GOVERNMENT SPENDING PRIORITIES:
A CROSS-NATIONAL PERSPECTIVE
By
Niccole M. Pamphilis

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Political Science
2012

ABSTRACT
GOVERNMENT SPENDING PRIORITIES:
A CROSS-NATIONAL PERSPECTIVE
By
Niccole M. Pamphilis
In the work to follow, I examine government spending priorities across 25 democratic
nations from 1990-2009. The goal of this research is to provide a better understanding of how
and why governments spend different amounts of money on similar types of public policies.
Specifically, I look at how expenditures are divided across a range of policy and how this
translates into interpreting government spending patterns. I further explore how commonly
found influences on government expenditures relate to spending priorities. Finally, I consider
how the number of institutional constraints present in a nation interacts with both mass and elite
preferences to decrease the responsiveness of democratic governments.
Using expenditure data on ten different policy areas, I construct a single measure of
government spending. This measure is more encompassing than prior measures that use fewer
policy areas or combine items that represent different aspects of the policy process. To do this, I
apply a unidimensional, metric, least-squares unfolding technique to the data. I find that policies
group into two distinct clusters. The results show a simple-to-interpret dimension of spending
where governments trade-off between particularized benefits that target specific groups within a
society, like the elderly, and collective goods that are intended to benefit society in more general
terms through areas such as education or economic development. The measure also captures
compromise by governments on its outputs as it expresses how governments spend scarce
resources across a range of policy domains.

After establishing how governments spend, I show why governments allocate their
resources to different policies. I argue that previous works use a combination of misspecified
models and measures of government outputs to explain government spending. The spending
priorities variable offers an improvement for examining the public policy outputs of
governments. I merge several arguments regarding spending patterns and find that the available
resources, what citizens want and need, as well as the individual institutions present in a nation
shape spending priorities. The results show how aspects from each separate theory influence
spending when analyzed in a more fully specified model.
The final section of this dissertation examines how the separate components of the
political system in a nation have a cumulative influence on government spending, expanding on
the individual effects explored in the literature to date. Institutions in a nation that increase the
number of actors involved in the decision making process, referred to as institutional constraints,
decrease the ability of governments to spend on policy areas that target particular groups, like the
unemployed. Instead, these attributes shift spending in a direction that favors society in broader
terms with spending on areas such as defense or environmental protection. The constraints in a
nation also mitigate the influences elite and mass preferences play in shaping government
spending, thereby making governments less responsive to demands. This finding suggests that
the exclusion of the institutional constraints from models may overstate the role citizens play in
shaping government outputs.

To my husband, Steven, your constant encouragement and support gave me the strength to press
on to the end.

iv

ACKNOWLEDGEMENTS

I would sincerely like to thank Saundra K. Schneider for her invaluable assistance and guidance
that has helped me to grow as a researcher, a political scientist, and an instructor. I would like to
thank William G. Jacoby, whose comments and suggestions, over the years have helped me to
become a better researcher. I would also like to thank the other members of my dissertation
committee, Ani Sarkissian and Christopher Maxwell, for their time and assistance with my work.
The chapters comprising this dissertation also benefited greatly from feedback from fellow
graduate students at Michigan State University, including Kurt Pyle, Robert N. Lupton, Seo
Youn Choi, Petra Hendrickson, Dominique Lewis, and Daniel Thaler.

v

TABLE OF CONTENTS

LIST OF TABLES .......................................................................................................... viii
LIST OF FIGURES .......................................................................................................... ix
CHAPTER 1
INTRODUCTION ............................................................................................................. 1
OVERVIEW .......................................................................................................... 6
CHAPTER 2
A REVIEW OF THE LITERATURE ............................................................................... 10
EXPENDITURES AS GOVERNMENT OUTPUTS ............................................. 10
SOCIO-ECONOMIC INFLUENCES ................................................................... 17
POLITICAL PREFERENCES AS INFLUENCES ................................................ 22
INSTITUTIONS AS INFLUENCES ..................................................................... 30
INSTITUTIONAL CONSTRAINTS AND POLICY RESPONSIVENESS ............. 35
CONCLUSION.................................................................................................... 39
CHAPTER 3
GOVERNMENT SPENDING PRIORITIES .................................................................... 41
DATA SELECTION ............................................................................................. 42
UNFOLDING ...................................................................................................... 53
Details of the Unfolding Procedure .......................................................... 54
WHY UNFOLDING? ........................................................................................... 56
Data Reduction ........................................................................................ 56
Original Data ............................................................................................ 57
Single Dimension ..................................................................................... 58
No A Priori Assumptions .......................................................................... 61
Reliability ................................................................................................. 61
RESULTS OF THE UNFOLDED EXPENDITURE DATA ................................... 62
Differences between Nations’ Spending Priorities and Policy Points ...... 68
Differences between Nations’ Spending Priorities .................................. 70
Close Examination of Spending Priority Scores ................... ……………..73
CONCLUSION.................................................................................................... 75
CHAPTER 4
DATA AND HYPOTHESES ........................................................................................... 77
FACTORS INFLUENCING GOVERNMENT SPENDING ................................... 77
POLICY RESPONSIVENESS HYPOTHESES ................................................... 95

vi

CHAPTER 5
TRADITIONAL INFLUENCES AND SPENDING PRIORITIES ................................... 100
SPENDING PRIORITIES MODEL .................................................................... 101
Socio-Economic Factors and Spending Priorities .................................. 101
Group Preferences and Spending Priorities........................................... 106
Institutions and Spending Priorities........................................................ 110
Country Examples.................................................................................. 113
OLD MODELS, NEW MEASURE ..................................................................... 115
CONCLUSION.................................................................................................. 124
CHAPTER 6
INSTITUTIONAL CONSTRAINTS AND POLICY RESPONSIVENESS ...................... 125
THE ROLE OF INSTITUTIONAL CONSTRAINTS ........................................... 128
MODEL ............................................................................................................. 130
Fixed and Random Effects ..................................................................... 135
RESULTS ......................................................................................................... 136
Controls ................................................................................................. 138
Institutional Constraints.......................................................................... 140
IMPLICATIONS FOR NATIONAL SPENDING PRIORITIES ............................ 150
CONCLUSION.................................................................................................. 154
CHAPTER 7
CONCLUSION ............................................................................................................ 156
GENERAL FINDINGS ...................................................................................... 156
EXTENSIONS AND IMPLICATIONS ................................................................ 159
CONCLUSION.................................................................................................. 164
APPENDICES ............................................................................................................. 167
APPENDIX A: DISTRIBUTION OF SPENDING BY POLICY AREA ................ 168
APPENDIX B: SEPARATE SPENDING MODELS BY SET OF INFLUENCES
......................................................................................................................... 175
APPENDIX C: DIAGNOSTIC TESTS AND MODEL SELECTION ................... 178
Time Dummies ....................................................................................... 178
Lags ....................................................................................................... 178
Transformation ....................................................................................... 180
Multicollinearity ...................................................................................... 183
Residuals ............................................................................................... 183
Influential Observations ......................................................................... 184
REFERENCES ............................................................................................................ 198

vii

LIST OF TABLES

Table 3.1 Democratic Nations and Time Periods .......................................................... 44
Table 3.2 Examples of Expenditures by Policy Area ..................................................... 45
Table 3.3 Average Error in Capturing Actual Spending with Unfolding ......................... 58
Table 3.4 Policy Typology using Lowi’s Categories....................................................... 59
Table 3.5 Exploratory Factor Analysis of Policy Areas .................................................. 60
Table 4.1 Results of Factor Analysis for Interest Groups .............................................. 89
Table 4.2 Summary Statistics ........................................................................................ 98
Table 4.3 Summary of Hypotheses ............................................................................... 99
Table 5.1 Spending Priorities Model............................................................................ 102
Table 5.2 Replication of Milesi-Ferretti et al. Model using Spending Priorities ............ 117
Table 5.3 Replication of Huber and Stephens Model using Spending Priorities .......... 120
Table 6.1 Traditional Influences of Government Spending Priorities ........................... 131
Table 6.2 The Effect of Institutional Constraints on Policy Responsiveness ............... 137
Table 6.3 Government Composition and Spending Priorities in the United Kingdom .. 142
Table 6.4 Observations for the United Kingdom from 1990-2009 used in the
Interaction Model ......................................................................................... 150
Table B.1 Influences of Socio-Economic Factors on Spending Priorities .................... 175
Table B.2 Influences of Group Preferences on Spending Priorities ............................ 176
Table B.3 Influences of Institutions on Spending Priorities .......................................... 177
Table C.1 Variance Inflation Factor Scores ................................................................. 186
Table C.2 Correlations Matrix for Independent Variables in Interaction Model ............ 187

viii

LIST OF FIGURES

Figure 3.1 Distribution for the Proportion of Spending on Health ................................. 46
Figure 3.2 Distribution for the Proportion of Spending on Social Protection ................. 47
Figure 3.3 Distribution for the Proportion of Spending on Public Order and
Safety ........................................................................................................... 48
Figure 3.4 Location of Unfolded Policy Points ............................................................... 63
Figure 3.5 Distribution of Spending Priorities within Nations over Time ........................ 69
Figure 3.6 Distribution of Spending Priorities over Time ............................................... 72
Figure 3.7 Distribution of Spending Priorities across Nations........................................ 74
Figure 6.1 Distribution of Spending Priorities by Nation over Time ............................. 127
Figure 6.2 Distribution of Spending Priorities over Time for Nations with Three or Four
Institutional Constraints .............................................................................. 128
Figure 6.3 Number of Years by Nation in the Panel Data............................................ 133
Figure 6.4 Predicted Spending Priorities for Government Composition ...................... 144
Figure 6.5 Predicted Spending Priorities for Role of Government ............................... 146
Figure 6.6 Predicted Spending Priorities for Public Opinion ........................................ 147
Figure 6.7 Predicted Spending Priorities for Interest Groups ...................................... 149
Figure 6.8 Predicted Spending Priorities for the United Kingdom with Zero and Three
Institutional Constraints .............................................................................. 152
Figure 7.1: Nation Gini Coefficients against Spending Priorities ................................. 162
Figure A.1 Distribution for the Proportion of Spending on Defense ............................. 168
Figure A.2 Distribution for the Proportion of Spending on Economic Development .... 169
Figure A.3 Distribution for the Proportion of Spending on Education .......................... 170
Figure A.4 Distribution for the Proportion of Spending on Environmental Protection .. 171
ix

Figure A.5 Distribution for the Proportion of Spending on Government Operations .... 172
Figure A.6 Distribution for the Proportion of Spending on Community Development .. 173
Figure A.7 Distribution for the Proportion of Spending on Recreation ......................... 174
Figure C.1 Component plus Residual Plot for the Natural Log of GDP/Capita ............ 188
Figure C.2 Component plus Residual Plot for the Natural Log of Unemployment ....... 189
Figure C.3 Component plus Residual Plot for the Aged Population ............................ 191
Figure C.4 Component plus Residual Plot for Government Composition .................... 192
Figure C.5 Component plus Residual Plot for Role of Government ............................ 193
Figure C.6 Component plus Residual Plot for Public Opinion ..................................... 194
Figure C.7 Component plus Residual Plot for Interest Groups .................................... 195
Figure C.8 Scatter Plot of Residuals against Fitted Values………………… ...……...…196
Figure C.9 Plot of Leverage and Residuals by Observation ........................................ 197

x

CHAPTER 1
INTRODUCTION
All democratic societies are faced with a multitude of demands from citizens, ranging
from building a strong and growing economy, to providing emergency services, to helping those
in need. In order to address these expectations, governments spend on a variety of policies to
alleviate or prevent the cause of societal strife including spending on economic development, on
fire and police services, and on areas of social protection. While nations face similar issues, not
all governments prioritize the problems in the same manner nor do they always respond to the
same issue in the same manner.
Looking at program expenditure data from democratic nations provides initial evidence
that there is a great deal of variation in how governments address societal problems. Variation is
evident in both the proportion of total spending across policy areas and in the level of spending
within similarly ranked policy areas. Twenty-two of the twenty-five nations examined in this
analysis spend the most on social protection in terms of total spending, which includes programs
such as survivor benefits, old-age pensions, and unemployment insurance than any other
alternative policy areas (i.e., defense or education).

1

Even among nations that dedicate the majority of their total l expenditures to social
protection, there is a sizable degree of variation. For example, Austria and Iceland both spent the
most on social protection relative to other policy areas in 2002, but they differ in the percentage
1

The three exceptions include South Korea, Canada, and the United States. South Korea spends
more on economic development, which focuses on aspects such as fuel and energy,
transportation, and communication, with social protection typically ranking sixth in terms of the
percentage of total expenditures. Canada spends more on government operations, which
includes administrative costs and foreign economic aid. And since 2004, health expenditures
have replaced social protection as the number one spending area in the United States at
approximately 20% of total expenditures.
1

of spending devoted to social protection relative to their overall spending profiles. In 2002,
Austria dedicated approximately 40% of its total expenditures to the area of social protection,
while Iceland spent only about 20% of its total expenditures on the area of social protection.

2

Furthermore, Iceland appears to have a relatively balanced spending profile with the top
three expenditures of social protection, health, and education each receiving roughly 18-20% of
total expenditures. Meanwhile in Austria, social protection received a much larger share
compared to the second highest spending area government operations, which received
approximately 14% of total expenditures. This example demonstrates that although nations may
share similar rank-order spending preferences across various policy areas, there are noticeable
differences in their resource allocations to particular policy areas.
Prior work examining the variation in government spending has resulted in a variety of
incorrect measures of government outputs. Previous studies have used measures on government
transfer payments, changes in spending on policies, spending on sets of policy areas in isolation
from one another, and on specific expenditures within policy areas (such as pension plans or
unemployment insurance). Research that focuses on an individual policy domain ignores the
possible connections that might exist between policy areas, where increasing spending on any
one policy area, like defense, reduces the resources available to spend on alternative policies,
such as education.
Studies that use composite measures of governmental activity do increase the number of
policy areas examined, but they also make implicit assumptions about what policy areas can be
grouped together. For example, Huber and Stephens (2001) place old-age pensions and health
2

For the 25 OECD nations examined in this analysis from 1990-2009, the range of spending on
social protection for nations, where it was the number one spending area, ranged from roughly
18% of expenditures to 47% with a mean value of 36% and a standard deviation of 6%.
2

care expenditures in the same category. Such categorizations, however, may mask important
differences between program in terms of their intended beneficiaries or the scope of their efforts.
In order to address these issues, I create a measure of government spending priorities that
encompasses a number of policy areas without making a priori assumptions about how the policy
areas should be categorized. To produce a measure of government spending priorities, I apply an
unfolding technique, developed by Jacoby and Schneider (2001) for use on the American states,
to create a single measure of government spending priorities that captures expenditures over ten
policy areas for 25 democratic nations. The priorities variable results in any easy-to-interpret
dimension that distinguishes between policies that favor specific groups in society like the
unemployed, referred to as particularized benefits, versus policies that provide broader collective
goods across society, such as defense and economic development.
The measure of spending priorities provides an answer to the first question my
dissertation addresses: Can government activities be captured in a more parsimonious, reliable,
and encompassing manner than in previous works? The findings not only produce a single
variable that is capable of expressing expenditures across a range of policy areas in a
parsimonious and easy-to-interpret manner, it also expands the work by Jacoby and Schneider
(2001, 2009) and Schneider and Jacoby (2006) on government spending priorities in the
American states and the theoretical work on spending trade-offs (Banks and Duggan 2000, 2005;
Lizzeri and Persico 2001; Volden and Wiseman, 2007).
Using this new measure of spending priorities for democratic nations, the second
question I address is: Do factors traditionally found to shape government spending patterns still
influence an encompassing measure of government spending? The evaluation of prior work in
relation to the new dependent variable includes measures from functionalist arguments that focus

3

on economic wealth, market openness, and the size of dependent populations; indicators
involving the preferences of different groups in society, such as political parties, the general
public, and organized interests; and the role of institutions in shaping the behavior of political
actors and general citizenry. These separate approaches, however, result in misspecified models
as researchers typically omit the influence of concepts presented in alternative arguments. I
combine these three sets of factors to create a better specified model that controls for a number
of variables that are argued to influence spending. Through this approach I am able to determine
the correct influence of each variable in relation to spending without having to question how
omitted variables are biasing the estimates of the coefficients in the model. Additionally, I am
better able to explain the variation in government spending priorities than any one set of factors
tested separately.
I further expand the understanding of variation in expenditure patterns by exploring how
institutions that constrain governments’ abilities to act can impede democracy. Institutions that
can constrain the ability of governments to act include presidential systems, electoral formulas
such as proportional representation and high district, bicameral legislatures, and federal systems
(Tsebelis 1995, 2002; Huber and Stephens 1993, 2000, 2001; Cox and McCubbins 2001). All of
these constraints increase the number of actors who have preferences over policy outputs in the
decision making process, making agreement on any issue more difficult.
Tsebelis (1995, 2002) argues that as the number of institutions constraining government
actions increase in a nation, the range of policy decisions over which agreement can be reached
can only get smaller. The decrease in viable policy solutions that can be agreed upon makes it
more difficult to enact policies that diverge from the status quo. Similar findings have been
found in the work of Cox and McCubbins (2001) where, as decision making power is separated

4

amongst multiple actors, it become more difficult for actors in the policy making process to
agree and the status quo becomes more resolute.
The prior research regarding the effect of institutions on legislation has generally been
restricted to the ability to pass legislation and has not focused on its relationship with the
resulting outputs of government. In an attempt to push the understanding of institutional
constraints further, I argue that constraints also shape the ability of different actors to obtain their
preferred spending priorities based on the number of institutional constraints present in a nation.
This leads into the third question I address: Do constraints alter the role of preferences in shaping
spending priorities in democratic systems?
Institutions that are labeled as constraints are those that increase the number of actors
with ideal policy outcomes present in the decision making process. As the number of constraints
increase, the number of preferences in the decision making process increase, and so does the
difficulty of reaching agreement on policy. In order for any policy to be enacted as the number
of constraints increases, compromises and bartering will have to occur in order to reach
agreement. I argue that the act of bartering and compromise prevents governments from
spending more on policy areas that target particular groups within the population and may
provide little to no benefit for some actors present in the decision making process. Instead,
government with more constraints will focus spending on policy areas that target society in more
general terms, and provide benefits for all the actors with preferences for government spending.
Further, through this process, no particular group is in a position to obtain its ideal policy outputs
from government. Therefore, as the number of constraints increases, any groups’ preferences
should matter less for what governments do, in terms of matching spending priorities to different
groups’ expectations. These findings are important for understanding why governments in

5

similar situations can have drastically different spending patterns and why governments with
citizens who demand one policy output, such as more spending on unemployment benefits, may
end up doing something entirely different, such as focusing on promoting economic growth
through tax cuts and stimulus packages.
OVERVIEW
The analyses presented in this work covers 25 democratic nations that are members of
the Organization for Economic Co-operation and Development from 1990-2009. In order to
answer the three questions set out above, I use a measure of government spending priorities
based on expenditure data for ten policy areas: government operations, social protection, health,
education, economic development, community development, defense, public order and safety,
environmental protection, and recreation. Expenditure data are useful to capture what
governments do in a given year as expenditures represent the end product of how scarce
resources are allocated across a variety of policy areas (Garand 1985). For example, Obinger
and Wagscahl (2010) use social expenditures as a means of understanding the mix of social
policies in place in different nations. I use government expenditures across a range of policy
areas as a means of understanding the overall policy mix present in nations. Taken as a whole,
expenditures show the pattern of behavior by a government and how it prioritizes its actions
(Dean 2006). Additionally, money represents a government’s commitment to a policy area
(Jacoby and Schneider 2001), and as Klingemann et al. (1994) aptly state, “money is not all there
is to policy, but there is precious little without it” (41).
Chapter 2 provides a review of prior work regarding why expenditures are an appropriate
avenue for evaluating government actions and compares this approach to alternative measures
that are used. This chapter also discusses the factors that are found to influence the variation in

6

spending priorities within and across nations, including socio-economic conditions, mass and
elite preferences, and institutional arrangements. I also examine the literature on institutional
constraints and explain my argument for how and why constraints should influence the spending
priorities of democratic nations. Chapter 2 also includes a discussion of the theoretical
justifications for the measures that are used to examine potential influences on governmental
actions.
In Chapter 3, I present the dependent variable used to capture government spending
priorities. This builds on the work done on spending priorities in the American states by Jacoby
and Schneider (2001, 2009) by extending their work to government spending across a number of
democratic nations. Chapter 3 explains the unidimensional, least-squares, unfolding technique
that is used to produce the measure of government spending priorities and provides an
explanation of how to interpret the values of the variable. The chapter concludes with a
discussion of the additional findings from the unfolding and answers the first question of the
dissertation: Can government activities be captured in a more parsimonious, reliable, and
encompassing manner than in previous works?
Chapter 4 translates the expectations from the prior findings presented in Chapter 2 into
testable hypotheses that are used in the subsequent analyses sections. The two sets of hypotheses
focus on the expectations for how influential factors should affect the new dependent variable,
and how institutional constraints affect government outputs and the role of preferences in
shaping policy expenditures. Additionally, following each hypothesis, the chapter presents how
each factor is operationalized, including the sources of the data and a discussion of any
additional calculations that are applied to the data to create the variables used in the following
chapters.

7

Chapter 5 provides an answer to the second question of the dissertation: Do factors
traditionally found to shape government spending patterns still influence an encompassing
measure of government spending? Chapter 5 provides empirical analyses on how socioeconomic factors, mass and elite preferences, and institutions previously found in the literature
affect the new measures of government spending priorities created in Chapter 4. Prior research
examines a number of variables but tends to focus on more limited measures of government
activities, usually expenditures on individual policy areas such as welfare, health, education, or
transportation. The new government spending priorities variable captures a range of government
expenditures across ten policy areas, providing a more parsimonious measure that can be used to
examine the potential influential factors shaping the variation in government spending
allocations.
In Chapter 6, I explore how institutional constraints might alter the spending allocations
of democratic governments. The analysis reveals that as the number of institutional constraints
increases, nations spend more on collective goods that are intended to provide benefits to all
members of society. Further, institutional constraints reduce the ability of democratic
governments to respond to the demands from both elites and masses. The interaction between the
number of institutional constraints and preferences reduces the cumulative influence of both
mass and elite expectations.
Chapter 7 concludes the work presented throughout the dissertation, provides
implications for past findings, and gives points to consider in future analyses. I discuss how the
measure of government spending priorities can be used in future empirical analyses, particularly
as an explanatory factor of policy outcomes. I explain how factors typically argued to affect
government spending patterns work when policy areas are examined simultaneously instead of in

8

isolation of one another. The findings suggest that certain variables representing influences like
globalization, may not alter government activities as prior studies suggests, highlighting the
importance of reliable, encompassing measures when attempting to understand government
spending patterns. I end with a discussion on how the omission of the interaction between
institutional constraints and preferences can actually overstate the role citizens’ preferences play
in democratic nations.

9

CHAPTER 2
A REVIEW OF THE LITERATURE
What causes the variation in spending patterns across democratic nations? A number of
scholars examine the variation in policy outputs in nations around the world and over time. The
different studies approach the examination of government activities in a variety of manners,
including the types and volume of legislation enacted, expenditures on individual policy areas,
and changes in spending. Regardless of how government actions are measured, a number of
factors are continually argued to shape the differences that exist both within and across nations.
Some of the most frequently used indicators include the socio-economic climate, the distribution
of power among members of society, and the institutional arrangements that exist within
particular nations.
EXPENDITURES AS GOVERNMENT OUTPUTS
Drawing on both Easton (1953) and Salisbury (1968), public policy represents the end,
aggregate product of what a government does with its time in office. As I wish to understand
the variation in government spending and what causes the differences seen in government
activities, examining public policy as an output in relation to traits in a nation presents itself as a
starting point. Salisbury notes that public policy “is patterns of behavior, rather than separate,
discrete acts” (153). As public policy represents an array of actions in relation to one another, I
need a means of capturing the variation in actions of democratic governments in a concise
manner.
One approach that is used to examine and capture the variation in government outputs is
through the use of typologies. Typologies are used to breakdown the different policy decisions a
government makes, from the groups in society who have access to services, who pays for the
policy, and what level of government is responsible for particular policy areas (Lowi 1964, 1972;
10

Salisbury 1968; Peterson, 1995; Savas 2000; and Wilson 1989). Based on historical patterns,
Lowi (1964) originally argued that governments create three types of policies. These policy
types are categorized according to their “impact on society” in the form of distributive policies
where no group is deprived of benefit (like national defense), regulatory policies that determine
what can and cannot be done and by whom (like labor laws), and redistributive policies that
focus on the divide between the “haves” and “have-nots” (like housing vouchers). When,
however, Salisbury (1968) re-examined the different policy decisions, he concluded that there
are four types of policies based upon the fragmentation of the demand pattern and the decision
system. Even Lowi (1972) later expanded his typology to include a fourth category referred to
as constituent policies.
While typologies may be useful to present a common understanding and simplify
complex topics, individuals may understand the issues presented in the typologies differently
(Baumgartner and Jones 1993). Depending on the dimensions used to create the typology,
different researchers can arrive at contradictory conclusions. For example, contrast Lowi’s
(1964) typology that focuses on who is affected in society, to Salisbury’s typology (1968) that
looks at how fragmented demands are versus the decision making body, and finally Peterson
(1995) who emphasizes the role of spending to either promote the economy or to address
divisions between the haves and have-nots.
Beyond the conflict over what the correct dimensions are, typologies can lose some of the
more intricate details of specific policies. For instance, take the example of Austria and Iceland
presented in Chapter 1. Spending on social protection would be categorized as a redistributive
policy area using Lowi’s typology. However, the variation in the level of resources dedicated by

11

each nation to social protection, as well as the variation in spending profiles for each nation is
missed (or disguised) if Lowi’s typology is used.
Even when researchers are able to create a typology that they believe represents the
events they are attempting to classify, the ability to place real world, complex events into
simplified typologies rarely results in clean classifications. Salisbury (1968) argues that it does
not matter what categories a researcher uses as the real world is fluid. Further, the use of
typologies decreases the level of measurement for expenditures from interval level variables to
nominal variables, reducing the information available for testing. Once the spending data are
classified into categories, variations over time and within policy areas are lost. For example, the
change in spending priorities for the United States in 2004 when health spending surpassed
social protection would not be evident.
An alternative to fitting policies within the confines of a typology has been to examine
the activities of the legislature in terms of time spent handling particular policy issues and
changes to legislation (Page and Shapiro 1983; Baumgartner and Jones 1993; Kingdon 1984;
Heller and McCubbins 2001; Haggard and Noble 2001). Examining the amount of time a
government spends talking about a policy area or issue can provide insights into what factors
brought the topic to the forefront of discussion; however, attention to a policy area or issue does
not necessarily imply a change in terms of how a government responds to these issues. For
example, in the United States from 1993-94, health care received a high level of attention;
however, in the end, attempts at reform were unsuccessful at changing how the United States
handled health care.
Additionally, a number of the studies that examine media, public, and government
attention to policy areas and resulting changes in specific legislation do not present predictive

12

models to understand how government is acting now or will act in the future given a set of
circumstances (Baumgartner and Jones 1993; Kingdon 1984). Typically, models of this form are
retroactive. For example, Kingdon’s (1984) three streams involving problems, solutions, and
policy entrepreneurs and windows of opportunity can help explain why policies change at one
point in time versus another, but do not offer the ability to predict when future policies or
changes to current policies will occur and what the final policies will contain.
Government expenditures lend themselves as an avenue to capturing what governments
do in a given year. Expenditures tell us how governments allocate scarce resources across a
variety of policy areas (Garand, 1985). As such, a growing consensus has emerged in the
literature that, “expenditures across substantive areas provide accurate representations of
governmental commitments to address various problems” (Jacoby and Schneider 2009, 3).
Government activity has been measured using individual expenditures like health, welfare,
education, defense, and transportation area (Obinger and Wagschal 2010; Chang 2008; Ringquist
1999; Budge and Keman 1990; Huber and Stephens 2001; Bräuninger 2005; Garand and
Hendrick 1991; Shelton 2007; Garand 1985; Soroka and Wlezien 2005; Penner, Blidook, and
Soroka 2006) and changes in expenditures (Klingemann et al. 1994; Haggard and McCubbins
2001; Garand 1985; Baumgartner and Jones 1993; Hofferbert and Budge 1992; Breunig et al.
2009; Soroka and Wlezien 2005; Persson and Tabellini 2003; and Jones et al. 2009). The use of
a single indicator to measure policy priorities can help determine what factors affect a given
policy area in isolation of the other policy decisions. On the other hand, a single indicator is not
capable of showing the effect of factors on a system of policy priorities. By using a single
indicator, the researcher makes the assumption that policy areas are not linked together.
However looking at real governments and decisions they must make, systems produce a full

13

range of policy expenditures which co-exist, and increasing spending on one policy area deprives
alternative policy areas of monetary resources.
The use of composite measures, such as additive scales and factor analysis, is another
approach that is used to study government spending priorities (Hofferbert 1974; Erikson, Wright,
and McIver 1989; Klingman and Lammers 1984). Hofferbert (1974) uses a factor analysis
combining both a variety of policy expenditures and policy outputs resulting in two dimensions
of choices in the American states composed of a welfare-education dimension and a highwaysnatural resources dimension. Klingman and Lammers (1984) use a principal components
analysis on six policy areas that mix expenditure data and non-fiscal measures involving policy
outputs over time. Though not related to expenditures, Erikson, Wright, and McIver (1989) use
an additive scale based on legislation representing policy liberalism to create a composite
measure of policy in the American states.
While the development of composite measures moves in the right direction by creating
more encompassing measures of government activity, the use of factor analysis and additive
scales to construct such measures has several limitations. Though factor analysis can reduce the
number of variables needed to capture information, it tends to produce at least two dimensions to
represent the underlying structure of policy areas. For example, in Hofferbert’s research (1974)
factor analysis was used to reduce four policy areas to only two dimensions.
At the same time, factor analysis is based around observed correlations being a product of
unobservable variables. The correlations between the observed variables are used to create the
underlying dimension(s). The variables that are seen to load on the same factor in the analysis
are then assumed to move together as change occurs, which is not an intuitive finding (Jacoby
and Schneider 2001). In the case of Hofferbert, (1974) highways and natural resources load

14

together on the same factor, implying that as expenditures increase for highways so do
government expenditures on natural resources, which is not necessarily true. Nor is it intuitively
logical that as spending on highways increases or decreases spending on natural resources will
mirror those changes in spending.
When using a factor analysis or additive scale, the researcher is responsible for selecting
the variables that will represent the underlying dimension and typically involves the combination
of data representing different parts of the policy process including inputs, outputs, and outcomes.
The range of indicators representing different stages of the policy making process makes it
difficult to determine what part of the policy process is being influenced in any analysis (Jacoby
and Schneider 2009). Not only do researchers combine a variety of measures, the variables may
come from different time periods. Using data from different time points, like combining data
from different parts of the policy process, can increase the difficulty of determining how factors
that influence policy work across time. In the case of Hofferbert (1974) and Klingman and
Lammers (1984), the analyses combine variables that represent both outputs in regards to
expenditure levels and outcomes in regards to quality of the policy areas such as high school
completion rates, infant mortality rates, and policy innovations in the form of enacted policies. If
these measures are used in empirical models it would be impossible to tell if the independent
variables are influencing the outputs of expenditures or the outcomes of the policy decisions.
The selection of variables in the composite measure can also omit categories that are of
importance in the policy process. For example, Hofferbert (1974) excludes health policy issues
and general spending because the variables do not load onto the two factors solution. However,
expenditures on health in the United States include Medicaid which is traditionally seen as part
of welfare and would therefore be expected to be a part of the welfare-education dimension.

15

Additionally, Klingman and Lammers (1984) omit social services, and Erikson, Wright and
McIver (1989) omit policy areas that are not believed to have a partisan interest, such as
highways.
If a democracy is a system which translates preferences into policy, and if expenditures
are linked to government actions, spending priorities should reflect preferences for government
actions. Examining the percentage share of total expenditures a policy area receives; its relative
importance compared to other areas can be found and used to determine a government’s
spending priorities. Spending priorities are the rank order of spending on policy areas. A policy
area receiving the largest portion of total expenditures in a given year will be ranked higher in
terms of priorities than alternative policy areas (Garand 1985; Hofferbert and Budge 1992;
Ringquist and Garand 1999; Jacoby and Schneider 2001, 2009).
A better way of capturing governmental outputs involves combining expenditure data,
which are argued to provide a measure of what governments do, over a variety a of policy areas
in a manner that is reliable, parsimonious, and substantively meaningful based on the data. This
is exactly what is done in the case of the American states to examine government spending
priorities (Jacoby and Schneider 2001, 2009; Schneider and Jacoby 2006). Jacoby and Schneider
use a spatial proximity model referred to as a unidimensional, metric, least-squares unfolding
analysis on government policy expenditures to create a measure of government policy priorities
that represents a continuum of policy packages. This approach allows for testing across multiple
policy areas at one time in a cohesive manner that is not based on correlations and retains the
uniqueness of the individual observations that are used to create the measure of government
policy priorities, compared to individual policy areas. Further, the data are not pre-selected
based on how the policies should group together or what the policies should represent. Instead,

16

the unfolding allows the data itself to determine what the underlying dimension of spending
involves, opposed to composite measures resulting from factor analysis, additive scales, or
principal components analysis.
When studying the pattern of government behavior, researchers apply a variety of labels
to different groups of policy areas including: targeted goods, public goods, distributive policy,
redistributive policy, rents, purchases, transfers, particularized benefits, and collective goods.

3

Moving forward throughout the dissertation, policy groupings are referred to as either
particularized benefits or collective goods. The labels I apply throughout this work are based on
a policy dichotomy that has emerged in the literature (Volden and Wiseman 2007; Jacoby and
Schneider 2001, 2009). Particularized benefits are defined as policies intended to benefit
specific subgroups within a population and include items such as old-age pensions and
unemployment insurance. Collective goods are policies intended to benefit the more general
population and include areas such as defense and environmental protection.
SOCIO-ECONOMIC INFLUENCES
There are a number of indicators that are repeatedly argued and found to influence the
variation in government activities. One set of factors falls under the functionalist argument, that
it is the socio-economic climate that shapes what governments do. Under a functionalist
argument, policies of any type are the product of economic resources and the demands for those
resources. In particular, “social policies are the unmediated response to social and economic
pressures. . .” and “…intervening forces such as the political organization of social demands or
3

While these labels have been repeatedly used, there is wide variation in what is included under
the heading from research to research. While at times there is overlap, policy areas have also
been seen to switch sides; take, for example the use of the term public goods. Public goods have
been shown to include healthcare, welfare, and education in one instance (Edwards and Thames
2006) and spending on bridges in another (Milesi-Ferretti et al. 2002), where welfare fell under
transfers.
17

governmental institutions are assumed to be either neutral towards or fully determined by the
socio-economic change” (Zutavern and Kohli 2010, 173). A range of socio-economic factors are
said to influence how governments spend money and prioritize policy areas in relation to one
another including: wealth, inflation, unemployment, the size of the dependent population, female
labor force participation rates, and the openness of a nation’s economy.
Wagner’s Law suggests that as a nation’s wealth increases so too will its role in the
public sector. Changes and growth in the economy result in governments taking on new
functions beyond traditional roles of providing defense and public order. The expansion of the
role of the governments includes providing educational services and welfare assistance to
address new and growing issues within nations due to economic development and growth
(Peacock and Scott 2000). Changes in the economy are a result of shifts from agrarian to
manufacturing societies with industrialization, and the growth of the service sector. Greater
levels of wealth are then associated with increases in spending on particularized benefits such as
unemployment insurance, pensions, and daycare services to accommodate the needs of the public
due to changing economic conditions. Evidence suggests wealthier nations are associated with
governments that spend more on particularized benefits, typically in the form of welfare
spending (Huber and Stephens 1993, 2001, 2003; Crepaz, 1998; Milesi-Ferretti et al. 2002;
Bräuninger 2005; Iversen and Soskice 2006; Shelton 2007; Brook and Manza 2007).
As the wealth of a nation rises, government can accommodate the needs of multiple
subgroups, as it has more resources at its disposal to spend on particularized policy areas;
however, inflation is found to limit a government’s resource pool. The higher the level of the
inflation rate, the more money is required to obtain the same level of goods and services than
before. In such situations, when inflation is high, a government has less money to spend on

18

particularized policy areas and can please fewer subgroups within the population, resulting in a
decrease in spending on particularized policy areas relative to collective goods that benefits the
broader community. General support has been found for inflation limiting government spending
on particularized policy areas (Huber and Stephens 1993, 2000; Crepaz 1998; Chang 2008).
Government actions are also shaped by the social pressures that determine what demands
governments face, including unemployment and the size of the dependent population comprising
young children and the elderly. As the level of unemployment in a nation rises, the proportion of
individuals who are in need of assistance to maintain a minimum standard of living also
increases. The rise in unemployment then challenges governments to provide goods and services
in the form of particularized benefits like unemployment insurance. Studies examining the effect
of unemployment find that government spending patterns are influenced by the levels of
unemployment in the nation (Crepaz 1998; Huber and Stephens 2000, 2001; Bräuninger 2005;
Iversen and Soskice 2006; Shelton 2007).
Similar to the argument in place for unemployment, the size of the dependent population,
consisting of the elderly and the young, is found to shape government spending patterns. Larger
elderly populations are related to more people in a nation that are typically no longer working
and are in need of aid from the government. The increase in a subgroup of the population
requiring government assistances shifts government priorities towards particularized spending
such as old age pensions. Additionally, as the proportion of the elderly increases, so should their
influence over policy outputs that favor their group, including increasing pension benefits (Huber
and Stephens 2001). Indeed, increases to the proportion of the aged population are shown to
increase government spending on particularized policy areas such as spending on pensions

19

(Huber and Stephens 1993, 2001; Scartascini and Crain 2002; Milesi-Ferretti et al. 2002;
Bräuninger 2005; Hay 2006; Shelton 2007; Chang 2008).
Increases in the size of the youth population are also suggested to increase government
spending on particularized policy areas. As the number of children increase, household incomes
are less able to provide for basic needs. As families are less able to provide themselves with
basic goods, government services are necessary in the form of items like daycare so parents can
work, after school services, food, and housing assistance. Studies examining the influence of the
proportion of the youth population in a nation have found it to be associated with greater
spending on particularized policies (Huber and Stephens 2000, 2001; Chang 2008).
Prior work shows that female participation in the workforce increases the level of
spending in different policy areas including welfare services (Huber and Stephens 2000, 2001;
Iversen and Soskice 2006). However, there are different arguments put forward on why female
labor force participation increases spending on welfare policy issues. One argument is based on
the notion that as women enter the workforce they require assistance to replace their traditional
care-giving duties in the form of particularized spending on such services as day care (Huber and
Stephens 2000). Another theory revolves around the increase to the number of workers. As the
size of the workforce increases, so does the number of workers who are entitled to benefits. If
women do not enter the workforce, they would not have access to certain particularized benefits
such as unemployment compensation (Iversen and Soskice 2006). Regardless of the different
theories, female participation in the workforce is shown to affect government spending patterns
(Huber and Stephens 2000, 2001; Iversen and Soskice 2006).
Risks posed by external factors through economic openness from trade and globalization
are suggested to shape government spending patterns and behavior. There are two opposing

20

arguments for the effect of economic openness in terms of trade and globalization on government
spending patterns. The first argument has openness resulting in governments that spend more on
social protection (Shelton 2007). As nations’ economies become more open, their domestic
economies are at increased risk for shocks posed by external factors and external economic
crises. In order to compensate for the increased risks that are posed by opening domestic
economies, governments increase the social safety nets in place and spend more on particularized
policy areas (Cameron 1978; Rodrik 1998; Shelton 2007; Swank 2010).
An alternative argument revolves around the idea of a “race to the bottom.” In this
context governments are unable to “sustain generous systems of public social protection” (Swank
2010, 319) because manufacturers rush to produce in the least costly nation. As a result, nations
that have generous welfare systems and strict laws in place, find businesses shifting production
to less costly nations and domestic unemployment rising. Therefore, governments make cuts to
welfare provisions to maintain competitive environments for producers. While more open
economies are suggested to influence government activity, the findings are far from conclusive.
Both arguments have found support, while other works produce null findings (Swank 2010;
Shelton 2007; Huber and Stephens 1993, 2000, 2001; Castles 2006; Scharpf 2000).
Another consideration regarding the openness of a nation’s economy involves
membership in the European Union. Countries that belong to the European Union have made
efforts toward economic integration among fellow member nations with the goal to prevent
future conflict (Europa.eu). This interconnectedness relating to the economies and the mix of
binding/ non-binding policies should influence the spending priorities of European Union
members. Connections in terms of policy outputs can be expected where supra-national
agreements in 2009 total roughly “eighty binding norms…in the three main fields of European

21

Union social policy regulation: health and safety, other working conditions, and equality at the
workplace and beyond…approximately ninety amendments [and]…approximately 120 nonbinding policy outputs” (Falkner 2010, p 293).
As the economies of the member nations of the European Union are connected through
policy agreements and a shared currency, a crisis in one economy can and does affect all
members. Consider the case of Greece in terms of its inability to finance its own government
operations and pay back loans. In order to continue operations, the European Union and the
International Monetary Fund have provided a series of bailouts to Greece in order to prevent the
Euro and the European Market from collapsing. Similar threats to European Union members’
economic security have come from Portugal, Ireland, and Italy. Economic consequences for
other nations that have dedicated money to help alleviate the debt of the aforementioned nations
include new austerity measures such as reduced salaries, higher taxes, and fewer employment
benefits (Associate Press: Austerity in Europe).
POLITICAL PREFERENCES AS INFLUENCES
Moving beyond the consequences of the socio-economic influences within nations, mass
and elite preferences are shown to alter a government’s policy priorities. Preferences linked to
spending priorities include the parties that comprise the government (Garand 1985; Hofferbert
and Budge 1992; Huber and Stephens 1993, 2000; Klingemann et al. 1994; Bräuninger 2005;
Breunig 2006), citizen mobilization (Baumgartner and Jones 1993; Huber and Stephens 1993,
2000, 2001; Hill and Hinton-Anderson 1995; Ringquist and Garand 1999; Lijphart 1997;
Jackman 1987), the general culture of expectations for the role of government (Almond and
Verba 1965; Inglehart 1990; Inglehart and Abramson 1995; Goren 2004; Norris 2004), public
opinion (Page and Shapiro 1983; Kingdon 1984; Erikson, Wright and McIver 1989;

22

Baumgartner and Jones 1993; Raimondo 1996; Ringquist and Garand 1999; Jacoby and
Schneider 2001, 2004; Soroka and Wlezien 2004, 2005), and interest groups (Schattschneider
1975; McConnell 1970; Lehmbruch and Schmitter 1982; Wilson 1982; Kingdon 1984;
Baumgartner and Jones 1993; Gray and Lowery 1996; Ringquist and Garand 1999; Jacoby and
Schneider 2001; Schneider and Jacoby 2006).
The composition of the government denotes what political parties hold office in
government. The parties that hold office are able to transfer their policy preferences into
governmental spending priorities. Research shows that political parties follow through on
implementing their party platforms as public policy once in office (Klingemann et al. 1994;
Hofferbert and Budge 1992). Political parties on the left tend to emphasize social services while
rightist parties have been found to emphasize areas of defense and order (Klingemann et al.
1994). Governments dominated by leftist parties are shown to spend more on particularized
benefits such as pensions and housing benefits for low-income groups, while those controlled by
rightist parties are shown to spend less on particularized benefits relative to collective goods
(Huber and Stephens 1993, 2000, 2001; Budge and Keman 1990).
The argument behind the effect of political parties finds support in both power resource
theory and partisan theory. Power resource theory argues that as the laborers organize and gain
strength more leftist party members will be elected to government (Korpi 1983). As leftist party
strength in government increases, there should be an increase in spending on policy areas
promoted by leftist parties including spending on particularized policy areas, such as retirement
benefits, unemployment insurance, and disability benefits. Partisan theory suggests that political
parties provide a number of policies to win elections and “implement policies favoring their core
constituencies” (Hibbs 1992). Leftist parties tend to be supported by the economically insecure

23

while rightist parties are supported by the more prosperous (Iversen and Soskice 2006). A
number of studies show that the divide between leftist and rightist parties in government shapes
spending patterns (Klingemann et al. 1994; Budge and Keman 1990; Huber and Stephens 2001;
Garand 1985; Hibbs 1992; Chang 2008; Korpi 1983, 1989).
Citizen mobilization is another potential factor that is suggested to influence government
priorities. Citizen mobilization can be measured in terms of voter turnout. In the end, voters are
responsible for electing officials to office, which, in turn, produces the makeup of government
parties in office. Citizen income within democratic nations tends to be right skewed with the
average worker earning less than the median income (Austen-Smith 2000). As a larger portion
of the population becomes active in voting, representation of lower income citizens increases
(Austen-Smith 2000) and as the lower income bracket has a high preference for redistribution
there will be an increase in parties elected who emphasize spending on particularized policy
areas (Hill and Hinton-Anderson 1995).
An alternative argument for the role of voter turnout suggests that as turnout increases
government will spend more on collective goods. Lijphart (1997) notes non-turnout rates are
higher among the poor. Therefore, as turnout increases, it is more likely to be voters from higher
income brackets voting for parties that favor less redistribution found in particularized policy
spending (Iversen and Soskice 2006). Mixed findings for the role of voter turnout exist
throughout the literature, supporting increases in spending for particularized policies, increases
in spending on collective goods, or no effect on government spending (Iversen and Soskice
2006; Chhibber and Nooruddin 2004; Crepaz 1998; Huber and Stephens 1993, 2000, 2001). The
mixed results may be a product of examining nations that have compulsory systems. Lijphart
(1997) notes that compulsory voting alters the probability of voting based on income.

24

Preferences can also manifest through public expectations on the role of government.
The general culture of public expectations about the proper role of government represents the
underlying preferences of citizens within a nation, setting the acceptable boundaries of
governmental activity. Public expectations about government responsibilities are akin to
political culture and are based upon the argument that, “culture is a system of attitudes, values,
and knowledge that is widely shared within a society and transmitted from generation to
generation” (Inglehart 1990, 18). Who people are and what people do are contingent on the
culture that surrounds them and as a culture changes over time, so will the nature of people’s
preferences (Easton 1953). Inglehart (1990) argues that there are enduring differences across
cultures and that these differences are tied to political outcomes. Political culture involves “a set
of orientations towards a special set of social objects and processes” and “patterns of orientation
toward political objects among the members of the nation” (Almond and Verba 1965, 12-13).
As such, citizens’ expectations about governmental responsibilities represent a set of orientations
that should affect subsequent governmental policy activities.
Essentially, there is a general sentiment in a nation that orients the public in viewing
political issues encountered over time. For “cultures are theories; they organize experience. If
everything is possible without constraint, there is no need to choose and no way to think, because
no act interferes with any other. If nothing is possible, everything being constrained, there is
also no way to choose and no point in thinking” (Wildavsky 1998, 196). As the expectations for
government represent the public’s preferences for political involvement, they set the boundaries
of what should be possible for government to do and what is not. Beliefs about the role of
government should then also influence the allocation of government resources across policy
areas with higher allocations going towards issues the public feels the government should be

25

responsible for addressing. As an example, if the public wants the role of government limited to
economic stability and defense, there should be greater government attention to these policies as
opposed to other areas where the public feels government should not be involved in, such as
foreign aid. If cultural norms influence policy preferences in a democracy, then these
preferences should be transferred into government priorities.
Similarly, public opinion represents how citizens feel the government should handle
specific societal problems, as well as the priority that government gives to various governmental
actions to address these problems (Baumgartner and Jones 1993). Public opinion plays an
important role as it can have both positive and negative consequences for policy outputs: “(I)t
might thrust some items onto the governmental agenda because the vast number of people
interested in the issue would make it popular for vote-seeking politicians” (Kingdon 1984, 65);
alternatively, it could also prevent some issues from ever getting on the governmental agenda for
policy action. In democracies, politicians are rewarded for responding to public preferences; if
public opinion goes unheeded during a politician’s time in office it is likely the incumbent will
be voted out during the next election. Because of politicians’ need for public support to stay in
office, public opinion should affect government priorities. A number of scholars have found that
public opinion in different manifestations influence government spending. Jacoby and Schneider
(2001) show that public opinion measured through citizen ideology and partisanship affects
government spending priorities with more liberal public opinion increasing government spending
on particularized benefits in the United States. Other research has shown that governments do
respond, in more general terms, to public opinion in the form of policy outputs (Soroka and
Wlezien 2004, 2005; Page and Shapiro 1983; Hill and Hinton-Anderson 1995; Penner, Blidook,
and Soroka 2006).

26

The culture of expectations for governmental action and public opinion are two separate
components regarding policy preferences. Knowing or identifying the expectations for
government in a society therefore, does not guarantee the ability to predict public opinion on
issues. For example, while expectations may set the boundaries of government involvement to
include economic stability, public opinion may favor handling unemployment with greater
unemployment assistance, a particularized benefit, or alternatively through economic stimulus
packages and tax cuts, which are closer to collective goods. Alternatively, expectations may
favor the promotion of equality and an egalitarian society. To get there, public opinion may
prefer equality in outcomes, with greater spending on social protection like housing and food
benefits (particularized benefits), or equality of opportunity, with greater spending on economic
development to promote increased employment a collective good.
Prior research suggests that public opinion may also be influenced by governmental
policy outputs, such that; citizens are responding to what governments do, instead of
governments responding to what the public wants. Page and Shapiro (1983) find that public
opinion typically moves before policy change. In situations where change occurs first, Page and
Shapiro (1983) suggest that public opinion may still be the driving force and that policy makers
are acting on anticipation of changes in public opinion. Soroka and Wlezien (2004) find a
feedback loop in public opinion and policy change in Canada, where changes in spending result
in changes in public preferences for government spending. However, after the public reevaluates its preferences, governments are found to respond to the new preferences for
government actions, such that when citizens want less spending on a policy area governments are
found to decrease spending, after which governments respond to the new evaluations of the

27

spending level by reducing spending further if cuts were not enough or increasing spending if
cuts went too far.
Members of the public may also collectively express their preferences to the government
in the form of interest groups. Interest groups/organizations represent, “highly detailed, nuanced
signals about the specific problems citizens face, as well as potential solutions” (Gray and
4

Lowery 1999, 242). As opposed to beliefs on the role of government or public opinion that
pervade society, interest groups represent particular sets of individuals in society and typically
target specific policy areas and push for their preferred policy and influence government actions
(McConnell 1970; Ringquist 1999; Crepaz 1998; Gray and Lowery 1999). Interest groups have
also been noted to play a strong role in shaping policy in corporatist nations where interest
groups replace, “representation based on geographic units or units of approximately equal
number of voters” (Wilson 1982, 219-220; Lehmbruch and Schmitter 1982).
Although interest groups do represent particular sets of preferences within a nation, more
interest groups do not always correspond with increases in particularized policy spending.
Instead, increases in the number of interest groups, measured as the number of state government
employees in the American states, are associated with greater spending on collective goods
(Jacoby and Schneider 2001). When there are fewer interest groups pushing for particularized
interests, it is easier for governments to accommodate their demands. However, as the number
of interest groups increases it becomes difficult to appease all interests at the same time. Instead
governments are in a better position to move forward on collective goods that benefit many
groups within the population. As Schattschneider (1975) noted, “If there are twenty thousand
4

Gray and Lowery prefer the use of the term interest organizations in reference to the
composition of units whose interests are being represented as opposed to interest groups, which
are restricted to membership groups.
28

pressure groups and two parties, who has the favorable bargaining position? In the face of this
ratio it is unlikely that the pressure groups will be able to play off the parties against each other”
(56).
For example, “there is no one organization that can speak for employers in the United
States,” making it difficult to best serve the interest of employers in policy as there is no
agreement on what employers want (Wilson 1982, 222). Additional work has shown that as
more preferences are introduced into the decision making process by using proportional
representation, the role of interest groups in shaping government outputs is diminished (Crepaz
1998). A summary report by Kenworthy (2003) on measures of interest groups shows that
different measures of corporatism have yielded both increases and decreases in the level of
redistribution in a nation, thereby affecting spending on particularized policy areas.
The importance of interest groups in shaping government actions can be seen when
looking at the process of health care reform under both Presidents Clinton and Obama. Under
Clinton, interest groups were able to turn public opinion against health care reform by using their
resources to distribute information suggesting that health care reform would limit/restrict the
rights of individuals to choose their own medical care. Ultimately, interest groups were able to
mobilize public opinion against health care reform. Learning from the reform attempts of 199394, Obama attempted to co-opt strong interest groups because “major economic-interests groups
with profits at stake would be much more vigilant, motivated, and organized than the diffuse
public” (Jacobs and Skocpol 2010, 69).
Unification of purpose allows organized interest groups to influence government more
effectively, with tangible consequences for policy outputs. Another example of organized
interests shaping policy can be found in Sweden during the 1990s. As the Swedish government

29

attempted to reduce spending on social protection, particularly pensions and unemployment
insurance, organized groups representing the interests of companies and union workers guided
the reform efforts (Anderson 2001). In this case, without a unified voice, it would be difficult for
employers and employees to shape policy in a manner that benefits their respective
memberships.
INSTITUTIONS AS INFLUENCES
As researchers seek to understand the causes of different policy outputs across
governments, another line of research examines institutional differences between nations. Here,
the argument focuses on how certain institutions create different incentives for politicians and
voters which, in turn, leads to varying policy outputs (Austen-Smith 2000). Influential political
institutions include majoritarian versus proportional representation (Persson and Tabellini 1999;
Ringquist and Garand 1999; Austen-Smith 2000; Tabellini 2000; Cox and McCubbins 2001;
Milesi-Ferretti et al. 2002; Iversen and Soskice 2006; Shelton 2007; Persson et al.. 2007),
presidential versus parliamentary systems (Huber and Stephens 1993; Lijphart 1999; Persson and
Tabellini 1999; Tabellini 2000; Tsebelis 2000; Haggard and McCubbins 2001; Shugart and
Haggard 2001; Scartascini and Crain 2002; Lienert 2005; Edwards and Thames 2006; Iversen
and Soskice 2006), district magnitude (Huber and Stephens 1993, 2000; Hill and Anderson 1995;
Ringquist and Garand 1999; Cox and McCubbins 2001; Milesi-Ferretti et al. 2002; Edwards and
Thames 2006; Persson et al.. 2007), bicameralism versus unicameral legislatures (Huber and
Stephens 1993, 2000; Immergut 2010), and federalism versus unitary governmental structures
(Obinger, Leibfried, and Castles 2005; Bednar 2009; and Immergut 2010). Each institutional
attribute affects how governments allocate their resources toward either particularized benefits or
collective goods.

30

The electoral systems that are used to determine which candidate wins office are argued
and have been found to shape government spending. The overwhelming finding is that
proportional representation systems favor spending on particularized policy areas and
majoritarian systems spend more on collective goods (Scartascini and Crain 2002; Chhibber and
Nooruddin 2004; Chang 2008). One of the general arguments focuses on the campaign strategy
adopted by candidates to win elections. In a majoritarian system, the candidate with the most
votes wins, so it is the candidate’s goal to please a majority of voters (Tabellini 2000). In a
given district voters generally benefit from the same form of collective goods, but have different
preferences for particularized benefits. Therefore, it is in the interest of the candidate to promote
collective goods to gain a chance at the majority vote share. Therefore, candidates aim policy
platforms at more collective goods; like community or economic development, to avoid
alienating a subgroup of voters that could prevent them from winning the election.

5

Under proportional representation systems, candidates have different incentives than in
majoritarian systems. In a proportional representation system, a candidate can win with just a
few votes as seats go to more than just the single candidate with the most votes (Tabellini
6

2000).

Proportional representation creates districts where voters have similar preferences for

collective goods but different preferences for particularized benefits. Proportional representation
produces candidates then that promote more particularized benefits like pensions or family and

5

Similar to what Norris (2004) refers to as a bridging strategy. Bridging strategies involve
bringing together individuals with heterogeneous interests to form a broad coalition.
6
Unless the district magnitude is equal to one, in which case it is essentially a majoritarian
system.
31

children benefits, to target subgroups.

7

Using proportional representation, if a candidate

promotes collective goods, voters opt for the candidate whose platform caters to their groups’
particular preferences.
In addition to changing the motives from the candidate’s, voters also have different
preferences for parties under the majoritarian/proportional representation divide (Lijphart 1999;
Iversen and Soskice 2006). Assuming there are three classes within a society, Iversen and
Soskice (2006) argue that the middle class has different motives for aligning with either the
upper or lower income brackets based on the electoral structure. In a system using proportional
representation, all three groups will have representation in office. In this situation, the middle
class will tend to form coalitions with the lower income bracket to tax the upper class and
redistribute the benefits. Here, if the lower income bracket attempts to tax the middle class as
well as the upper class, the middle class can leave the coalition and join with the upper class to
prevent taxation. Proportional representation then creates a government favoring particularized
benefits that provide redistributive goods.
Under a majoritarian system, typically only two parties hold office and the middle class is
faced with joining either the upper or lower income brackets’ parties. Unlike proportional
representation systems, once elected, the middle class cannot prevent the lower income group
from taxing the middle class through defections. In a majoritarian system, if the lower income
bracket’s party wins, the party controls government without requiring a coalition for support, and
can tax the middle class and the upper class and redistribute more benefits solely to the lower
income bracket. Under a majoritarian system, out of fear of taxation, the middle class will
support the upper class parties and at least avoid taxation at the loss of some redistributive
7

Similar to what Norris (2004) referred to as a bonding strategy. Bonding implies bringing
together individuals with homogenous preferences on certain issues (Norris 2004).
32

benefits. Lijphart (1999) finds similar evidence that proportional representation systems support
more leftist parties that favor more particularized benefits and majoritarian systems support more
rightist parties that favor more collective goods.
Presidential and parliamentary systems have a similar divide in outputs to that of
majoritarian and proportional representation. Under a presidential system, the executive is
elected at large by the nation and holds power independently of the nation’s legislature, and
while the executive cannot dissolve the legislature, the executive typically possesses some form
of veto power over legislation (Shugart and Haggard 2001). Under a parliamentary system, the
executive is appointed by the majority party or governing coalition in the legislature; both the
executive and the legislature have the power to dissolve government, and typically legislation
passes with a majority vote (Shugart and Haggard 2001).
These two systems are found to produce governments with different public policy
patterns. The underlying argument to these differences is based on the constituencies to which
the executives in the different contexts are accountable. Under a presidential system, the
executive is accountable to the public at large and must seek out policy areas that benefit the
largest number of voters. In this context, policies that are voted on by the legislature and favor
more particularized groups can be vetoed in some manner by the executive.
An executive in a parliamentary system is accountable to the legislature that he/she is
appointed by and typically represents more particularized interests depending on its party
composition. In a parliamentary system it is expected that more particularized policies favoring
the parties in office would dominate over collective goods policy areas. Prior works shows that
presidential systems do spend more on collective goods and parliamentary nations spend more

33

on particularized benefits, including spending on welfare issues (Lijphart 1999; Scartascini and
Crain 2002; Edwards and Thames 2007; Crepaz 1998).
Crepaz’s (1998) work on political institutions and welfare expenditures highlights an
example of how presidential systems behave relative to parliamentary system. Crepaz argues
that parliamentary systems are in a better position to incorporate the needs of groups that benefit
from increased spending on social protection and unemployment insurance than presidential
systems. However, in presidential systems, the need to win the majority of votes to obtain office
produces spending that benefits broad communities in the form of collective goods instead of
particular groups across regions. Looking at welfare expenditures as a proportion of gross
domestic product, that represents spending on particular groups of individuals in a society,
Crepaz (1998) finds parliamentary systems spend more on welfare than presidential systems.
Examining district magnitude and its effect on policy is similar to the comparison of
majoritarian and proportional representation. As district magnitude increases the number of
votes required to win offices decreases. As a candidate becomes less tied to pleasing as many
people as possible the candidate can go after subgroups in the population that are large enough to
ensure a successful election campaign with promises of particularized policies. High district
magnitudes can also constrain the ability of governments to enact or pass legislation. As the
number of candidates than can be elected in a particular district increase, so do the number of
viable candidates for office and the political parties they represent (Cox 1997). A variety of
works find that higher district magnitudes increase government spending on particularized policy
areas (Edwards and Thames 2007; Milesi-Ferretti et al. 2002).
When Milesi-Ferretti et al. (2002) look at proportional representation versus majoritarian
systems, they argue that spending that targets particular groups is more prevalent in nations that

34

use proportional representation. While majoritarian and proportional representation determine
the rules for elections, district magnitude sets the level of proportionality. As district magnitude
increases, candidates need fewer votes to win and will use spending that favors set groups within
the society to increase their likelihood of winning office. Higher district magnitudes are found to
be associated with greater spending on particularized benefits, such as social security payments
and less spending on collective goods like building bridges.
INSTITUTIONAL CONSTRAINTS AND POLICY RESPONSIVENESS
In addition to examining the role of institutions independently from one another,
researchers look at how institutions work together to constrain policy outputs (Tsebelis 1995,
2000; Cox and McCubbins 2001; Iversen and Soskice 2006). These studies focus on how
institutions create veto players (Tsebelis 1995), the role of the institutions in the form of veto
points (Immergut 1990), and the ability of institutions to promote policy decisiveness or
8

resoluteness (Cox and McCubbins 2001). Institutional constraints are features of the
governmental/political system that increase the number of preferences present in the decision
making process by separating decision making power across multiple actors. Researchers
examine a variety of different institutional characteristics in their examination of constraints
including: presidential systems, bicameralism, federalism, and proportional representation
systems (Tsebelis 1995; Huber and Stephens 1993, 2000, 2003; Cox and McCubbins 2001;
Immergut 2010).
Constraints are found to affect the ease with which policy agreement can occur. As the
number of constraints increases, the ability for policy agreement to be reached decreases and
governments become more resolute in their policy actions (Cox and McCubbins 2001). As the
8

Decisiveness is defined as the ability to reach and enact policies whereas resoluteness is
defined as the level of commitment to the current status quo (Cox and McCubbins 2001).
35

number of constraints increases, the ability for incumbents to reach agreement on particularized
policy areas decreases due to a greater number of actors with preferences over governments
actions present in the decision making process.
While majoritarian and proportional representation can shape the spending patterns of a
government, proportional representation also serves to constrain the ability of governments to
reach policy agreements (Immergut 2010; Lijphart 1999). Proportional representation tends to
promote multiparty systems and simple majoritarian systems support two party systems (Riker
1982). As the number of political parties increases so do the number of preferences over policy
outputs. The increase in the number of demands constrains a government from acting quickly as
bartering and negotiations on policy must first occur to appease the multiple actors involved in
the decision making process. Not only does the increase in the number of preferences decrease
the ability of government to reach agreements, it also leads to greater compromises among the
various groups involved in policymaking. Greater compromise means no particular group is able
to obtain its ideal policy outputs. Thus, in a two party system, however, one party will have
greater control over policy and is therefore less constrained in its ability to act with fewer
opponents to appease. Under a majoritarian system, political parties should be able to achieve
policy outputs closer to their desired expectations.
Presidential systems, as opposed to parliamentary systems, can serve to constrain a
government’s ability to act (Immergut 2010; Bradley et al 2008; Tsebelis 2000; Huber and
Stephens 2000, 2001). As noted earlier, in a presidential system the executive is elected
separately from the legislature and tends to have different expectations for policy outputs.
Additionally, presidents typically possess some form of veto power over legislation before it can
be enacted. Under this context, presidential systems can constrain government action by vetoing

36

legislation that does not meet with the executive’s preference for policy outputs. In a
parliamentary system, however, the executive’s fate is tied to the legislature which can end
government with a vote of no confidence, reining in the executive’s ability to prevent legislation
from passing and from constraining government’s ability to act.
Similar to the argument for proportional representation, as the number of parties with
demands present in the decision making process increases, so does the difficultly of enacting
legislation that appeases enough voters in the decision making process. Therefore, as district
magnitude increases, more parties enter government due to the lower threshold to win office, and
governments are more constrained in their ability to act. Prior research supports the argument
that increases in district magnitude makes it more difficult for government to reach an
agreement on policy outputs as it diffuses decision making power across more actors with
different preferences (Tsebelis 2000; Immergut 2010)
In unicameral systems, only one house is responsible for writing and passing legislations.
However, in a bicameral legislature, the ability to write legislation is divided between two houses
that both need to agree for a policy to be passed. The two houses then increase the difficulty of
reaching agreement. For example, each house may be able to reach compromise within itself on
legislation, and yet unable to reach a compromise with the other house. Therefore, bicameral
legislatures are found to constrain a government’s ability act compared to unicameral legislatures
(Cox and McCubbins 2001; Shugart and Haggard 2001; Huber and Stephens 2000, 2001;
Immergut 2010).
With a federal system, decision making power is dispersed across different levels of
government and is found to decrease a government’s ability to reach policy agreements (Huber
and Stephens 2000; Lijphart 1999; Cox and McCubbins 2001; Shugart and Haggard 2001;

37

Immergut 2010). This dispersion of power is shown to constrain governments from being able to
reach agreements over policy (Obinger, Leibfried, and Castles 2005; Immergut 2010). Examples
of how federal systems are constrained compared to unitary systems can be seen through looking
at the “old” and “new” politics of the welfare state (Obinger, Leibfried, and Castles 2005).
When welfare states were on the rise, nations with federal systems were unable to rapidly shift
spending towards particularized benefits like social protection. However, in the “new” politics
of the welfare state, federal systems are unable to cut funding to welfare policy issues compared
to unitary systems and have higher levels of spending on areas of social protection (Obinger,
Leibfried, and Castles 2005). Additionally, the dispersion of power can allow lower levels of
government to pre-empt action at the higher levels on policy areas, making it more difficult for
the national governments to create legislation as it infringes on the rights of the lower levels of
government (Obinger, Leibfried, and Castles 2005).
The argument I test in Chapter 6 is based on the institutional constraints in democratic
nations. As the number of constraints increase, the number of actors/groups with preferences
present in the decision making process increases, and so does the difficulty of reaching
agreement on policy outputs. In order to reach an agreement on policy, as the number of
constraints increase, compromises and bartering among the actors will have to occur. The
process of bartering will increase spending on collective goods that provide benefits to all
members involved in the decision making process. There should be a decrease in spending on
particularized policies as the number of actors present in the decision making process increases
because particularized spending will only benefit some members at the expense of others who
have the ability to block legislation. Through this process, no particular group is in a position to
implement its ideal policy from government. Additionally, as the number of constraints

38

increases, the ability of any preference to drive, or influence, policy outputs will decrease,
making government less responsive to actors in terms of matching policy outputs to preferences,
where responsiveness is defined as, “the degree to which policy choices follow public
preferences” (Roberts and Kim 2011, 819).
CONCLUSION
A number of approaches are used to explain government outputs including examining
specific acts of legislation and changes to particular policy areas; however, expenditure data are
suggested as the primary way to examine governmental commitments as it shows the “tangible
distribution of public resources and not merely the intention of politicians and office holders”
(Jacoby and Schneider 2001, 546). Scholars have shown that the socio-economic climate, the
preferences of different groups, and political institutions do influence the policy outputs of
governments. However, most of the factors that are studied in relation to government priorities
have taken place in separate analyses, looking only at the role of political institutions or the
affect of socio-economic conditions without also examining the institutional make-up of
governmental systems. The omission of critical explanatory factors needs to be addressed in
order to properly understand the relationship between the explanatory variables and government
spending. The priorities model in Chapter 5 presents a more fully specified model of
government spending. The fully specified model allows me to explain why factors like citizen
mobilization has mixed findings in prior models and why the elderly are the only dependent
population found to influence government spending.
In addition to looking at the individual factors discussed in this chapter, I will also
examine the relationship between political institutions in nations and citizen preferences.
Although prior research indicates that the combined institutional factors can make political

39

systems more resolute in terms of enacting policies different from the status quo, the interaction
between institutional design and citizens preferences has not been examined. In this research, I
show how institutional factors alter the role mass and elite preferences play in shaping policy
outputs in democratic governments.

40

CHAPTER 3
GOVERNMENT SPENDING PRIORITIES
Prior studies on government spending priorities tend to divide policies into two
categories: policy choices that favor subsets or particular group of the population referred to here
as particularized benefits and policy choices that favor the majority of citizens or the general
population within a nation that are referred to as collective goods (Persson and Tabellini 2003;
Hofferbert 1974; Klingemann et al. 1994; Huber and Stephens 2001; Iversen and Soskice 2006;
and Milesi-Ferretti et al. 2002; Penner, Blidook, and Soroka 2006; Jacoby and Schneider 2001;
2009). However, a number of studies do not show that governments actually make policy
decisions along a single dimension of choice or that spending is based on the groups of citizens
intended to be affected by the policies. Instead, the two types of policies are examined in
separate models using total spending on particular sets of policies relative to gross domestic
product (Huber and Stephens 1993, 2001; Chhibber and Nooruddin 2004), using the proportion
of total spending (Garand 1985; Garand and Hendrick 1991; Hofferbert and Budge 1992;),
change in spending (Jones et al. 2009; Soroka and Wlezein 2004) or using a composite measure
(Hofferbert 1974; Klingman and Lammers1984; Erikson Wright and McIver 1989).
In this research, I use an unfolding model to operationalize government spending
priorities. The unfolding model shows that the spending patterns of democratic government
across a number of policy areas can be represented using a single dimension. As governments
choose policy packages along a single dimension, the unfolding model depicts the variation in
program expenditures found in the original data by policy area. Further, I provide support to the
literature on the pattern of spending regarding the dichotomy between particularized benefits and
collective goods at the state level in the United States. The results lend support to my first
hypothesis:
41

H1: Government expenditures can be captured by a single policy dimension that
represents two types of policies: those that benefit particular groups (particularized
benefits) and expenditure that benefit society more generally (collective goods).
The spending priorities variable can then be used to properly test the effect of factors that
have been argued to influence the variation in spending priorities across and within nations and
over time. Importantly, the unfolding produces a more ideal measure of spending priorities than
alternative approaches using single indicators, typologies, and composite measures.
DATA SELECTION
In order to create a measure of spending priorities I use government expenditures across a
wide range of policy areas. Government expenditures represent spending by the general
government and include national, state, and local government expenditures where information is
relevant. Although it has been noted that expenditures do not cover all policy decisions, such as
regulations (Hofferbert 1974), it has also been pointed out that the majority of policy debates and
decisions focus on the distribution of funds (Hofferbert and Budge 1992). Government
expenditures also reflect a degree of commitment by the government to a policy (Schneider and
Jacoby 2006). Expenditures serve as a central component of what governments’ do, where
agencies and policies grow and contract based on how much money they are allotted (Kingdon
1984). As such, the variation in shares of spending over time provides a measure for tracking the
rise and fall of policy areas in importance on government agendas (Hofferbert and Budge 1992).
Government expenditures by policy areas act as tangible representations of what
governments actually do, versus the promises of what they will do or of what they would like to
do (Jacoby and Schneider 2001). The reality is that governments cannot freely increase spending
across all policy areas without increasing deficits. As such, governments must make

42

compromises on how to allocate the resources they have across the issues they face. If
governments do increase spending on all policy areas without making trade-offs, they run the
risk of economic collapse. For example, Greece’s choices in spending on social protection,
particularly its high levels of spending on pensions with eligibility at the age of 57, helped to
push the government into bankruptcy and forced it to seek financial assistance from other
members of the European Union (OECD-Pensions at a Glance). Limitations on spending force
governments to trade spending more or less on one policy area in relation to the other policy
options, reflecting the spending priorities of a government (Ringquist and Garand 1999).
Expenditures by policy area are measured as a percentage of the total spending across a
range of program areas. I use the percentage of spending because my interest is in the relative
rank ordering of expenditure allocations, and not specific spending levels. The use of specific
program expenditures would not control for the total size of the government across time or
across countries and would prevent comparisons between nations in the form of a pooled
analysis (Huber and Stephens 2001). The percentages are also based on total expenditures
because I am interested in the percentage of the money that is spent on policies and how it was
spent to different areas instead of the amount spent relative to what could have been spent. Other
works have examined spending as a percentage of gross domestic product or gross national
product, implying an examination of the generosity of government with the denominator
9

representing total resources at a government’s disposal. Additionally, changes in spending
measured this way can be misleading if the level of spending does not change but the economy
expands or contracts; it can appear as if the government is spending more or less on goods and
services when nothing has actually changed regarding spending patterns.
9

If an unfolding were performed based on the expenditures as a proportion of GNP or GDP a
similar policy dimension would be produced.
43

Table 3.1

Democratic Nations and Time Periods
Country
Austria
Belgium
Canada
Czech
Denmark
Finland
France
Germany
Greece
Hungary
Iceland
Ireland
Italy
Japan
Korea
Luxembourg
Netherlands
Norway
Poland
Slovakia
Slovenia
Spain
Sweden
UK
US

Time-Span
1995-2009
1990-2008
1990-2006
1995-2008
1990-2009
1990-2008
1995-2008
1991-2008
2000-2008
1995-2008
1998-2007
1990-2008
1990-2008
1996-2007
2000-2008
1990-2009
1995-2009
1990-2008
2002-2008
1995-2007
1999-2008
1995-2008
1995-2008
1990-2008
1990-2008

The expenditure data by policy area are available from the Organization for Economic
Co-Operation and Development (OECD) starting as early as 1990 for some nations and running
up through 2008 for most nations with 2009 data available for a several others as of December
2010 (Table 3.1). The general government accounts are based on total government spending by
all levels of government and are divided into ten different expenditure areas by function:
government operations, defense, public order and safety, economic development, environmental

44

protection, health, community development, recreation, education, and social protection. An
overview of spending by policy area can be found in Table 3.2.
Table 3.2

Examples of Expenditures by Policy Area

Activities by Policy Area
Government Operations
Education
Executive and legislative organs, financial
Pre-primary and primary education
and fiscal affairs, external affairs
Secondary education
Foreign economic aid
Post-secondary non-tertiary education
General services
Tertiary education
Basic research
Education not definable by level
Public debt transactions
Subsidiary services to education
Defense
Health
Medical products, appliances and
Military defense
Equipment
Civil defense
Outpatient services
Foreign military aid
Hospital services
R&D Defense
Public health services
Public order and safety
Recreation, culture and religion
Police services
Recreational and sporting services
Fire-protection services
Cultural services
Law courts
Broadcasting and publishing services
Prisons
Religious and other community services
Economic Development
Social protection
Economic, commercial and labor affairs
Sickness and disability
Agriculture, forestry, fishing and hunting
Old age
Fuel and energy
Survivors
Mining, manufacturing and construction
Family and children
Transport
Unemployment
Communication
Housing
Environmental Protection
Community Development
Waste management
Housing development
Wastewater management
Community development
Pollution abatement
Water supply
Protection of biodiversity and landscape
Street lighting

45

Figure 3.1

Distribution for the Proportion of Spending on Health

Note: The histogram is based on the percentage of spending out of total spending across
the ten policy areas for all 379 nation years in the dataset.
A brief examination of a few of the policy areas serves to highlight the variation in
expenditures. Figure 3.1 shows the proportion of total spending on health across the 25 OECD
nations from 1990-2009. The unimodal distribution shows most nations spending roughly 14%
of total expenditures on health care services. The right tail of the distribution is slightly longer
and is capturing the United States spending on health care, a nation that spends more on health

46

care than any other in the dataset and is its top ranked expenditure area for five of the years
examined.
Figure 3.2

Distribution for the Proportion of Spending on Social Protection

Note: The histogram is based on the percentage of spending out of total spending across
the ten policy areas for all 379 nation years in the dataset.
A different form of variation can be observed by looking at the proportion of spending on
social protection in Figure 3.2. Here there are two clusters of spending. The low end shows the
variation in spending for the nations that did not have social protection as the number one ranked
policy area in terms of expenditures; including Canada, the United States and South Korea. The
47

high end of distribution expresses the variation for nations that spent more on social protection
than the alternative policy areas. Both groups show differences in expenditures for nations that
did and did not rank social protection as the top expenditure area.
Figure 3.3

Distribution for the Proportion of Spending on Public Order and
Safety

Note: The histogram is based on the percentage of spending out of total spending across
the ten policy areas for all 379 nation years in the dataset.
The final example of variation in spending can be seen in the area of public order and
safety (Figure 3.3). Here the proportion of expenditures ranges from 1.5% to 6.5%. There is a

48

higher proportion of nations spending across the entire range of public order and safety, unlike
health (Figure 3.1) and social protection (Figure 3.2) that each had unimodal distributions of
spending.

10

The distribution is indicative of the nations spending relatively consistently on

public order and safety during the time period examined.
The expenditure data taken from the OECD’s Annual National Accounts statistics
classify government spending into the ten different policy areas listed above. The expenditures
include goods and services that take the form of cash benefits such as pensions or unemployment
and in-kind benefits, for example housing vouchers or education services.
Similar to the literature already discussed, the OECD attempts to break down the policy
areas as either providing collective goods or individual goods. The OECD defines a collective
good as “a good that benefits society as a whole” and an individual good as “a good that
primarily benefits individual citizens” (OECD 2009, 61). The OECD’s definition for collective
goods is a close match for the definition I use here; however, the individual goods definition is
slightly broader than the one for particularized benefits. According to the OECD, individual
goods include spending on health and education, while I label these policy areas as collective
goods. Although individuals are deriving direct benefits from spending on these policy areas,
through universal health care coverage and public education programs all individuals in a society
benefit from these goods and services, while particularized policies focus spending only on
specific sets of individual who are intended to receive benefits such as disability or survivor
benefits.

10

To see the distributions for the remaining policy areas (defense, economic development,
education, environmental protection, government operations, community development, and
recreation) see Appendix A.
49

What spending is represented by the policy areas? Government operations includes day
to day financing for operating the executive and legislative branches of government, such as the
office of the executive, town councils, political staffs, libraries and other reference services and
ad hoc commissions and committees. The funding of external affairs, such as office abroad and
cultural services extending beyond a nation’s borders and economic aid to developing countries,
are other examples of government operations. I would argue that the direct benefits of salaries
and government expenditures go to a particular group of individuals who are employed by
government or are not members of the society in terms of foreign aid and should be classified as
particularized benefits. A similar argument is made in the case of the American states by
Schneider and Jacoby (2006) who argue government salaries represent a distinct group.
Spending on defense covers a range of services to protect a nation from external threats.
Expenditures include funding the military operations for land, sea, air, and space defense.
Defense spending also includes aspects of civil defense and foreign military aid. Where defense
spending focuses on threats from outside a nation’s borders, public order and safety expenditures
address domestic issues. Police services, fire protection, law courts, and prisons make up the
expenditures for public order and safety.
Economic development deals with general spending concerning the economy, commerce,
and labor. This policy domain includes aspects like equal opportunity employment and labor
mobility. Expenditures also address the conservation of arable land and flood control issues for
agriculture, production, resources, and the distribution of fuel and energy. Transportation
services like road maintenance, waterways, railways, and air transit that can affect economic
activities are included in economic development. These are services that benefit the entire
community.

50

Environmental protection handles spending on issues related to creating a clean society.
Spending from waste management such as local street cleanings and public parks to the disposal
of physical, chemical and biological waste is covered. Additionally, spending on waste-water
management and pollution abatement to ensure clean air, soil and ground water are part of
environmental protection. The benefits here are intended to benefit society in broader terms
versus particular groups of individuals.
Community development focuses on spending for public and neighborhood
improvements. Expenditures include the promotion and monitoring of housing developments
and slum clearances to rebuild communities. Issues that relate to clean water supplies and
streetlights are also covered. The benefits of community development are intended to benefit
society in general terms versus particular types of individuals.
Spending on health includes a number of services: pharmaceuticals; medical appliances
and equipment; outpatient services like specialized medical care and dental; hospital services and
public health services like vaccines, disease detection and blood banks. While individuals may
be the direct recipients, society as a whole generally has access to these services making this
domain a collective good. Furthermore, many citizens within the nations examined here find
health care to be a right of citizenship, particularly in European nations where “they [Europeans]
see universal access to health care as a social right, a crucial element of a decent society” (Okma
2011). Note that all OECD nations examined here except for the United States had universal or
near-universal health care coverage as of 1990 (Gurría 2008).

11

As such, in these nations health

care is provided to all citizens with a “core” set of services covered (OECD Health at a Glance
11

The OECD notes that three nations do not have forms of universal health care coverage in
place during the time period under study from 1990-2009 include the United States, Turkey, and
Mexico.
51

2009). Although coverage reaches almost all of the citizens within these nations, forms of
funding vary and include a mix of both public and private funding and insurance markets where
citizens may still be responsible for co-payments, deductibles, and supplementary insurance
coverage (Docteur and Oxley 2003).
Recreation as a policy area includes spending on sporting services like sports facilities,
fields, parks, and campgrounds. Further, spending covers cultural services such as libraries,
zoos, aquariums, and museums. Additionally, broadcast and publishing services, including
regulations, are covered by recreational expenditures. The use and access of the goods and
services provided under recreation are for societies in general terms.
Spending on education covers all levels of schooling. Primary, secondary, and college
educational expenditures are included in this area. Scholarships and grants for educational
purposes as well as vocational training make up part of the total spending. Primary education is
compulsory in the democratic nations where everyone in society benefits from spending
(UNESCO, 2009). Further, many nations offer assistance for higher education, such as the
United States that awards Pell Grants for students attending college.
The last policy area is social protection and includes a number of expenditures that are
aimed at particular groups in the society. Expenditures focus on disability and sickness, old age
pensions, unemployment insurance, and survivor benefits for those who need assistance
providing for themselves. Additionally, spending on families and children through housing
benefits, food vouchers, orphanages, foster families, and nursing care for children are included
under social protection. The spending areas in social protection are aimed at particular groups of
individuals in society. The direct nature of the spending for specific subsets of the population is
why social protection is labeled as a particularized benefit.

52

Based on the spending categories for each policy area, I expect defense, public order and
safety, environmental protection, economic affairs, education, health, community development,
and recreation to form a cluster at one end of the policy continuum representing collective goods,
and government operations and social protection to form a cluster at the opposite end
representing particularized benefits.

12

UNFOLDING
The unfolding technique is a means of locating an underlying, latent dimension based on
rank order preferences. The unfolding analysis produces two sets of points, one set of points
representing the decision making actors, such as individuals, corporations, or as is the case here,
governments. The other set of points represents the outputs of the decisionmaking process. In
the unfolding analysis performed here, the outputs are the proportion of spending on particular
program areas. The unfolding analysis arranges the two sets of points along a continuum such
that the distances between an actor point and an output point correspond to the relative
preference by that actor for that particular output. Smaller distances represent a higher
preference for a certain output and larger distances represent a lower preference for that
particular policy output.
The unidimensional, metric, least-squares unfolding approach applied to the expenditure
data from 25 democratic governments provides more information than the rank order preferences
of spending on policy areas. The interval level nature of the expenditure data are maintained and
the distances between nation spending priorities and policy areas capture the percentage of

12

Based on the OECD coding, health, social protection, education and market subsidies that
comprise economic development are individual goods, while defense, public order and safety,
environment protection, government operations, community development, and recreation are
considered to be collective goods.
53

spending by a particular nation on each policy area. A discussion of how to obtain the original
spending by nations on policy areas is provided later in the chapter.
Details of the Unfolding Procedure
The premise of unfolding is based foremost on the assumption that there is an underlying
dimension of choice that exists and is consistent across actors. Although the unfolding process
itself produces a dimension of choice based solely on the data, it falls on the researcher to
determine what the substantive order and groupings of the output points represents. In this
context, I use nations that have been repeatedly classified as democratic over the time period
from 1990-2009 as the decision-making actor.
The unfolding process I use to determine the location of the nations’ spending priority
points and the policy area stimulus points is the metric, least-squares unfolding method used by
Jacoby and Schneider (2001, 2009) when examining government spending priorities across the
American states. The approach attempts to place the output points (representing policy areas)
and the actor points (representing a nation’s spending priorities) to minimize the squared errors
between the actual percentage of nation spending (xijt) for nation i, on policy j, at time t, and the
*

predicted percentage of spending based on the unfolding for nation i, on policy j, at time t (dijt ):
2

*

min ∑eijt =

(dijt - xijt)

(3.1)

The predicted level of spending based on the unfolding is found using the distance
between nation points (nit) for nation i, at time t, and the points for each policy area (pj).
However, distances between a nation’s spending priority point and a policy point are inversely
proportional to the amount of spending from the government that policy receives. The closer the
policy point is to the nation’s spending priority point, the more funds that nation spends on that
54

particular policy area. If for example the distance between France in 2008 and the point
representing social protection is zero then France would be predicted to spend 100% of its total
expenditures on social protection and 0% on the alternative policy areas in 2008. To convert the
distance into the corresponding percentage of spending on the policy area the distance between
nation points and policy points must be subtracted from the total proportion of spending:
*

dijt = 100 - |nit - pj|

(3.2)

In order to locate the policy points and the nation spending priority points that provide us
with the minimum squared errors the process uses partial derivatives to find conditional global
minima.

13

The process begins by holding the nation points fixed and placing all the policy points

to the left of the nation points. Next the first policy point has its error calculated when it is to the
left of the first nation point. The first policy point then proceeds through each interval between
nation points and has its error calculated. After the first policy point has moved through the
nation ideal points and had its error calculated, it is placed within the interval that produced the
smallest squared error. This process is then repeated for each of the remaining nine policy areas.
After the policy points have moved through the nation points, the new locations of the
policy points are fixed and the process begins moving each of the nation points. The first nation
point moves through the intervals created by the policy points with its squared error calculated in
each interval. After moving through the intervals, each nation point is placed in the interval that
minimizes the squared error of that nation point.

13

The approach uses conditional global minima to maximize the variance explained. It is a
minimum in that the squared error is the smallest, conditional on the fixed set of nation or policy
points at the time of the calculation. Once a fixed policy or nation point is moved, the points
locations do not represent the minimum squared error that is possible.
55

An iteration of this process involves moving all the policy points through the fixed nation
points and moving all the nation points through the fixed policy points. The entire process is
repeated until further iterations no longer increase the variance explained. Regardless of the
starting points for the policy and the nation points, the process is quick to converge, with each
additional iteration decreasing the amount of total squared error present (Poole 1984). The
process results in policy points that represent the location of the policy area stimuli and nation
points that represent the spending priorities for nations in given years.
WHY UNFOLDING?
Unlike the alternative measures discussed in Chapter 2, unfolding presents a number of
benefits that the other approaches cannot. The unfolding provides one variable that can capture
all of the policy areas. Further, the spending priorities variable combines information on all
available policy areas, measured in the same manner, and from the same time points to produce a
measure that can be represented in only one dimension. The unfolding also produces a variable
that shows change over time, across nations, and within policy areas.
Data Reduction
One of the benefits of this approach is its ability to compress a large amount of information into
a small number of values. The original dataset used here includes all available information from
1990-2009 and contains 3,790 data values. The original dataset requires 10 data values for each
of the 25 countries, for each year, to convey the spending information across the ten policy areas.
After performing the unfolding analysis, each nation’s spending by year can be represented by
one data value, reducing the number of required data values by 90% to 379 data values.
The data values that capture spending across the range of policy areas are referred to as
nation spending priorities. Individually each nation spending priority can indicate the relative

56

spending by a nation in a given year on the two main sets of spending priorities. The average
nation spending priority for the OECD data is 53.36 and can be used to provide insight into
general spending patterns across the nations examined. For example, a nation with values lower
than the mean, such as Belgium in 1990 which scores a 50.90, spends more on the particularized
policy areas like social protection than the average nation from 1990-2009. Whereas a nation
with values higher than the mean spending priority, such as Ireland in 1999, which has a
spending priority value of 54.29, spends more on collective goods such as education than the
average nation during this time period.
Original Data
Another benefit of the unfolding analysis is that the original data values that go into the
technique can be obtained after the analysis has been completed. The following formula takes
the output of the unfolding technique and recreates the original data values if so desired and
prevents a loss of data from the process.
c - xijt = |nit - pj| + eijt

(3.3)

Because the spending values by policy areas were in percentages c=100, which is the
total percentage to be distributed across the spending areas, xijt represents the original percentage
of spending by nation i, on policy j, at time t; pj is the location of policy j; n is the nation
spending priority point for nation i, at time t; and eijt is an error term for nation i, on policy j, at
time t. This formula can be used on the nation points and policy points from the unfolding
technique to reproduce the original data, unlike factor analysis, additive scales, or typologies
where the original data cannot be reconstructed from the assigned values alone after the
techniques have been applied.
57

An example of how this process works is as follows. Austria spent about 9.71% of its
total expenditures on economic development in 2005. Using the equation above and the
information from the unfolding analysis that follows, I can recreate the original percentage of
spending. The above equation can be rearranged such that:
xijt = c - |nit - pj| + eijt

14

(3.4)

Where c=100 and based on the unfolding analysis nAustria,2005=52.57 and peconomic =142.77
which gives us an xAustria,Economic,2005=9.80. The unfolding predicts the actual spending for
Austria on economic development in 2005 to be 9.80% of total expenditures, resulting in an error
of only 0.09. The average difference between the re-derived expenditures for nations by policy
and the actual spending is zero (Table 3.2).
Table 3.3

Average Error in Capturing Actual Spending with Unfolding
Policy Area
Government Operations
Defense
Order and Safety
Economic Development
Environmental Protection
Community Development
Health
Recreation
Education
Social Protection

Average
Error
0.00018
-0.00018
-0.00018
-0.00019
-0.00018
-0.00018
-0.00019
-0.00018
-0.00018
0.00018

Single Dimension
The unfolding analysis provides a single dimension to test theories across a set of
spending priorities. Prior research has tested theories on separate policy areas in isolation from
14

Equation 3.4 is the same as equation 3.2 with an error term.
58

each other. Using the original dataset here, this approach would require ten different models
where the ten separate variables would fail to captures the relationship between policy areas.
Alternative methods have attempted to combine policy areas that were hypothesized by the
researcher to go together creating a variety of dimensions including Hofferbert’s two dimensions
of education/welfare and highways/natural resources, and Erikson, Wright, and McIver’s
measure of policy liberalism, which fails to account for policy areas that do not have an obvious
partisan slant.
Another approach used to explain policy outputs from government actions is the use of
typologies. While typologies can provide a simplified approach to viewing complex topics, the
categories used vary based on the objective of the researcher, often focusing on aspects such as
accessibility by use, who pays, and public versus private goods. Applying a common typology
derived by Lowi (1964, 1972), the ten policy areas I study here would fall into two of Lowi’s
four categories: distributive and redistributive policies (Table 3.4). Distributive policies offer
benefits to a wider community while redistributive benefits provide services to groups with
particular needs.
Table 3.4

Policy Typology using Lowi’s Categories

Distributive Policy
Health
Community Development
Economic Development
Environmental Protection
Education
Defense
Public order and Safety
Recreation

Redistributive
Policy

Government Operations
Social Protection
59

Applying a typology to the data however, forces the researcher to place policies into
categories that are not a perfect fit. For example, when fitting the ten policy areas into Lowi’s
typology (Table 3.4) environmental protection is categorized as a distributive good as
expenditures are used to promote clean air, clean water, and proper disposal of toxic waste have
benefits that spread throughout a society. However, environmental protection will also involve a
number of regulations, which means it carries characteristics of other categories in Lowi’s
typology. Furthermore, models that use typologies to produce a categorical dependent variable
will only be able to examine how the probability of spending on a set of policies is affected as a
set, and cannot predict the percentage of spending for each area. Once assigned to a category,
the resulting data points will be unable to identify the spending that belongs to each policy area.
Table 3.5

Exploratory Factor Analysis of Policy Areas
Policy Area
Government Operations
Defense
Public Order and Safety
Economic Development
Environmental Protection
Community Development
Health
Recreation
Education
Social Protection

Factor
1
0.9483
0.9690
0.9605
0.9055
0.9670
0.9338
0.9920
0.9460
0.9933
0.9801

Factor
2
0.1298
-0.1541
0.2389
0.2544
0.0840
-0.2800
-0.0087
-0.2154
-0.0900
0.0494

Running an experimental factor analysis on the raw spending data produces a six factor solution
for the ten policy areas when unconstrained and a one factor solution when constrained (Table

60

15

3.5).

Once the factor variables are created the original spending data cannot be retrieved. The

one factor solution does not provide intuitive insight into the pattern of government spending,
and appears to represent a factor covering government expenditures. Additionally, when using a
factor as a dependent variable, the predictions in spending will be for the set of policies and the
actual levels of spending by policy cannot be determined.
No A Priori Assumptions
Instead of pre-selecting policy areas and a variety of indicators that I believe may
represent a predetermined dimension of choice, the unfolding analysis allows the data to speak
for themselves. The unfolding analysis uses data measuring all available policy areas in the
same manner instead of combining a range of indicators representing inputs and outputs of
government and presents a dimension without a pre-specified label. After the unfolding analysis
has been executed, substantive interpretations of the underlying dimension can be discussed
based on what the data reveal. Although the assignment of descriptive label for the underlying
dimension may be subjective, it is based on what the data shows after the method is applied and
cannot bias what the unfolding technique produces.
Reliability
2

The reliability of the unfolding process can be determined by calculating the R value
using the unfolded spending priorities, the policy points, and the original data (Jacoby and
Schneider 2009). This is because the original expenditures can be re-calculated with equation
3.3 and simplified using equation 3.2 such that:

15

As the spending data are in the currency for each nation, only those nations that belong to the
Euro zone are included in the exploratory factor analysis: Austria, Belgium, Finland, France,
Germany, Greece, Ireland, Italy, Luxembourg, The Netherlands, Slovakia, and Spain (OECD
2009).
61

*

xijt = dijt +eijt

(3.5)

The resulting equation represents an OLS equation where the intercept equals zero and
the slope equals 1.
*

xijt = 0.00 + 1.00dijt + eijt

(3.6)

2

The R of this OLS model shows the amount of variance from the original expenditure
dataset that is explained by the unfolding technique, where “reliability is defined as the
proportion of a measure’s variance that corresponds to variance in the phenomenon being
measured” (Hand 2004). If xijt represents the phenomenon being measured (actual spending by
2

policy area), then the R value shows what percentage of the variance in the phenomenon is
explained. The unfolding analysis explains about 91% of the variance for the ten policy areas.
This high degree of reliability demonstrates that a single dimension of spending can capture
government spending patterns. The reliability of the measure supports the first part of the
hypothesis that a single dimension of policy can be used to represent government expenditures in
a parsimonious, encompassing, and reliable manner.
RESULTS OF THE UNFOLDED EXPENDITURE DATA
Applying the unfolding technique to the expenditure data from 1990-2009, establishes the
location of the policy points (Figure 3.4). The unfolding results show a clear and simple
distinction between the policies located at the two ends of the continuum. The results of the
unfolding show that government spending can be expressed as a trade-off in spending on policies
that target particular groups versus spending that targets the community in more general terms.
Increases in relative spending on one set of these policy areas will result in governments that
spend less on the other set of policy areas.
62

Figure 3.4

Location of Unfolded Policy Points

Environmental Protection
Community Development
Recreation
Public Order and Safety
Defense
Economic Development
Education
Health
Social Protection
Government Operations
0

50

100

150

Policy Points

The resulting spending priority measure for each nation can also be taken to represent
government compromise, where the priority values represent the final spending package
approved by government across a range of policy areas. This measure can then be used to test
what factors result in government spending patterns and the resulting compromises of
government outputs. Further, as the spending priority variable represents government
compromises it could be used to test the resulting policy outcomes of government, and how
successful a policy was at achieving its desired goal.

63

The values assigned to the policy areas serve to provide an interpretable meaning to the
distance between themselves and the spending priority points. Additionally, the relative
distances of the policy points in relation to one another can also be interpreted. The data
produced a dimension that has two main clusters with one at the far left and one at the far right.
On the left end of the continuum are a set of policy areas that affect or are intended to benefit
particular subsets of the population. At the other end of the continuum are a group of policy
areas that are intended to benefit more general groupings of the population.
The data suggest spending patterns are based on the targeted population that is intended
to be affected by the policy area. At the far left are policy areas that benefit particular groups
within a society referred to as particularized policy areas. At the right are policy areas that are
geared towards the population of a society more generally and are referred to as collective
16

goods.

Starting at the left end of the policy dimension are government operations and social
protection. The citizens who directly benefit from the policy expenditures under government
operations represent a fairly limited subset of the population, including non-citizens in foreign
countries and government employees. Government operations include foreign aid for economic
issues in developing and transitioning countries of which the citizens of the lending government
receive no direct benefit. Additionally, salaries and expenditures for running different levels of
government administration are included in this category. Therefore, unless a citizen is employed
in a section of the government, policy expenditures on this area do not directly affect them.

16

The policy dimension found for the 25 democracies here is similar in nature to the latent
policy priority dimension found for the American states by Jacoby and Schneider, which they
labeled particularized benefits and collective goods (2001, 2009).
64

Next to government operations is social protection, which consists of expenditures
received by particular subsets of a nation’s population. Social protection is typically one of the
largest expenditure categories for the nations in the dataset across time and includes expenditures
for unemployment insurance, old-age pensions, family allowances, children services, disability
compensation, housing support, and survivor benefits. Importantly, not everyone in a nation
meets the criteria to receive benefits from social protection, particularly in terms of services that
are referred to as “welfare,” and as Bahle, Pfeifer and Wendt (2010) explain, “[it] does not
17

include the concept of an unconditional basic income for all” (348).

Examples of types of

individuals who do qualify for benefits under social protection include the elderly who have
worked a certain number of years, war veterans, and families that fall below minimum income
levels established by the government.
At the right end of the continuum is a grouping of policies that represent more collective
goods. These are policies that are accessible to broader groups of a nation’s population,
especially in terms of programs like health and education that all citizens expect to make use of
at some point in their lives (Dean 2006). Health expenditures include spending on equipment,
hospital services, and public health services such as vaccines, blood banks, and disease detection,
which benefit the majority of citizens. As noted earlier, all of the nations during the time period
examined this research except for the United States have universal or near universal health care
coverage which provides access to at least a minimum set of health care services across most (if
not all) of the population.
Education expenditures include spending on primary through secondary and college
education, as well as vocational training. Primary education expenditures include expenditures
17

Emphasis added.
65

on literacy programs for individuals who have not met primary school literacy standards. All
levels of education include scholarships and grant funding expenditures. All citizens have access
to public education through at least high school and many individuals have access to funding
through the government for higher education.
Economic development expenditures focus on a variety of issues facing a nation’s
economy including trade, the prevention of discrimination in the workforce, agricultural
protection, obtaining and developing fuel and energy sources, and the development/ maintenance
of infrastructures. Economic development affects national populations in more general and
broad terms. Travel, goods and services for purchase, power and fuel, and employment are all
tied to growing economies.
Defense expenditures include spending on military protection by land, sea, air, or space,
civil protection for civilian institutions, and foreign military aid such as peacekeeping forces.
Everyone in a nation is expected to benefit from defense programs, not just certain social or
political groups, geographic regions, or policy interests.
Public order and safety is the next policy point on the continuum and encompass a
number of public programs. These expenditures include the provision and maintenance of police
and fire protection services, law courts, and prisons. All the expenditures under public order and
safety are provided to citizens within a nation. If citizens from one city travel into another city,
they are still able to receive emergency assistance. A police department will respond even if a
citizen is not a resident of their respective jurisdiction.
Recreational expenditures include spending on cultural activities, sports and recreation
services, and other community programs. Cultural expenditures go to support libraries,
museums, zoos, concert events, art galleries, and historic sites. Sports and recreation

66

expenditures are used to maintain playing fields, courts, tracks, golf courses, pools, and parks.
Recreation expenditures are collective goods because the expenditures are targeted at
communities in broad term to citizens, and not at particular groups of individuals within the area.
Community development expenditures are used for housing developments, to provide
sewage and water supplies to communities, and to develop/maintain public transportation
systems. All individuals within a community benefit from such services, like housing
developments improving to the quality of living in neighborhoods and raising property values
and those traveling through benefit from better lighting and the water supply. Note that
expenditures providing short and long term housing solutions for individuals unable to meet a
minimum standard of living are included in social protection. Unlike the housing expenditures
covered by social protection, the housing expenditure category itself focus on community
improvement and is not directed at specific individuals.
Finally, at the far right end of the unfolded policy dimension is the policy area of
environmental protection that deals with the quality of the environment. Expenditures focus on
waste management including the disposal of nuclear material, wastewater management, pollution
abatement directed at air and climate protection, and the protection of biodiversity and
landscapes that includes the protection of endangered flora and fauna. Environmental protection
and maintenance is benefits all citizens in a nation who breathe the air, drink the water, and
interact with the environment around them. As such these expenditures are the most collective
goods in the unfolded spending priority dimension.
The location of the policy points within the dimension also indicates which policy areas
are ranked higher in spending than others. The policy areas that are located closer to the center
of the policy continuum, represent policy areas that typically receive more spending than the

67

policy areas that are father away from the center. The policy areas of health, education, and
economic affairs represent policies that receive slightly higher proportions of spending than the
other collective goods, such as defense and environmental protection, located at the right of the
policy area continuum.
One more subtle note can be observed by examining the collective goods grouping at the
right end of the continuum. Within the cluster of collective goods, there are two separate
groupings, one containing health, education, and economic development and the other comprised
of defense, public order and safety, community development, and environmental protection. As
discussed earlier, the OECD labels health, education, and portions of economic development as
individual goods, while I contend that these areas are collective goods. The grouping of health,
education, and economic development is intended to benefit the community in broader terms
even though the expenditures targeted at specific individuals such as students in school or
individuals with medical needs. The spending on defense, public order and safety, and
environmental protection are closer to being pure collective goods where benefits are dispersed
across the society and not directed as specific individuals.
Difference between Nations’ Spending Priorities and Policy Points
The general position of a government spending priority within the cluster of spending
priority points indicates which nations have a preference for particularized policy areas over
collective goods relative to one another (Figure 3.5). Nations that are located towards the left,
have lower spending priority scores and are closer to the particularized benefits policy cluster,
representing higher spending on particularized policy over collective goods compared to nations
with higher scores. This includes Greece, Italy, and Belgium which have lower than the average
priority scores for the nations and time period examined. In 1990, Belgium had the lowest

68

spending priority score at 50.9, indicating it spent the most on particularized benefits compared
to any other nation-year in the dataset.
Figure 3.5

Distribution of Spending Priorities within Nations over Time

Greece
Germany
Denmark
Italy
Belgium
Sweden
Finland
Austria
Netherlands
France
Poland
Luxembourg
Hungary
Slovenia
United Kingdom
Canada
Spain
Noway
Ireland
Slovakia
Japan
Czech Republic
United States
Iceland

50

52

54
56
Spending Priority

58

Moving towards the right, nations have higher spending priority scores, placing these
nations closer to the collective goods end representing greater expenditures on policy areas such
as defense, economic development, or education. This includes nations such as the United
States, Iceland, and South Korea, which have the highest priority scores for the nations and time
69

period examined. South Korea had the highest spending priority score at 59.2 in 2003,
representing the nation-year with the highest priority for spending on collective goods over
particularized benefits.
The interpretation of particularized benefits and collective goods assigned to this
unfolded dimension serves to support my hypothesis regarding how governments spend and
conforms to expectations found in prior research. Based on the number of citizens who are
affected by the policy areas and how the policy areas are arrayed it appears that governments do
select policy packages based on the grouping of citizens who are intended to receive direct
benefits by the policy expenditures.
Differences between Nations’ Spending Priorities
Figure 3.6 shows the variation in policy priorities from 1990 through 2009. The variation
within a given year indicates that there are factors unique to the nations in this study that cause
the democratic governments to differ from each other in spending priorities at the same point in
time. The centers of the boxes represent the mean spending priority point for that given year.
Over the course of the twenty years, variation in spending priorities occurs repeatedly and is
substantial in size. In 2009 there is a more concentrated degree of variation, which is a product
of the limited number of nations with data available at the time of the analysis. While the
average range of spending priorities in a year may seem small at 7% of total government
spending, this is roughly equal to a €73 billion shift in spending for a nation like Germany or a
£23 billion shift in spending for the United Kingdom.
Nations’ spending priority points can also be interpreted in relation to one another. The
difference between two nations’ spending priority points represents the percentage point
difference between those two nations for spending on the different clusters of policies. An

70

example of this can be seen by examining the spending priority scores for Germany and Iceland
in 2001. In 2001, Germany has a spending priority score of 51.85 and Iceland has a spending
priority score of 56.74. What do these two values imply? Individually, the spending priorities of
these two nations can be compared to the mean value of spending priorities, which is 53.36, it
can be seen that Iceland spends more on particularized policy areas such as social protection than
the average nation while Germany spends more on collective goods. In relation to each other,
the difference between the spending priorities reflects that Germany spent 4.89% more of its
total spending on particularized benefits such as pensions and unemployment than Iceland.
Conversely, Iceland spent 4.89% more on collective goods including items involving
environmental protection than Germany did in 2001. The 4.89% difference in 2001, is
equivalent to a shift in spending between these two sets of policy areas of 16 billion Icelandic
Krona or €49 billion in Germany.
The differences help reveal which nations spend more on certain types of policy areas
compared to others. Nations with spending priority scores that are closer together, where there
are smaller distances between points, have similar spending patterns than nations that are further
apart and have greater differences in terms of spending on policy areas. Looking at the
distribution of spending priorities by nation (Figure 3.6) it is clear that nations like Germany and
Italy spend more on particularized benefits than nations like Iceland and Japan. Additionally,
nations like Germany and Italy which have priority values closer to each other, will spend on
policy areas in a similar manner, where as the United States, which has much higher spending
priority scores compared to Germany, will exhibit a much different spending profile in terms of
spending on the same set of policy areas. As the spending priorities variable captures these

71

differences, it is possible to test what influences the variation in policy expenditures across
nations.
Figure 3.6

Distribution of Spending Priorities over Time

The differences between spending priorities can also be used to measure changes in
spending patterns over time within nations. Again looking at the distribution in spending
priorities (Figure 3.6), nations’ spending patterns are not constant from one year to the next and
this measure captures and retains these differences. In the same manner that different nations’
spending priorities can be compared, the differences within a nation, over time, can be examined.
72

For example, the United Kingdom had a spending priority of 52.95 in 1998 and 54.88 in 2008.
The difference between these two time points suggests a 1.93 percentage point or roughly £7.8
billion shift in spending from particularized benefits, such as children and family benefits, to
collective goods, like education and economic development.
Close Examination of Policy Priority Scores
In order to gain a better understanding of what the spending priority scores look like,
three nations representing the lower, middle, and upper end of the spending priority spectrum are
selected (Figure 3.7). Based on the arguments presented in Chapter 2, it is suggested that the
differences in spending patterns across nations are a result of political institutions. For example,
Austria represents a nation that has spending priority scores at the lower end of the policy
continuum indicating that it has higher expenditures on particularized benefits like
unemployment or housing vouchers than other nations. Austria is a country that has a
parliamentary system and uses proportional representation to elect officials for office. Austria
also possesses a system that has a relatively high district magnitude of about 20 seats per district.
The initial examination here supports the findings in the literature that nations with
parliamentary systems, proportional representation, and larger district magnitudes have higher
expenditures on particularized benefits.
Canada, however, is a nation that represents a government whose policy packages come
close to representing the mean of spending priorities of the democratic nations. Canada, like
Austria, uses a parliamentary system. But, unlike Austria, Canada has a majoritarian system for
electing officials to office and has a district magnitude size of one. Compared to Austria, Canada
has larger spending priority scores, indicating that Canada spends more on collective goods than

73

Figure 3.7

Distribution of Spending Priorities across Nations

74

does Austria. The greater spending collective goods in Canada may be a product of it using a
majoritarian system with a district magnitude of one for electing officials to office.
At the far right end of the spending priority spectrum is South Korea. South Korea
represents a nation that has relatively high expenditures on collective goods. South Korea has a
presidential system and uses a mixed form of voting to elect officials to office. Additionally,
South Korea has a district magnitude of about 8.6. Unlike Austria and Canada, South Korea’s
use of a presidential system may be influencing its focus on collective goods. At the same time,
South Korea has a relatively smaller district magnitude than Austria but one larger than Canada
and uses a mixed system for elections.
While political institutions may help to understand differences across nations, these
systems cannot explain the variations within nations as political institutions do not change
frequently over time. The variation occurring within nations may be a product of other
conditions with nations like inflation, unemployment, or government composition. The potential
effect of different combinations of political institutions cannot be fully worked out with the three
examples presented here and requires a more detailed analysis that follows in Chapter 5.
CONCLUSION
While single indicators, composite measures, and typologies have been used previously
as a means of capturing policy decisions, these measures have limitations that prevent a full
understanding of the outputs the measures are used to test. The single indicator approach risks
omitting policy areas that are related to the policy under examination. While composite
measures tend to combine either variables that represent different aspects of policy making or
different views on what the dimension of choice regarding policies is, before the technique is
applied. The spending priorities variable is a better measure of government outputs. The

75

variable is capable of tracking changes over-time and across nations. Further, it has a readily
interpretable meaning; lower scores represent nations that spend more on particularized benefits
while larger values indicate greater spending on collective goods. More detailed information on
the proportion of expenditures dedicated to each area can be obtained through simple subtraction
as well, allowing for detailed predictions.
The new measure of democratic government spending priorities provides support for the
argument that policy priorities can be shown along a single dimension and that the choices made
by governments are a product of the group of citizens who are intended to benefit from these
policies. Having demonstrated how governments spend using spending priorities, I can examine
what factors influence these priorities. Compared to prior research that examines the influence
of different indicators, like the state of the economy, on policy areas separately or through
composite measures that encompass different aspects of the policy process, I can now analyze
the t the factors that influence the overall pattern of spending priorities in democratic nations
across time within a single model.

76

CHAPTER 4
DATA AND HYPOTHESES
Prior literature suggests that three main types of variables influence government spending
priorities. The first set of factors includes the socio-economic conditions that indicate who in
society needs government assistance, like the elderly population and the unemployed, and the
resources available for government to spend. Further, as a democratic government is expected to
have policy outputs that correspond to the expectations of the people it represents, the second set
of factors addresses how the preferences of different groups, including political parties and
citizens, affect government spending. The final set of variables I test focus on different types of
political institutions measured separately and in combination with each other. The role of
institutions is suggested to shape the behavior of incumbents and as such alter policy outputs in
predictable ways.
FACTORS INFLUENCING GOVERNMENT SPENDING
The socio-economic make-up of the society in which governments operate is found to
influence policy expenditures; this includes the level of wealth in a nation and the composition of
different groups in society like the elderly, the youth, and the workforce. Based on the socioeconomic condition of a nation, governments have various resources to work with and problems
to solve. In a nation where unemployment is high, a government has a greater need to address
social welfare issues to appease its citizens. As unemployment increases, more individuals are
unable provide basic necessities for themselves and their families. The expectation then is that
when unemployment in a nation is high, a nation will spend more on particularized policy areas
such as housing subsidies to help alleviate issues resulting from unemployment. However, in a
nation where unemployment is lower, the government will have less pressure from society to
alleviate problems and will spend less on particularized policy areas relative to collective goods.
77

H2a: Increases (decreases) in unemployment produce governments that spend more on
particularized benefits (collective goods).
In order to operationalize unemployment, I use data available from the World Bank based
on the percent of unemployed out of the total work force population. Unemployed individuals
are considered as those who are out of work but are actively seeking employment. This
information is available for all years of the analysis and for all nations.
Tied to employment and policy expenditures is the number of women in the workforce.
As more women enter the workforce the dynamics of society change because the needs of those
who are employed change. The more women who participate in the workforce, the more aid
governments provide to make it easier for women to enter and remain in the workforce. In order
to help support female participation in the labor force, nations are expected to increase spending
on particularized policy areas that help reduce the barriers of female participation, such as
providing day care services for children. Therefore, the expectation is that as the percentage of
women who are a part of the workforce increases, the more a nation will spend on particularized
policies.
H2b: Increases (decreases) in the percentage of women in the workforce produce
governments that spend more on particularized benefits (collective goods).
Information regarding the percentage of female participation in the workforce is obtained
from the World Bank. The World Bank provides data on the workforce participation rate for
females over the age of 15 as a percentage of the total female population in a nation. This
information is available for all nations in the analysis for all time periods.
Another characteristic of a nation found to affect government expenditures is the size of
the dependent population, which involves the percentage of youth and elderly populations in

78

nations. As the percentage of elderly increases in a nation, fewer citizens are working and able
to provide a minimum standard of living for themselves. Based on the increase in aid demanded
of a government for elderly citizens, the expectation is that as the percentage of elderly increases
in a nation, so will the nation’s expenditures on particularized policy areas such as pensions.
However, as a nation’s elderly population decreases, there is less strain placed on the
government to provide for the elderly population and it is expected that the nation will have
greater relative expenditures on collective goods.
While prior work at times has grouped the youth population together with the elderly as a
measure of “dependent populations” (Huber and Stephens 2000, 2001), the needs of the two
groups differ. The elderly may require assistance to maintain a minimum standard of living,
whereas younger populations require a different set government services, such as educational
spending. Because the needs of the youth population focus on services that are collective goods,
larger proportions of the youth population are expected to increase spending on collective goods.
As the needs of the two dependent populations differ, I will examine the effects of the aged and
youth populations separately.
H2c: Increases (decreases) in the percentage of elderly in the population produce
governments that spend more on particularized benefits (collective goods).
H2d: Increases (decreases) in the percentage of youth in the population produce
governments that spend more on collective benefits (particularized benefits).
Variables representing the size of the youth and aged population are created with data
from the World Bank. The aged population variable is based on data from the World Bank on
the population above the age of 65 as a percentage of the total population. The youth population

79

variable is based on the percentage of the population under fifteen years of age. This
information is available for all nations in the analysis across all time periods.
Another trait in society that is expected to alter policy expenditures is the wealth of the
population in a nation. As wealth in a nation increases the demands placed on the government
by the people to provide more goods and services increases. The form of the goods and services
demanded by the people as development occurs take on the form of particularized spending such
as unemployment and retirement benefits. Wealth is expected to increase spending on
particularized benefits because governments are in a position to provide services to particular
groups in the community without depriving the general population of services through collective
goods. When governments have fewer resources, relative spending is expected to favor more
collective goods that serve the broader community at the expense of particularized benefits such
as housing or daycare.
H2e: Increases (decreases) in national wealth produce governments that spend more on
particularized benefits (collective goods).
I operationalize wealth following standard practices in the literature and use real gross
domestic product per capita (GDP/Capita). This variable is created using data from the World
Bank and the Bureau of Labor Statistics (BLS). Information regarding the current gross
domestic product per capita in nominal US dollars, is obtained from the World Bank. As the
gross domestic product per capita figures are in nominal dollars adjustments are made for
inflation to obtain real gross domestic product per capita figures. In order to obtain real gross
domestic product per capita figures information regarding the Consumer Price Index (CPI) for
the US is taken from the BLS. Using the nominal gross domestic product per capita and the CPI,
the real gross domestic product per capita figures are calculated for all the nations in the study

80

were in real 2008 US dollars. The following formula is used to change the nominal gross
domestic product per capita figures into real gross domestic product per capita. The nominal
gross domestic product per capita value at year t, is multiplied by the 2008 CPI value divided by
the CPI value for year t:
Real GDP/Capitat = Nominal GDP / Capitat * ( CPI2008 / CPIt )

(4.1)

In regards to the economy, the level of inflation in a nation alters the resources a nation
has to work with when dealing with policy expenditures. As the level of inflation increases, it
costs a government more to provide the same level of goods and services. Based on the effect of
inflation, it is expected that as inflation increases, governments are less able to spend on
particularized policy areas due to increased costs. As inflation decreases, governments are able
to supply more goods and services with the same amount of resources and will spend more on
particularized policy areas without having to decrease the level of collective goods provided to
the general public.
H2f: Increases (decreases) in the inflation rate produce governments that spend more on
collective goods (particularized benefits).
Inflation is examined in the analysis with data from the World Bank. The inflation
variable is measured as the annual percent of inflation for each nation year in the analysis. The
data are available for each year of the analysis for all nations.
The effect of trade openness due to globalization in relation to government spending has
varied in the literature. Some work has argued that increases in trade and globalization will
increase government spending on particularized benefits to offset the costs and risks due to more
open economies involving issues like unemployment or housing benefits (Cameron 1978; Rodrik
1998). Alternatively, other research suggests that increased openness will decrease government
81

spending on particularized benefits because policies favoring particularized benefits make
nations less attractive to businesses, as it is more costly to produce in nations with higher taxes
and better protected workforces with generous unemployment and disability benefits (Scharpf
2000). Therefore, I test the effect of openness on government spending with no previously
determined directional expectation.
To test the effect of openness on government spending patterns I use data on imports and
exports from the OECD. Following the approach used by Huber and Stephens (2001), I
operationalize trade openness as the total imports and exports as a percentage of GDP. Higher
proportions are suggested to represent greater openness in terms of a nation’s economy.
A final socio-economic indicator expected to shape government actions is membership in
the European Union. Nations that belong to the European Union are intentionally integrating
their economies and creating binding and non-binding policies. As such, the nations that are
members of the European Union are expected to behave differently than the nations that are not
members. Nations that belong to the European Union have traditionally had higher levels of
spending on particularized policy areas that focus on welfare. Nations that belong to the
European Union have attempted to establish minimum standards involving elements of social
protection for workers; promoting attention to particularized policy spending that includes
spending on disability, sickness, survivor benefits, pensions, and unemployment. Therefore, the
expectation is for nations that belong to the European Union to spend more on particularized
benefits than non-European Union member nations.
H2g: Members of the European Union will spend more on particularized benefits than
non-member nations.

82

In order to capture the effect of the European Union, I create a variable based on
membership in the European Union with information obtained from Europa.org. The EU
variable is a dummy variable that receives a one if a nation was a member of the European Union
in a given year. If a nation was not a member of the European Union in a given year it is coded
as a zero.
Different forms of political preferences are found to shape government actions. As
discussed in Chapter 2, the political parties in office, voter turnout, role of government, public
opinion, and interest groups all influence government actions. The composition of the
government in terms of which parties are in office affect the decisions a government makes.
Political parties located on the ideological left are typically associated with greater emphasis on
policies which benefit particular groups in society, generally those who may be less well off and
in need of assistance. As such, governments dominated by left parties are expected to have
greater spending on particularized policy areas like housing or food benefits. Conversely,
political parties on the ideological right are less likely to be associated with greater spending on
policies aimed at particular groups in the society. Instead, the expectation is that governments,
dominated by political parties on the right will spend less on particularized policy areas and
relatively more on policy that target the general public in the form of collective goods, such as
economic development or defense.
H2h: Ideologically left (right) dominated governments spend more on particularized
benefits (collective goods).
The composition of the government is based on information available from the Database
of Political Institutions (DPI), political parties’ websites, and the European Elections Database
(EED). The DPI provides information about the number of seats held by the top three parties in

83

control of the government and the number of seats held by the top three opposition parties, in
addition to the total number of seats in the lower house with its gov#me and gov#seat
variables.

18

At the same time, the DPI provides Left, Right and Center designations for each of

the major parties in office with its gov#rlc variables. Because the focus here is on expenditures
and the DPI coded Center parties as those that are fiscally conservative; not based on overall
Left/Right policy positions, Center designations are reclassified here as Right. For political
parties where the DPI did not provide a Left, Center, or Right position, information available
from the EED was used to determine the political parties designation based on names in the cases
of Green and Communist parties or the EED’s classification of Left or Right. For the smaller
parties where data was not available from the DPI or the EED, the political parties’ websites are
used and the Left/Right leanings of the parties label are assigned.

19

When coding is a result of

personal judgment, attempts are made to follow the DPI coding as closely as possible with
“Right: for parties that are defined as conservative, Christian democratic, or rightwing…[and]…Left: for parties that are defined as communist, socialist, social democratic, or
left-wing” (Keefer 2009, 6).
The variable representing the composition of the government is calculated as the
percentage of seats held by leftist parties. The percentage of seats belonging to leftist parties is
calculated based on the number of seats held by leftist party members and the total number of
seats in the lower house for which information is available for each year.

18

st

nd

rd

The # sign takes the place of 1 , 2 , and 3 largest party numbers. The average number of
seats captured by the top three parties is 94.5%.
19
In situations where the party websites are not posted in English, Google Chrome is used to
translate the websites into English.
84

In addition to the parties in government, who votes has been suggested to influence
government actions. Citizen mobilization has been argued to shape government activities as it
determines what political parties are in office. However, there is disagreement on what increased
voter turnout implies: more low income voters electing left party members into government or
more wealthy voters electing right party candidates. The mixed findings for voter turnout may
be a product of the previous measures used to capture government activities. As such, I test for
the effect of voter turnout in Chapter 5. Therefore, while I expect voter turnout to influence
government spending patterns, there is no directional expectation for its influence.
Citizen mobilization is captured using a measure of voter turnout. The voter turnout
variable is created using data available from Institute for Democratic Election Assistance
(IDEA). The variable is created using data on the total number of votes cast (valid or invalid)
divided by the number of names on the voters' register. The level of turnout is held constant
between election years based on the percentage at the time of the last election year.
As democratic governments are expected to be responsive to the preferences of their
citizens, the expectations for the role of government should affect a nation’s policy decisions.
Some nations have expectations that emphasize a more limited role for governments, such as
focusing greater efforts on ensuring a growing economic climate and protecting its citizens. In
nations that exhibit beliefs with more limited roles governments are predicted to spend more on
collective goods that include policy areas like defense and economic development. Some
nations, however, have expectations of a greater role for government, such as ensuring freedoms.
In nations with a broader view of government’s role, the prediction is for government spending
to promote basic levels of equality that would require spending on more particularized benefits.

85

H2i: Expectations of government dominated by views of a limited (expansive) role for
government produce governments that spend more on collective goods (particularized
benefits).
The role of government is measured here using Inglehart’s Post-Materialism index
available through the World Values Survey (WVS) and the European Values Survey (EVS). The
questions used to create the four-item index can be seen as representing the different
expectations for government involvement in more aspects of life or less. Of the four goals for
their country people are asked to select among, two are maintaining order in the nation and
fighting rising prices, which can be seen as more limited roles of government. Meanwhile,
giving the people more say in important government decisions and protecting freedom of speech
and can be taken as expanding the role of government. Based on respondents’ selections of the
top two priorities the variable for role of government is created based on the percentage point
difference between those who selected the two values of expanding government’s involvement
and those who selected the two values that focus on government having a more limited role. As
survey data are not available for every year of the analysis, interpolation is used between
available data points.

20

In situations where the observations fall after the last available year of

data, the last known interpolated rate of change is calculated and is used going forward. In years

20

Data for the United Kingdom was not available in a United Kingdom aggregated format.
Therefore when the data are available for Northern Ireland and Great Britain the data are
recombined to create a United Kingdom data value. In 1998, data for Northern Ireland and 1999
data for Great Britain were combined and used as both 1998 and 1999 values for the United
Kingdom. In 2006, only Great Britain has available data which are used as the last known data
point for the United Kingdom.
86

before the first known data point, the first known data point is used as a constant going back in
time as the rate of change is unknown.

21

Role of Government = (Post-Materialist Values%) – (Materialist Values %)

(4.3)

Even though societies tend to have established boundaries on the role of the government,
public opinion on how governments should address particular issues can still vary. As such, the
expectation is that when liberal preferences dominate citizens’ public opinion in a nation, the
government will spend more on particularized benefits that include welfare spending. However,
when conservative opinions dominate, the expectation is to see government focus shift to greater
spending toward collective goods. For example, when expectations limit the role of government
actions to areas like the economy a more liberal public opinion may prefer government
addressing unemployment through increased spending on unemployment benefits or housing
subsidies (particularized benefits), while more conservative publics may prefer governments to
use tax cuts to stimulate business growth to reduce unemployment (collective goods).
H2j: Public opinion dominated by liberal (conservative) attitudes produces governments
that spend more on particularized benefits (collective goods).
In order to capture public opinion, information from the WVS, waves one through five,
regarding what political party a respondent would vote for if an election was held today are used.
Klingemann et al. (1994) argue that political parties represent packages of policies that they
present to the voters through their party manifestos. If parties represent packages of policies to
21

Interpolation is an approach that can be used to calculate new data points within a range of
known data points. The interpolations were calculated in the following linear manner:
yt+1=yt+(yb-ya)/(b-a)
Here, a and b represent the values at the first (a) and last (b) year in the range of interpolation
and t=a. This approach is applied between all known data points where there is no available
data. This method of interpolation is applied to all survey data questions used in the analysis
where change occurs between time points.
87

be implemented upon obtaining office, then the political party individuals would vote for at any
given time should correspond to the policy packages individuals prefer. This variable is used as
a proxy for public opinion as it represents the preferences of individuals across a number of
policy areas based on the party platform they would support.
The political party respondents state they would vote for are coded as left or right using
the same coding scheme used to classify the parties for the composition of the government
variable. After adjusting for respondents who did not have an answer for the question, the
difference is calculated between the sum of those who would vote for a left party and those who
would vote for a right political party using the following formula for each nation year:
Public Opinion =

(4.4)

Again, as information is not available for all years, values are interpolated between years of
available data.

22

The last interpolated rate of change for observations is used to extrapolate past

the last known data point to calculate values going forward, while the first known data point is
used as a constant for years occurring before the first data point as the pattern of change is
unknown prior to available information.
An additional factor that affects government actions in a nation is interest groups.
Interest groups organize to promote specific agendas based on the aligned preferences of their
members. If there are a limited number of interest groups in a nation, the groups may be capable
of capturing the attention of government and swaying officials to increase spending on particular
policy areas favoring particular groups’ interests. However, as the number of interest groups in a

22

Data for South Korea were not available for any time periods. Additionally, parties receiving
less than 0.2 percent of the respondents vote in the survey were omitted from the study as a result
of obscure and difficult to locate information on party positions.
88

nation increase, it becomes difficult for a government to respond to the demands of interest
groups. Therefore, the expectation is that as the density of interest groups in a nation increase,
the ability of the groups to influence a government to spend more on their particular interests’
decreases; and as such the expectation is for greater spending on collective goods.
H2k: Increases (decreases) in the density of interest groups in a nation produce
governments that spend more on collective goods (particularized benefits).
I use a factor analysis on two variables representing interest group strength to create the
interest group variable. The first variable included in the factor analysis is the number of
business associations present in a nation based on information from the World Guide to Trade
Associations. The second variable is the number of public sector employees in thousands based
on data from the International Labour Organization’s Labour Statistics Database. Government
employees are used as a measure for interest group strength because government employees can
act as advocates for interest groups and have previously been used as a measure of interest group
strength (Jacoby and Schneider 2001). The factor analysis produced a single factor representing
23

interest group strength (Table 4.1).
Table 4.1

Results of Factor Analysis for Interest Groups
Variable
Business Associations
Government Employees

Factor 1 Uniqueness
0.7527
0.4335
0.7527
0.4335

As an additional check on the influence of interest groups, I use Lehmbruch’s (1984)
measure of corporatism. The higher the degree of corporatism on the 1 to 5 scale, the more
23

Additional model specification runs captured interest groups as: count of business
associations, count of government employees, and GDP/business association to capture resources
available to groups. The alternative measures resulted in models with similar results in terms of
signs, magnitude, and statistical significance of coefficients.
89

influence interest groups have over government actions. However, the data on this variable are
more limited in terms of the number of coded countries and only overlap with 214 of the nationyears in my dataset. Therefore, I use this to confirm the results of prior work using this measure,
but also run alternative models using the variable produced by the factor analysis discussed
above.
Prior research has led to the conclusion that institutions should matter for policy outputs
in a nation (Iversen and Soskice 2006; Persson et al. 2007; Huber and Stephens 1993; Immergut
2010; Lijphart 1999). Institutions found to influence government actions include presidential
versus parliamentary systems, majoritarian versus proportional representation systems, district
magnitude, bicameralism, and federalism.
Democratic nations typically use either a presidential or parliamentary system to
determine the executive. Both presidential and parliamentary systems are associated with
different expectations about the behavior patterns of the executive. In a presidential system, the
executive is elected separately from the legislature, and must obtain a majority of the votes in
order to win office. Therefore, a presidential system promotes an executive who attempts to run
on a policy platform to appeal to as many citizens as possible (using collective goods). If a
presidential candidate runs on a policy package that is targeted towards too specific of a group in
society the candidate risks pushing away potential supporters.
Unlike a presidential system where the executive is elected separately from the
legislature, under parliamentary systems the executive is appointed by the legislature. Typically,
the controlling party or coalition within the legislature in a parliamentary system selects the
executive. In this situation, the executive serves to promote the interests of the group that
appointed the executive. The dynamics of a parliamentary system promote an executive who

90

represents the interests of the ruling party or coalition. The expectation is for nations with
parliamentary systems to favor spending on more particularized policy areas as compared to
presidential systems. Conversely, in a presidential system, the expectation is for greater emphasis
on spending towards collective policy areas such as education and economic development.
H2l: Presidential (parliamentary) systems produce governments that spend more on
collective goods (particularized benefits).
In order to capture whether a nation has a presidential or parliamentary system, data from
the DPI are used. The DPI notes if in a given year a nation has a presidential or parliamentary
system with its system variable that I use to create a dummy variable representing the presence of
a presidential system. A one for the presidential systems variable represents a nation with a
presidential system and a zero indicates a parliamentary system in the nation, in a given year.
In a nation using a majoritarian system for its electoral formula, the candidate with the
most votes will gain office. Majoritarian systems promote candidates that appeal to as many
citizens as possible in order to win elections. Candidates in this situation are expected to run on
platforms that appeal to citizens in general and not to particular groups of individuals. If
candidates in a majoritarian system opt for particularized policy platforms they risk alienating a
voter base of citizens who either do not receive benefits from the particularized policies or are
made worse off by policy platforms. As a result, it is expected that nations that use majoritarian
systems will have greater spending on collective goods.
In a proportional representation system, candidates do not necessarily need to win a
majority of votes to win office; instead, candidates can win by appealing to subgroups within
society. Candidates then use particularized policy packages to appeal to targeted groups that are
large enough to win office. The particularized policies may alienate other groups within the

91

society, but as a candidate does not need a majority to win an election, particularized policy
packages, such as family or disability benefits are used to secure a reliable vote base. Based on
the expected behavior of candidates in nations with proportional representation systems, the
expectation is for a greater emphasis on spending in particularized policy areas.
Some nations do not use either a pure majoritarian or pure proportional representation
system, but instead opt for an electoral system that uses a combination of the two formulas.
Compared to a pure proportional representation system, in a mixed system some candidates are
elected under majoritarian rules and others through proportional representation. The mix
produces some candidates then whose policy platforms target particular subgroups within the
population and some who run on more collective policy platforms. In this situation it is expected
that mixed systems will have greater spending on collective goods as compared to proportional
representation systems as some candidates aim for majority votes with collective goods.
Relative to majoritarian systems, nations with mixed systems will have greater spending on
particularized policy areas with some candidates targeting subgroups within a population to win
elections.
H2m: Proportional representation, compared to majoritarian systems, produces
governments that spend more on particularized benefits.
H2n: Mixed electoral systems, compared to majoritarian (proportional) systems, produce
governments that spend more on particularized benefits (collective goods).
A dummy variable representing the voting structure in a nation in a given year is created
using information from the DPI variables plurality and proportional representation. Using the
information on the voting structures I create three dummy variables called majoritarian,

92

proportional representation, and mixed voting based on the DPI. A one represents the presence
of the attribute for which the variable is named and a zero represents the absence of the attribute.
While the electoral formula determines how votes result in winning candidates, district
magnitude determines how many candidates can win an election. As district magnitude
increases, the number of votes a candidate needs to win an election decreases. In a district with a
magnitude of one, only the candidate with the most votes will win the seat. When more seats are
available in a district, candidates can win a seat without obtaining the most votes. Increases in
district magnitude are expected to mirror a proportional representation system, where candidates
turn towards targeted groups within a population in the hopes of obtaining a secure vote base that
can be relied on at election time. In order to win the support of targeted groups within the
population, candidates focus on policy areas that are aimed at the unique characteristics of a
group towards which other candidates are not catering. Based on the argument regarding district
magnitude, the expectation is that increases in district magnitude move candidates towards
catering to particular groups within the population, producing a government that will spend more
on particularized policy areas than proportional representation policy areas.
H2o: Larger district magnitudes produce governments that spend more on particularized
benefits.
In order to capture district magnitude size I use information from the DPI regarding mean
district magnitude for the lower house. As not all nations have a bicameral legislature data for
the district magnitude size of the upper house do not always exist, unlike information regarding
the mean district magnitude size for the lower house. The DPI calculates mean district
magnitude based on available data involving the number of representatives for each
constituency.

93

While bicameralism and federalism are both argued to act to constrain the ability of
governments to act in terms of reaching policy agreements, there are no standing arguments for
how these institutions should or would individually shape expenditures in democratic nations.
Particularly for federal systems, as the division of policy responsibilities is not universal and
varies widely in terms of the unique policy domains. However, as bicameral systems increase
the number of preferences present in the decision making process by dividing power across two
houses, it increases the difficultly for any particular actor to move policy in a direction that
favors a particular group at the expense of other groups. Therefore, I expect bicameral systems
to spend more on collective goods that benefit groups more broadly compared to unicameral
systems where fewer actors need to agree over policy.
H2p: Bicameral (unicameral) systems produce governments that spend more on
collective goods (particularized benefits).
Bicameral legislatures are captured by creating a dummy variable with information from
the Inter-Parliamentary Union’s PARLINE database on national parliaments. The bicameralism
variable is coded as a one when a nation has a bicameral legislature and a zero when a nation has
a unicameral system
Additionally, federal systems have been unable to reduce spending on areas of social
protection as opposed to unitary systems during the era of welfare retrenchment during the time
period examined (Obinger, Leibfried, Castles 2005). Therefore, I expect federal systems to
spend more on particularized benefits compared to unitary systems in the time period examined.
This expectation is based on their inability to shift spending priorities as quickly as unitary
systems, and is not based on federal systems having a particular nature for spending on one
policy area over another.

94

H2q: Governments with federal (unitary) systems spend more on particularized benefits
(collective goods).
Nations with federal systems are captured using information from Bednar’s work on
federalism. Bednar breaks down nations that have unitary, quasi-federal, and federal systems. I
create a federalism dummy variable where a one corresponds to a nation with a federal system
based on Bednar’s coding and a zero in all other situations. Bednar’s coding produces three
quasi-federal systems in my dataset that are coded as zeros: Italy, Spain, and the United
Kingdom. Each quasi-federal system lacked one of Bednar’s three defining criteria to be labeled
as federal. Italy lacked direct governance where “authority is shared between the nation and the
national governments: each governs its citizens directly, so that each citizen is governed by at
least two authorities. Each level of government is sovereign in at least one policy realm. This
policy sovereignty is constitutionally declared” (Bednar 2009, 18). The inability to have unique
policy domains prevents federalism from acting as a constraint as discussed earlier because the
national government can intervene on any policy area without overstepping its political limits.
Spain and the United Kingdom both lack geopolitical division, where the “territory is divided
into mutually exclusive nations (or provinces, Länder, etc). The existence of each state is
constitutionally recognized and may not be unilaterally abolished” (Bednar 2009, 18). Both
Spain and the United Kingdom do not divide their territories into fully autonomous regions that
can act to constrain the national government’s actions and as such are coded as zeros.
POLICY RESPONSIVENESS HYPOTHESES
The second set of expectations involves institutions and the interaction between
institutional constraints and preferences. This set of hypotheses is used to address the third
question set out at the beginning of this dissertation: Does the institutional design of a nation

95

alter the role of preferences in shaping government spending? As the number of institutional
constraints increases there are two anticipated results. The first expectation is that an increase in
the number of constraints will shift spending towards collective goods that are intended to
benefit more general groups within the population. This is a product of constraints introducing
more actors, with preferences over policy outputs, into the decision making process. As the
number of groups present in the decision making process increases, it becomes harder for any
actor to increase spending on policy areas that benefit their particular interest at the expense of
others. The result then is greater spending on collective goods that benefit groups more broadly.
H3a: Nations with more institutional constraints spend more on collective goods, relative
to nations that have fewer institutional constraints.
In order to capture the number of constraints in a nation, information regarding the
previously generated institutional variables is used. The constraint variable is an additive index
based on the count of institutional constraints present in a nation in a given year (alpha=0.7291).
For the purpose of the count of constraints, the district magnitude variable discussed above is
converted into a dummy variable where a one represents average district magnitudes greater than
one and a zero represents district magnitudes equal to one. I also collapse the electoral
institutions variable into a majoritarian/non-majoritarian dummy variable where a one represents
the presence of a proportional representation or mixed voting system. Both proportional
representation and mixed voting systems increase the number of actors present in the decision
making process and are expected to serve as constraints.

96

Constraints = President + Non-majoritarian + District Magnitude Dummy
+ Bicameralism + Federalism

24

(4.5)

Further, as the constraints increase in number in a nation, policy becomes more difficult
to change and is more resolute. The increased policy resoluteness is a product of the increase in
the number of actors with preferences that need to be accommodated for policies different than
the status quo to be enacted. The increased difficulty of agreement should decrease the ability of
any group/actor to obtain its ideal policy whether it is for greater spending on particularized
benefits or for collective goods. Building upon the analyses in Chapter 5 that test the second set
of hypotheses presented earlier in this chapter, I expand the study of influences on government
spending patterns in Chapter 6. I argue that as the number of constraints increase, governments
respond less to the preferences of different groups. I test this argument using the following
hypothesis:
H3b: Increasing the number of institutional constraints decreases the policy
responsiveness of governments to different groups’ preferences for government outputs.
To test the effect of institutional constraints on preferences in Chapter 6, I interact the
constraints variable with each of the four measures of preferences: government composition, role
of government, public opinion, and interest group density. I use interactions between the number
of constraints and measures of preferences because I hypothesize that the effect of each group
should not only be a product of its respective expectations, but that its effect will be contingent
upon the constraints present. Therefore, as the number of institutional constraints changes, the
ability of government spending to respond to each group should also change.

24

Creating an additive index of institutional constraints is also the approach used by Huber and
Stephens (1993, 2000, 2001) and Brooks and Manza (2007).
97

Table 4.2 provides a brief look at the data used in Chapter 5 and 6 to test the hypotheses
presented in this chapter. The first variable in the table provides descriptive statistics for the
spending priorities variable that is created in Chapter 3. A brief summary of the expectations
regarding the variables from this chapter is available in Table 4.3.
Table 4.2

Summary Statistics

Variable
Government Spending Priority
Presidential System
Majoritarian System
Proportional Representation
Mixed Voting
District Magnitude
Bicameralism
Federalism
Constraints
Government Composition
Voter Turnout
Role of Government
Public Opinion
Business Associations
Government Employees
Corporatism
Female Participation Rate
Unemployment Rate
Aged Population
Youth Population
GDP/Capita
Inflation Rate
Openness
EU Members

N
369
369
369
369
369
369
369
369
369
369
369
369
369
369
369
214
369
367
369
369
369
369
369
369

Mean
53.36
0.07
0.19
0.55
0.26
13.37
0.57
0.24
2.47
46.60
74.92
-1.57
0.96
874.98
2674.36
3.07
50.30
7.88
14.71
18.23
33332.55
3.41
88.03
0.79

98

Standard
Deviation
1.45
0.26
0.39
0.50
0.44
26.85
0.50
0.43
0.98
16.83
11.11
16.45
26.94
1291.71
4415.92
1.43
8.81
4.14
2.05
2.54
15576.12
3.76
51.50
0.41

Minimum Maximum
50.90
57.48
0.00
1.00
0.00
1.00
0.00
1.00
0.00
1.00
1.00
150.00
0.00
1.00
0.00
1.00
0.00
4.00
5.61
94.82
40.57
93.38
-57.90
29.30
-71.11
60.08
1.00
5773.00
26.79
21974.00
1.00
5.00
33.20
75.60
1.50
23.90
10.68
19.92
13.81
28.02
4547.40 110933.00
-1.88
30.62
16.01
319.55
0.00
1.00

Table 4.3

Summary of Hypotheses

Directional Tests
Particularized Collective
Benefits
Goods
+
+

Non-Directional
Test

Presidential System
Majoritarian System
Proportional
Representation
+
Mixed Voting
+
District Magnitude
+
Bicameralism
+
Federalism
+
Constraints
+
Government Composition
+
Voter Turnout
+
Role of Government
+
Public Opinion
+
Business Associations
+
Government Employees
+
Corporatism
+
Female Participation Rate
+
Unemployment Rate
+
Aged Population
+
Youth Population
+
GDP/Capita
+
Inflation Rate
+
Openness
+
EU Members
+
Note: For the directional tests, + represent an increase in spending on for either
particularized benefits or collective goods in relation to the presence of the
characteristic for categorical variables and increases for the interval level
variables.

99

CHAPTER 5
TRADITIONAL INFLUENCES AND SPENDING PRIORITIES
Government expenditures are divided between policies that target particular groups
within the population, like the elderly or the poor, or the society more broadly through spending
on areas such as economic development or health. Where some nations like Greece and
Denmark spend more on particularized benefits and others such as Japan and the Czech Republic
spend more on collective goods.
Prior research provides several different sets of influences that are expected to shape
government spending patterns and outputs; including socio-economic factors, mass and elite
preferences, and political institutions. However, the work on these various arguments generally
results in misspecified models. Previous studies typically analyze one or two sets of influences
such the role of socio-economic influence and institutions (Crepaz 1998; Scartascini and Crain
2002; Shelton 2007; Edwards and Thames 2007; Chang 2008) or preferences and the socioeconomic climate (Bradley et al. 2003; Bräuninger 2005; Soroka and Wlezein 2005); however,
these studies show, all three sets of factors influence government activity after controlling for
25

each other.

To address the concern of misspecified models I run a spending priorities model

using a number of measures from all three perspectives on what shapes spending. This approach
helps confirm how expectations for factors work when controlling for other theoretically
influential variables.
Even research that uses more complete model specifications produces questionable
results based on the dependent variables that are used to measure government outputs. For
example, Huber and Stephens (2001) combine health care and pension spending into a single
25

To see model runs for the three sets of influences (socio-economic, mass and elite preferences
and political institutions) see Appendix B.
100

variable. However, this is directly at odds with the results of the unfolded spending priorities
variable that demonstrates these two items belong to policy areas that target two different types
of populations and increases in spending on one area (health/pensions) will decrease spending in
the other (pensions/health). By using the spending priorities variable I am able to test the
influence of multiple variables on a measure of government outputs that correctly captures the
relationship between policies. Through the priorities model I show what factors influence
government spending patterns and determine if findings from prior work are maintained in a
correctly specified model.
SPENDING PRIORITIES MODEL
Model 1, referred to as the spending priorities model, shows the results of the fully
26

specified spending model (Table 5.1).

The findings of the priorities model are broken down by

the three types of influences shaping government spending: socio-economic factors, citizens’
preferences, and political institutions.
Socio-Economic Influences and Spending Priorities
Regarding the socio-economic variables, the expectations from Chapter 4 predict that high levels
of wealth, unemployment, female participation in the workforce, dependent populations
comprising of the elderly and youth populations, and membership in the European

26

The models run in this chapter are time-series, cross-sectional analyses with panel corrected
standard errors. Each model includes year dummies that are omitted from the table of results for
clarity of reading. Each independent variable is lagged by two years to present a more accurate
picture of the expenditure process. The models follow the same arguments as the ones
presented in more detail in the next chapter. For more details on the model specification please
see Chapter 6 and Appendix C.
101

Table 5.1

Spending Priorities Model
Coefficient
(s.e.)

Variable
a
GDP/Capita
Unemployment

-0.59
(0.129)

b

European Union
Government Composition
Voter Turnout

b

Role of Government
Public Opinion
Interest Groups
President
PR
Mixed Voting
District Magnitude

a

Bicameralism
Federalism
2

R
N

0.01
(0.001)
-1.67
(0.099)
0.001
(0.002)

0.000

0.467

0.34
(0.042)
0.80
(0.059)
-1.14
(0.096)

Aged Population

0.000

-0.0003
(0.005)
-0.009
(0.003)
-0.01
(0.003)
0.61
(0.093)
-0.14
(0.245)
-1.42
(0.161)
-0.89
(0.157)

Inflation

0.000

-0.51
(0.104)
0.04
(0.007)
0.02
(0.011)
-0.13
(0.025)

a

Female Participation

Openness

p-value

0.000

0.000
0.027
0.000

0.000
0.366

0.001
0.000
0.000
0.283
0.000
0.000

0.000
0.000
0.7647
367

102

Table 5.1 (cont’d)
Note:

a Indicates that the natural log of the
original variable was used in the model.
b Indicates that p-value for the variable is
for a non-directional test, all other
variables based on directional tests.

Union would increase government spending on particularized benefits.

27

Higher levels of

inflation are expected to increase spending on collective goods, while the levels of trade
openness do not have a predicted directional expectation as past works have shown mixed
findings. Overall, the variables in the priorities model conform to the expectations found in the
previous literature.
Economic resources are related to government spending priorities. Specifically, wealthier
nations (represented by higher values for gross Domestic Product per capita) allocate greater
resources toward particularized benefits ( i.e., family benefits and housing assistance) as
represented by the negative coefficient. Conversely, nations with less economic resource are
more likely to spend money on collective goods. The effect of gross domestic product per
capita is constantly negative and is statistically significant at the 0.05 level for directional tests.
This result indicates that economic does, indeed, affect the spending priorities of democratic
nations.
As the percentage of the population unemployed increases, the expectation is that nations
will need to spend more on programs that assist individuals who cannot provide for themselves
or their families. As expected, the findings show that higher levels of unemployment are re
associated with greater spending on particularized benefits. The unemployment rate variable has
27

Studies that use either exclusively or predominately socio-economic indicators include:
Rodrik 1998; Cameron 1978; Shelton 2007; Bradley et al. 2003; Huber and Stephens 1993; and
Chhibber and Nooruddin 2004.
103

a negative sign, indicating that as more individuals become unemployed governments do spend
more of their resources on particularized benefits like unemployment benefits.
When inflation in a nation increases, it becomes more expensive for nations to provide
the same level of goods and services. So, nations with higher inflation rates will spend less on
areas that benefit only certain groups of the population and promote spending of more limited
resources that are intended to provide broader benefits to society as a whole. The results of the
priorities model shows that higher levels of inflation are associated with greater spending on
collective goods. The coefficient for the inflation rate is positive, indicating that nations with
higher levels of inflation are more likely to spend more on collective goods cluster of policies.
The result is statistically significant for a directional test at the 0.05 level.
Female participation in the workforce are expected to increase government spending on
particularized benefits; however, the results suggest that greater female participation rates are
associated with higher spending priorities values which represent greater spending on collective
goods such as education and economic development. The increase in spending on collective
goods may imply that as more women enter the workforce, they are better able to provide for
themselves and their families, requiring less assistance from the government in the form
particularized benefits. The impact of female participation is statistically significant for a
directional test at the 0.05 level.
Larger shares of the population that are dependent are expected to increase spending on
particularized benefits to address the needs of vulnerable population groups. Several models
were initially examined to test the influence of the dependent population on government
spending patterns. The first model combined the affect of the youth and aged groups in society.
Here the results indicated that the dependent population did not have a statistically significant

104

affect on spending at the traditional levels of significance. Since the old and the young have
different types of needs, this is not a surprising result. The elderly would require greater
assistance in the form of particularized spending on pensions, while the youth would have a
greater demand for services like education, a collective good. These opposing needs may result
in the effect of these groups cancelling each other out when the two are combined into a single
variable.
The second model examines the influence of the youth and aged populations, separately,
to determine if either group has a unique affect on spending priorities. The results of this model
show that higher levels of the aged population are associated with greater spending on
particularized benefits, like pensions, with the negative sign of its coefficient. However, the
effect of the youth population failed to reach statistical significance at either the 0.05 or 0.10
levels for a directional test, indicating that the young in a population do not contribute to the
spending patterns. This finding may be a result of the youth population who are younger than 15
and unable to vote and influence politicians and spending, while the elderly can and do vote.
The specification of the dependent population used in the priorities model presented here
omits the youth population that was not found to shape government spending priorities and still
shows that higher percentages of the aged population increase government spending on
particularized benefits. The results in Table 5.1 show that nations with larger aged populations
spend more on particularized benefits than collective goods. The finding supports the
expectation in the literature that the larger this group is, governments spend more on items like
pensions to address the needs and demands of this particular segment of society.
Prior research suggests that the degree of trade openness should affect government
spending patterns. However, the expectation associated with trade openness has produced mixed

105

results. Several studies indicate that more open economies should produce governments with
higher levels of spending on particularized benefits to protect workers from external economic
shocks (Cameron 1978; Rodrik 1998; Crepaz 1998; Shelton 2007), while other studies suggest
that greater openness should decrease spending on particularized benefits in order to create more
desirable economic conditions for producers (Swank 2010; Hay and Rosamond 2002). The
priorities model shows greater trade openness is associated with greater spending on collective
goods, represented by the positive coefficient and statistically significant for a non-directional
test at the 0.05 level. This finding suggests that more open economies actually spend less on
areas of social protection and supports the “race to the bottom” argument.
The final socio-economic variable examined in Table 5.1 is the effect of membership in
the European Union on spending priorities. The effect of the European Union conforms to
expectations that nations that belong to the European Union have higher levels of spending on
particularized benefits represented by the negative coefficient. The effect of the European Union
is negative and statistically significant at the 0.05 level. The result implies that European Union
member nations do behave differently from non-member nations in terms of spending and more
specifically, these nations spend more on areas such as social protection than non-member
nations.
Group Preferences and Spending Priorities
While the socio-economic climate can determine the level of resources governments have
to work with and what groups within the population may need particular services, the
preferences of both elites and the masses are also believed to shape government actions.

28

28

The

Prior studies that focused exclusively or predominately on different measure of preferences to
explain government outputs include: Erikson, Wright and McIver 1989; Hofferbert and Budge
106

expectations or demands of the masses include the role of government and public opinion, where
the beliefs in the role of government indicate the boundaries on issues the public feels the
government should be involved in and public opinion represents the desired actions by the public
on specific issues. Voter turnout and the resulting composition of the government produce
another set of preferences for government actions. The power resource literature argues that as a
larger portion of the working population mobilizes and votes the greater their power is in shaping
political actions. The resulting composition of government contains elite actors with their own
set of ideal political outputs.
A final set of demands can be found in organized interests. Two different approaches to
examining the effect of interest groups were tested. One approach looked at the level of
corporatism in a nation, where higher levels of corporatism correspond to nations that are more
directly influenced by pressure groups when making policy decisions. However, as the number
of pressure groups increase in a nation, it becomes more difficult for any group to speak on
behalf of the entire population it represents. The alternative approach I use to examine the effect
of interest groups is based on a factor analysis for the count of business associations and
government employees in a nation. More pressure groups should correspond to nations that are
29

less corporatist in nature and governments that are less influenced by pressure groups.

Table

5.1 shows the results using the density of pressure groups in a nation as this operationalization
provides more observations.
The expectation for government composition is that as the percentage of seats held by
leftist parties increases, governments should spend more on particularized benefits associated

1992; Page and Shapiro 1983; Garand 1985; Burstein 2006; and Penner, Blidook and Stuart
2006.
29
The correlation between the corporatism variable and the interest group variable is -0.5103.
107

with leftist party platforms like unemployment, pensions, and sickness benefits. The priorities
model shows that more leftist parties in office increase spending on collective goods that include
spending on education and health. However, the coefficient for government composition in the
priorities model examining fails to reach statistical significance for directional tests at the 0.05 or
0.10 level.
While voter turnout is predicted to influence spending priorities, no directional affect is
assigned. Prior literature indicates that increases in voter turnout may be associated with greater
portions of low-income individuals turning out who have expectations for greater particularized
benefit spending (Jackman 1987; Powell 1986; Crepaz 1998) or more high-income voters turning
out who prefer greater spending for collective goods spending compared to particularized
benefits (Lijphart 1997; Iversen and Soskice 2006). However, the mix in results may be a
product of the dependent variables that are used to study the effect of voter turnout. If a concept
was mislabeled to represent particular group spending like health or education it may have biased
the results, as the unfolding in Chapter 3 showed these to be collective goods. Using the
spending priorities variable that captures a range of policy areas and how expenditures are
connected between policies, shows that higher levels of voter turnout are associated with greater
spending on particularized benefits like disability benefits. This effect is represented by the
negative coefficients but fails to reach statistical significance at the 0.05 or 0.10 level for nondirectional test.
The lack of statistical significance suggests the mixed finding in the literature may be a
product of the dependent variable that are used and whether they represent particularized benefits
like social protection or collective goods like economic development. While voter turnout does
not appear to statistically influence the pattern of spending, it may influence the overall level of

108

total spending which is not captured by the priorities variable that I use. If this is the case testing
voter turnout in relation to separate policy areas would show similar changes in spending on
social protection and areas like economic development, producing the mixed findings on what
types of policies are favored as turnout increase.
Higher values of the expectations for government correspond to demands for a more
expanded role and are predicted to increase spending on particularized benefits as larger portions
of the population want greater government involvement in promoting aspects like equality.
Greater spending on collective goods is predicted when beliefs support a more limited view on
the role of government, like focusing on activities such as economic development and defense.
The priorities model in Table 5.1 show that preferences for more government action are
associated with higher levels of spending on particularized benefits like housing and food
subsidies which are represented by the negative coefficients and are statistically significant at the
0.05 level for directional tests.
More liberal preferences for government actions are predicted to increase government
spending on particularized benefits on areas like social protection. The results of the model
show more liberal opinions in society are associated with greater spending on particularized
benefits such a pensions, represented by the negative and statistically significant coefficient at
the 0.05 level for a directional test. This implies that as public opinion favors demand greater
government attention to issues such poverty and equality of outcomes governments spend more
on items to address these demands through areas like children benefits and housing vouchers.
Greater numbers of interest groups are expected to increase government spending on
collective goods, where larger numbers of interest groups would decrease the ability of any one
group to speak on the behalf of everyone else, making each group less influential. Higher values

109

of the interest group variable represent the presence of more pressure groups in a nation and
should produce a cacophony of demands that the government cannot accommodate. The interest
group variable should carry a positive sign indicating that more interest groups push government
spending towards collective goods that benefit all groups instead of spending that favors
particular groups.
The priorities model shows that as the density of interest groups increases, governments
spend more on collective goods represented by the positive signed coefficients. This finding
suggests that as more groups issue demands for spending, governments spend more on collective
goods that benefit broader communities like education and economic development than on
particularized policy areas. The effect of interest groups is statistically significant the 0.05 level
for a directional test.
Institutions and Spending Priorities
The final set of factors captures the political institutions present in a nation.

30

Presidential systems, majoritarian systems, low district magnitudes, bicameral legislatures, and
unitary systems are expected to increase spending on collective goods, while parliamentary
systems, proportional representation, high district magnitudes, unicameral legislatures, and
federal systems are expected to increase spending on particularized benefits. Table 5.1 shows
the effect of institutions in combination with the socio-economic climate, and mass and elite
preferences.
Expectations predict that presidential systems should produce governments that spend
more on collective goods. In a presidential system the executive is accountable to the entire
30

Prior studies that focused exclusively or predominately on institutional variables in relation to
government outputs include: Jackman 1987; Edwards and Thames 2006; Immergut 1990; Chang
2008; and Scartascini and Crain 2002.
110

nation and needs to ensure spending benefits the nation in broad terms to avoid upsetting voters.
In a parliamentary system the executive is accountable to the controlling party or coalitions and
will approve legislation that supports the controlling group’s particular interests. Table 5.1
shows that presidential systems are not found to influence spending priorities in a statistically
meaningful manner. Here the result may be a product of having only two nations that have
presidential systems in place (Poland and the United States). However, in a separate model run
looking only at the political institutions in place, presidential systems are shown to behave as
expected, where presidential systems spend more on collective goods than parliamentary systems
(Appendix B). Alternatively, this finding suggests that the expected relationship may be a
product of model misspecification in the literature that looks at political institutions in isolation
from other influential variables.
Electoral formulas that use proportional representation and mixed voting should spend
more on particularized benefits than nations than use purely majoritarian systems. Under
proportional representation candidates appeal to specific vote bases with policy platforms
targeted at particular group needs as opposed to majoritarian system candidates who use
collective goods to appeal to broader vote bases. Mixed systems use a combination a
proportional representation and majoritarian rules producing some candidates that target
particular groups and some candidates that need to appeal to broader vote bases. The priorities
model supports the expected relationships between electoral formulas and spending priorities.
Nations that use proportional representation spend more on particularized benefits like sickness
benefits and maternity leave than nations that use majoritarian systems, represented by the
negative coefficient that is statistically significant at the 0.05 level for a directional test. Mixed
voting systems also spend more on particularized benefits than pure majoritarian nations as

111

represented by the negative coefficient that is statistically significant at the 0.05 level for a
directional test.
As district magnitude increases, more candidates can win office with a smaller
percentage of votes, allowing candidates to target particular groups in the population. As more
candidates can be elected in a district the expectation is for governments to increase spending on
particularized benefits to cater to subsets of the population representing their constituencies. The
spending priorities model shows that higher levels of district magnitude are associated with
greater spending on collective goods such as economic development, as represented by the
positive coefficient and is statistically significant at the 0.05 level for a directional test. The
increase in spending on collective goods may imply that district magnitude effects post-election
behavior, unlike proportional representation, which affects pre-election behavior. Proportional
representation determines how candidates are elected, but district magnitude determines how
many incumbents need to agree before legislation can be enacted. As district magnitude
increases, more incumbents need to agree, meaning each incumbent can obtain less for their
constituents in the form of particularized benefits.
Prior literature does not lend specific expectation about the separate influence of
bicameralism and federalism on government spending patterns; however, the required agreement
between two separate houses in a bicameral legislature forces two separate entities to reach an
agreement before spending can occur. That, as I hypothesized in Chapter 4, should decrease
spending on particularized benefits that may serve the needs of members of one house at the
expense of the other, producing governments that spend more collective goods. The results of
the priorities model support the hypothesis where nations with bicameral legislatures spend more

112

on collective goods than nations with unicameral legislatures. The positive coefficient is
statistically significant at the 0.05 level for a directional test.
Federal systems are expected to spend more on particularized benefits than unitary
systems. The expectation is not based on behavioral patterns associated with incumbents in
federal systems but on the inability of federal systems to move quickly to change spending
allocations as found in the welfare state literature. As a result, federal systems lag behind unitary
systems in their effort to dismantle or decrease the scope of government programs and they end
up spending more on particularized policy areas. The findings of the spending priorities model
lends support to this hypothesis, showing that federal systems do spend more on particularized
benefits than unitary systems. The coefficient for the variable captures the organizational
structure of the governmental system (federal versus unitary) is positive and statistically
significant at the 0.05 level for a directional test.
Country Examples
What do the results of the model suggest for spending patterns within the nations
examined? The general finding is that the demands of the general public and groups in need like
the elderly or poor, the resources at the government’s disposal, and the institutions that
governments operate in all influence the resulting outputs of government spending. A more
specific example can be observed by looking at Greece’s current economic crisis and how
different factors could shape its future spending priorities. Greece has been faced with budget
cuts in order to receive its financial bailouts. Additional recommendations have been presented
by the OECD to address Greece’s debt. In order to reduce expenditures by the government,
among the many areas that could be changed, the OECD recommended reform of Greece’s oldage pension system including an increase in the effective age of retirement to 65 (Greece at a

113

Glance 2009). Changes to the retirement age would help decrease the proportion of individuals
receiving already generous pensions, where retirees can expect old-age pensions equivalent to
96% of previous earning compared to the OECD average of 52% (Greece at a Glance 2009).
By encouraging individuals to remain in the workforce longer, there should be a decrease
in spending attributed to particularized benefits that include pension benefits as fewer adults
would be drawing pensions. This would have a strong influence on Greece’s spending priorities
as the nation spends more on pensions than other OECD governments; in 2005, Greece spent
roughly 11.5% of its GDP on pensions compared to the OECD average of 7.2%. This reduction
alone would shift Greece’s spending priority closer to collective goods as there would be a
decrease in the size of the dependent population requiring assistance from the government during
retirement.
An alternative example comes from the United States. Economic problems in the United
States may see expenditures shift to favor greater relative spending on particularized benefits.
Since 2008, the United States has seen an increasing level of unemployment peaking at 10% in
2009 and as of May 2012 had not yet dropped below 8% nationally (Bureau of Labor Statistics).
High levels of unemployment have resulted in increases in the number of individuals who
require assistance from the government (in the form of unemployment benefits, housing and food
vouchers, and family and children benefits) in order to help meet their every day, basic needs.
The higher unemployment rate then should result in an increase in government
expenditures on particularized benefits and shift the United States priority score closer to that
end of the spending priorities continuum. In the last year of the sample data examined here, the
United States had an unemployment level of 4.6% for 2006; in 2010 the unemployment rate had

114

31

increased to 10%.

Holding all else constant, this change in unemployment would shift

government spending allocations in the United States by three percentage points towards
particularized benefits in. This would be the equivalent of a $1.6 billion shift in spending based
on 2008 total expenditures from collective goods (like education) in favor of greater spending on
social protection (such as unemployment benefits and housing vouchers). However, this output
is contingent on all other influential factors remaining constant which is not the case as
demonstrated by changes to public opinion with the emergence and rise of the Tea Party,
changes to the composition of the government with the election in 2010 and the approaching
2012 election, and shocks posed by external economies facing economic recessions.
OLD MODELS, NEW MEASURE
Having explored a range of indicators previously examined in the literature, I now turn to
how the results prior models compare to government spending priorities, assuming the old
model specification was valid, in order to look at how limited measures of prior dependent
variables altered findings. I examine the relationship between spending priorities and the factors
of influence used in two commonly cited works: one that focuses predominately on a limited set
of economic variables by Milesi-Ferretti et al. (2002) and one by Huber and Stephens (2001) that
uses a range of socio-economic, preference, and institutional variables.
In Electoral Systems and Public Spending, Milesi-Ferretti et al. (2002) examine the effect
of electoral systems in relation to the level of government transfers and spending on public
goods. Milesi-Ferretti et al. (2002) argue that proportional representation should produce
31

As will be discussed in more detail in Chapter 6, all independent variables in the model are
lagged by 2 years to ensure a more accurate representation of the expenditure process. As such,
2006 unemployment rates are used to predict the 2008 spending priorities variable and the 2010
unemployment rate would be used to predict the 2012 spending priorities value for the United
States.
115

governments that have higher levels of transfer spending and majoritarian systems should have
higher levels of spending on public goods. In order to capture government spending MilesiFerretti et al. use two main dependent variables, one representing government transfers, “defined
as the sum of social security payments and other transfers to families, plus subsidies to firms”
and the other public goods, “defined as the sum of current and capital spending on goods and
services” (629).
Regardless of the measure of government activity, the same four influences are controlled
for: district magnitude, the aged population, gross domestic product per capita, and OECD
membership. Milesi-Ferretti et al. (2002) use three different measures to operationalize the
degree of proportionality in a nation: average district magnitude, standardized district magnitude,
and the average deviation from proportionality (the average difference between the proportions
of seats each party holds versus the proportion of votes won); however, the models using the
three different measures of proportionality yield similar results in regards to coefficients’ signs,
magnitudes, and statistical significance. As a result, I only use average district magnitude to
measure proportionality when comparing findings from Milesi-Ferretti et al. (2002) to a
replication using the spending priorities variable.
Across the two models, higher district magnitudes are found to increase government
spending on transfers and decrease spending on public goods. The aged population was
repeatedly found to increase spending on transfers but not to have a statistically significant effect
on public goods. Gross domestic product per capita and OECD nations also showed mixed
results across the variety of models. In Table 5.2 I present the results from Milesi-Ferretti et al.
(2002) Model 4, Table V on “Primary Spending, Transfers, Public Goods, and Electoral Systems
(Full-Sample)” for transfer spending. I then run a similar model using spending priorities as the

116

dependent variable in Model 8 the Milesi-Ferretti et al. replication. However, as all the nations
used in my analysis are members of the OECD, I include the EU dummy variable as an
alternative.
Table 5.2

Replication of Milesi-Ferretti et al. Model using Spending Priorities

District Magnitude
Aged Population
GDP/Capita

a

OECD

a

Milesi-Ferretti et al.
Coefficient
(t-statistic)
p-value
1.70
(3.49)
1.25
(4.04)

**

Replication
Coefficient
(s.e.)
p-value

European Union
2

0.000

-1.91
(0.148)

2.19
(1.37)
1.20
(0.37)

0.314

-0.40
(0.091)

**

-0.05
(0.046)
-0.27
(0.027)

0.000

0.000

R
0.84
0.5490
N
40
369
Note: a Indicates that the natural log of the original variable was used in the
model.
** Indicates the coefficient is statistically significant at the 0.05 level;
based on
Model 8 uses OLS following Milesi-Ferretti et al. (2002).
The results from my replication of Milesi-Ferretti et al.’s (2002) model show similar
results to that of the model run by Milesi-Ferretti et al. where the transfer spending falls under
social protection and is associated with particularized spending in the spending priorities
variable. Higher district magnitudes in both models correspond to governments that spend more
on particularized benefits. In Milesi-Ferretti et al. (2002)’s model it is associated with greater
spending on transfers, and in my model it is associated with greater spending on items such as
pensions or unemployment benefits. In the replication, district magnitude fails to reach

117

statistical significance at the 0.05 level. The results of my replication imply that when examined
across a range of policies simultaneously, greater wealth, aged populations, and membership in
the European Union increase government spending on areas like social protection, whereas
looking solely at transfers produces inconclusive results.
Unlike the Milesi-Ferretti et al. (2002) model shown in Table 5.2, the socio-economic
variables included are all statistically significant at the 0.05 level, where only the proportion of
the aged population was found to affect the level of government transfers in the model from
Milesi-Ferretti et al. (2002). The larger proportion of the aged population, gross domestic
product per capita, and membership in the European Union are all found to produce government
that have higher levels of spending on particularized benefits as indicated by the negative
coefficients.
This work serves as an example of models that looks at a small set of influential factors
when examining government activities. The model only controls for three socio-economic
indicators and looks at a single institutional factor, proportionality. Correcting for the limited
dependent variable yields different results. Further, compared to the priorities model with the
appropriate model specification shows even more changes to the results. In the priorities model,
district magnitude shows nations spending more on collective goods which runs counter to
Milesi-Ferretti et al.’s argument. Additionally, all the socio-economic variables have a statically
significant effect on government spending patterns in the priorities model unlike Milesi-Ferretti
et al.
Another frequently referenced work regarding government spending patterns is Huber
and Stephens’s (2001) Development and Crisis of the Welfare State. Huber and Stephens (2001)
argue that the distribution of power affects both the creation and maintenance of strong welfare

118

states. Their work examines a number of measures to capture spending associated with welfare
states including: spending on pensions, health, mother’s employment and youth, a measure of
decommodification, poverty, inequality, and measures of redistribution.
Going beyond more simplified models, like those used by Milesi-Ferretti et al. (2002),
Huber and Stephens (2001) use more expanded specifications to capture welfare state activity.
Huber and Stephens examine a range of socio-economic influences, preferences of the electorate
and elites, historical ties, and institutional designs. As there are a number of models used by
Huber and Stephens (2001), here I look at their model for the effect of influential measures on
the public share of health expenditures and pension generosity (76).
Unlike Huber and Stephens (2001), I do not include a measure of military spending in my
replication, as defense spending is included in the measure of spending priorities. The number of
observations available for comparison to the work of Huber and Stephens is drastically limited
compared to other models I have run. The 90 observations are a product of using variables
coded by Huber and Stephens for authoritarian legacy and strikes. There are several countries
that do not overlap between the two samples of data as well as there being differences in time
points examined. In order to mirror the approach used by Huber and Stephens, my replication is
a time-series, cross-sectional analysis with panel-corrected standard errors (Table 5.3).
In the case of the spending priorities variable, spending on health and pensions are
divided between the two clusters of particularized benefits and collective goods. Spending on
pensions fall with particularized benefits as only a certain group within the population is eligible
to receive benefits from spending on this area. Spending on health services fall within collective
goods as the majority of nations that are examined have universal health care coverage. The
combination of these two policy areas into a single measure will result in increases in one area

119

Table 5.3

Replication of Huber and Stephens Model using Spending Priorities

Government Composition
Christian Democratic Cabinet
Constitutional Structure
Female Participation

Huber and
Stephens Model
Replication
Coefficient
Coefficient
(s.e.)
p-value
(s.e.)
p-value
0.47
**
-0.17
0.012
(0.073)
0.30
-0.01
0.001
(0.005)
-5.19
***
0.32
0.000
(0.086)
0.25
*
-0.14
0.042
(0.082)

Government Composition x
Female

-0.04

Voter turnout

0.09

Aged Population

1.22

Strikes

-0.01

Authoritarian Legacy

0.78

GDP/Capita

a

***

0.003
(0.001)
0.003
(0.009)
-0.02
(0.048)
0.001
(0.001)
-1.04
(0.163)

0.013

-0.61
(0.041)
0.01
(0.041)
-0.17
(0.028)

0.069

0.374
0.323
0.256
0.000

1.18

**

Consumer Price Index

-1.63

*

Unemployment

0.21

Military Spending

-0.58

FDI Out

-0.36

0.02
(0.007)

0.001

-0.03

-0.03
(0.003)

0.000

Openness

b

2

R

2

Adjusted R
N

0.452
0.000

0.9002
0.7
416

120

0.8806
90

Table 5.3

(cont’d)

Note: a Indicates that the natural log of the original variable was used in the model.
b Indicates that p-value for the variable is for a non-directional test, all other
variables
*, **, *** Indicates the coefficient is statistically significant at the 0.10, 0.05,
or 0.001 level for a directional test based on information in the original
work by Huber and Stephens (2001).
canceling out the decreases in spending that are occurring in the other. Increases in pensions will
increase spending on social protection and will take away resources that could be spent on
collective goods including health. Likewise, increases in health care spending will take away
resources from social protection that includes pension spending. Therefore, the results using the
additive spending on these two policy areas will be misleading in terms of how factors influence
higher or lower levels of spending on these two opposing policy areas.
In my replication of Huber and Stephens model, higher percentage of seats held by leftist
and Christian democratic parties are found to increase spending on particularized benefits. This
implies an increase in spending on social protection which includes pensions; however, it would
indicate a decrease in spending on health, which is again a collective good. However, in Huber
and Stephens’s model, Christian Democrats do not have a statistically significant affect on
spending for health and pensions. The lack of statistical significance in Huber and Stephens
model may serve as an example of how the two policy areas that target different groups cancel
each other out in terms of differences in spending.
Constitutional structure is an additive measure used to capture the number of institutional
constraints present in a nation. Huber and Stephens’ measure is based on the presence of the
following institutions: bicameralism, presidentialism, federalism, and referenda. The measure of
constitutional structure I use was defined in Chapter 4 and includes: presidentialism, proportional

121

representation, district magnitude, bicameralism, and federalism.

32

In both models, more

institutional constraints decrease spending on pensions which fall under particularized benefits in
the replication, and at a statistically significant level. However, my results show an increase in
spending on health compared to Huber and Stephens’s model which shows a decrease in
spending on health care.
Female participation in the workforce has a statistically significant effect across both
Huber and Stephens’s model and my replication. Increases in the proportion of women in the
workforce increase spending on particularized benefits and is at a statistically significant level
for a directional test. This again, though, implies a decrease in spending on health counter to the
Huber and Stephens’s model, which shows an increase in both types of spending. The
interaction between women in the workforce and left parties shows increases in the percentage of
left seats and women in the workforce collectively increase spending on collective goods and are
statistically significant for directional tests resulting in an increase in health care spending and a
decrease in social protection.
Gross domestic product per capita shows a different effect across both models in Table
5.3. In Huber and Stephens’s models higher levels of gross domestic product per capita are
associated with greater spending on pensions and health at a statistically significant level for a
directional test. In my replication, higher level of gross domestic product per capita is associated
with greater spending on particularized benefits that includes pensions but less spending on
health, and is statistically significant for a directional test.
The effect of unemployment found in Huber and Stephens’ model is again partially
supported by my replication. In both models, higher levels of unemployment increase spending
32

I omitted referenda from the institutional constraint index as it was found to decrease the
reliability of the index.
122

in the dependent variable. In both cases, the effect of unemployment is statistically significant
for a directional test for Huber and Stephens this would be for both pension and health where I
find an increase in social protection but a decrease in health. Further, in both models, voter
turnout, the aged population, and strikes fail to reach statistical significance.
Increases in the consumer price index (CPI) are similar across the two models, where
increases in the CPI are associated with greater spending on collective goods or less spending on
pensions and health services. This finding is statistically significant across both models for a
directional test. Therefore, as goods and services become more expensive nations provide less in
terms of spending for particular groups in society capturing social protection but my model
would imply a relative increase in health expenditures.
However, unlike Huber and Stephens’s model, an authoritarian history was found to be
statistically significant. Nations that had an authoritarian legacy in the replication spend more on
particularized benefits like pensions than the nations that did not have an authoritarian legacy.
Another difference was in the effect of foreign direct investment outward (FDI out). In Huber
and Stephens’s model FDI out was not found to be statistically significant. However, in the
replication, higher levels of FDI out are associated with greater spending on collective goods
which include health, at a statistically significant level for a directional test. Trade openness is
also found to affect spending priorities in the replication, as opposed to Huber and Stephens’s
model. Higher degrees of economic openness are associated with greater spending on
particularized benefits which includes pensions, which is at odds with the priorities model run in
this chapter adding to the mixed findings in the literature on the role of globalization and its link
to government activity.

123

The results from my replication model are similar to Huber and Stephens only in terms of
how the variables affect pensions, as health is shown to target a different group in society and
moves in an opposite direction to social protection in terms of increases and decreases in
expenditures. In Huber and Stephens’s model, the researchers combined spending based on the a
priori assumption that spending on health and pensions were similar in nature and could serve as
a measure of welfare state spending. However, the unfolding in Chapter 3 shows that the two
policy areas are different in nature based on the clustering of policy areas. The use of the
spending priorities variable then allows for a better understanding of the actual spending
allocations of nations on a priori notions about what policy areas ought to belong together.
CONCLUSION
Compared to previous research, the findings in this chapter produce a richer explanation
of the factors that influence the policy outputs of democratic political systems. Studies that use
individual measures and composite measure produce conflicting and contradictory results
pertaining to the influence of globalization and voter turnout in relation to spending patterns.
The priorities model improves upon previous research that examines how and why
certain policy areas fit together. For example, Huber and Stephens (2001) combine social
protection and health expenditures to represent welfare spending, whereas the unfolding in
Chapter 3 shows the two policy areas represent two different forms of expenditures:
particularized benefits and collective goods. Instead of allowing the a priori assumptions to
determine which policy areas belong together, the unfolding model allows the data to determine
how the policy areas are actually related to one another. This approach produces a better
understanding of the factors that influence government spending patterns based on the empirical
data.

124

CHAPTER 6
INSTITUTIONAL CONSTRAINTS AND POLICY RESPONSIVENESS
While individual political institutions are shown to influence government spending
behaviors, the literature also suggests that national political institutions affect the ability of
governments to reach policy agreement. If this is true, then national institutions exert a
cumulative effect on government spending priorities, representing policy compromise across a
range of public programs.
A number of political institutions are said to have separate affects on government
spending patterns. Presidential systems are argued to promote executives who pursue more
collective policies compared to parliamentary systems (Lijphart 1999; Tabellini 2000; Persson
and Tabellini 1999). Proportional representation increases spending on particularized policy
areas that benefit candidates’ constituent bases relative to majoritarian systems (Persson and
Tabellini 1999; Austen-Smith 2000; Cox and McCubbins 2001; Iversen and Soskice 2006;
Milesi-Ferretti et al. 2002). And district magnitude, produces candidates who represent the
interests of subsets of citizens and favor spending that serves these groups (Hill and Andersen
1995; Persson et al. 2007; Milesi-Ferretti et al. 2002). Previous research has shown that these
institutional features do influence the spending patterns of nations. But these features are also
part of a set of institutions within a nation, referred to as institutional constraints, that increase
(or decrease) the ability of governments to reach policy agreements, impeding or inhibiting the
process of government action.
Institutional constraints are described as systems or rules that separate decision making
power across different actors within a political jurisdiction. They include: presidential systems,
proportional representation, high district magnitudes, bicameral legislatures, and federal systems.
Briefly, presidential systems divide policy objectives between the executive and the legislature in
125

which the president is responsible to the entire nation and represents one set of interests, and
members of the legislature represent smaller constituencies with different policy preferences.
Proportional representation and high district magnitude increase the number of actors with a vote
in the legislature that can influence government actions. Bicameral systems separate decision
making power across two houses that need to reach agreement before legislation can be enacted.
Federal systems divide the decision making power on a range of issues across various levels of
government, increasing the difficulty of achieving coordinated government action.
However, the variation in spending priorities within nations over time shows that
regardless of the number of national institutional constraints, governments are able to reach
policy agreement and establish clear spending priorities, and also make changes to these
priorities over time. This point can be observed initially by looking at the range of spending
priorities over time across all nations in Figure 6.1, which includes nations that have as few as
zero constraints (e.g. the United Kingdom) and as many as four institutional constraints (e.g.
Germany). A more explicit illustration of the variation in spending priorities can be observed by
looking exclusively at the nations that have three or four institutional constraints, which are
shown in Figure 6.2. Even nations exhibiting the maximum number of constraints show
spending priorities that change over time. For example, Belgium, which has a proportional
electoral system, a district magnitude of 13.63, a bicameral legislature, and a federal system, not
only reaches policy outputs captured by the priorities measure, but also shows variation over
time, with a spending range equivalent to roughly €4.5 billion in 2008.

126

Figure 6.1

Distribution of Spending Priorities by Nation over Time

Greece
Germany
Denmark
Italy
Belgium
Sweden
Finland
Austria
Netherlands
France
Poland
Luxembourg
Hungary
Slovenia
United Kingdom
Canada
Spain
Noway
Ireland
Slovakia
Japan
Czech Republic
United States
Iceland

50

52

54
56
Spending Priority

58

While prior literature on institutions suggests how these constraints increase policy
resoluteness—the inability of governments to enact policies different that the current status quo
(Tsebelis 1995, 2000; Cox and McCubbins 2001; Shugart and Haggard 2001; Immergut 1990,
2010)—history and current events show that even nations with multiple institutional constraints
are able to reach legislative agreements. For example, Germany has four institutional
constraints: proportional representation to elect a number of its legislators, an average district

127

magnitude of 3.6 representatives per district, a bicameral legislature, and a federal system, and is
still in a position to react to problems domestically and internationally. This leads to my final
question: How do institutional constraints alter government outputs?
Figure 6.2

Distribution of Spending Priorities over Time for Nations with Three
or Four Institutional Constraints

THE ROLE OF INSTITUTIONAL CONSTRAINTS
Prior research argues that as the number of institutional constraints increases, more actors
with preferences over policy outputs enter the decision making process (Tsebelis 1995, 2000;

128

Cox and McCubbins 2001; Shugart and Haggard 2001; Immergut 1990, 2010). As more actors
have a vote on policy outputs, the area of consensus or area of overlapping policy preferences
gets smaller, making it more difficult to reach an agreement. This result is referred to as policy
resoluteness (Cox and McCubbins 2001). But the question left untested by previous work is how
policy decisions that are reached are altered by the institutional constraints.
My argument for the influence of institutional constraints arises from the policy
resoluteness framework. Each additional institutional constraint results in more actors—each
possessing often divergent ideal policy outputs they would like to see implemented by the
government—participating in the policy process. Each actor prefers policy outputs as close as
possible to own ideal results. As more actors enter the decision making process, it becomes
more challenging for any actor to shift policy in a direction that favors his or her ideal interests.
In order to reach agreement when multiple actors with preferences over policy outputs are
present, negotiations and bartering over policy choices will occur.
Actors involved in the decision making process want to maximize their gains and
minimize their losses for the groups they represent. Phrased slightly differently, each actor does
not want other groups to get more than his or her own group. Therefore, when there are multiple
preferences for policy outputs, negotiating and bartering should result in greater spending on
collective goods that provide at least a small benefit to all parties involved and less spending on
particularized policies that may give zero benefits, or even penalize certain groups.
H3a: Nations with more institutional constraints spend more on collective goods, relative
to nations with fewer institutional constraints.
An extension of the above argument involves the role preferences play in shaping the
spending priorities of government. I argue that constraints should not only have a direct effect

129

on government actions, but should also reduce the ability of governments to incorporate both
mass and elite preferences into policy. If constraints induce bartering and compromise over
policy outputs, everyone should get less of what they want. Each group will have to forgo some
portion of their ideal outcome in order to get any portion of the outputs they desire. Therefore,
no group gets everything it would like and policy is less responsive to all parties involved in
order for anyone to get anything. Responsiveness is defined here as “the degree to which policy
choices follow public preferences” (Roberts and Kim 2011).
H3b: Increasing the number of institutional constraints decreases the policy
responsiveness to different groups’ preferences for government outputs.
MODEL
In order to test these two hypotheses I begin with the spending priorities model from
Chapter 5 explaining levels of national spending across policy areas. I replace the individual
variables representing national political institutions with an additive index representing the total
number of institutional constraints within a nation. (Table 6.1)

33

This variable allow for testing

of the direct effect of constraints on spending priorities. In other words, rather than testing each
institution separately, this model conceptualizes institutional constraints as interchangeable in
order to test the hypothesis that more constraints of any type should produce greater spending on
collective goods relative to particularized benefits. Support for my first argument is found if the
coefficient for the institutional constraints variable is positively signed and statistically
33

Several variables that are statistically significant in the priorities model are not statistically
significant in the initial constraint model: trade openness, inflation, female participation rates,
and voter turnout. In order to ensure a parsimonious model, I tested the explained variance of
the constraint model against the priorities model. The empirical F-statistic calculated based on
priorities model and constraint model is 0.827 which is less than the critical value of 2.40 at the
95% level, meaning I fail to reject the null that the priorities model explains more variance in
than the constraints model.
130

Table 6.1

Traditional Influences of Government Spending Priorities
Priorities Model
(Model 1)

GDP/Capita

Coefficient
(s.e.)

a

Unemployment

a

Aged Population
European Union
Openness

b

Inflation Rate
Female Participation
Government Composition
Role of Government
Public Opinion
Interest Groups
Voter Turnout

b

Institutional Constraints
2

p-value

Constraint Model
(Model 2)
Coefficient
(s.e.)
p-value

-1.05
(0.109)

0.000

-0.99
(0.080)

0.000

-0.72
(0.110)
-0.24
(0.030)
-1.61
(0.124)

0.000

-0.73
(0.086)
-0.23
(0.012)
-1.65
(0.079)

0.000

-0.001
(0.001)
-0.002
(0.013)
0.01
(0.009)
-0.01
(0.002)
-0.01
(0.003)
-0.01
(0.002)
0.56
(0.045)

0.541

-0.01
(0.003)
-0.01
(0.002)
-0.01
(0.002)
0.53
(0.031)

0.005

0.01
(0.006)
-0.20
(0.04)

0.201
-0.19
(0.023)

0.000

0.000
0.000

0.000
0.000

0.442
0.284
0.002
0.005
0.000
0.000

0.000

0.000
0.000
0.000

R
0.7265
0.7238
N
367
367
Note: a Indicates that the natural log of the original variable was used in the
model.
b Indicates that p-value for the variable is for a non-directional test, all
other variables

131

significant, representing greater spending on collective goods, such as economic development
and education that benefit the society in broader terms, as the number of institutional constraints
increases.
Then, I examine whether or not institutional constraints reduce the ability of governments
to respond to mass and elite preferences. Here, I specify an interaction model that includes
interaction terms to determine the relationship between institutional constraint and the four
different measures of elite and mass preferences (government composition, role of government,
public opinion, and interest group density). If the number of national institutional constraints
decreases the policy responsiveness of governments to preferences, then the interaction terms
should carry the opposite sign of each preference thereby decreasing the cumulative effect of the
preference in influencing government spending priorities.
In the interaction model, I use a time-series, cross-sectional analysis with panel-corrected
standard errors.

34

I use a time-series approach in order to ensure that the relationships between

the variables hold over time and are not a reflection of a specific time point. It has been argued
that a period of at least ten years is preferred when examining patterns in policy areas (Kingdon
1984; Baumgartner and Jones 1993). The unbalanced panel dataset covers the time period from
1990 through 2009. The shortest time period in the analysis is for Poland, with seven years, and
the longest time period is for Luxembourg and Denmark, at twenty years each. The average
number of nation-years for a nation in the sample is fifteen years. Figure 6.3 shows the number
of years of data for each nation.

34

See Appendix C for an explanation of the diagnostic tests performed to assess non-linearity in
the independent variables, multicollinearity, heteroskedasticity and influential observations, and
corrections that were made in response to these conditions.
132

Figure 6.3

Number of Years by Nation in the Panel Data

Instead of looking at only one nation’s spending over time, I use a cross-sectional
approach. By examining multiple nations over time, I am able to study factors that are constant
over the time period within nations but vary across nations. The examination of the institutional
constraints variable requires a cross-sectional approach, as none of the nations in the sample
experience a change in the number of constraints for the time period in the analysis.

133

Due to the nature of the data, traditional OLS assumptions regarding the error process
may be problematic. The observations may not have constant error variance (heteroskedastic
errors). Furthermore, as the variables observed are at the nation level, there is the risk that the
interactions among groups of nations may influence data points within other nations; for
example, nations that frequently interact as members of the European Union may influence each
other on policy areas relating to unemployment (the problem of spatial correlation). In order to
address these two issues, I use panel-corrected standard errors. Panel-corrected standard errors
assume that the variance of the error term is not constant and that the variance of the error term
across nations may be related (Beck and Katz 1993).
Another concern to address at the outset of this analysis is serial correlation, where the
errors for each observation are correlated with each other over time. Situations where serial
correlation exist and are not addressed can produce inflated t-values and deflated standard errors,
leading to overly confident estimates and Type I errors in which the null hypotheses concerning
the coefficients are wrongly rejected. I use Wooldridge’s (2002) test for first order correlation in
panel data and reject the null hypothesis that there is no first order correlation in the constraints
model. In order to address this issue, I re-specify the interaction model assuming an AR1
process (Model 4). The results between the interaction model and the model with the AR1
process are similar for most variables, except the constraint variable by itself that fails to reach
statistical significance at the 0.05 level for a directional test; however, as the results between the
two models are similar in terms of signs, magnitude, and statistical significance, I discuss the
results from the interaction model that are more consistent with the approach used in the
literature to examine government spending patterns.

134

Fixed and Random Effects
When dealing with cross-sectional data there are factors unique to each case (i.e. nations)
that may influence the regression results. Under certain situations using fixed effects may help
address some of the issues of case uniqueness. Fixed effects assume that there are omitted
variables α, but that these variables are time-invariant (αi), or constant across time. Fixed effects
essentially create separate intercepts for each of the cases in the data, but slopes for each of the
independent variables are assumed to be constant across cases. The separate intercepts are used
to control for the factors unique to the cases across time.
While fixed effects may help address factors unique to the cases, I do not specify a fixed
effects model because the process creates an issue for the political institutions under study in the
analysis. The institutional constraints and the European Union variables included in the analysis
are time-invariant for all nations during the time period in the analysis. In this situation, when
estimating a fixed effects model, the institutional constraints and the European Union variables
would be omitted and the effects would be grouped with unobserved country specific factors.
Even if there were a few instances where the institutional constraints and the European Union
variables changed within nations, the estimates would be based on the few instances of change
and would result in imprecise estimates of the effect of the institutions under study. For
example, if the institutions were examined separately, small changes such as Italy’s switch from
proportional representation to a mixed voting system in 1994 would incorrectly drive the
estimates of the institutional constraint coefficient.
Furthermore, a number of the variables in the data that do vary over time mostly occur
across cases rather than over time. Cameron and Travedi (2005) note that when variation is
cross-sectional instead of over time, estimates using fixed effects will also be imprecise.
135

Because of the nature of the variables under study and the conflict posed by using fixed effects, I
do not use fixed effects in the model.
Another common modeling approach to examine panel data is random effects. Unlike
fixed effects, which assumes α is time-invariant and correlated with the variables in the model,
random effects assumes that factors unique to the country α, are not correlated with the observed
independent variables, x. Under random effects, α is assumed to be independent and identically
2

distributed with mean of zero and a variance of σ . If however, α is correlated with x, then the
coefficient estimates will be inconsistent (Cameron and Travedi 2005).
I do not specify a random effects model because the variables included in the analysis
include aspects of societies that are highly likely to be correlated with factors omitted from the
model. For example, the estimates of public opinion are related salient political issues, such as
focusing events that are omitted from the model, which in turn will affect policy priorities.
Therefore, using random effects with the foreknowledge that measures omitted from the model
are correlated with included variables would be to produce inconsistent parameter estimates
intentionally.
RESULTS
Table 6.2 shows the results of the constraint model that include the interaction terms
between the number of national institutional constraints and the four different measures of elite
and mass preferences. I refer to this model as the interaction model. The inclusion of the
interactions between the institutional constraints variable with the preferences measures
increases the variance explained in the dependent variable to 79.71%, a statistically significant
increase from the constraint model omitting the interactions which explains 72.38% of the total

136

Table 6.2

The Effect of Institutional Constraints on Policy Responsiveness
Interaction Model
(Model 3)

GDP/Capita

Coefficient
(s.e.)

a

Unemployment

a

Aged Population
European Union
Institutional Constraints (IC)
Government Composition
IC x Government Composition
Role of Government
IC x Role of Government
Public Opinion
IC x Public Opinion
Interest Groups
IC x Interest Groups
2

p-value

AR 1 Model
(Model 4)
Coefficient
(s.e.)

p-value

-0.89
(0.090)

0.000

-0.67
(0.165)

0.000

-0.52
(0.062)
-0.28
(0.012)
-1.64
(0.067)
0.53
(0.118)
0.02
(0.005)
-0.01
(0.003)
-0.03
(0.008)
0.01
(0.007)
-0.06
(0.008)
0.02
(0.003)
2.22
(0.157)
-0.59
(0.050)

0.000

-0.42
(0.116)
-0.29
(0.034)
-1.63
(0.163)
0.23
(0.165)
0.01
(0.006)
-0.01
(0.003)
-0.03
(0.012)
0.01
(0.005)
-0.04
(0.011)
0.01
(0.004)
2.09
(0.348)
-0.56
(0.128)

0.000

0.000
0.000
0.000
0.000
0.000
0.000
0.026
0.000
0.000
0.000
0.000

0.000
0.000
0.079
0.011
0.018
0.004
0.050
0.000
0.001
0.000
0.000

R
0.7971
0.9963
N
367
367
ρ
0.7570
Note: a Indicates that the natural log of the original variable was used in the
model.

137

variance in government spending.

35

Positive coefficients represent greater spending on collective goods such as defense and
economic development relative to particularized benefits such as social protection and foreign
aid. Negative coefficients represent greater spending on particularized benefits relative to
collective goods.
Controls
Table 6.2 omits the coefficients for the year dummies. The year dummies do not carry
directional hypotheses and were included to control for time trends. A joint F-test of the year
dummies shows that the effect of year dummies was statistically significant at the 0.05 level with
2

a χ value of 5191.52 and a p-value of 0.000. The positive values on the year dummies indicate
that each year nations spent more on collective goods like defense and economic development
compared to nations in 1990.

36

After 1990, the nation dummies carry positive coefficients

indicating that each year in the model, relative to 1990, spent more on collective goods. This
shift to greater spending on collective goods relative to 1990 suggests support for a retrenchment
of the welfare state argument, which argues nations moved to reduce spending on welfare items
like housing vouchers, unemployment benefits, and old-age pensions over the past two decades
(Pierson 1996; Starke 2006).

35

The empirical F-statistic for the interaction model compared to the constraint model is 28.89
which is larger than the critical F-statistic of 2.37 necessary to reject the null hypothesis that the
interaction model does not explain more variance in the spending priorities variable than the
constraint model.
36
An alternative specification using a single variable to express the year yields similar results to
using the year dummy in terms of signs, magnitudes, and statistical significance of the
independent variables. In this form, the year variable carries a positive coefficient representing
nations spending more on collective goods each successive year after 1990, which follows the
retrenchment argument in the literature.
138

The traditional socio-economic factors concerning wealth, the unemployment rate, and
the aged population all exhibit the predicted relationships with government spending priorities
outlined in Chapter 4 and supported in the spending priorities model in Chapter 5. Greater
wealth is associated with governments that have more resources to spend on particularized
benefits like family and children’s benefits without decreasing the goods and services provided
to society more broadly. In the interaction model (Model 3), higher levels in the natural log of
gross domestic product per capita are associated with governments that spend more on
particularized benefits, such as pensions and spending on families and children represented by
the negative coefficients. The relationship is statistically significant at the 0.05 level for a
directional test.
Additionally, as the unemployment rates increases, the expectation was that governments
spend more on services addressing the needs of the unemployed, such as unemployment benefits
or housing vouchers. The coefficient shows increases in the natural log of the unemployment
rate are associated with greater spending on particularized benefits confirms this hypothesis.
The proportion of the aged population in a nation should increase government spending
on particularized benefits that address the needs of the elderly population. The interaction model
shows that as the percentage of the aged population increases, governments spend more on
particularized benefits like unemployment. The effect of the aged population is based on the
negative coefficient and represents greater spending on particularized benefits and is statistically
significant for a one-tailed test at the 0.05 level.
Nations belonging to the European Union are predicted to behave differently in terms of
spending patterns based on their integrated economies as noted by the OECD, “common policy
goals regarding economic growth, agriculture, energy, infrastructure, and research and

139

development (among others) may also affect the structure of expenditures” (At a Glance 2009).
The interaction model indicates that member nations of the European Union spend more on
particularized benefits, including aspects such as housing assistance and unemployment benefits,
compared to non-member nations of the European Union. This relationship supports the
hypothesis for European Union nations outlined in Chapter 4 and is statistically significant for a
directional test at the 0.05 level.
Institutional Constraints
As the number of national constraints increase, it should be harder for any group to obtain
its ideal spending priorities. This expectation extends previous work showing constraints alter
government spending priorities, specifically, increasing spending on collective goods relative to
particularized benefits. My argument is based on the effect of increasing the number of actors
involved in the decision making process. As the number of actors increases, single preference
holder should find it more difficult to increase spending on policy areas that may benefit their
group at the expense of others. Therefore, more national institutional constraints should be
associated with spending on policy areas that benefit the entire society at least somewhat versus
particularized areas that only benefit parts of it. The positive and statistically significant
coefficient for a directional test at the 0.05 level in the interaction model, which includes the
additive index based on the number of institutional constraints present in a nation and the
interaction between the number of constraints in a nation with different measures of elite and
mass preferences, lends support to the hypothesis that more institutional constraints increase
government spending on collective goods.
The results from the interaction terms involving each of the four measures of elite and
mass preferences provide additional insights on the question presented at the beginning of the

140

chapter: Do the number of institutional constraints alter government policy outputs? The
interaction model demonstrates that governments are less responsive to both mass and elite
preferences as the number of institutional constraints present in a nation increases. Again,
responsiveness is measured as the degree to which policy outputs match expectations and
demands on government, which are captured here by the composition of political parties’ in
office, public opinion, general beliefs about the role of government and interest group strength.
My expectation is that as the number of institutional constraints increases and thus more groups
exist that must reach agreement, bartering will intensify and each group will obtain less spending
on any particular policy area than they would if they had sole discretion over spending. The
resulting increase in bartering and compromise produces less responsiveness, as policy outputs
will not match the demands of different groups. The results of the interaction terms show that
regardless of the measure of preferences, increasing the number of institutional constraints
decreases the degree to which governments respond to groups’ expectations; measured as the
cumulative effect of each preference based on the number of existing constraints.
The total effect of institutional constraints on government composition indicates that
increasing the number of institutional constraints decreases the effect of left parties in
government on spending priorities. Previous research suggets that more seats held by leftist
party members is associated with greater spending on particularized benefits, which conforms to
prior expectations. However, when I include the interaction term between government
composition and institutional constraints, in the presence of zero institutional constraints, the
results show that governments spend more on collective goods. The interaction term for the
coefficient is statistically significant at the 0.05 level for a directional test.

141

What this result may indicate is that when leftist parties have fewer actors to appease in
the decision making process they are able to spend on a range of policy areas that support their
party platforms, including spending on education and health. However, as the number of
constraints increase requiring more bartering to pass legislation, leftist parties forgo policy areas
that are more collective in nature and fight for spending on policies that would receive less
spending without leftist parties, like social protection.
Table 6.3

Government Composition and Spending Priorities in the United
Kingdom

Year
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008

Percentage Spending
of Left
Priority
Seats
Score
39.28
54.49
39.28
54.00
39.28
53.67
39.28
53.36
39.28
53.23
45.64
53.05
45.64
53.10
45.79
52.95
46.42
52.95
46.73
53.34
72.61
53.58
72.61
53.80
72.61
54.08
72.61
54.27
73.00
54.28
73.00
54.36
73.00
54.56
73.00
54.50
66.83
54.88

The result under zero institutional constraints is highlighted by examining the United
Kingdom during the time period examined (Table 6.3). From 1990-97, increases in the
142

percentage of seats held by leftist party members are associated with greater spending on
particularized benefits. During this period the left was not in control of the government and the
best the party could do was fight for a few of their causes. When Prime Minister Tony Blair took
office in 1997, the Labour party presented a new party manifesto to the public that focused on
education, crime, health, jobs and economic stability (Labour.org.uk). Four of the five focuses
of the traditionally leftist party correspond with collective goods policies, namely education,
public order and safety, health and economic development, and with control over the parliament,
the left was able to shift priorities to match their full policy agenda. With control of parliament
and no institutional constraints to promote compromise, the left could implement its full policy
agenda.
The spending pattern of leftist parties, in the presence of zero institutional constraints,
was similarly demonstrated in the American state context (Alt and Lowry 2000). In unicameral
states when Democrats held office, there is greater spending on the public sector than when
Republicans held office. However, in a bicameral systems the constraint, “… induces bargaining
when different parties control the legislature and executive” (Alt and Lowry 2000, 1039). The
product of bargaining forces the parties in office to choose among the policy objectives that are
feasible for it to achieve.
Regardless of the spending priority under zero institutional constraints, as the number of
constraints increases, government policy responsiveness to party preferences decreases in
relation to spending priorities. This result is shown by the negative coefficient for the interaction
term between government composition and the institutional constraint variable and is statistically
significant at the 0.05 level for a directional test.

143

Figure 6.4 provides a visual display of the effect of government composition under zero
and four institutional constraints, respectively. Holding all else constant, when a nation has four
constraints, leftist parties are unable to obtain the same level of spending on policy areas as they
would under zero constraints. For example, moving from a nation with zero institutional
constraints (e.g., the United Kingdom) and a legislature dominated by leftist parties, switched to
a nation with four institutional constraints (e.g., Germany) holding all else constant, the model
predicts a 3% shift in spending from collective goods to particularized benefits; 3% is equivalent
to roughly 170 billion 2008 US dollars.
Figure 6.4

Predicted Spending Priorities for Government Composition
Collective Goods

Particularized Benefits

Note: Predicted values across the range of government composition from left parties
holding between 0 and 100 percent of the seats in the legislature. Based on
coefficients from the interaction model, with the remaining independent variables
held at their respective mean values, for the year of 1995.
144

Particularized benefits are expected to increase when the population favors more
expansive roles for government involvement in a society (higher values on the role of
government variable) and increase spending on collective goods when views favor a more
limited role focusing on defense and the economy (lower values on the role of government
variable). In the interaction model, when there are zero institutional constraints, the negative
coefficient for the role of government indicates governments spend more on particularized policy
areas such as pensions and unemployment as the proportion of the population favoring more
expanded roles of government is higher and is statistically significant at the 0.05 level for a
directional test. Therefore, as the public expects government to be involved in areas such as
promoting equality and creating a humane society, governments spend more on particularized
policy areas like housing, and family and child benefits to address issues in these areas.
As the number of national institutional constraints increase, however, the effect of
government expectations diminishes, which is represented by the positive coefficient for the
interaction term between role of government and institutional constraints. As the number of
constraints increases, the cumulative effect of beliefs regarding government on spending
priorities decreases and is statistically significant at the 0.05 level for a directional test. The
decrease in government policy responsiveness to expectations can be examined graphically in
Figure 6.5, which depicts the predicted spending priorities across the range of expectations for
government under zero and four institutional constraints, respectively. When there are zero
institutional constraints, there is greater government responsiveness to beliefs compared to when
there are four constraints. The decrease in policy responsiveness to public perception for the role
of government under four constraints can be seen by comparing the absolute slopes under zero
and four constraints.

145

Figure 6.5

Predicted Spending Priorities for Role of Government

Collective Goods

Particularized Benefits

Note: Predicted values across the range of expectation from societies dominated by
restricted views on governments role (-100) to expansive views on the role of
government (100). Based on coefficients from interaction model, with the
remaining independent variables held at their respective mean values, for the year
of 1995.
In nations with more liberal publics, the interaction model shows that in the presence of
zero institutional constraints, government spending priorities favor spending on particularized
benefits, indicated by the negative coefficient. The relationship between spending priorities and
public opinion supports the hypothesis from Chapter 4 and is statistically significant at the 0.05
level for a directional test. The result implies that when governments are not forced to barter to
pass legislation they can better represent and incorporate the preferences of public opinion into

146

policy expenditures.
Figure 6.6

Predicted Spending Priorities for Public Opinion

Collective Goods

Particularized Benefits

Note: Predicted values across the range of public opinion from societies dominated by
right political priorities (-100) to societies dominated by left priorities (100).
Based on coefficients from interaction model, with the remaining independent
variables held at their respective mean values, for the year of 1995.
Similar to government composition and expectations regarding the role of government, as
the number of institutional constraints increases, government responsiveness to public opinion
decreases. The diminished response to public opinion is shown in the positive interactive
coefficient between the number of institutional constraints and public opinion. As the number of
institutional constraints increases, holding all else constant, the total effect of public opinion on
government spending priorities decreases. The interaction term is statistically significant for a
directional test at the 0.05 level. Figure 6.6 depicts the responsiveness of government to public
147

opinion when there are zero and four institutional constraints, respectively. The absolute slope
under zero institutional constraints is larger than when there are four constraints, implying a
diminished role for public opinion in shaping spending priorities compared to when there are
zero constraints. The interaction term shows that as the number of institutional constraints
increase public opinion, like the preferences of the actors themselves will play a smaller role in
influencing the final policy outputs of government as a result of bartering and negotiations.
The final variable examined in the interaction model is the role of interest groups in
shaping spending priorities. Higher densities of interest groups in nations are expected to
increase government spending on collective goods; more interests mean more preferences to
appease, which overwhelm the system such that no group ultimately gets what it wants. When
there are zero institutional constraints in a nation, the interaction model shows that higher levels
of interest group density are associated with greater spending on collective goods, indicated by a
positively signed coefficient. The effect of interest groups, when there are zero constraints, is
statistically significant at the 0.05 level for a directional test. This finding supports the
hypothesized relationship between interest groups and government actions. In nations where
fewer pressure groups exist, it is easier for government to respond to these groups’ demands. As
the number of interest groups increases, however, it becomes difficult for governments to
respond to the multiple groups because resources are limited. Additionally, it is more
challenging to determine what organized interest is speaking for what sets of individuals in the
public.
As the number of constraints increases, the effect of interest groups on spending priorities
is diminished, represented by the negative coefficient for the interaction term between interest
group density and institutional constraints. Decreased government responsiveness to interest

148

Figure 6.7

Predicted Spending Priorities for Interest Groups

Collective Goods

Particularized Benefits

Note: Predicted values across the range of interest group density from low levels of
density (-1) to high levels of interest group density (3). Based on coefficients
from interaction model, with the remaining independent variables held at their
respective mean values, for the year of 1995.
group density can be seen in Figure 6.7. The decrease in the absolute slope from zero constraints
to four constraints demonstrates the diminished policy responsiveness to organized interests as
the number of institutional constraints increases. The smaller slope could also be interpreted as
interest groups acquiring greater spending on the particularized benefits they desire. Institutional
constraints create access points to more decision makers that the organized interests can target
and push their spending agenda upon; however, as the graphs of spending priorities show, the
effect of interest groups at four constraints is not statistically different than zero, and it may

149

actually indicate that governments cease to be influenced by interest groups as more actors with
preferences over spending enter the decision making process.
IMPLICATIONS FOR NATIONAL SPENDING PRIORITIES
What does the interaction model imply for national spending priorities? In order to
understand how institutional constraints can alter national spending priorities, I begin by
examining the priorities within the United Kingdom. Over the time period from 1990 through
2008, the United Kingdom saw its spending priorities influenced by a range of factors favoring
spending on both collective goods, such as education and health, and factors fostering greater
spending on particularized benefits, such as housing vouchers.
Table 6.4

Observations for the United Kingdom from 1990-2009 used in the
Interactions Model

GDP/Capita
27,210.54
26,144.00
29,137.02
29,065.17
29,096.55
25,326.70
26,627.25
28,175.56
28,770.50
31,261.40
32,881.25
33,090.05
31,369.53
30,252.46
32,519.36
36,552.37
41,922.89
41,736.57
42,987.19

Unemployment
Aged
Government
Role of
Public
Rate
Population Composition Government Opinion
9.00
15.51
39.28
-5.21
-3.53
7.40
15.63
39.28
-3.31
-4.55
7.00
15.73
39.28
-1.41
-5.58
8.60
15.81
39.28
-1.26
-3.64
9.80
15.87
39.28
-1.10
-1.71
10.30
15.90
45.64
-0.95
0.22
9.70
15.93
45.64
-0.80
2.16
8.70
15.94
45.79
-0.64
4.09
8.20
15.93
46.42
-0.49
6.03
7.10
15.92
46.73
-0.33
7.96
6.20
15.89
72.61
-0.18
9.89
6.00
15.88
72.61
-0.18
9.82
5.60
15.88
72.61
1.83
9.75
4.70
15.89
72.61
3.84
9.67
5.00
15.93
73.00
5.85
9.60
4.80
15.98
73.00
7.87
9.53
4.60
16.04
73.00
9.88
9.45
4.80
16.10
73.00
11.89
9.38
5.40
16.16
66.83
13.90
9.31
150

Interest
Groups
0.72
0.70
0.68
0.66
0.63
0.61
0.59
0.57
0.56
0.56
0.56
0.56
0.57
0.59
0.61
0.63
0.65
0.67
0.68

The United Kingdom saw increases in wealth, lower levels of inflation, a larger aged
population, greater expectations for the role of government and more liberal public opinions over
the time period. All of these variables are expected to result in governments spending more on
items covered by social protection with greater resources to spend and a greater demand for these
goods and services. At the same time, lower levels of unemployment and a left party in office
favoring a policy agenda that included education, public order and safety concerns, and
economic development also served to shape the spending patterns within the United Kingdom
during these years (Table 6.4).
The United Kingdom also serves as a case with zero institutional constraints. Although
the United Kingdom does have a bicameral legislature, the House of Lords does not have the
ability to veto legislation passed by the House of Commons, in essence creating a unicameral
legislature (Lijphart 1999). In Figure 6.8 the solid black line represents the predicted spending
priorities for the United Kingdom based on the interaction model. The dashed line represents the
spending priorities for the United Kingdom assuming it had three institutional constraints over
the same time period, retaining all other values for the variables in the model. The interaction
model predicts that if the United Kingdom had three institutional constraints during this period,
such as a nation like the United States or Italy, it would have spent more on particularized
benefits than it did.
Why is the United Kingdom predicted to spend more on particularized benefits with three
institutional constraints than zero? There are three primary explanations for this outcome. First,
while under zero constraints a more liberal public opinion will result in governments spending
more on particularized benefits, whereas with three institutional constraints, the effect of public
opinion is negated. Similar to public opinion, when the public expects the government to be

151

Figure 6.8

Predicted Spending Priorities for the United Kingdom with Zero and
Three Institutional Constraints

Predicted Spending Priority

0 Constraints
3 Constraints

55

54

53

1990

1995

2000

2005

Year
more involved on issues such as social protection, governments are predicted to spend more on
particularized benefits, under zero institutional constraints. However, expectations on the role of
government are cancelled out with three institutional constraints.
A second reason for expecting greater spending on particularized benefits in this situation
is the relationship between constraints and government composition. Left parties in government
152

carry policy agendas that favor spending on a range of policy areas that include a number of
collective good like education and health. When left parties controlled government in the United
Kingdom under zero institutional constraints, they were able to implement a full range of policy
objectives and higher levels of spending on collective goods. However, the increase in the
number of institutional constraints predicts that the same left parties in office would not be in a
position to obtain their ideal spending outcomes and instead would barter away spending on
areas like education and health in favor of retaining spending on social protection.
A third reason for the constraints pushing spending toward particularized benefits under
higher levels of institutional constraints relates to the number of interest groups in the nation.
The initial argument around interest groups centered on more interests overwhelming the system,
making it difficult for government to cater to the needs of all the groups. However, an increase
in the number of institutional constraints provides more venues or access points for interest
groups to target with their agendas. The model predicts that for the United Kingdom, which was
already seeing a decrease in interest group density during this period (which would increase
spending on particularized benefits), an increase in the number of points for these interest groups
to target and sway regarding policy outputs would lead to further increases in spending on
particularized benefits. However, the influence of interest groups with three institutional
constraints on spending is very modest, at 0.2% shift in spending.
In this example, the predicted result of increasing the number of institutional constraints
is to shift spending towards particularized benefits. The United Kingdom is predicted to
experience an approximate 2.6% point shift from spending on collective goods like education
and defense to particularized benefits such as old-age pensions and unemployment insurance; a
difference equivalent to £17.8 billion in 2008. The influence of preferences on government

153

spending priorities was altered under three institutional constraints. Additionally, the changes
under three constraints over time can be seen in the smaller changes to the predicted spending
priorities in Figure 6.8 compared to when there are zero institutional constraints.
CONCLUSION
The results confirm the hypotheses that institutional constraints influence the policy
process more than the prior literature suggests. Institutional constraints alter the final product of
government. As the example with the United Kingdom highlights, changing the number of
institutional constraints shifts the spending priorities of government. Not only do institutional
constraints shift spending priorities, but they also serve to moderate the ability of governments to
incorporate the preferences of both the elites and the mass publics into the decision making
process. The more actors that are introduced into the policy making process, the harder it will be
for any one actor to move spending in a direction that favors their ideal policy outputs or the
ideal outputs of any particular group.
The findings here are also applicable to different levels of government, particularly
within federal systems where different levels of government must make policy decisions within
specific policy domains. All levels of government face institutional constraints that should alter
their policy outputs. For example, Erikson, Wright and McIver (1989) find that within the
American states, once in office, parties move towards the center regarding nation policy outputs.
The findings in this chapter suggest that once in office, these same parties may still hold the
same preferences for policy outputs that are a part of their party platform and do not move in
terms of their policy preferences, but rather are forced into compromises that benefit all actors in
the decision making process based on the constraints present at the nation level. Within the
American state context, constraints may be found in bicameral state legislatures, within divided

154

governments and in the form of bureaucrats whose careers are not tied to the election process and
represent a separate set of interests from elected incumbents.
A similar result is found in Alt and Lowry’s (2000) work on budgets within the American
states focusing on the ability of parties in office to change revenue dedications under unified and
divided governments between the legislature and executive branches and within bicameral
legislatures. Unified governments at the legislative level provide parties with a greater ability to
change revenues than when the houses are divided. In this context, divided government at the
legislative level serves as a constraint forcing compromise on ideal spending levels. Therefore,
while parties in office may change policy outputs, constraints limit their ability to completely
overhaul spending.

155

CHAPTER 7
CONCLUSION
The work in this dissertation makes three strong contributions to the literature regarding
government spending priorities. First, the spending priorities variable examined in Chapter 3
expands upon the work done in the American states by Jacoby and Schneider (2001, 2009). The
results show that the particularized benefits/collective goods dimension, found at the state level
in the United States over time, also exists across 25 OECD member nations. Additionally, the
priorities model fills a gap in the literature by examining the factors that shape spending patterns.
This results in a model, which is more fully specified relative to previous work. Further, the
interaction model extends prior research on institutional constraints showing how constraints
slow down or impede the policy process. This research indicates that institutional constraints
also alter the final outputs of government as well as their ability to incorporate the will of its
citizens into the decision making process.
GENERAL FINDINGS
The spending priorities variable demonstrates that even though the process of
government spending across democratic nations appears to be a relatively complex, it can
actually be interpreted in a relatively simple, straightforward manner. Government spending is
based on whether expenditures target particular groups in society, through particularized benefits
(e.g., housing vouchers and unemployment benefits), or provide goods and services to the
society in more general terms through collective goods (e.g., defense and education). The
spending priorities variable shows an increase in spending on one set of policies decreases
spending on the other. For example, higher levels of spending on defense will decrease the
resources available to for particularized benefits, such as social protection.

156

The spending priorities variable represents a single dimension of spending and explains
over 90% of the variance in government expenditures. Previous research in this area generally
examines one policy area (or at most, a few policy areas) pre-selected by the researcher based on
a priori assumptions about what policy areas represent similar types of government outputs. On
the other hand, the priorities variable combines expenditures across a fairly wide range of policy
domains. This includes such policies as government operations, social protection, health,
community development, education, economic development, defense, public order and safety,
recreation, and environmental protection. The unfolding analysis allows the expenditures to
determine how the policy areas relate to one another, based on the share of total government
expenditures policies receive relative to one another. Thus, government spending across a
variety of policy domains can be reliably measured and compared across nations in an
encompassing manner.
Furthermore, I establish how different factors traditionally characterized as influencing
government spending patterns affect a reliable measure of government outputs. Three different
sets of factors are examined, including socio-economic influences (predominant in the
functionalist literature), the role of mass and elite preferences for government actions, and the
influence of political institutions. Unlike prior work that tends to focus on one or two groups of
factors, I examine a more fully specified model of government spending that includes variables
from all three theories.
The results show that resources, public demands, and institutions all collectively shape
government spending priorities in democratic nations. Socio-economic variables represent the
resources available to the government and the influence of the societal groups dependent on
government to meet a minimum standard of living (like the elderly and unemployed). As

157

expected within democratic nations, the preferences of both the elites in office and the masses
are mirrored in the policy outputs of government spending priorities. In addition, institutional
designs within a nation result in different expenditure patterns between nations.
The priorities model also shows that the fully specified model is able to account for more
variance in nations’ spending priorities than the individual arguments separately (See Appendix
B). The spending priorities model explains almost 80% of the variance in the priorities variable
which is 20 percentage points higher than the next best model running each set of factor
separately (Model 3, Appendix B). Further, the separate model specifications have omitted
variables producing biased coefficients. This has led to mixed findings in the literature about the
relationship between globalization and citizen mobilization relative to government outputs.
Expanding the understanding of government spending and institutions, I show how
institutional constraints influence expenditure patterns. Institutional constraints increase the time
it takes for governments to reach policy agreements because bartering and negotiations are
required to appease various expectations for policy outputs.

This compromise affects the

responsiveness of governments to different groups’ preferences for government outputs. My
examination extends the work previously done regarding the nature of institutional constraints
and the ability to enact policies different than the status quo (Tsebelis 1995, 2000; Immergut
1990, 2010; Cox and McCubbins 2001; Shugart and Haggard 2001).
My findings suggest the bartering required to reach policy agreement in nations with
multiple institutional constraints also influences the spending priorities of government.

In

nations that have more institutional constraints, actors in the decision making process are in less
of a position to obtain their ideal policy outputs. The bartering results in greater spending on
collective goods that provide all citizens makers with some general level of benefits for all

158

groups. This is the case as collective goods benefit society more broadly, by funding programs
to protect the environment or promote a nation’s defense capabilities.

On the other hand,

spending on particularized benefits would provide goods and services to only some sections of
the population, while leaving out direct benefits to others. My argument is supported with the
interaction model, where more institutional constraints are found to increase spending on
collective goods like education, environmental protection, economic development, and national
defense.
The interaction model also shows that as the number of constraints increase, the ability of
governments to respond to mass and elite preferences diminishes. Again, governments are
forced to compromise in the face of multiple decision makers with expectations for government
spending patterns; no group is in a position to obtain its ideal spending patterns. The effect of
constraints on preferences in society over policy outputs is evident in the diminished cumulative
effect for each set of expectation when interacted with the institutional design present in the
nation.
EXTENSIONS AND IMPLICATIONS
While the focus of this dissertation is on understanding how and why democratic nations
spend resources on different policy areas, the spending priorities variable has various
implications for future work. First, the unfolded policy dimension provides insight into research
on the modern welfare state. These studies have tended to group expenditures on education,
health, and social protection into one category representing “welfare” spending (Huber and
Stephens 2001). However, I show health and education are collective goods that provide
benefits to the community, more broadly. All individuals have access to primary schooling and
universal health care coverage in the majority of democratic nations. Grouping together policy

159

areas that provide benefits to different segments of the population may provide confounding or
contradictory empirical results. For example, when social protection, education and health are
used as a single indicator of welfare a variable (such as globalization) might be found to decrease
spending resulting in support for the “race to the bottom” argument, but when multiple policy
areas are examined together with the spending priorities variable, globalization does not have a
statistically significant affect.
Additionally, the spending priorities variable allows for an examination of government
actions across an array of policy areas, simultaneously. The ten policy areas included in the
development of this variable capture most (if not all) of the spending commitments by the
general government, allowing use to control for any possible relationships between policy areas.
Previous research demonstrates that, the analysis of individual policy areas can produce results
that distort the influence of individual independent variables. Recall the Huber and Stephens’
(2001) model that uses a combination of health and pension spending as the dependent variable
and the replication of this model which includes the composite spending priorities measure as the
dependent variable. Adding together two expenditures like health and pension, which the
unfolding shows cover different types of government spending, resulted in explanatory factors
that are predicted to increase (or decrease) spending on these two different program areas in a
similar manner. The priorities variable preserves the actual relationship, in terms of spending
between these policies, and produces a different picture. In the replication, factors that increase
spending on pensions plans the aged population in a nation actually decrease spending on health
services.
The approach used to create the priorities variable allows researchers to predict examine
levels on individual policy areas after the analyses are run. Predictions are possible because the

160

spending priorities variable preserves the proportion of spending on each policy area based on
the distances between nation spending priority points and the location of the policy points.
Using equation 3.4 discussed in Chapter 3, a researcher can use the predicted spending priorities
values from a model then to determine the proportion of spending on the individual policy areas.
In this manner, researchers who are interested in spending on a specific policy area are able to
use the priorities variable to analyze spending and then derive how their models affect spending
within the domain of interest.
The two sets of policies also support the depiction of government expenditures that is
becoming more prevalent in the literature (Banks and Duggan 2000, 2005; Lizzeri and Persico
2001; Volden and Wiseman 2007). The work done by Jacoby and Schneider (2001, 2009) and
Schneider and Jacoby (2006) shows similar results for the American states over time. Chapter 3
establishes that the same pattern exists for democratic nations. This increases the
generalizability of the particularized benefit/collective goods policy spending dimension.
Since the spending priorities variable captures government activities, it can also be used
to explain policy outcomes in democratic nations. For example, when examining how effective a
policy is at reducing the level of income inequality within nations, the priorities variable can be
used to operationalize government spending in relation whether expenditures promote policies
such as social protection or economic development. Figure 7.1 shows a scatter plot between the
Gini Index, which represents inequality in democratic nations, and policy spending priorities.
The graph shows that as governments spend more on particularized benefits (lower values of the
spending priorities variable) for example, social protection in terms of housing, unemployment,
and pensions—the level of income inequality in a nation decreases. However, as nations spend

161

more on collective goods (higher values of the spending priorities variable), the level of income
inequality in a nation rises.
FIGURE 7.1 Nation Gini Coefficients against Spending Priorities

Data sources: Gini coefficient data were obtained from the CIA World Factbook and
the World Bank.
Note: The slope estimate for the OLS regression line is 0.91 with a standard error of
0.434.
The analyses in Chapters 5 and 6 indicate that institutions matter for government
spending patterns; however, studying institutions, alone, in relation to policy outputs, does not
provide the full picture of government behavior, particularly in regards to political
162

responsiveness. Each individual institution provides insights into how actors behave in this
process. For example, presidential systems are more likely to promote spending on goods that
benefit society as a whole compared to parliamentary systems. However, the cumulative effect
of the increased number of actors introduced into the decision making process given the presence
of constraining institutions is overlooked in relation to policy outputs. Looking at the total effect
of constraints on government spending shows that the outputs of government vary when
bargaining must occur, under multiple constraints, and to when there is a single set of decision
makers with unified goals.
Studies that look at the role of institutional constraints acknowledge that these institutions
slow down the legislative process; however, the indirect effect of constraints on other influential
indicators has not been examined in the literature. By failing to look at how constraints alter the
role of preferences in shaping government activities, studies over state the role of citizens in the
democratic process. The interaction model run in Chapter 6 shows that expectations for
government outputs at both the mass and elite level still influence governmental activity;
however, the nature of the institutions present in the nation can diminish the responsiveness of
governments to translate citizens’ preferences into policy outputs.
The relationship between institutional constraints and societal demands or expectations
for government actions ties back to Lijphart’s (1999) examination of consensual versus
majoritarian systems. Consensual systems attempt to incorporate many preferences to make
democracy the “rule of as many people as possible;” on the other hand majoritarian systems
promote majority rule. Consensual institutions are the institutions that represent constraints in a
nation. These include presidentialism, proportional representation, increased district magnitude,
bicameralism, and federalism. The interaction model shows that by incorporating as many

163

preferences as possible no group in society is in a position to get what they want from
government.
The finding that institutional constraints influence the role of preferences is not limited to
the national government. Constraints also exist at the state and local levels of government. For
example, states may have bicameral legislatures. States may also have officials who are not
elected and have their own preferences over policy outputs. The constraints will again force
compromise on policy and alter government activities as shown in Chapter 6. Therefore, when
examining factors that influence government actions at any level of government, the constraints
present will affect how other factors influence policy outputs.
This point may offer an alternative explanation to the work done on the American states
by Erikson, Wright, and McIver (1989). Erikson, Wright, and McIver argue that once in office,
the political parties move to the center regarding policy decisions. However, the interaction
model would suggest that this is not necessarily true. Instead, the political parties may still hold
the same preferences for government spending as presented during elections, but as a result of
state constraints (such as divided government, bicameral legislatures, and lifelong bureaucrats),
parties may be forced to compromise on government outputs in order to reach agreements while
in office.
CONCLUSION
Democratic nations conceive of policies in a similar ways based on whether expenditures
target particular groups or the community in broader terms. However, governments have
different spending patterns on policy in accordance with the following: what they have to work
with, what citizens want, and how many people have to agree for government to act. No single
policy can capture the full range of what governments do. By examining the influence of

164

variables like national wealth and unemployment on expenditures for social protection, the
connection to other policy areas like economic development is omitted. Models only considering
an individual policy cannot confirm that the increases (or decreases) in spending explained by
the independent variables are unique to that policy area. For example, testing the effect of trade
openness on expenditures for economic development, alone, may show that greater openness
results in higher levels of spending on the economy. However, the same relationship may exist
between trade openness and spending on health, recreation, education, social protection, and
environmental protection. By examining policy areas in isolation from each other, the influence
of explanatory variables, revealed in the model may not be unique. In fact, these relationships
may be occurring across the range of government expenditures in various policy areas (i.e.
increases across total government spending instead of shifts in spending patterns across policies).
Without applying the link between preferences and constraints, the effect of citizens’
expectations and demands for government action can be overstated. The stronger association
between citizens’ preferences and government actions may lead researchers to underestimate the
trade-offs between institutional designs. For example, Lijphart (1999) suggests there are no
socio-economic trade-offs when adopting more consensual institutions. As previously noted,
consensual institutions increase the number of actors/groups present in the decision making
process (i.e. institutional constraints). Instead, Lijphart (1999) argues that the constraints create
a kinder form of democracy represented by greater spending on elements of social protection and
lower crime rates. However, the interaction model I present shows how Lijphart’s argument
misses political trade-offs in terms of how much the public can influence government activities.
This leads to a false sense of security that democracies promote the will of the people. Instead,

165

government outputs are based on compromise that does not mirror any groups’ ideal preferences
for government actions.

166

APPENDICES

167

APPENDIX A

DISTRIBUTIONS OF SPENDING BY POLICY AREA

Figure A.1

Distribution for the Proportion of Spending on Defense

Note: The histogram is based on the percentage of spending out of total spending across
the ten policy areas for all 379 nation years in the dataset.

168

Figure A.2

Distribution for the Proportion of Spending on Economic
Development

Note: The histogram is based on the percentage of spending out of total spending across
the ten policy areas for all 379 nation years in the dataset.

169

Figure A.3

Distribution for the Proportion of Spending on Education

Note: The histogram is based on the percentage of spending out of total spending across
the ten policy areas for all 379 nation years in the dataset.

170

Figure A.4

Distribution for the Proportion of Spending on Environmental
Protection

Note: The histogram is based on the percentage of spending out of total spending across
the ten policy areas for all 379 nation years in the dataset.

171

Figure A.5

Distribution for the Proportion of Spending on Government
Operations

Note: The histogram is based on the percentage of spending out of total spending across
the ten policy areas for all 379 nation years in the dataset.

172

Figure A.6

Distribution for the Proportion of Spending on Community
Development

Note: The histogram is based on the percentage of spending out of total spending across
the ten policy areas for all 379 nation years in the dataset.

173

Figure A.7

Distribution for the Proportion of Spending on Recreation

Note: The histogram is based on the percentage of spending out of total spending across
the ten policy areas for all 379 nation years in the dataset.

174

Appendix B

SEPARATE SPENDING MODELS BY SET OF INFLUENCES

Table B.1

Influence of Socio-Economic Factors on Spending Priorities
Model 1
Coefficient
(s.e.)
p-value

Variable
a
GDP/Capita
Unemployment

a

Inflation
Female
Participation
Dependent
Population

0.000

-0.83
(0.111)

0.000

-0.77
(0.087)

0.000

-0.65
(0.077)
0.000
(0.017)

0.000

-0.84
(0.083)
-0.02
(0.014)

0.000

-0.79
(0.077)
-0.02
(0.013)

0.000

0.02
(0.007)

0.001

0.005
(0.012)

0.335

0.007
(0.010)

0.252

0.02
(0.060)

0.347
-0.21
(0.032)

0.000

-0.26
(0.040)

0.000

0.05
(0.045)

0.142

-0.002
(0.001)
-1.62
(0.068)

0.195

-0.002
(0.001)
-1.63
(0.065)

0.189

0.500

Youth
Population
b

European Union

Model 3
Coefficient
(s.e.)
p-value

-1.00
(0.160)

Aged Population

Openness

Model 2
Coefficient
(s.e.)
p-value

0.004
(0.001)
-2.20
(0.103)

0.006
0.000

2

0.065

0.000

0.099

0.000

R
0.5152
0.5992
0.5967
N
367
367
367
Note: a Indicates that the natural log of the original variable was used in the model.
b Indicates that p-value for the variable is for a non-directional test, all other variables
based on directional tests.

175

Table B.2

Influence of Group Preferences on Spending Priorities

Variable
Government Composition
Voter Turnout

b

Role of Government
Public Opinion
Corporatism

Model 4
Model 5
Coefficient
Coefficient
(s.e.)
p-value
(s.e)
p-value
-0.003
0.245
0.001
0.386
(0.004)
(0.005)
-0.10
(0.006)
-0.01
(0.008)
0.005
(0.003)
-0.12
(0.032)

2

0.036
0.065

-0.03
(0.004)
-0.02
(0.005)
-0.02
(0.003)

0.000

0.52
(0.414)

Interest Groups

0.000

0.000

0.000
0.000

0.000

R
0.5707
0.2548
N
214
369
Note: b Indicates that p-value for the variable is for a non-directional test, all
other variables based on directional tests.

176

Table B.3

Influence of Institutions on Spending Priorities

Variable
President
PR
Mixed Voting
District Magnitude

a

Model 6
Coefficient
(s.e.)
p-value
1.93
0.000
(0.262)
-1.79
0.000
(0.107)
-1.47
0.000
(0.124)
0.34
(0.053)
0.37
(0.068)
-0.75
(0.073)

Bicameralism
Federalism
2

0.000
0.000
0.000

R
0.2484
N
369
Note: a Indicates that the natural log of the original
variable was used in the model.

177

Appendix C

DIAGNOSTIC TESTS AND MODEL SELECTION

Time Dummies
Year time dummies are included in the model in order to address any potential time
trends present in the spending priority variable. The year dummies do not carry directional
hypotheses. 1990 is used as the base category for the time dummies and the remaining time
dummies would be interpreted in relation to 1990. After the analysis a joint F-test is performed
to test for joint significance of the time dummies. If the year dummies are omitted the results
estimates from the resulting model are similar to the estimates in interaction model in terms of
signs, magnitude, and statistical significance. The only change is that the interaction term
between institutional constraints and role of government is not statistically significant at the 0.05
or 0.10 level for a directional test.
Lags
The independent variables included in the analysis are lagged two time periods. The
lagging of the independent variables is theoretically justified based on the general process of how
government expenditures occur. National governments’ expenditures tend to run on a two year
lag in order to established the budget and execute spending (Hofferbert and Budge 1992;
Klingmann et al. 1994). One year is required to discuss and approve spending and an additional
year is required for spending to occur. As there is a two year lag, it would not make theoretical
sense to determine the relationship between current events and observations with spending
patterns that are established and executed previously.

178

The process of a two year lag also helps to alleviate concerns over the direction of the
relationship between the dependent and independent variables. As I am using independent
variables that have been lagged two years, it would not be expected to find the actual
expenditures at time t affecting what happened at time t-2.
A lag of the spending priorities variable is not included for theoretical reasons. If a lag of
the spending priorities is included it would explain a large portion of the variance seen in
spending priorities the following year. However, from an information standpoint, this approach
fails to provide insight into what influences the actual pattern in the level of spending as it does
not contribute to the understanding of why governments spend. Further, there is degree of
multicollinearity between the lagged dependent variable and the other independent variables
increasing the difficultly of separating the relationships between the variables in the model. As
the focus here is on explaining spending priorities and not predicting spending priorities, I have
chosen to omit the lag of spending priorities that would improve predictions in favor of variables
that are expected to influence spending.
I also do not include a lag of the spending priorities in the model as it would theoretically
support an incrementalist argument in regards to the spending process. Incrementalism is based
on the assumption that a full review of the budget each year requires too much time, therefore
small changes are made each year but complete overhauls do not occur. It has been repeatedly
shown, however, that while this approach can explain the majority of changes that are minor
increases and decreases in spending that occur, it cannot account for the rapid changes in policy
explained by punctuated equilibriums (Baumgartner and Jones 1993; Baumgartner, GreenPedersen, and Jones. 2006; Baumgartner, Foucault, and Francois 2006; Baumgartner et al. 2009;
Jones et al. 2009). Therefore, if I include a lag of spending in the model I could predict the small

179

changes from year to year, but not large changes, nor could I explain why governments are
spending at the level they are on different policy areas.
Achen (2000) makes the point examining models that predict budgets, that, “budgets are
generally quiet stable over time and well predicted by the prior year, but nothing follows about
the bounded rationality or incrementalist thinking of the decisionmaker” (11). The expenditures
that are studied here behave similarly to budgets where the prior year’s values can explain
roughly 95% of the variance in the current year’s expenditures. However, this does not provide
and insight into what resulted in last year’s expenditures beyond what was spent the year before,
and so on. Achen (2000) also argues that, “lagged budgets will falsely appear to be the sole
cause of future budgets when the political environment is stable, but not otherwise,” applying
this to the expenditure data, when the influences of expenditures remain relatively similar to last
year, last year’s expenditures will appear to be a good predictor of the current year’s
expenditures(10). However, when the influences on expenditures change dramatically, the
ability of last year to predict would not accurately reflect why governments are spending. For
example, if extreme conservatives like Tea Party members or Libertarians gained office in the
United States, the ability of the prior years to predict expenditures patterns would likely fall
short. Instead models that focus on what factors influence spending and test the effect of
political parties in office and public opinion; that would have changed to elect these candidates,
the spending priorities would be more appropriately modeled.
Transformation
Having established the estimation procedure, I begin with the final model examined in
Chapter 5 that includes the socio-economic, elite and mass preferences and institutional variables
(priorities model) and replace the separate institutional variables with the institutional constraint

180

variable (constraint model). After examining the results of the constraint model, I rerun the
model omitting the variables that failed to reach statistical significance at the 0.05 level which
include trade openness, the inflation rate, female participation, and voter turnout (Model 2).
After running Model 2, I compare the variance explained by the two models and find the
variance explained by the model that includes the variables that failed to reach statistical
significance is not statistically different than the model that does not (Model 2, Table 6.2). The
empirical F-statistic calculated based on Model 1 and Model 2 is 0.827 which is less than the
critical value of 2.37 at the 95% level, meaning I fail to reject the null that Model 1 explains
more variance in spending priorities than Model 2.
Having established the base model, I run a full model that includes the interaction effects
between the institutional constraints and the measure of elite and mass preferences (interaction
model, Table 6.3). In order to confirm that openness, the inflation rate, female participation, and
voter turnout would not affect the finding in interaction model, I ran an additional model that
included the openness inflation rate, female participation, voter turnout, and voter turnout
interacted with the number of institutional constraints. The model finds produced similar results
in terms of signs, magnitude, and statistical significance as interaction model. Further, openness,
the inflation rate, voter turnout and the interaction between voter turnout and constraints were
not statistically significant either the 0.05 or the 0.10 level. Female participation carried a
positive coefficient and was statistically significant at the 0.05 level for a directional test.
However as the model does not increase the variance explained in a statistically significant
manner I chose to use the more parsimonious interaction model.
Because the models I use are based on the assumption of linearity I examine the
relationship between spending priorities and the independent variables before moving forward.

181

Figures 6.2 through 6.8 present the component plus residual plots for each of the interval level
variable in the model, each figure includes a line for the linear fit and a lowess curve
representing the relationship between the independent variable and government spending
priorities.
Examining each of the component plus residual plots shows that the relationships
between the independent variables and the spending priorities variable exhibit predominately
linear relationships, controlling for the other variables in the model. The natural logs of gross
domestic product per capita and unemployment were logged in the prior models from Chapter 5,
following the conventions of prior works that had examined the relationship between gross
domestic product per capita and unemployment, and government spending. Figures C.1 and C.2
show the natural logs of gross domestic product per capita and unemployment produce a linear
relationship with the spending priorities variable, controlling for the other variables in the model.
The aged population (Figure C.3), government composition (Figure C.4), role of
government (Figure C.5), public opinion (Figure C.6), and interest groups (Figure C.7), also
produce linear relationships with the spending priorities variable. The relationship between
expectations for government and spending priorities in Figure C.5 shows a small divergence
from linearity at lower values of expectations; however, at the lower end there are fewer
observations to use to establish the relationship. A linear relationship exists through the majority
of the data between role of government and spending priorities suggesting a transformation is
unwarranted; as the relationship is statistically significant in the model and a transformation
would only increase the difficulty of interpreting the effect of expectations for government on
spending priorities.

182

Multicollinearity
The nature of the data indicates prior to the analysis that multicollinearity is likely to be
present. In situations where multicollinearity is high, the standard errors will be inflated and
may produce estimates that appear to fail to reach traditional levels of statistical significance. In
order to address any issues of multicollinearity I examine the variance inflation factors after the
analysis in interaction model is performed. In situations where multicollinearity appears to be
high and the variable fails to reach statistical significance I note the potential for the standard
errors to have been inflated by the other variables in the model. In the future as more data
become available, additional analyses will help sort out the effects of different variables that may
suffer under current conditions of multicollinearity.
After running the interaction model, the variance inflation factors appear high; however,
as the four key interaction terms are a product of four preference variables in the model and the
institutional constraints variable this is not surprising. Table C.1 shows the variance inflation
factors for interaction model and the variance inflation factors for Model 2 where the interactions
are omitted. The remaining multicollinearity in Model 2 exists between the socio-economic
variables that are expected to be related to one another. Table C.2 shows the basic correlations
between the independent variables and also demonstrates that the majority of multicollinearity
occurs between the interaction terms, preferences, and the institutional constraint variables.
Residuals
After running the analysis I examine the residuals to test that there is constant error
variance. The examination of the residuals begins with looking at scatter plot of the residuals
versus the fitted values (Figure C.8). The scatter plot shows that the residuals appear to be

183

randomly distributed around zero. However, around both the minimum and maximum fitted
values, the residuals appear to deviate from a random distribution around zero.
In order to confirm what is seen in the scatter plots if Figure C.8, I run a White test for
heteroskedasticity. The null hypothesis for the White test is that homoskedasticity exists in the
residuals. After running the White test on residuals from the analysis I am unable to reject the
null hypothesis with a p-value of 0.1108. Using these two approaches to examine the residuals I
am comfortable stating that the residuals are homoskedastic. Had heteroskedasticity been an
issue, however, the panel-corrected standard errors would have accounted for it in the estimation
of the standard errors in the model.
Influential Observations
In order to ensure that influential observations are not affecting the results I create a plot
of the leverage values against the normalized residuals squared (Figure C.9). After examining
the leverage versus residuals plot in Figure C.9, there do not appear to by any particular
observations that have both high residuals and high leverage to bias the results of the interaction
model.
In addition to examining the plot of leverage against normalized residuals squared, I also
examine the actual values of the leverage statistics and the residuals. There were no observations
that had a both a leverage statistic greater 0.1744 and a residual greater than 1.312 or less than 37

1.312.

To further ensure that there are no influential observations I also examine the Cook’s

Distance (or Cook’s D) values. None of the Cook’s D values are close to 1, indicating that the

37

Here the threshold for the leverage value is greater than 2*32/367=0.174386921 and the
threshold for the residuals is absolute value is greater than 1.312052 (+/- two standard deviations
away from the mean).
184

influence of any individual observation is minimal.

38

I use this as further evidence that there are

no individual observations influencing the results in the model.

38

The largest cook’s D value is 0.0635107.
185

Table C.1

Variance Inflation Factor Scores

GDP/Capita
Unemployment
Aged Population
European Union
Institutional Constraints (IC)
Government Composition
IC x Government Composition
Role of Government
IC x Role of Government
Public Opinion
IC x Public Opinion
Interest Groups
IC x Interest Groups
Mean VIF

186

VIF
Model 3
55.03
22.90
78.76
7.01
75.70
53.48
67.27
14.66
13.76
29.69
25.59
20.86
22.55
16.89

VIF
Model 2
44.45
18.44
73.88
6.80
8.11
10.17
1.42
1.63
1.34
7.74

Table C.2

Correlations Matrix for Independent Variables in Interaction Model

GDP/Capita
Unemployment
Aged Population
European Union
Institutional Constraints
Government Composition
IC x Government
Composition
Role of Government
IC x Role of Government
Public Opinion
IC x Public Opinion
Interest Groups
IC x Interest Groups

GDP/
Capita Unemployment
1.0000
1.0000
-0.5143
0.1909
0.0787
-0.2927
0.3174
-0.0888
0.0974
-0.2328
0.0902
-0.2322
0.5622
0.5342
-0.3271
-0.2335
0.2303
0.2356

0.1680
-0.2371
-0.2497
0.2706
0.1681
-0.0194
-0.0505

Aged
Population

European
Union

Institutional
Constraints

IC x
Government Government
Composition Composition

1.0000
0.2575
0.0365
0.1309

1.0000
0.0307
0.0045

1.0000
-0.0795

1.0000

0.0947
0.1562
0.1520
0.1835
0.1937
0.0186
-0.0093

0.0279
-0.1035
-0.0817
0.0739
0.0158
-0.3221
-0.3424

0.7330
0.0510
0.0652
-0.1850
-0.1233
0.1141
0.2376

0.5775
0.1225
0.0775
0.3880
0.3776
0.0702
0.0673

187

1.0000
0.0892
0.0770
0.1403
0.1832
0.1385
0.2367

Table C.2

(cont’d)
Role of
Government

GDP/Capita
Unemployment
Aged Population
European Union
Institutional Constraints
Government Composition
IC x Government Composition
Role of Government
IC x Role of Government
Public Opinion
IC x Public Opinion
Interest Groups
IC x Interest Groups

IC x Role of
Government

1.0000
0.9520
-0.2269
-0.1554
0.2156
0.2072

1.0000
-0.1561
-0.0956
0.2404
0.2429

188

Public
Opinion

1.0000
0.9672
0.1749
0.1630

IC x Public
Opinion

1.0000
0.1850
0.1975

Interest
Groups

1.0000
0.9636

IC x Interest
Groups

1.0000

Figure C.1

Component plus Residual Plot for the Natural Log of GDP/Capita

189

Figure C.2

Component plus Residual Plot for the Natural Log of Unemployment

190

Figure C.3

Component plus Residual Plot for the Aged Population

191

Figure C.4

Component plus Residual Plot for Government Composition

192

Figure C.5

Component plus Residual Plot for Role of Government

193

Figure C.6

Component plus Residual Plot for Public Opinion

194

Figure C.7

Component plus Residual Plot for Interest Groups

195

Figure C.8

Scatter Plot of Residuals against Fitted Values

196

Figure C.9

Plot of Leverage and Residuals by Observation

197

REFERENCES

198

REFERENCES
Achen, Christopher H. 2000. “Why Lagged Dependent Variables Can Suppress the Explanatory
Power of Other Independent Variables.” Prepared for the Annual Meeting of the Political
Methodology Section of the American Political Science Association.
Alt, James E. and Robert C. Lowry. 2000. “A Dynamic Model of State Budget Outcomes under
Divided Partisan Government.”Journal of Politics. 62.4:1035-69.
Anderson, Karen M. 2001. “The Politics of Retrenchment in a Social Democratic Welfare State:
Reform of Swedish Pensions and Unemployment Insurance.” Comparative Politics. 34.9:
1063-91.
Almond, Gabriel and Sidney Verba. 1965. The Civic Culture: Political Attitudes and Democracy
in Five Nations, an Analytic Study. Boston: Little Brown.
Associate Press. “Austerity in Europe: what does it mean for ordinary people?” accessed 12 May
2012.
Austen-Smith, David. 2000. “Redistributing Income under Proportional Representation.”
Journal of Political Economy. 108.6: 1235-69.
Bahler, Thomas, Michaela Pfeifer and Claus Wendt. 2010. “Social Assistance.” in The Oxford
Handbook of the Welfare State. eds. Francis G. Castles, Stephen Leibfried, Jane Lewis,
Herbert Obinger, and Christopher Pierson. USA: Oxford University Press.
Banks, Jeffrey S. and John Duggan. 2000. “A Bargaining Model of Collective Choice.”
American Political Science Review. 94:73–88.
Banks, Jeffrey S. and John Duggan. 2005. “A General Bargaining Model of Legislative Policymaking.” Quarterly Journal of Political Science. 1: 49–85.
Baumgartner, Frank R. and Bryan D. Jones. 1993. Agendas and Instability in American Politics.
Chicago: The University of Chicago Press.
Beck, Nathaniel and Jonathan N. Katz. 2001. “Time-Series-Cross-Section Data: What Have we
Learned in the Past Few Years.” Annual Review of Political Science. 4:271-93.
Bednar, J. 2009. The Robust Federation: Principles of Design. Cambridge University
Press.
Bradley, David et al. 2003. “Distribution and Redistribution in Postindustrial Democracies.”
World Politics. 55.2: 193-228.

199

Bräuninger, Thomas. 2005. “A Partisan Model of Government Expenditure.” Public Choice.
125.3/4:409-29.
Breunig, Christian. 2006. “The More Things Change, The More Things Stay the Same: A
Comparative Analysis of Budget Punctuations.” Journal of European Public Policy.
13.7: 1069-85.
Brooks, Clem and Jeff Manza. 2007. Why Welfare States Persist: The Importance of Public
Opinion in Democracies. The University of Chicago Press Books.
Bureau of Labor Statistics. “Databases & Tools.” http://www.bls.gov/.
Budge, Ian and Hans E. Keman. 1990. Parties and democracy: Coalition Formation and
Government Functioning in 22 Democracies. Oxford University Press.
Cameron, David R. 1978. “The Expansion of the Political Economy: A Comparative Analysis.”
American Political Science Review. 72.4:1243-61.
Cameron, Colin A. and Pravin K. Trivedi. 2005. Microeconometrics: Methods and Applications.
Cambridge University Press.
Castles, Francis G. 2006. “A Race to the Bottom?” in The Welfare State Reader. eds.
Christopher Pierson and Francis G. Castles. Policy Press.
Chang, Eric. 2008. “Electoral Incentives and Budgetary Spending: Rethinking the Role of
Political Institutions.” Journal of Politics 70.4: 1086-97.
Chhibber, Pradeep and Irfan Nooruddin. 2004. “Do Party Systems Count? The Number of
Parties and Government Performance in the Indian States.” Comparative Political
Studies. 37: 152-87.
Cox, Gary. 1997. Making Votes Count: Strategic Coordination in the World’s Electoral Systems.
Cambridge University Press.
Cox, Gary W. and Matthew D. McCubbins. 2001. “The Institutional Determinants of Economic
Policy Outcomes.” In Presidents, Parliaments, and Policy. ed. Stephan Haggard and
Mathew D. McCubbins. Cambridge University Press.
Crepaz, Markus M.L. 1998. “Inclusion versus Exclusion: Political Institutions and Welfare
Expenditures.” Comparative Politics. 31.1:61-80.
Dean, Hartley. 2006. Social Policy. USA: Policy Press.
Docteur, Elizabeth and Howard Oxley. 2003. Health-Care Systems: Lessons from the Reform
Experience. OECD Health Working Papers.

200

Easton, David 1953. The Political System. New York: Alfred A. Knopf Inc.
Edwards, Martin S. and Frank C. Thames. 2006. “District Magnitude, Personal Votes, and
Government Expenditures.” Electoral Studies. 26: 338-45.
Erikson, Robert S., Gerald C. Wright Jr. and John P. McIver. 1989. “Political Parties, Public
Opinion, and State Policy in the United States.” American Political Science Review.
83.3:729-50.
European Union. “Countries.” http://europa.eu/about-eu/countries/index_en.htm.
Falkner, Gerda. 2010. “European Union.” in The Oxford Handbook of the Welfare State. eds.
Francis G. Castles, Stephen Leibfried, Jane Lewis, Herbert Obinger, and Christopher
Pierson. Oxford University Press.
Garand, James C. 1985. “Partisan change and Shifting Expenditure Priorities in the American
States, 1948-78.” American Political Science Research. 13:355-391.
Garand, James C. and Rebecca M. Hendrick. 1991. “Expenditure Tradeoffs in the American
States: A Longitudinal Test, 1948-1984.” Western Political Quarterly. 44.4:915-40.
GESIS Data Archive for the Social Sciences in Cologne European Values Study.
http://www.europeanvaluesstudy.eu/
Goren, Paul. 2004. “Political Sophistication and Policy Reasoning: A Reconsideration.”
American Journal of Political Science. 48.3: 462-78.
Gray, Virginia and David Lowery. 1996. The Population Ecology of Interest Representation:
Lobbying Communities in the American States. Ann Arbor: The University of Michigan
Press.
Gurría, Angel. 2008. “Developing a Health Care System Benefiting All.” China Development
Forum. OECD.
Haggard, Stephens and Matthew D. McCubbins. 2001. “Introduction: Political Institutions and
the Determinants of Public Policy.” In Presidents, Parliaments, and Policy. ed. Stephan
Haggard and Mathew D. McCubbins. Cambridge University Press.
Haggard, Stephens and Gary Noble. 2001. “Power Politics: Elections and Electricity Regulations
in Taiwan.” In Presidents, Parliaments, and Policy. ed. Stephan Haggard and Mathew D.
McCubbins. Cambridge University Press.
Hand, David J. Measurement: Theory and Practice. Oxford University Press.
Hay, Colin and Ben Rosamond. 2002. “Globalization, European Integration, and the Discursive
Construction of Economic Imperatives.” Journal of European Public Policy. 9.2:147-67.
201

Heller, William and Matthew D. McCubbins. 2001. “Political Institutions and Economic
Development: The Case of Electric Utility Regulation in Argentina and Chile.” In
Presidents, Parliaments, and Policy. ed. Stephan Haggard and Mathew D. McCubbins.
Cambridge University Press.
Hibbs, Douglas A. 1992. “Partisan Theory after Fifteen Years.” European Journal of Political
Economy. 8.3:361-373.
Hill, Kim Quaile, and Angela Hinton-Anderson. 1995. “Pathways of Representation: A Causal
Analysis of Public Opinion-Policy Linkages.” American Journal of Political Science
39.4: 924-35.
Hofferbert, Richard I. 1974. The Study of Public Policy. Macmillan.
Hofferbert, Richard I. and Ian Budge. 1992. “The Party Mandate and the Westminster Model:
Election Programmes and Government Spending in Britain, 1948-85.” British Journal of
Political Science. 22.2:151-82.
Huber, Evelyn and John D. Stephens. 1993. “Political Parties and Public Pensions.” Acta
Sociologica. 36.4:309-25.
Huber, Evelyn and John D. Stephens. 2000. “Partisan Governance, Women’s Employment, and
the Social Democratic Service State.” American Sociological Review. 65.3: 323-42.
Huber, Evelyn and John D. Stephens. 2001. Development and Crisis of the Welfare State:
Parties and Policies in Global Markets. University of Chicago Press.
Immergut, Ellen M. 1990. “Institutions, Veto Points, and Policy Results: A Comparative
Analysis of Health Care.” Journal of Public Policy. 10.4:391-416.
Immergut, Ellen M. 2010. “Political Institutions.” in The Oxford Handbook of the Welfare State.
ed Francis G. Castles, Stephen Leibfried, Jane Lewis, Herbert Obinger, and Christopher
Pierson. Oxford University Press.
Inglehart, Ronald. 1990. Culture Shift in Advanced Industrial Society New Jersey: Princeton
University Press.
Inglehart, Ronald and Paul Abramson. 1994. Value Change in Global Perspective. USA:
University of Michigan Press.
Institute for Democratic and Electoral Assistance. “Voter Turnout.”
http://www.idea.int/vt/index.cfm.
Inter-Parliamentarian Union. PARLINE database on national parliaments.
http://www.ipu.org/english/home.htm.

202

Iversen, T. and D. Soskice. 2006. “Electoral Institutions and the Politics of Coalitions: Why
Some Democracies Redistribute More than Others.” American Political Science Review.
100.2:165-81.
Jackman, Robert W. 1987. “Political Institutions and Voter Turnout in the Industrial
Democracies.” American Political Science Review. 81.2: 405-24.
Jacobs, Lawrence and Theda Skocpol. 2010. Health Care Reform and American Politics: what
Everyone Needs to Know. Oxford University Press.
Jacoby, William G. and Saundra K. Schneider. 2001. “Variability in State Policy Priorities: An
Empirical Analysis.” Journal of Politics 63.2:544-68.
Jacoby, William G. and Saundra K. Schneider. 2009. “A New Measure of Policy Spending
Priorities in the American States.” Politica Analysis. 17: 1-24.
Jones, Bryan D. et al. 2009. “A General Empirical Law of Public Budgets: A Comparative
Analysis.” American Journal of Political Science. 53.4:855-73.
Keefer, Philip. 2010. Database of Political Institutions. Development Research Group. The
World Bank.
Kenworthy, Lane. 2003. “Quantitative Indicators of Corporatism.” International Journal of
Sociology. 33.3:10-44.
Kingdon, John W. 1984. Agenda, Alternatives, and Public Policies. Boston: Little, Brown.
Klingemann, Hans-Dieter, Robert I. Hofferbert, and Ian Budge. 1994. Parties, Policies, and
Democracy. Boulder: Westview Press.
Klingman, David and William W. Lammers. 1984. “The ‘General Policy Liberalism’ Factor in
American State Politics.” American Journal of Political Science. 28.3: 598-610.
Korpi, Walter. 1983. The Democratic Class Struggle. Routledge & Kegan Paul.
Korpi, Walter.1989. “Power, Politics, and State Autonomy in the Development of Social
Citizenship: Social Rights during Sickness in Eighteen OECD Countries since 1930.”
American Sociological Review. 54:309-29.
Korpi, Walter.2006. “Welfare-State Regress in Western Europe: Politics, Institutions,
Globalization, and Europeanization.” In The Welfare State Reader, eds. Pierson C and
F.G. Castles. Polity Press.
Labour. “History of the Labour Party.” http://www.labour.org.uk/history_of_the_labour_party.

203

Lehmbruch, Gerhard. 1982. “Introduction: Neo-Corporatistm in Comparative Perspective.” in
Patterns of Corporatist Policy-Making. eds. Lehmbruch, Gerhard and Philippe C.
Schmitter. Sage Publications.
Lienert, Ian. 2005. “Who Controls the Budget: The Legislature of the Executive.” IMF Working
Papers.
Lijphart, Arend. 1997. “Unequal Participation: democracy’s Unresolved Dilemma.” American
Political Science Review. 91.1: 1-14.
Lijphart, Arend. 1999. Patterns of Democracy. New Haven and London: Yale University Press.
Lizerri, Alessandro, and Nicola Persico. 2001. “The Provision of Public Goods under Alternative
Electoral Incentives.” American Economic Review 91: 225–39.
Lowi, Theodore J. 1964. “American Business, Public Policy, Case-Studies, and Political
Theory.” World Politics. 16.4: 677-715.
McConnell, Grant. 1970. Private Power and American Democracy. Vintage Books.
Milesi-Ferretti, Gian. M., Roberto. Perotti, and Massimo. Rostagno. 2002. “Electoral Systems
and Public Spending.” Quarterly Journal of Economics. May:609-57.
Norris, Pippa. 2004. Electoral Engineering: Voting Rules and Political Behavior. Cambridge
Studies in Comparative Politics.
Norweigen Social Science Data Service. European Elections Database.
http://www.nsd.uib.no/european_election_database/.
Obinger, Herbert, Stephan Leibfried & Francis G. Castles, Eds. 2005. Federalism and the
Welfare State. New World and European Experiences, Cambridge University Press.
Obinger, Herbert and Uwe Wagschal. 2010. “Social Expenditures and Revenues.” in The Oxford
Handbook of the Welfare State. ed Francis G. Castles, Stephen Leibfried, Jane Lewis,
Herbert Obinger, and Christopher Pierson. Oxford University Press.
Okma, Kieke G.H. 2011. “Reflections from Afar: Obama’s Health Reform in European
Perspective.” Journal of Politics.36.3.
Organization of Economic Co-operation and Development. OECD stats. National Accounts:
Government Expenditures by Function 1990-2009.
Organization of Economic Co-operation and Development. OECD. 2009. Government at a
Glance.
Organization of Economic Co-operation and Development. OECD. 2009. Health at a Glance
2009: OECD Indicators.
204

Page, Benjamin J. and Robert Y. Shapiro. 1983. “Effects of Public Opinion on Policy.”
American Political Science Review. 77.1: 175-90.
Peacock, Alan and Alex Scott. 2000. “The Curious Attraction of Wagner’s Law.” Public Choice.
102:1-17.
Penner, Erin, Stuart Soroka, and Kelly Blidook. 2006. “Legislative Priorities and Public Opinion:
The Representation of Partisan Agendas in the Canadian House of Commons.” Journal of
European Public Policy. 13.7: 1006-20.
Persson, Torsten and Guido Tabellini. 2003. The Economic Effects of Constitutions. The MIT
Press.
Peterson, Paul E. 1995. The Price of Federalism. The Brookings Institute.
Pierson, Paul. 1996. “The New Politics of the Welfare State.” World Politics. 48.2: 143-79.
Poole, Keith. "Least Squares Metric, Unidimensional Unfolding." Psychometrika, September.
311-23.
Powell, G. Bingham Jr. 1986. “American Voter Turnout in Comparative Perspective.” American
Political Science Review. 80.1:17-43.
Raimondo, Henry J. 1996. “State Budgeting: Problems, Choices, and Money.” In The State of
the States. eds. Carl E. Van Horn, 3rd ed. Washington D.C.: Congressional Quarterly
Press.
Riker, William H. 1982. “The two-party System and Duverger’s Law: An Essay on the History
of Political Science.” American Political Science Review. 76.4: 753-66.
Ringquist, Evan J. and James C. Garand. 1999. “Policy Change in the American States.” In
American States and Local Politics: Directions for the 21st Century. ed. Ronald E. Weber
and Paul Brace. Chatham House Publishers and Seven Bridges Press, LLC.
Roberts, Andrew and Byung-yeon Kim. 2011. “Policy Responsiveness in Post-communist
Europe: Public Preferences and Economic Reform.” British Journal of Political Science.
41:819-39.
Rodik, Dani. 1998. “Why do More Open Economies have Bigger Governments?” Journal of
Political Economy. 106.5: 997-1032.
Salisbury, Robert H. 1968. “The Analysis of Public Policy: A Search for Theories and Roles.” In
Political Science and Public Policy. ed. Austin Ranney. Chicago: Markham Publishing
Company.
Savas, E.S. 2000. Privatization and Public-Private Partnerships. CQ Press.
205

Scartascini, Carlos G. and W Mark Crain. 2002. “The Size and Composition of Government
Spending in Multi-Party Systems.” Social Science Research Network.
Scharpf, Frtiz. 2000. “The Viability of Advanced Welfare States in the International Economy:
Vulnerabilities and Options.” Journal of European Public Policy. 7.2: 190-228.
Schattschneider, E.E. 1975. The Semisovereign People: A Realist’s View of Democracy in
America. The Dryden Press.
Schneider, Saundra K. and William G. Jacoby. 2006. “Citizen Influences on State Policy
Priorities: The Interplay of Public Opinion and Interest Groups” in Public Opinion in
State Politics. ed. Jeff Cohen. Stanford University Press, pp. 183-208.
Shelton, Cameron A. 2007. “The Size and Composition of Government Expenditure.” Journal of
Public Economics. 91:2230-60.
Shugart, Matthew S. and Stephen Haggard. 2001. “Institutions and Public Policy in Presidential
Systems.” In Presidents, Parliaments, and Policy. eds. Stephan Haggard and Mathew D.
McCubbins. Cambridge University Press.
Soroka, Stuart N. and Christopher Wlezien. 2004. “Opinion Representation and Policy
Feedback: Canada in Comparative Perspective.” Canadian Journal of Political Science.
37.3: 531-59.
Soroka, Stuart N. and Christopher Wlezien. 2005. “Opinion Representation and Policy
Feedback: Canada in Comparative Perspective.” Canadian Journal of Political Science.
37.3:531-60.
Starke, Peter. 2006. “The Politics of Welfare State Retrenchment: A Literature Review.” Social
Policy & Administration. 40.1: 104-20.
Swank, Duane. 2010. “Globalization.” in The Oxford Handbook of the Welfare State. ed Francis
G. Castles, Stephen Leibfried, Jane Lewis, Herbert Obinger, and Christopher Pierson.
Oxford University Press.
Tabellini, Guido. 2000. “Constitutional Determinants of Government Spending.” CESifo
Working Paper Series.
Tsebelis, George. 1995. “Decision Making in Political Systems: Veto Players in Presidentialism,
Parliamentarism, Multicameralism, and Multipartyism.” British Journal of Political
Science. 25: 289-326.
Tsebelis, George. 2000. “Veto Players and Institutional Analysis.” Governance: An Institutional
Journal of Policy Administration. 13.4: 441-74.

206

Tsebelis, George. 2002. Veto Players: How Political Institutions Work. Princeton University
Press.
United Nations. International Labour Organization. http://www.ilo.org.
United Nations. United Nations Educational, Scientific and Cultural Organization. “Enrolment
and Alternative Solutions.”
http://www.unesco.org/education/educprog/50y/brochure/tle/130.htm
Volden, Craig and Alan E. Wiseman. 2007. “Bargaining in Legislatures over Particularistic and
Collevtive Goods.” American Political Science Review. 101: 79-92.
Wildavsky, Aaron. 1998. Culture and Social Theory. eds. Sun-Ki Chain and Brendon Swedlow
Transaction Publishers.
Wilson, James Q. Bureaucracy: What Government Agencies Do and Why They do it. Basic
Books Inc.
Wooldridge, Jeffery. 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press.
World Bank. “Data” http://data.worldbank.org/.
World Guide to Trade Associations. 1985, 1995, 1999, 2001. New York: New York, R. R.
Bowker Co.
World Values Survey Assocation. World Values Survey waves 1-4.
http://www.worldvaluessurvey.org/.
Zutavern, Jan and Martin Kohli. 2010. “Needs and Risks in the Welfare State.” in The Oxford
Handbook of the Welfare State. ed Francis G. Castles, Stephen Leibfried, Jane Lewis,
Herbert Obinger, and Christopher Pierson. USA: Oxford University Press.

207