points. This implies that the percent of voters over 65 years old who choose not to vote on the
township tax referendum averages 14.57 and this percentage is 15.27 for voters under 65 years of
age.15 Interestingly, adding these voter composition covariates do not appreciably change the rank
coeﬃcient estimate.

We next consider whether the eﬀect of ballot position on the decision to vote on the referendum
varies across voter types by interacting our voter composition measures with our rank variable. The
estimates associated with these interaction terms in Column (4) suggest that how ballot location
aﬀects the probability of voting does not vary signiﬁcantly based on fraction over age 65 or party
aﬃliation. More interestingly, when interactions between renewal and percent over 65 years of age
15The percent of voters choosing not to vote on the referendum is a weighted average of the percentage among voters
over age 65 and the percentage among voters under age 65. Table 3.2 indicates that the fraction of voters choosing
not to vote on the township referendum averages 0.151 and the fraction of voters over 65 years of age averages 0.239.
Therefore, the coeﬃcient of -0.6953 in Table 2.3 implies that the percent of voters over 65 years of age who don’t vote
on the township referendum is 15.1-(1-0.239)(0.6953) = 14.57 and this percentage for voters under 65 years of age is
14.57+0.6953 = 15.27.

66

and between renewal and percent registered democrat/republican are added as covariates, the eﬀect
of a tax renewal on the probability of voting appears to vary dramatically with voter age and party
registration. The statistically signiﬁcant coeﬃcient estimates in Column (5) indicate that voters
over age 65 are much more likely than younger individuals to vote on a referendum involving a
tax renewal relative to one involving a new tax and voters registered to a particular political party
(democrat or republican) are much less likely to vote on a referendum involving a tax renewal
relative to one involving a new tax. The eﬀect of the referendum involving a tax renewal on the
decision of a speciﬁc voter type to vote on the referendum appears not to depend on where the
referendum appears on the ballot, based on the estimates in Column (6) of Table 3.2.

The estimates in Table 3.2 suggest that the prior ﬁndings documenting roll-oﬀ are not the result
of unobservable heterogeneity in voter characteristics. However, the estimates do raise concern that
unobserved heterogeneity in voter composition may be a contributing factor in A&Ns empirical
analysis testing for a status quo bias. If voters over the age of 65 are less likely and those who
register a party aﬃliation are more likely to vote on a referendum involving a renewal (where passing
would be maintaining the status quo), and if these voter composition measures are correlated with
the contests location on a ballot, then the evidence consistent with status quo bias could actually
be attributable to unobserved voter heterogeneity. The large within-jurisdiction, across-election
variation in these voter characteristics (See Figure 1) causes further concern.

To provide further evidence that unobservable voter heterogeneity may be responsible for the
results attributable to status quo bias, we estimated the same speciﬁcation in Table 3.2 with share
yes votes (i.e., yes votes divided by total votes) as the dependent variable while still using only
township referenda. Table 3.3 contains these estimates which reinforce our concern regarding voter
heterogeneity. The estimates in the ﬁrst three columns suggest that the fraction voting yes: (i)
increases as the referendum is located further down ballot; (ii) is greater for referenda involving
a tax renewal; and (iii) increases with the fraction of voters over the age of 65. The estimates in
Column (4) suggest that, like the decision to vote on the referendum, how ballot location aﬀects the
probability of voting does not vary signiﬁcantly based on fraction over age 65 or party aﬃliation.

67

While the estimates in Column (5) indicate that the eﬀect of a referendum involving a tax renewal
on the fraction voting yes does not vary appreciably with voter party aﬃliation, the statistically
signiﬁcant estimate of 0.2899 associated with the interaction of percent over 65 and tax renewal
referendum indicates that senior citizens, relative to younger voters, are almost 30 percent more
likely to vote yes for a tax renewal than a new tax. The estimates in Column (6) of Table 3.3 indicate
that the eﬀect of a referendum involving a tax renewal on the share voting yes does not depend on
where the referendum appears on the ballot for diﬀerent age and party aﬃliation voters.

Table 3.4 contains estimates from the same speciﬁcations as in Table 3.3 using the 5,286 county,
school district and municipality referenda. These estimates indicate that when we consider what
aﬀects the share who vote yes in these other jurisdictions, similar results are obtained as for the
township referenda. Speciﬁcally, Table 3.4 indicates that the share voting yes is greater for items
that are further down the ballot and for referenda that involve a tax renewal. The estimates in
Columns (5) and (6) suggest that, like for township referenda, senior citizens are much more likely
to vote yes for a tax renewal than a new tax compared to younger voters but that this diﬀerence does
not depend on where the referendum is located on the ballot. The portions of voters aﬃliated with
the democratic and republican parties appears not to aﬀect share of yes votes. Columns (3) and (4)
indicate that, unlike township referenda, the share voting yes decreases with the fraction of voters
over the age of 65 and this aﬀect is greater the further down the ballot the referendum is located.

4.4 Conclusion

A broad range of empirical research suggests that the sequence of choices inﬂuences an indi-
vidual’s decisions and use the concept of “choice fatigue” to explain this phenomenon. Some of
the most interesting papers use ballot location to ascertain how choice fatigue aﬀects not only the
decision to vote but also how an individual votes. The best paper in this area is arguably Augenblick
and Nicholson [2016] who look at contests in San Diego County between 1992 and 2002. They
ﬁnd that the lower the contest is on a ballot, the larger fraction of individuals will choose not to
vote and, for those who do vote, more will vote no on a proposition and more will vote for the

68

ﬁrst candidate list. They argue that this result is due to a “status quo bias” that is more prevalent
when an individual experiences choice fatigue. Their identiﬁcation strategy is quite ingenious but
relies on the assumption that unobserved heterogeneity in voter composition within a jurisdiction
across elections is not correlated with either ballot location and voter preferences. Augenblick and
Nicholson’s empirical analysis considers this selection issue but, unfortunately, they do not have
voter level demographic characteristics to directly account for voter heterogeneity.

This study’s contribution to the literature is based on our ability to directly account for the
heterogeneity in voters’ age and political aﬃliation. Using election results and voter information
for local Ohio tax referenda from 2004 to 2012, our results suggest that controlling for voter age
and political aﬃliation is important when explaining how ballot location eﬀects whether and how
individuals’ vote. While Augenblick and Nicholson results suggesting status quo bias may not
be attributable to unobserved voter heterogeneity in their San Diego County contest data, voter
heterogeneity clearly has important implications for how people vote on Ohio tax referenda.

69

APPENDICES

70

APPENDIX A

NOTE ON PROJECT LEVEL STATISTICS

Table 2.8 shows summary statistics at the project level. The ﬁrst row summarizes number of work
locations involved in the project (projects along a stretch of road may be listed as separate work
locations in the ﬁle). The third and fourth rows summarize number of house districts and number
of senate districts that each project overlaps, while the ﬁfth and sixth rows show number of areas
of intersection between pre-redistricting and post-redistricting districts that each project overlaps.1
Estimated total cost, summarized in the seventh row, is the closest available approximation to the
value or size of the project. Average estimated total cost per project is about $2.36M.

The remainder of the table summarizes average estimated total cost and number of projects
within diﬀerent categories of sponsoring agency and type of work. I have created ten categories
grouping diﬀerent sponsors of the projects. Most projects list the ODOT itself as a sponsor, while a
number list one of the ODOT’s twelve districts along with one of ﬁve funding areas: bridges, traﬃc,
maintenance, production, and planning. Some list a municipality or township2 (I have grouped
these together as “municipality sponsor”) while others list a county as the sponsor (I have grouped
all other special purpose local governments in the state, such as port authorities, with counties).
Finally, some list a non-DOT state agency as a sponsor, or an agency in another state (“other state
sponsor”). The ﬁfth panel shows average cost and number of projects by primary work area. I
created the work area categories by grouping expressions from a verbal description of the primary
work area for each project: for example, the intersection variable groups the string “INTER” and
thus includes projects involving an intersection or interchange.

1I describe how I construct geographic area of overlap between old and new districts in the text.
2According to ODOT employees working with these shapeﬁles, the shapeﬁles I am using include all projects in
the state with some state or federal funding. Thus, a project sponsored by a county or municipality will be included if
it is partly funded by the state or federal government.

71

APPENDIX B

TABLES

Table 2.1: Summary Statistics, by Party-Area of Intersection

Estimated total cost

Number of projects

Number of large projects

Expenditure, large projects (Thousands)

Expenditure, Ohio is main sponsor (Thousands)

Republican

County & township public works exp. (Thousands)

Local government capital outlay (Thousands)

Municipal transportation spending (Thousands)

Gas tax rev. dist., municipalities (Thousands)

Gas tax rev. dist., townships (Thousands)

Average millage rate devoted to roads

Employment (Thousands)

Employed population (Thousands)

Population (Thousands)

Suﬃciency Rating, bridges

Daily vehicle miles traveled, non-local roads

Observations

Mean

(Standard Deviation)

Total
6772.5
(17821.6)

Democratic Republican

6460.2
(17780.8)

7024.3
(17855.4)

22.52
(42.99)
2.980
(6.052)
3840.1
(16137.2)
4908.3
(16553.0)

38.44
(69.64)
4.348
(8.755)
3011.8
(12361.5)
5611.3
(17380.7)

31.34
(59.76)
3.737
(7.697)
3381.4
(14174.8)
5297.6
(17017.3)

0
(0)

2200.6
(3017.6)
6542.6
(7136.0)
3765.3
(5649.3)
945.5
(855.3)
154.1
(343.9)
0.423
(0.808)
19.61
(21.81)
16.63
(14.17)
37.17
(31.52)
69.64
(11.88)
56.22
(49.13)
1419

1
(0)

4295.6
(4306.8)
5923.9
(7580.0)
2713.7
(3420.0)
867.5
(883.2)
435.9
(526.3)
1.016
(1.117)
18.30
(17.45)
20.82
(16.24)
45.45
(35.28)
72.87
(8.734)
71.26
(59.25)
1761

0.554
(0.497)
3339.8
(3914.7)
6206.2
(7385.8)
3193.5
(4602.9)
903.1
(871.3)
307.3
(473.5)
0.746
(1.031)
18.90
(19.57)
18.91
(15.47)
41.75
(33.90)
71.42
(10.39)
64.40
(55.37)
3180

Years Available

2007-2018

2007-2018

2007-2018

2007-2018

2007-2018

2007-2018

2007-2015

2007-2015

2007-2015

2007-2015

2007-2015

2007-2015

2007-2015

2007-2015

2010

2007-2015

2007-2015

There are 265 areas of intersection between old and new districts over twelve years (2007-2018), for a total
of 3180 observations. Control variables are lagged one year in the regressions and not available for all years
- table shows the years when control variables are used in the regressions.
†This includes only distributions from Fund 7060, which weakly increased with number of registered vehicles
in the township. Townships also received distributions from Fund 7068, but these were allocated evenly
among townships and were not available for all years.

72

Table 2.2: Summary Statistics, by Party-District

Estimated total cost, (Thousands)

Number of projects

Number of large projects

Expenditure, large projects (Thousands)

Expenditure, Ohio is main sponsor (Thousands)

Republican

County & township public works exp. (Thousands)

Local government capital outlay (Thousands)

Municipal transportation spending (Thousands)

Gas tax rev. dist., municipalities (Thousands)

Gas tax rev. dist., townships (Thousands)

Average millage rate devoted to roads

Employment (Thousands)

Employed population (Thousands)

Population (Thousands)

Suﬃciency Rating, bridges

Daily vehicle miles traveled, non-local roads

Observations

Mean

(Standard Deviation)

Democratic Republican
17747.1
(27370.2)

18670.1
(30382.2)

Total
18128.5
(28644.6)

65.08
(76.66)
8.611
(9.968)
11097.9
(28526.7)
14185.0
(28828.5)

0
(0)

5418.5
(4921.2)
16110.1
(13435.7)
9271.5
(9165.3)
2328.2
(1159.1)
379.3
(693.8)
0.507
(0.732)
48.29
(33.30)
40.94
(17.49)
107.4
(9.803)
70.25
(6.854)
138.4
(61.20)

491

97.12
(136.4)
10.98
(15.65)
7609.4
(19684.2)
14177.2
(27108.8)

1
(0)

8880.8
(6904.5)
12247.3
(11488.9)
5610.4
(5128.7)
1793.5
(1230.3)
901.3
(925.4)
0.962
(0.760)
37.84
(23.87)
43.05
(21.58)
114.8
(9.230)
73.43
(6.341)
147.3
(78.17)

697

83.88
(116.5)
10.00
(13.64)
9051.2
(23792.4)
14180.4
(27820.4)

0.587
(0.493)
7449.9
(6392.1)
13843.8
(12471.4)
7123.5
(7304.3)
2014.5
(1229.4)
685.6
(875.7)
0.767
(0.781)
42.16
(28.61)
42.18
(20.01)
111.8
(10.15)
72.06
(6.749)
143.7
(71.75)
1188

Years Available

2007-2018

2007-2018

2007-2018

2007-2018

2007-2018

2007-2018

2007-2018

2007-2015

2007-2015

2007-2015

2007-2015

2007-2015

2007-2015

2007-2015

2010

2007-2015

2007-2015

There are 99 districts over twelve years (2007-2018), for a total of 1188 observations. Control variables are
lagged one year in the regressions and not available for all years - table shows the years when control variables
are used in the regressions.

73

Table 2.3: Representative Transitions and Relationship with Other Gov. Branches

Transition
R to R
R to D
D to R
D to D
Same Rep., GOP
Same Rep., Dem.
New Rep., Same Party, GOP
New Rep., Same Party, Dem.
Incumbent Lost General Election: GOP
Incumbent Lost General Election: Dem.
Incumbent Did not Run:† GOP
Incumbent Did not Run: Dem.
Incumbent in Diﬀerent Dist.: GOP
Incumbent in Diﬀerent Dist.: Dem.
Representative and other branches
Same Party, House and Senate
GOP, House and Senate
Dem, House and Senate
Representative Same Party as Gov.

†Includes incumbents who lost a primary election.

2008-2010; 2014-2016 2012 (Redistricting year) T-stat
1.55
-6.26
-2.74
2.43
6.45
6.42
-6.50
-5.13
0.47
2.52
1.74
2.25
-21.80
-21.37

0.46
0.13
0.08
0.33
0.18
0.12
0.28
0.21
0.01
0.00
0.10
0.07
0.26
0.26

0.51
0.04
0.04
0.41
0.39
0.31
0.12
0.10
0.02
0.02
0.14
0.12
0.00
0.00

0.80
0.51
0.30
0.51

N = 1060

0.85
0.52
0.33
0.54

N = 265

-1.62
-0.29
-1.07
-1.16

74

Table 2.4: Eﬀect of Redistricting Party Change on Road Construction Spending, 2007-2018

(1)

(2)

(3)

-3462.6∗∗
(1470.9)
-973.9
(1729.6)
-238.0
(1600.2)
-1171.1
(1112.3)
301.9
(1918.9)
-112.5
(1593.4)

-2566.5
(1612.0)
-923.0
(1811.6)
30.37
(2069.7)
32.28
(1078.3)
-5.551
(2081.5)
598.9
(1677.0)

Construction spending ($1000s)
(R to D) × 1(year > 2012)

(D to R) × 1(year > 2012)

(D to D) × 1(year > 2012)

(R to D) × 1(year = 2007-2010)

(D to R) × 1(year = 2007-2010)

(D to D) × 1(year = 2007-2010)

(R to R) × year

(R to D) × year

(D to R) × year

(D to D) × year

-6616.2∗∗
(3038.0)
-2087.2
(3345.2)
-5616.7∗
(3088.8)
4028.9∗∗
(2019.8)
1111.3
(2386.2)
6087.8∗∗
(2825.3)
48.34
(153.5)
1385.7∗∗
(556.1)
424.3
(920.5)
1914.7∗∗∗
(706.9)

0.110
(0.580)
0.620
(0.376)
-10.85
(13.38)
-0.979
(35.25)
-1007.4
(897.5)
-189.1
(297.3)
-213.4
(337.1)
-92.19∗∗∗
(30.19)
69.32
(97.13)

Yes
Yes
Yes
Yes
OLS
2500
0.013

(4)

-0.659∗∗
(0.256)
-0.0170
(0.337)
-0.00809
(0.213)
-0.342
(0.216)
0.0660
(0.435)
-0.0193
(0.249)

(5)

-0.561
(0.361)
-0.0848
(0.426)
0.0430
(0.336)
-0.103
(0.241)
-0.156
(0.604)
0.161
(0.330)

(6)

(7)

-3440.7∗∗
(1529.2)
-934.1
(1725.6)
-324.2
(1591.3)
-1171.1
(1112.7)
301.9
(1919.5)
-112.5
(1593.9)

-2308.9
(1753.0)
-857.6
(1807.8)
171.2
(2084.5)
26.44
(1076.3)
-11.48
(2081.2)
593.8
(1678.5)

2305.7
(2363.4)
3567.8
(3264.4)

Yes
Yes
No
No
OLS
3180
0.007

2296.8
(2538.8)
2105.3
(3389.3)
0.0886
(0.569)
0.575
(0.375)
-11.58
(13.43)
-19.24
(36.95)
-1294.2∗
(759.9)
-202.1
(298.9)
-228.4
(328.4)
-86.96∗∗∗
(29.68)
68.69
(92.37)

Yes
Yes
Yes
No
OLS
2500
0.011

0.0000101
(0.0000474)
0.0000374
(0.0000306)
-0.000315
(0.00153)
-0.000198
(0.00372)
-0.534∗∗∗
(0.204)
-0.0200
(0.0264)
-0.0315
(0.0404)
-0.0463∗∗∗
(0.0137)
0.00995
(0.0111)

Yes
Yes
Yes
No

Yes
Yes
No
No

Poisson QMLE Poisson QMLE

3144

2440

-12008506

-9578191

GOP Rep., GOP Sen. year after redist. × (year > 2012)

Dem. Rep., Dem. Sen. year after redist. × (year > 2012)

L.Municipal transportation spending

L.County & township public works exp.

L.Gas tax rev. dist., municipalities (Thousands)

L.Gas tax rev. dist., townships (Thousands)

L.Average millage rate devoted to roads

L.Employment

L.Employed population

L.Suﬃciency Rating (bridges)

L.Daily vehicle miles traveled, non-local roads

Area of intersection ﬁxed eﬀects
Year ﬁxed eﬀects
Controls for demand
Party-change trends
Estimation
Observations
R2
Log Pseudolikelihood

0.0842
(0.574)
0.576
(0.374)
-11.64
(13.45)
-19.59
(37.19)
-1300.1∗
(762.1)
-202.1
(298.1)
-226.0
(335.3)
-86.27∗∗∗
(29.70)
65.49
(97.08)

Yes
Yes
Yes
No
OLS
2500
0.011

Yes
Yes
No
No
OLS
3180
0.006

Robust standard errors in parentheses - OLS standard errors are clustered at the area of intersection level. ∗ (p < 0.1), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01). The
dependent variable is estimated total cost (in thousands of dollars) of all construction projects in the area that received state funding for columns without
controls, and total cost of projects sponsored by the state for columns with controls.

75

Dependent Variable
(R to D) × 1(year > 2012)

(D to R) × 1(year > 2012)

(D to D) × 1(year > 2012)

(R to D) × 1(year = 2007-2010)

(D to R) × 1(year = 2007-2010)

(D to D) × 1(year = 2007-2010)

Area of intersection ﬁxed eﬀects
Year ﬁxed eﬀects
Controls for demand
Party-change trends
Estimation
Observations
Log Pseudolikelihood

0.0658
(0.136)
0.243∗
(0.145)
-0.0169
(0.0802)
-0.141
(0.177)
0.0602
(0.125)
-0.247∗∗∗
(0.0665)

Yes
Yes
No
No

3144
-18656

-0.478∗∗
(0.216)
-0.258
(0.260)
-0.0104
(0.177)
-0.0510
(0.262)
-0.0783
(0.198)
0.291∗
(0.176)

Yes
Yes
No
No

2964
-8082

-0.0378
(0.112)
0.0298
(0.169)
-0.167∗
(0.0951)
-0.169
(0.173)
-0.0676
(0.136)
-0.389∗∗∗
(0.0785)

Yes
Yes
Yes
No

2440
-14426

-0.430∗∗
(0.192)
-0.0839
(0.295)
0.0586
(0.168)
-0.0214
(0.263)
-0.127
(0.196)
0.306∗
(0.177)

Yes
Yes
Yes
No

2430
-6313

-0.351∗∗
(0.177)
-0.200
(0.319)
-0.298∗
(0.156)
0.140
(0.227)
0.164
(0.260)
-0.260∗∗
(0.120)

Yes
Yes
Yes
Yes

2440
-14387

-1.074∗∗∗
(0.374)
-0.282
(0.416)
-0.314
(0.229)
0.606∗
(0.368)
0.0608
(0.322)
0.654∗∗
(0.281)

Yes
Yes
Yes
Yes

2430
-6291

Table 2.5: Eﬀect of Redistricting Party Change on Number of Projects, 2007-2018

(1)

(2)

(3)

(4)

(5)

(6)

Num. Proj. Num. Large Proj. Num. Proj. Num. Large Proj. Num. Proj. Num. Large Proj.

Pois. QMLE

Pois. QMLE

Pois. QMLE

Pois. QMLE

Pois. QMLE

Pois. QMLE

Robust standard errors in parentheses. ∗ (p < 0.1), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01). The dependent variable is number of construction projects
(or number of all large construction projects) in the area that received state funding for columns without controls, and number of projects
(large projects) sponsored by the state for columns with controls. Controls include employment and employed population; public works
expenditure by counties and townships, transportation expenditure by municipalities, capital outlay by townships, cities, and counties, and
population weighted average local government property tax millage rate dedicated to roads; gas tax distributions to municipalities and to
townships; daily vehicle miles traveled; and average FHWA suﬃciency rating for bridges.

76

Table 2.6: Eﬀect of Party and Alignment with Gov. on Construction Project Funding, Compare
Voting Changes to Redistricting Changes 2007-2018

Construction expenditure ($1000s)
GOP Representative, GOP Gov.

GOP Representative, Dem. Gov.

Dem. Representative, GOP Gov.

(R to D) × 1(year > 2012)

(D to R) × 1(year > 2012)

(D to D) × 1(year > 2012)

Incumbent Representative, Gov. same party

Incumbent representative, Gov. diﬀerent party

Representative beat incumbent, Gov. same party

Representative beat incumbent, Gov. diﬀerent party

(1)

(2)

(3)

(4)

(5)

(6)

2574.3∗
(1433.6)
985.6
(1079.6)
1346.7
(1198.2)

4990.7∗∗
(2200.2)
1309.4
(1244.8)
3640.3∗
(1918.0)

2982.2
(1921.8)
1043.6
(1520.1)
1828.2
(1645.7)
-1835.1
(1356.9)
-2069.6
(2193.5)
0.536
(1473.0)

5353.1∗∗
(2682.6)
1421.3
(1788.9)
3641.7
(2297.0)
-1414.1
(1860.4)
-2550.1
(2574.9)
749.7
(1972.7)

5585.2
(3497.7)
2280.2
(1886.9)
5794.1∗
(3048.0)
-1967.8
(1371.7)
-1985.6
(2157.0)
-411.9
(1507.5)
1648.2
(1118.6)
124.4
(962.9)
3292.6∗∗
(1357.9)
-7094.4∗∗
(3500.9)

6087.3∗∗
(2869.0)
2932.5
(2310.6)
5976.9∗∗
(2483.1)
-1496.4
(1812.0)
-2324.9
(2547.0)
332.7
(2025.8)
2199.1
(1374.7)
286.8
(1233.9)
3254.8∗∗
(1407.6)
-8158.6∗
(4559.2)

Area of intersection ﬁxed eﬀects
Year ﬁxed eﬀects
Controls for demand
Estimation
Observations
R2

Yes
Yes
No
OLS
3180
0.008

Yes
Yes
Yes
OLS
2500
0.017

Yes
Yes
No
OLS
3180
0.006

Yes
Yes
Yes
OLS
2500
0.014

Yes
Yes
No
OLS
3180
0.006

Yes
Yes
Yes
OLS
2500
0.015

Standard errors, clustered at the area of intersection level, in parentheses. ∗ (p < 0.1), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01). The dependent variable
is estimated total cost of all construction projects (in thousands of dollars) in the area that received state funding for columns without controls,
and total cost of projects sponsored by the state for columns with controls. Controls include employment and employed population; public
works expenditure by counties and townships, transportation expenditure by municipalities, capital outlay by townships, cities, and counties,
and population weighted average local government property tax millage rate dedicated to roads; gas tax distributions to municipalities and to
townships; daily vehicle miles traveled; and average FHWA suﬃciency rating for bridges.

77

Construction expenditure ($1000s)
(R to D) × 1(year > 2012)

(D to R) × 1(year > 2012)

(D to D) × 1(year > 2012)

(R to R)×(Same Representative)×(year > 2012)

-1436.3
(1643.8)
1052.5
(1879.1)
-531.1
(1873.2)
5045.2∗∗
(2195.6)
(D to D)×(Same Representative)×(year > 2012) 6509.3∗∗∗
(2459.7)
60.75
(1346.1)
1533.7
(2064.0)
-42.47
(1967.2)
3066.9∗
(1744.9)
3260.3
(2826.2)

(R to R)×(Same Representative)×(year < 2011)

(D to D) × 1(year = 2007-2010)

(D to R) × 1(year = 2007-2010)

(R to D) × 1(year = 2007-2010)

(D to D)×(Same Representative)×(year < 2011)
Area of intersection ﬁxed eﬀects
Year ﬁxed eﬀects
Controls for demand
Estimation
Observations
R2
Log Pseudolikelihood

Yes
Yes
No
OLS
3180
0.010

-88.74
(1831.9)
1532.5
(2019.1)
-692.8
(2258.4)
6479.4∗∗
(2859.9)
8547.4∗∗
(3722.8)
1233.7
(1317.8)
1148.5
(2238.9)
259.7
(2019.2)
3039.3∗
(1702.2)
4199.5
(3392.8)

Yes
Yes
Yes
OLS
2500
0.016

-3728.5
(3445.2)
780.2
(3821.3)
-7337.6∗
(3838.0)
8004.6
(5078.9)
12154.0∗∗
(4889.4)
4837.8∗
(2577.7)
1868.9
(2962.0)
6724.1∗
(4062.7)
1622.1
(3344.5)
636.3
(4946.6)

Yes
Yes
Yes
OLS
2500
0.020

-0.405
(0.311)
0.237
(0.381)
-0.107
(0.347)
0.460∗
(0.266)
0.630∗∗
(0.305)
-0.0976
(0.272)
0.311
(0.466)
-0.0539
(0.407)
0.445∗
(0.250)
0.512
(0.419)

Yes
Yes
No

3144

-0.100
(0.460)
0.358
(0.501)
-0.154
(0.527)
0.791∗∗
(0.398)
1.142∗∗
(0.518)
0.239
(0.304)
0.182
(0.632)
0.123
(0.430)
0.636∗∗
(0.283)
0.745
(0.563)

Yes
Yes
Yes

2440

Table 2.7: Eﬀect of Redistricting Party Changes on Construction Project Funding, Controlling for
Areas that Changed Representative 2007-2018

(1)

(2)

(3)

(4)

(5)

Robust standard errors, clustered at the area of intersection level for OLS regressions, in parentheses. ∗ (p < 0.1), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).
Controls include employment and employed population; public work expenditure by counties and townships, transportation expenditure by
municipalities, capital outlay by townships, cities, and counties, and population weighted average local government millage rate dedicated to
roads; gas tax distributions to municipalities and to townships; daily vehicle miles traveled; and average FHWA suﬃciency rating for bridges.

Poisson QMLE Poisson QMLE

-11929099

-9405760

78

Table 2.8: Summary Statistics, Projects

No. work areas per project
No. house districts per project
No. house district overlap areas per project

Estimated total cost
Spons. by County
Spons. by other state
Spons. by ODOT
Spons. by ODOT Dist. Planning
Spons. by ODOT Dist. Maintenance
Spons. by ODOT Dist. Bridges
Spons. by ODOT Dist. Production
Spons. by ODOT Dist. Traﬃc
Spons. by Municipality
Landscaping or Noise Wall
Safety Features
Signs or signals
Sidewalk
Construction
Maintenance
Intersection/Interchange
Culvert, Drainage or Erosion
Widening
Road Surface
Bridge or Tunnel
Miscellaneous
Observations

Mean (Count)

3.248
1.486
1.747

Std. Dev. (Count)

Number
9780
9780
9780
Mean (Cost, $1000s) Std. Dev. (Cost $1000s) Number

6.887
1.389
2.380

10022.4
2744.0
7888.9
11227.5
9147.0
1254.6
575.1
18329.7
237.0
4779.6
1362.3
2518.1
1413.9
3023.2
43473.1
8363.3
9807.4
568.5
16834.7
2213.5
10981.1
27298.6

9780
1305
16
4493
1333
235
20
753
11
1646
155
306
388
291
105
1255
387
518
328
3556
2186
250

2357.8
1437.5
4420.7
2641.8
2590.8
867.1
464.4
3359.3
419.1
1848.1
964.3
1140.8
823.9
904.9
29798.8
2055.6
3612.5
463.8
7871.4
1794.1
2132.1
3924.7
9780

79

Table 3.1: Descriptive Statistics

Variable

Income Tax
Has income tax
Income tax base (millions)
Municipal income tax rate
Property Tax
Real property value (millions)
Res. and ag. value (millions)
Comm. and ind. value (millions)
Tangible personal property value (millions)
Municipal property tax rate (mills)
Prop. tax rate (mills), overlapping jurisdictions
Municipality Characteristics
Population
Employment
Land area
Has a charter?
Has council-manager gov.
Layoﬀs
Mass layoﬀ reported (%)
Number of mass layoﬀs occurring
Number of jobs lost in mass layoﬀs
Proportional job loss (= jobs lost
population
Revenue
Property tax revenue (100,000s)
Income tax revenue (100,000s)†
Revenue from services (100,000s)
Revenue from ﬁnes and fees (100,000s)
Intergovernmental rev. (100,000s)
Own source revenue (100,000s)
Total revenue (100,000s)
Expenditure
General Government (100,000s)
Public Safety (100,000s)
Health (100,000s)
Roads and Utilities (Infrastructure)
Leisure and Community Environment (100,000s)
Cap. Outlays (100,000s)
Debt Expenditure (100,000s)
Other Expenditure (100,000s)
Total Expenditure (100,000)††

)

Mean

Std. Dev.

0.75
445
1.07

195
139
56.3
17.7
7.11
54.5

11326
6041
5.62
0.37
0.16

9.7
1.69
327
0.03

24.3
58.9
7.71
5.55
25.6
96.9
123

18.8
46.4
3.25
17.0
14.9
19.3
21.8
0.24
134

0.43
1620
0.74

665
416
257
72.2
4.79
25.9

41605
25181
12.32
0.48
0.37

29.6
1.94
1270
0.16

147.9
300.9
38.50
27.67
125.1
464.6
588

63.5
235.5
24.9
91.8
94.6
97.4
234.6
7.15
650

N

10115
7275
10115

10115
10115
10115
8330
10115
10115

10115
7140
10115
10115
10115

10085
983
983
983

8925
8925
8925
8925
8925
8925
8925

8925
8925
8925
8925
8925
8925
8925
8925
8925

80

Table 3.1 (cont’d)

Debt, Assets, and Fund Balance (Cities Only)
General Obligation Bonds (100,000s)
2172
Notes and loans payable (100,000s)
2172
Unreserved Fund Balance (100,000s)
2172
Cash and Investments (100,000s)
2172
General Obligation Bonds $> $ 0
2172
Notes and Loans Payable $> $ 0
2172
Unreserved Fund Balance $> $ 0
2172
General Obligation Bonds per capita
2172
Notes and loans payable per capita
2172
Unreserved Fund Balance per capita
2172
Cash and Investments per capita
2172
† This row includes all municipalities, not only municipalities with an income tax.
†† Total revenue exceeds total expenditure by roughly $1.08M. Debt principal repayment,
averaging $960,000 per year, is included in the total expenditure ﬁgure, while debt issue
is not included on the revenue side. The $120,000 diﬀerence remaining implies that net
borrowing increased $120,000 per municipality between 1999 and 2012.

748
59
208
462
0.38
0.49
0.25
472
190
1225
1911

145
19
91
158
0.83
0.62
0.93
353
87
520
762

81

Table 3.2: The Eﬀect of Job Loss in a Mass Layoﬀ on Tax Base

Inc. tax base, log Employment, log Real prop. value, log

Proportional job loss, lead

Proportional job loss

Proportional job loss, lag

Proportional job loss, two lags

Proportional job loss, three lags

Proportional job loss, four lags

L5.Population, log

L5.Res. and ag. value, log

L5.Real prop. value, log

L5.Aggr. prop. tax rate (mills)

L5.Muni. inc. tax rate

Observations
R2

0.293∗∗∗
(0.112)
0.0805
(0.111)
-0.459∗∗∗
(0.101)
-0.163∗∗∗
(0.0408)
-0.129∗∗∗
(0.0409)
-0.163∗∗∗
(0.0402)
0.188∗∗∗
(0.0399)
-0.0984∗∗
(0.0406)
0.392∗∗∗
(0.0554)
-0.000663
(0.000443)
-0.0869∗∗∗
(0.0209)

5172
0.174

0.283∗∗
(0.127)
0.127
(0.125)
-0.264∗∗
(0.114)
-0.0989∗∗
(0.0459)
-0.0949∗∗
(0.0461)
-0.0883∗
(0.0453)
0.217∗∗∗
(0.0450)
-0.0258
(0.0457)
0.218∗∗∗
(0.0625)
0.0000106
(0.000500)
0.0592∗∗
(0.0235)

5160
0.040

0.00552
(0.0227)
0.00477
(0.0222)
-0.00827
(0.0227)
-0.0173
(0.0229)
-0.0325
(0.0248)
-0.0486
(0.0377)
0.208∗∗∗
(0.0404)

-0.000838∗∗
(0.000354)
-0.0113
(0.0131)

7140
0.612

Estimated by ﬁxed-eﬀects OLS, with a full set of year dummies. Robust standard errors,
clustered at the municipality level, in parentheses. ∗ (p < 0.1), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).

82

Table 3.3: The Eﬀect of Job Loss in a Mass Layoﬀ on Tax Rates

Muni. inc. tax rate Municipal property tax rate (mills)

0.570
(0.561)
0.290
(0.506)
0.0983
(0.363)
-0.0158
(0.0673)
0.0530
(0.0478)
-0.0333
(0.0856)
0.00667
(0.00509)
-0.401
(0.735)
-0.172
(0.510)
-0.260
(0.344)
-0.341
(0.224)
7140
0.034

Proportional job loss, lead

Proportional job loss

Proportional job loss, lag

Proportional job loss, two lags

Proportional job loss, three lags

Proportional job loss, four lags

L5.Aggr. prop. tax rate (mills)

L5.Real prop. value, log

L5.Res. and ag. value, log

L5.Population, log

L5.Muni. inc. tax rate

Observations
R2

0.127
(0.0903)
0.175∗
(0.0969)
0.197∗
(0.111)
0.0421∗∗
(0.0207)
0.0390∗
(0.0201)
0.0492
(0.0329)
0.000210
(0.000625)
-0.0272
(0.0825)
0.00641
(0.0530)
-0.00992
(0.0471)

7140
0.113

Estimated by ﬁxed-eﬀects OLS, with a full set of year dummies. Robust standard errors, clustered at the
municipality level, in parentheses. ∗ (p < 0.1), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).

83

Table 3.4: The Eﬀect of Job Loss in a Mass Layoﬀ on Revenue

Inc. tax rev., log

Prop. tax rev., log

Services, fees, and ﬁnes rev.

Intergovernmental rev. Total rev., log

0.232
(0.554)
-0.237
(0.396)
-0.935∗∗
(0.412)
-0.0492
(0.0738)
-0.00366
(0.0525)
0.00100
(0.0379)

Yes
Yes

0.120
(0.179)
0.0509
(0.233)
-0.291
(0.186)
-0.131
(0.0819)
-0.0903
(0.0724)
-0.115
(0.0829)

Prop. job loss, lead

Prop. job loss

Prop. job loss, lag

Prop. job loss, two lags

Prop. job loss, three lags

Prop. job loss, four lags

0.402∗∗∗
(0.127)
0.236
(0.149)
-0.211
(0.199)
-0.115∗∗∗
(0.0434)
-0.105∗
(0.0623)
-0.126∗∗
(0.0563)

0.160
(0.300)
-0.353
(0.318)
-0.520∗
(0.294)
-0.198
(0.131)
-0.160
(0.113)
-0.104
(0.124)

0.238
(0.268)
0.115
(0.340)
0.166
(0.287)
-0.0322
(0.0360)
-0.0542∗∗∗
(0.0192)
-0.0458
(0.0320)

7104

Municipality ﬁxed-eﬀects
Year dummies
Estimation
Observations
R2
Includes controls for log of real property value, log of residential and agricultural property value, log of population, municipal income
tax rate, property tax rate of overlapping jurisdictions, and municipal property tax rate, each lagged by ﬁve years. Robust standard errors,
clustered at the municipality level for OLS equations, in parentheses. ∗ (p < 0.1), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).

Yes
Yes
OLS
7140
0.063

Yes
Yes
OLS
5410
0.197

Yes
Yes
OLS
7140
0.097

Pois. QMLE

Yes
Yes

Pois. QMLE

7140

84

Prop. job loss, lead

Prop. job loss

Prop. job loss, lag

Prop. job loss, two lags

Prop. job loss, three lags

Prop. job loss, four lags

Observations
Municipality ﬁxed-eﬀects
Year dummies
Estimation
R2

0.319
(0.240)
0.267
(0.232)
-0.0993
(0.202)
-0.0545
(0.0364)
-0.0167
(0.0322)
-0.0526
(0.0469)

7140
Yes
Yes

Pub. Safety
0.226∗∗
(0.0967)
0.251∗∗
(0.101)
0.140∗
(0.0787)
-0.0167
(0.0119)
-0.0129
(0.0103)
-0.0399
(0.0270)

7116
Yes
Yes

0.348
(0.323)
-0.339
(0.415)
-0.466∗∗
(0.204)
-0.00259
(0.0263)
-0.0265
(0.0183)
-0.0272
(0.0284)

7140
Yes
Yes

1.701∗∗∗
(0.558)
0.891
(0.558)
0.811∗∗
(0.394)
-0.0517
(0.158)
0.0687
(0.0707)
0.0563
(0.106)
6960
Yes
Yes

0.868
(0.938)
1.124
(0.867)
-0.719
(0.713)
-0.149
(0.108)
-0.121
(0.0984)
-0.00225
(0.0701)

7020
Yes
Yes

Health
-0.424
(0.571)
-1.202∗
(0.725)
-1.318∗∗
(0.646)
-0.136
(0.106)
-0.0105
(0.0251)
0.0662∗∗∗
(0.0241)

6324
Yes
Yes

Debt Exp.

0.337
(1.338)
3.416∗∗∗
(0.707)
2.092
(1.309)
0.107
(0.342)
0.00957
(0.330)
0.423
(0.349)
6696
Yes
Yes
OLS

Total
0.120
(0.229)
0.379
(0.339)
-0.107
(0.166)
-0.136
(0.108)
-0.0896
(0.0777)
-0.0623
(0.0790)

7140

0.038

Table 3.5: The Eﬀect of Job Loss in a Mass Layoﬀ on Expenditure

Gen. Gov.

Infrastructure Leis. and comm. Cap. outlays

Pois. QMLE Pois. QMLE Pois. QMLE

Pois. QMLE

Pois. QMLE Pois. QMLE

Includes controls for log of real property value, log of residential and agricultural property value, log of population, municipal income tax rate,
property tax rate of overlapping jurisdictions, and municipal property tax rate, each lagged by ﬁve years. Robust standard errors, clustered at the
municipality level for OLS regression, in parentheses. ∗ (p < 0.1), ∗∗ (p < 0.05), ∗∗∗ (p < 0.01).

85

Table 3.6: The Eﬀect of Job Loss in a Mass Layoﬀ on Debt and Asset Balance

G.O. Bonds per cap. Notes and loans payable per cap. Unres. Fund Bal. per cap. Cash and Invest. per cap.

3166.8∗∗∗
(1128.4)
2160.1∗∗∗
(629.9)
-2338.0∗∗∗
(465.7)
-32.28
(23.08)
-41.00
(31.52)
-48.29∗
(27.65)
2172
0.060

2299.3∗∗∗
(810.6)
1668.5∗∗
(753.9)
297.2
(653.2)
-70.73∗∗∗
(22.95)
-53.49∗∗∗
(17.81)
-75.14
(49.32)
2172
0.021

Prop. job loss, lead

Prop. job loss

Prop. job loss, lag

Prop. job loss, two lags

Prop. job loss, three lags

Prop. job loss, four lags

Observations
R2

-1062.8∗
(551.1)
-1051.3∗∗
(492.3)
-1283.0∗∗∗
(351.0)
26.81
(17.58)
16.52
(19.11)
-29.92
(55.21)
2172
0.053

1203.3∗∗
(464.9)
1631.3∗∗∗
(321.5)
73.31
(220.7)
-5.808
(4.119)
9.830∗∗∗
(3.500)
19.77∗
(11.22)
2172
0.042

Estimated by ﬁxed-eﬀects OLS, with a full set of year dummies. Includes controls for log of real property value, log of residential and
agricultural property value, log of population, municipal income tax rate, property tax rate of overlapping jurisdictions, and municipal
property tax rate, each lagged by ﬁve years. Robust standard errors, clustered at the municipality level, in parentheses. ∗ (p < 0.1), ∗∗ (p <
0.05), ∗∗∗

86

Table 4.1: Summary Statistics

Mean (Standard Dev.)

Township County, School District,

Municipality

734
(1394)
490
(1004)
0.620
(0.107)
0.858
(0.350)
0.151
(0.184)
2.495
(1.077)
0.483
(0.5)
0.550
(0.498)

Yes Votes

No Votes

Share Yes [Yes Votes/Total Votes]

Referendum Passes

Fraction of Voters not Casting Referendum Vote

Rank

Tax Renewal

Police, Fire and EMS Indicator

Community or Senior Center Indicator

Health Services Indicator

Child Services Indicator

School Capital Expenditure Indicator

School Current Expenditure Indicator

School Emergency Expenditure Indicator

School Permanent Improvement Indicator

87

4720
(17741)
3623
(11035)
0.552
(0.119)
0.670
(0.47)

2.468
(1.124)
0.413
(0.492)
0.148
(0.355)
0.027
(0.162)
0.046
(0.21)
0.013
(0.115)
0.103
(0.304)
0.152
(0.359)
0.13
(0.337)
0.095
(0.293)

Table 4.1 (cont’d)

Mean (Standard Dev.)

Roads Indicator

Current Expenditure Indicator

Parks and Recreation Indicator

Cemetery Indicator

Proposed Property Tax

Millage Rate

Proposed Sales Tax

Sales Tax Rate

Proposed Income Tax

Income Tax Rate

Number of State Issues

Presidential Election

Presidential Primary

Gubernatorial Election

Gubernatorial Primary

0.231
(0.422)

0.108
(0.311)

1
(0)
1.52
(1.034)

2.25
(2.082)
0.287
(0.453)
0.039
(0.193)
0.187
(0.390)
0.057
(0.231)

88

Township County, School District,

Municipality

0.047
(0.212)
0.168
(0.374)
0.032
(0.176)
0.013
(0.112)
0.83
(0.376)
3.351
(2.662)
0.074
(0.262)
0.456
(0.222)
0.110
(0.312)
0.693
(0.379)
1.923
(2.089)
0.243
(0.429)
0.067
(0.25)
0.154
(0.361)
0.092
(0.289)

Township County, School District,

Municipality

0.444
(0.497)
0.712
(0.453)
0.067
(0.250)
0.222
(0.415)
0.255
(0.058)
0.18
(0.113)
0.331
(0.130)
5,186

Table 4.1 (cont’d)

Mean (Standard Dev.)

Non-Presidential and Gubernatorial Election

November Election

March Election

May Election

Fraction of Voters over 65 years of age

Fraction of Voters registered as a Democrat

Fraction of Voters registered as a Republican

Observations

0.430
(0.495)
0.843
(0.364)
0.039
(0.193)
0.118
(0.323)
0.239
(0.060)
0.155
(0.098)
0.346
(0.120)
3,504

89

Table 4.2: Share of Voters who did not Vote on Referendum

(1)

0.0197
(0.0163)
-0.0082
(0.0228)

(2)

0.0114
(0.0111)
0.0117
(0.0178)

(3)

0.0117
(0.0114)
0.0144
(0.0182)
-0.6953∗∗
(0.2852)
0.3842∗∗
(0.1611)
0.1941
(0.2058)

(4)

-0.0397
(0.038)
0.0144
(0.0177)
-0.9681∗∗
(0.4006)
0.3723∗
(0.2133)
0.0619
(0.3185)
0.0854
(0.1051)
0.0192
(0.0698)
0.0642
(0.0987)

(5)

-0.042
(0.0374)
0.0541
(0.0661)
-0.4317
(0.2992)
0.1448
(0.195)
-0.1773
(0.2772)
0.0681
(0.0976)
0.0147
(0.0674)
0.0832
(0.0962)
-1.0214∗∗
(0.3165)
0.3688∗∗
(0.1374)
0.4566∗∗
(0.1992)

Rank

Renewal Indicator

% of Voters over 65 years of age

% of Voters Registered as Democrat

% of Voters Registered as Republican

% over 65 years of age * Rank

% of Voters Registered as Democrat* Rank

% of Voters Registered as Republican* Rank

% over 65 years of age * Renewal

% of Voters Registered as Democrat* Renewal

% of Voters Registered as Republican* Renewal

Renewal*Rank

% over 65 years of age * Renewal*Rank

% of Voters Registered as Democrat* Renewal*Rank

% of Voters Registered as Republican* Renewal*Rank

Township Fixed Eﬀects
R-squared
Observations

No
0.034
3,504

Yes
0.84
3,504

Yes
0.842
3,504

Yes
0.843
3,504

Yes
0.85
3,504

(6)

-0.0215
(0.0512)
0.168
(0.1562)
-0.3546
(0.3562)
0.1699
(0.2755)
-0.1072
(0.3506)
0.0357
(0.1485)
0.0008
(0.1116)
0.0521
(0.1413)
-1.2815∗∗
(0.6164)
0.321
(0.3015)
0.3359
(0.3934)
-0.0489
(0.0528)
0.1034
(0.1972)
0.0235
(0.1156)
0.0565
(0.149)

Yes
0.85
3,504

Covariates include millage rate, number of state issues on ballot and indicators for purpose of the tax revenue, for whether
referendum stipulated bonds, for election characteristics (presidential/gubernatorial and whether a primary), and for election
month. ∗∗ indicates statistically signiﬁcant at ﬁve percent level and ∗ indicates statistically signiﬁcant at ten percent level.

90

Table 4.3: Share Yes for Township Referenda

(1)

0.0054
(0.0041)
0.1010∗∗
(0.0046

(2)

0.0087∗
(0.0047)
0.1081∗∗
(0.0056)

(3)

0.0099∗∗
(0.0046)
0.1069∗∗
(0.0055)
0.2580∗
(0.1398)
0.1317
(0.1063)
-0.0697
(0.08)

(4)

0.0183
(0.0157)
0.1067∗∗
(0.0055)
0.2438
(0.1691)
0.1523
(0.1263)
-0.0173
(0.1018)
0.0098
(0.0434)
-0.0112
(0.0348)
-0.0244
(0.0259)

(5)

0.0211
(0.0151)
0.0480∗∗
(0.02)
0.1213
(0.186)
0.2093
(0.1366)
-0.005
(0.1046)
0.0096
(0.0414)
-0.0137
(0.033)
-0.0292
(0.0254)
0.2899∗∗
(0.1188)
-0.0689
(0.0777)
0.0006
(0.0569)

Rank

Renewal Indicator

% of Voters over 65 years of age

% of Voters Registered as Democrat

% of Voters Registered as Republican

% over 65 years of age * Rank

% of Voters Registered as Democrat* Rank

% of Voters Registered as Republican* Rank

% over 65 years of age * Renewal

% of Voters Registered as Democrat* Renewal

% of Voters Registered as Republican* Renewal

Renewal * Rank

% over 65 years of age * Renewal*Rank

% of Voters Registered as Democrat* Renewal*Rank

% of Voters Registered as Republican* Renewal*Rank

Township Fixed Eﬀects
R-squared
Observations

No
0.381
3,504

Yes
0.705
3,504

Yes
0.709
3,504

Yes
0.709
3,504

Yes
0.713
3,504

(6)

0.0169
(0.02)
0.0182
(0.0528)
0.1166
(0.2197)
0.174
(0.1656)
-0.0022
(0.1294)
0.0116
(0.0598)
-0.0006
(0.0512)
-0.0296
(0.0386)
0.3399
(0.235)
-0.002
(0.1555)
-0.0052
(0.119)
0.0127
(0.0209)
-0.0204
(0.078)
-0.0274
(0.0612)
0.0008
(0.0455)

Yes
0.713
3,504

Covariates include millage rate, number of state issues on ballot and indicators for purpose of the tax revenue, for whether
referendum stipulated bonds, for election characteristics (presidential/gubernatorial and whether a primary), and for election
month. ∗∗ indicates statistically signiﬁcant at ﬁve percent level and ∗ indicates statistically signiﬁcant at ten percent level.

91

Table 4.4: Share Yes for County, School District, and Municipality Referenda

(1)

0.0028
(0.0029)
0.0994∗∗
(0.0066)

(2)

0.0039∗
(0.0022)
0.1062∗∗
(0.007)

(3)

0.0038∗
(0.0022)
0.1077∗∗
(0.0067)
-0.3871∗∗
(0.1702)
0.0571
(0.1083)
0.1322∗∗
(0.0635)

(4)

0.0280∗∗
(0.0107)
0.1083∗∗
(0.0066)
-0.1063
(0.2118)
0.0642
(0.1408)
0.1229
(0.0992)
-0.1090∗∗
(0.0512)
0.0002
(0.0289)
0.0105
(0.0247)

(5)

0.0276∗∗
(0.0105)
0.0455
(0.033)
-0.2474
(0.2097)
0.1184
(0.1408)
0.1138
(0.1035)
-0.1058∗∗
(0.0492)
0.0018
(0.0289)
0.0074
(0.0242)
0.3080∗∗
(0.1461)
-0.1032
(0.0709)
0.0177
(0.0589)

Rank

Renewal Indicator

% of Voters over 65 years of age

% of Voters Registered as Democrat

% of Voters Registered as Republican

% over 65 years of age * Rank

% of Voters Registered as Democrat* Rank

% of Voters Registered as Republican* Rank

% over 65 years of age * Renewal

% of Voters Registered as Democrat* Renewal

% of Voters Registered as Republican* Renewal

Renewal * Rank

% over 65 years of age * Renewal*Rank

% of Voters Registered as Democrat* Renewal*Rank

% of Voters Registered as Republican* Renewal*Rank

Township Fixed Eﬀects
R-squared
Observations

No
0.427
5,285

Yes
0.664
5,285

Yes
0.667
5,285

Yes
0.669
5,285

Yes
0.672
5,285

(6)

0.0352∗∗
(0.0141)
0.1257∗
(0.0726)
-0.1836
(0.2366)
0.1092
(0.1429)
0.1469
(0.1137)
-0.1322∗∗
(0.0626)
0.0133
(0.0389)
-0.0023
(0.0337)
0.0147
(0.2573)
-0.0184
(0.1491)
-0.0357
(0.1193)
-0.0273
(0.0301)
0.1114
(0.1114)
-0.0383
(0.0588)
0.0154
(0.0458)

Yes
0.673
5,285

Covariates include millage rate, number of state issues on ballot and indicators for purpose of the tax revenue, for whether
referendum stipulated bonds, for election characteristics (presidential/gubernatorial and whether a primary), and for election
month. ∗∗ indicates statistically signiﬁcant at ﬁve percent level and ∗ indicates statistically signiﬁcant at the ten percent level.

92

APPENDIX C

FIGURES

Figure 2.1: Partisan Control of Ohio Government

93

Figure 2.2: House Districts in Ohio

94

Figure 2.3: Summed Road Construction Funding, 2007-2012 vs. 2013-2018

95

Figure 2.4: Trends in Expenditure

96

Figure 3.1: Job Loss in Mass Layoﬀs as Percent of Population, Ohio Municipalities 2009

97

Figure 3.2: Percentage Change in Income Tax Base and Employment after Mass Layoﬀ

98

Figure 4.1: Voter Characteristics by Election Date and Jurisdiction

(a) Fraction of Voters over 65

(b) Fraction of Voters Registered as Republicans

(c) Fraction of Voters Registered as Democrats

99

BIBLIOGRAPHY

100

BIBLIOGRAPHY

David Albouy. Partisan Representation in Congress and the Geographic Distribution of Federal
Funds. Review of Economics and Statistics, 95(1):127–141, 2013. doi: https://doi.org/10.1162/
REST_a_00343.

Stephen Ansolabehere, James M. Snyder, and Charles Stewart. Old Voters, New Voters, and the
Personal Vote: Using Redistricting to Measure the Incumbency Advantage. American Journal
of Political Science, 44(1):17–34, 2000.

Stephen Ansolabehere, James M. Snyder, and Charles Stewart. Equal Votes, Equal Money: Court-
Ordered Redistricting and Public Expenditures in the American States. The American Political
Science Review, 96(4):767–777, 2002. doi: https://doi.org/10.1017/S0003055402000448.

Ned Augenblick and Scott Nicholson. Ballot Position, Choice Fatigue, and Voter Behaviour. The
Review of Economic Studies, 83(2):460–480, 2016. doi: https://doi.org/10.1093/restud/rdv047.
Identifying and Coping with Fiscal Emergencies in Ohio Local Gov-
doi:

International Journal of Public Administration, 27(8-9):615–630, 2004.

Jane Beckett-Camarata.

ernments.
https://doi.org/10.1081/PAD-120030258.

Christopher R. Berry, Barry C. Burder, and William G. Howell. The President and the Distribution
of Federal Spending. American Political Science Review, 104(4):783–799, 2010. doi: https:
//doi.org/10.1017/S0003055410000377.

Shaun Bowler, Todd Donovan, and Trudi Happ. Ballot Propositions and Information Costs: Direct
Democracy and the Fatigued Voter. Western Political Quarterly, 45(2):559–568, 1992. doi:
https://doi.org/10.1177/106591299204500216.

Emanuele Bracco, Ben Lockwood, Francesco Porcelli, and Michela Redoano. Intergovernmental
Grants as Signals and the Alignment Eﬀect: Theory and Evidence. Journal of Public Economics,
123:78 – 91, 2015. doi: https://doi.org/10.1016/j.jpubeco.2014.11.007.

Martin Browning and Eskil Heinesen. Eﬀect of Job Loss Due to Plant Closure on Mortality and
Hospitalization. Journal of Health Economics, 31(4):599 – 616, 2012. doi: https://doi.org/10.
1016/j.jhealeco.2012.03.001.

Anne Case. Election Goals and Income Redistribution: Recent Evidence from Albania. European
Economic Review, 45(3):405 – 423, 2001. doi: https://doi.org/10.1016/S0014-2921(00)00078-7.
Antoni Castells and Albert Solé-Ollé. The Regional Allocation of Infrastructure Investment: The
Role of Equity, Eﬃciency and Political Factors. European Economic Review, 49(5):1165 – 1205,
2005. doi: https://doi.org/10.1016/j.euroecorev.2003.07.002.

Jowei Chen. The Eﬀect of Electoral Geography on Pork Barreling in Bicameral Legislatures.
American Journal of Political Science, 54(2):301–322, 2010. doi: https://doi.org/10.1111/j.
1540-5907.2010.00432.x.

101

Howard Chernick, Adam Langley, and Andrew Reschovsky. The Impact of the Great Recession and
the Housing Crisis on the Financing of America’s Largest Cities. Regional Science and Urban
Economics, 41(4):372 – 381, 2011. doi: https://doi.org/10.1016/j.regsciurbeco.2011.04.002.

Erich Cromwell and Keith Ihlanfeldt. Local Government Responses to Exogenous Shocks in
Revenue Sources: Evidence From Florida. National Tax Journal, 68(2):339–376, 2015. doi:
dx.doi.org/10.17310/ntj.2015.2.05.

David Daley. Ratf****d: The True Story Behind the Secret Plan to Steal America’s Birthright.

Liveright, New York, 2016.

Avinash Dixit and John Londregan. Ideology, Tactics, and Eﬃciency in Redistributive Politics.

The Quarterly Journal of Economics, 113(2):497–529, 1998.

Richard Dye. State Revenue Cyclicality. National Tax Journal, 57(1):133–145, 2004. doi:

dx.doi.org/10.17310/ntj.2004.1.07.

Richard Dye and Andrew Reschovsky. Property Tax Responses to State Aid Cuts in the Recent
Fiscal Crisis. Public Budgeting & Finance, 28(2):87–111, 2008. doi: https://doi.org/10.1111/j.
1540-5850.2008.00906.x.

Leo Feler and Mine Z. Senses. Trade Shocks and the Provision of Local Public Goods. American
Economic Journal: Economic Policy, 9(4):101–43, 2017. doi: https://doi.org/10.1257/pol.
20150578.

Ronald Fisher. Property Taxes for Local Finance: Research Results and Policy Perspectives

(Reconsidering Property Taxes: Perhaps not so Bad after All.), 2009.

John N. Friedman and Richard T. Holden. Optimal Gerrymandering: Sometimes Pack, but Never
Crack. American Economic Review, 98(1):113–44, March 2008. doi: https://doi.org/10.1257/
aer.98.1.113.

Geraldine Gambale.

Cautious Opti-
mism Reﬂected. http://www.areadevelopment.com/Corporate-Consultants-Survey-Results/Q1-
2016/corporate-executive-site-selection-facility-plans-441729.shtml, 2016.

31st Annual Survey of Corporate Executives:

Geraldine Gambale. 31st Annual Survey of Corporate Executives: Conﬁdence in U.S. Economy,
Need for Investment in Infrastructure Reﬂected. http://www.areadevelopment.com/Corporate-
Consultants-Survey-Results, 2017.

GASB, Governmental Accounting Standards Board. Fund Balance: It May Not Be What You

Think It Is. The User’s Perspective, 2006.

Thomas W. Gilligan and John G. Matsusaka. Structural Constraints on Partisan Bias under the
Eﬃcient Gerrymander. Public Choice, 100(1):65–84, 1999. doi: https://doi.org/10.1023/A:
1018344022501.

102

Government Accounting Standards Board. Statement Number 54 of the Governmental Account-
ing Standards Board: Fund Balance Reporting and Governmental Fund Type Deﬁnitions.
https://www.gasb.org/jsp/GASB/Document_C/DocumentPage?cid=1176159972156&
acceptedDisclaimer=true, 2009. Accessed 7/28/2018.

Michael Greenstone, Richard Hornbeck, and Enrico Moretti. Identifying Agglomeration Spillovers:
Evidence from Winners and Losers of Large Plant Openings. Journal of Political Economy, 118
(3):536–598, 2010. doi: https://doi.org/10.1086/653714.

Daniel E Ho and Kosuke Imai. Randomization Inference With Natural Experiments. Journal of
the American Statistical Association, 101(475):888–900, 2006. doi: https://doi.org/10.1198/
016214505000001258.

Daniel E. Ho and Kosuke Imai. Estimating Causal Eﬀects of Ballot Order from a Randomized
Natural Experiment: The California Alphabet Lottery, 1978–2002. Public Opinion Quarterly,
72(2):216–240, 2008. doi: http://dx.doi.org/10.1093/poq/nfn018.

Jesse Edgerton and Andrew F. Haughwout and Rae Rosen. Institutions, tax structure and state-local
ﬁscal stress. National Tax Journal, 57(1):147–158, 2004. doi: dx.doi.org/10.17310/ntj.2004.1.
08.

Eva Johansson.

Intergovernmental Grants as a Tactical Instrument: Empirical Evidence from
Swedish Municipalities. Journal of Public Economics, 87(5):883 – 915, 2003. doi: https:
//doi.org/10.1016/S0047-2727(01)00148-7.

Andreas Kuhn, Rafael Lalive, and Josef Zweimüller. The Public Health Costs of Job Loss. Journal
of Health Economics, 28(6):1099 – 1115, 2009. doi: https://doi.org/10.1016/j.jhealeco.2009.09.
004.

Valentino Larcinese, Leonzio Rizzo, and Cecilia Testa. Allocating the U.S. Federal Budget to
the States: The Impact of the President. The Journal of Politics, 68(2):447–456, 2006. doi:
https://doi.org/10.1111/j.1468-2508.2006.00419.x.

Byron Lutz. The Connection Between House Price Appreciation and Property Tax Revenues.

National Tax Journal, 61(3):555–572, 2008. doi: dx.doi.org/10.17310/ntj.2008.3.13.

Byron Lutz, Raven Molloy, and Hui Shan. The Housing Crisis and State and Local Government
Tax Revenue: Five Channels. Regional Science and Urban Economics, 41(4):306 – 319, 2011.
doi: https://doi.org/10.1016/j.regsciurbeco.2011.03.009.

A. Craig MacKinlay. Event Studies in Economics and Finance. Journal of Economic Literature,

35(1):13–39, 1997.

Justin Marlowe. Fiscal Slack and Counter-Cyclical Expenditure Stabilization: A First Look at the
Local Level. Public Budgeting & Finance, 25(3):48–72, 2005. doi: https://doi.org/10.1111/j.
1540-5850.2005.00367.x.

Justin Marlowe. What’s the Point of Rainy Day Funds? Governing the States and Localities, 2013.

103

Therese McGuire, Leslie Papke, and Andrew Reschovsky. Local Funding of Schools: The Property
Tax and Its Alternatives. In Helen Ladd and Margaret Goertz, editors, Handbook of Research in
Education Finance and Policy. Routledge, New York, 2014.

John Mikesell. The Contribution of Local Sales and Income Taxes to Fiscal Autonomy. In Gregory
Ingram and Yu-Hung Hong, editors, Municipal Revenues and Land Policy. Lincoln Institute of
Land Policy, Cambridge, MA, 2010.

Ohio Moraine. Oﬃcial Statement: Various Purpose Infrastructure and Improvement Bonds, Series
2010. https://emma.msrb.org/EP382306-EP300578-EP696385.pdf, 2010. Accessed 7/29/2018.

Richard G Newell and Daniel Raimi. Shale Public Finance: Local Government Revenues and Costs
Associated with Oil and Gas Development. Working Paper 21542, National Bureau of Economic
Research, September 2015.

Ohio Department of Taxation. Tax Data Series.

Ohio Secretary

of State.

2013 Ohio Ballot Questions

and

Issues Handbook.

www.sos.state.oh.us/SOS/Upload/elections/EOresources/general/2013QandIUpdated2014-
07.pdf, 2013.

Guillermo Owen and Bernard Grofman. Optimal Partisan Gerrymandering. Political Geography

Quarterly, 7(1):5 – 22, 1988. doi: https://doi.org/10.1016/0260-9827(88)90032-8.

James M. Poterba. State Responses to Fiscal Crises: The Eﬀects of Budgetary Institutions and
Politics. Journal of Political Economy, 102(4):799–821, 1994. doi: https://doi.org/10.1086/
261955.

Jørn Rattsø and Per Tovmo. Fiscal Discipline and Asymmetric Adjustment of Revenues and
Expenditures: Local Government Responses to Shocks in Denmark. Public Finance Review, 30
(3):208–234, 2002. doi: https://doi.org/10.1177/109114210203000303.

Danielle Sandler and Ryan Sandler. Multiple Event Studies in Public Finance and Labor Economics:
A Simulation Study with Applications. Journal of Economic and Social Measurement, 39(1-2):
31–57, 2013. doi: https://doi.org/10.3233/JEM-140383.

Peter Selb. Supersized Votes: Ballot Length, Uncertainty, and Choice in Direct Legislation Elec-
tions. Public Choice, 135(3):319–336, 2008. doi: https://doi.org/10.1007/s11127-007-9265-7.
Katerina Sherstyuk. How to Gerrymander: A Formal Analysis. Public Choice, 95(1):27–49, 1998.

doi: https://doi.org/10.1023/A:1004986314885.

Mark Skidmore and Eric Scorsone. Causes and Consequences of Fiscal Stress in Michigan Cities.
Regional Science and Urban Economics, 41(4):360 – 371, 2011. doi: https://doi.org/10.1016/j.
regsciurbeco.2011.02.007.

Russell Sobel and Randall Holcombe. Measuring the Growth and Variability of Tax Bases over the

Business Cycle. National Tax Journal, 49(4):535–552, 1996.

104

Albert Solé-Ollé and Pilar Sorribas-Navarro. The Eﬀects of Partisan Alignment on the Allocation of
Intergovernmental Transfers. Diﬀerences-in-Diﬀerences Estimates for Spain. Journal of Public
Economics, 92(12):2302 – 2319, 2008. doi: https://doi.org/10.1016/j.jpubeco.2007.06.014. New
Directions in Fiscal Federalism.

State of Ohio. Ohio Revised Code. LAWriter® Ohio Laws and Rules.

United States Census Bureau.

LEHD Origin and Destination Employment Statistics.

lhttps://lehd.ces.census.gov/data/#lodes.

United States Census Bureau. Annual Surveys of State and Local Government Finances, and Cen-
sus of Governments: Finance, (2014). https://www.census.gov/govs/local/historical_data.html,
2017a.

United States Census Bureau. Annual Surveys of State and Local Government Finances, and Census

of Governments: Finance, (2015). https://www.census.gov/govs/local/, 2017b.

United

States Department

of Labor.

ing Notiﬁcation Act:
https://www.doleta.gov/programs/factsht/warn.htm. Accessed: 7/1/2016.

A Guide

The Worker Adjustment
to Advance Notice of Closings

and Retrain-
and Layoﬀs.

United States Government Accounting Oﬃce. The Worker Adjustment and Retraining Act: Re-
vising the Act and Educational Materials Could Clariy Employer Responsibilities and Employee
Rights. https://www.gao.gov/assets/240/239817.pdf, 2003. Accessed 7/28/2018.

Jeﬀrey Wooldridge. Econometric Analysis of Cross Section and Panel Data. The MIT Press,

Cambridge, Massachusetts, 2 edition, 2010.

Sen-Yuan Wu and Hyman Korman. Socioeconomic Impacts of Disinvestment on Communities in
New York State. The American Journal of Economics and Sociology, 46(3):261–271, 1987. doi:
https://doi.org/10.1111/j.1536-7150.1987.tb01964.x.

105