Fiscal Centralization: Theory and

Evidence from the Great Depression

Daniele Coen-Pirani

Michael Wooley

March 30, 2015

Abstract
The Great Depression produced a profound and lasting influence on the
structure of U.S. government. We study theoretically and empirically
the increased centralization of revenue collection and expenditures in the
hands of the states relative to local governments during the 1930s. In our
political-economy model the income decline of the Depression causes a
rise in property tax delinquency, undermining the main source of revenue
for local governments and leading to increased political support for sales
taxation and centralization by the states. Empirical evidence based on
cross-state variation in the severity of the Depression is consistent with
the models key predictions.
Keywords: Centralization, Fiscal Federalism, Great Depression, Property
Tax, Sales Tax
JEL Classification: H77, H71, N42

We thank Robert Fleck for generously sharing his data with us. Thanks to Dennis Epple,
Marty Gaynor, Bob Miller, Lowell Taylor, Marla Ripoll and seminar participants at the
University of Pittsburgh and Carnegie Mellon for useful comments. The usual disclaimer
applies.

Department of Economics, University of Pittsburgh. E-mail: coen@pitt.edu. Corresponding author.

Department of Economics, Northwestern University.

E-mail:
wooley@u.northwestern.edu.

Introduction

The Great Depression produced a profound and lasting influence on the size
and structure of U.S. government. The size of the government grew as programs
in areas like social insurance were expanded. Its structure also changed, as the
federal government grew in importance relative to state and local governments.
For example, the federal governments share of non-defense expenditures went
from about 27 percent on the eve of the Great Depression to almost 45 percent
on the eve of World War II in 1940. The big losers in this period were local
governments - cities, counties, school districts - whose share of expenditures fell
dramatically from about 54 percent in the late 1920s to 31 percent in 1940.1
While the New Deal and associated policies are usually credited with the
rise of the federal government in this period (Rockoff, 1998), a less well-known
but important development of the 1930s was the rise of state governments
relative to local governments (see Wallis, 1984 for an early paper, discussed
below, on this topic). Specifically, state governments share of combined state
and local expenditures and revenues on average increased by over 20 percentage
points, the largest increase experienced in any decade of the 20th century.
The process of fiscal centralization by state governments is important beyond
its historical interest because the new fiscal arrangements that emerged from
the Great Depression were long-lasting and still exert a powerful influence in
todays world. Two of these long-lasting developments are worth mentioning
at the outset. First, general sales taxes, which now represent the single most
important source of revenues for state governments, were first introduced in the
U.S. by 28 states during the Depression.2 Second, the current involvement of
state governments in elementary and secondary education funding has its origins
in the Great Depression. Prior to the Great Depression, local governments
1

Here and in the rest of the paper we define expenditures to include both direct and indirect
expenditures of a given level of government. Indirect expenditures include intergovernmental
grants to other levels of government. The numbers cited in the text are based on our
computations using data from the Historical Statistics of the U.S. The pre-Depression data
refers to the year 1927.
2
On the importance of sales taxes in present times, see, for example, the summary report
of the 2012 Census of Finance, available at http://www.census.gov/govs/local/.

accounted for over 80 percent of education revenue, with the states providing
the rest. By 1950, the share of local governments had declined to 60 percent;
states provided most of the remaining 40 percent. The further expansion
of states funding share after the mid-1970s is dwarfed by the increase that
occurred during the 1930s.3
In this paper we seek to explain the rise of state governments during the
1930s. In doing so, the paper makes two contributions to the topic of fiscal
centralization. First, it presents a novel theory of centralization, which is
used to interpret the events of the Depression. We develop a positive theory
of centralization. As such, our paper complements the mostly normative
literature on fiscal federalism. The latter discusses the trade-offs between the
costs of centralized public good provision - such as uniformity of provision as
in Oates (1972) or conflict of interest between citizens in different jurisdictions
as in Besley and Coate (2003) - and the externalities across local jurisdictions
associated with decentralized provision. The second contribution of our paper
is to test empirically some of the key implications of this theory using data on
U.S. states.
Our theory of fiscal centralization is based on the observation that local
governments have historically relied on property taxation to fund their expenditures.4 By contrast, at the onset of the Depression the role of property
taxation in states budgets was minimal and had been in decline for decades.5
3
See, for example, National Center for Education Statistics (1993, Figure 12). The
Federal government did not play a significant role in K-12 education until the Elementary
and Secondary Education Act of 1965.
4
The property tax has provided about two-thirds of own revenue for local governments
both before and during the Depression. Local sales and income taxes never accounted for
more than one or two percent of all locally-collected revenues during the 1930s. Wallis (2001)
provides a detailed explanation for local governments reliance on the property tax. Teaford
(2002, 122) remarks that many states barred local governments from collecting all but a
select set of levies.
5
In the first three decades of the 20th century, the states reduced their dependence on
the property tax and adopted a number of new taxes such as excises on gasoline and motor
vehicle registration fees. On the expenditure side, local governments main expenditure
function was K-12 education which accounted for about one-third of their outlays, followed
by highway expenditures, accounting for about one-sixth of total expenditures. States spent
directly and through grants to local governments on highways and on education (including
tertiary education).

Local governments heavy reliance on the property tax made them particularly
sensitive to the sharp and sudden income decline at the onset of the Great
Depression that led to a large increase in property tax delinquency rates. For
example, in 1933 more than a quarter of property taxes levied in cities with
a population greater than 50,000 were delinquent (Bird, 1934 cited in Beito,
1988). The rise in property tax delinquency rates early in the Depression plays
a central role in our theory, as it made it difficult for local governments to fund
ordinary expenses such as education, let alone the rising demand for social
spending brought about by the Depression. This crisis in the ability of local
governments to collect taxes provided the impetus towards states introduction
of new taxes, such as sales taxes. Tax delinquency was much less of a problem
with sales taxes but it was only cost-effective for state or national governments
to collect these levies (Wallis, 2001 and Nechyba, 1997). Though sales taxes
addressed the local revenue problem, they inevitably eroded local fiscal autonomy. The widespread rise in property tax delinquency in the 1930s received
much attention from contemporary economists and policymakers. For example,
Fred Rogers Fairchild, an economist writing in the American Economic Review
in 1934, described the connection between income decline and property tax
delinquency in the context of the Depression:
The property tax, though a direct tax levied upon capital value,
is intended, not to deprive the taxpayer of a part of his capital
funds, but to be paid out of income. But the tax is levied obviously
without reference to current income; though the capital value is
there, present income may be lacking. It may thus occur that a
property tax, legally levied, finds the taxpayer without money to
pay or so short of cash that he feels forced to put other obligations
ahead of taxes. So he postpones payment, and delinquency results,
as does not ordinarily occur in connection with other forms of
taxation (1934, 143).
We embed these elements in a simple model of public good provision to
illustrate our argument and then proceed to test some of its implications using
4

a variety of state-level data. On the model side, we build on Besley and Coate
(2003), Fernandez and Rogerson (2003) and others and consider an economy,
meant to represent a state, divided into many geographically distinct localities
populated by agents with heterogeneous income. We consider two alternative
systems of public good provision. The first is a decentralized system in which
each locality chooses its own level of the public good and pays for it using a
locally administered head tax, our proxy for a property tax. An important
feature of the head tax is that agents might choose not to pay, suffering a utility
cost. The second system is a centralized system in which an economy-wide level
of the public good is financed by means of a regressive sales tax. We show how,
in the benchmark pre-Depression economy, the decentralized system is preferred
by a majority of the population if the sales tax is sufficiently regressive.
After establishing this benchmark, we use the model to illustrate the impact
of the Great Depression on the politico-economic support for centralization.
We model the Depression as an income decline that affects a fraction of agents
in each locality. This shock leads to a rise in head-tax delinquency under
decentralized provision of the public good. In turn, the rise in tax delinquency
increases support for centralized provision financed by sales taxation. The main
intuition for this result is that the sales tax allows the government to expand
its tax base, effectively forcing otherwise (head) tax delinquent individuals to
contribute to the financing of public goods.
This theory is consistent with a number of accounts of these events by
economic historians and economists alike. For example, the historian Jon
Teaford in his 2002 book The Rise of the States ascribes the process of centralization by the states to taxpayer discontent towards the property tax and the
emergency created by the inability of local governments to adequately fund
expenditures during the Depression.6 Among economists, in a related empirical
paper, Rueben (1994) focuses on the determinants of the adoption of state sales
6

Though he recognizes the large role played by the Federal government in providing relief
after 1933, Teaford emphasizes the independent centralizing policies enacted by the states.
For example, at least seven states began providing emergency relief prior to 1933 and almost
every state centralized school finance - an area in which the federal government was inactive
- to some extent during the decade. We return to this latter point in Section 2.

taxes during the 1930s. She finds support for economic stress theories of the
introduction of sales taxes, according to which the economic downturn induced
state governments to seek out new sources of revenue to finance expenditures.
Hartley et al. (1996) study voter support for the Riley-Stewart Amendment,
a major fiscal reform passed by California voters in 1933.7 Its passage led
the state government to adopt sales and income taxes while local property
tax levies were decreased. Based on their empirical investigation, Hartley et
al. (1996, 666) conclude that the passage of Riley-Stewart...emerged as a
response to growing voter discontent over the property tax during the Great
Depression.
While our interpretation is consistent with the studies cited above, other
economists have emphasized other forces at work. Specifically, Wallis (1984)
and Wallis and Oates (1998) argue that a key feature of the New Deal was
its use of state - rather than local - matching grants. Matching grants gave
incentives to state governments to raise new revenues while local governments
were able to dial back their revenues and expenditures. Thus, according to this
view, it was not the Great Depression per se that led to centralization but the
reaction of the federal government to it. It will be useful here and elsewhere to
apply labels to these two channels. We label the Wallis (1984) and Wallis and
Oates (1998) thesis the top-down channel, which emphasizes the idea that the
federal governments matching grant program led to state-level centralization.
The channel that we model focuses instead on local fiscal distress as a cause
of centralization so we label it the bottom-up channel. It should be stressed
early on that these two mechanisms are not necessarily inconsistent with one
another. Quantifying their relative importance is an empirical matter to which
we devote the second part of the paper.
On the empirical side, we test the key prediction of our bottom-up mechanism that, all else equal, the magnitude of the peak-to-trough income decline
experienced by a state during the Depression should be systematically associ7

The amendment required the state to expand its support for public education, authorized
the state to raise new revenues to finance this expenditure, limited state and local expenditure
increases, and returned revenues from public utility property to local governments.

ated with the likelihood that the state subsequently passed a blanket limitation
on property taxes, introduced a sales tax, and increased its share of combined
state and local tax revenues or expenditures.8 Overall, we find strong support
for these key predictions of the model. We account for the top-down channel
of Wallis and Oates by controlling for measures of federal aid received by a
state during the 1930s. In order to address the potential endogeneity of federal
aid, we instrument it using political and land variables that strongly predict
the allocation of New Deal money to the states (Wright, 1974; Wallis, 1984;
and Fleck, 2008). Controlling for federal aid in this fashion has little effect on
the estimated magnitude of our bottom-up channel. In addition, federal aid
does not have an economically or statistically significant impact on any of our
dependent variables.
In the rest of the introduction we briefly place our paper in the context
of the related literature. Among the empirical papers on fiscal policy during
the Great Depression, our work is most strongly related to Wallis (1984) and
Rueben (1994). We discuss in detail the relationship between our results and
theirs in Sections 6.3.3 and 6.3.2 of the paper, respectively. While our results
are consistent with Ruebens, we extend her work in three new directions.
First, we interpret the introduction of sales taxes within the broader context
of discontent with the property tax and changing intergovernmental relations.
Thus, in addition to sales tax adoption we consider the effect of our bottom-up
mechanism on measures of revenue and expenditure centralization and passage
of blanket property tax limitations by the states. Second, in our regressions for
sales tax adoption we control for the effect of federal aid, explicitly accounting
for the top down channel of Wallis and Oates. Last, we use a different
proxy for a states economic distress than Rueben. Instead of employment
decline between 1929 and 1932 we focus on decline in income per capita in
the same period. We find that the latter variable is more strongly associated
with the adoption of sales taxes than the employment-based measure. In
terms of theory, our model is mostly related to the work of Besley and Coate
8

See Section 2 for a definition of blanket tax limitations and a discussion of each of these
variables.

(2003). Differently from them, we emphasize the role played by property tax
delinquency and assume that in the centralized economy public good provision
is financed through a sales tax instead of a uniform head tax. Thus, the shift
from decentralized to centralized provision is accompanied by a shift in tax
instrument, i.e., from property to sales taxation. Nechyba (1997) provides the
basic argument justifying this assumption.9
The rest of the paper is organized as follows. Section 2 discusses the
empirical trends and historical backdrop that are the focus of the paper.
Section 3 introduces the basic model. Sections 4 and 5 evaluate the effect of
the Great Depression on the models equilibrium and the politico-economic
support for a more centralized system of public good provision. The empirical
analysis is presented in Section 6. Section 7 concludes. The appendices contain
the proofs of propositions and details on the data used in the empirical part of
the paper.

Basic Trends and Institutional Setting

In this section we further motivate the paper by providing additional details

on government revenues and expenditures prior to the Great Depression and
the dramatic changes that occurred during the 1930s.

2.1

Before the Great Depression

As the statistics mentioned in the opening paragraph of the paper suggest,

local governments were the most important level of government in terms of
expenditures and revenue prior to the 1930s. Figure 1 shows the evolution
of state governments share of combined state and local own revenue and
9
The paper is also related to the positive literature on the determinants of centralization.
In this literature, Baicker, Clemens, and Singhal (2012), document the rise of the states in the
U.S. since 1952. Strumpf and Oberholzer-Gee (2002) test the hypothesis that heterogeneous
preferences are associated with more decentralized policy-making using data on liquor control.
Matsusaka (1995, 2000) finds that U.S. states with voter initiatives exhibit lower levels of
expenditure centralization over the twentieth century. Panizza (1999) and Arzaghi and
Henderson (2005) study the determinants of fiscal decentralization across countries.

Figure 1: States Share of Total State and Local Revenue and Expenditures

Rev.
.55

.55

Rev.

.6

(b) Excluding Highways

.6

(a) Including Highways

Exp.

1990

1980

1970

1960

1950

1942

1932

1927

1922

year

1913

.15
1902

1990

1980

1970

1960

1950

1942

1932

1927

1922

1913

1902

.15

.2

.2

.25

.25

.3

.3

.35

.35

.4

.4

.45

.45

.5

.5

Exp.

Highway revenues are defined as motor fuel taxes and fees from motor vehicle
and operators licenses. Own revenue refers to total revenue less revenue from
intergovernmental grants.Source: Historical Statistics of the U.S.

expenditure.10 At the turn of the 20th century state governments accounted

for less than 20 percent of aggregate local and state revenue and expenditure.
Figure 1(a) shows that the states share of both variables had been increasing
starting in 1913. However, panel (b) of Figure 1 makes it clear that much of
this early increase is attributable to rising expenditures and revenue collections
for new roads and highways.11

2.2

The Great Depression and the Developments of the

1930s

The Great Depression started in the late summer of 1929 and was characterized
by a very sharp contraction in production and income. From its August 1929
peak to its trough in early 1933, industrial production declined by 47 percent
and real Gross Domestic Product fell by 30 percent (Romer, 2004). In this
10

Own revenue refers to total revenue minus revenue from intergovernmental grants.
Expenditures include both direct expenditures and intergovernmental expenditures.
11
The advent of the automobile led to an expansion of both local roads and roads connecting
different cities and localities within a state. The states played an important role in financing
road construction using gasoline taxes and licenses fees to finance this expansion (Wallis,
2001).

section we document the process of fiscal centralization by the states that

occurred during the 1930s, emphasizing the key developments that led to the
rise in state centralization according to our bottom-up interpretation.
2.2.1

State-Local Centralization

It is evident in both panels of Figure 1 that the greatest increase in centralization

occurred from about 1927 to 1942.12 In those fifteen years alone, the states
share of combined state and local revenues rose by about 23 percentage points.
State-level data show that all state governments centralized revenue and
expenditures during the 1930s (see Table 2 in Section 6.2). The extent of
centralization varied across states. Some of this variation is illustrated in
Figure 2. The figure uses state-level data to show the evolution of the average
state share of state and local expenditures as well as the 25-75 percentile
range.13
2.2.2

Key Elements of the Bottom-Up Interpretation

The bottom-up interpretation of the process of state centralization has, as

its starting point, the effect that the contraction in income at the onset of the
Great Depression had on the ability of local governments to collect revenue
through the property tax. Symptoms of the increased burden of the property
tax were the rise in delinquency and the success of referenda that imposed
strict restrictions on property tax rates in a state. The latter are typically
referred to as blanket property tax limitations. Sales taxes were adopted by
states in order to combat this crisis in local government finances that became
acute around 1932. By that time, tax delinquency, tax limits, and the growing
demand for unemployment relief and social services - which were at that time
a local government responsibility - had pushed many local governments to a
12

We use 1927 because it is the last pre-Depression year for which aggregate data is
available.
13
Figure 2 reports fewer observations than Figure 1 because it draws on data that is
aggregated at the state rather than national level. The state-level data is, in fact, available
for fewer years.

10

1960
1962

1950

1940
1942

1930
1932

1920

1913

1910

1902

.2

.3

.4

.5

.6

.7

Figure 2: State Share of State and Local Expenditures

Light blue area represents 25-75 percentile interval of cross-sectional distribution of

states. For expenditure share definition see Data Appendix. Source: Sylla, Legler,
Wallis (1995).

breaking point.14 In what follows we present and discuss the relevant data on
each of the key points summarized above.
Property Tax Delinquency. When the Depression hit, incomes fell
further and faster than property tax levies. As the quote in the Introduction suggests, many taxpayers simply lacked the funds to pay their tax bills.
Widespread property tax delinquency resulted. According to one study that
compares pre-Depression delinquency rates to those observed during the Great
Depression in cities with a population greater than 50,000, the median tax
delinquency rate rose from 10.1 percent in 1930 to 26.3 percent in 1933 (Bird,
1934, quoted in Beito, 1988).15 Beito (1988, 8) notes that - prior to the De14

While, in principle, local government could have issued debt, new local government
debt issues fell 65.7% from 1929 to 1933. This is partly explained by the fact that many
local governments were already deep in debt in 1929. Moreover, the decline in property tax
revenues undermined lenders expectations that bonds would be paid back. In 1933, 3.7% of
all municipal debt issuers were in default (Fons et al. 2008). Furthermore, different state
laws on borrowing were added or strengthened. For example, Ohio increased the majority
needed to issue new school debt from a simple majority to 55% in 1931 and 65% in 1933
(Tostlebe, 1945, 27).
15
Detailed data on tax delinquency is scattered and incomplete. The first attempts
at systematic collection of data on tax delinquency began in response to Depression-era

11

pression - governments could quickly collect delinquent revenues by selling tax

titles. The tax title market, however, ceased to function during the Depression.
Lent D. Upson, Chief Statistician of the Census Bureaus Division of Real
Estate Taxation observed in 1934 that: Today, however, the professional tax
title buyer has generally retired from the market owing to curtailed bank credit
and deflated realty values (1934, 363).
Blanket Tax Limitations. The decline in income forced taxpayers to
devote a larger share of their income to property taxes. This development
spurred the rise an the anti-property tax movement organized around the goals
of achieving economy in government, widening the tax base, and decreasing
property taxes. The most important consequence of this movement was the
passage of blanket property tax limitation laws (Beito, 1988). Blanket property
tax limits set a cap on the combined millage that all jurisdictions within a state
could levy on a single piece of property.16 Blanket limitations were effective at
curbing property tax collections and, therefore, at decreasing local government
revenues.17 Prior to the Depression, only North Dakota had a law resembling
a blanket limitation. Between 1932 and 1933, however, blanket property tax
limitations were adopted by popular vote in seven states, listed in Table 1.
We therefore use the adoption of blanket limits as an indicator of popular
discontent with the property tax in the empirical analysis of Section 6.
Sales Tax Adoption. Prior to 1932 no state had a general sales tax.
By 1938 the sales tax was in place in 28 states, listed in Table 1, together
with the year of adoption. The importance of the sales tax in state and local
delinquency. The Census Bureaus Division of Real Estate Taxation issued a county-level
tabulation of tax delinquency on the levy of 1932-33. However, this data suffers from issues
of comparability across states (Woodworth,1934, 332). Moreover, there does not appear to
be state-level data on tax delinquency prior to the 1932-33 levy.
16
For example, a 10 mill (1%) blanket property tax limit meant that a property owner
paid no more than 10 mills in total in taxes. This meant that different jurisdictions had
to agree to divide the available millage among themselves. The school district may get 3.5
mills, the city another 3.5, and the county 3.
17
Though almost every state had some sort of law limiting property tax collections, these
laws were widely acknowledged to be ridden with loopholes. The negative effect of blanket
property tax limits on municipal revenues is vividly captured in the essays on tax limits by
municipal administrators and public finance experts collected in Leet and Paige (1934).

12

Table 1: State Blanket Property Tax Limitations and Sales Tax Adoptions
Blanket Property Tax Limit
Michigan
Washington
Indiana
West Virginia
Ohio

Oklahoma
New Mexico

1932

Sales Tax Adoption

Mississippi Pennsylvania*

1933

Arizona
Utah
Oklahoma
New York*

Illinois
Michigan
California
South Dakota

Indiana
North Carolina
Washington

1934

Missouri
Kentucky*

Iowa

West Virginia

1935

Ohio
Wyoming
Maryland*

Colorado
Arkansas
New Mexico

Idaho*
North Dakota

1936

Louisiana

1937

Alabama

Kansas

Blanket Property Tax Limitation: North Dakota instituted a blanket property tax
limitation in 1919. Sales Tax Adoption: denotes states that allowed the sales tax
to expire.
Source: Tax limits: Mott and Suitor (1934). Sales Taxes: Jacoby (1938, Table 14)

13

.05 .1 .15 .2 .25 .3 .35 .4 .45 .5 .55

Figure 3: State Share of State and Local Own Expenditures

Other Taxes

Income Tax

1960

1950

1942

1932

1927

1922

1913

1902

Sales Tax

The upper border of the Other Taxes area (i.e. the orange area) is the same revenue
share series that is displayed in figure 1a. The sales tax share includes both general
sales and excise taxes, which explains why the share is non-zero prior to 1932.
Source: Historical Statistics of the United States.

public finance is illustrated by Figure 3. Sales taxes (including excise taxes

on gasoline) accounted for 5.7 percent of all state and local revenues in 1927
and 18.0 percent in 1942. This growth implies that sales taxes accounted for
approximately 53 percent of the growth in revenue centralization observed from
1927 to 1942. By contrast, the share of combined state and local revenues
collected via state income taxes only grew by 2.1 percentage points from 1927
to 1942, which implies that growth in income tax revenues only accounted for
about 9 percent of revenue centralization.18
2.2.3

Challenges to the Top-Down Mechanism

The empirical analysis of Section 6 attempts to disentangle the effects of the

bottom-up and top-down mechanisms on centralization. Here we discuss
two independent pieces of evidence that suggest that the top-down channel is
unlikely to represent the entire explanation for the evolution of state-local fiscal
18

Sales tax and income taxes play a similar role in our bottom-up channel. The model
of Section 3 allows for both.

14

relations during the 1930s. The first piece of evidence is that several states
enacted centralizing policies prior to 1933, the year in which the New Deal
policies were voted by Congress and started to be implemented. Teaford (2002)
mentions at least seven states whose governments initiated emergency relief
programs in response to the Depression prior to 1933.19 Similarly, the early
adoption of sales taxes by many states (see Table 1) in 1932 or 1933 casts some
doubt on the view that they represented a response to New Deal policies. The
second piece of evidence has to do with elementary and secondary education
financing. In 1927, the states provided on average less than 15 percent of funds
for education. Most of education funding was provided by local governments
(school districts) and financed through property taxation. Figure 4 shows that
the share of funding coming from the states increased dramatically during the
1930s, reaching a peak of over 35 percent in 1942. This trend is important
because elementary and secondary education was never a focus of the New
Deal nor of any other federal program during this period.20 Therefore, the
increased provision of education funds by the states cannot be ascribed to their
reaction to New Deal matching grants in this area.

The Model Economy

We illustrate our bottom-up channel through a simple model of decentralized

and centralized provision of local public goods. We focus on the interaction
between income decline, property tax delinquency, and preferences for centralization through a sales tax. In what follows we first present the model and
then in Sections 4 and 5 use it to analyze the effect of the Depression on the
political-economic support for sales taxation and centralization.
19

New York, New Jersey, Rhode Island, Wisconsin, Pennsylvania, Illinois, and North
Carolina (Teaford, 2002, 123-25). Other states studied centralizing policies early in the
decade but didnt pass them until years later. See, e.g., Lutz (1930) on Georgia, the
[Michigan] State Commission of Inquiry into Taxation (1930), and the Ohio School Survey
Commission (Mort, 1932).
20
On the New Deals neglect of public education, see the discussion in Tyack et al. (1984,
ch. 3) and Teaford (2002, 129-130).

15

1960

1950

1942

1932

1927

1922

1913

1902

.15

.2

.25

.3

.35

.4

Figure 4: State K-12 Education Grants as Share of State and Local K-12
Education Expenditures

Source: Historical Statistics of the U.S.

3.1

The Framework

The model economy, which is meant to represent a U.S. state, is populated by a

measure one of agents. Agents are differentiated by their income level y, which
is distributed in the population according to the density f (y).21 Let y denote
average income in this economy and assume that it lies above median income.
The economy is divided into a continuum of geographically distinct localities or
districts. Each district is assumed to be initially populated by a unit measure
of individuals with the same income level y (see e.g. Fernandez and Rogerson,
2003).22 We sometimes refer to district y as the district inhabited by individuals
with income y. Agents have preferences defined over a private consumption
goods c and a public good g according to the following logarithmic utility
function:
U = ln c + ln g,
(1)
21
It is worth pointing out that allowing for preference, instead of income, heterogeneity, as
in Besley and Coate (2003), would not change the mechanism we seek to emphasize, although
some of the details of the analysis would be different.
22
Similarly to Besley and Coate (2003) and Fernandez and Rogerson (2003), the sorting of
individuals across localities is exogenously given here.

16

where the parameter > 0 indexes the relative weight of the public good.23
Within this basic framework, we distinguish between two distinct approaches
to selecting and financing the provision of the public good g. The first approach
consists of district-level (local) choice of the quantity of the public good, financed
through a head tax. We refer to this arrangement as decentralized provision
and we discuss its properties in Section 3.2. The second approach consists of
state-level choice of the uniform quantity of the public good, financed by a sales
tax on the consumption good. We refer to this arrangement as centralized
provision and we discuss its properties in Section 3.3. Section 3.4 discusses the
politico-economic support for the decentralized system in the context of the
model, prior to the shock of the Great Depression.

3.2

Decentralized Provision

In this section we consider the decentralization case in which each district

selects its own policy, consisting of a head tax t and a level of the public good g.
We interpret the head tax as a property tax in this setting.24 For our purposes
the key aspect of the property tax that is conveniently captured by a head tax
is that when an individuals income falls, its tax bill does not fall automatically,
unless the agent moves to a cheaper house. This feature of property taxation
- emphasized in the quote from the American Economic Review reported in
the Introduction - leads to the emergence of tax delinquency in the Great
Depression scenario considered in Section 4.1.
Given policies (t, g) , an agents private consumption in a district with
income y is:
c = y t.
(2)
Individuals may opt for not paying the head tax, in which case they cannot
23

Notice that, differently from Besley and Coate (2003)s model, there are no spillovers
across districts associated with the consumption of the public good in our model. Spillovers
of this sort would amplify the effect of our bottom-up mechanism, so we do not include
them in our analysis in order to keep the model as simple as possible.
24
Hamilton (1975, 1976) argues that the property tax is equivalent to a head tax if localities
can impose zoning restriction on the size of houses within their boundaries.

17

be excluded from the consumption of the public good, but suffer a utility cost
k. This cost captures the consequences of tax delinquency for an individual,
from social stigma to the threat of losing the title to ones home.25 The key
assumption here is that the cost of tax delinquency is a utility cost that does
not represent revenue for a local government.26 Tax delinquency plays an
important role in the Great Depression scenario. Here we establish a preDepression benchmark in which tax delinquency does not occur in equilibrium.
For future reference, it is convenient to express the utility cost k in equivalent
consumption terms. To do so, define to be the maximum fraction of income
y that an agent is willing to give up to avoid being tax-delinquent:
ln (y y) = ln y k.
Solving this expression for yields:
1 exp(k) < 1.

(3)

In what follows we restrict (and thus indirectly k) to be sufficiently large so

that before the Great Depression agents find it to be optimal to pay the head
tax in equilibrium:
Assumption 1. > / (1 + ) .
The equilibrium level of the head tax in each district is determined by its
representative agents preferences over policies (t, g) , subject to the budget
constraint:
t = g.
(4)
The following proposition summarizes the politico-economic equilibrium.
25

See Luttmer and Singhal (2014) for a discussion of nonpecuniary factors such as social
stigma in tax compliance decisions.
26
Property tax delinquency usually involved local governments selling the title to ones
house after some time. As explained by Beito (1988, 8), the tax title market effectively
ceased to function during the Depression. Hence, an individuals failure to pay the property
tax did result in a loss of revenue for local governments.

18

Proposition 1 (Equilibrium with decentralized provision). Under Assumption

1, the unique politico-economic equilibrium of the economy with decentralized
provision is such that the head tax and public good spending in a district y are
given by:

t = g =
y.
(5)
1+
There is no tax delinquency in equilibrium.
Proof: See Appendix.
The condition that the utility cost of tax delinquency is high enough
(Assumption 1) guarantees that individuals prefer to pay the tax given by (5)
rather than be delinquent. Notice that, in equilibrium, public good provision
and the head tax are increasing in the districts income, with richer localities
taxing and spending more.

3.3

Centralized Provision

As an alternative arrangement, which we label centralization, the public good

is financed at the state level and expenditures are equalized across localities.27
In addition, the state employs a sales tax on consumption, rather than the
head tax, to finance expenditures. The tax-expenditure mix at the state level
is decided by state-wide majority. We assume that individuals cannot avoid
sales taxes so there is no issue of delinquency in this context. This assumption
captures an important asymmetry between property taxes and sales taxes and
deserves some discussion. The nature of this asymmetry reflects two aspects of
these taxes. The first aspect is the magnitude of the payments involved. Yearly
property tax payments by the owner of property to the (local) government were
large relative to the day-to-day transactions (such as purchasing food) that were
subject to sales taxation. Sales tax evasion, if detected, would have entailed a
fine whose magnitude was probably disproportionate relative to the amount of
27

Besley and Coate (2003) argue that centralization does not have to imply uniformity
and highlight, instead, misallocation and uncertainty as other costs associated with it.
For our purposes, either of these two alternative costs of centralization would give rise to
implications similar to those implied by uniform provision.

19

the tax due. Therefore, a cost-benefit analysis would have discouraged sales
tax evasion because of the relatively small size of the transactions involved.
The second important asymmetry between property and sales taxes in terms
of opportunity for delinquency is the fact that, while property tax payments
consisted of direct transfers by the owner of property to the (local) government,
the sales tax was, instead, collected by a retailer who then transferred it to the
(state) government. The retailer acted as a tax collector on behalf of the state
government and might have refused a sale if a customer failed or refused to
pay the sales tax. More generally, tax delinquency would in this case require
the willing participation of buyers and sellers, instead of the unilateral decision
of the property owner.
We model the sales tax as a regressive tax on consumption.28 The regressive
nature of the sales tax plays an important role in determining the political
support for its adoption and, more generally, for centralization. Formally, we
capture this essential feature of sales taxation while keeping the model analytically tractable by adopting a formulation originally introduced by Benabou
(2002) and recently employed by Heathcote et al. (2014). According to this
formulation, the consumption of an agent with income y is specified as:
c = y 1+ ,

(6)

so that the tax paid by the agent is equal to:

T (y) = y y 1+ .

(7)

The parameters > 0 and 0 in equation (6) determine the overall level
of the tax and its degree of regressivity, respectively.29 This tax scheme is
28

The regressive nature of sales taxes was clearly emphasized by economists and commentators in the debates that occurred in the early 1930s. For example, Jacoby (1938, Table 31)
lists food, water, gas, electricity as some of the items within the scope of retail sales taxes,
and education tuition, domestic services, association and club dues as some items not within
the direct scope of such taxes.
29
We implicitly assume that the income distribution and tax parameters are such that
agents with higher income pay a higher sales tax, or T 0 (y) > 0 for all y in the support of
f (y). We do not make these conditions explicit in order to avoid burdening the notation and

20

regressive - in the sense that the marginal tax rate is smaller than the average
tax rate - as long as > 0. When = 0 the tax scheme is equivalent to a
linear income tax.30 The degree of regressivity is taken as a given feature of
sales taxation and is not chosen by agents in the politico-economic equilibrium.
Agents, instead, are assumed to choose the parameter , which determines the
overall level of taxation, jointly with the public good level g, subject to the
states budget constraint:

g=

T (y)f (y) dy.

(8)

We are agnostic about the exact mechanism to aggregate individual preferences over the policy (, g) because the logarithmic structure of utility
implies that voters are unanimous. Thus, we only require that a policy that is
unanimously preferred is adopted. The following proposition summarizes the
politico-economic equilibrium in this case.
Proposition 2 (Equilibrium with centralized provision). In an equilibrium
with centralized provision, the quantity of the public good is given by:
g =

y.
1+

(9)

The associated tax parameter is reported in the proof of the proposition.

Proof: See Appendix.
Comparing equations (5) and (9), notice that the quantity of the public good
under centralized provision is equal to the average (across localities) quantity
under decentralized provision. Thus, decentralized provision offers a smaller
(larger) consumption of the public good to agents with income below (above)
the mean relative to centralized provision. An equally important consideration,
in order to determine the politico-economic support for decentralization, is
how the tax burden is shared among agents with different incomes. The next
exposition of the model any further. None of our results depends on these conditions.
30
As discussed in Section 2.2.2, income taxes played a more limited role than sales taxes
in generating revenue for the states during the 1930s.

21

section summarizes our findings on this question.

3.4

Comparing the Two Systems

In this section we determine how the politico-economic support for decentralized

provision depends on the models parameters. As anticipated in the last
section, the relative appeal of decentralization for an agent depends on how her
consumption of the public good and her tax burden vary across the two systems.
The trade-off involved is as follows. A decentralized system allows high-income
agents to benefit from a higher consumption of the public good relative to
a centralized one. This gain is larger the larger is the preference parameter
in the utility function (1). However, the regressive nature of the sales tax
implies that high-income agents might benefit from centralized financing of
the public good, since their share of aggregate sales taxes would be smaller
than their share of aggregate income. The gap between these shares increases
with the degree of regressivity of the sales tax, indexed by the parameter
in equation (7). It follows that political support for decentralization depends
on the relative magnitude of the parameters and , as summarized by the
following proposition.
Proposition 3 (Comparing the decentralized and centralized systems). The
relative preference for decentralization depends on the models parameters and
on agents income in the following way:
1. If = , then the decentralized system is preferred by all agents.
2. If > , then all agents whose income is below a cut-off y d (defined in
the proof of the proposition) prefer the decentralized system. This is a
majority of the population.
3. If < , then all agents whose income is above the cut-off y d prefer the
decentralized system. This group is a majority of the population if is
close to and becomes a minority as converges to zero.
Proof: See Appendix.
22

Notice that the case = represents the situation in which, from a pure
redistributional perspective, the two systems are equivalent. The tax price of
the public good is independent of income as the regressivity of the sales tax
system fully offsets the potential from redistribution due to uniform provision.
Given this, all agents prefer a decentralized system because, in this case, they
retain control over the quantity of the public good they consume. In other
words, local provision is preferred because the quantity of the public good
selected locally exactly matches the preferences of each agent - a pure Tiebout
type of argument.
When the sales tax is sufficiently regressive relative to the taste for the
public good ( > ), it is low income agents who prefer decentralization, while
the opposite is true when the sales tax is not very regressive ( < ). In the
former case it is possible to conclude unambiguously that decentralization is
preferred by a majority because the agent with average income pays a larger
tax under the regressive sales tax system than under the decentralized one,
while consuming the same quantity of the public good. However, when <
decentralization is preferred by a majority of high-income agents only if is
close to .31 As 0, the sales tax becomes a linear income tax, and the
comparison between the two systems is the same as in Fernandez and Rogerson
(2003). In this case the cut-off y d y, and a majority of agents with income
below the mean prefer centralization because their public good consumption is
larger and their tax payments are lower than under decentralized provision.
Prior to the Great Depression the decentralized system was the most
common form of provision of non-defense public goods in the U.S. We take this
situation to be our benchmark. Proposition 3 states that our model is able to
rationalize the popularity of decentralized provision among (at least) a majority
of the population as long as the sales tax regressivity parameter is not too
close to zero. The next section shows how the events of the Great Depression
- specifically the sharp income decline and the related rise in property tax
delinquency - created a situation in which the politico-economic support for
31

Since when = , decentralization is unanimousy preferred, this is also the case when
is close to by a continuity argument.

23

the decentralized system declined.

Great Depression Experiment

We formalize the Great Depression as a negative income shock that affects a

fraction < 1/2 of the population in each locality. Specifically, an agent who
suffers a negative income shock experiences an income decline from y to y,
for some < 1. Notice that, after the Great Depression shock, each districts
population is divided into two groups based on income. For a fraction 1 of
the districts population income remains unchanged at y, while for a fraction
it declines to y. This ex-post heterogeneity in income is a necessary ingredient
to generate some tax delinquency in the decentralized equilibrium (see Section
4.1).32 As a matter of notation, in what follows we keep referring to district
y to indicate a district whose pre-Depression income was y, and where now a
fraction of its inhabitants has experienced a drop in income to a new level y.

4.1

Decentralized Equilibrium and Delinquency

The option to not pay the head tax becomes more appealing for an agent as
her income declines because of the higher marginal utility of consumption. The
equilibrium nature of the model is such that while individual tax delinquency
decisions depend on the magnitude of the head tax, the latter is endogenous
and therefore depends on aggregate delinquency. In what follows we analyze
each of these two channels in order to characterize the new politico-economic
equilibrium.
Consider first the decision to be delinquent for an agent with income y,
given a tax t.33 The tax delinquency decision is based on the comparison
between the utility associated with being delinquent while consuming the entire
32

Notice that we are keeping the distribution of population across localities constant after
the shock occurs.
33
The decision of an agent with income y is, of course, analogous. However, agents whose
income does not decline do not find it optimal to be delinquent in equilibrium by virtue of
Assumption 1.

24

income y:
ln (y) k,
and the utility associated with paying the head tax:
ln (y t) .
It follows that an agent chooses to be delinquent if the tax is sufficiently large:
t > y.

(10)

While delinquency depends on the tax t, the converse is also true, through
local voting over taxes and public good provision. The assumption that < 1/2
guarantees that agents whose income does not decline are decisive in the choice
of (t, g). Consider the policy choice problem of these agents. They choose (t, g)
in order to maximize their utility subject to the following government budget
constraint:

(1 ) t if t > y,
g=
(11)
t
if t y.
This constraint reflects the fact that if the head tax is too large agents hit by
the income shock will choose to be delinquent. In what follows we impose a
second restriction on the parameter so that head tax delinquency emerges in
equilibrium.
Assumption 2. < / (1 + ) .
The following proposition summarizes the politico-economic equilibrium.
Proposition 4 (Decentralized provision in the Great Depression). Define the
following cut-off for the parameter :
1
1

b 1 (1 ) (1 + ) +1 .

b under Assumption 2, the unique equilibrium tax and public good

1. If < ,

25

provision in a district y are given by:

t () =

y, g () =
y (1 ) .
1+
1+

(12)

In such an equilibrium, agents whose income declines to y choose to be

delinquent on the head tax.
b the unique equilibrium is such that:
2. If ,
t () = g () = y.
In such equilibrium, no agent is delinquent.
Proof: See Appendix.
In case 1 of this proposition, agents who are not affected by the recession are
willing to keep taxing themselves in the same amount as before the Depression
if the fraction of the population that is delinquent is not too large. The
fact that the tax itself does not depend on is a feature of logarithmic utility.
Assumption 2 guarantees that the equilibrium tax in equation (12) satisfies the
inequality in equation (10), inducing tax delinquency of agents with income
y. Alternatively, if is relatively large (case 2), the politically decisive agents
prefer to reduce the level of taxation to the point where all agents are willing
b because tax
to pay the head tax. In what follows we focus on case 1 ( < ),
delinquency was, indeed, a prominent feature of the Great Depression.34
b
Assumption 3. < .
In the next section we consider the centralized equilibrium in the Great
Depression scenario.

4.2

Centralized Equilibrium

In the centralized equilibrium with a regressive sales tax, the consumption of

agents whose income y does not decline is still given by equation (6), while
34

Notice that this assumption is consistent with our previous one that < 1/2.

26

consumption of the agents hit by the negative income shock and living in the
same district is:
c = (y)1+ .
(13)
Per capita tax collection in district y is now:




T (y) = (1 ) y y 1+ + y (y)1+ ,

(14)

reflecting the ex-post heterogeneity in income within the district. The budget
constraint for the state as a whole takes the same form as in equation (8), with
T (y) given by equation (14).
The preferred quantity of the public good for an agent with given income is
obtained by maximizing her utility function (1) with respect to (, g), subject
to the relevant budget constraint (6) or (13).35 The logarithmic specification of
preferences implies that all agents in the economy, irrespective of their income
level, prefer the same public good provision. This result is summarized in the
following proposition.
Proposition 5 (Centralized provision in the Great Depression). In the Great
Depression version of the economy, all agents agree on the following level of
public good provision:
g () =

(1 + ) y.
1+

(15)

The corresponding tax parameter () is reported in the proof of this proposition.

Proof: See Appendix.
Comparing equations (9) and (15), we conclude that the quantity of the
public good under centralization declines proportionately to average income
relative to the pre-Depression scenario. However, comparing Proposition 4
(case 1) with Proposition 1 shows that under decentralized provision the
economy-wide provision of the public good declines proportionately more than
35

Recall that the degree of regressivity of the sales tax is taken as given here.

27

average income.36 This observation provides the key insight about the increased
support for centralization in our models Great Depression scenario relative
to the benchmark economy. A more detailed comparison is undertaken in the
next section.

Rising Support for Centralization during the

Great Depression

In this section we ask two questions related to our model and relevant for
the empirical analysis that follows. First, does the fall in income associated
with the Depression increase the political support for centralization relative
to the pre-Depression scenario? Second, does support for centralization in the
Depression scenario increase more the larger the income drop? The following
proposition summarizes our answer to the first question, while Proposition 7
provides the answer to the second.
Proposition 6 (Increased support for centralization in the Great Depression).
1. A larger fraction of agents whose income does not decline supports centralized provision of the public good in the Great Depression scenario than
in the benchmark economy.
2. A larger fraction of agents whose income declines supports centralized
provision of the public good in the Great Depression scenario than in the
benchmark economy if and only if the degree of regressivity of the sales
tax is not too high:
ln (1 + ) (1 )
<
.
(16)
ln
Proof: See Appendix.
36

Specifically, while average income declines to a fraction (1 + ) of its pre-Depression

value, the average quantity of the public good under decentralized provision declines to a
fraction (1 ) of its pre-Depression value.

28

Points 1 and 2 of Proposition 6 compare the politico-economic support

for centralization in the pre-Depression and the Depression scenarios. Point 1
deals with agents whose income remains constant and who therefore are not
delinquent in the Great Depression scenario. Those agents, who retain the
political control over the head tax in each locality, choose to keep the head
tax constant (Proposition 4, part 1). However, they are worse-off because the
head tax delinquency by agents whose income falls leads to a reduction in local
revenue and expenditures. By contrast, the sales tax cannot be avoided and
it allows the (state) government to tax individuals who would not otherwise
contribute to pay for the public good due to their delinquent status. In other
words, the sales tax forces delinquent agents to pay for their share of the
public good. This feature of the sales tax induces a greater portion of nondelinquent agents to support centralization of revenue collection through the
sales tax in the Depression scenario.37
Point 2 of Proposition 6 considers the preferences over centralization of
agents whose income declines due to the Depression and who become tax
delinquent. Under condition (16), in all localities in which non-delinquent
voters support centralization, tax delinquent voters also support it. In addition,
there are some localities in which non-delinquent voters oppose centralization,
while delinquent ones support it. The reason why tax delinquent agents are
more strongly in favor of centralization than non-delinquent ones is that, while
the former do not contribute to public good provision in the decentralized
equilibrium of the Great Depression scenario, they suffer the utility costs k
associated with being delinquent. If this utility cost is large relative to the
degree of regressivity of the sales tax - that is if equation (16) is satisfied then tax delinquent agents are more strongly in favor of centralization than
non-delinquent agents with the same pre-Depression income.38
37

More specifically, if > ( < ), statement 2 (3) of Proposition 3 applies with a cut-off
level of income y d () strictly smaller (greater) than y d , the cut-off in Proposition 3. If = ,
in the Great Depression scenario, agents unanimously support either decentralization (as in
Proposition 3) or centralization, depending on parameters.
38
The degree of regressivity of the sales tax cannot be too large because in that case
agents whose income declines would see their marginal tax rate increase by a large amount
under centralization.

29

Our empirical analysis builds on the results of Proposition 6. However,

instead of relying on a simple pre-post Depression comparison, we focus on
a comparison across states between the magnitude of the income decline
experienced by a state and the likelihood of it passing a number of centralizing
reforms. We view the model economy described above as representing a
U.S. state. The magnitude of the income decline in a state is captured by the
parameter , with higher values of leading to larger declines in average income.
Proposition 7 then summarizes the prediction of our theory for the cross-state
pattern of correlation between income decline and support for centralization.
Proposition 7 (Monotonic support for centralization). Voters support for
centralization increases monotonically with the extent of the decline in average
income, indexed by .
Proof: see Appendix.
Intuitively, as the magnitude of the income decline increases, more agents
become delinquent, exacerbating the problems with the head tax system
and increasing political support for centralization. The next section tests
this implication of the model using state-level data and various measures of
centralization.

Empirical Analysis

Proposition 7 provides the foundation of our empirical analysis. Since we do

not directly observe voters support for centralization, we proceed by relating
the extent of the decline in per-capita income experienced by a state at the
onset of the Great Depression with the subsequent adoption of the sales tax
and with the increase in indicators of fiscal centralization. In addition, we
provide some more direct evidence for our mechanism and the role it ascribes
to property taxation by showing that states that experienced a more severe
income decline were also more likely to adopt blanket property tax limitations,
a strong indicator of popular discontent with the property tax.

30

6.1

Data and Descriptive Statistics

We now provide a brief description of the data used.39 Throughout our analysis,
the basic unit of observation is a U.S. state. The data can be conveniently
organized into four categories: outcome variables, regressors of interest, other
covariates, and instrumental variables. Summary statistics of the data, organized in this fashion, are presented in Table 2. In this section we discuss the
first two categories of data and postpone the discussion of additional controls
and instruments to Subsections 6.3.1 and 6.3.2.
Outcome variables. We consider three outcome variables. The first is
the adoption of blanket tax limit during the 1930s. This takes the form of
a zero-one dummy variable, with one denoting the adoption of a blanket tax
limitation by a state during the 1930s. As shown in Table 2, one in six states
adopted a blanket property tax limitation during the Depression.40 The second
policy variable - a dummy as well - is the adoption of a general sales tax
during the 1930s. Almost 60 percent of the states adopted a sales tax in this
period. Third, we seek to explain the increase in revenue and expenditure
centralization - measured by a states share of combined state and local revenues
and expenditures - between 1932 and 1942.41 The choice of these two years
reflects constraints on data availability. The state-level fiscal data, provided by
Legler, Sylla, and Wallis (1995) and based on the Census of Governments, is
only available for these two years. The earliest year of data available before
1932 is 1913. The statistics on these indicators of fiscal centralization in Table 2
reiterate the message of Figure 2: the states share of revenue and expenditures
increased by about 27 percentage points from 1932 to 1942.42
39

The data is described in full in Appendix B.

The list of states that adopted blanket tax limitations or sales taxes is contained in
Table 1.
41
Our definitions of revenue and expenditures are consistent with our previous analysis
in Section 2 and Wallis and Oates (1988). Specifically, revenue refers to own revenues, i.e.
revenues raised directly by the government and not derived from intergovernmental grants.
Expenditures include grants made by the government to another level of government but
subtracts out the total grants received from other levels of government.
42
Notice that this increase is somewhat larger than the corresponding increase (of about
23 percentage points) between 1927 and 1942 based on national (as opposed to state) level
data documented in Section 2.2.1. Notice also that in the interest of space, in Table 2 we
40

31

Regressors of interest. There are two regressors of interest. The main

driving force of our model and bottom-up mechanism is the income decline
experienced by a state. We measure the latter as the percent decrease in a
states income per capita from 1929 to 1932. We compute the latter as the
difference in the logarithm of income per capita between 1932 and 1929 using
the Bureau of Economic Analysis Regional Data. The income variable is
converted in real terms by dividing it by the national Consumer Price Index.43
We consider the decline from 1929 to 1932 because the National Bureau of
Economic Research dates the peak of the business cycle in the third quarter
of 1929 and the trough in the first quarter of 1933. Since we only have yearly
income data at the state-level, we consider 1929-32 to be the contraction phase
of the cycle. This is consistent with, for example, Garrett and Wheelock
(2006). Moreover, the policies we focus on - tax limitations and sales tax
adoptions - were all enacted during or after 1932. Similarly, the revenue and
expenditures data we use to measure centralization refer to the 1932-1942
period. To economize on language we will oftentimes just call this variable the
state income shock or income decline.
In addition to the decline in per capita income, we also consider an alternative driving force of centralization, the New Deal grants received by a state
(Wallis, 1984). We measure federal aid as the (logarithm of) the total real
aid per capita received by a state from the federal government in the period
19331939. To construct this meaure of federal aid we use data from the Report
of the Secretary of the Treasury.44
only report the change in the measures of centralization between 1932 and 1942. Although
not reported in Table 2, our three outcome variables tend to be positively associated in
the cross-section of states. For example, all seven states that adopted a blanket tax limit
subsequently adopted a sales tax. In addition, the average increase in the states share of
combined state and local expenditures from 1932 to 1942 was 11.7 (7.8) percentage points
greater in states that adopted a tax limit (sales tax).
43
Notice that since we are using the log difference in income, any variation in the price
level common to all states between 1929 and 1932 is absorbed by a regression constant
or time effect. Thus, dividing the income data by the national CPI has no effect on the
regression results. The inflation adjusted data of Table 2 are, however, more informative
because they provide a sense of the magnitude of the income decline. As far as we know,
consumer price indices for individual states are not available during this period.
44
We have also employed a second measure of federal aid reported by Reading (1973), who

32

Table 2: Summary Statistics.

Mean

SD

Min

Max

Outcome Variables
Tax Limit Dummy
Sales Tax Dummy
Difference State Revenue Share, 32-42
Difference State Expenditure Share, 32-42

0.167
0.583
0.274
0.272

0.377
0.498
0.072
0.082

0.000
0.000
0.131
0.112

1.000
1.000
0.432
0.495

Regressors of Interest
Log Difference in Per Capita Income, 29-32
Log Federal Aid Per Capita, 33-39

-0.364
4.974

0.101
0.428

-0.595
3.995

-0.166
6.175

Covariates
% Renters, 1930
% School 27-28
% Urban 30
Log Income Per Capita 29
% Dem. 28
Non-White Pop. 30
Initiative State Dummy
Debt Restriction Dummy
Same Party Control Dummy
Republican Control Dummy

0.494
0.193
0.460
6.347
0.434
0.106
0.438
0.833
0.729
0.125

0.085
0.063
0.199
0.377
0.125
0.132
0.501
0.377
0.449
0.334

0.361
0.102
0.166
5.583
0.271
0.002
0.000
0.000
0.000
0.000

0.679
0.335
0.924
7.048
0.914
0.503
1.000
1.000
1.000
1.000

Instrumental Variables
Federal Land per capita
Non-Federal Land per capita
Electoral Votes per capita
SD Dem. Vote, 96-32

0.039
0.043
0.006
10.175

0.148
0.056
0.004
4.326

0.000
0.002
0.004
2.500

0.998
0.246
0.033
18.100

33

Table 3: Average Outcome Measures by Percent Growth in Per Capita Income.

Change in States Share
Tax Limit

Sales Tax

Revenue

Expenditure

Highest 3rd (26%)

31

26

25

Medium 3rd (36)

13

56

28

28

Lowest 3rd (48%)

38

88

29

29

Growth in per capita income

Statistics in columns 1 and 2 are percentages while columns 3 and 4 are percentage
point differences.

6.2

Preliminary Analysis

Before moving to a formal regression analysis, we briefly illustrate the role of

income decline in fostering centralization through simple conditional means.
Specifically, we first partition the 48 states in three groups of 16 states each,
based on their 1929-32 per capita income growth.45 We then summarize the
mean of each outcome variable for each group of states. The results are reported
in Table 3.
Notice that states differ greatly in terms of the decline in income they
experienced. The average real income decline in the hardest-hit states was
48 percent over the period 1929-1932 (see the bottom row of Table 3). By
contrast, it amounted to 26 percent in the states least-affected by the Great
Depression (top row of Table 3). The magnitude of the income decline bears
a systematic association with our outcome variables. For example, none of
the least-affected states passed a blanket property tax limitation, against 38
percent of the most-affected states. Similarly, states with the highest decline
in income were more likely to adopt a sales tax and centralize revenues and
expenditures than states with medium and lowest income declines. In the next
section we analyze these patterns in a regression framework, which allows us
drew on a different government document, with nearly identical results. The correlation
between the two variables is 0.93. A more detailed comparison of our aid variable and
Readings one is provided in appendix B.2.2.
45
Recall that Alaska and Hawaii became states only in 1959 so data on these two states
became available only at that time.

34

to control for other economic forces.

6.3

Regression Results

In this section we present a number of regressions to illustrate and quantify

the magnitude of the bottom-up mechanism relative to the top-down one
and other economic forces in accounting for variation in our three outcome
variables. For clarity of exposition we consider each of these three in separate
subsections.
6.3.1

Blanket Property Tax Limitations

Property tax limitations were almost always adopted or strengthened through

voter initiatives. A blanket property tax limitation, then, may be interpreted as
a strong and direct indicator of voter discontent with the property tax. In this
section we ask whether states with larger decline in per capita income between
1929 and 1932 were more likely to pass a blanket property tax limitation.
Specifically, we run state-level probit regressions for the probability of passing
tax limitations during the 1930s on measured income decline from 1929 to 1932
and a number of control variables. We exclude from the latter the federal aid
variable both because the top-down mechanism does not naturally account
for voters discontent with the property tax and because blanket tax limitations
were passed relatively early in the Depression, in 1932 and 1933, so it is highly
unlikely that they are explained by New Deal policies enacted in 1933 and
implemented over a number of subsequent years.46
Table 4 reports marginal effects of these regressions. Column 1 of Table
4 suggests that the percent growth in per capita income from 1929 to 1932
is indeed negatively correlated with the probability of adopting a sales tax.
According to this estimate, a one (cross-state) standard deviation decrease in
1929 to 1932 income growth - corresponding to about 10 percentage points increases the probability that a blanket tax limitation will be adopted by 14.7
46

Including federal aid in the regressions does not affect our results and the aid variables
are not significant.

35

Table 4: Marginal Effects for Property Tax Limitation Adoption (Probit).

% Growth in Per Capita
Income, 29-32
% Renters 30

(1)

(2)

-1.471***
(0.511)

-1.356***
(0.433)
-0.0859
(0.992)
-1.350**
(0.642)
1.040**
(0.478)
-0.530*
(0.299)
-1.506**
(0.628)
-0.850
(1.008)
0.0985
(0.0891)

48

48

% School 27-28
% Urban 30
Log Income 29
% Dem. 28
Non-White Pop. 30
Initiative State

Observations

Robust standard errors reported in parentheses. Significance: *** 99% level, ** 95%
level, * 90% level. Variable definitions: see text and data appendix.

percentage points.
The effect of income growth on the adoption of property tax limitations is
robust to including a number of control variables emphasized in the literature on
property tax limitations (e.g. Vigdor, 2001) that account for voters preferences
for public good spending. They include the share of households in the state
that were renters in 1930 (% Renters 30), the percent of voters that voted for
the Democratic nominee in 1928 (% Dem. 28), the share of the population ages
10-19 enrolled in grades 7-12 in 1927-28 (% School 27-28), the share of the
population living in an urban area (% Urban 30), log real 1929 state income
per capita (Income 29), and a dummy for whether the state permitted voter
initiatives.47 Results for this specification are reported in column 2 of Table 4.
47

In many states, blanket tax limitations seem to have been a populist or grass roots
movement (Leet and Paige 1934). If voter initiatives were permitted in a state it may have
been more likely that a property tax limitation vote would occur and - perhaps - be passed.

36

Adding these controls does not affect significantly the estimated marginal effect
of income growth on the adoption of tax limits. Thus, the results of Table 4
are consistent with our hypothesis that the income decline of the Depression
led to a rising discontent with the property tax. The next section shows that
the income decline induced governments to look for new sources of revenue in
sales taxes.
6.3.2

Sales Tax Adoption

Our model of sales tax adoption predicts that states experiencing larger income
shocks are more likely to adopt a sales tax. Table 5 displays marginal effects
from probit regressions in which the dependent variable is sales tax adoption.
In all regression specifications the income growth variable displays a statistically
and economically significant association with the probability of adopting a sales
tax. States that experienced larger declines in per capita income were more
likely to adopt sales taxes. In column 1 of Table 4, the only explanatory variable
is the percent growth in income. The marginal effect of income growth from
1929 to 1932 can be interpreted by considering the effect on sales tax adoption
of reducing a states income growth by one standard deviation, or about 10
percentage points (Table 2). This additional income decline is associated with
an increase in the probability of adopting a sales tax by about 26 percentage
points. In column 2 of Table 5 we add the federal aid variable to the regression.
The latter is neither statistically nor economically significant. Moreover, the
marginal effect of income growth is unaffected by the inclusion of federal aid.
Last, column 3 adds the variables included in Rueben (1994) in order
to facilitate a comparison with her results.48 The estimated effects of these
control variables on sales tax adoption are consistent with Ruebens. Examining
The effect of voter initiatives during the first half of the twentieth century is the subject of
Matsusaka (2000), which is also the source of data on voter initiatives.
48
These variables include a dummy (Debt Restriction) for whether the state government
was legally prohibited from raising debt; a second dummy (Same Party Control) indicates
states where a majority of both houses of the state legislature and the governor were in
the same political party; a dummy (Republican Control) set equal to one if Republicans
controlled both houses of the legislature and the governors office; a dummy variable for
southern states (Southern State); the logarithm of 1929 state income per capita.

37

Table 5: Marginal Effects for Sales Tax Adoption (Probit).

% Growth in Per Capita

Income, 29-32

(1)

(2)

(3)

-2.599***
(0.423)

-2.606***
(0.413)
-0.045
(0.041)

-1.799***
(0.497)
-0.003
(0.042)
0.212*
(0.112)
0.461***
(0.130)
-0.115
(0.144)
-0.044
(0.159)
0.417*
(0.218)

48

48

48

Log Federal Aid

Debt Restriction
Same Party Control
Republican Control
Southern State
Log Income per capita,
29
Observations

Robust standard errors reported in parentheses. Significance: *** 99% level, ** 95%
level, * 90% level. Variable definitions: see text and data appendix.

38

column 3 of Table 5 we notice that, while the marginal effect of the state income
shock has decreased somewhat relative to column 2, it is still statistically and
economically significant. According to the specification in column 3, a one
standard deviation decrease in income from 1929 to 1932 is associated with
an increase in the probability of adopting a sales tax by about 18 percentage
points. The marginal effects of both federal aid variables remain economically
small and insignificant.
There are two main differences between our specification in column 3 of
Table 5 and Rueben (1994)s, who first studied empirically the question of sales
tax adoption by the states. First, we use income growth between 1929 and
1932, rather than employment growth for this same period as our economic
shock variable. We prefer to use income growth because it corresponds directly
to the shock considered in our model. Moreover, the income variable includes
- but is not limited to - labor income and thus is likely to provide a more
accurate measure of the resources available to households in a given state than
employment growth. We have also run the regression for sales tax adoption
using employment growth (measured by the employment index in Wallis, 1989)
instead of income growth. In the regression specification in column 3, we find
that a one standard deviation (10 percentage points) decline in employment
growth increases the probability of adopting a sales tax by 12 percentage
points.49 By contrast, a one standard deviation decline in income growth
leads to an increased probability of passing a sales tax by about 18 percentage
points, according to column 3 of Table 5. Therefore, the results associated with
the income growth measure are quantitatively larger than those obtained by
Rueben (1994). The second difference with respect to Ruebens specification is
that we include a federal aid variable in our regressions in order to account for
the top-down channel discussed in the introduction. This variable turns out
not to matter for sales tax adoption and does not affect the importance of the
income shock or employment index variables.
49

This number is consistent with Ruebens (1994, 12) finding that a decline in employment
growth by 5 percentage points increases the probability of passing the sales tax by 5 percentage
points.

39

The last issue we address in this section is the potential endogeneity of

federal aid in the regressions of Table 5, which would lead to inconsistent
estimates.50 The ideal instrument exploits variation in federal aid received
by a state that is uncorrelated with the unobserved determinants of sales tax
adoption. Following the literature, we consider two instruments. The first
is based on the work of Wright (1974), who argued that federal grants were
allocated across states so as to increase Franklin Delano Roosevelts probability
of reelection in 1936. Wright (1974) showed that the number of electoral
votes per capita and the within-state (time-series) standard deviation of the
Democratic vote share in presidential elections from 1896 to 1932 were powerful
predictors of New Deal spending in a state.51 The second instrument is based
on the land area of a state. Reading (1973) hypothesizes that more grants
flowed to states with large amounts of federal land because relief projects on
federal land both improved the land and minimized the amount of bureaucratic
machinery necessary to get relief projects underway. More recently, Fleck (2008)
points out that the amount of non-federal land in a state was an important
determinant of the amount of federal highway grants received by a state. We
therefore consider both federal and non-federal land per capita as instruments
for federal aid.
Table 6 presents instrumental variable probit regressions using these instruments. Both political and land-based instruments affect the allocation of federal
aid across states in the predicted way in first-stage regressions. The p-value
50

While such endogeneity might occur for various reasons, an obvious one is reverse
causation from sales tax adoption to federal aid.
51
The premise of Wrights (1974) argument is that the Roosevelt administration sought
to distribute discretionary aid dollars among states so as to maximize electoral votes while
minimizing costs. Assuming that a given amount of federal aid could buy a vote and that
the distribution of this cost is equal across states, states with more electoral votes per capita
could be bought relatively more cheaply since each voter has more influence over which
candidate the states electoral votes goes. States in which the standard deviation of the
Democratic vote share was low were either solidly Democratic or Republican; an aid dollar
spent there would do little to change the outcome of the election. High standard deviation
states, however, could potentially be swung into the Democratic camp with more aid dollars.
Notice that we would obtain similar results using Wrights political productivity index
(which is itself based on the electoral votes measure) as an instrument, rather than the
electoral votes variable.

40

Table 6: Marginal Effects for Sales Tax Adoption (IV Probit).

% Growth in Per Capita Income,
29-32
Log Federal Aid

(1)

(2)

(3)

(4)

-2.605***
(0.412)
-0.041
(0.046)

-2.568***
(0.422)
-0.012
(0.049)

-1.777***
(0.473)
-0.021
(0.061)
0.200*
(0.120)
0.436***
(0.128)
-0.110
(0.140)
-0.079
(0.166)
0.367*
(0.209)

-1.797***
(0.547)
0.031
(0.042)
0.227*
(0.120)
0.513***
(0.126)
-0.111
(0.145)
0.032
(0.152)
0.503**
(0.220)

Debt Restriction
Same Party Control
Republican Control
Southern State
Log Real Income per capita, 29

First-stage Coefficients
Electoral Votes per capita
SD Dem. Vote, 96-32

209.983***
(57.585)
0.062**
(0.027)

Federal Land per capita

1.454**
(0.543)
16.454***
(2.503)

Non-Federal Land per capita

Observations
p-value

197.214***
(50.166)
0.061*
(0.033)

48
0.000

48
0.000

1.995***
(0.573)
14.860***
(2.497)
48
0.000

48
0.000

First stage regressions also include % growth in per capita income,29-32 in columns
1-2 and % growth in per capita income, 29-32, Debt Restriction, Same Party
Control, Republican Control, Southern State, and Log Real Income 1929 in columns
3-4. Robust standard errors reported in parentheses. Significance: *** 99% level, **
95% level, * 90% level. Variable definitions: see text and data appendix.

41

on the significance of the instruments rules out weak instruments problems.

Examining the second stage regressions, we find that the estimated marginal
effects of both the federal aid variable and the income growth one on sales
tax adoption are very similar to the non-IV probit estimates of Table 5. We
therefore confirm that the possible endogeneity of the federal aid variable is
unlikely to affect our main results.
6.3.3

Fiscal Centralization

Our previous results concern the effect of income decline on discontent with
property taxation and the adoption of the sales tax. In this section we consider
two direct indicators of state fiscal centralization - the state governments
share of combined state and local revenue and expenditure - and assess their
association with the decline in state per capita income and New Deal aid. The
state-level fiscal data are available in 1932 and 1942, allowing us to estimate a
difference-in-difference regression of the following form:
zst = s + d1942 (0 + 1 ys + 2 aids ) + st ,

(17)

where zst is the share of revenue or expenditures in state s at time t = 1932, 1942;
s is a state fixed effect; d1942 is a dummy for t = 1942; and 0 is a year effect.
The coefficients of interest are 1 and 2 , which correspond to the effect of
1929-32 income growth (denoted by ys ) and federal aid (denoted by aids ) on
the change in the dependent variable between 1932 and 1942.
OLS estimates of equation 17 are presented in column 2 (revenue share) and
4 (expenditure share) of Table 7. Columns 1 and 3 show analogous results for
the case in which we omit the federal aid variable from these regressions. The
effect of the state income shock is negative in all four columns. The estimated
effects are statistically significant at least at the 10 percent level. The point
estimates in columns 2 and 4 suggest that a one standard deviation decrease
in income per capita was associated with about 1.72 and 2.05 percentage
point increase in the growth of the states share of revenues and expenditures
between 1932 and 1942, respectively. This corresponds to about a quarter of

42

Table 7: Fiscal Centralization (OLS).

Revenue Share
(1)
(2)
% Growth in Per Capita Income,
29-32 1942
Log Federal Aid 1942

Observations
R2
State Fixed Effects
Year Fixed Effects

-0.160*
(0.093)

96
0.954
Yes
Yes

-0.172**
(0.082)
-0.017**
(0.008)
96
0.959
Yes
Yes

Expenditure Share
(3)
(4)
-0.195*
(0.109)

-0.205*
(0.102)
-0.014
(0.009)

96
0.945
Yes
Yes

96
0.948
Yes
Yes

Robust standard errors reported in parentheses. Significance: *** 99% level, ** 95%
level, * 90% level. Variable definitions: see text and data appendix.

the cross-sectional standard deviation in these variables (see Table 2).

Federal aid enters with the wrong (from the perspective of the top-down
view) sign in all regressions and the estimated effects are small. As for sales
tax adoption, one may be concerned that the results in Table 7 might fail to
account for all factors that influence both centralization and federal aid. We
therefore instrument federal aid using the political and state land variables
discussed in the previous section.
The results of the instrumental variables regressions are presented in Table
8. The point estimates on 1929-32 income growth are essentially the same as
in Table 7 but are estimated much more precisely. Now, all the coefficients
on income growth are significant at the one percent level. The estimated
magnitude of 2 remains quite small and with the wrong sign. For example, a
one standard deviation increase in federal aid is associated with a decrease in
a states revenue share of about half of a percentage point.
The small estimated effect of federal aid and its incorrect sign in regressions
for states expenditure and revenue shares raises the question of why our
conclusions are different from those reached by Wallis (1984) and Wallis and
Oates (1998). The latter paper refers to the former, so we limit our comments
to the first paper here. In the interest of space we summarize our main
argument here while relegating a more detailed discussion to Appendix C.
43

Table 8: Fiscal Centralization (IV).

Revenue Share
(1)
(2)
% Growth in Per Capita
Income, 29-32 1942
Log Federal Aid 1942

-0.170***
(0.057)
-0.014**
(0.006)

-0.168***
(0.058)
-0.011**
(0.006)

Expenditure Share
(3)
(4)
-0.199***
(0.072)
-0.006
(0.006)

-0.200***
(0.072)
-0.007
(0.006)

First-stage Coefficients
Electoral Votes per capita
1942
SD Dem. Vote, 96-32
1942
Federal Land per capita
1942
Non-Federal Land per capita
1942
Observations
p-value
State Fixed Effects
Year Fixed Effects

209.983***
(57.585)
0.062**
(0.027)

209.983***
(57.585)
0.062**
(0.027)
1.454**
(0.543)
16.454***
(2.503)

96
0.000
Yes
Yes

96
0.000
Yes
Yes

1.454**
(0.543)
16.454***
(2.503)
96
0.000
Yes
Yes

96
0.000
Yes
Yes

First stage regressions also include % Growth in Per Capita Income, 29-32. Coefficients from IV regressions reported. Robust standard errors reported in parentheses.
Significance: *** 99% level, ** 95% level, * 90% level. Variable definitions: see text
and data appendix.

44

The regression in Wallis (1984) that is closest to our equation (17) is his
regression for local expenditures (page 157). The latter is also a difference-indifference specification for the change in local expenditures between 1932 and
1942 across states. Instead of using federal aid as independent variable as we
do, Wallis main regressor of interest is the intergovernmental aid received by
local governments in a state.52 The latter is instrumented using Wrights (1974)
political variables, the same variables we use to instrument federal aid. He
finds that states in which local governments received more intergovernmental
aid experienced a relative reduction in local governments own expenditures. In
Table C1 of Appendix C we show that we can replicate this result adopting a
simplified version of Wallis regression equation. That is, intergovernmental aid
to local governments has a negative and statistically significant effect on the
share of expenditures accounted for by local governments.53 Two issues, however,
make us think that this analog of Wallis result should be interpreted with
caution. First, when we introduce our income decline variable in this regression,
the intergovernmental aid variable loses its statistical significance at the 5
percent level, while the income decline variable is significant with the predicted
sign. Second, and more important, we find that in the first-stage regression of
intergovernmental aid on Wright (1974)s political variables, states with more
electoral votes per capita experienced a smaller increase in intergovernmental
aid received by local governments. This result questions the direction of the
connection between federal aid to the states and intergovernmental aid received
by local governments postulated by the top-down view. Overall, our results
suggest that the evidence in favor of the top-down mechanism for state
centralization is not robust.
52

A second difference between Wallis (1984)s regression specification and ours is that he
considers as dependent variable the level of local expenditures from own revenue, rather than
the share of expenditures accounted for by local governments. Notice also that our results are
not comparable with Wallis regression for state (rather than local) expenditures (equation 1
in Wallis, 1984, page 150, with results in his Table 3) because the latter does not include
either state or year dummies and therefore is not a difference-in-difference specification.
53
We use the share rather than the level of local expenditures to be consistent with the
rest of our analysis.

45

Conclusion

In this paper we have analyzed one of the most striking changes in intergovernmental fiscal relationships in U.S. history. During the Great Depression the
states share of combined state and local revenue and expenditures increased
greatly. Our main hypothesis was that this shift was brought about by the
different mix of taxes available to state and local governments and by how
these tax instruments interacted with the sharp income decline of the early
part of the 1930s. We introduced a model of fiscal centralization to illustrate
the mechanism through which income decline leads to increased support for
sales taxation and fiscal centralization. Finally, we demonstrated empirically
that states that experienced larger declines in income in 1929-32 were more
likely to to pass blanket limitations on property taxation, introduce a general
retail sales tax, and, more generally, to centralize revenue and expenditures in
the hands of state governments during the 1930s.
The lesson of the Great Depression is still relevant to the current world for
at least two reasons. First, many of the fiscal policies that emerged from it such as sales taxation and state centralization - have persisted for many decades
after the end of the Depression. While a complete account of this hysteresis
is outside the scope of the paper, one hypothesis is that the severity of the
Depression enabled the type of institutional reform that requires a very broad
political consensus. After the economic crisis passed, support for centralization
might have declined, but the politico-economic coalition favoring a return to
the pre-Depression status-quo was not sufficiently large.54
A second reason for why the lesson of the Great Depression is still relevant is
that politico-economic mechanisms reminiscent of those of the 1930s continue to
operate decades later. For example, during the 1970s, California experienced a
sharp rise in housing values that outpaced many households income growth. As
a consequence, property tax bills increased as a fraction of income, forcing many
households to sell their homes and move elsewhere. According to OSullivan,
54

In Section D of the Appendix we offer a formalization of this argument based on the

model of Sections 35.

46

Sexton and Sheffrin (1995) these events were a powerful force behind the
passage of Proposition 13, which marked the beginning of modern property
tax revolts in the U.S.55 Since the property tax is the main tax controlled by
local governments, the passage of Proposition 13 and equivalent measures in
other states was instrumental in further increasing fiscal centralization by state
governments at the expense of local ones. Thus, the mechanisms we focus
on in this paper, linking income decline and property tax discontent to swift
politico-economic change, are still relevant to understand more recent fiscal
developments.

References

Arzaghi, Mohammad, and J. Vernon Henderson. Why Countries Are Fiscally

Decentralizing." Journal of Public Economics 89, no. 7 (2005): 1157-1189.
Baicker, Katherine, Jeffrey Clemens, and Monica Singhal. The Rise of the States:
US Fiscal Decentralization in the Postwar Period." Journal of Public Economics
96, no. 11 (2012): 1079-1091.
Beito, David T. 1989. Taxpayers in Revolt. Chapel Hill, NC:University of North
Carolina Press.
Bnabou, Roland. Tax and Education Policy in a Heterogeneous-Agent Economy:
What Levels of Redistribution Maximize Growth and Efficiency?" Econometrica
70, no. 2 (2002): 481-517.
Besley, Timothy, and Stephen Coate. Centralized Versus Decentralized Provision of
Local Public Goods: A Political Economy Approach." Journal of Public Economics
87, no. 12 (2003): 2611-2637.
Bird, Frederick L. Four Year Trend in Tax Delinquency." National Municipal Review
23, no. 2 (1934): 110-116.
Fairchild, Fred Rogers. The Problem of Tax Delinquency." The American Economic
Review (1934): 140-150.
Fernandez, Raquel, and Richard Rogerson. Equity and Resources: An Analysis of
Education Finance Systems." Journal of Political Economy 111, no. 4 (2003):
858-897.
55

Proposition 13 limited property tax rates at one percent of assessed values and stated
that the latter could not grow at more than two percent per year if the property was not
sold.

47

Fleck, Robert K. Voter Influence and Big Policy Change: The Positive Political
Economy of the New Deal." Journal of Political Economy 116, no. 1 (2008): 1-37.
Fons, Jerome S., Thomas Randazzo, and Marc D. Joffe. 2011. An Analysis of
Historical Municipal Bond Defaults." Kroll BondRatings.
Ford, Robert S. 1902-1970. Realty Tax Delinquency In Michigan. Ann Arbor: The
Horace H. Rackham School of Graduate Studies, University of Michigan, 1937.
Garrett, Thomas A., and David C. Wheelock. Why Did Income Growth Vary Across
States During the Great Depression?" The Journal of Economic History 66, no.
02 (2006): 456-466.
Haines, Michael R., and Inter-university Consortium for Political and Social Research.
Historical, Demographic, Economic, and Social Data: The United States, 17902002. ICPSR02896-v3. Ann Arbor, MI: Inter-university Consortium for Political
and Social Research [distributor], 2010-05-21.
Hamilton, Bruce W. Zoning and Property Taxation in a System of Local Governments." Urban Studies 12, no. 2 (1975): 205-211.
. Capitalization of Intrajurisdictional Differences in Local Tax Prices." The American Economic Review (1976): 743-753.
Hartley, James E., Steven M. Sheffrin, and J. David Vasche. Reform During Crisis:
The Transformation of Californias Fiscal System During the Great Depression."
The Journal of Economic History 56, no. 03 (1996): 657-678.
Heathcote, Jonathan, Kjetil Storesletten, and Giovanni L. Violante. Consumption and
Labor Supply with Partial Insurance: An Analytical Framework." The American
Economic Review 104, no. 7 (2014): 2075-2126.
Jacoby, Neil H. Retail Sales Taxation. Chicago: Commerce Clearing House (1938).
Leet, Glen, and Robert Myron Paige. Property Tax Limitation Laws: The Evidence
and the Arguments for and Against Them by Twenty-four Authorities. Public
Administration Service, 1934.
Luttmer, Erzo FP, and Monica Singhal. Tax Morale." Journal of Economic Perspectives 28, no. 4 (2014): 149-68.
Lutz, Harley L. The Georgia System of Revenue. Atlanta: Foote and Davies Company
(1930).
Matsusaka, John G. Fiscal Effects of the Voter Initiative: Evidence from the Last 30
years." Journal of Political Economy (1995): 587-623.
. Fiscal Effects of the Voter Initiative in the First Half of the Twentieth Century."
The Journal of Law and Economics 43, no. 2 (2000): 619-650.
Michigan. State Commission of Inquiry into Taxation, 1929. Report of the State
Commission of Inquiry Into Taxation. 1930.: Authorized by Act 122 of the Public
Acts of 1929. Lansing: Franklin DeKleine company, printers and binders (1931).
Mott, Rodney L. and W.O. Suiter. The Types and Extent of Existing Tax Limitations." in Property Tax Limitation Laws: the Evidence For and Against Them
48

by Twenty-Four Authorities, Edited by Glen Leet and Robert M. Paige, 41-46.

Public Administration Service, 1934.
Nechyba, Thomas J. Existence of Equilibrium and Stratification in Local and
Hierarchical Tiebout Economies with Property Taxes and Voting." Economic
Theory 10, no. 2 (1997): 277-304.
Oates, Wallace E. Fiscal Federalism. New York: Harcourt Brace Jovanovich (1972).
Mort, Paul R. Equalizing Educational Opportunity In Ohio: Report of a Survey of
State And Local Support of Public Schools. Columbus, OH: Ohio School Survey
Commission, 1932.
OSullivan, Arthur, Terri A. Sexton, and Steven M. Sheffrin. Property Taxes and Tax
Revolts: The Legacy of Proposition 13. New York: Cambridge University Press,
1995.
Panizza, Ugo. On the Determinants of Fiscal Centralization: Theory and Evidence."
Journal of Public Economics 74, no. 1 (1999): 97-139.
Reading, Don C. New Deal Activity and the States, 1933 to 1939." The Journal of
Economic History 33, no. 04 (1973): 792-810.
Rockoff, Hugh. By Way of Analogy: The Expansion of the Federal Government in
the 1930s." In The Defining Moment: The Great Depression and the American
Economy in the Twentieth Century, pp. 125-154. University of Chicago Press,
1998.
Romer, Cristina D. Great depression." Encyclopedia Britannica (2004).
Rueben, Kim S. The Political Economy of State Sales Tax Adoptions: 1930-1938."
Paper presented at All-UC Conference on Fiscal Crises in Historical Perspective,
M.I.T. 1994.
Snyder, Thomas D. ed. 120 Years of Education: A Statistical Portrait. Washington,
DC: NCES, 1993.
Strumpf, Koleman S., and Felix Oberholzer-Gee. Endogenous Policy Decentralization:
Testing the Central Tenet of Economic Federalism."Journal of Political Economy
110, no. 1 (2002): 1-36.
Sylla, Richard E., John B. Legler, and John Wallis. Sources and Uses of Funds In
State and Local Governments, 1790-1915: [United States]. 1991. Ann Arbor, MI:
Inter-university Consortium for Political and Social Research [distributor], 1993.
http://doi.org/10.3886/ICPSR09728.v1
Teaford, Jon C. The Rise of the States: Evolution of American State Government.
Baltimore, MD: JHU Press, 2002.
Tostlebe, Alvin Samuel. Local Government Debt in Rural Counties of Ohio." Bureau
of Agricultural Economics, U.S. Department of Agriculture, Washington D.C.
(1945).
Tyack, David B. Public Schools in Hard Times: The Great Depression and Recent
Years. Cambridge, MA: Harvard University Press, 1984.
49

U.S. Treasury Department. Annual Report of the Secretary of the Treasury on the
State of the Finances." Yearly: 1932-1940. Washington DC: GPO.
Upson, Lent D. Tax Delinquency: Administration and Legislation." Paper presented
at the Twenty-Seventh Annual Conference on Taxation Under the Auspices of
the National Tax Association, Boston, MA, October 1-5, 1934.
U.S. Advisory Commission Intergovernmental Relations. Significant Features of Fiscal
Federalism, 1993.
Vigdor, Jacob L. Other Peoples Taxes: Nonresident Voters and Statewide Limitation
of Local Government." Journal of Law and Economics 47, no. 2 (2004): 453-476.
Wallis, John J., and Wallace Oates. The Impact of the New Deal on American
Federalism." In The Defining Moment: The Great Depression and the American
Economy in the Twentieth Century, pp. 155-180. University of Chicago Press,
1998.
Wallis, John Joseph. The Birth of the Old Federalism: Financing the New Deal,
1932 1940." The Journal of Economic History 44, no. 01 (1984): 139-159.
. Employment in the Great Depression: New Data and Hypotheses." Explorations
in Economic History 26, no. 1 (1989): 45-72.
. The Political Economy of New Deal Spending Revisited, Again: With and
Without Nevada." Explorations in Economic History 35, no. 2 (1998): 140-170.
. A History of the Property Tax in America." Property Taxation and Municipal
Government Finance: Essays in Honor of C. Lowell Harris, ed. Wallace E. Oates,
Cambridge: Lincoln Institute of Land Policy (2001).
. Federal, State, and Local Government Finances Census of Governments. Table
Ea1-583 in Historical Statistics of the United States, Earliest Times to the Present:
Millennial Edition, edited by Susan B. Carter, Scott Sigmund Gartner, Michael
R. Haines, Alan L. Olmstead, Richard Sutch, and Gavin Wright. New York:
Cambridge University Press, 2006.
Woodworth, Leo Day. Collection of Property Taxes With Special Reference to Real
Estate." Paper presented at the Twenty-Seventh Annual Conference on Taxation
Under the Auspices of the National Tax Association, Boston, MA, October 1-5,
1934.
Wright, Gavin. The Political Economy of New Deal Spending: An Econometric
Analysis." The Review of Economics and Statistics (1974): 30-38.

50

(Appendices not meant for publication.)

Appendix: Proofs of Propositions 16

A.1

Proposition 1

Assuming no agent is delinquent, the governments budget constraint is

g = t.
Using this relationship, the optimal policy choice solves the problem:
max {ln (y g) + ln g}
g

yielding the solution in equation (5). Given this solution, an agent chooses not
to be delinquent if
ln (y g ) ln y k
which holds under Assumption 1.

A.2

Proposition 2

The preferred policy bundle (, g) by an agent with income y solves the problem:


max ln y 1+ + ln g
,g

s.t.
g = y e
y,
where the term ye is defined as follows:

ye

y 1+ f (y) dy.

51

The first-order condition is

e
y
1
=

y e
y
yielding as a solution = y/ [(1 + ) ye] and g given by equation (9).

A.3

Proposition 3

To compare the two systems, compute the indirect utility associated with each.
The decentralized one yields:
V dec = ln

y
y
+ ln
1+
1+

while the centralized one:

V cen = ln

y
+ (1 + ) ln y + ln
y.
(1 + ) ye
1+

Comparing these two, we obtain that V dec > V cen if and only if:
ye

y <

1+

(18)

If = , this inequality is always satisfied because, by Jensens inequality:

ye > y 1+ .
Thus, in this case there is unanimous support for decentralization.
If > , the inequality (18) becomes
y < yd
where the term

ye

ye

y 1+

1+

1


(19)

>1

because of Jensens inequality since > 0. It follows that the cut-off in equation
52

(19) is larger than y. Thus, given that median income is less than mean income, a
majority of the population with relatively low income supports the decentralized
system.
Finally, if < , the inequality (18) becomes:
y > yd
where


ye

1


< 1.

y 1+
Thus, if

y d < median income,

(20)

a majority of the population with relatively high income supports decentralization, while if the inequality in equation (20) does not hold, supporters of
decentralization are a minority of the population with relatively high income.
As , the term:
1


ye
y 1+
tends to zero and (20) is always verified and a majority of the population favors
decentralization. As 0, it is straightforward to verify that y d y. Since
median income is always assumed to be below mean income, the inequality in
equation (20) is never verified as 0, and a majority supports centralization.

A.4

Proposition 4

We assumed that < 1/2, which implies that the decisive voter at the district
level will have income y. We derive the optimal (t , g ) pairs for this decisive
voter with and without a positive delinquency rate and then derive conditions
under which it is optimal to choose a policy (t , g ) that induces delinquency
in equilibrium.
In an equilibrium in which all agents whose income declines are delinquent

53

on their head tax, the decisive voter achieves the following utility:
V1 max
ln(y t) + ln[(1 )t].
t
Note that this problem is identical to that faced by the voter in Proposition
1 with the exception of an added constant. It can be shown that the solution
takes the same form as in Proposition 1:
t () =

y.
1+

Under Assumption 2, t () > y which leads to tax delinquency by agents

whose income is y. Thus, t () is the optimal tax rate in the equilibrium with
positive delinquency. The expression for g () is then obtained from the local
governments budget:

g () =
y(1 ).
1+
Consider now an equilibrium with no delinquency. The decisive voter solves:
V0 max ln(y t) + ln t.
t[0,y]

The objective function is monotonically increasing for all t in the constraint

set. Therefore, the optimal tax in this equilibrium is t () = y and public
good provision is just g () = y.
The decisive voter chooses the equilibrium that gives her the highest utility
level. Thus, the equilibrium with tax delinquency is chosen if and only if
V1 > V0 . It is straightforward to verify that this condition is equivalent to the
following constraint on :
1

< 1 [(1 + )(1 )]

(1 + ) .

Alternatively, if the equilibrium without delinquency is selected.

54

A.5

Proposition 5

Under centralized provision, the public good consumed by each agent is given
by the following state governments budget constraint:


g() = ((1 ) + ) y (1 ) + 1+ y.
An agent with income y thus solves


max ln y 1+ + ln g()

to determine her preferred level of (, g) . Solving the first-order condition for

yields
1 +
y
(21)
() =
1+
1 +
(1 + )
y
and replacing () into g() yields:
g () =

(1 + )
y.
1+

Notice that these quantities are independent of income. An agent with income
y solves:


max ln (y)1+ + ln g().

This objective function is the same as that of an agent with income y,

up to an additive constant. Thus, also this type of agent prefers the policy
(g (), ()).

A.6

Proposition 6

We provide a separate proof for each of the two statements. It is convenient to

define the following terms:
1 + 1+
,
1 +
1
2
.
1 +
1

55

Note that since < 1 and > 0 the following are true:
Fact 1. 1 , 2 < 1
Fact 2. 1 / < 0, 2 / < 0.
Consider now each claim of the proposition.
A.6.1

Support for centralization increases among agents whose income does not decline

The indirect utility functions of agents whose income does not decline under
decentralization and centralization are:

Vydec = ln(y y
) + ln((1 )y
),
1+



1 +
1
y

cen
1+
Vy = ln
y
(1 + )
y .
+ ln
1 (1 + )
y
1+
Combining the log terms and taking a positive monotonic transformation
yields equivalent indirect utility forms:
Vydec =

(1 + )1+

(1 ) y 1+

1 (1 + ) y1+ 1+
cen

y
Vy =
1 (1 + )1+
y
There are two cases of interest.
Case > Centralization is preferred to decentralization if and only if
Vycen > Vydec . That is, if
y

1+

(1 + )1+ y1+ 1+
(1 ) <
y
1 + 1+ y
1



1
y

y > 1 2
y
y1+

1
y > y d () 1 2 y d ,

56

(22)

where y d is defined in equation (19) in Proposition 3. Notice that Fact 1

1
and the condition > imply that the term 1 2 < 1. It follows that
y d () < y d , so support for centralization increases in the Great Depression
economy.
Case > Using the same setup as in the previous case, it is straightforward to derive an analogous condition:
y < y d () = 1 2

1

yd.

1
The discussion from the previous subcase implies that 1 2 > 1. Thus,
in this subcase the range of income such that agents support centralization
in the Great Depression economy is wider than in the benchmark economy.
Support for centralization is therefore higher in the Great Depression economy.
Case = . Centralized provision is preferred to decentralized provision
< Vycen ) if:


1
y
< 1.
(23)
1
1+
1 + y

(Vydec

We know that

1

1
1 +


<1

and that y > y1+ , so condition (23) might or might not be satisfied depending
on parameters. That is, it is not always the case that decentralization is always
preferred as was the case in the pre-Depression economy (case = 0, in which
case the inequality (23) is always satisfied).
A.6.2

If the sales tax is not too regressive, then support for centralization is higher among agents whose income declines in
the Great Depression

If there is positive delinquency then it must be the case that t () > y.

The indirect utility functions of an agent whose income declines to y under

57

decentralized and centralized public good provision are:

) k,
Vydec = ln(y) + ln((1 )y
1
+





1 +
y

1+
cen
Vy = ln
(y)
+ ln
(1 + )
y .
1 + 1+ (1 + )
y
1+
By combining the log terms and using a positive monotonic transformation
we get equivalent formulations:
Vydec


=

(1 )
1+

ek y 1+ ,

1 [(1 + ) ] y1+
(y)1+ .
Vycen =
1 (1 + )1+
y
Case > A delinquent voter supports centralization if Vydec < Vycen ,
that is, if:
 1

(1 + ) ek d
y>
y () .
(24)

Notice that (24) is (22) times a constant. Equation (16) in the main text
of the paper implies that (1 + ) ek / < 1. Thus, the income cut-off for
delinquent agents to support centralization is lower than for non-delinquent
agents. The claim in the proposition therefore holds.
Case >

The analogous condition for support of centralization is:


y<

(1 + ) ek

1


y d () ,

where the term that multiplies y d () on the right-hand side of this inequality is
now larger than one because > . Thus, as in part A.6.1 the range of incomes
such that delinquent agents support centralization in the Great Depression
economy is wider than the range in the benchmark economy. In other words,
support for centralization is strictly larger in the Great Depression economy.

58

Case =

Centralization is preferred by delinquent agents when:


1

1
1 +

y (1 + ) ek
< 1.
y1+

This condition is implied by the analogous condition for non-delinquent agents

(equation 23) since (1 + ) ek / < 1 by equation (16).

A.7

Proposition 7

If =
6 , we need to show that the income cut-off y d () defined in equation
(22) is monotonically decreasing in when > and monotonically increasing
in when < . This is straightforward to verify because Fact 2 guarantees
that 1 2 is decreasing in . When = , the propositions statement requires
the term


1
1
1 +
to be declining in . This is the case because 1 declines in (Fact 1).

Data Appendix and Definitions

We discuss the data and variable definitions in terms of: outcome variables
(B.1); regressors of interest (B.2); control variables (B.3); instruments for
federal aid (B.4). The sources of all variables are summarized in Table B3 of
B.5.

B.1

Outcome Variables

B.1.1

State Budget Shares

State revenue and expenditure shares were both derived from the 1932 and
1942 Census of Governments, which were collected by Sylla, Legler, and Wallis
(1995). State revenue shares in our analysis are defined ownrevenue shares,
i.e. they only include revenue raised by the government directly and do not

59

include intergovernmental grants. The data that we used, instead, includes

intergovernmental grants in the total revenue figure. Thus, some adjustments
were necessary. Suppose RL and RS represent total revenues for local and
state governments, respectively. Denote intergovernmental grants to local
governments (by state governments) and to state governments (by the federal
government) by IL and IS . Then, the states share of combined state and local
own revenues, RSO , is given by
RSO

RS IS
.
(RS IS ) + (RL IL )

In constructing the state expenditure shares we attribute intergovernmental

grants to the granting government. The data that we used counts grants
to other governments as a part of total expenditures but we still need to
subtract out expenditures paid for by grants from other levels of government to
avoid double counting. Let EL and ES denote expenditures by state and local
governments, respectively, and use the same notation for intergovernmental
grants as above. Then the state expenditure share, ESO is given by
ESO

ES IS
.
(ES IS ) + (EL IL )

The 1932 Census of Government reports expenditure and revenue figures

for states, counties, cities and towns, school localities, other civil divisions, and
townships. We sum expenditures, revenues, and grants across counties, cities
and towns, school districts, other civil divisions, and townships to generate
EL , RL , and IL , respectively. State expenditures, revenues, and grants only
include the state government figures. The ISO codes used by Sylla, Legler, and
Wallis (1995) for each of these variables for 1932 is given in appendix Table B1.
The 1942 Census of Government only reports total local government and
state government figures. The total expenditure figure includes Provision for
Debt Repayments, an item not included in other years. Sylla, Legler, and
Wallis (1995) advise that this item be removed from expenditure totals for
1942 in order to make them comparable across years. We follow this advice in

60

Table B1: ISO Codes for Computing Variables From Legler, Sylla, Wallis
(1995)
Variable
E
R
I

Year
(
1932
1942
(
1932
1942
(
1932
1942

ISO Code
0003
0003
0001
0001 4100
2300
2350 + 2361

For state governments ISO = 2361 (Aid From State Government Only) is always
equal to zero in 1942.

computing expenditures for 1942. The 1942 Census also provided more detail
on intergovernmental grants received. In particular, it includes total grants and
grants from state governments. This is reported in Sylla, Legler, and Wallis
(1995) as aid From Federal Government (ISO = 2350) and Aid From State
Government Only (ISO = 2361). The data providers assume that the states
only received aid from the federal government, which - in their words - is an
inaccurate assumption, but not too far wrong (Sylla, Legler, and Wallis, 1995).
For our purposes this does not present a problem since aid From Federal
Government is the same as total aid for states and it is total aid that we are
after. These details are documented in appendix Table B1.
B.1.2

Sales Taxes

There are two ways to count sales tax adoptions. The first only counts those
states that instituted a permanent sales tax. There were twenty-three such
states. This is the criteria used by the Advisory Commission on Intergovernmental Relations (1993). The second approach - used by Jacoby (1938) and
followed by Rueben (1994) - counts both permanent and temporary sales tax
adoptions as sales tax adoptions. When this criteria is used twenty-eight states
are counted as sales tax states. We list all states who adopted a sales tax at
any time from 1932-1940 in Table 2 of the main text. States that only had
61

temporary sales taxes are marked with an asterisk. Our results hold whether
we use twenty-three or twenty-eight states.
B.1.3

Tax Limitations

Tax limitations were adopted en masse during the early years of the Depression.
They are detailed in tables 1, 2 and 3 of Mott and Suiter (1934). Though this
document was compiled in the midst of the Depression, the New York Times
(1939) confirms that the states mentioned in the Mott and Suiter article were
the only states to pass tax limitations during the Depression.

B.2
B.2.1

Regressors of Interest
Economic Indicators: State-Level Income and Employment

From 1929 onward, personal income data are available on an annual basis from
the Bureau of Economic Analysis. We use this data to generate our percent
decline in personal income variable for 1929-1932. Wallis (1989) presents indices
of employment, manufacturing, and non-manufacturing for 1929-1940. All
series for all states have values of 100 in 1929.
B.2.2

Federal Aid to States

Our variable for federal aid to states is taken from the Report of the Secretary
of the Treasury for the years 1933 to 1939.56 For all years and for each state
we collect the grand total of all federal aid to the state. We then correct the
figure for inflation (using 1932 as a base year). We sum the annual aid figures
over all years. Finally, to put the variable in per capita terms, we divide total
aid by the total population of the state over the same period.
The decision to count all federal aid rather than only those that were
explicitly for New Deal programs was prompted by this observation from
Reading (1973, 793):
56

The exact name and table number differs by year. We provide the table numbers here:
1933: 49, 1934: 47, 1935: 48, 1936: 52, 1937: 54, 1938: 59, 1939: 61.

62

Some New Deal Programs were established to meet a specific emergency; others were directed toward aiding depressed areas for the
duration of the depression; still others were permanent and lasting.
Some programs were the creation of New Deal planners; others
were holdovers from the previous administrations. The Roosevelt
Administration viewed many well-established programs (the Bureau of Public Roads, the Veterans Administration, the Bureau of
Reclamation) as vehicles for and methods of increasing employment.
Reading followed this reasoning in constructing his own set of federal aid
figures from a report by the Statistical Section of the Office of Government
Reports entitled Federal Loans and Expenditures (1940). Subsequent work on
New Deal Aid has used Readings data, generally dividing the total aid figure
by each states 1930 population to get a per capita figure. This work includes
Wright (1974), Wallis (1984, 1987, 1998), Anderson and Tollison (1991), and
Fleck (2001, 2008). Wallis (1998) provides an insightful and entertaining
overview of the history of this data.
Our decision to use a slightly different aid variable was prompted by a few
concerns. First, by collecting yearly data we are able to account for the wide
swings in prices that occurred during the Depression. The Reading data is in
nominal terms. Yearly data also allows us to correct for population growth.
Finally, we were unable to locate the original document used by Reading (only
aggregate figures are reported in the 1973 paper). Thus, a key advantage of
the Treasury data is that we can see the breakdown of federal aid by program.
Wallis (1984) used the same Treasury data used here to supplement Readings
data.
It should be noted that our federal aid data is not merely a transformed
version of Readings. It was not possible to replicate the Reading figures by
varying the process we used to generate total per capita federal aid. Each
variable is summarized in Table B2.
The correlation between these two variables is strong (0.9333) but the
Treasury data has consistently smaller means. If we exclude Nevada - a

63

Table B2: Federal Aid Variables

Source
Office
of
Government
Reports

Construction
Mean
Sum of federal aid 293.44
to state (1933-39,
nominal) divided by
1930 population.

U.S. Trea- Sum federal aid to

sury
state (1933-39, real
1932 dollars) divided by weighted
average 1930/1940
population.

159.7

SD
178.14

Min
147.31

Max
1130.76

82.73

54.33

480.8

consistent outlier - the estimated fitted line between these two variables is
AidReading = 17.36 + 1.34 AidT reasury .
The robust 95% confidence interval on the slope coefficient is [1.04, 1.65]; the
difference in the aid variables is larger for states that received large amounts of
aid.57
Although these differences in federal aid variables are important they do
not affect our results. The general finding that the amount of federal aid
provided by the government was not significantly correlated with the change in
centralization in a state holds whether we use our preferred variable, Readings
variable, or New Deal spending only.

B.3
B.3.1

Control Variables
Debt Limitations

We count only constitutional debt limitations. There are two conflicting sources
for this data. Shawe (1936) reports that all states except Delaware, Mississippi,
57

The coefficient is larger if Nevada is included.

64

Vermont, Massachusetts, Tennessee, Maryland, New Hampshire, Connecticut,

North Carolina, and North Dakota had a tax limitation. Heins (1963) presented
a list equivalent to Shawe but also counted North Carolina and North Dakota
as having a tax limitation. This discrepancy might be attributed to differing
opinions of what constitutes a debt limit.58 North Carolina was limited to
borrowing 7.5% of its assessed valuation but borrowing below that limit could be
authorized by the legislature, a far less stringent process than the constitutional
amendment required by other states. The North Dakota legislature was also
allowed to issue debt but Not in excess of $2,000,000 unless secured by a first
mortgage (Shawe 1936, 125). Rueben (1994) uses the list provided by Heins
(1963). Our results are similar using either list.
B.3.2

Single Party Control

Single party control variables are computed from data provided by Burnham
(1985). Following Rueben (1994), we say that a state exhibits single party
control if a majority of both houses of the legislature and the governorship are
occupied by politicians from the same party in 1932. The majority of states
hold elections on even years and, therefore, report the relevant figures in 1932.
Three states - Mississippi, Kentucky, and Virginia - held elections in odd years
and only report the relevant figures in those years. When this is the case we
use the figures from 1931. One state - Maryland - only reported the relevant
figures for 1930; the unified variable for Maryland is constructed from this
data. Finally, Nebraska and Minnesota had non-partisan legislatures during
this period. These states are counted as not having single party control.

B.4

Instruments: Political and Land Variables

The political variables used as instruments to predict New Deal aid were first
introduced by Wright (1974). Robert Fleck (2008) was kind enough to provide
us with this data as well as the land variables data.
58

Heins (1963) also reports the year of adoption of tax limitation so the discrepancy is not
attributable to a change in the law between the thirties and the sixties.

65

B.5

Source Summary

Table B3 summarizes the sources used for all variables in the empirical analysis
presented in section 6.

Detailed Comparison with Wallis (1984)

In this appendix we compare in detail our results with those in Wallis (1984),
trying to reconcile our different findings. Recall that Wallis hypothesis is
that exogenous variation in federal grants - especially matching grants - to
the states, would lead to an increase in expenditures by state governments.
Further, state governments transferred some of the funds received by the federal
government to localities inducing them to reduce expenditures financed through
local taxation.
Consistent with this narrative, in his paper Wallis runs two types of regressions. The first type of regression (Table 3, page 154) links state expenditures
in a given year to federal aid received by that state in that year, for the years
19371940.59 The federal aid variable is instrumented using Wrights political
productivity measure and the standard deviation of the Democratic vote share.
Wallis finds that an exogenous increase in federal aid is associated with an
increase in state expenditures (see column 2 of Table 3). In the second set of
regressions, appearing on page 157 of his paper, Wallis relates the 19321942
change in aggregate local expenditures in a state to total intergovernmental
transfers received by local governments in that state. Notice that while local
governments received transfers from both states and the federal government, the
great majority of these transfers came from the states. In this second regression,
Wallis instruments the intergovernmental transfers variable using the same
instruments he had used for federal aid to a state in the first set of regressions
discussed above. The underlying logic here is that politically motivated (and
59

Specifically, the equation being estimated is equation (1) in Wallis (1984, page 150)
The choice of years was dictated by data availability issues, as discussed by Wallis (1984,
page 152): data pertaining to state government expenditures is available for 1932 and then
between 1937 to 1940. No local government expenditure or revenue data is available between
1933 and 1941.

66

Table B3: Source Summary

Variable
Tax Limit
Sales Tax
Difference State Revenue Share,
32-42
Difference State Expenditure
Share, 32-42
% Change in per capita income,
29-32
Federal Aid per capita, 33-39
(Treasury)
Federal Aid per capita, 33-39
(Reading)
% Renters, 1930
% School 27-28
% Urban 30
Income 29
% Dem. 28
Non-White Pop. 1930
Initiative State
Debt Restriction
Same Party Control
Republican Control
Federal Land per capita
Non-Federal Land per capita
Electoral Votes per capita
SD Dem. Vote, 96-32
Wright Index

Source
Mott and Suiter (1934)
Jacoby (1938), SFFF (1993)
Sylla, Legler, and Wallis (1995) [ICPSR 6304]
Sylla, Legler, and Wallis (1995) [ICPSR 6304]
BEA
U.S. Treasury (various years)
Reading (1973)
Haines (2010) [ICPSR 2896, part 29]
U.S. Office of Education (1930, pp. 984ff.)
Haines (2010) [ICPSR 2896, part 26]
BEA
Leip (2014)
Haines (2010) [ICPSR 2896, part 28]
Matsusaka (2000)
Heins (1963), Shawe (1936)
Burnham (1985) [ICPSR 00016, DS4]
Burnham (1985) [ICPSR 00016, DS4]
Fleck (2008) from Rand McNally (1992) and United
States Committee on Appropriations (1939)
Fleck (2008) from Rand McNally (1992) and United
States Committee on Appropriations (1939)
Fleck (2008) from Wallis (1998)
Fleck (2008) from Wright (1974)
Fleck (2008) from Wright (1974)

67

therefore plausibly exogenous) transfers from the federal government to the

states induced the latter to increase aid to local governments. Wallis finds (see
page 157 of his paper) that local spending declined in relative terms between
1932 and 1942 in states that were exposed to larger exogenous increases in
federal aid (to state governments).
These two sets of results appear consistent with Wallis top-down channel.
Moreover, they appear inconsistent with our finding (Table 8) that exogenous
variation in federal aid does not appear to affect the relative expenditure share
of state governments.
The following three observations should help to explain why our findings
are different from Wallis.
1. Wright (1974)s instruments for federal aid vary only across states by
not over time. Hence, the only variation that identifies the effect of
federal aid on state spending in Wallis first regression (Table 3 of his
paper) discussed above is cross-sectional. In other words, despite using
data for multiple years (19371940) for each state, Wallis is not using a
difference-in-difference estimator in this regression (Table 3).60 It follows
that his first set of results cannot be compared directly with ours. Hence,
we focus our discussion on his second set of results (those at page 157 of
his paper).
2. Wallis second estimated equation (appearing on page 157 of his paper)
has a difference-in-difference form, in that he regresses the change in
expenditures by a states local governments between 1932 and 1942 on
the change in intergovernmental transfers received by local governments
from (mostly) state governments. The latter is instrumented using the
political variables that generate exogenous variation in federal aid to
the states. Here, we show that we can in fact replicate Wallis second
set of results by running a version of his regression (page 157 of his
paper). The version we run relates the local of spending (relative to the
60

Notice that the fact that the instruments do not vary over time prevents Wallis from
including state fixed effects in the regression.

68

Table C1: Expenditure Centralization (IV).

(1)
Log Intergovernmental Transfers to Local
% Growth in per capita Income, 29-32

Electoral Votes per capita

SD Dem. Vote, 96-32

Oberservations
R2

(2)

-0.054**
(0.023)

-0.040*
(0.024)
0.164**
(0.065)
First-stage Coefficients
-74.973***
(13.206)
0.024
(0.022)

-75.692***
(15.731)
0.025
(0.023)

96
0.953

96
0.955

First stage regressions also include % growth in per capita income as well as state
and year fixed effects. Robust standard errors reported in parentheses. Significance:
*** 99% level, ** 95% level, * 90% level.

state-local combined share) to the intergovernmental transfers per capita

received by local governments in that state, using Wrights variables as
instruments. Notice that, while the explanatory variable is the same as
Wallis, we consider the local expenditure share, instead of the level of
local expenditures.61 Table C1 shows our results.
The table shows that an exogenous increase in intergovernmental transfers
to local governments is associated with a decline in local governments
expenditure share. Here, consistent with our analysis (see discussion in
the paper in Section 6.3.2), we use the states electoral votes per capita
and the standard deviation of the Democratic vote share as instruments
for intergovernmental transfers. The estimated coefficient on intergovernmental transfers to local governments is -0.054 with a p-value of 0.02.
In column 2 of this regression we include our 1929-32 income decline
61

The share approach is consistent with our previous analysis. Our results do not depend
on focusing on the share rather than the level. Notice that the local expenditure share is
one minus the state share.

69

measure in this regression. Notice that the coefficient on intergovernmental transfers drops from 0.054 to 0.04 and its p-value increases to
0.09. Also, the coefficient on the income decline has the expected sign, a
magnitude comparable to that reported in Table 8, and a p-value of 0.01.
Thus, part of the reason for Wallis second set of findings is the omission
of the 1929-32 income decline variable from the regression equation he
estimated.
3. The third observation is the most important. As described above, in Wallis second set of regressions, intergovernmental aid to local governments
is instrumented using Wrights political productivity measure and the
standard deviation of the Democratic vote share. The logic behind this
instrument is that politically motivated transfers to the states by the
federal governments should result in higher intergovernmental transfers
from the states to their local governments. In other words, in the firststage regression, higher values of the instruments should be associated
with higher values for the intergovernmental transfer variable. Table C1
reports the results of this first-stage regression. The analogous result is
not reported in Wallis (1984).
Notice that the only statistically significant instrumental variable is the
electoral vote one. Its t-stat is about 5.7. However, this variable enters
with the wrong sign in the first-stage regression. That is, everything
else equal, states with higher electoral votes per capita were characterized
by smaller increases in intergovernmental transfers per capita to local
governments. This negative partial correlation fundamentally questions
the logic of the top-down channel because it leads to the conclusion that
states that received more federal aid for political (and therefore plausibly
exogenous) reasons were characterized by a smaller increase in transfers
to local governments, and therefore (second-stage of the regression in
Table C1) by smaller declines in the share of expenditures accounted for
by local governments.

70

Hysteresis

Sales taxation and state centralization persisted long after the economic effects of the Depression lessened and income recovered and surpassed its preDepression peak. Few of the states that adopted sales taxes during the 1930s
subsequently abolished them.62 It thus appears that a temporary, although
very severe, economic downturn led to long-term institutional changes that
still persist after many decades. In this section we use the framework we
introduced in Sections 35 to discuss this phenomenon, sometimes referred to
as hysteresis. Our argument is based on the intuitive idea that institutional
change of the extent experienced during the Great Depression requires convergence of a broad coalition of political actors. In other words, while intensive
margin variation in a policy instrument, say a tax rate, might require only a
simple majority of votes, an extensive margin shift toward a different tax or
centralization of expenditures decisions will require a broader consensus.
We capture this idea in our model environment by postulating that a shift
from decentralization to centralization, or viceversa, requires support from a
fraction p of the population where p > 1/2. In the pre-Great Depression period,
the decentralized system represented the status-quo and might have enjoyed a
certain popular support of say q percent of the population with 1/2 < q < p.
In words, a majority of the population preferred the decentralized system,
but support for it did not exceed the threshold p. Proposition 3 shows that
this situation is consistent with a broad set of parameters, only ruling out a
situation in which and are sufficiently close to one another to generate a
unanimous support for decentralization.
Proposition 6 shows that the Great Depression increased political support
for sales taxation and centralization. From the perspective of the model, this
shift must have been supported by at least a fraction p of the population in
62

Five states (17%) that adopted a sales tax in the thirties allowed them to expire (see
asterisks in Table 1). From 1940 to 1995 seventeen further statesincluding the five states
that allowed their 1930s-vintage taxes to expireadopted the sales tax. These included
Connecticut (1947), South Carolina (1951), and Idaho (1965) (U.S. Advisory Commission
1993).

71

order to displace the old set of institutions.63 Thus, the economic shock of
the Great Depression gave rise to a new status-quo in which the states had
much more power vis-a-vis local governments. The key question is then why
this balance of power did not revert itself back to the original pre-Depression
situation after the end of the economic depression in the 1940s and beyond. In
the context of our model, we interpret the end of the Depression as a return
to an economic situation in which = 0. Under this circumstances, head tax
delinquency rates should be zero and the decentralized system enjoys support
of q percent of the population. While this is a majority, it is not broad enough
to reach the minimum support needed (p percent) to revert back to an economy
characterized by local control and financing of public goods, because q < p.
Therefore, the new status-quo persists despite the fact that more than half of
the population would prefer the alternative regime.
This narrative is consistent with the related and often-mentioned idea that
a status-quo might persist simply because an organized minority of individuals
are able to block a reform that would benefit a majority (e.g. Olson, 1965).
We capture this view, in the context of our model, by requiring a support of p
percent of the population for a reform to occur. We add to this idea the insight
that large economic shocks - such as the Great Depression - might temporarily
lead to an increased support for certain reforms which then persist long after
the economic shock and the additional political support for reform are gone.

Additional References
Anderson, Gary M., and Robert D. Tollison. Congressional Influence and
Patterns of New Deal Spending, 1933-1939." Journal of Law and Economics
34 (1991): 161.
Burnham, W. Dean. Partisan Division of State Governments, 1834-1985.
Conducted by Massachusetts Institute of Technology. ICPSR ed. Ann
Arbor, MI: Inter-university Consortium for Political and Social Research
[producer and distributor], 1984.
63

In the data, referenda that limited the property tax passed in many states with very
large majorities.

72

Fleck, Robert K. Population, Land, Economic Conditions, and the Allocation

of New Deal Spending." Explorations in Economic History 38, no. 2 (2001):
296-304.
Heins, A. James. Constitutional Restrictions Against State Debt. University of
Wisconsin Press, 1963.
Leip, David. 1932 Presidential General Election Data - National by State.
Accessed July 9, 2014. http://uselectionatlas.org/RESULTS/.
Office of Government Reports, Statistical Section. Federal Loans and Expenditures, Vol. II, Washington D.C. 1940.
Olson, Mancur. The Logic of Collective Action: Public Goods and the Theory
of Groups. Cambridge, MA: Harvard University Press, 1965.
Rand McNally. The New Cosmopolitan Worl Atlas. Chicago: Rand McNally,
1992.
Shawe, Earle K. An Analysis of the Legal Limitations on the Borrowing Power
of the State Governments." FERA Monthly Report, June 1936.
Tax Limit Trend Noted in 8 States." The New York Times, March 26, 1939,
160."
US Department of the Interior, Office of Education. Biennial Survey of
Education, 1936-1928." Washington, DC: GPO, 1930.
US Senate Committee on Appropriations. Work Relief and Public Works
Appropriation Act of 1939. Hearings, 76th Con., 1st sess. Washington, DC:
GPO, 1939.
Wallis, John J. Employment, Politics, and Economic Recovery During the
Great Depression." The Review of Economics and Statistics (1987): 516-520.

73