Emu-m

 

 

 

KTHESH

 
    
 
 

HERA!“
’ Michigan Etate
‘ University

O . I ,
‘p" W"""

This is to certify that the

thesis entitled

STRATEGIC BEHAVIOR IN SINGLE ISSUE
COLLECTIVE CHOICE SETTINGS

presented by

DANIEL B. STENGEL

has been accepted towards fulfillment
of the requirements for

Ph D degree in Political Science

    

ajor professor

31 July, 1984 Gary J' M11191
Date

 

0-7639 MS U is an Affirmative Action/Equal Opportunity Institution

 

 

 

 

 

 

MSU

LIBRARIES
‘35—»

 

 

RETURNING MATERIALS:
Place in book drop to
remove this checkout from
your record. FINES wilI
be charged if book is
returned after the date
stamped below.

 

 

 

 

 

STRATEGIC BEHAVIOR IN SINGLE ISSUE COLLECTIVE CHOICE
‘- ' ' SETTINGS“‘
AN EXPERIMENTAL ANALYSIS WITH APPLICATION TO MILLAGE ISSUE
‘ " ELECTIONS ‘ ' '

HY

DANIEL B. STENGEL

A DISSERTATION

Submitted to
E Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Political Science

1984

ABSTRACT
STRATEGIC BEHAVIOR IN SINGLE ISSUE COLLECTIVE CHOICE
‘ SETTINGS:
AN EXPERIMENTAL ANALYSIS WITH APPLICATIONS TO MILLAGE ISSUE
' ELECTIONS ‘
BY

DANIEL B. STENGEL

Early formal models of collective choice were explicitly
unidimensional (i.e., concerned with a single issue). Later
analysis has shown that the initial success of these models
depended essentially upon their unidimensional form and that
this form was not appropriate to most collective choice
settings. One exception is the use of public referenda to
set local tax rates for the provision of public education.
Given this, much ink has been shed by scholars of economics
and political science in an effort to test empirically
notions derived from.forma1 modeling in this environment.
The results of these efforts have been largely inconclusive.
This dissertation argues that much of this work has been
misdirected and that a primary failure has been the
inappropriate specification of certain crucial variables
within these models. Experimental designs are used to
isolate key factors such as the relationship between turnout
and intensity of preference, relative levels of
sophistication on the part of key actors within the
environment, and the effects of different institutional
constraints on participant interaction. The data derived

from these experiments are analyzed using Probit and

analysis of variance techniques. The respecified variables
are then applied in a case study of a recent millage

election in a Michigan school district.

ii

TABLE OF CONTENTS

INTRODUCTION... ....... ...... .............................. 1
CHAPTER 1 ........................ .... .................. ..11
CHAPTER 2. ................. . ........... . ................. 44
CHAPTER 3.. ..... . ....................................... .74
CHAPTER 4 ................................................ 92
CHAPTER 5. ...................... . ......... . ............. 112
EPILOGUE .......... . ................................. ....137
APPENDIX... ........... .... .............................. 111

REFERENCES .............................................. iix

INTRODUCTION
"All the matters about which science speaks, whatever the
science be, are abstract, and abstract things are always
clear. 80 that the clarity of science is not so much in
the heads of scientists as in the matters of which they
speak. What is really confused, intricate, is the concrete
vital reality, always a unique thing. The man who is
capable of steering a clear course through it, who can
perceive under the chaos presented by every vital
situation the hidden anatomy of the movement, the man, in
a word, who does not lose himself in life, this is the man
with a really clear head.”-Jose Ortega y Gasset
This dissertation is an experimental analysis of certain formal
models and hypotheses concerning single issue collective choice
situations. The modifier ‘formal’ has a precise meaning here. By
‘formal’ I mean that these structures are derived from.an axiomatic
base by virtue of a ‘first-order predicate calculus with equality’,
or, in less grandiose terms, they are grounded in logical or
mathematical principles rather than empirical experience. This
condition of formalism in scientific practice was initially specified
by the ‘Vienna Circle’ at the birth of the logical-positivist
philosophy of science. Even in so doing, the ‘founding fathers’ of
logical-positivism.reserved the determination of the meaningfulness of
such structures to observational reference. The working out of that
connection between the formal and the empirical has been the
fundamental problem for philosophy of science ever since. I mm
certainly not proposing to solve this problem in this dissertation.
However, what I do propose has a very real claim to a piece of that
problem; at least with respect to the practice of political science.

Formal modeling in political science is still a fairly recent

phenomenon. It is hard to fix a date for its beginning. Still, whether

2
we choose Arrow [1963], Black [1958], Downs [1957], or some other
auspicious point, certain characteristics stand out in most of the
candidates. Notably, all of the authors just mentioned are trained as
economists rather than political scientists. The theoretical
orientation of their models, understandably, reflects concerns
borrowed from microeconomics (e.g., equilibria and efficiency). What
makes their models ‘political’ is the subject matter to which they
are applied. The universe of discourse is the public choice setting
rather than the private market. The translation of microeconomic
principles and designs to this environment has been anything but

smooth.

The branch of political science most concerned with formal modeling
often goes by the nom de plume of ‘positive theory’. Given that the
major results published under this heading are profoundly ‘negative’,
a more ironic title is hard to imagine. As Aldrich and Ostrom
[1980:8701 put it:

"Perhaps the single most important goal for anyone
interested in the substance of the [formal modeling]
enterprise is the search for the nonobvious conclusion. In
political science, the search for the nonobvious has often
revealed not just the surprising discovery, but the
paradoxical, the anomalous -indeed, even the ethically and
philosophically disturbing."

The paradoxical situations confronted by formal modelers in
political science are almost always two-sided. 0n the one hand,
structures which, by virtue of their roots in microeconomic theory,
champion equilibrium.and efficiency and up predicting disequilibria

and inefficiency. 0n the other hand, these dire predictions are seldom

borne out empirically -at least not to the extent that might be

3
expected from the models. This double-bind has led some commentators
to dismiss the entire enterprise of formal modeling as irrelevant to
the concerns of political science. While such an assessment seems
rather extreme, it cannot be denied that this is a curious state of
affairs indeed.

However tempting it may be simply to say that economics are
economics and politics are politics and ‘never the twain shall meet’,
this ignores the real issue of formalism. Certainly, political
questions require political answers. Still, to accept this does not
carry with it a necessary rejection of formal modeling. The canons of
logic and mathematics are the same for both political scientists and
economists. The ‘truth’ of formal models is a definitional truth, and
is, therefore, not subject to empirical refutation. Thus, the source
of paradox does not lie in the practice of formalism.itself.

If political scientists have erred in borrowing the techniques of
formal modeling -and I am.not altogether sure that they have - perhaps
the source of that error is given in the fact that with these
techniques they have also borrowed the values and expectations of
economics. The desire for market optimality and equilibrium is
valuable, but it should not be allowed to supplant the fundamental
questions which undergird the discipline of political science and
distinguish it from other disciplines. These questions harken back, in
various forms, to Lasswell’s notion that politics is the science of
‘who gets what, when and how’. We, as political scientists, should not
apologize for our own disciplinary orientation to these issues, nor
should we seek to be anything other than what we are. To be a

political scientist involves a concern for ‘who’ benefits from.the

4
political process and ‘how’ they achieve their benefits. Economists.
may be content to regard the outcomes of public choice as anonymous
mathematical artifacts to be evaluated in terms of their mathematical
properties, we cannot do so.

Viewed from.this perspective, the exposition of apparent paradox has
a certain heuristic utility. When formal models demonstrate that
‘market-like’ equilibria and optimality are rarely reproduced in the
public choice setting the lesson to be learned from this is found in
the examination of the institutional differences between the two
environments. Rather than rejecting the entire enterprise of
formalism, we reject the exclusive reliance on those standards and
assumptions of microeconomics which are specific to market exchange.
In so doing, we not only acknowledge that political environments are
somehow different from the private market, we further allow for the
development of new standards and assumptions which better reflect the
relevant institutional characteristics of the public choice setting.
While this step is a necessary precondition for the exposition of
models of public choice which are at once both ‘formal’ and truly
‘political’, it is, nonetheless, a radical step.

A concern for the inclusion of relevant institutional
characteristics in the design of public choice models is not new
(despite the fact that such concerns are frequently attributed to a
school of thought with the title:”the new institutionalism"). Buchanan
and Tullock [1962] present early arguments for consideration of
institutional factors in public choice. Rae [1971] compares various
electoral settings in an effort to disentangle the particular effects

of different voting procedures. Niskanen [1971] develops a model of

5
bureaucratic/legislative interaction which explicitly includes
institutional factors as essential explanatory variables. To this list
we might add the names of Shepsle [1979], Fiorina [19781, Weingast
[1979], and many others. Still, despite this large and growing
literature, resistance to an ‘institutionalist’ view of public choice
is well entrenched and not easily dislodged. Indeed, some of the works
just cited, while including institutional factors in the models,
persist in the use of notions of market-like equilibrium and

efficiency as the standards of success.

The -often vain- quest for market-like mechanisms which yield
stable, Optimal equilibria in the public choice setting continues
space. It is buoyed up by the apparent stability manifested
empirically in many public choice environments. While the desire to
provide market-style explanations of this stability has occasionally
led to errors in method, the more common result is a peculiar tendency
toward ‘theoretical gymnastics’ -often bordering on the absurd. In
reading much of the public choice literature one often gets the
feeling that a modification here or an altered assumption there would
render visible some sort of political ‘invisible hand’. In any case,
this devotion to a purely economic analysis of the public choice
setting is a persistent ‘blind-spot’ in many scholarly pieces.

With respect to these matters, this dissertation can be regarded as
an ‘essay in craft’. Selecting a particular public choice environment,
I propose to demonstrate the increased analytic power afforded by the
rejection of a narrow economistic style of interpretation. In place of

this style, I offer what I regard as a step toward a truly ‘political’

6
treatment of that environment. Of course, a direct comparison between
these two approaches would be misleading. The development of a
‘political’ style of analysis represents a quantum jump from earlier
economic interpretations. This is because that which is problematic to
a political scientist may not be so to an economist. Nonetheless, I
freely admit that my idea of a ‘political’ approach to public choice
can do no better at finding the answers to standard micro-economic
questions than the best efforts of a corps of welfare economists. More
to the point is the fact that both these questions and their possible
answers are irrelevant. The argued superiority of a ‘political’
approach is established with respect to its ability to deal with
‘political’ questions.

I have chosen to examine the literature associated with the public
selection of levels of taxation for the provision of schools and other
collectively consumed goods and services. While this area of study
provides a fairly narrow focus, there are many factors present in this
environment which are favorable to a ‘purely economic’ approach and
are not present in other public choice settings. The domain of
alternatives is, at least overtly, unidimensional; thus avoiding the
complications and disequilibria found in multidimensional settings
(see: McRelvey [1976] and Plott [1967]). Within this domain, the range
of alternatives manifests a natural ordinality expressed in mills of
tax per dollar of assessed property value. It is reasonable to assume
that this ordinality extends to the preference profiles of individuals
within the electorate. This factor serves to further stablize the
environment by reducing the likelihood of a problem of cyclical

majorities. Moreover, this area of study has attracted the attention

7
of authors from the ‘institutionalist’ camp as well. Thus, the
selection of this narrow area may provide a crucial test of the
relative power of the two perspectives.

In final form, this dissertation is organized into five chapters and
an epilogue. In the first chapter I present a critical analysis of
many of the published arguments concerning the tax referendum/public
choice environment. My analysis is not be limited to a simple review
of the literature. Rather, I attempt to highlight the strengths and
weaknesses in many trains of thought. I am particularly interested in
the rare attempts to apply theoretical models to empirical data. In
these cases knowledge claims have been made of an explanatory nature.
What I question is the plausibility of these explanations. From.this
analysis I attempt to locate crucial assumptions which may appear
implausible or, at least, vulnerable. Once located, these assumptions
are abstracted from the literature and provide the basis for a series
of theoretical models.

An account of the development of these models and their rendering
into a set of testable hypotheses is given in the second chapter. In
the third chapter I present a description of the research designs and
analytic methods employed in the actual testing of the hypotheses
developed in the second chapter. Both of these chapters are relatively
short and their contents are complementary. Nonetheless, between them
they capture the essence of this research. While the first chapter
yields little more than an outline of the argued weaknesses
encountered in a narrow micro-economic interpretation of the tax
referendum setting, the second and third chapters must make a case for

the promised increase in analytic power afforded by the abandonment of

8
that approach and sets the terms under which this argument may be
judged.

The testing procedures I have in mind involve experimental designs.
The experimental setting provides a unique opportunity to isolate the
pertinent assumptions of any theoretical model; they can stand out in
a sharp relief which is seldom afforded in ‘real world’ data.
Nonetheless, I freely acknowledge that my experimental results are not
‘real world’ data and, as such, cannot establish the final empirical
truth of the matter. Whether we take this as a strength or weakness of
experimentation depends upon the motives of the experimenter. In one
sense, as stated above, any well derived formal model is true before
it encounters the ‘real world’. Thus, in the analysis of these formal
models, truth is not the issue. What matters here is just how useful
these structures prove to be in the task of interpreting that ‘real
world’. This usefulness is precisely what is often lost when formal
models are applied empirically.

The criteria of ‘usefulness’, themselves, reflect the motives of the
researcher. My primary motive here is to develop a better
understanding of the tax referendum/public choice setting as a
political environment. The study of politics is, at root, a science of
‘winners’ and ‘losers’. That is to say that, a political understanding
bespeaks a game-theoretic orientation. various actors fill separate
and sometimes multiple roles within the environment. To each actor we
may attribute a set of goals which can be summarized in a utility
function. Individual behavior in pursuit of these goals reflects the
strategic interaction between actors. The institutional conditions of

the environment impose limits on this behavior and, thereby, set the

9
strategic balance. Given this, I judge a model to be useful if it
illuminates certain characteristics of the strategic balance in the
tax referendum setting and brings into focus various possiblities of
strategy possessed by various actors. This notion of usefulness
undergirds my presentation of the results of my experiments in chapter
four.

Just as ‘the proof of the pudding is in the eating’, so the proof of
the usefulness of my experimental results can only be ultimately
established through the successful application of the concepts and
patterns found there in an actual tax referendum environment. Choosing
the appropriate format for such a setting is a ponderous task in and
of itself. Were I to follow the general practice of most works in the
literature, I would design a broad based econometric model to be
applied to a large data set. However, application to a large data set
might only serve to obsure strategic and case specific characteristics
which may prove crucial in determining the election results. At this
stage in the development of our understanding of the tax referendum
environment I believe that a carefully chosen case study will yield
more information than an aggregate design.

The fifth chapter of this dissertation presents such a case sudy. In
looking for a ‘crucial case’ many legal and institutional factors
recommend school districts in the state of Michigan. Even though the
models developed here take a single election format, the selection
criteria included consideration of the electoral history of the
particular school district. Also included was the notion of a good
balance between urban and rural populations within the electorate so

as not to introduce a bias from this factor in the analysis. With all

10
these factors in mind, the September 2, 1980 millage election of the

Sault Area Public School District was chosen as an appropriate case.

CHAPTER 1

A CRITICAL OVERVIEW OF THE LITERATURE
The Continental School

The first formal analyses of the use of public choice mechanisms in
the selection of levels of taxation are Wicksell [1896] and Lindahl
[1919]. While these two essays make distinct contributions to our
understanding of taxation, they share a central theme. Both authors
view taxation as the ‘price’ paid by members of the polity for the
output of government. This linkage implies that the selection of
levels of taxation should be viewed as an exchange between government
and individual citizens and can be analyzed in terms of ‘benefit/cost’
calculations. Any conceptual scheme for the taxation decision which
fails to include consideration of how the tax revenues are used to
provide public benefits is regarded as fundamentally deficient.

In this day and age the demand that levels of taxation be linked to
levels of government output does not seem very radical. Nonetheless,
the Anglo-American economics literature failed to give serious
consideration to this linkage until the period following the second
world war (see, Buchanan [1968]). Thus, interest in the
english-speaking countries for an exchange theory of taxation must be
regarded as a fairly recent phenomenon. Still, if anglophones were
slow to appreciate the contributions of Wicksell and Lindahl, their
zeal for the program was no less once it caught on. Indeed, the

postwar public finance literature reflects the values and aspirations

ll

12
of the ‘continental school’ (as the Wicksell/Lindahl view came to be
called) in many important respects.

Both Wicksell and Lindahl are concerned that systems of taxation be
‘just’. For Wicksell, any taxation system is ‘just’ if and only if it
is capable of gaining -at least in principle- the unanimous support of
those individuals subject to the tax. As strange as this notion may
sound, Wicksell presents a persuasive case for it. In Wicksell’s view,
a ‘just’ government functions as the representative of the collective
interests of the polity. Its purpose is to provide those goods and
services which address collective needs. If the benefits to be had
from such collective action equal or exceed the ‘costs’ of the
taxation for each and every taxpayer, then unanimous consent is
possible. Moreover, anything short of unanimous consent leaves open
the possibility that some subset of the polity will be coerced into
suffering the costs of government activities from which they derive no
benefit. As Wicksell (in Musgrave and Peacock [19671:89) puts it:

"It would seem to be a blatant injustice if someone should

be forced to contribute toward the costs of some activity which
does not further his interests or may even be diametrically
opposed to them.”

Wicksell allows that there are immediate practical difficulties with
the use of a unanimous decision rule. Every public good would have to
be financed separately, and in some contexts, the unanimous decision
rule may be wholly inappropriate. Even so, Wicksell maintains that an
ideal taxation mechanism.which is capable of gaining unanimous support
is a real possibility and worthy of exploration. It was Lindahl who,
building on Wicksell’s work, first designed such a mechanism. The

Lindahl tax scheme is both clever and conceptually simple. Lindahl is

able to satisfy Wicksell’s unanimity criterion through the device of a

13
tax pricing system which provides for the adjustment of individual
fractional tax shares and total output of a public good.

An individual’s tax share (‘ti’) is that fraction of the per-unit
expenditure for the public good (‘x’) which is borne by the
individual. An individual’s total tax is found by multiplying the
individual’s tax share by the quantity of the public good which is
actually produced: ti(x). We assume that each individual has a
utility function which includes a functional statement of the personal
utility derived from.consumption of the public good as well as other
sources of utility:

ui - vi(x) + yi
We further assume that each individual’s resources are finite and that
this is expressible in terms of a budget constraint:

bi ' ti(x) + yi
Given the standard rationality assumption (i.e., under any given state
of the world an individual will seek to maximize utility), we may
identify the quantity of the public good demanded by an individual at
any given tax price by taking the first derivative of the individual’s
utility function with respect to the public good and setting it equal
to the assigned tax share.

Assuming that we wish to finance the public good entirely from.tax
revenues, the sum of the fractional tax shares must be greater than or
equal to one. Assuming that we wish to avoid surpluses, we may ignore
the ‘greater than’. Allowing for divergent tastes, an equal
distribution of the tax burden is likely to yield markedly different
levels of demand for the public good. The Lindahl tax scheme provides

for adjustments in individual tax shares to correct this. Those

14
individuals who demand relatively less of the public good have their
tax share decreased. This, in turn, should lead them to demand more of
the public good. In order to maintain the sum of tax shares at unity,
those who demand relatively more of the public good must have their
tax share increased. This, of course, should decrease their demand.
These adjustments continue until all members of the polity who pay
taxes demand the same amount of the public good. This amount of the
public good (called the ‘Lindahl equilibrium’) is capable of gaining
unanimous support when the tax burden of its provision is ‘justly’
distributed on the basis of individual benefits through these
adjustments in individual tax shares.

By casting the taxation decision in terms of exchange theory
Wicksell and Lindahl borrow the values of the private market. The
first fundamental value of market exchange is Pareto optimality. The
aspiration to Pareto Optimality is the soul of Wicksell’s notion of
‘justice’. In the private market, Pareto optimal outcomes result from
voluntary exchange. Individuals exchange resources up to a point at
which further exchange would result in a loss of utility for some
party. Wicksell maintains the volunteerism of the private market in
the public goods environment through the mechanism of unanimous
consent. Nonetheless, Wicksell’s pronouncements on the subject are
informal. The first formal specification of a sufficient condition for
the Pareto optimal provision of public goods appears in Samuelson
[1954].

The Samuelson condition for Pareto optimal output of a public good
requires that the sum of the marginal valuations of the good (the

first derivative of each individual’s utility function with respect to

15
the good) equal the marginal costs of producing the good. Samuelson
notes that the Lindahl equilibrium meets this condition by tying tax
shares to individual marginal valuations of the good. Since the
Lindahl tax shares must sum to unity, this property extends to the
marginal valuations of all members of the polity. Hence, the Lindahl
equilibrium is on the Pareto frontier. However, the explication of
this property for the Lindahl equilibrium does not lead Samuelson to
an endorsement of the Lindahl tax scheme.

The primary flaw which Samuelson locates in the Lindahl tax scheme
stems from the fact that the scheme requires absolutely honest demand
revelation. Even so, the scheme provides a definite incentive for
misrepresentation of demand to individual members of the polity. By
definition, public goods are accessible to members of the polity on a
non-exclusive basis. Thus, an individual’s access to the benefits of
the public good is not dependent upon the size of that individual’s
contribution toward its provision. Since the size of an individual’s
tax contribution is determined by a self-reported level of demand, a
clear incentive for underrepresentation of demand exists. This
incentive is most pronounced when two factors are in play. First, if
an individual is reasonably assured that other members of the polity
have relatively less elasticity of demand, the underreporting of the
individual’s preferences will have minimal impact on the level of the
public good provided. Similarly, if the polity is relatively populous,
again, the impact of an individual’s misrepresentation will be
‘minimal. However minimal these effects may be, it is clear that less
of the public good will be produced and this entails fewer benefits

for individual consumption. Nonetheless, this loss will be more than

16
offset by the individual’s tax savings. The flaw in the Lindahl tax
scheme is precisely the so-called "free rider" problem. On the basis
of his analysis of the Lindahl system, Samuelson concludes that any
decentralized tax pricing system will be vulnerable to such
manipulations.

Faced with the immediate practical difficulties of the Wicksell I
Lindahl program, serious scholars of public finance have been forced
to accept more modest aspirations. Still, the basic tenets of the
Wicksell argument have maintained a significant influence on the
development of later models. The primary criterion of success remains
Pareto optimal output from the public sector. Following Wicksell’s
lead, the use of consentual mechanisms, such as voting, appears to be
the most promising avenue for the attainment of this result. By
allowing those who are taxed to have some say in the level of
taxation, government is held accountable for the use of tax revenue.
This provides the link between taxation and expenditures which is
established as a necessary condition for Pareto optimal provision of
public goods by the Wicksell logic. If unanimity is too strict a
requirement, then the question as to what is possible with less than

unanimous consent becomes paramount.

Majoritarianimm and the Bowen equilibrium

The most common decision rule in the public choice setting is the
rule of the simple majority. Justifications for this rule can be made
on both practical and normative grounds. Buchanan and Tullock [1962]

provides an interesting argument for simple majority decision making

17
on the basis the associated types of decision costs under different
decision rules. These authors devise a continuum of decision rules
with one pole allowing a single voter to decide for the collective and
the other pole requiring unanimous consent. At the first pole, the
‘costs’ of this type of decision rule, which the authors label
‘external’, involve the ability of some subset of the papulation to
impose their will upon the whole. A continuum of these ‘costs’
represents a scalar view of Wicksell’s notion of ‘blatant injustice’.
Given Wicksell’s logic, such ‘costs’ are eliminated at the extreme
pole of the unanimous decision rule. Even so, it is at this opposite
pole that another type of ‘costs’ are at their most extreme. This
second type of ‘costs’ can be expressed in terms of the time and
trouble required to reach a decision. Buchanan and Tullock argue that
these two types of costs tend to balance out at the middle of the
continuum. While the exact location of the ‘optimal decision rule’
will vary under different types of decisions and distributions of
preferences in the electorate, the authors claim that the simple
majority decision rule stands out as roughly ‘optimal’ for a broad
class of decision settings.

Whether or not Buchanan and Tullock succeed in making a case for the
simple majority is far from clear. Moreover, we run the risk of
infinite regress when the question of the ‘optimal decision rule’ must
be decided prior to the question of the ‘optimal decision’. I bring
these matters up merely to illustrate the extremes possible when
notions of what is desirable become the only appropriate topics for
formal works in public choice. In so doing I wish to set the tone for

the following sections of this chapter. Having said this, it is

18
sufficient for all practical purposes to recognize that the rule of
the simple majority is prevalent, and build a body of theory from this
basis.

The first serious analysis of the taxation decision under simple
majority rule appears in Bowen [1943]. Bowen argues that demand for
public sector output is directly analogous to demand for private
sector output, but collective consumption mandates that a single level
of output be selected. This can best be done through a voting
mechanism. Assuming that all possible levels of provision for some
public good are available to the electorate, under a simple majority
decision rule, "...one immediate amount ...will be voted for by more
individuals than any other single amount."[p.37] This level of output
is the quantity most preferred by the individual at the exact median
of the distribution of preferences for the entire electorate. Thus,
the ‘Bowen equilibrium’ is a ‘median voter’ result.

Political scientists usually associate the median voter result
with Downs [1957] and Black [1958], but rarely acknowledge Bowen’s
contribution. This is understandable given that Bowen’s treatment of
the median voter result is fairly informal. Indeed, Bowen seems to
simply assume its validity as a preface to a broader theory of the
taxation decision. A formal understanding of the conditions for the
median voter result is provided by Black. The primary condition for
the existence of this result is given in Black’s notion of
‘single-peaked’ preferences. By definition, "...a single-peaked curve
is one which changes its direction at most once, from up to
down.”[1958:7] If the alternatives available to the electorate can be

ordered along a continuum such that the preferences of the electorate

19
can be represented as a series of single-peaked curves, the highest
point on the preference curve of that voter at the exact median of the
distribution of preferences is a Condorcet winner (i.e., it is the
only alternative which can defeat all others in pairwise elections
using majority rule).

With respect to the taxation decision, a natural continuum exists
from lesser to greater levels of taxation. This, in turn, is taken as
representative of lesser to greater levels of public goods provision.
If an individual ‘most prefers’ some particular level of the public
good, it is reasonable to assume that any movement away from that
level in either direction along the continuum.will lead to decline in
preference. Hence, voter preferences will naturally tend to be
single-peaked with respect to taxation. This usually cannot be said
about other public choice environments. Preferences within the
electorate for candidates, ideologies, or specific platforms seldom
lend themselves to such orderly analysis. Since the problematic nature
of the single-peakedness property is less apparent in tax referenda,
Bowen’s failure to discuss it formally is perhaps understandable.

As should be clear from the above, Bowen’s primary concern is not
the identification of the equilibrium point that bears his name,
rather, most of his discussion focuses upon the specification of
conditions under which this result will be economically efficient
(read Pareto optimal). The Bowen model assumes: 1) a single dimension
of choice given in separate taxation to finance each separate public
good; 2) all members of the polity vote and vote sincerely in accord
with their personal demand for the public good; 3) all costs of

providing the public good are known; 4) whatever level of the public

20
good is eventually chosen, the costs of its provision will be divided
equally amoung all members of the polity; and, 5) demand for the
public good is normally distributed throughout the polity. From this
assumptive base Bowen provides three results. If the per-unit cost of
providing the public good is constant, the median voter result will be
Pareto efficient. Given economies of scale, if the cost of provision
is decreasing, again, efficiency will be a characteristic of the
median voter result. However, if costs are increasing, efficiency is
not assured.

We need not dwell on the means by which Bowen reaches these
conclusions since it is clear that none of his five assumptions are
above reproach. Still, despite the implausibility of Bowen’s
assumptions, his argument contains some interesting and useful
insights into the practical environment of the taxation decision.

Bowen acknowledges that, in the ‘real world’, the electorate is not
presented with a choice between all possible levels of taxation or
public sector output in any single election. For the most part, the
voter’s decision involves simply voting ‘yes’ or ‘no’ to a total
package which is proposed by government. Since governments usually
propose only one such package in any given election, the ability of
the procedure to locate the median voter equilibrium is seriously
constrained. However, if we look beyond the single election format and
view the taxation decision as a process of incremental adjustment, the
median voter equilibrium can be established in the long run.

Bowen’s incremental adjustment model views the taxation decision as
a series of elections. Assuming that the polity begins the process

with some initial allocation of the public good (which may be as

21
little as none), each election represents a pair-wise comparison
between this level and some new level proposed by government. The
winning package in an initial contest may be subjected to further
challenges by other alternatives in the continuum. With repeated
challenges, the median voter equilibrium, being a Condorcet winner,
may be located by a process of elimination.

The exposition of this incremental adjustment model is an important
step for the Bowen argument and an equally important contribution to
our general understanding of the tax referendum environment. In one
stroke it provides the basis for empirical application of Bowen’s
theory and raises crucial questions with respect to the institutional
factors which govern tax referenda. Within this model, government does
not passively respond to the expressed demand for the public good in
the polity. Rather, government must take an active role in structuring
the agenda of alternatives through a series of elections. What
procedures do or should governments follow in the fulfillment of this
function? What motivates the activities of government in this role?
What effect does the location within the continuum of that initial
allocation have upon the playing out of the process? At what point
should the process stop? What effect does the order in which
alternatives are considered have upon the playing out of the process?
Regrettably, these questions are left unasked in the Bowen piece.
Indeed, Bowen seems to treat the entire issue of incremental
adjustment as an aside -a begrudged concession to the ‘real world’

which distracts attention from his quest for Pareto efficiency.

22

Empirical Application of the Median voter Model

The search for appropriate empirical environments to apply the Bowen
model (and other median voter models) has focused upon local community
decision making with respect to expenditures. The preference for the ‘
local government setting is keyed to several simplifying elements
which are found there and not at the higher levels (i.e., national
governments). The relative homogeneity of local populations, the
directly quantifiable nature of local expenditures, the limited array
of such expenditures, and the immediacy of the impact of such
expenditure decisions are often given as reasons for this preference.
Still, there are deviations from this theme. To a certain extent,
Bowen’s article can be viewed as the first attempt to use a median
voter model to interpret empirical data. In support of his notion that
demand for public goods is normally distributed throughout the polity
he examines data relating to demand for private goods; finds that
marginal valuations tend to be normally distributed; and concludes
that this pattern can be reasonably assumed for public sector output
as well. Even so, Bowen’s work, at best, should be regarded as a
partial and tentative application of the logic of the model to
empirical data. Bowen uses the data to shore up an assumption of the
model, but, otherwise, refrains from.aubjecting the model to critical
testing.

A more elaborate empirical exposition of the median voter model is
given in Barr and Davis [1966]. These authors employ the model as a
descriptive device ininterpreting data concerning local government

expenditures in various communities of Pennsylvania. Arguing that

23

local revenues are primarily derived from local property taxes and
assuming that the preferences of the median voter are decisive in
setting local expenditure policy, Barr and Davis use per capita
assessed valuation of taxable real property and the percentage of
property owners as surrogates for the tax price of public goods
provision. Relative increases in the proportion of the electorate who
own taxable property increase the probability that the median voter is
a property owner and must bear the costs of public goods provision.
This should, therefore, lead to a relative decrease in median demand
and, hence, lower expenditures. On the other hand, relative increases
in assessed valuation of taxable property within a community indicate
greater wealth. If the median voter is relatively wealthier, he or she
should have greater demand and, hence, higher levels of local
expenditure should result. Both of these expectations are borne out
empirically with significant coefficients of the expected sign.

However gratifying it may be to have their expectations confirmed in
the data, the Barr and Davis work cannot be regarded as a test of the
median voter model. The results indicate that local expenditures tend
to rise and fall in the expected directions with the surrogates for
tax price. While these results are consistent with the expectations of
the median voter model, they are also consistent with any number of
other possible designs. No attempt is made to locate median preference
within the communities. Thus, we do not know if the ultimate levels of
expenditure are greater or less than the median preferred level.
Moreover, Barr and Davis simply assume that the median preference is
decisive in determining these ultimate levels. It is quite possible

that some other group in the electorate is actually decisive. Given

24

similar variation in the tax price surrogates, the results would be
consistent with the mean income voter -or, indeed, any member of the
electorate- being the actual decisive voter. Thus, the Barr and Davis
piece does little to advance our practical understanding of the
workings of the median voter model. Nonetheless, as Homer and
Rosenthal put it: ”The value of the Barr-Davis contribution is not in
the empirical results, but in the stimulus provided by their attempt
to link empirical work with a formal model of the political process.”
[1979:1491

Most of the works which have attempted to apply the median voter
model to the taxation decision after Barr and Davis can be grouped
into two categories. The first such grouping shares the telos of
attempting to demonstrate the decisiveness of median voter preference
in empirical data. Alternately, another set of authors have tried to
make some determination of the relative efficiency of a median voter
result. I shall briefly summarize these works, dealing with the former

category first.

111' 1...,

Bergstrom and Goodman [1973] present a substantially refined view of
the local expenditure demand function which makes use of notions of
median preference as essential elements of the model. The model take a
log-linear form with the key variables being household income and tax
price. In this form elasticities are assumed to be constant and equal
for all members of the polity. In so doing, the authors identify the

median voter as having median income. They further assume that "...the

25
citizen with median income lives in and owns the house with median
value and that his house constitues his entire holdings of real
property."[p.284] Since the majority of local expenditures are
financed via local property tax, the authors use a calculation based
on median assessed valuation within a community to stand for their tax
price variable.

Bergstrom and Goodman use data derived from ten states to apply
their model. In deference to institutional differences in taxation
procedures between states, data from each state is first treated as a
separate sample, then pooled. Moreover, the authors consider three
types of local expenditures separately: general (excluding education
and welfare); police; and parks and recreation. The results offer
modest support for the notion that median voter preferences (as
defined by the the authors’ assumptions) are decisive in the final
expenditure decision. The regression coefficients for the crucial
variables, which stand as estimates of income and price elasticities,
tend to behave in the expected manner. In the pooled series, estimates
of income elasticity are always significant and positive, while price
elasticity is always significant and negative. Breaking the data up
by the separate states yields occasionally insignificant results, but
the authors are quick to point out that "...[i]n no case was an
estimated elasticity coefficient both significant and of perverse
sign."[p.287]

Given the apparent success of the Bergstroeroodman model, it became
the basis for a series of later works by other authors (e.g.,
Pommerehne and Prey [1976], Pomerehne [1978], Pommerehne and

Schneider [1978], and Inman [1978]). Nonetheless, the model is not

26
without its problems. On the face of it, the model might be judged
superior to the Barr-Davis formulation in that notions of median
preference are explicitly included in the explanatory variables rather
than simply assumed. Even given this, we might quibble with the
Bergstroeroodman specification of median tax price. Since most
renters are on the low end of the income distribution, the proportion
of the polity who are homeowners should be shifted up the income
scale. Thus, the median income individual probably owns a house of
less than median value. This could introduce a systematic bias in the
estimates of price elasticity. That the estimates behave well despite
this does not reflect well on the notion of median decisiveness.
Moreover, the model does not provide for direct tests against specific
alternative hypotheses. To a certain extent, Pommerehne and Frey
[1976] and Pommerehne [1978] attempt to correct this by estimating the
model using statistics based on the mean. The comparisons between
models based on the mean and the median are not statistically
conclusive however and much of the justification for the use of the
median relies upon a preference for theory-based research over
atheoretical specification.

On another plane, the results of comparisons between the mean and
median in Pommerehne [1978] are, themselves, interesting. Using data
from.various Swiss municipalities, Pommerehne found that median
statistics performed better in 48 direct democracies. However, in 62
representative democracies, the measures using the mean proved
statistically superior. Homer and Rosenthal [1979], in their own
critique of the Pommerehne results, suggest that this discrepancy may

be attributed to differences in measurement. The median statistics

27
were calculated using only Swiss citizens, while the mean statistics
used the entire population. Noting that about 202 of the population
are foreign residents, Homer and Rosenthal argue that the results
indicate that the preferences of non-citizens get little consideration
in direct democracies. Still, this explanation assumes that the
preferences of foreign residents with respect to local expenditures
are substantially different from those of the citizen population, and
that representative democracies are somehow more responsive to these
preferences. Neither assumption is overly persuasive in that: 1) alien
or citizen, both receive similar benefits from the expenditures at
stake; and, 2) since aliens do not vote in either case, it is hard to
see why elected representatives would champion alien interests to the
detriment of their own constituency.

Perhaps a better explanation for the discrepancies in the Pommerehne
results is found in the differences between the types of collective
decision making employed in the two environments. While Bowen takes
pains to argue for the applicability of his median voter model in
representative or indirect decision settings, it is clear, even in his
original article, that the institutional conditions of direct
democracies provide a more solid basis for the expectation of the
Bowen equilibrium. In the American setting, given the decline of the
town meeting, there are very few instances of direct democracy. One
possible exception is the use of p0pular referenda to set tax rates
for the provision of public education. Since educational expenditures
generally take up the lion’s share of local budgets, this particular
decision setting is clearly important. Moreover, these tax referenda

are usually limited to educational matters. Hence, a single issue

28
format supports the notion of unidimensionality required by the median
voter models. All this has led Romer and Rosenthal to proclaim: "...if
researchers are ever to find an empirical context that is
approximately unidimensional and single-peaked, school expenditures
appear to be an appropriate choice." [1979:1471 While the
Bergstrom-Goodman piece explicitly excludes educational expenditures
from the analysis, later works which apply the Bergstroeroodman model
to American data limit themselves to this area of application. A case
in point is Inman [19781. Inman is concerned with directly testing an
ancillary set of assumptions within the Bergstroeroodman model. In so
doing, Inman uses tax referenda results from 58 Long Island school
districts as a data set.

The original specification of the Bergstroeroodman model does not
allow for great diversity of tastes with respect to expenditures
between communities. That is to say that, by explaining levels of
expenditure in terms of median income and tax share other cultural and
psychological factors which might effect median preference are
overlooked. Bergstrom.and Goodman recognize this weakness in their
model and develop a theorem to account for taste specific differences
between communities. The theorem assumes that any given community may
be disaggregated into socio-cultural subgroups. These subgroups may
differ in tastes from one another, but are assumed to have consistent
tastes across communities. Thus, the impact of diversity of tastes can
be specified as a shift factor derived from the proportional make-up
of various subgroups within a given community. This shift factor will
have a multiplicative effect upon the median level of demand.

Nonetheless, rather than explicitly trying to incorporate this notion

29
in their model, Bergstrom and Goodman simply include a set of
demographic variables to stand for the shift factor.

While the Inman study attempts to directly estimate the shift factor
described by Bergstrom.and Goodman, Inman’s motives for doing so may
not fit precisely with the Bergstroeroodman plan. Rather than using
the estimated shift term to improve the predictive power of the model,
Inman attempts to demonstrate the shift term’s irrelevance. In so
doing, he hopes to make the case for the use of median income,
unmodified, as the ideal surrogate for median voter preference. The
means Inman employs to make his case come down to standard hypothesis
tests of significance on the estimated shift term, ‘s’. If s - 1, then
ln(s) ' 0. An ‘s’ equal to one would provide no multiplicative effect
in the predicted expenditure outcome using median income. Thus, ln(s)
. 0 is used as the null hypothesis. Inman estimates ln(s) using
ordinary least squares and two-stage least squares for each of his 58
school districts. None of the 58 OLS estimates prove significant at
951, and, while 13 TSLS estimates acheive 951, they fail at 99!. Given
these results, Inman boldly claims to have ‘buried politics’,
asserting:

"...we can model local fiscal choices as an individual utility
maximizing problem and transfer to government behavior the
usual theorems of consumer choice. Politics becomes economics.
This is a powerful conceptual advantage, and one which
justifies much of the theoretical work to date on the behavior
of local governments.” [p.591

Despite such claims, it is clear that in Inman’s zeal to ‘bury
politics’ he has forgotten the elementary logic of hypothesis testing.
In failing to reject a null hypothesis, at whatever level of

significance, we are not allowed thereby to accept the null hypothesis

as the truth of the matter. To do so would commit the fallacy of

30
affirming the consequent. At best we may allow that the null
hypothesis is one of a theoretically infinite number of plausible
candidates. In short, no substantive conclusion can be made. To
paraphrase Mark Twain, the reports of the death of politics appear to
be premature.

Inman’s results are further complicated by specification problems in
the model itself. As in the Bergstroeroodman model, all income
distributions are assumed to be lognormal and each subgroup
distribution is proportional across cities. Inman further assumes that
subgroup distributions are proportional to the overall distribution
for each city as a whole. Romer and Rosenthal [1979] demonstrate that
these assumptions trivalize the use of median statistics since, given
them, parameter estimates (excluding the intercept) would be unchanged
regardless of which fractile of the income distribution is used. We
might as well use the voter at the 55th (or any other) percentile.
Homer and Rosenthal rightly conclude that:

"...Inman’s method is caught in a crossfire. If a median
regression proves superior to the use of other fractiles, the
assumptions used to justify the model are contradicted. On the
other hand, if all fractiles give similar results, then any
voter appears ‘as if’ decisive." [p.153]

The discussion above should serve to highlight some of the
difficulties encountered when we try to assess the notion of median
decisiveness empirically. Several problems stand out as apparently
insoluble in the literature discussed so far. First, of course, no
means has been found to demonstrate median decisiveness empirically.
The best efforts of Bergstrom and Goodman [1973] and Inman [1978],

while explicitly including median statistics in the specification of

the model, ultimately can do no better than simply assuming the

31
dominance of median preference. Moreover, even assuming median
decisiveness, these models do not demonstrate that the expenditure
level selected is that level most preferred by this pivotal voter. The
most reasonable assumption is that people vote in accord with their
preferences, but we lack the omnipotence to measure these preferences
directly. In place of such measurements these authors have used
surrogates such as income, assessed valuation of housing, and
proportional membership in various socio-cultural subgroups (to stand
for divergent tastes). The resulting regressions, at best, night
demonstrate a functional relationship between the selected level of
expenditure and these variables (perhaps even median values of these
variables). Nonetheless, this is a far cry from the assertion that the
median voter receives his or her most preferred outcome. The function
in question might yield outcomes which are ten times greater or twice
as small as the median voter’s ideal.

Is the median voter decisive? If so, does the median voter recieve
his or her ideal outcome? The authors cited above have not answered
these questions, and, in truth, we may never be able to answer them in
empirical terms. Still, the pursuit of these answers does not exhaust
the possible empirical applications of median voter models of the tax
decision setting. At this point, I shall briefly discuss another
parallel set of empirical research which has been oriented toward a
different question. Specifically, this set of authors have tried to
assess the relative efficiency of a median voter result in the tax

referendum.environment.

32
I I] 1. 1 . 17

Barlow [1970] presents a truly seminal attempt to assess empirically
the relative efficiency of a median preferred allocation of local
educational expenditures. In doing so, it should be noted that Barlow
does not succeed in demonstrating median decisiveness or that median
ideal preference is guaranteed. Rather, Barlow simply assumes these
notions in order to examine the efficiency characteristics of such a
result. In a similar fashion to those authors cited above Barlow
assumes that preference is primarily a function of income and tax
price (as measured by assessed valuation cf real property). Thus,
individual demand for education is summarized by the equation:

a b
Qi a kYi Pi

Where ‘Q’ denotes a preferred quantity of education; ‘Y’ denotes
income; ‘P’ denotes tax price; and ‘a’ and ‘b’ denote elasticities of
demand. Demand elasticities with respect to price and income are
assumed constant, as is the term ‘k’. Given this equation, the
marginal benefit for any individual ‘i’ at any given ‘Q’ can be found
by solving for the price ‘P’ that this individual would willingly
spay: I

a l/b
Pi - (Qi I kYi)

Further, any individual ‘i’ receives a share of marginal social

benefit as defined by the equation:

-a/b -a/b
Pi/ ZPi ' Yi ' IZYi

Barlow notes that the share of social benefit received by ‘i’ does not

33
depend upon the particular ‘Q’ selected.[p.1033] This is the result of
assuming that demand elasticities with respect to price and income are
constant.The implication is that shares of social benefits also remain
constant for any given ‘Q’ and are set by an exogenously determined
tax pricing scheme. Thus, an individual’s share of marginal social
benefit would be the same for an efficient ‘Q’ as for an inefficient
‘Q’. This becomes the crucial factor in assessing the efficiency of a
median preferred result.

Assuming median voter decisiveness and given Samuelson’s conditions
for optimal allocation in a public good, Barlow’s analysis shifts to
the question of whether this crucial voter will select a socially
optimal allocation. Barlow argues that the median voter will select a
socially optimal allocation of the public good if and only if the
voter’s share of the tax burden of providing the good is the same
fraction as this voter’s share of marginal social benefit. Thus, a
ratio between the median voter’s share of the tax burden and his or
her share of marginal social benefit must equal unity. Moreover, the
extent to which this ratio varies from.unity (in either direction)
will indicate the extent to which the good is over or under provided.
By making the assumption that the median voter is the individual with
median income, Barlow is ready to estimate the efficiency of the
median voter result empirically. All that is needed is a data set
which provides information concerning income, tax burden and
the size of the tax base for any given polity.

Barlow makes use of data provided by Musgrave and Diacoff [1958] for
Michigan in 1956. The data is grouped by income into seven groups.

Using proportions within groups as provided by 1960 census data,

34
Barlow is able to estimate the location of the median voter through
interpolation of the data. Estimates of elasticities of demand are
made from.cross sectional data for Michigan school districts. Given
this, Barlow is able to produce estimates of median tax burden, and
the median share of marginal social benefit. The resulting ratio is
estimated at roughly 0.5. Thus, Barlow concludes that the median voter
result for Michigan in 1956 would provide education at a level below
the social optimum. Given an estimated price elasticity (b) of -0.34,
Barlow calculates that the change in median tax burden required to
raise the ratio to unity would produce a 26 percent increase in median
demand. This defines the extent to which the median voter result falls
short of efficiency.

Barlow’s model provoked a series of comments, modifications and
critiques. Bergstrom [1973] noted that the Barlow results are
crucially dependent upon the shape of the income distribution and the
relative progressiveness of the prevailing tax structure. Given
modifications in either or both of these factors different results
would maintain. These observations, of course, hardly count as a
biting critique of Barlow’s work. Indeed, Barlow is quite aware that
the final estimates reflect the standard notion that "...residential
property tax is steeply regressive with respect to income..."
[1970:1035]. Nonetheless, by allowing for variation in the tax
structure, Bergstrom is able to specify sufficient conditions for
results which both yield and exceed the social optimum. In so doing,
Bergstrom.aot only demonstrates that allocations below the social
optimum do not necessarily follow from.majority choice systems, he

further underscores the importance of prevailing institutional

35
factors in determining the outcomes.

On another tack, Edelson [1973] questions Barlow’s specification of
the problem on several levels. Edelson notes that public education is
not entirely financed by taxes levied on voting members of any given
school district. Rather, in most states, various government general
funds provide both lump sum and matching grants which supplement the
resources generated by local property taxes. In addition, business
property provides yet another source of support from.aonvoters. The
existence of such external sources of financial support should effect
the ultimate allocation of education in the district and, thereby,
have an impact upon calculations with respect to relative efficiency.
Further, Edelson argues for the abandonment of the individual
household as the unit of analysis in favor of calculations relating to
individual children of school age within the district. This notion
reflects Edelson’s rejection of the treatment of publiceducation as a
Samuelsonian "pure public good".

In principle, a "pure public good" is held in common and one
individual’s consumption does not limit the consumption of others. In
this sense, it should be both nonexclusive and inexhaustible. Despite
collective and egalitarian provision, public education does not meet
either condition. Thus, Edelson argues, calculations with respect to
public demand (particularly income and price elasticities) should
reflect individual consumption. Since elasticities of demand play a
crucial role in the estimation of the benefit/burden ratio for the
median voter, any shift to calculations on the basis of ‘per pupil
expenditures’ could substantially change estimations of the relative

efficiency of the median result.

36

The move toward parameter estimates which reflect individual
consumption introduces a profound degree of complexity in the model.
As Edelson notes:

”It is obviously unrealistic to assume that each voter shares
the same fraction of total output. Certain voters do not use
the public schools at all (parents with children in parochial
or secular private schools, aged couples, unrelated
individuals), while others have different numbers of children
enrolled in public schools." [p.168]
Edelson disregards the second source of variation by simply assuming
that the average number of children per voter is constant across
income classes. However, the proportion of voters who derive no direct
benefit from public education proves to be a thorny problem indeed. To
deal with it, Edelson imposes a set of assumptions concerning voter A
behavior which he supports through reference to demographic data from
the U.S. Department of Commerce. Given this data, 13 percent of the
voters in each income class are assumed to vote against any school
budget beyond a constrained minimum. Since there is a greater
proportion of private school users in the upper extreme of the income
distribution, an another 17 percent of the highest income class is
added to this corps of ‘no’ voters. Armed with this new battery of
assumptions, Edelson dutifully cranks out new estimates of the model
parameters using the Barlow data.

While the notion that some voters do not derive direct benefits from
the public school system presents little more than a complicating
factor in Edelson’s respecification of the Barlow model, Barzel [1973]
makes it the keynote of a fundamental attack upon Barlow’s entire

enterprise. Barzel reasons that voters who make use of private or

parochial schools will have little or no demand for the public school

37
system. The Barlow model assumes that demand for public education will
always increase with increases in income. However, given that use of
private schools also tends to increase with income (particularly in
the highest income group). this assumption cannot hold. As a result
the use of median income to define median preference is precluded; the
crucial voter should, in general, have less than median income.

How does the shifting of the location of the crucial voter effect
calculations with respect to the efficiency of the median voter
result? 0n the surface we might conclude that, if the median preferred
quantity of public education under the Barlow model was less than
optimal, any shift to lower levels of the income distribution should
yield an even greater shortfall. Still, Barzel rejects this
interpretation out of hand:

"When the higher~income households move out of the public
school system, they themselves now obtain whatever amount of
education they desire. Their tax contribution then is basically
a transfer payment- entirely from.the rich to the rest of the
community. The clear-cut regressive nature of financing public
schools through property taxes present in Barlow’s model no
longer holds." [p.178]

Does this imply that the ultimate allocation is, in fact, closer to
efficiency? Barzel points to several factors which inhibit any
straightforward answer to this question. Lacking specific information
with respect to the income distributions within the higher income
groups, we cannot locate the "break-even” point at which households
opt for private alternatives to public education. Moreover, given
diverse populations between the several school districts in Michigan,
any pronouncements concerning the relative efficiency of allocations

can only be made at the school district level and not for Michigan as

a whole. Still, even if we were able to overcome these obstacles,

38
there remains one element which maybe fundamentally damning to the
entire enterprise.

The presence (and, indeed, the success) of the private education
alternative leads Barzel to argue that education itself is in
principle a private good. Barzel is quick to point out that, even on
the basis of the rudimentary data available, there is a clear
incentive for a majority of the polity to favor collective supply.
Nonetheless, collective supply does not change the nature of the good.
This notion of schooling as a private good is borne out empirically by
the lack of uniform consumption even at the individual level.
Accepting this, Samuelsonian interpretations of public sector
optimality may not be applicable. As Barzel puts it: ”[tlhe concept of
the ‘efficient quantity’ is itself meaningful only when unifonmity is
imposed."[p.1851 Thus, Barzel concludes his critique of Barlow’s work
by pointing out that questions with respect to the relative efficiency
of median educational preference may themselves be theoretically
meaningless, regardless of the manifold practical difficulties

involved in answering them.
On the formulation of an inquiry

For the purposes of this dissertation, one is tempted to regard the
foregoing discussion (indeed, this entire chapter) as a cautionary
tale. What has been demonstrated, after a promising start, is the
potential for studies of the tax referendum environment to move
rapidly toward an seemingly endless series of dead ends. If this is

typical of the ways in which sciences advance, the prevailing

39

statistics on mental health among academics may be well justified.
Nonetheless, I have no intention of abandoning this enterprise in
favor of the plumbing trade. In any case, there is much to be learned
from these apparent failures. Where, then, have these authors failed?

To paraphrase the German metaphysician Martin Heidegger, every
inquiry begins with the formulation of a question. The formulation of
the question presupposes the nature of the answer in that, by its
formulation, the question structures the methods of the inquiry and
sets out standards for judging the relative success of proposed
answers. The lines of inquiry discussed above proceed from questions
formulated chiefly by Bowen: Is the median preference decisive?; Is
the median preference efficient? These formulations remain unchanged
despite the contributions of the several authors. After following the
tortuous chains of argument presented above it seems that neither
formulation allows for successful answers. Indeed, given Barzel’s
argument, we might conclude that the latter cannot be meaningfully
asked. If the authors have failed, perhaps the single most important
source of their failure is the nature of the questions they seek to
answer. What remains for this inquiry is a reformulation at the most
basic level. What is needed here are questions that can be both asked
and answered. Moreover, to formulate these questions we need not
wander far from these failed efforts. The basic source of such a
reformulation can be found in the very works criticized above.

The various analyses of the tax referendum environment discussed
above have emphasized the outcomes of a policy making environment. In
so doing, they remain true to the committments of the discipline of

economics in so far as it has made a contribution to the development

40
of policy science. Policy science, which I take to be the study of
public policy making and implementation from various aspects, is not
the sole province of any particular academic discipline. Nonetheless,
it has long been dominated by economics. Political scientists, for
their part, have dutifully adopted techniques of analysis borrowed
from economics (e.g., econometrics and welfare economic modeling) as
tools in their own participation in the practice of policy science.
Still, if political approaches to policy science cannot be
distinguished from economic approaches in terms of subject matter or
technique, such a distinction can be made in terms of the relative
emphasis placed upon process and outcome.

The ‘bottom line’ for economic approaches to policy science is the
location and evaluation of predicted outcomes in a policy environment.
The processes involved in public policy making are important in so far
as they provide the means by which policy outcomes are generated, but
primary emphasis is reserved for the outcomes themselves. Given this
emphasis, the processes themselves are seldom treated as problematic.
Rather, they are incorporated in the axiomatic base of a given model
and, thereby, assumed as given a priori. The various basic models
cited above well exemplify the economic approach to policy science.
Institutional and environmental conditions governing voting, the
distribution of costs and benefits, and the motives of various actors
are assumed at the outset. Once these assumptions are stated, little
attention is given to them. Nonetheless, in terms of the critiques
above, this is the precise location of where these models fail.

The relative neglect of these ‘process’ variables reflects the

general disdain in which economists hold political questions. Politics

41
is, after all, a messy business. The stability and parsimony of the
marketplace is wholly replaced by the chaos of political gamesmanship.
While few of the authors cited above are as outspoken in their
denunciation of politics as Inman, most, I suspect, would share his
general regard for politics and would rather not delve into such murky
matters. Still, the environment they model is profoundly political.

It is here that a political approach to modeling can make a unique
contribution to our understanding of this particular policy setting.
As indicated in the introduction to this dissertation, political
approaches are distinguished from economic approaches in that they are
guided by the fundamental Lasswellian question of ‘who’ benefits and
‘how’ they achieve their benefits. This involves a greater emphasis
upon the behaviors of individual actors, their motives, and the
strategic possibilities available to them as given in the institutions
and processes extant in the environment. While an analysis which
reflects concerns for these elements may prove more rewarding than a
narrow economic approach (indeed, this is exactly what this
dissertation is designed to demonstrate), it is, nonetheless, a much
more detailed treatment of any particular policy environment. The tax
referendum setting is no exception to this requirement of greater
detail.

The tax referendum environment includes school boards, school
administrators, organized interest groups (particularly faculty and
parent groups), and, of course, the electorate. While these sets of
actors must collectively select a single alternative level of taxes
and expenditures, they individually do not evaluate this result from

the same or similar utility functions. Moreover, each separate group

42
plays a different role in the selection process with different means
of influencing the final selection in accord with their own
preferences. The interplay of strategies may, indeed, be a messy
affair, but an understanding of the dynamics of this conflict is a
necessary prerequisite for the type of analysis proposed by the
authors cited above. Thus, this chapter concludes with a call for a

‘political’ analysis of the tax referendum environment.

In the next chapter I define the parameters of this analysis and lay
out new formal models to be tested. The emphasis of the analysis is
the dynamics of strategic interaction in the tax election setting.
Still, the notions of equilibria and stability championed in the works
cited above are not absent from.the inquiry. There is certainly a
place for consideration of equilibria in political approaches to
policy science. If nothing else, the potential for stable equilibria
in any public policy setting acts as a constraint upon the behavior of
the parties to the policy making process.

Occasionally, these limitations are overwhelmingly important.
Holcombe [1977:80-821 describes a system of collective tax choice
which constrained the outcome to a median voter equilibrium regardless
of the preferences of various organized interest groups. The system
was used in Florida between 1939 and 1968. Voters indicated their
preferred level of expenditure directly on their ballots. These levels
were rank ordered and (assuming single-peaked preferences) the median
outcome was found by selecting the quantity desired by the median
individual in the distribution. Here is the economists’ dream: a
straightforward political system which yields a stable equilibrium.

Nonetheless, the very fact that Florida has abandoned this system

43

points to the relevance of political variables even here.

CHAPTER 2
THEORETICAL ORIENTATION

While the discussion presented in chapter 1 deals with several
approaches and models, the themes and primary research questions
addressed in most of them reflect the concerns articulated in
Bowen[1943]. In closing the last chapter I have argued that the
failures of these lines of inquiry stem from.the limited nature of
those ‘research questions’ as formulated by Bowen. The task in this
chapter is to reformulate the inquiry so as to include certain
elements of the political environment which were either overlooked or
pointedly ignored by the authors discussed in chapter 1. That is, here
I shall define a new set of parameters within the political context of
tax referenda. These are to be incorporated into the research
enterprise with the hope that they may afford a broader understanding
of this particular environment than was possible under the assumptive
base of the Bowen model.

It should be noted at the outset that this reformulation marks a
major redefinition of the research task itself. This dissertation will
not ‘rework’ Bowen yet again. Rather, what is presented here is best
described as a first tentative attempt to formulate a ‘political’
model of the tax referendum environment as distinct from the
‘economic’ approaches employed by Bowen and those which followed him.
As noted in the last chapter, this distinction is not made in terms of
subject matter or methods, but, rather, in terms of what is taken to
be problematic. Where the ‘process’ variables are subordinate to the

evaluation of potential outcomes in the ‘economic’ models discussed in

44

45
chapter 1, a ‘political’ model champions the primary salience of the
processes themselves. While both approaches include both process and
outcome as elements in their basic design, the relative emphasis
placed on each is profoundly different. Hence, the questions addressed
are themselves conceptually distinct. Nonetheless, given the common
elements between the two approaches, the reformulation proposed here
may rely upon the prior work (particularly Bowen’s model) as a guide.
In defining what parameters are relevant here we might consider what
is assumed away previously. This, at least, seems a reasonable place
to start .

As noted in the last chapter, Bowen’s attention to the ‘real world’
issues of the process by which an electorate selects the median
preferred quantity of tax and/or expenditure is given in his
exposition of an incremental adjustment model of tax choice. Within
this model a series of alternatives are pitted against the status quo
and the Condorcet winner is located by a process of elimination. While
I have argued that many questions concerning the practical
difficulties of this approach remain unasked in the Bowen piece, one
point stands out as immediately relevant to the task here. In
presenting this incremental adjustment model Bowen relies upon some
agency of government to structure the agenda of alternatives available
to the electorate. For the purposes of this chapter, perhaps the
single most important assumption of the Bowen design is given in the
manner in which Bowen addresses the motivations and incentives which
affect this government agency in terms of its behavior as an agenda
setter. While Bowen does not deal with these issues in any direct

sense, this is not to say that such motives are not implied.

46

Clearly, in order for the median preferred allocation to prevail it
must be proposed by the agenda setter. The agenda setting agency
described in Bowen’s incremental adjustment model uses the electoral
mechanism itself to identify and propose the median preferred
alternative. The fact that this agency must actively prosecute this
task implies that it is not simply an inert instrument of the
electorate. Rather, from.this we may infer that such agenda setters
behave as the active agent of median preference. In principle, Bowen
assumes that this set of actors share, or at least behave as if they
share, the preferences of the median voter. While such an assumption
certainly acts to simplify the model, the normative commitment
involved is far from.trivial.

What would be the effect of attributing other particularized motives
to the behavior of Bowen’s agenda setter? A growing body of literature
(e.g., Niskanen[1971], Plott and Levine[l975], Mackay and Weaver[l978]
and others) suggests a broad array of opportunities for strategic I
manipulation of the outcome in various collective choice environments
with structured agendas. Nonetheless, the idea of external agenda
control used for selfinterested motives does not preclude the
possibility of stable equilibria. Indeed, the fact that agenda control
itself is an institutionally relevant condition bespeaks the presence
of other potentially salient institutional factors which may constrain
the agenda setter to a single outcome or limited range of outcomes.
Yet another growing body of literature (e.g., Shepsle[1979],
Riker[19801 and others) suggests the possibility of institutionally
enforced equilibria even in cases where no equilibrium would otherwise

exist. The key to the identification of such results lies in the

47
careful articulation of the relevant institutional conditions and
motives extent in each institutional setting.

This chapter of the dissertation recasts the issues of the tax
referendum setting in terms of these ‘institutionalist’ concerns.
These are the ‘process’ variables from.which I shall reformulate the
inquiry. The first issue to be dealt with involves the motives which
guide agenda control. From this point I shall turn to consider other
relevant factors which may serve to shape and structure the possible

outcomes from a formal vantage.
Problems of Agenda Control

Some twenty-two states use tax referenda for the financing of public
education. This allows a wide variety of legal and organizational
procedures governing the selection of alternatives. Even so, by far
the most common source of taxation proposals is the local school
board. Given this, we may consider several possible explanations of
why and how school boards make these proposals.

Since school board members are usually elected representatives, it
might seems reasonable to assume that their attitudes on taxation
issues should, at least, not be inconsistent with those of the
electorate at large. The representative character of the school board
may provide a decisive bias towards the median preference of the
electorate. This is to say that, since a school board acts as a
committee (usually operating under majority rule) and assuming that
the attitudes of the board stand as a representative sample of the

attitudes of the electorate, the median preference within the school

48
board should be roughly equivalent to the median preference of the
electorate at large. This is, in effect, the argument presented by
Bowen. Holcombe[1975] takes the notion of representation within the
school board even further to argue that electoral sanctions will be
imposed upon members of any school board which attempts to propose
excessive levels of taxation. As elected politicians mindful of the
problems of reelection, school board members will toe the median
line.

This line of reasoning fails to take into account the fact that
candidates for local school boards seldom if ever run for office on
the basis of their attitudes toward taxation. Moreover, the image of
local school boards as populated by Downsian politicians of Holcombe’s
sort, seems well out of step with most local communities. Perhaps in
large urban settings the political professionalism.of school board
members is relatively acute, but for the most part school board
members tend to be little more than ‘involved’ citizens who care
enough about the schools in a community to seek what is often actually
or essentially a volunteer position. Given this, candidate platforms
tend to reflect the motives of the ‘pure of heart’.

Attitudes toward education in general, specific education policies,
and personal experience are common themes in bids for election to the
school board. Taxation policy is rarely a factor. Nonetheless, there
is a case to be made which casts doubts upon the notion that the tax
preferences of the typical school board should simply mirror those of
the electorate at large. In saying that most school board members are
simply ‘involved’ citizens we should consider the nature of their

involvement. While school boards contain individuals of various

49
backgrounds, those individuals who are attracted to the job in the
first place tend to be those who see immediate intrinsic value in the
goods and services provided by the educational system. It is no
accident that parents and professional educators occupy a large share
of the seats on local school boards across the country. These groups
can be classed as ‘high demand’; which is not to say that they are
insatiable in their demand for education, merely that they place a
greater value upon this particular public good than might other groups
within the electorate. On the other hand, ‘low demand’ groups are
typically underrepresented on most school boards. Given this imbalance
of representation, there seems little hope that the median preference
of the typical school board is in synch with the preferences of the
median voter.

In noting that attitudes toward taxation policy is seldom.offered
as a qualification for election to a school board it should be
recognized that matters of school finance are largely regarded as an
administrative task. In these matters the guidance, resources and
support of educators and school administrators play a vital role.
While there is considerable room to debate the actual budgetary
preferences of the typical school board members, such is not the case
with this group of actors. Those whose livelihood is predicated upon
the size of the local school operating budget have a clear incentive
to lobby for maximum allocations. Whether this group is capable of
bending the school board to its will (the ‘pure of heart’ are often
easily conned) or, alternately, the ‘high demand’ characteristics of
typical school board members provide a sympathetic audience to the

budgetary demands of administrators, the outcome is the same. In

50
keeping with this logic Barlow[l970:10301 dismisses the notion that
school boards would hesitate to propose high levels of taxation.

If we accept that school board behavior in proposing levels of
taxation is likely to be more representative of the interests of
budget maximizing administrators than those of the electorate at
large, the theoretical structure of the tax referendum environment is
profoundly changed. The school board is viewed as a monopoly supplier
of a public good with control of the agenda of alternatives available
to the electorate and an incentive structure which, essentially,
favors progressively increasing levels of taxation. The electorate,
for their part, function as a monopoly consumer of the public good (a
monopsonist). The preferences of the electorate favor the median
preferred alternative, but this result cannot be enforced without the
cooperation of the school board. The fact that school board members
are themselves voters is trivial. For the sake of analytic simplicity
we can regard the two groups as separate and their relationship as
adversarial within the context of a bilateral monopoly bargaining
game. The preferences of the electorate constrain the school board’s
ability to maximize the school budget since their proposals are
subject to voter approval. The ability of the electorate to enforce
its preferences is checked by the school board’s agenda control.

The situation described above mirrors the model of congressional
decision making with regard to bureaucratic budgets presented in
Niskanen[1971] in every important respect. This similarity led
Holcombe[l975] to propose a model based on Niskanen’s work as a
possible alternative to a Bowen-style median voter model in analysis

of results of school millage elections in Michigan. Holcombe argued

51

that the legal and institutional characteristics of Michigan are
particularly favorable to a Niskanen-style interpretation. In testing
his model, Holcombe was able to identify a limited set of specific
elections in which he claimed the school board’s agenda control had
yielded millage rates in excess of the median preferred level. Still,
Holcombe’s general assessment of the predictive success of the
Niskanen model is negative. In interpreting the aggregate results he
concluded that the average millage rate in Michigan more closely
resembled that which would be predicted from a median voter model.
This, of course, implies that some districts have millage rates below
the median predicted level to offset those districts which he
identified as being in excess of that level. Nonetheless, Holcombe
champions the median voter model as a better general rule of thumb.

We need not accept Holcombe’s conclusions as the final word on the
comparative success of the Niskanen model as applied to tax referenda.
If for no other reason, his operationalization of several key
parameters is open to critique. Still, those who would accord
significance to agenda control and Niskanen-style arguments must both
account for the failure of Holcombe’s efforts and provide a basis for
a more adequate specification of the model. To do so we must carefully
examine and test the various implications of the Niskanen approach as

applied to this setting.

52

Niskanen, Sophistication, and the Bargaining Range

The Niskanen model of congressional decision making with regard to
bureaucratic budgets is quite simple in design. Niskanen acknowledges
that the model is weighted toward the supply side of the public goods
allocation process. Collective demand for the output of some bureau is
summarized by the median legislative preference profile which is
assumed at the outset. This ‘total evaluation curve’ is given by the
equation:

2
B - aQ - bQ

(where:‘B’ - the bureau’s budget; ‘Q’ . level of bureau output; and,
‘a’ and ‘b’ are greater than zero). The costs of producing any given

level of output are defined by the equation:

2
C - cQ + dQ

(where:‘C’ - total cost; and, ‘c’ and ‘d’ are greater than zero). The
budget maximizing bureaucrat, being privy to this information and
having complete agenda control, faces a simple constrained
optimization problem. Once the bureaucrat locates the maximum budget
that the legislature will bear, all that remains is to propose that
level. Assuming, as the model does, that legislators are passive
participants who vote sincerely according to their preferences, the
bureaucrat is assured of the proposal’s success.

The Niskanen predicted level of output is always in excess of the
socially optimal level. Thus, the model has gained favor with several

circles of public welfare economists and provides major theoretical

53
support for those who would argue for the privatization of the supply
of public goods and services. Nonetheless, it is oddly constructed in
many respects. The empirical conditions of bureaucratic/legislative
interaction correspond to a bilateral monopoly bargaining game.
However, as these interactions are depicted in Niskanen’s model, no
bargaining is involved. The legislature makes no use of its position
as a monopsonist. In effect, the legislature simply gives over its
total evaluation curve to the bureau and allows the bureau to make the
final budgetary choice. The fact that bureaucrats have no actual vote
in the final decision is excluded from the model. Given the assumption
that legislators are passive and vote sincerely, the actual voting is
relegated to a nggufggmg_exercise.

Legislative behavior, as depicted in the model, is not irrational in
a strict sense. After all, the model assumes that legislators always
vote in a manner absolutely consistent with their expressed
preferences. The fact that these preferences still serve as a
constraint upon the success of the budget maximizing bureaucrat
indicates that the legislature has not totally abdicated from its role
as a counterforce. Still, if legislative behavior cannot be said to be
irrational within the model, it is, at least, viewed as relatively
unsophisticated.

In game theoretic terms, sophisticated play involves a willingness
to discard strategies which would follow from a straightforward
adherence to the institutional ‘rules of the game’ in favor of more
complex, but potentially more successful, strategies of a less obvious
sort. Moreover, sophisticated play usually involves a modicum of risk.

The success of a sophisticated strategy often rests upon control of

54
uncertainties over various aspects of the gaming environment. Such
control is afforded by the careful gathering of information and the
formulation of assumptions concerning the expected behavior of other
players. Information may be incomplete and assumptions may fail. The
choice of which risks are acceptable and the ability to develop
complex strategies depend primarily on skill and experience within the
gaming environment.

The attention paid by game theorists to the concept of
sophistication has, for the most part, been limited to the cataloging
of various techniques and strategies available to players in specific
gaming environments. Farquarson11969] applied the concept to simple
committee voting over discrete alternatives. Samuelson119541 made a
case for sophisticated behavior in decentralized tax pricing systems
through the false revelation of preferences. Plott and Levine119751
demonstrate the potential for sophisticated play through the careful
ordering of the agenda of alternatives available to a committee. The
Gibbard-Satterthwaite theorem established the potential for
sophisticated strategies in virtually any collective choice setting.
(Gibbard119731 and [1977]; Satterthwaite[l9751)

In the Niskanen model a high level of sophistication is assumed on
the part of the bureau. Similar sOphistication is denied to
legislators by assumption. It is this imbalance which allows Niskanen
the simplicity of his model despite his admission of the bilateral
monopoly characteristics of the environment he is modeling. The
assumption that legislators lack sophistication is not well justified.
Indeed, the notion of passive naivete as characteristic of legislators

does not square with the common image of the politican as a well oiled

55
Machiavellian machine. Miller and Moell981] justly take Niskanen to
task on this basis. They argue that if the budgetary interactions
between legislatures and bureaus truly yield suboptimal results the
fault lies with the legislature rather than the institutional
structure of those interactions.

In the tax referendum setting it is the electorate, rather than the
legislature, which acts as the monopsonist. In this decision setting
the assumption of relatively low levels of saphistication on the part
of the monopsonist appears much more plausible. The ability of the
electorate to coordinate a sophisticated strategy to counter the
school board’s agenda setting power requires a level of organization
which is seldom seen empirically. On the other hand, school boards
have the resources necessary to gather information concerning the
budgetary preferences of the electorate and are generally expected to
do just that. Thus, the criticisms of the Niskanen model offered by
Miller and Moe carry less weight in this setting. Moreover, the use of
the Niskanen model in this dissertation does not involve efforts to
attribute blame for potentially suboptimal results (real or imagined)
in any case. It is hoped that the Niskanen model will provide a useful
heuristic device to better understand the dynamics of the tax
referendum setting. Given the Miller-Moe critique, the Niskanen model
may be more apprOpriate to this task and to this environment in
general than to the purposes for which it was originally designed.

Saying that the assumption of low levels of sophistication on the
part of the monopsonist is more plausible in the tax referendum
setting does not imply that an absolute pronouncement can be made even

here. Despite all obstacles, there are rare cases in which low demand

56
groups within the electorate have been able to organize effective
campaigns against taxation proposals. As Coleman[19571 reported, the
presence of organized opposition during tax elections tends to
decrease the probability of the passage of a proposal. Such behavior
on the part of any subgroup of the electorate indicates a
sophisticated understanding of the bargaining game, however modest.
Moreover, the fact that school board members seldom campaign on the
basis of tax policy may have implications for any claim that they
possess high levels of sophistication. If they do not ordinarily
attend to matters of tax policy we can hardly expect them to be
experts on these matters. This possible lack of expertise might lead
to more cautious behavior on their part, or, alternately,
mismanagement of the tax proposal campaign. Either case would tend to
bias the outcome toward the preferences of the electorate.

If the concept of sophistication is to be meaningfully applied to
tax referenda it must be dealt with in relative terms. The Niskanen
model attributes all sophistication to the monopoly supplier in the
bargaining game. Even if this proves to be mostly true, the
possibility that some level of saphistication is manifest in the
behavior of the monopsonist could lead to outcomes at variance with
the Niskanen predicted result. Similar, if not greater, variation
should be seen if the behavior of the monopoly supplier demonstrates
less than perfect sophistication. Thus, the Niskanen model should be
regarded as a ‘polar’ type. The results predicted there define one
pole of a range of possible outcomes. If the median voter result is
predicted when the potential influence of the monopoly supplier is

excluded from.the model, this result defines the alternate pole of the

57
bargaining range. A greater level of sophistication on the part of one
group relative to the other should bias the outcome toward the

interests of that group within this limited range.
Managing Risk and Reversion

Keeping in mind that the Niskanen model does not exclude the effects
of the preferences of the monopsonist, these effects must carry over
in its application to the tax decision setting. In most tax referenda
these preferences are best represented by the demand curve of the
median voter. There are two exceptions to this generalization:
Nebraska requires a 551 majority for all levies over 12 mills; and,
Washington requires a 602 majority for levies over 7 mills. While the
institutional differences in these two states change the location of
the crucial swing-voter and thereby shift the bargaining range
downward, the essential structure of the Niskanen model does not
change. The Niskanen predicted result remains the highest levy capable
of being supported by the swing-voter.

The Niskanen model locates its predicted result through reference to
the allocation level given if the monopoly supplier’s proposal fails.
In his original statement of the model Niskanen assumes that this
reversion level is ‘zero’. That is to say that, the bureau makes its
budget proposal to the legislature on an all-or-nothing basis. He
further assumes that the bureau’s proposal will fail if and only if
the swing-voter.gtgigglxupggfiggg.‘nothing’. Thus, the Niskanen
predicted result is that point in the swing-voter’s demand curve at

which the voter is indifferent with respect to the choice between the

58

proposed allocation and no provision at all. Given single-peakedness,
this is a unique point. Even allowing for the cases of Nebraska and
Washington, there are several institutional considerations which must
modify this basic model in application to the tax referendum setting.

With respect to educational finance the relevant reversion level is
never actually ‘zero’. Every state in the union receives some
appropriation of revenue for public instruction from the federal
government. Moreover, most states provide some additional revenue for
the provision of public education from sources other than earmarked
property taxes (usually sales and income taxes). In those states which
use tax referenda to assess property taxes for public education the
percentage of the total school district revenue derived in this
fashion varies from a low of 161 in Delaware to a high of 802 in
Nebraska. While it is clear that a levy proposal failure in a state
such as Michigan (where the average school district derives 582 of its
total revenue from local property taxes) could be devastating -even to
the point of closing schools- the fact that a substantial portion of
the district’s operating budget is provided without voter approval
should significantly alter the location of the Niskanen predicted
result.

This situation is further complicated by the fact that voters may
not be able to fully appreciate the relative value or amount of
non-matching grants from other sources in making a tax voting
decision. Often the impact of non-property tax funding sources is
underestimated. Thus, voters may assume that the failure to pass a tax
prOposal will have more dire consequences than is actually the case

(an assumption with is encouraged by most school boards). This failure

S9
to perceive the effect of outside funding sources makes the task of
estimations based upon the reversion level difficult and uncertain.
The situation was first discussed in Gramlich and Galperll973] and
labeled the ‘flypaper effect’ by Courant et a1.[l979]. Since these
initial studies several authors have attempted to develop means to
estimate the magnitude of the effect (e.g., Oatesll979], Fisherl1982l,
Fossettll983] and, particularly, Boner and Rosenthalll9801), but to
date no reliable method has been demonstrated.

Other institutional characteristics may further complicate the
location of the Niskanen predicted result. Particularly, about half of
the states which use tax referenda allow school districts to maintain
any successful levy perpetually. This means that once a levy proposal
is passed by the voters it never expires and the district is allowed
to assess taxes at that level indefinately. The remaining tax
referendum states require periodic renewals of all previously passed
levies. Some states allow school districts a given level of taxation
without voter approval, requiring referenda only when the school
board’s proposal exceeds this given level. Some states even allow
periodic ‘cost of living’ increases in these base levels. For
instance, Colorado school districts may levy up to 1062 of the
previous year’s base rate without voter approval (see Hamilton and
Cohen[1974]). As above, all these factors should influence the
location of the reversion level and thereby affect the Niskanen
predicted result.

Romer and Rosenthall1978] present a careful theoretical analysis of
the effects of high reversion levels on Niskanen style arguments,

particularly with respect to tax referenda. Assuming that the median

60
voter result is potentially more efficient and socially preferable to
the aspirations of a budget-maximizing agenda setter; these authors
reach what they call a "seemingly perverse and paradoxical result".
They argue that the interests of the polity are better served if the
agenda setter is allowed high reversion levels:
"The reason for this is that, in effect, zero taxes also mean
zero expenditure. The setter can then use his monopoly power
over the agenda to threaten the voters with facing the
consequences of zero expenditure if they fail to approve a high
level of expenditure."[p.29]
Thus, as the reversion level approaches zero, the threat potential of
the agenda setter is enhanced. 0n the other hand, those situations in
which the reversion level is fairly high allow the electorate to
resist the demands of a budget-maximizing agenda setter more
effectively.

Romer and Rosenthal’s conclusions receive initial support from
Meufeldll977]. Neufeld’s study was limited to Michigan school
districts; a state with relatively high use of earmarked property
taxes and requiring periodic renewal of all levies. These
characteristics combine to produce relatively low reversion levels,
which, in turn, should make the electorate more favorably disposed
toward the proposals of the school board. As expected under the Romer
and Rosenthal model, Meufeld found a high correlation between
decreases in the reversion level and increases in the percentage of
‘yes’ voters. Similarly, Holcombell975] found that millages which
passed in the 1972-73 school year in Michigan tended to pass by wider
margins than similar cases in other states (the average percentage of
‘yes’ voters in Michigan: 65.3 [p.1081). While results such as these

tend to confirm the basic conclusion of Romer and Rosenthal, they do

not speak well for the sophistication of Michigan school boards. If

61
the school board’s proposal passes by a wide margin, say greater than
601, it is likely that an even higher proposal could have passed by a
more narrow margin. What, then, is holding these school boards back?

A key component of the concept of sophistication, as presented
above, is the management of risk. While it may be true that lower
reversion levels provide a school board with a more potent threat, we
should not forget that this threat also threatens the school board
itself. School board members might not be elected on the basis of
their budgetary attitudes, but if these attitudes lead to school
closings one could hardly expect such results to enhance reelection
chances. In the extreme case, with a zero reversion level, the school
board’s threat potential may indeed be at its greatest. However, if
the school board were to propose the Niskanen predicted level under
these conditions it would, in effect, be putting a gun to its own
collective head, hoping against hope that the electorate will not pull
the trigger. While the failure of a levy proposal under these
conditions (i.e., the success of a zero reversion level)is a
suboptimal result for all concerned, the real brunt of the failure is
borne by the school system and, thus, the school board.

In keeping with the notion of a bargaining range, the reversion
level establishes the location of the Niskanen predicted result and,
thus, the upper extreme of the range. Low reversion levels yield
higher Niskanen results, and thereby provide for larger bargaining
ranges. Since the lower extreme of the range is fixed, any increase in
the reversion level should contract the bargaining range. Still, the
risks to the agenda setter are not constant over different bargaining

ranges. As bargaining ranges increase in size, so do these risks.

62
Increased risks can lead to greater care in the proposing behavior of
the agenda setter and, perhaps, bias that behavior away from the upper

extreme of the bargaining range.

Serial Bargaining and Multiple Elections

The above discussion has continually referred to the tax referendum
setting as a ‘bargaining’ game with a ‘bargaining’ range..Bargaining,
of course, involves communication and it is not clear from.the
foregoing how the various parties communicate with each other.
Certainly, school boards might communicate with the electorate through
the news media and public hearings of various types. The electorate
might communicate their preferences to the school board through public
comment, ‘letters to the editor’, and responses to polling
questionnaires. Still, the most definitive act of information exchange
between these two groups is given in the actual proposition and voting
over levy proposals. Herein lies a problem. There seems little room
for ‘bargaining’, per se, if the object of the bargaining is contained
in the act of communication itself.

Throughout this discussion the single election has been taken as the
unit of analysis. Of course, the single election format overlooks the
fact that each election is merely a single point in an ongoing
interaction between the school board and the electorate. The success
of the Niskanen model is predicated upon the assumption that the
agenda setter’s proposal is regarded as an all-or-nothing, take-it-
or-leave-it offer. For the agenda setter’s threat to be credible, the

monopsonist must regard the proposal as final. If, however, failure to

63

approve the proposal is immediately greeted by a new lower proposal,
the credibility of the threat is lost. Even so,the basic power
relationships remain unchanged. Both sides maintain their particular
power and neither can unilaterally enforce an outcome. What has
changed is that suddenly there is room for bargaining over a series of
proposals and votes. What is the expected effect of this bargaining?

The effect of multiple elections on the outcome in tax decision
settings has been an important analytic theme since Bowenll943]. Of
course, Bowen regarded the use of a multiple election format as
intended to yield the median voter equilibrium. As a result, the idea
that multiple elections normally shift the final outcome toward the
median voter result is common in the public finance literature. The
studies of Neufeld and Holcombe, while giving due consideration to the
preferences of the school board and other institutional factors, end
up uncritically accepting the assumption that any multiple election
format will ultimately produce the median preferred alternative. Only
Ladha et al.[1982] seriously attempt to test this notion empirically.

The Ladha study uses a sample of 111 Oregon school districts which
held budget elections in 1971. Of these, 69 budget proposals passed on
the first try, 25 passed on the second try, 13 on the third, and 4
districts used a total of four elections to finally pass a budget
proposal. The authors expected to see a drawn out bargaining process.
Assuming that a school board would offer a potentially high proposal
in order to ‘test the waters’, the board would then allow itself to be
‘bargained’ down (all this while carefully monitoring the proportions
of ‘yes’ and ‘no’ voters in an effort to estimate the extent to which

they would ultimately have to make such concessions). This ‘learning’

64
and adjustment effect was not reproduced in a consistent fashion in
the data. Of the 42 districts which held a second election, 16 did not
change the initial proposal and 3 actually raised it. Of those who
lowered the proposal for a second try, 13 lowered it by only a tiny
increment (less than 21). As a result, the statistics on the
‘learning’ effect were not significant.

Despite these results, other evidence tended to support the notion
that the bargaining process was simply more complex than was
originally assumed. Several cases in the sample showed the absolute
number of ‘no’ votes falling even when proposals were unchanged and
turnout figures increased between elections. This was taken to mean
that "... some voters who initially vote ‘no’ may change their votes
as the closing of the schools becomes a more imminent
possibility.”[p.26]

A school board which allows itself to be ‘bargained down’ may be
showing little or no sophistication. By serially adjusting a proposal
downward, the school board may actually encourage ‘no’ voters to hold
out for a better offer. On the other hand, the use of a multiple
election strategy can aid in the gathering of information and,
thereby, the control of risk. By gaining a better sense of the
preferences of the electorate, the school board’s proposal at the next
iteration is better informed and the probability of success is
enhanced. Where the Ladha group failed was in assuming that the
appropriate response to this information was always a lowering of the
proposal. In fact, under certain circumstances this may be precisely
the wrong response. What must be remembered is that the information

gained by the experience of an election becomes common knowledge. Low

65
demand groups within the electorate may be encouraged by these
results. In such cases the reassertion of agenda control and the
reliance upon other resources to shift the balance may be the only
possible check on the growing confidence of ‘no’ voter groups.

In many tax referendum states the possibility of serial bargaining
over a number of elections is constrained by statutory limits on the
number of such elections within a given fiscal year. Arkansas and
Florida limit their school districts to one election per year.
Delaware and New Jersey allow two such elections, and Ohio has a limit
of three. Michigan allows a school board to call an election once
every six months, but this limit may be exceeded if additional
elections are qualified for the ballot through petition drives
(Michigan School Code [l955:sec.340.511]). From the point of view of
the school board, the Michigan case may present the best combination
of possibilities for serial and single election strategies. Here, the
‘take-it-or-leave-it’ threat of the Niskanen model is not wholly
abandoned. Nonetheless, there is an escape clause in case of
emergency.

In sum, the possibility of serial bargaining over several elections
complicates any model of the tax referendum setting. In comparison to
the original Niskanen model, the agenda setter’s threat potential is
somewhat curtailed. However, so are the risks associated with an
aggressive use of agenda control. We might expect more elaborate
bargaining strategies and a share of experimentation under a multiple
election format. This should contribute to greater variance within the
bargaining range. Still, the notion of a bargaining range as

established by the poles of the Niskanen and Bowen results remains

66
intact. There seems no reason to conclude that the presence of

multiple elections is sufficient to insure the median voter result.
Managing Turnout

In discussing the possible responses to an initial failure of a
levy proposal above it was noted that lowering the proposal might not
be appropriate and that other resources might be used to change the
outcome of a subsequent vote. Perhaps the most highly valued resource
available to the school board is given in the actual management of who
votes and who does not. Even a casual perusal of the literature
available to professional educators and school administrators leaves
the impression that voter turnout is a key factor in the successful
management of a levy campaign. Tax elections are seldom held
concurrently with other elections. Their isolation tends to lead to
lower turnout for these elections than for other types of referenda.
Piele and Hallll973] note that normal turnout for school tax referenda
is about 251 of the electorate -about half what is expected in
national elections.[p.55] Far from being disappointed, educational
practitioners applaud these low figures. Several articles in trade
journals champion the notion that lower turnout means higher
proportions of ‘yes’ voters (e.g., Crosbyll963], Harringtonll959],
and Hansonl19691).

The veracity of this piece of conventional wisdom is well supported
by a barrage of empirical studies on the matter. Piele and Hall cite
no fewer than fifteen separate studies which show a statistically

significant relationship between low turnout and favorable voting in

67
school finance elections.[p.64] What is so striking about these
results is the fact that this conclusion has been reached in a wide
variety of empirical settings with several different models and
measurement techniques. While it is true that some studies have found
no significant relationship between turnout and favorable voting
(e.g., Stone[l965] and Hahnll968l), these are distinctly in the
minority. The relationship appears to be quite robust.

The importance of this relationship not withstanding, consideration
of turnout is conspicuous by virtue of its absence in the public
finance literature. One likely explanation of this neglect harkens
back to the original notions of Wicksell and Lindahl. The linkage
between taxation and public goods allocation depends on an absolute
identity between voters and public goods consumers. The necessity of
this identity led these authors to advocate a unanimous decision rule.
The impracticality of this rule led Bowen to relax it to a simple
majority, but the identity in question is partially preserved in the
Bowen model by the assumption of 1002 turnout. Barr and Davis[1966].
Barlowll970], Borcherding and Deaconll972] and Bergstrom.and Goodman
[1973], all share the ‘everybody votes’ assumption. Inmanll978] is
virtually unique in the attempt to include turnout effects in his
model.

It is not the case that these scholars are unaware of the
characteristically low turnout of local tax referenda. Rather, they
seem to regard it as a trivial detail to be explained away. Holcombe
acknowledges low turnout, but argues: ”If the voters who cast ballots
are an unbiased sample of the papulation of registered voters, then

the ballots of the sample can be considered representative of the

68
population."[1975:53] Thus, Holcombe justifies his use of the
‘everybody votes’ assumption as a useful fiction. It is particularly
important to his model that he do so. Holcombe estimates median demand
on the basis of the closeness of the election. All other things being
equal, if the election is close,the results should reflect the
preferences of the median voter. If this median voter is also the
median taxpayer, then all properties of the Bowen equilibrium
maintain. Unfortunately, the evidence presented by Piele and Hall
suggests that this final identity may not hold. In the face of this
possibility, Holcombe’s estimates are questionable at best.

Assuming for the moment that the conventional wisdom which posits a
relationship between low turnout and favorable voting is valid, the
question remains as to why this should be so. In the past half century
political scientists have proposed several explanations of why some
pe0p1e vote and others do not. A variety of attitudinal and socio-
economic factors such as education, age, income, sense of duty,
efficacy, and costs of voting are included in these various models.
Still, the narrow focus and unidimensionality of school tax referenda
may provide for a more simplified understanding of turnout in this
environment.

Piele and Hall suggest such a simplified model based on the concept
of a ‘normal vote’ (see Converse et a1.[1961] and Campbell et al.
[1966]). Included in the group identified as "normal participants" in
school finance elections are owners of extensive property, parents of
school aged children, civic elites, and high interest groups such as
educators. A crucial factor which separates "normal participants" from

infrequent or nonparticipants is the intensity of their preferences

69
with respect to the outcome of the referendum. Piele and Hall’s notion
of intensity boils down to a simple, individual level cost/benefit
calculation. In effect, as the absolute value of the difference
between the costs and benefits of providing public education at some
proposed level approach infinity, the probability of voting approaches
one. Thus, "normal participants" tend to be those who hear
significantly high costs or receive relatively high benefits from the
provision of public education.

The relative size of ‘normal’ participant groups as opposed to
‘normal’ nonparticipant groups is an important element of the Piele
and Hall argument. Given that the ‘normal’ participant groups make up
a minority of the general electorate (say, less than 302), there
remains a large pool of nonparticipants whose relative disinterest is,
in theory, attributable to little preceived difference between the
costs and benefits of the provision of public education at a given
level. The asymmetry of size between the participant and
nonparticipant groups can be explained in terms of a similar asymmetry
in the distribution of costs and benefits throughout the polity.

While a weak case can be made that most members of the polity
receive some benefits from the provision of public education, it is
clear that the direct benefits of such provision are fairly
concentrated amoung such groups as parents of school aged children and
professional educators. For the most part, the direct benefits of
public education are virtually nonexistent for individuals outside of
these groups. Still, these low benefits are matched by relatively low
costs. Assuming that the costs of providing public education are

fairly evenly distributed throughout the polity, relatively

70
substantial increases in the basic operating budget of a school system
tend to translate into minimal changes in individual assessments once
they are absorbed by the mass of taxpayers as a whole.

The asymmetry in the distribution of costs and benefits goes a long
way toward explaining the fundamental logic of the relationship
between low turnout and the success of taxation proposals. Given that,
in general, benefits are fairly concentrated and costs are fairly well
distributed, we should expect the ‘normal’ voter group to be dominated
by those whose intensity is based upon the perception of relatively
high benefits. These, then, are mostly ‘yes’ voters. On the other
hand, if the ‘normal’ nonparticipants were somehow forced to
participate, it is likely that most would be poorly disposed to
increases in tax levies. Their most immediate experience of the costs
and benefits of public education is weighted toward the cost side.
While their intensity, as measured in absolute values, is relatively
small, the basic difference is predominantly negative in most cases.
Therefore, the probability of passing a levy proposal is highest when
turnout is limited to the ‘normal vote’ group, and tends to decline as
regularly low participation groups are brought into the voting
population.

If intensity of preference proves to be the crucial factor‘
regulating turnout, and intense groups tend to be more favorably
disposed to tax proposals than the electorate as a whole, perhaps the
best strategy available to the school board is not to campaign at all.
After all, intense groups should more closely monitor the budgetary
actions of the school board. Hence, they are likely to know of

impending elections and will turnout in greater numbers than the

71
inattentive public. Any campaigning would just encourage more ‘no’
voters. This is precisely the strategy advocated in Crosbyll963].
Still, there are several reasons why a school board might elect to
campaign anyway. First of all, failing to campaign may be illegal.
Garberll962] documents several cases where courts have thrown out the
results of underpublicized referenda. Moreover, this approach is a
misreading of the basic logic provided by the Piele and Hall design.
What matters is not simply low turnout, but the actual make up of the
‘normal vote’ group. It is not low turnout, but selective turnout that
makes the difference.

Rubinfeldll977] illustrates the importance of selective turnout with
an empirical example. The study focused on the voters of a single
Michigan school district who faced two identical millage proposals in
the same year. In the first election the measure was defeated by a
modest margin. In the second election the turnout increased, yet the
measure passed. Hubinfeld used survey data to identify vote groups in
both elections and found that the crucial difference was made in terms
of increased participation by women with school aged children. Herein
lies the strategic potential given to a school board in the careful
management of turnout. This is a factor which transcends all of the
institutional conditions so far discussed in this chapter. By
successfully managing turnout, a school board can theoretically ‘stack
the deck’ in its own favor and shift the location of the median voter
up the scale.

All of this, of course, rests on the assumption that the Piele and
Hall design accurately describes the forces at work. While the data

presented by these authors ably demonstrates the basic asymmetry in

72
the distribution of costs and benefits, the link between this and the
idea of a ‘normal vote’ group is decidedly post hoc. As such, this
data does not provide a test of the basic theoretical structure of
their design. What remains to be demonstrated is the notion that
their concept of intensity is the salient factor in determining the
probability of voting in single issue, unidimensional choice

settings.
On the testing of theory

In keeping with the stated intentions of this chapter, several key
institutional factors which may influence the applicability of some
potentially relevant theoretical models have been located and
discussed. In general, two types of models appear relevant to the tax
referendum environment: median voter models (which emphasize the
demand side of school board/voter interaction); and, Niskanen-style
models (which emphasize the supply side of that interaction). These
two types of models should not be regarded as contradictory. Indeed,
there seemm adequate reason to believe that the forces described in
both may be in play simultaneously in any given empirical case.
Nonetheless, both basic models are deterministic in nature and lead to
dramatically different predictions. I have argued that the two models
should be regarded as polar types and that their expected results
establish the extremes of a ‘bargaining range’. Factors such as the
institutional character of agenda control, relative sophistication,
serial bargaining, and the management of risk, reversion and turnout,

have been offered to explain variance in (and, indeed, the location

73
of) that bargaining range. Theoretical arguments and empirical
evidence have been mashalled to justify the relevance and expected
effects of these factors. Nonetheless, these justifications have been
presented on a piecemeal basis.

Any adequately specified model of the dynamics of the tax referendum
process must pay due heed to the interaction between these factors and
the impacts of that interaction upon expectations arising from any
subset of them. The task of the next chapter is to outline a research
design which will, once implemented, provide the information necessary

to specify such a model.

CHAPTER 3

RESEARCH DESIGN, OPERATIONALIZATIONS AND METHODS

A set of hypotheses

The second chapter of this dissertation took the basic design of the
median voter model proposed by Bowen and offered a series of
modifications which reflect institutional factors relevant to the tax
referendum.eontext. Accepting that the original Bowen model depends
upon governmental agenda setting to produce its median preferred
result, I have argued that these agenda setters are likely to prefer
an outcome which sets tax rates at a higher level and that agenda
control provides them with leverage to aid in the achievement of such
a result. The decision context takes the form of a bilateral monopoly
bargaining game with the preferences of the governmental agenda setter
pitted against the median preferences of the electorate. This
situation is analytically similar to what is assumed by Niskanen in
his model of legislative bargaining over bureaucratic budgets.Thus,
the Niskanen model was proposed as an alternative to the Bowen model.

Both the Bowen and Niskanen models assume a unidimensional choice
space, a majority decision rule, and single-peaked preferences. Both
are deterministic in that they predict a unique outcome. Nonetheless,
these predictions differ. Their differences can be attributed to the
relative emphasis each puts upon the effects of agenda control. I have
argued that we need not regard these models as contradictory, nor

should we view their deterministic characteristics as absolute.

74

75
Rather, the two models should be regarded as polar types. The Niskanen
result should be regarded as the expected outcome when the control of
the agenda setter is overwelmingly dominant. The absolute absence of
the effects of agenda control should yield the Bowen result. Thus, the
differing predictions establish a ‘bargaining range’ which exists by
virtue of the effects of agenda control. Here stands a testable
hypothesis.
HYPOTHESIS 1: Agenda control establishes the possibility of an array
of results with extremes at the Bowen and Niskanen predicted
alternatives. In the absence of agenda control, this range should
collapse to the Bowen predicted result.

Assuming for the moment that existence of the bargaining range is
not problematic, additional hypotheses should address the factors
which would influence variance within that range. I have argued that
the primary factor influencing outcomes within the range is the
management of risk, particularly on the part of the agenda setter.
Assuming that the agenda setter’s preferences correspond to Niskanen’s
notion of budget maximization, the Niskanen extreme of the bargaining
range is calculated by locating the point at which the median voter is
indifferent between some proposed expenditure alternative and a
‘reversion level’ which is given at the outset of bargaining. As the
reversion level decreases the Niskanen predicted outcome increases.
Thus, lower reversion levels provide the agenda setter with greater
leverage in gaining higher expenditure levels (as demonstrated
empirically by Homer and Rosenthalll978l). On the other hand, as the
difference between the Niskanen predicted outcome and the reversion
level increases, so do the risks involved with proposing such a large

expenditure level. If the agenda setter’s proposal fails, the

76
resulting reversion level exacts a penalty which is primarily borne by
the agenda setter.

I have argued for two factors relevant to the abilities of various
‘players’ within the bargaining game to effectively manage risk. One
is a factor which I have broadly labeled ‘sophistication’. In this
setting, sophisticated bargaining on the part of the agenda setter
would be manifest in the use of information concerning the preferences
and likely behavior of the electorate to structure the strategy of
play (both in terms of making the proposal and arguing for it).
Sophisticated bargaining on the part of members of the electorate
would be manifest in efforts to convince the agenda setter that the
‘likely behavior’ in question will lead to rejection of ‘excessive’
proposals. The use of a threat to enforce the reversion level should
lead to a lowering of the agenda setter’s proposal. Given that there
is a potentialeor sophisticated behavior on both sides, an hypothesis
concerning the effects of sophistication should allow for biases in
the outcome which favor the more sophisticated group.

HYPOTHESIS 2: All other things being equal, greater relative
sophistication in bargaining, on the part of either electorate groups
or the agenda setter, should bias the outcomes within the bargaining
range toward that extreme of the range which favors the interests of
that group.

A second factor which is relevant to the notion of risk management
is the potential for serial bargaining over a series of elections. I
have argued that this has the effect of lowering the aggregate risk
for all parties. The stakes of any individual decision are less given
that a failed proposal does not necessarily enforce the reversion

level for good and all. Nonetheless, with risk decreased the relative

threat potential of both sides of the bargaining game is similarly

77
decreased. My expectation is that a multiple election format will
lead to greater experimentation and more elaborate strategies. This
will contribute to greater variance in outcomes.
HYPOTHESIS 3: The possibility of serial adjustment in bargaining over
a series of elections should produce greater variance within the
bargaining range than in institutional settings where multiple
elections are precluded.

Finally, there is the Piele and Hall hypothesis concerning a ‘normal
vote’ interpretation of local tax referenda. If turnout in single
issue tax referenda is primarily a function of intensity of
preference the strategic potential for the agenda setter in managing
turnout may be the single most important element of a sophisticated
strategy.

HYPOTHESIS 4: In the tax referendum setting, turnout is primarily a
function of intensity of preference. The probability that an
individual voter will vote increases as the net personal valuation of
the goods and services provided increases.

These are the basic hypotheses of this dissertation. Now I am left

with the task of demonstrating how they can be tested.

Operationslizstions and Experimentation

In order to test these hypotheses empirically several elements need
to be operationalized in definite terms. Since both the Niskanen and
Bowen models rely upon the structure of median preference to determine
their results, the preference profile of the median voter must be
known at the outset before we can determine the size and limits of the

‘bargaining range’ or interpret results with respect to this range. To

78
a great extent, the empirical work on tax referenda discussed in
Chapter 1 can be viewed as little more than efforts to operationalize
the concept of median preference.

The technique used by these authors followed a standard form. The
concept was operationalized through the a priori specification of
various measurement instruments and the relationships between them.
Once specified, the authors inserted the data and cranked away. The
results of this exercise were subjected to statistical tests from
which the authors hoped to judge the value and appropriateness of the
hypotheses which guided the research in the first place. However,
judgements on these matters are clouded by issues of the adequacy of
the initial specifications. Do negative results indicate a failure of
the theory to describe empirical reality, or, do they reflect the
inadequacy of the operationalization of the theory? If the results
tend to confirm the hypotheses, would an alternative specification
have done even better? The ‘trial-and-error’ nature of this type of
a priori specification provides little basis for the answering of
such questions.

The essential element of operationalization is the categorization of
a concept. Categorization involves defining the limits of a
measurement instrument. We select what portions of empirical reality
count as part of the concept and set aside all other aspects of that
reality. The discarded portions of reality are assigned to the error
term of any theoretical model and are governed by a ceteris paribus
clause. In effect, we assume that ‘all other things are equal’ and
have no relevance to the problem.under study. In fact, we know that

this is almost never the case. We hope that ‘error’ is normally

79
distributed and that things will work themselves out without our
ministries. Still, we can seldom be sure of this assumption. Moreover,
in light of my critique of the empirical work discussed in Chapter 1,
I argue that we can be fairly sure that ‘all other things’ are not
equal in most cases.

A central theme of this dissertation is that previous analyses of
the tax referendum environment have been misdirected with respect to
the ultimate goals of the research and that this misdirection has led
to inadequate specification of the various models of this environment.
In truth, I have serious doubts as to whether we can ever empirically
identify median preference structures in the data used in these
studies -at least, not in the absolute terms required by these
authors. If we are ever to make meaningful use of the concept of
median preference in the tax referendum setting, we must develop a
better grasp of the limits of this concept and the nature of its
relationship to other factors within the environment. To establish
this, we need a research setting in which allows us to have faith in
the ceteris paribus clause and in our operationalizations of the key
elements in our theoretical models (including notions of median
preference).

Controlled experimentation is one research setting which provides
reasonable assurance that the ceteris paribus clause is tenable. The
use of control groups for comparison and the careful management of the
experimental environment provide fairly reliable guarantees that ‘all
other things’ are equal. Moreover, the control over the environment
afforded by the experimental setting provides the researcher with

certain knowledge of crucial factors within the design. Where other

80
researchers must make heroic assumptions concerning the personal
utility functions of various actors, the experimentalist knows these
functions precisely by virtue of his or her having designed them in
the first place. Because of these attributes, I have chosen to use
controlled experimental designs to test the hypotheses listed above.
Only after the careful analysis of the results of this experimentation
will I attempt to apply the theoretical concepts developed here to a

‘real world’ setting.
A Set of Experimental Designs

The basic design for the testing of my first three hypotheses is
partially derived from earlier experimental work on committee decision
making, specifically Isaac and Plottll977] and Eavey and Millerll982].
Each experiment involved four subjects: three ‘committee members’ who,
as a collective monopsonist, were empowered to vote over various
alternatives; and, one ‘convener’ who had no formal vote in any
experimental setting. The set of alternatives available for committee
choice were given by an abstract continuum of numbers (0 through 110).
Preferences over this continuum.were generated by means of individual
assigned dollar payoff functions. These individual payoff functions
were provided to each subject in the form of an easily read, two
dimensional graph (see APPENDIX). All payoff functions were single-
peaked and varied between individuals. Player number 1, a committee
member, was given a payoff graph which designated him or her as a ‘low
demand’ voter. This function peaked at the alternative (20) for a

payoff of $11.00. The payoff function of Player number 2, the ‘median

81
voter’, peaked at alternative (50) for a payoff of $13.00. Player
number 3 had a peak at alternative (100) for a payoff of $16.00.

This technique of using assigned dollar payoffs to generate
preferences has been used in several experimental studies and has been
shown to be successful in that subjects do reveal the induced
preferences in their nominating and voting behavior (see Fiorina and
Plottll9781). While, obviously, not every member of the committee can
receive his or her ‘ideal’ payoff, all players were guaranteed at
least $2.00 by their own payoff function. The fourth subject, the
‘convener’, had a payoff function which was increasing throughout the
continuum, from $2.00 for alternative (0) to $20.00 for alternative
(110). This was done to Operationalize Niskanen’s notion of budget
maximization.

The assymetry of information which undergirds the Niskanen model was
operationalized through the distribution of payoff information. The
‘convener’ was provided with complete information on the individual
payoff functions of all subjects in the experiment. Committee members
received only their own payoff information. Moreover, committee
members were not allowed to discuss the actual monetary value of any
alternative with other subjects. Thus, their discussion was limited to
relative pronouncements of ‘better’ or ‘worse’ when collectively
evaluating the alternatives. This was done to further frustrate any
efforts to organize a collective strategy in opposition to the
‘convener’, and to enforce reliance upon the convener’s special
information.

The design included several standard features of this style of

collective choice experimentation. Subjects were prohibited from

82
making any deals to pool individual payoffs and divide the profits
equally. Similarly, side-payments between subjects were explicitly
prohibited. All discussions between subjects were monitored by a
researcher to ensure that these rules were observed. The general
subject pool was recruited from the undergraduate population at
Michigan State university. Potential subjects were approached in lower
division general education courses (primarily social and natural
science courses). Care was taken to try to recruit subjects who did
not know each other in other contexts, and all subjects were sworn to
not reveal the nature of the experiment or their experience to others
outside the experimental setting.

In all experiments, the selection of an alternative from the
continuum was done under a majority vote of the committee members
(i.e., at least two votes of the three). Failure to select an
alternative resulted in the default selection of alternative (0), the
reversion level. Taken together with the median voter’s preference
profile, the Niskanen predicted result was located at alternative
(90). Thus, the hypothesized ‘bargaining range’ is the interval
between alternative (50) and alternative (90), inclusive. Tests of
hypotheses with respect to this range, particularly the first and
third hypotheses, were afforded by the use of three separate designs
which varied institutional rules governing agenda control and the
procedures for making proposals.

under the first set of institutional rules, the convener was given
exclusive control of the agenda and the number of formal proposals
which were ultimately available for committee approval was limited to

one. This is theoretically the most extreme form of agenda control and

83
was meant to operationalize Niskanen’s notion of a ‘take-it-or-leave-
it’ offer. Any failure of a majority of the committee to approve the
convener’s formal proposal meant enforcement of the reversion
alternative (0). A second set of institutional rules preserved
convener agenda control (i.e., the committee could only vote on
alternatives other than ‘0’ which were formally proposed by the
convener), but failure to approve any convener proposal did not
automatically enforce the reversion alternative. The convener was
allowed to make as many or as few formal proposals as he or she
wished. The viability of the reversion level was maintained in this
set of experiments through a set of adjournment rules. Either the
convener alone or a majority of the committee could adjourn the
experiment without the passage of an alternative. This would result in
the enforcement of the reversion alternative (0). In the final set of
institutional rules, the convener could neither propose nor vote on
any alternative. The agenda was open to any and only members of the
committee. Here the convener’s bargaining power was limited to
persuasion based on the information asymetry.

Comparisons of the results between experiments using the first and
second sets of institutional rules serve to test the third hypothesis
defined above, while comparisons of both these results with those
under the third set serve to test the first hypothesis. The hypothesis
concerning the relative effects of sophistication, however, requires a
further modification of the basic design. The concept of
sophistication as defined above is context specific. What stands for
sophisticated behavior in one environment does not carry over directly

to another. Thus, sophistication here must be operationalized to

84
reflect the conditions of the experimental context rather than the tax
referendum environment. Such an operationalization was afforded by the
recruitment of a special subject pool. These subjects were graduate
students with course work in micro-economics, game theory, or positive
political theory in their background. Given this background, these
subjects could be assumed to possess the requisite skill and
experience to develop and implement a saphisticated strategy in the
experimental setting.

To allow for sophistication effects on both sides of the ‘bargaining
range’, subjects from the special pool were introduced into the
experimental setting to play one of two key positions. In one set of
experiments, the ‘sophisticate’ was assigned to the role of the
‘convener’ (Player number 4). In a second set, the ‘s0phisticate’
played the role of the median voter (Player number 2). All other
subjects in these experiments were taken from the general subject pool
and pains were taken to make the inclusion of the ‘sophisticate’ as
unobtrusive as possible. While the subjects from the special pool were
not ramdomly selected, neither were they true confederates in the
style of social pychological experimentation. These subjects received
no special instructions. They were not told the purpose of the
experiments, and their sole incentive was the same monetary
compensation provided to subjects from the general pool in their
particular role.

To test the interaction between sophistication and the various
institutional arrangements, experiments were run using the special
subjects for each set of institutional rules and with ‘sophisticates’

in either the median voter or ‘convener’ role. Thus, the ultimate

85
experimental design includes nine different sets of experimental
results: one set of three institutional designs with a ‘sophisticate’
in the median voter role; one set with no ‘sophisticates’; and, one
set with the ‘sophisticate’ taking the role of the ‘convener’. In each
of these particular experimental settings four actual runs of the
experiment were conducted. Thisprovided a total sample of thirty six
separate experimental results.

The results of these experiments will be grouped into a three by
three matrix of analysis. The dimensions of this matrix are defined by
the hypotheses under test. The horizontal dimension, which defines the
columns of the matrix, is partitioned by the sophistication conditions
in play. The vertical dimension, which defines the rows of the matrix,
is partitioned by the operant institutional rules. This grouping will
facilitate the testing of the various hypotheses through the use of
standard analysis of variance techniques (chiefly, difference of means
tests using the F-distribution). These results are reported and

discussed in the next chapter of this dissertation.
Testing the Piele and Hall Hypothesis

The final hypothesis suggested above relates to turnout. Turnout
characteristics cannot be meaningfully addressed in a smallgroup
experimental setting. The basic hypothesis to be tested grows out of
the Piele and Hall arguments for a ‘normal vote’ interpretation of
turnout in local tax referenda. According to these arguments, those
voters who trouble to turnout for such elections do so because they

have intense preferences with respect to that single issue addressed

86
in the election. Those voters who normally do not vote in these
elections are not regarded as indifferent with respect to the outcome.
Indeed, these nonvoters might have very distinct preferences. However,
their preferences are relatively less intense than those of the voting
groups. It is argued that most nonvoters tend to be poorly disposed to
the passage of tax referenda, while ‘normal’ voters tend to be more
favorably disposed to such passage. Thus, the ‘normal’ low turnout in
these elections should bias the outcome toward the interests of high
demand groups in the electorate.

To model these conditions in an experimental setting requires a
relatively large sample design so as to allow for greater variance
with respect to intensity of preference. The design used here involved
a sample of 207 subjects. The subjects were offered a chance to vote
for one of two abstract alternatives, ‘X’ and ‘Y’. Preferences over
the alternatives were generated by an assigned dollar payoff which
varied between different groups in the sample. Specifically, subjects
were randomly assigned to one of ten groups, each containing roughly
20 subjects. Groups 1 through 6 -or roughly 602 of the sample-
received no payoff at all if ‘Y’ were voted in. Their payoffs for a
victorious ‘X’ varied from $0.25 to $1.50 in increments of $0.25
between groups. Groups 7 through 10 -or roughly 402 of the sample-
were paid only if ‘Y’ was victorious. Their payoffs started at $2.00
for Group 7 and increased by an increment of $1.50 to $6.50 for
subjects in Group 10. Of course, the decision rule was a majority of
those voting. Since the issue at stake was to be regarded as a public
good, victorious subjects were paid regardless of whether or not they

voted. The issue of interest in the experiment was not which of the

87
alternatives won, but, rather, who ultimately voted.

Subjects for the experiment were recruited from a large enrollment
introductory political science course at Michigan State university.
The course in question was geared to the saphomore level, but was
taken by students at virtually any stage of their undergraduate
program. Since the course credits were applicable to any of a wide
variety of undergraduate majors, the subject pool gave a fairly good
cross section of the general student body. Several socio-ecomonic
factors which are arguably relevant to turnout research should be
fairly constant within this sample. One of these is education. The
variation in this factor is institutionally constrained by the
requirements of the university. Since Wolfinger and Rosenstonel1980]
found that education was the single most important variable in the
explanation of turnout in general election data, control of this
factor is particularly important to the tests conducted here. Other
socio-economic variables, such as personal income and social class
membership, may not be so well behaved in the sample. Nonetheless, the
assumption of a normal distribution appears quite plausible.

Since the early days of voting research (i.e., since Ihgpgmggigan_
39Lg2119601) several attitudinal variables have been proposed and
measured in models of voter behavior. Of those which relate to models
of turnout, notions of efficacy, partisanship, and duty have figured
prominently. Thus, this study of turnout attempted to tap these
variables within the sample. To do so, each member of the sample
completed a questionnaire. Included in this questionnaire were the
standard questions designed by the Survey Research Center of the

Inter-University Consortium for Political Research to measure these

88
variables (see APPENDIX for questionnaire). The data from these
questions were assembled into the standard indices. In addition to
this data, the questionnaire contained questions with respect to the
distance the subject lived from campus, the subjects primary means of
transportation to campus, and the number of classes the subject had on
Fridays. Since the actual voting was scheduled on a Friday at a
location in the center of the campus, these questions were meant to
provide some information concerning the individual costs of voting. A
final question inquired as to the subjects past play of the state
lottery. This was included to provide a crude measure of willingness
to accept risks.

Once the election was held, information on the turnout of individual
subjects was coded in machine readable form along with the data
supplied by the questionnaire. Initial analysis of the relationships
between the various independent variables and the dependent variable,
turnout, was done using simple bivariate crosstabulation technique.
This approach provides summary information and several useful
statistics which may serve as a general guide to interpreting the
relevance of each of the independent variables as predictors of the
dependent variable. Even so, a multivariate technique which allows for
analysis of the simultaneous effects of the independent variables is
called for here. Given the complexity of the environment, ordinal
analysis using crosstabulation would prove unwieldy. A common
alternative would be the use of Ordinary Least Squares estimation via
a linear regression technique. However, the non-interval measurement
of the dependent variable precludes such an approach.

The dependent variable, turnout, was coded as a binary; (0) for

89
nonvoters and (1) for voters. The use of a binary endogenous variable
presents problems for analysis using OLS regression techniques. In
particular, the Gauss-Markov assumption of homoskedasticity becomes
untenable in that error variance is dependent upon values of the
exogenous variables (see Pindyck and Rubinfeldl1981:276]). As a result
of this failure, estimates using the linear model will be both biased
and inefficient. If this were the only problem encountered here,
several generalized least squares techniques might be applied to
correct for it. However, the source of the problem is in the
dichotomous nature of the dependent variable. In theory we have a true
dichotomy, yet any linear estimation technique will treat this
variable as if it were continuous. In so doing, such techniques render
the theorized [0,1] interval meaningless. The theorized relationship
is , itself, nonlinear. Thus, to specify this relationship as a linear
function is the most substantial violation of the assumptions of
linear analysis. Clearly, a nonlinear technique which provides for a
transformation of the data such that estimates are bounded by the
[0,1] interval is required.

One technique which is particularly appropriate to the problem at
hand is Probit. The logic of the Probit technique applied to this
setting is fairly straight forward. We assume at the outset that an
individual will vote if the expected utility of doing so is "high
enough". Determinations of when utility is "high enough" are up to the
individual. Nonetheless, we assume that this determination is based
upon a theoretical (though unobserved) index: ‘Ii’. We further assume
that the index ‘Ii’ is a linear function of a series of explanatory

factors, or, in matrix notation:

90
Ii = Xi’B
where: ‘Xi’ denotes a vector of independent explanatory
factors; and, ‘B’ denotes a vector of weights which are
constant across individuals.

Each individual makes the choice between voting and not voting by
comparing the value of ‘Ii’ to a threshold level, say, ‘Ii*’, so that
if ‘Ii’ is greater than or equal to ‘Ii*’, the individual votes. The
actual value of ‘Ii*’ for any individual may be based on many factors,
but, in the limit, we assume that it is a normally distributed random
variable. Based on this assumption, we may calculate the conditional
probability of an individual voting (i.e., Ii greater than or equal
to Ii*) via transformations based on the cumulative standard normal
distribution, or:

Pr(voting: Ii) - F(Ii) ' F(Xi’B)
where: ‘F’ denotes the value of the cumulative standard
normal distribution.

From this theoretical base, a Probit procedure estimates the
probability coefficients, the vector ‘B’, attached to a specified set
of independent explanatory variables, the matrix ‘X’, using Maximum
Likelihood Estimation. Provided that the normality assumption holds,
these estimates are unbiased and efficient. Moreover, using them to
solve for the estimated individual probabilities of voting yields
estimates which are bounded by the interval [0,1]. The coefficient
estimates appear to be similar to those generated by OLS regression
estimation. However, this similarity is deceptive. As Aldrich and
Cnudde point out:

"...the fact that only an ordinal dependent variable is

observed limits and weakens the straightforward
interpretation of the coeffecients. For example,...

91

"slope" or "b" coefficients cannot be interpreted (as in
OLS regression) as the amount of change in the dependent
variable for a one-unit change in an independent variable.
Or, since the concept of variance is undefined for an
ordinally measured variable such as [the] dependent
variable, there is no analogue to the "standardized beta"
of 0L8 regression. In fact, since the dependent variable
is only ordinally measured, estimates are only unique up
to positive linear transformations... Nonetheless one
obtains an estimate of the best, weighted linear
combination of the independent variables."[l975:580-58l]

As a result of these effects, there is uncertainty as to what
statistical hypothesis tests are appropriately applied to the
individual coefficients. Still, appropriate statistics are available
for tests of Probit models taken as a whole and for statistical
comparisons between models. Thus, the results reported in the next
chapter will rely upon comparisons between models to test the notion

that intensity of preference, as measured by the payoff group

variable, is the most salient factor in explaining turnout.

CHAPTER 4

EXPERIMENTAL RESULTS

A Guide to Interpretation

In the third chapter of this dissertation a set of hypotheses was
outlined and several experimental designs were proposed to test them.
In this chapter the results of these experiments will be reported and
interpreted. Initially, interpretation will be limited to the
theoretical and statistical implications of the experimental results.
No immediate attempt will be made to extend inferences from these
results directly to the tax referendum setting. Such extentions are
not strictly justifiable in any case. The experimental setting is not
the ‘real world’ of a tax election, and the experimental designs used
here do not contain sufficient controls for external validity to argue
that these experiments simulate that ‘real world’. Nonetheless, the
experiments do provide an adequate test of the hypotheses in
abstraction. Indeed, given the controls afforded in the experimental
setting, if these models do not perform here, it is hard to believe
that they will do any better in the ‘real world’. The analysis
reported here will begin with the small group, committee decision
making experiments, results and discussion of the large group turnout
experiment will follow. The chapter will close with a discussion of
the applicability of these results to interpretations of actual tax

referenda.

Small Group Experimental Results

The raw outcomes of the thirty six small group experiments are

92

93
categorized according to the conditions of the matrix proposed in
chapter 3 (see FIGURE 1). At first glance, certain cells stand out as

particularly striking.

WW‘J‘g4‘4‘44’"

COLUMN 1 COLUMN 2 COLUMN 3
SOPHISTICATED NEITHER SOPHISTICATED
MEDIAN VOTER SOPHISTICATED CONVENER

A444¥4g4444444444444444JA444AAAAA444J###‘4‘444

 

 

 

ROW 1 I AI BI CI
AGENDA I so, 55 I 60, 65 I 90, 90 I
CONTROL I 58, 60 I 70, so I 100, 100 I
SINGLE I (55.75) I (68.75) I , (95) I
W14‘444; 44444444 I 4444444444444 I
now 2 I DI EI FI
AGENDA 1’ so, 50 'I 55, 70 I 73, 90 'I
CONTROL I 56, 65 I 70, 100 I 90, 105 I
MULTIPLE I (55.25) I (73.75) I (89.5)‘ I

Lacigr 44444444 IAAAAALAAAAAAAAIAAAAAA,A,;A,,I
ROW 3- I GI HI II
OPEN I 50, 50 ‘I 50, 50 I 50, 50 I
AGENDA I 50, 60 I 50, 65 I 70, 100 I
MULTIPLE I (52.5) I (53.75) I (67.5) I
PROPOSAL Idi‘ggég“““‘144-AI,A,,AAIA,I,,4,,P,l;;,,,I

 

The cells containing those results which are most consistent with
Niskanen-style expectations are CellC (mean I 95) and CellF (mean -
89.5). Indeed, the saphisticated conveners achieved an outcome above
the ‘bargaining range’ twice out of four experiments in CellC. As
interesting as these results are, they are balanced by the outcomes in
CellA (mean I 55.75) and CellD (mean - $5.25) which share the
institutional conditions of CellC and CellF, respectively. Of
particular interest are the results in CellA (50,55,58,60). Here the
convener possesses the theoretically strongest form of agenda control
-the single ‘take-it-or-leave-it’ offer. By the logic of the Niskanen

model, these institutional conditions are the most favorable to the

94
convener’s bargaining position. Nonetheless, the sophisticated middle
demand committee members were able to keep the range of outcomes
within ten points of their ideal.

Examining the results reported in CellA, evidence of the
sophistication of the middle demand committee members is manifest in
the effective management of risk. The ability to impose a two-point,
‘take-it-or-leave-it’ choice on the committee greatly enhances the
agenda setting power of the convener. This notion is essential to the
logic of the Niskanen model. Nonetheless, there is always the
possibility that the committee will choose to ‘leave-it’. The payoffs
associated with the zero option (the reversion level) were universally
low in these experiments. The middle demand committee member received
only four dollars for this outcome; the convener received two dollars.
While the differences between these payoffs is fairly slight, the
balance favors the middle demand committee member. By definition, the
Niskanen predicted result is the point at which the middle demand
individual is indifferent between the proposal and the reversion
level. In effect, conveners who make such proposals are asking the
middle demand voter discount their voting authority while, at the same
time, providing this voter with the means to punish them for this
affront. While the middle demand voters in these experiments were not
provided with information concerning the convener’s payoffs, they
certainly knew their own payoffs. Given these institutional rules,
there are no risks associated with a threat to enforce the zero option
in the face of the possibility of a Niskanen convener prOposal.

Notes taken of the discussion during and after each experiment

reveal how the sOphisticated middle demand committee members developed

95
and employed their bargaining strategies. In all cases, these subjects
recognized that their ability to overcome the convener’s absolute
agenda control was keyed to their ability to enforce the zero option.
By threatening this result repeatedly during discussions of the
alternatives, these subjects were able to gain concessions from the
convener. In one case reported in CellA, the middle demand voter
doggedly refused to consider any option above ‘50’, the middle demand
ideal point. This persistence was rewarded, though not until after a
prolonged period of haggling. The convener finally and absolutely gave
in; prOposing the ultimate outcome of ‘50’.

Clearly, the ability to enforce the zero option provided the middle
demand committee members with a potent threat. Even so, the discussion
notes reveal that, if the convener anticipates this threat at the
outset, there exists a viable counterstrategy. The potency of the
threat is keyed to the ability of the middle demand voter to ‘punish’
the convener. To undercut the threat, the convener must convince the
committee that the zero option is a personally acceptable alternative
and, thus, does not punish the convener. This was done by falsely
portraying the convener’s own preferences as ‘dual-peaked’. In every
case reported in COLUMN 3, the sophisticated conveners, at one point
or another during the experiment, made such a claim. When pressed for
more specific information, these subjects usually discribed their
payoff function as ‘U-shaped’ with one peak at zero and another at
‘110’. The greatest weakness in this strategy, at least as far as it
was employed in these experiments, is the fact that it is a falsehood.
If the committee does not believe the convener’s revelations, the

convener may be forced to make more concessions. Thus, the

96
credibility of the convener’s information must be established.

In order to establish their own credibility, the sOphisticated
conveners relied upon their special information. In every case
reported in COLUMN 3, the convener claimed a special expertise on the
basis of this information. In ten out of the twelve cases, this claim
was explicitly linked to appeals to the ‘common good’. While
acknowledging that the committee members knew their own preferences,
these sophisticated conveners argued that, in lacking the ability to
directly compare payoffs for any given alternative, the committee was
incapable of arriving at a result which was truly fair to all. Thus,
the committee must trust its convener to do what is right.

When these appeals to fairness and the ‘common good’ were coupled
with claims that the convener’s own preferences were ‘dual-peaked’,
the conveners made themselves appear as impartial arbiters. In agenda
control situations, the conveners argued that their proposals were
fair, based on a careful consideration of all the available
information with the notion of collective optimality guiding the
analysis. In seven out of the twelve cases reported in COLUMN 3, the
convener argued against the Condorcet alternative of ‘50’ by pointing
out that it was the middle demand voter’s ideal point; thus portraying
the middle demand voter as greedy -being willing to sacrifice the
‘common good’ for the sake of self-interest. MOreover, the conveners
frequently argued that, since the zero option was personally
acceptable, the use of it as a threat was further evidence of the
avarice of the median voter; its enforcement would be devastating to
the high demand committee member. In situations without agenda

control, this strategy proved to be the only one available. Indeed, in

97
one case reported in CellI, the low demand committee member was so
moved by the convener’s appeals to collective optimality that she
endorsed the high demand committee member’s proposal of ‘100’ over the
loud protests of the middle demand voter.

The discussion above has focused upon the particular results
reported in the matrix. It was presented to illustrate the dynamics of
the bargaining in these experiments and to highlight the common
strategies used by committee members and conveners. However, such
piecemeal evidence is not sufficient to judge the various hypotheses
proposed in chapter 3. There is a great deal of variation in the
results reported in FIGURE 1. This alone leads me to conclude that the
bargaining setting is clearly more complicated than either the
Niskanen or Bowen models would allow. Nonetheless, statistical tests
of the various hypotheses (using the F-distribution) should shed
further light on these matters. The results of these tests are

reported in TABLE 4-1.

4’4A44‘44#4_4444A##_4_44‘_4#444444444#A44444A4444AAAJJJA4444444#44

[F(1,22)]
ROW 3 vs. now 2: 0.41 now 2 vs. ROW 1: 0.002

COLUMN 1 VS. COLUMN 2: 5.98* COLUMN 3 VS. COLUMN 2: 7.23*

#44!44¥4J44444444444444!44#4##4#4J4#4AA4##4##14JA4J4444J44444J

 

The test used in this analysis is commonly called a ‘test of the
difference in means’. The null hypothesis suggests that the true
population means are equal. Thus, this hypothesis assumes that the
variables which separate the various groups produce no effect, or,

alternately, that the samples could be drawn from the same theoretical

98
population. The F-ratio is calculated by partitioning the sums of
squared deviations from the ‘grand mean’(the mean value of all
individual cases included in the test). The nature of these
partitions determines the degrees of freedom.used in calculating the
F-ratio. In the statistics reported in TABLE 4-1, a single partition
was used to calculate the F-ratio. For instance, to test the
difference in means between ROW 3 and ROW 2, the ‘within groups’ sum
of squares was calculated by summing the squared differences between
each individual case and its row mean. In effect, this tests the
independent effects of the variable which separates the two rows and
regards all other sources of variation as irrelevant to the analysis.
Thus, the statistics for tests between rows consider only the effects
which can be attributed to the institutional differences in the agenda
control variable; tests between columns consider only the effects of
the sophistication variable.

With respect to the hypotheses proposed in chapter 3, the results
reported in TABLE 4-1 are mixed. The first hypotheses, which argues
that the absence of agenda control should lead to a significant bias
in the outcomes, is tested by the comparison of ROW 3 to ROW 2. Here,
the F statistic does not allow us to reject the null hypotheses.
Similarly, the third hypothesis, which argues for significant
differences between the single proposal and multiple proposals
settings, is not supported by the comparison statistics between ROW 2
and ROW 1. Only the statistics relevant to tests of the second
hypotheses, the biases resulting from the location of ‘sophisticated’
subjects, allow a rejection of the null hypothesis at 95% confidence.

It should be noted that the statistical tests reported in TABLE 4-1

99
are, perhaps, the most naive tests of the proposed hypotheses in that
they assume no interaction between variables. That is to say that,
when testing the hypotheses concerning the effects of agenda control,
the potential confounding effects of sophistication are ignored. Of
course, the same can be said for tests of the sophistication effects
ignoring confounding variations from the institutional variables.
There, however, the sophistication effects still proved significant.
In sum, these statistics do provide evidence to support the hypotheses
concerning the relevance of sophistication, but to dismiss the
relevance of the various types of agenda control on the basis of these
results may be premature. At the very least, the agenda control
variables may influence the nature of the sophistication effect.

To test these notions, a new set Of F statistics were calculated on
the basis of different partitions Of the sums of squares. In these
calculations the confounding effects of the variables not under test
were eliminated by partitioning the matrix so that these variables
were held constant within the test. Thus, the sophistication
hypotheses was tested by comparing cell means within each separate row
of the matrix. Similarly, the possibility of any independent effect of
the agenda control variables was tested by comparisons of cell means
within each separate column of the matrix (thus holding the '
sophistication conditions constant). The results of these calculations

are reported in TABLE 4-2.

A444.14444.4444g44444444444A!44J4#g4444jl4444J444J

[P(2,9)]
ROW 1: 38.30** now 2: 6.11* now 3: 1.30

COLUMN 1: 0.39 COLUMN 2: 2.63 COLUMN 3: 3.33
(*) indicates significance at .05 or better

    
   

I. . I- I -I
, v a d'uv 4—1'

Taking each row of the matrix as a whole and comparing the cell
means, the null hypotheses can be rejected with at least 951
confidence in ROW 1 and ROW 2. Here, sophistication appears to have
produced significant effects. The null hypothesis cannot be rejected
at this level of significance from the F statistic reported for ROW 3.
Itshould be noted that the experiments in this row of the matrix
employed an open agenda and that the results reported within the row
appear to be heavily weighted toward the Condorcet alternative, ‘50’.
Other experimental work (e.g., Fiorina and Plott[l978] and Berl et a1.
[1976]) has repeatedly confirmed the predictive power of the Condorcet
winner under similar circumstances. These results add to that
confirmation and further illuminate the relative importance of the
institutional condition to the success of the sophistication effect
proposed here.

For the first time in this analysis, the hypothesized effects of
sophistication have failed to be significant. This failure can be
directly attributed to the absence of agenda cOntrol in ROW 3. Thus,
the pattern established in the statistics reported in TABLE 4-2 is
consistent with the notion that the institutional variables are
relevant to the outcome. That relevance corresponds to the role of a

prior specifying condition of the sophistication effect. Nonetheless,

101
the notion of an independent significance for the institutional
variables is again not supported; the F statistics within the columns
of the matrix are universally insignificant at the .05 level.
Accepting that these variables are important as a specifying factor to
the primary effects of sophistication, further analysis can unlock the
nature of this specification effect. Another series of F tests
performed between specific cells within each row reveal a pronounced
pattern within the data. Using the central column as a control
comparisons of cell means were made between adjacent cells. These

statistics are reported in TABLE 4-3.

 

, i'SJIE ,,,,
‘[F(1.6)]'

CELLS A and B: 7.35* CELLS B and C: 25.94**
CELLS D and E: 3.37‘ CELLS E and F: 1.88 "

CELLS G and H: 0.08 CELLS H and I: 1.23

4‘44! 444444444!44444#4444444444)4J444A444J4444!4#

(:1 indicates significance at .05 or better
i I . I. . .E. Q] ] II

In the statistics reported in TABLE 4-3, the null hypothesis can be
rejected with 95% confidence between adjacent cells in ROW 1 only. It
is here that agenda control is employed in its theoretically strongest
form. Relaxing the agenda setting power of the convener to allow for
multiple proposals in ROW 2, the null hypotheses can be rejected with
95% confidence for the row as a whole, but not between adjacent cells
within the row. Here, for the first time in this analysis, is evidence
that the possibility of multiple proposals may have a specifying
effect upon sophistication. This would support the third hypothesis
proposed in chapter 3. Of course, nothing proved statistically

significant from.the tests performed on ROW 3.

102

Taking the results reported in TABLE 4-2 and TABLE 4-3 together, a
pattern of interaction becomes clear. As we move upward through the
matrix, the relationship between sophistication and its hypothesized
effects tends to increase in strength. While all of this tends to
support arguments for the relevance of agenda control and multiple
proposals (as hypothesized), this support is, nonetheless, indirect. I
have yet to provide direct evidence to confirm the initial hypotheses
with respect to these factors. Such evidence, however, can be
provided. The hypothesis tests reported in TABLE 4-1 assumed no
interaction between the variables which structure the dimensions of
the matrix. The results reported in TABLE 4-2 and TABLE 4-3 indicate
that this assumption is untenable. Assuming that such interactions do
affect the outcomes, a more appropriate set of statistical tests can
be performed. Such tests are known as ‘multiple comparisons’ tests. In
using these techniques, the ‘within groups’ sum of squares is
calculated using partitions which reflect the effects of several
variables simultaneously. In this case, for instance, to test the
difference in means between ROW 3 and ROW 2, the ‘within groups’ sum
of squares is calculated by summing the squared differnces between
each individual case and its cell mean. The results of ‘multiple

comparisons’ tests on the hypotheses are reported in TABLE 4-4.

 

 

ROW 3 VS. ROW 2: 2.98* ROW 2 VS. ROW 1: 9.48**

COLUMN 1 VS. COLUMN 2: 3.32* COLUMN 3 VS. COLUMN 2: 4.45**

AAAJJJ#4!444444‘J."4#_‘#_‘4#AA#AA‘A‘4!!!4444#444¥441A4¢4J#AAAAJA

(*) indicates significance at .05 or better

«.4.me

103

The results reported in TABLE 4-4 do provide support for the
hypothesized relevance of agenda control and multiple proposals.
Still, it is important to note that they do so only in terms of
interactions with the sophistication variable. Taking the four tables
of statistics together, the primary effect, which has maintained
throughout these analyses, is the sophistication effect. The agenda
control and multiple proposal variables are relevant in that they act
as prior specifying conditions for this effect. Another way of
putting this would be to say that their relevance is on the order of
the old adage that ‘you cannot make bricks without straw’. Without the
power of agenda control, a sophisticated convener is hard put to
defeat the Condorcet alternative. Without a sophisticated convener,
the power of agenda control -even given the single proposal setting—
cannot guarantee the Niskanen predicted result. Moreover, given a
sophisticated middle demand voter, these same conditions can be turned

as tools against the convener.

Testing the Piele and Hall Hypothesis

The actual turnout for the Piele and Hall experiment was 16.91 of
the sample, or 35 out of 207 subjects. Data concerning the independent
characteristics of the subject pool were coded and organized into nine
variables:
BENI - membership in a particular payoff group
(0 - 0.25,1 - 0.50, 2 - 0.75, 3 - 1.00, 4 - 1.25, 5 - 1.50, 6 B 2.00,
7 - 3. 50, 8 - 5. 00, 9 - 6.50) ‘ ‘ ' '

PID - party identification
(1 = Republican, 2 - Independent, 3 - Democrat)

104
PART - partisanship strength index
(1 I strong Republican, 2 I Republican, 3 I Independent, 4 I Democrat,
5 I strong Democrat)

EXEFF - index of external efficacy
(1 I very high, 2 I high, 3 I medium, 4 I low, 5 I very low)

DUTY - index of sense of citizen duty
(1 I very high, 2 I high, 3 I medium, 4 I low, 5 I very low)

DIST - distance subject lived from.campus [a ‘cost’ variable]
(1 I over 5 miles, 2 I 2 to 5 miles, 3 I less than 2 miles,
4 I on campus resident)

TRANS - primary means of transportation to class [a ‘cost’ variable]
(1 I car, 2 I bus, 3 I bicycle, 4 I walk)

CLASS - number of other classes subject had on fridays [ a ‘cost’
variable]
(1 I 2 or more, 2 I 1 class, 3 I none)

RISK - frequency of play on the state lottery
(l I regularly, 2 I occasionally, 3 I never)

Once the data was coded, a series of bivariate crosstabulations
between each of the independent variables and the turnout variable
were run. TABLE 4-5 reports summary statistics for each Of these
crosstabulations. Chi square is used as a measure association and

Kendall’s tau ‘c’ is used as a measure of strength and direction.

 

 

IND . VAR . I BENI PID PART

CHI SQUAREI 19.05* 0.39 2.10
df- 9 ~ 2 4
TAU ‘c’- .20* —.03 -.05
IND . VAR . I EXEFF DUTY DIST
CHI SQUAREI 3.40 4.31 5.13
de 4 4 3
TAU ‘c’- -.05 .03 .05
IND. VARI TRANS CLASS RISK
CHI SQUAREI 1.99 0.02 2.78
de 3 2 2:
TAU ‘c’- -.06 .00 -.07

44444444J!J414AJAJAggJ!444‘4J‘44JJ44‘A44

105

The statistics reported in TABLE 4-5 are generally consistent with
the hypothesis under test. Of the nine independent variables, only
BENI, the variable meant to stand for the induced intensity of
preference with respect to the outcome, proves to be significantly
related to the turnout variable. While the tan ‘c’ is fairly low for
this crosstabulation, it, too, is significant at .05 or better. Of the
other variables, the results Of the crosstabulations are mixed. The
partisanship variables (PID and PART), while not significantly related
to turnout, do behave in a fashion consistent with the turnout
research from.the Agggiggnulggggll960] on; Republicans do vote more
often than Democrats. The crosstabulation with DUTY reports a tau ‘c’
which has a countertheoretical direction. In the case of DUTY, we
might doubt the accuracy of measure given a univariate distribution of
the variable; 562 of the sample were classed as ‘very high’, 02 of the
sample scored as ‘very low’. Of the variables, CLASS, a measure of the
number of other classes the subject attended on fridays, was the least
successful. Indeed, the tau ‘c’ for this crosstabulation shows no
relationship at all. All told, there is little to discuss of the
statistics reported in TABLE 4-5 apart from the performance of BENI.

A more thorough analySis of the bivariate relationship between
intensity of preference (as measured by membership in a particular
payoff group) and turnout is afforded by a serial breakdown of turnout
by each group. Such a breakdown is provided in TABLE 4-6. Included in
this table are the within group percentages which will make the nature
of the trend towards increased turnout through the groups more

obvious.

 

 

 

(10.0) (90.0)

0.50 2 20 22
( 9.1) (90.9)

0.75 2 17 19
(10.5) (89.5)

1.00 1 20 21
( 4.8) (95.2)

1.25 2 17 19
(10.5) (89.5)

1.50 1 21 22
( 4.5) (95.5)

2.00 5 17 22
(22.7) (77.3)

3.50 8 13 21
(38.1) (61.9)

5.00 6 15 21
(28.6) (71.4)

6.50 6 14 20

MAO). 112.92 a...

35 ‘172 207

“IAJEEE4744174’444447’147 444444444444444444444

The results reported in TABLE 4-6 indicate that the trend toward
increased turnout as the payoffs increase is far from smooth. Turnout
in the first six groups is fairly constant, averaging 8.12. Once the
payoff increases to $2.00, the within-group turnout leaps to 22.72.
The next increment increse in payoff produces a similar jump in
turnout (up to 38.12), but, thereafter, the percentage turnout
declines to a fairly stable level (averaging 29.32). Thus, the trend

established in TABLE 4-6 seems to correspond more to a step function

107
than a monotone increase. It is interesting that the crucial break in
the trend comes at the point at which the payoffs, themselves, take a
step increase in value; the increment increases from $0.25 to $0.50,
and, thereafter, to $1.50. That the increases in turnout correspond to
the increases in the increments which separate the groups is
encouraging. However, the decline in turnout in the last two groups
leads me to proclaim a perfect correspondence to theory. Nonetheless,
the relationship between intensity of preference and turnout (and the
absence of such a relationship with the other variables) is, at least,
theoretically gratifying.

In keeping with the analysis proposed in chapter 3, multivariate
analysis of the predictive power of the various independent variables
was done using Probit estimation techniques. The limitations of Probit
do not allow for a meaningful interpretation Of the significance of
Probit coefficients for individual variables. Thus, analysis of the
particular significance of groups of independent variables was
provided by comparisons between models with different configurations
of exogenous variables. Four separate models were actually estimated.
The first of these included all the independent variables (choosing
the strength of partisanship index in preference to party
identification):

MODEL 1: Pr(VOTE) - F(b + bBENI + bPART + bEXEFF + bDUTY + bDIST +...
bTRANS + bCLASS + bRISR + e)

The second model removed the standard Survey Research Center variables
related to partisanship and attitudes of efficacy and citizen duty,
but retained ‘cost’, ‘risk’, and intensity variables:

MODEL 2: Pr(VOTE) I F(b + bBENI + bDIST + bTRANS + bCLASS + bRISK + e)

The third model removed the ‘cost’ variables, but retained the ‘risk’

108
and intensity variables: A
MODEL 3: Pr(VOTE) - F(b + bBENI + ansx + e)
Finally, a model was run using just the intensity variable:
MODEL 4: Pr(VOTE) - F(b + bBENI + e)

A standard test of the overall significance of a Probit model is
afforded by the calculation of -2.0 times the log of the likelihood
ratio provided by the model. This statistic distributes as a Chi
Square with (R - 1) degrees of freedom, and can be interpreted as

such. The results Of these calculations are reported in TABLE 4-7.

_‘#A44.4.;4444.4444J#A44¥A4_‘J¥1‘!444A4A4A4444¥44 4.4.4

 

TABLEEQEL
MODEL 1: ' MODEL 2:
19.39* 17.69**
df I 8 df I 5’
MODEL 3: MODEL 4: .
15.57*** 11.46***
df I 2" df I'l“

4444A!!44#44#4!4J4#44#4444_‘4AA‘A‘AA444444g444J44

(*) indicates significance at .05 or better
(**) indicates significance at .01 or better

‘***2 infllfiflififl EIEHILISEHSE it QQ] 9‘ EEIIEE

While all of the models estimated prove significant at .05 or
better, it is interesting to note that none prove more significant
than MODEL 4. Given the results of the bivariate crosstabulations, it
is reasonable to assume that BENI is the primary factor which
maintains the significance across the models. Indeed, given the
general decline in significance as variables are added to the
equation, it appears that the additional variables do little more
than increase the aggregate noise in the system. Still, better means
of comparing the models should be discussed before passing judgement

on their relative predictive power. One such technique involves a case

by case analysis of correct versus incorrect predictions generated by

109
each model. The models are then compared through calculations Of rank
correlations between actual and predicted values of the dependent
variable.

In performing this analysis, the low turnout within the sample
(16.92) creates problems for the predictive power of the dichotomous
Probit package used to estimate the models. Since the dependent
variable is a strict dichotomy, a case is normally predicted to vote
if the probability of doing so exceeds .5 (based on combined additive
effect of the various variables in the model). However, the
estimations of these probabilities are derived from maximum likelihood
estimations of the fit to the variation within the dependent variable.
The low turnout of subjects in the experiment produced so little
variation in the dependent variable that individual predictions were
systematically skewed toward not voting. In order to overcome this
skew, a Baysian adjustment was made to force the models to predict
the distribution of voters to nonvoters in roughly equal proportions
to the actual distribution in the turnout variable. Such adjustments
are common when using Probit and Discriminant Functions analysis
techniques, and have the effect of shifting the probability curve
toward the origin in the same manner as one might force an OLS
regression line through the origin. Having made such an adjustment,
crosstabulations of actual versus predicted values of the turnout
variable were generated and a Goodman and Eruskal’s tau was calculated
as an asymmetric measure of rank correlation. These results are

reported in TABLE 4-8.

 

 

MODEL 1: MODEL 2:

PREDICTED PREDICTED
EDIE w 2913. w
w ‘ ‘
VOTE: 10 25 9 26
31 176 30 177
TAU: .030 .021
MODEL 3: MODEL 4:
PREDICTED PREDICTED
w W w W
m
VOTE: 10 25 12 23
NOT VOTE- ‘21... “129...“. ‘22".
32 176 166
TA“. 0027 ' 0027

4A44444444AJ¥444A#4444J44#4444444;1444A‘4444444J444

What stands out most about the results reported in TABLE 4-8 is the
lack of differences between the four models. This is rivaled only by
the generally low tau statistics. The interpretation of these results,
taken together with those reported earlier, is both straightforward
and inconclusive on various levels. On the straightforward side, it
can be said that the payoff variable is a significant factor in
accounting for turnout. Thus, there is support provided for the
hypothesis under test. Nonetheless, the Probit analysis indicates that
a substantial amount of the variance in turnout is not explained by
the payoff variable or any other variable measured in the experiment.
This may be the fault of the design of the measurement instruments or
it may indicate that other factors, which remain unmeasured, are at
work. In either case, while these results do provide some support for
the Piele and Hall hypothesis, the evidence here does not lead to the

type of certainty that was hoped for at the outset.

111

Application to the ‘Real WOrld’

The weak case provided for the Piele and Hall hypothesis by the
experimental results might be aided by additional evidence taken from
the strategies and behaviors of actual organizers in a tax referendum.
If the proof of the ‘pudding’ is in the ‘eating’, perhaps the best
evidence of the relevance of intensity of preference to the management
of turnout is in ‘real world’ examples of its use. This is, in fact,
the aim of the next and final chapter of this dissertation. There,
too, the results of the small group experiments will be applied in
interpreting the dynamics of an actual referendum campaign. Just as
sophistication proved of primary importance in the experimental
results, so too its importance can be demonstrated in the playing out
of actual strategies in the ‘real world’ setting. The components of a
sophisticated strategy will be different between the two environments.
This is expected given the context specific nature of the concept of
sophistication. Nonetheless, there exist several plays and stratagems
which have analogous roles between the two environments. These
parallels will be addressed, as will the immediate relevance of the

analytic models themselves, in the next chapter.

CHAPTER 5
MAKING SENSE OUT OF ACTUAL TAX REFERENDA

In chapter 4 of this dissertation the results of several series of
experiments designed to test the hypotheses proposed in chapter 3 were
reported and discussed. So far, interpretations of these results have
been limited to discussions of their implications in terms of the
theoretical expectations given in the experimental setting.
Nonetheless, the ultimate purpose of both the hypotheses and the
experiments is to provide a basis for the interpretation of ‘real
world’ tax referenda. To this end, this chapter will present an
analysis of an actual tax referendum. The election was held on
September 2, 1980 in Chippewa County, Michigan (Saute Ste. Marie
area). The issue included two separate proposals of millage taxation
to finance the operation of the local public schools. The history of
events leading up to September 2nd, the design and implementation of
campaign strategies on the part of the school board, and the actual
results and outcome of the referenda, will be the subject matter to be
interpreted here. It is in this interpretation that the relevance of
the experimental results must be established.

The small group experiments provide a plausible case for the
salience of agenda control, multiple elections and sophistication in
the performance of the theoretical models in abstraction. In applying
the Niskanen and Bowen models to the ‘real world’ setting, we must, at
least, be mindful of the possible effects of these factors. Moreover,
the strategies observed in the experimental setting may have certain

analogues in the behavior of actors in actual tax referenda. Still, a

112

113
direct extention of the experimental results to the ‘real world’ is
neither methodologically justified nor possible. Thus, care must be
taken to define the factors -particularly ‘sOphistication’- in context
specific terms before they can be applied in this case study.

The evidence presented in the Piele and Hall experiment presents a
relatively weaker case in support of the notion that turnout
management may be a crucial factor affecting the outcomes of actual
tax referenda. Nonetheless, the discussion presented here will argue
that this factor, coupled with sophistication, may prove to be the
most important strategic asset available to a governmental agenda
setter in such environments. It is important to note that the
salience of turnout management will be discussed in tandem with
sophistication. Given the small group experimental results, the same
can be said for the salience of multiple elections and agenda control.
Sophistication appears to be the glue that holds all these hypotheses
together. With that in mind, this chapter will begin with a discussion

of the meaning of sophistication in the tax referendum setting.

What does sophistication mean?

In the experimental setting, sophistication was operationalized in
terms of familiarity and experience with abstract analytic systems and
gaming environments similar to those used in the experimental design.
At that time, it was argued that sophistication is a context specific
concept. What stands for sophisticated behavior in the experiments
does not carry over into other environments. While the behavior
manifest by the SOphisticates in these experiments tended to support

the operationalization of the concept in that environment, it cannot

114
be said that these ‘sophisticates’ would fare equally well in the tax
referendum setting. How, then, should sophistication be defined in
this case study?

A straightforward transposition of sophistication between the two
environments might define the concept as familiarity and experience
with the management of tax election campaigns. Nonetheless, most tax
election campaigns are managed directly by local school boards or
administrations. As important as the results of these elections may be
to the wellbeing and success of these agencies, campaign management is
secondary to the primary tasks of setting and administering policy for
the schools. Thus, despite their experience, these actors tend to
remain relative amateurs when it comes to the intricacies of political
campaign management. On the other hand, a tax referendum is a fairly
unique campaign setting. It does not immediately follow that
professional campaign managers, who work for the election of
candidates and party slates, would have the requisite experience to
design and implement a sophisticated strategy in a tax election. Such
experience may, indeed, be helpful, but it is not sufficient to define
the concept of sophistication.

When it comes to the conditions of actual tax elections, perhaps the
best way to develop an understanding Of the concept of sophistication
is in terms of the actual strategies implemented by various actors. In
these terms, analogies between the experimental setting and the ‘real
world’ can be drawn. For instance, the ‘SOphisticated’ conveners in
the small group experiments designed and implemented their respective
strategies through the careful analysis and use of their ‘special

information’ concerning the distribution of preferences within the

115
committee. Evidence of sophistication in the management of actual tax
elections should start with a similar analysis of the distribution of
preferences within the electorate. Without such knowledge at the
outset, it would be difficult to design an effective campaign strategy
which makes Optimal use of the various resources available to the
campaign manager.

As in the Niskanen model, analysis of the distribution of
preferences is a key factor in determining ‘what the traffic will
bear’ in terms of actual proposals of different levels of taxation.
Moreover, from the basis provided in this analysis, a campaign manager
may identify different demand groups within the electorate and design
selective appeals for each, or simply choose to ignore certain groups.
Here too, effective management of turnout must be based upon knowledge
of the distribution of preferences. All these elements will figure
into the discussion of sophistication presented here. To illustrate
the importance of the gathering and analysis of information concerning
electorate preferences, this case study will begin by examining the
situation in the Sault district prior to the organization of the
September 2nd campaign and then compare past efforts to the plan

developed for that particular election.

The background of the Sault campaign

An examination of ten years of tax referenda in the Sault Ste. Marie
Area Public School District provides a basis from which to judge the
relative sophistication of the local school board in its role as a
governmental agenda setter. In the period between June 1971 and

September 1980, seventeen separate millage proposals were offered to

116
the Sault area electorate. All but the September 1980 election were
managed in-house by the school board itself. Ten of these seventeen
proposals were successfully passed, which, on the surface, seems like
a fairly good report card on the electoral skills of the school board.
Still, closer examination of this record stands in opposition to such
a rosey appraisal.

While Michigan law provides that operational millages for local
public schools must be periodically renewed, the period of renewal can
be as long as three years. Nonetheless, all but one of the seventeen
proposals offered by the Sault Area school board during the 1970’s
were single year proposals. This requires the school board to come
back to the electorate every year; repeatedly taxing the resources of
the board. The reasons for going with this single year period are hard
to fathom. Indeed, the June elections of 1971, 1972, and 1973 all
offered the same levy of 15 mills. If nothing else, a single three
year proposal would have been less demanding in terms of campaign
resources. With the election of June 1974, the millage proposal was
increased to 16 mills. Nonetheless, the proposal of the following June
returned with the same 16 mills. Given this, it is unlikely that the
school board, in proposing a one mill increase, was ‘testing the
waters’ or attempting to develop a more elaborate bargaining strategy
with the electorate. Rather, it appears that the board was simply
adjusting the old 15 mill proposal for inflation. If the board feared
the use of a three year period on the grounds that inflation might
have unpredictable effects, the option remains of proposing additional
mills at any time within the three years. While Michigan law limits

the number of tax referenda per fiscal year to two, this leaves five

117
opportunities for adjustments within the three year period. Thus,
through the first five years of the 1970’s, the Sault Area school
board seemed to have little grasp of the strategic potential given in
their role as agenda setter.

The proposal behavior of the school board goes through a series of
changes in the second half of the 1970’s. Given cuts in state lump sum
grants and subsides in 1976, the June 1976 proposal was raised from
the previous budget of 16 mills to 19.3 mills. This dramatic increase
was not well recieved by the electorate, which turned down the
proposal by a wide margin. Fearing school closings, the board came
back with a more modest proposal in August. Michigan law allows a
school board to partition its proposals separate ballot issues. So
long as all the parts are voted on on the same ballot, the election
counts as only one referendum against the statutory limit. The August
proposal made use of this by offering two separate issues: 16 mills
and 2 mills. The board felt confident that the 16 mill proposal would
pass; it was simply a reproposal of the status quo. Thus, the
campaign focused on the additional 2 mills. Taken together, the two
proposals were less than the original 19.3 mill request. This was
argued to be a gesture of good faith bargaining with the electorate.
The additional millage was then defended as absolutely necessary given
inflation and changes in state funding. The strategy was successful,
but, more importantly for this study, it marks the first recognition
on the part of the school board of the adversarial nature of its
relationship with the electorate.

The school board’s increased sensitivity to the strategic

characteristics of the tax referendum environment, which was manifest

118

in the proposing behavior of the 1976 elections, carried over into the
the next round of elections. In June 1977, the board again used a two
issue proposal, setting the first at the new status quo (18 mills) and
seeking another 1 mill in addition as a separate issue. Unlike
previous elections, however, the 18 mills were offered as a three year
levy. With the uncertain economic climate of 1977, the board felt that
setting a sold base for three years would minimize the risks of school
closings by limiting bargaining to small increments of additional
millage. In keeping with this notion, the 1 mill proposal was offered
as a one year levy. In this case, the use of a one year period had
some strategic value. The school board was able to portray the 1 mill
proposal as a temporary measure necessitated by declines in state
funding due to the growing economic recession in Michigan. This
strategy appears to have met with some success. While the initial 1
mill proposal failed, a subsequent ‘bargaining down’ to 0.8 mills
passed in August. 4

Given that the economic conditions in Michigan did not improve, the
argument of temporary additional millages ran quite thin. Indeed,
progressive declines in state support coupled with a growing sense of
economic vulnerability on the part of the electorate had the dual
effect of making additional millages more necessary and less likely to
pass. Two one year proposals in June 1978 and 1979 failed, leaving the
district with an 18 mill operating budget through the bulk of the
three year period. Interestingly, these two proposals show a certain
amount of strategic planning despite their failures to pass. With a
sense of security provided by their 18 mill base budget, and with

further declines in state support, the school board showed no

119
willingness to bargain through a series of elections. The June 1978
proposal of 1.5 mills was not followed by an August attempt, rather,
the following June election offered another increase to 2.5 mills.
Nonetheless, by early 1980, the board recognized that the margin of
safety provided by the three year proposal was coming to an end. It
was clear to the board that a new strategy of campaign management was
required.

At this point in the narrative the stage has been set for a
discussion of the three elections of 1980. It is there that the best
case can be made for an analysis of the essential elements of a truly
sophisticated strategy in the tax referendum setting. The discussion
above illustrates the gradual development of a modicum of
sophistication on the part of the Sault Area school board in both
their proposing behavior and their ability to exercise and exploit
measures designed to contain the risks associated with different
proposals. Indeed, the fact that there were three millage elections in
1980 without a petition drive to qualify the third bears witness to
this development. Since the fiscal year of the school district runs
from June to June, the first election of 1980 was called in March.
This election was, therefore, charged to the 1979 - 1980 fiscal year;
leaving two opportunities before the expiration of the 18 mill budget.
Even so, the type of sophistication demonstrated up until this point
was limited to the use of institutional conditions governing agenda
control.

In structuring the actual proposals the school board relied on the
notion of incremental change from a given status quo. With the

elections of 1980, a modest increase beyond the 18 mill base was

120
precluded by three years of deteriorating economic conditions. The
board felt that it would have to pass a minimum 6 mill increase just
to meet the basic operating expenses for the next fiscal year. The
difference between this strategic situation and those encountered in
the past ten years meant another shift in the development of board
strategy. Given this, the experiences of the 1980 elections should be

discussed separately.

The elections of March and June, 1980

In planning strategy for the March 1980 election there was general
consensus within the school board on the necessary size of the
prOposed levy (24 mills) and the need to use a single issue format for
the ballot design. The 24 mill figure was provided by the
superintendent, based on the projected budgetary requirements of the
district at full operation. The single issue ballot design reflected a
general agreement on the type of strategy used to persuade voters. It
was felt that the 24 mill levy was a functional minimum, and could be
campaigned for as such. Given the superintendent’s budgetary
projections, nothing less than 24 mills would be sufficient. Moreover,
given recent electoral history, partitioning of the levy into
separate issues would do little beyond increasing the probability that
only a portion of the 24 mill figure would pass. Thus, the decision
was made to go with an ‘all-or-nothing’ proposal.

As discussed in chapter 2 of this dissertation, the use of an ‘all
-or-nothing’ agenda setting strategy (similar to that which is used in
the Niskanen model) increases the level of risk to the agenda setter.

This notion was recognized and accepted by the board. While it was

121
felt that a partitioned proposal would hurt the chances of passing the
full 24 mill levy, the board preserved its maximum.opportunities for
serial bargaining over multiple elections by calling the election
within the old fiscal year. Thus, the board hedged its bet; what
appeared to be a ‘take-it-or-leave-it’ offer was, in fact, still open
to bargaining. In a sense, the board hoped to combine the advantages
of the single election format and a multiple election setting.

A related aspect of the board’s attempts to manage risk is given in
the selection of a one year term for the proposal. While arguments
concerning the temporary nature of the school district’s fiscal crisis
had failed in the past, board members continued to have faith in its
effectiveness as a tool of persuasion. The failures of this approach
in the past two millage elections were attributed to the relatively
passive style in which previous campaigns had been run. With a more
aggressive persuasion campaign, the board hoped to make effective use
of the ‘temporary crisis’ argument. Before judging these hopes as
naive, it should be noted that such arguments were standard political
rhetoric in Michigan at that time. The recession of the late 1970’s
was, after all, a national problem. Moreover, Michigan had been one of
the hardest hit states in the union. Cutbacks in state support to
public education had been debated and passed in the state legislature
as ‘temporary measures’. This style of argument had not abated despite
repeated use over the last few years. Further, there was growing
sentiment in the state capitol for a restoration of some of the
cutbacks to education (at the expense of the social services budget)
if the fiscal crisis persisted.

Clearly, much of the school board’s strategy in managing the millage

122
campaign was keyed to the notion Of aggressive persuasion. The board
felt that the only way to pass a 24 mill levy would be to convince the
electorate that the school district’s problems were dire and not
solved by partial measures. To do this, the board felt that the
standard media were not sufficiently reliable to be trusted to get
their message across. Therefore, a plan was developed to supplement
the usual media campaign of press releases and canvassing with what
amounted to a traveling road show. Approximately two thousand dollars
were set aside to develop a multimedia presentation which was to be
given at public meetings in various areas of the school district. The
presentation included a slide show with a taped narration and a
battery of speakers.

The nature of the appeal in this presentation was directly analogous
to the arguments used by the ‘sophisticated’ conveners in the small
group experimental setting. The presentation stressed the value of the
public school system as a public good shared by all of the community.
Apart from cataloging the various activities of the public schools,
the program went on to stress the importance of a good school system
to the community as a whole and to individuals without school aged
children. In short, the maintenance of a good public school system was
portrayed as a policy which was fair and beneficial to all members of
the polity. A consistent theme running through the presentation was
the notion that the only way to preserve such a good school system was
the passage of this particular 24 mill levy. Nothing less would be
fair or socially efficient. This is an analogue to the convener claims
of ‘dual-peaked’ preferences in the small group experiments. The

school board was essentially arguing that it would prefer a zero

123
provision to a compromise appropriation below the 24 mill request.

Along with the actual design of this multimedia presentation, a plan
was developed for its deployment to particular precincts within the
school district. This plan was based upon an examination of the
marginal turnout by precinct in the June 1979 election, and marks the
first instance in which the turnout factor was taken seriously as an
aspect of the board’s strategy. However, the strategy that was
ultimately implemented showed little grasp of the relevance of turnout
in determining the outcome.

Since the multimedia presentation was a persuasion focused campaign,
the board chose to emphasize its use in those precincts which had been
least favorably disposed to the 1979 millage proposal. The logic
behind this approach argued that most ‘no’ votes in the 1979 election
were based in ignorance of both the value and fiscal needs of the
local school system. If these voters could be shown the truth of the
situation, the potential for converting large blocks of them into
‘yes’ voters would turn the tide in favor of the proposal. Even if the
precincts ultimately cast a plurality of ‘no’ votes, if the campaign
kept that margin small, the ‘yes’ votes in stronger precincts might
carry the day.

Several factors point to the potential fallacies in the lOgic behind
the school boards use of the persuasion campaign. First, the board
probably over estimated the number of voters who could be persuaded to
change their minds from their previous revealed preferences. Since no
effort was made to estimate this potential via survey research, the
board was building its strategy on what amounted to idle speculation.

Also, the precincts which had shown the least support for the 1979

124
millage proposal were largely in the rural areas of the district.
These were sparcely populated. Even a major shift in support in the
rural precincts would not generate enough ‘yes’ votes to justify the
expense of resources to mount an aggressive campaign there. Since the
1979 millage had gained a plurality of support in only one of the
urban precincts, a substantial effort was required to shift the
balance in other close urban areas. In the long run, the potential for
aggregate vote gain was greater in these areas than in the rural
communities. Finally, if the Piele and Hall hypothesis holds, any
attempts to mount a high profile publicity campaign in areas know to
be poorly disposed to millage proposals are likely to increase both
the aggregate turnout in those areas and the turnout of ‘no’ voters in
the bargain.

Given the above, the results of the March election reflected the
flawed logic of the media campaign. Aggregate turnout declined between
the June 1979 election (4219 votes or 362) and the March 1980 attempt
(3978 votes or 352). The bulk of this decline was experienced in the
urban precincts. Since the primary support for the earlier proposal
was found in these areas, this had the effect of actually decreasing
the raw total of ‘yes’ votes from 1668 in June 1979 to 1639 in March
1980. The only precincts which experienced increases in turnout were
inthe targeted rural areas. As expected under the logic of the Piele
and Hall hypothesis, the majority of this increase went to the ‘no’
vote total. Thus, the final result was a rejection of the proposal by
a 695 vote margin.

In assessing the outcome of the March election, the school board was

most troubled by the decline total ‘yes’ votes derived from the urban

125
precincts. In light of this result, it was decided that the mutlimedia
campaign should be retargeted to these areas for the June 1980
election. The rural strategy was wholly abandoned. Also, themillage
proposal was decreased to 23.6 mills. This would be offered as a good
faith effort to bargain with the electOrate by cutting the proposed
Operating budget to a bare subsistence level. In all other aspects,
the board’s strategy was unchanged from the March campaign.

The results of the June election proved more favorable to school
board’s position. While the aggregate turnout was essentially
unchanged from the March results (4072 votes or slightly more than
352), turnout within precincts showed increases in the urban areas and
decreases in the rural communities. The proposal gained a plurality of
support in four of the thirteen precincts within the district (all
in urban or suburban areas) and ran dead even in a fifth. The
aggregate ‘yes’ vote was increased to 1961, a gain of 322 votes over
the March total. Still, despite these improvements, the prOposal was

defeated by a 150 vote margin.

Designing the September campaign

In light of the June results, the school board was pessimistic with
respect to the possibility of ultimately passing a reasonable millage
proposal. The failure of the June campaign left them with little time
to develop a new strategy before the scheduled start of the new school
year. Still, without the passage of a new levy, there would be no new
school year. Faced with this dilemma, the school board chose to give
over the planning and design of the September campaign directly to the

superintendent.

126

The budgetary preferences of the superintendent were clear from the
outset. The 24 mill proposal was, after all, his original design.
Moreover, it was felt that the minimal change of the proposal to 23.6
mills in the June election had no effect on the outcome. Thus, the
September proposal would return to the 24 mill level. Even so, there
was something to be said for the appearance Of a willingness to
bargain. In aid of this, a second proposal was included on the ballot
asking for 1.5 mills additional levy. The 1.5 mill proposal was
regarded by both the superintendent and the board as a ‘pipe dream’.
Nonetheless, it made a good stalking horse for the 24 mill proposal.
It gave tax protesters something to vote against. Moreover, if debate
over the virtues of the 1.5 mill proposal moved to the fore, it might
divert negative attention from the 24 mill proposal. Further, if the
two proposals both passed, so much the better.

While the agenda setting stage of the September election strategy
appeared set, neither the superintendent nor the school board had a
clear idea of how the campaign should be run. To develop an
operational plan, the superintendent asked a member of his staff to
act as consultant. Within the school district administration was an
individual who had extensive experience in the operations of political
campaigns. This person had been active in Democratic party politics in
the Sault area at various levels including a term as the local party
chair. While he had never attempted to manage a tax election, he had a
working knowledge of the mechanics of any political campaign. The
superintendent asked that he put aside other duties and develOp an
analysis of the possibilty of passing either or both of the

proposals.

127

Much to the surprise of both the board and the superintendent, the
consultant’s report was favorable. In so doing, however, the report
was critical of the design of the past to campaigns. Both the March
and June campaigns had emphasized persuasion and high levels of media
exposure. These activities may have stimulated turnout, but there
seemed to be as many or more ‘no’ voters gained as ‘yes’ voters. At
this point, given the experience of two very similar proposals, it was
unlikely that many potential voters were undecided over the 24 mill
issue. Given fewer people to persuade, persuasion activities should be
deemphasized and limited to just those individuals who could be
identified as undecided on the 1.5 mill question. Thus, the first
major recomendation of the consultant was the abandonment of the
multimedia presentation and high media exposure in general.

Earlier efforts to manage the style of appeal by precinct were also
criticized. There appeared to be a certain amount of justification for
ignoring the rural precincts went targeting resources for the
campaign. Nonetheless, using the precinct as the unit of analysis to
manage a turnout campaign had led to mixed results. In comparing the
results of the last two elections, the consultant estimated that a
large pool of ‘yes’ voters had voted in only one of the two elections.
Indeed, despite the improvements in the June results, the cOnsultant
estimated that if just 602 of those who voted ‘yes’ in March but did
not vote in June could be turned out, the 24 mill proposal would win
by 394 votes. This estimate was based on a plan for turnout management
in which the individual voter was used as the unit of analysis, rather
than the precinct, and an expected decrease in ‘no’ voter turnout due

to deemphasis of generalized media. Passage of the 1.5 additional

128
mills was less likely under this plan.

Clearly, a large number of those who would vote ‘yes’ on 24 mills
would be undecided or negatively disposed toward the 1.5 mill
proposal. The key to victory would be the location and persuasion of
those undecided voters. Assuming 750 undecided 1.5 mill voters, a
targeted persuasion campaign was estimated to be able to generate a
70/30 split by election day. This would yield a hypothetical 300 vote
advantage for the 1.5 mill issue amoung previously undecided voters.
This margin might be sufficient to pass the issue, or at least keep
the results close. Moreover, even if the 1.5 mill proposal failed, the
type of targeted persuasion campaign advocated would have the residual
effect of shoring up support for the 24 mill proposal. This should
lead to an even greater margin of victory than that projected by just
the targeted turnout campaign.

In keeping with these goals, the consultant presented a campaign
plan with three integral stages of activities: Voter Identification;
Targeted Voter Persuasion; and, Targeted Voter Turnout. In order to
structure the targeted activities of persuasion and turnout
management, information concerning the distribution of individual
voter preferences had to be gathered and analyzed. This phase of the
campaign would take approximately two weeks to complete. Once the data
was available, the campaign would shift to persuasion tactics directed
toward those voters which had been identified as undecided on the 1.5
mill question while supporting the 24 mill issue. At the same time, a
campaign to motivate those voters who were identified as favoring
either 24 or 25.5 mills, but who had not voted in either of the last

two elections, was to be initiated. In addition, all voters who had

129
been identified as favoring the 24 mill proposal were to be contacted
with respect to their possible need for an absentee ballot. Finally,
as the campaign pulled within ten days of the election, activities
would shift to turnout management.

The entire campaign plan was distinctly labor intensive. The plan
called for an estimated 1000 person/hours of labor, but only $717
total material costs (most of these related to direct mailings). This
represented a major change from the previous two campaigns. Both of
these had been media intensive with relatively high associated
material costs, but only limited requirements in terms of labor.
Clearly, in order to implement the plan, a pool Of labor needed to be
recruited beyond the small group of community volunteers who had
participated in the previous campaigns. For this pool, the consultant
turned to the unions representing the teachers (the Michigan Education
Association) and the clerical and custodial staff (the United
Steelworkers). Since most of the labor involved ‘behind the scenes’
activities (i.e., phone calling, envelope stuffing, etc.), the use of
clearly self-interested personnel such as these was not expected to
seriously hurt the campaign’s appeals to ‘the common good’. Even so,
while roughly three quarters of the campaign staff were school
employees, only community volunteers were to be allowed to participate
in such visible activities as door to door canvassing and poll
watching.

Not surprisingly, the school board chose to implement the
consultant’s campaign plan as designed and assigned him the task of
managing the campaign. With less than two months remaining before the

election, the program began apace.

130

Implementing the Campaign Plan

The first task to be addressed was not the actual conducting of the
telephone survey to identify voters, but rather, the training of those
staff members who would conduct the survey. The training sessions were
conducted in the evenings and involved a series of role-playing
exercises. In this way, staff members were given the chance to work
through ways of handling various situations which might occur during
the survey. Staff members were taught how to avoid discussing the
issues of the campaign, deflect questions concerning who was taking
the survey, as well as proper means of gaining and coding the
information needed.

The targeted sample to be surveyed included 5700 of the
approximately 12,000 registered voters in the district. These
individuals were grouped into three classes: (A) those who had voted
in both of the last two millage elections; (B) those who had voted in
only one of the last two millage elections; and, (0) those who had
voted or registered in the last two years, but had not voted in either
millage election. The survey placed emphasis on the urban and suburban
areas of the district; coding each case by precinct number to aid in
the precinct level organization of canvassing and poll watching
activities. Opinions on the two millage proposals were coded as: (l)
supporter; (2) undecided; and (3) opposition. The case identification
used was the last five numbers of the respondent’s phone number (this
cut down on administrative time in phone canvassing). Thus, a typical
coded responce would be : 02A1253362. This is a person in the second

precinct who voted in both previous elections, favors the 24 mill

131
proposal, is undecided about the 1.5 mill proposal and lives at the
given phone number. Once so coded, opponents were removed from the
sample. The sample information was partitioned by precinct and then by
the alphabetical voter code and the preference codes. From this a
series of lists were drawn up for distribution to the local precinct
organizations.

Once the survey was completed, activities turned to persuasion and
motivation. These involved direct, targeted door to door canvassing.
Only community volunteers were used for the actual contacts with the
voters. These volunteers went out in pairs, armed with campaign
literature, only to houses and individuals assigned to them on
particular lists. Three types of lists were drawn up. One type of list
included ‘C’ voters (or, rather, nonvoters) who had been coded as
favoring both millage proposals. Here the only persuasion needed was
to get these people to vote. The canvassers were schooled in a variety
of arguments which were designed to convince these individuals of the
importance of their particular vote in determining the outcome of the
election. A second type of list included ‘A’ and ‘B’ class voters who
were coded as undecided on both millage proposals. The canvassers were
schooled in a different variety of arguments which emphasized the
schools and children of the district (avoiding the issues of taxation
itself as much as possible). The only time discussions were to
voluntarily focus on tax issues themselves was when talking to senior
citizens -these were entitled to a rebate of their school millage.
Finally, a third type of list included ‘A’, ‘B’, and ‘C’ voters who
were coded as favoring the 24 mill proposal but being undecided on the

1.5 mill proposal. Arguments here stressed the notion that the 1.5

132
mill proposal was not for ‘extra’ programs, but was needed for the
same basic budget items as the 24 mill proposal.

As canvassing activities progressed, lists of voters were
continually updated. The goal was to complete the canvassing by August
29th. Interestingly, no door to door contact was made of ‘A’ and ‘8’
class voters who were coded as favoring both proposals. Nonetheless,
these voters were not neglected. All voters who were coded as
supporting the full 25.5 mills were contacted by phone concerning the
likelihood that they might need absentee ballots. This phone campaign
served to reinforce the support of these voters as well as aid in the
turnout management of absentee voters. While the ‘C’ class voters were
included in both this group and the door to door canvassing, the
additional attention was deemed appropriate given their past history
of nonparticipation. The absentee ballot campaign was designed to
conclude by August 27th. As lists of absentee voters were made-up,
.poll watchers regularly checked the lists available through the
municipal government to make sure that their voters both received and
filed their ballots. Follow up calls were given to tardy voters.

On August 20th, a meeting was called of all campaign volunteers to
coordinate activities for the turnout management phase of the
campaign. Over 50 people attending the meeting were organized into
precinct teams which played various roles. Each precinct had two
captains, one was a member of the school administration the other was
a community volunteer. This division of labor at the top reflected the
division of labor throughout the precinct team. School employees were
assigned to the tasks of telephone canvassing and direct mail appeals,

while community volunteers acted as drivers, babysitters and poll

133
watchers. The first stage of the turnout management campaign involved
the use of direct mail appeals.

The direct mail campaign involved to separate mailings.One was sent
to those voters who were identified as supporting the the entire 25.5
mills. These voters received a signed letter thanking them for their
support and giving a general pep talk on the prospects of passage if
all true supporters come out to vote. This was mailed on August 22nd
and received that fourth week of the month. The second mailing was
sent to all potential supporters of either proposal. This was a more
modest affair than the first mailing, a simple ‘thankyou’ on note
sized paper. However, it was this mailing which was expected to have
the greater effect in that it was timed through the mail to be
delivered on election day.

_ The telephone campaign to manage turnout was organized in three
phases and coordinated with poll watchers, drivers and baby sitters.
The first battery of calls went out on Monday, September 1st (the day
before the election) between 6:30 and 9:30 in the evening. The
telephone script provided a simple reminder to supporters of either
proposal and provided rides to the polls and babysitters to those who
needed them. This round of calling was the most demanding in that all
the voters identified as supporters were actually phoned. The second
and third rounds of calling went out between 3:00 and 5:00 p.m. and
6:00 and 7:30 p.m. on election day. These calls were coordinated with
the poll watchers who kept track of which ‘yes’ voters had not voted
by 2:30 and 5:30, respectively. The first phone script informed voters

that the election was very, very close" and that "if too many people

stay home the ...millage won’t be passed". The second script spoke of

134
"very little time left". Both scripts repeated the offer to supply
rides and babysitters.

The election results were satisfactory but not wholly as projected.
The 24 mill proposal passed by a 175 vote margin, the 1.5 mill
proposal failed by a 887 vote margin. The reason for this breakdown
from the early estimates can be attributed to an idiosyncratic event.
A former school board member staged a virtually single-handed campaign
against the proposals through the ‘letters to the editor’ section of
the Sault Evening News. This individual had opposed increases in
teacher’s salaries during the last round of contract negotiations. As
a result, the MEA had Opposed her reelection to the board (to which
she attributed her defeat). When she discovered the use of MBA
personnel in the millage campaign, she launched a particularly bitter
series of written attacks against the proposals. The anomaly of a
former board member attacking her former colleagues attrached media
attention which undercut efforts to minimize turnout of ‘no’ voters.
This undoubtedly contributed to the curiously large aggregate turnout
in the election: 5376 votes or 442, a full nine percent increase over

the June results.

Reflection on the Hypotheses

This chapter of the dissertation has detailed the events of an
actual tax referendum with the purpose of demonstrating the relevance
of the formal structures and hypotheses developed in earlier chapters.
The narrative presented here used these hypotheses as a guide to
interpretation of the events. This presentation has arguably offered

positive instances Of the application of strategies and behaviors

135
which derive from.cr are analogous to those given in the hypotheses
and the experimental results. Over the ten year electoral history
the school board was argued to have developed several sophisticated
strategies in their proposal behavior which shared several theoretical
and practical features with strategies used by ‘sophisticated’
subjects in the experimental setting. The gradual development of this
type of sophistication tends to support the notion that sophistication
itself is, as hypothesized, a reflection of skill and experience which
may be learned over time.

With respect to the Piele and Hall hypothesis, the strategy employed
by the consultant/campaign manager reflects a faith in the truth of
the hypothesis, and the results of the election are certainly
consistent with what is expected under the hypothesis. The key to
success seemed to be the management of turnout which was predicated
upon the gathering and analysis of information concerning the
distribution of preferences within the electorate. Indeed, despite the
actual increase in aggregate turnout, effective management seems to
have actually shifted the location of the median voter to a higher
level of demand. While this result is consistent with the result
reported in the Rubinfeld11977] study, the analysis presented here
goes beyond Rubinfeld to attribute the outcome to particular behaviors
observed during the campaign.

It certainly can be argued that the narrative and discussion
presented in this chapter are somewhat of a self-fulfilling prophecy.
The evidence has been organized and interpreted in a way which
necessarily corresponds to the theory proposed. Nonetheless, there

were specific expectations that were capable of being falsified here.

136
In a Popperian sense, the fact that they were not falsified does not
lead to the conclusion that the hypotheses were necessarily correct.
Certainly, other interpretations could be offered and might provide a
better basis for analysis. Still, without the theory to guide the

analysis in the first place, nothing would be achieved.

EPILOGUE

In the introduction to this dissertation I referred to the project
as a ”study in craft". As such, I have labeled this closing section an
‘epilogue’ rather than a ‘conclusion’. The research presented and
discussed in the body of the dissertatiOn can be appreciated from
several prospects. On its most basic level, this is a dissertation
about tax referenda. On another level, the dissertation can be
understood as an effort to use and demonstrate the potential of a
particular methodology. That is, this could be viewed as a
dissertation about using experimental methods to form a bridge between
formal theory and interpretations of ‘real world’ phenomena.
Alternately, this dissertation might be viewed as an argument
supporting the importance of incorporating political understandings in
formal models of various policy environments -supporting the unique
contribution which political science can make to ‘policy science’.
While all of these interpretations (and others as well) have merit, it
is the last one which I shall focus on in this epilogue; and, so, I
come full circle to the issues discussed in the introduction.

The ‘uniqueness’ of the contribution that political science can make
to ‘policy science’ is established by reflecting upon the current
‘state of the art’. ‘Policy science’ is the study of the impact of
public policy as judged from.its design, selection and implementation.
While it is not the province of any single discipline, the dominance
of economics is manifest through the methods of analysis. The
literature is rife with studies employing cost/benefit analysis,

137

138
micro-economic theory and modeling, and econometric techniques.
Political scientists, as participants in ‘policy science’, dutifully
adopt these methods. Nonetheless, political approaches to ‘policy
science’ should be distinct from.economic approaches. The environment
in which public policy is made and implemented is inherently
political. The actors who participate in the public policy process are
motivated by incentives which must be defined in political terms in
order to be coherently understood. Given this, we might expect that a
‘political approach’ to policy science would be worthwhile.

In the introduction to this dissertation, I argued that the
distinction between political and economic approaches to policy
science was made in terms of ‘ends’ versus ‘means’. That is to say
that, a ‘political’ approach puts primary emphasis on the analysis of
the dynamics of conflict within the public policy process, while
economic approaches stress the location and evaluation of the outcomes
of the policy process. Nonetheless, concern for the outcomes of the
public policy process is not absent in political approaches, nor is
concern for the dynamics of conflict ignored in economic studies.
Indeed, the two approaches do not differ in terms of subject matter or
methodology. Thus, the argument made in the introduction is rather
complex and, on a superficial level, appears self-contradictory. In
light of these difficulties, the five chapters between this point and
the introduction can be interpreted as an illustration of the
argument.

The first chapter presented an overview of the literature concerning
tax referenda from.an economic point of view. True to form, the

emphasis of this body of research was given to two questions: ‘Is the

139
median preference dominant?’; and, ‘Is the median preference
efficient?’. The critique of these efforts wove its way through
several different attempts to answer these questions and closed the
chapter by suggesting that these authors had not only failed to
provide such answers, there is serious doubt as to whether these
questions can be meaningfully asked in the tax referendum context. The
next four chapters of this dissertation provided an alternative
approach to analysis of the same subject matter discussed in the first
chapter.

The problem was recast in political terms. The failures of the first
chapter were reinterpreted from a new perspective. For instance, the
Bowen model, which had inspired the largest body of studies discussed
in the first chapter, was examined in terms of the motives attributed
-perhaps inadvertently- to the governmental agenda setter of a tax
referendum. An alternative set of assumptions concerning such motives
was supplied by discussion of the Niskanen model and with them came a
new set of expectations. However, both the Bowen and Niskanen models
remain in the style of economic analyses. As such, their use in this
dissertation was modified to reflect a more ‘political’ style. This.
stylistic change is reflected in the emphasis placed upon 6
institutional variables such as serial bargaining, as well as a
discussion of the relevance of the personal capacities of actors to
control risk and develop sophisticated strategies. Finally, the
discussion turned to the relevance of turnout in determining the
outcome of a tax referendum, as well as the potential for actors to
control turnout as a strategic resource. With these factors included

in the analysis, the expectations for outcomes become far less

140
precise.

At this point, discussion shifted from.notions Of stable equilibria
to ranges of possible outcomes. Such a change is most indicative of
the ‘political’ approach. Indeed, as the precision of forecasting with
respect to the policy outcome declines, so does the ability of the
economist to evaluate its efficiency. Such a move stands as anathema
to the economic approach and those who practice it. Nonetheless, if
the modifications discussed in this dissertation are valid, the
apparent precision of previous models of the tax referendum setting is
no more than an artifact of an erroneous set of assumptions. Moreover,
by bringing these matters to the fore, the possibility for developing
better models is given in the grounding provided by a better
understanding of the dynamics of the environment under study. To these
ends, the third through fifth chapters of this dissertation sought to
demonstrate the validity of the arguments presented in the second
chapter.

I leave it to the reader to judge the relative success of this
effort. While the case study provided in the fifth chapter gives
illustration to the theoretical points made in earlier sections, it
falls far short of establishing the empirical basis for a broad
general theory of tax referenda. All that can be claimed for it is
that it appears to represent a more meaningful interpretation of the
events of a single case than can be provided in the theoretical models
discussed in the first chapter. We have not yet reached the point
where a general econometric model of tax referenda can be designed let

alone meaningfully interpreted.

A PPENDIX

111

 

ALTERNATIVE PLAYER 1 PLAYER 2 'PLAYER 3 PLAYER 4

11mm mmmm

2. 75' 4. 00 ‘ 3. 50 2. 00
10 5. 00 4. 50 4. 00 2. 50
20 11.00 6.00 4.50 3.00
30 9.00 7.50 5.00 3.25
40 5.50 8.00 5.50 3.75
50 3.50 13.00 6.00 ~ 4.00
60 3.25 10.00 7.00 6.00
70 3.00 8.00 11.00 8.25
80 2.75 5.50 14.00 12.00
90 2.50 4.00 15.50 15.00
100 2.25 2.50 16.00 17.00

MAJAJAJJAAW

 

iv

INSTRUCTIONS FOR SMALL GROUP EXPERIMENTS
COMMON INSTRUCTIONS FOR ALL THREE SETTINGS:

1. We would like to have you participate in a committee
process experiment. The purpose of the experiment is to help
us understand certain aspects of the generally complex ways
in which committees operate. The instructions are simple. If
you follow them carefully and make good decisions you might
earn a considerable amount of money. You will be paid in
cash.

2. All you have to do is to attend a committee meeting and
for this participation you will be paid. The purpose of the
meeting is to choose a number from the set of numbers from
zero through 110. Only one number will be chosen and the
payment you receive for participation depends gngi;glx_upon
which one it is. For example, on the enclosed compensation
graph, you can read up.from.the number 30 to the line

graph, and then ggg;_to the vertical axis to discover the
amount you will be paid if the number 30 is the committee’s
choice. The amount for other numbers can be discovered
similarly. The compensation tables may differ amoung

individuals . W

 

. You
should decide what choice you want the committee to make and
do whatever you wish within the confines of the rules to get

things to so your way. W422

. u. .. .3...
WWW Under no
circumstances may you mention anything quantitative about
your compensation. You are free, if you wish, to indicate
which ones you like best, etc., but you cannot mention
anything about the actual monetary amounts. Under no
circumstances may you mention anything about activities
which might involve you and other participants after the
experiment (i.e., no deals to split up afterwards and no
physical threats).

SPECIAL INSTRUCTIONS:

(OPEN AGENDA)

3. PROCEDURES. The following procedures must be followed.
Player 4 has been designated as a convener. The convener is
not a member of the committee and therefore cannot vote on
any of the proposals. The option chosen by the committee
will be the one proposed by some member of the committee and
ratified by a majority of the committee (two or more
people). Any failure of the committee to act automatically
results in a committee choice of option ‘0’.

The committee will have a discussion period lasting a
minimum of 5 minutes and a maximum of 15 minutes in which
the convener is not present. During this time, the

V

committee, operating under majority rule, should try to come
to some agreement as to their preferred option.

At the conclusion of the discussion period, the convener
will formally open the decision period for the committee.
Open discussion may continue; the convener is free to make
suggestions to the committee. At any time during the
decision phase, any committee member may make any formal
proposal (s)he likes. A committee member proposes options in
writing by completing and signing the forms provided. Any
proposal by a committee member automatically cancels any
previous, unratified proposals by that committee member. A
proposal is ratified if it is signed by a majority of the
committee; i.e., to be ratified, a proposal must be made by
a committee member and approved by at least one other
committee member. A proposal once ratified is final and all
participants will be paid accordingly.

(AGENDA CONTROL-MULTIPLE PROPOSAL)

3. PROCEDURES. The following parliamentary procedures must
be followed. Player 4 has been designated as a convener. The
convener is not a member of the committee and therefore
cannot vote on any of the proposals. But, the option chosen
by the committee will be the one proposed by the convener
and ratified by a majority of the committee (two or more
people). Any failure of the committee to act automatically
results in a committee choice of option ‘0’.

The committee...their preferred option.

At the conclusion of the discussion period, the convener
will address the committee. Committee members are free to
make suggestions to the convener. The convener may propose
any option (s)he wants, or (s)he can refuse to propose
options if (s)he so desires. (S)he proposes options in
writing by completing and signing the forms provided. Any
proposal by the convener automatically cancels any previous,
unratified proposals. A proposal is ratified if it is signed
by a majority of the committee; i.e., to be ratified, a
proposal must be signed by the convener and approved by at
least two committee members. A proposal once ratified is
final and all participants will be paid accordingly. A
proposal which is not ratified, and is then cancelled by
another proposal may be reproposed again and again.

4. TERMINATING THE MEETING. The meeting can be terminated in
any of three ways: "

i) If some option is prOposed and ratified, the meeting is
terminated and the committee members are paid on the basis
of the committee’s choice.

ii) If a motion to adjourn is approved by a majority of the
comittee, the meeting is terminated. Any committee member
is free to introduce such a motion at any time. The motion
must be submitted in writing and signed by a majority of the
committee. If the meeting adjourns without a choice, then
option ‘0’ is automatically designated as the choice and you
will be paid accordingly.

iii) The convener has the right to adjorn the meeting at any

vi

time. (S)he does so simply by submitting the anouncement in
writing to the experimenter. Again, if the meeting adjourns
without a choice, then option ‘0’ is automatically
designated as the choice and you will be paid accordingly.

(AGENDA CONTROL-SINGLE PROPOSAL)

3. PROCEDURES. The following parliamentary procedures must
be followed. Player 4 has been designated as a convener. The
convener is not a member of the committee and therefore
cannot vote on any proposal. But, the convener is the only
one who can introduce an alternative to be voted on against
the status quo. The status quo is option ‘0’. The committee
will vote on any alternative proposed by the convener
against the status quo.

The committee...their preferred option.

At the conclusion of the discussion period, the convener
will address the committee. Committee members are free to
make any suggestion to the convener. Discussion will be open
for as long as the convener wishes. Whenever (s)he chooses
to do so, however, the convener may officially propose an
option. The convener may propose any option (s)he wants.
(S)he proposes an option in writing by completing and
signing the form provided. The proposal is official and
discussion ends when the convener hands the proposal to the
experimenter. At that time, each of the three committee
members casts a secret ballot for either the status quo
alternative ‘0’ or for the alternative proposed by the
convener. The ballots will be counted by the experimenter.
If the convener’s proposal gets two or more votes, then it
wins. Otherwise, the status quo alternative wins. The
winning alternative is final and all participants will be
paid accordingly.

vii

QUESTIONNAIRE FOR LARGE GROUP TURNOUT EXPERIMENT

We would like you to answer a few questions about yourself.
The questions are similar to those which are used in
political research. Please try to answer them completely and
honestly. When you have answered all the questions return
this questionnaire to Dr. Miller.

1. NAME 2. STUDENT NUMBER
3. Generally speaking, do you usually think of yourself as a
Republican, a Democrat, an Independent, or what?
4. If you answered #3 as a Republican or a Democrat, would
you call yourself a strong supporter of the party, or a not
very strong supporter of the party?
(Please indicate whether you agree, disagree, or cannot
decide with each of the following eight statements.)
5. People like me don’t have any say about what the
government does.
6. VOting is the only way that people like me can have any
say about how the government runs things.
7. Sometimes politics and government seem so complicated
that a person like me can’t really understand what’s going
on.
8. I don’t think public officials care much what people like
me think.
9. It isn’t so important to vote when you know your party
dosen’t have any chance to win.
10. So many other people vote in the national elections that
it doesn’t matter much to me whether I vote or not.
11. If a person doesn’t care how an election comes out he
shouldn’t vote in it.
12. A good many local elections aren’t important enough to
bother with.
13. While in school, how far do you live from.campus?

(LIST OF DISTANCES)
14. How do you usually get to class?

(LIST OF MEANS OF TRANSPORTATION)
15. Besides PLSZOO, do you have any other classes on
fridays? ‘

(more than one;one;none)
16. During the past twelve months have you ever played the
state lottery?

(regularly;occasionally;never)
17. I am:(MALE;FEMALE) -

LIST 01“ REFERENCES

iix
REFERENCES

Aldrich, John H., and Charles F. Cnudde, 1975, "Probing the
bounds of conventional wisdom: a comparison of regression,
probit, and dicriminant analysis", American Journal of
Political Science, v.19,3.

LLLLLLLLLLLLLLLL . and Charles W. Ostrom, Jr., 1980,
”Regularities, verification, and systematization", American
Behavioral Scientist, v.23,6.

Arrow, Kenneth J., 1963, Social Choice and Individual
Values, John Wiley, New York, NY.

Barlow, Robin, 1970, "Efficiency aspects of local school
finance", Journal of Political Economy, v.78.

Barr, James L., and Otto A. Davis, 1966, "An elementary
political and economic theory of the expenditures of state
and local governments”, Southern Economic Journal, v.33.

Barzel, Yoram, 1973, "Private schools and public finance",
Journal of Political Economy, v.81

Bergstrmm, Theodore C., 1973, "A note on efficient
taxation", Journal of Political Economy, v.81.

LLLLLLLLLLLLLLLLLLLLL , and Robert Goodman, 1973, "Private
demand for public goods" , American Economic Review, v.63 ,3.

Black, Duncan, 1958, The Theory of Committees and Elections,
Cambridge University Press, London.

Borcherding, Thomas E., and R.T. Deacon, 1972, "The demand
for the services of non-federal governments”, American
Economic Review, v.62,5.

Bowen, Howard R., 1943, "The interpretation of voting in the
allocation of economic resources", Quarterly Journal of
Economics, v.58.

Buchanan, James M., 1968, The Demand and Supply of Public
Goods, Rand McNally, Chicago,IL.

LLLLLLLLLLLLLLLLLL , and Gordon Tullock, 1962, The Calculus
of Consent, University of Michigan Press, Ann Arbor, MI.

Campbell, Angus, et al., 1960, The American Voter, John
Wiley, New'York, NY.

LLLLLLLLLLLLLLLLLLLLLLL , 1966, Elections and the Political
Order, John Wiley, New York, NY.

ix

Coleman, James S., 1957, Community Conflict, The Free Press,
New York, NY.

Converse, Philip E., et al., 1961, "Stability and change in
1960: a reinstating election”, American Political Science
Review, v.55.

Courant, Paul, Edward Gramlich and Daniel Rubinfeld, 1979,
"The stimulative effect of intergovernmental grants", in
Mieszkowski and Oakland (eds), Fiscal Federalism and
Grands-inmAid, The Urban Institute, Washington, D.C.

Crosby, Otis A., 1963, "How to make bonds a winning issue",
Nation’s Schools, v.72.

Downs, Anthony, 1957, An Economic Theory of Democracy,
Harper and Row, New York, NY.

Eavey, Cheryl L., and Gary J. Miller, 1982, "Bureaucratic
agenda control: imposition or bargaining”, (mimeo) Michigan
State University, E. Lansing, MI.

Edelson, Noel M., 1973, "Efficiency aspects of local school
finance", Journal of Political Economy,v.8l.

Farquarson, Robin, 1969, Theory of Voting, Yale University
Press, New Haven, CT.

Fiorina, Morris P., 1978, "Legislative facilitation of
government growth: universalism.and reciprocity in majority
rule institutions" , Caltech Social Science WOrking Paper No.
228.

.................. , and Charles R. Plott, 1978, ”Committee
decision under majority rule: an experimental study",
American Political Science Review, v.72.

Fisher, Ronald, 1982, "Income and grant effects in local
expenditure", Journal of Urban Economics, v.12.

Fossett, James, 1983, "On confusing caution and greed: a
political explanation of the flypaper effect", prepared for
the annual meeting of the Midwest Political Science
Association, Chicago, IL.

Gerber, Lee 0., 1962, "School bond elections should be
publicized fully and fairly and displayed conspicuously",
Nation’s Schools, v.69.

Gibbard, Alan, 1973, "Manipulation of voting schemes: a
general result", Econometrics, v.41.

Gramlich, Edward, and Harvey Galper, 1973, "State and local
fiscal grant policy", Brookings Papers on Economic Activity No. 1

Hahn, Harlan, 1968, "Voting in Canadian communities: a
taxonomy of referendum issues", Canadian Journal of
Political Science, v.1.

Hanson, Carroll B., 1969, "How to pass a bond issue",
National Education Association Review, v.13.

Hanushek, Eric A., and John E. Jackson, 1977, Statistical
Methods for the Social Sciences, Academic Press, New York,

Hamilton, Howard D., and Sylvan R. Cohen, 1974, Policy
Making and Plebiscite: School Referenda, D.C. Heath,
Lexington, MA.

Harrington, John H., 1959, "Blueprint for a successful bond
or tax election”, Nation’s Schools, v.64.

Holcombe, Randall C., 1975, A Public Choice Analysis of
Millage Issue Elections for Financing Public Schools, PhD
dissertation, Virginia Polytechnic Institute and State
University. ' ‘

Isaac, R. Mark, and Charles R. Plott, 1977, ”Cooperative
game models of the influence of the closed rule in three
person, majority rule committees: theory and experiments",
Caltech Social Science WOrking Paper No. 181.

Inman, Robert P., 1978, "Testing political economy’s ‘as if’
proposition: is the median income voter really decisive?”,
Public Choice, v.33.

Ladha, Krishna, Thomas Homer, and Howard Rosenthal, 1982,
”If at first you don’t succeed: budgeting by a sequence of
referenda", Caltech Social Science WOrking Paper No. 449.

Lindahl, Erik, 1919, "Just taxation: 8 positive solution",
in Musgrave and Peacock (eds), Classics in the Theory of
Public Finance,1967, St. Martin’s, New York, NY.

MacRay, R.J., and C. Weaver, 1978, "Monopoly bureaus and
fiscal outcomes", in Tullock and Wagner (eds), Policy
Analysis and Deductive Reasoning, D.C. Heath, Lexington,
MA . '

McKelvey, Richard D., 1976, "Intransitivities in
multidimensional voting models and some implications for
agenda control", Journal of Economic Theory, v.12,3.

xii

Miller, Gary J., and Terry M. Moe, 1981, "Bureaucrats,
legislators, and the size of government", (mimeo), Michigan
State University, E. Lansing, MI. ‘

Neufeld, John, 1977, ”Taxrate referenda and the property
taxpayer’s revolt", National Tax Journal, v.30.

Niskanen, William A., Jr., 1971, Bureaucracy and
Representative Government, Aldine Atherton, Chicago, IL.

Oates, Wallace, 1979, "Lump-sum.intergovernmental grants
have price effects", in Mieszkowski and Oakland (eds),
Fiscal Federalism and Grants-ineAid, The Urban Institute,
Washington, D.C.

Piele, Philip R., and John S. Hall, 1973, Budgests, Bonds,
and Ballots, D.C. Heath, Lexington, MA.

Pindyck, Robert S., and Daniel L. Rubinfeld, 1981,
Econometric Models and Economic Forecasts, 2nd ed., McGraw
Hill, New York, NY. ‘

Plott, Charles R., 1967, "A notion of equilibrium.and its

possibility under majority rule", American Economic Review,
v.57.

LLLLLLLLLLLLLLLLL , and Michael E. Levine, 1975, "On using
the agenda to influence group decisions: theory, experiments
and application", Caltech Social Science WOrking Paper

NO. 66.

Pommerehne, W. W., 1976, "Budgetary redistrbution in
democracies: an empirical test of alternative hypotheses",
Discussion Paper No. 64, Konstanz.

AAAAAAAAAAAAAAAAA . 1978, ”Institutional approaches to public
expenditures: empirical evidence from.Swiss municipalities”,
Journal of Public Economics, v.9.

LLLLLLLLLLLLLLLLL , and B. S. Frey, 1976, "Two approaches to
estimating public expenditures", Public Finance Quarterly,
v.4.

Rae, Douglas W., 1971, The Political Consequences of
Electoral Laws, Yale University Press, New Haven, CT.

Romer, Thomas, and Howard Rosenthal, 1978, "Political
resource allocation, controlled agendas, and the status
quo", Public Choice, v.33,4.

11111111111111111111111111111111111 , 1979, "The elusive
median voter", Journal of Public Economics, v.12.

xiii

11111111111111111111111111111111111 , 1980, "An institutional
theory of the effects of intergovernmental grants", National
Tax Journal, v.33.

Rubinfeld, Daniel, 1977, "Voting in a local school election:
a micro-analysis", Review of Economics and Statistics,
v.59.

Samuelson, P. A., 1954, "The pure theory of public
expenditures”, Review of Economics and Statistics, v.36.

Satterthwaite, M. A., 1975, "Strategy-proofness and Arrow’s
conditions: existence and correspondence theorems for voting
procedures and social welfare functions", Journal of
Economic Theory, v.10.

Shepsle, Kenneth A., 1979, ”Institutional arrangements and
equilibrium.in multidimensional voting models”, American
journal of Political Science, v.23,1.

Slutsky, S., 1977, "A voting model for the allocation of
public goods: existence of an equilibrium", Journal of
Economic Theory, v.14.

Stone, Clarence N., 1965, "Local referendums: an alternative
to the alienated voter model", Public Opinion Quarterly,
v.29.

Weingast, Barry R., 1979, "A rational choice perspective on
congressional norms", American Journal of Political Science,
v.23,2.

Wicksell, Knut, 1896, ”A new principle of just taxation", in
Musgrave and Peacock (eds), Classics in the Theory of Public
Finance, 1967, St. Martin’s, New York, NY.

WOlfinger, Raymond E., and Steven J. Rosenstone, 1980, Who
Votes?, Yale University Press, New Haven, CT.

Wood, Robert C., 1958, Suburbia: It’s People and Their
Politics, Houghton Mifflin, Boston, MA.