Political Institutions and Policy Choices: Evidence from...

Journal of Economic Literature Vol. XLI (March 2003) pp. 7–73

Political Institutions and PolicyChoices: Evidence from

the United States

TIMOTHY BESLEY and ANNE CASE1

7

1. Introduction

IN RECENT YEARS, there has been increasedinterest in bringing economics and politics

together to understand policy. This resur-gence of political economy has generatednew questions and new models. Insightsfrom game theory, contract theory, andmechanism design theory now permeate ourunderstanding of the political sphere.2 Therecent literature has paid much more atten-tion to theory than empirics. This essay high-lights the progress that has been made inempirical political economy using variationsin institutional rules across the UnitedStates.

The essence of the literature is illustratedin the following example. A number of stateshave recently passed laws that tie voter reg-istration to motor vehicle registration, so-called “motor-voter” laws. The intention is topromote voting, particularly among disad-vantaged groups that have low rates ofturnout. Here, we might ask three questions.First, does the passage of such laws lead togreater turnout? Second, does this change in

turnout (if there is one) change the composi-tion of the legislature, for example, by in-creasing representation of Democrats?Third, does the increase in the extent of rep-resentation by Democrats (if there is one)have an influence on policy outcomes? Inprinciple, these questions can be addressedempirically using available data, and we dis-cuss the answers below. In thinking throughthese possible chains of influence (as docu-mented in the data), one is also drawn tothinking about the right theoretical contextin which to understand the policy process.

This paper studies a rather heterogeneousgroup of institutions. Thus, the term institu-tion should be understood in the ratherbroad sense suggested by Douglass North(1990) as “the humanly devised constraintsthat structure human interaction” (p. 3). Weuse the term institutional choice in prefer-ence to the narrower idea of a constitutionalrule, which suggests a formally encodedmandate. As observed by a number of com-mentators (such as Avinash Dixit 1996) itdoes not make an awful lot of sense to drawhard distinctions between terms like institu-tion, constitutional rule, and (even) policies.Moreover, in practice, those who study con-stitutional economics use an equally broadconception of the term constitution. For ex-ample, Geoffrey Brennan and Alan Hamlin(1998) characterize constitutional economicsas the study of the “basic rules under whichsocial orders may operate” (p. 401).

1 Besley: London School of Economics. Case:Princeton University. We are grateful to Jim Alt, MarkCrain, three referees, and John McMillan for helpfulcomments and advice. Erland Berg, Silvia Pezzini,Susanna Peralta, Marit Rehavi, and James Vere pro-vided excellent research assistance at various stages.

2 Torsten Persson and Guido Tabellini (2000) bringmany of these together in an excellent survey of thefield.

Article 1 2/21/03 4:17 PM Page 7

This survey will restrict itself to evidencebased on cross-state institutional differencesin the United States. This has costs and bene-fits. The main cost lies in the limited range ofinstitutional differences that can be consid-ered across U.S. states. While there are someinteresting and important differences avail-able, there are a host of common featureswhose implications cannot be studied in thisway. The United States cannot, for example,be used to test the difference between pro-portional and first-pass-the-post electoral sys-tems. However, the broadly common institu-tional and constitutional setting is also abenefit. There is real hope of isolating the truesource of differences, rather than attributingto a particular institution some effect that is, inreality, due to unobserved heterogeneity. Thiscontrasts with much work using cross-countrydata. A further benefit from the U.S. focus isthe availability of a wide range of existing stud-ies on all aspects of institutional differences.

In addition to reviewing the main contri-butions in the area, we also present new em-pirical results. By using a consistent set ofdata and methods, this will provide an addi-tional perspective on the issues in question.It will also help us to illustrate some of themethodological issues involved in makinguse of cross-state institutional variation tostudy policy outcomes.

Most of the work that is discussed below isby economists. There is also, however, a largebody of relevant work coming out of politicalscience departments. Within political sci-ence, there is far from universal acceptanceof the value of formal and quantitative rea-soning. The debate surrounding DonaldGreen and Ian Shapiro (1994) is indicative ofthis. Green and Shapiro criticize the unwill-ingness of formally inclined political scientiststo engage in empirical testing. This essay ispartly an effort to promote the work that hasbeen done. It is clear from the existing litera-ture, however, that the evolution of theoryand empirical evidence on understanding po-litical institutions has been unbalanced.Much of the empirical testing eschews inter-

pretation of results in terms of models, andhence fails to push forward the debates aboutmodeling. At the other end, a good deal oftheory seems to be lifting insights from gametheory, contract theory, and information eco-nomics without pushing testable implications.

The remainder of the paper is organized asfollows. The next section presents some back-ground discussion of the literature and theevolution of thinking in the field of politicaleconomy. Section 3 develops a framework forthe study of political competition and discussesits empirical implications for the study of howpolitical institutions affect policy outcomes.Section 4 reviews how institutional arrange-ments vary across U.S. states. Section 5 studiesinstitutions that directly affect the process ofpolitical representation, to see whether andhow they affect voter turnout and such legisla-tive outcomes as political ideology and theparty composition of the legislature and gover-nor. Section 6 then examines how political rep-resentation affects state policies, particularlythose of taxing and spending. Section 7 consid-ers institutional effects on policy outcomes, ei-ther through direct influence or through theirproximate effect on representation. Thesemiddle sections of our paper follow the planset out in figure 1. Section 8 turns to the ques-tion of how institutional rules are made andchanged. Section 9 concludes.

2. Background

Understanding how institutions changepolicy is an important intellectual end in it-self. It is also part of the kit bag that econo-mists need to improve the way in which theworld works. That a just polity involves thedesign of appropriate rules for policy forma-tion is discussed in James Buchanan (1967,1987), for example. A great deal of publiceconomics is about the choice of policies—the level of taxation, whether to mandatehealth insurance, and such. There are tworeasons, however, to think that this shouldbe augmented with studies that go back onestage prior and evaluate the process on the

8 Journal of Economic Literature, Vol. XLI (March 2003)


basis of which policy is made. First, institu-tional reforms are frequently on the agendaand we need to have a framework (and em-pirical knowledge) for judging them.Second, the policy advice and insights thateconomists offer are mediated through thepolitical system. It may be that once theworkings of the political system are under-stood, then we would change the policy ad-vice that we give. Policies that appeared sub-optimal may be desirable because of the wayin which they are operated in political equi-librium. But this raises the larger question ofwhether it is better to change the rules bywhich policies are formed than to advocatepolicy changes themselves. This, then, re-quires an understanding of the mappingfrom institutional rules to policy outcomes.

One rather grand view of policy making,suggested by Buchanan, is to think of therebeing two stages of analysis. At the firststage, a constitution is designed.3 This hastwo components. A procedural constitutionsets the terms by which decisions are made(electoral rules, term limits, the separationof powers, and such). A fiscal constitutionbuilds in constraints on the policies that canbe adopted within the framework of the pro-cedural constitution. This might include, forexample, limits on taxation or particularforms of public spending. After the constitu-tion is determined, policies are chosen.These are autonomous, however, and the

key role for the policy advisor at stage one isto anticipate the outcome at stage two.4

While useful as a benchmark, the distinc-tion between a rule and policy is quite narrow(the discussion in Dixit 1996, for example, ad-dresses this issue). A good example is the rulethat prohibited many U.S. states from levyingan income tax, but which was overturned bymost states during the twentieth century. Thiskind of fiscal rule then looks much closer towhat we would ordinarily call a policy than arule. In practice, there may be larger costs as-sociated with changing some aspects of thepolicy framework than others—the need forratification by two-thirds majorities is a goodexample. Thus, it is probably a little danger-ous to try to draw a hard-and-fast distinctionbetween the immutable constitution and thepliable policy arena.

The notion of designing an optimal consti-tution is tinged with hubris. In practice, theoptimizing approach to policy analysis canbe solved only under very stylized assump-tions about the economic environment andincentive problems. Moreover, the bewil-dering array of different policy issues need-ing to be solved makes the notion of specify-ing the optimal constitution a distant dream.On a more practical level, we might hope tounderstand the workings of particular insti-tutional changes. Much as in policy econom-ics, we can debate how particular interven-tions—such as minimum wages or publicly

Besley and Case: Political Institutions and Policy Choices 9

4 Timothy Besley and Stephen Coate (2001) developa very simple model that illustrates the issues thatmight arise in the design of a fiscal constitution.

3 Jean-Jacques Laffont (2000) provides a conceptionof the optimal constitution problem from a mechanismdesign perspective.

RepresentationSection 5 Section 6

Section 7

Institutions(Section 4)

Policies

Figure 1. Organization of the Text


funded broadcasting—affect the economy.The contributions surveyed here are more inpursuit of this agenda and make progress forprecisely this reason. While it is true thatthere are probably important interactionsbetween policy decisions—for example in-creasing the minimum wage may result inadjustments to welfare policy—we need tobegin where concrete progress can be made.

Most of the institutional changes analyzedhere are procedural, in Buchanan’s terminol-ogy, although there will be exceptions (bal-anced budget restrictions being a key exam-ple). Institutional rules are of two broad kinds:(i) electoral rules: restrictions on who canvote, whether proportional representation isused, and such, and (ii) decision-making rules:the use of line-item vetoes; whether certainagencies are independent. The distinction,however, is not a very precise one.

To understand the effect of an institutionalchange on policy outcomes requires anunderlying model of the policy process. Theimportance of providing theoretical founda-tions is twofold. First, the interpretation of aparticular effect is normally (consciously orunconsciously) tied to a model. Second, andmore importantly, is a concern about the po-tential generalizability of the findings. If thesole aim of a study is to answer a narrowlyposed question (did a particular historical in-stitutional change have some impact?), thenpresenting an underlying model may not benecessary. Most authors, however, would liketo claim a result of more wide-ranging signif-icance. This can most often be achieved byshowing that it illuminates some particulartheoretically validated relationship. Thestudies that are discussed below vary greatlyin the extent to which the authors spell outthe theory behind their results.

The political economy literature has de-veloped a plethora of models that can beemployed to address a wide range of ques-tions, and here is not the place to developany kind of survey of them. We will make afew comments, however, about the theoreti-cal literature that will be germane to what

follows. Persson and Tabellini (2000) pro-vide a comprehensive tour of the field.

A great deal of theoretical political econ-omy literature begins from the notion thatpolicies should line up with the preferencesof the median voter. Indeed, for a long periodthis was almost a caricature of economists’forays into political economy. The motivationcomes from two key contributions: (i) the ob-servation (most often attributed to DuncanBlack 1987) that restrictions on preferences(most often single-peakedness) imply that themedian voter’s preferred point is a Condorcetwinner,5 and (ii) the observation due toAnthony Downs (1957) that two parties whocare only about winning would pick out theCondorcet winner if they could commit topolicies during an election campaign andknow the electorate’s preferences. In spite ofthe model’s centrality in the literature, bothcomponents provide a deeply flawed basis forthinking through the implications of politicalcompetition for policy determination. First,policy environments with multiple issuesrarely have a Condorcet winner. Second, theassumptions about commitment and motiva-tion in the Downsian paradigm are unreason-able and outcomes are highly unrobust to de-viations from them. If parties care aboutpolicy, this may have significant implicationsfor the credibility of their ex ante policy pro-nouncements. For example, Alberto Alesina(1988) shows that whether complete conver-gence of parties’ policies is possible, even inan infinitely repeated game, will depend onsuch parameters as the parties’ discount rates,polarization of preferences, and the relativepopularity of the parties.

Progress in theoretical political economyhas had to wait for modeling approaches thatleave the straightjacket imposed by theDownsian paradigm behind them. Politicaleconomy thrived in some quarters by throw-ing out explicit electoral modeling com-pletely as, for example, in Sam Peltzman


5 A Condorcet winner is a policy that would beat allothers in binary comparisons based on majority rule.

Article 1 2/21/03 4:17 PM Page 10

(1976) and Gary Becker (1985). However,their reduced-form approaches do not pro-vide a way for thinking about the policy im-plications of institutional changes. It has alsobecome apparent that, if we are to have a re-alistic hope of generating predictions thatmight apply in reality, then modeling needsan approach that allows the simultaneousfunctioning of elections and interest groups.Progress on this front has been slow butGene Grossman and Elhanan Helpman(1996) and Persson and Tabellini (2000, sec-tion 7.5) point the way ahead.

Literatures have also sprung up that de-emphasize the spatial choice component ofelections, instead concentrating on the role ofelections in curbing the opportunistic behav-ior of politicians (as in Robert Barro 1973 andJohn Ferejohn 1986). Below, we will arguethat some of the cross-state literature fromthe United States speaks to the relevance ofthis approach for explaining the data.

Among political scientists, a large literaturedeveloped that focused on approaches tosolving the problems of multi-issue decisionmaking at the level of legislatures, but rarelyhas this been mapped back to overall elec-toral consequences. A key insight emergesfrom this that is useful in many other con-texts. Part of the difficulty in the Downsianparadigm is the fact that there is little institu-tional restriction on policy proposals. It isvery difficult to get a stable point when anypolicy can be proposed by any political actorat any time. Kenneth Shepsle and BarryWeingast (1981) discuss how restrictions ofthe structure of proposal power within alegislature can be used to generate a stablepoint in a multidimensional policy space.

Restricting proposal power can be valuablein other contexts. It is at the heart of the cele-brated model of agenda setting due toThomas Romer and Howard Rosenthal(1978). For example, Besley and Coate (1997)restrict policies to those that are optimal forsome citizen. John Roemer (2001) restrictsproposal power by modeling within-partyconflict. Such restrictions improve the odds of

developing a model that predicts an equilib-rium outcome in a particular policy context,providing a basis for empirical analysis.

The notion that the structure of the powerto propose policy is important is really partof a more generally valuable insight, that thestructure of political institutions affects po-litical outcomes. This is the core idea behindthe empirical literature we discuss below. Itshould also be clear that this is likely to reston departures from the median voter logicthat has dominated so much of the politicaleconomy literature. If policies we see in theworld really are responsive to the medianvoter’s preferred point in any kind of generalsense, then there is no good reason to thinkthat the institutional structure of decisionmaking would matter much at all.

One particular theoretical issue that ishighlighted by many empirical studies is theneed for a tractable framework to study theallocation of multiple issues. Most often ana-lysts are interested in studying the impact ofinstitutional rules that may affect only somepart of the policy space or, at least, would af-fect different parts in varying ways. Thus,consider the effect of increased voter regis-tration among minority voters. We wouldtypically want to consider the impact of thison general policies and those that particu-larly affect the groups in question. However,to think about this requires a model wherethe impacts on both group-specific and gen-eral policies are considered. It would notmake much sense, for example, to consider aworld where redistribution toward minori-ties (in the form, say, of affirmative action)were the only policy being determined inpolitical equilibrium. Political scientists arewell aware that changes in the salience of is-sues is central to understanding the evolu-tion of policy (see, for example, EdwardCarmines 1994).6 However, economists donot have well-developed frameworks for


6 Besley and Coate (2000a,b) also provide a way ofthinking about salience in a multiple issue frameworkand how institutions can change the salience of issues.

Article 1 2/21/03 4:17 PM Page 11

handling this. Some play has been obtainedfrom the multidimensional models of AssarLindbeck and Jorgen Weibull (1993), andDixit and John Londregan (1998). They en-visage two ideologically opposed parties whocourt groups of voters by promising trans-fers. While this yields useful insights, it can-not address the question of when the ideo-logical dimension may shift—as arguably hasbeen the case in a number of time periods inthe United States. It also seems odd to in-voke such strong assumptions about com-mitment to transfers without specifying themechanism by which credibility is achieved.

The post-Downsian literature in politicaleconomy suggests the following simplethree-way theoretical classification of theimportance of institutions in policy forma-tion. First, institutions may affect policypreferences directly, and their expression atthe ballot box. The simplest possibility istheir effect on who votes. However, policypreferences may change for a given votingpopulation. For example, a voter may bemore willing to tolerate a Democrat boundby a tax and expenditure limitation. Second,in a multi-issue world, institutions can affectpolicy priorities. Thus, incentives to targettransfer programs to particular groups maychange if that group is registered to vote. Asargued in Besley and Coate (2000b), allow-ing voters to place legislation directly on theballot via citizens’ initiatives can unbundle aparticular policy issue and change itssalience to voters. Third, institutions canchange the ability of the policy process andparticular politicians to commit. For exam-ple, a term-limit may reduce the credibilityof election promises since an incumbentmay not run again for office. These three ef-fects can go a good way towards interpretingmany of the findings below.

3. An Organizing Framework

This section serves two main purposes.We present a model of the different compo-nents of the policy process, which will serve

as a unifying framework in the subsequentdiscussion of the literature. In addition, wediscuss the kinds of empirical tests that fol-low naturally from the framework.

3.1 Components of the Policy Process

The policy process constitutes an embar-rassment of riches for applied research.We begin by laying out the components ofthat process that might be studied empiri-cally. While in practice most researcherstake a reduced-form approach, looking atthe components helps to lay bare the as-sumptions that implicitly go into the inter-pretation of the effects of institutions onoutcomes.

For the purposes of our theoretical dis-cussion, we partition the institutions thatvary across the U.S. states into two vectorsdenoted by I1st and I2st, with Ist denoting thefull institutional vector. The vector I1st con-tains those institutions that are typicallythought to affect the policy process for agiven structure of political representation.For the purposes of this essay, they are:

Policy Making Institutions (I1st). Restric-tions on the governor’s and legislators’ free-doms, including tax and expenditure limita-tions; supermajority requirements for taxincreases; the governor’s possession of aline-item veto; rules for appointing regula-tors and judges; rules governing whether astate permits direct democracy, such as citi-zens’ initiatives; and rules on whether gover-nors face term limits.

The vector I2st comprises that set of insti-tutions that affect the way in which electionsare conducted. In this essay, these are:

Electoral Institutions ( I2st). Rules affect-ing who can run for office and who can vote,including those affecting the costs of regis-tering to vote (such as poll taxes and literacytests); those regulating campaign contribu-tions in state elections; and those governingthe conduct of primary elections.

Both lists contain rules on whether a statepermits direct democracy, such as citizens’initiatives, and rules on whether a governor


Article 1 2/21/03 4:17 PM Page 12

faces a term limit. From a theoretical pointof view, a case could be made for them hav-ing an influence at both of the stages thatwe discuss below. More generally, the dis-tinction between policy making and elec-toral institutions need not be hard and fast.We use it primarily as a device for organiz-ing the literature. In principle, any of the in-stitutions we analyze could belong to bothI1st and I2st.

3.1.1 Preferences

Virtually all approaches to politicalprocesses take the preferences of votersand parties as given, and we shall do sohere. Suppose then that preferences ofvoters are defined over a policy spacexst ∈ t . This is a potentially wide-rangingdescription of all policies that can be con-trolled or influenced by state governmentsin the United States. Suppose also that het-erogeneity across the voting populationcan be parameterized for voter i by θi ∈ Θ .In practical terms, the vector θi would in-clude the ideological stance of each voterand their views on other key issues such asgun control or abortion. Preferences canbe written as:

v(xst, yst, θi)

where yst is a vector of state demographicand economic characteristics that affect pol-icy preferences.7 Let Θst parameterize thedistribution of voter tastes in the populationin state s at date t. There is plenty of evi-dence of substantial heterogeneity in distri-butions of preferences across states—see,for example, Gerald Wright, Robert Erik-son, and John McIver (1987).

Also relevant are the policy preferences ofthe Democratic and Republican party ineach state. We denote these by:

V (xst, χjst, yst) for j ∈ {D ,R} ,

where χjst parameterizes the distribution ofparty members and/or influential partyelites.8 It is widely documented that χDstand χRst are distinct. There may also besubstantial differences between states.Received wisdom among political scientistsalso suggests that elites are important in pol-itics and can hold views that are distinctfrom average opinion among voters. Thissuggests that χjst differs from Θst .

Data exist to characterize some elementsof party and voter preferences. Here, we usethe cross-state series created by WilliamBerry et al. (1998). However, there is poten-tially a richer set of data available throughpolling surveys. Such data form the basis ofthe work of Wright, Erikson, and McIver(1987) and Erikson, Wright, and McIver(1989).

3.1.2 The Post-Election Policy Process

To describe the post-election policyprocess, let �st be a variable that character-izes the political outcomes in state s at timet. This, for example, would include the com-position of the legislature and any relevantinformation about the governor. Our legisla-tive outcome function is:

(1)xst = G(�st,XDst,XRst, I1st, yst)

where Xjst is the “platform” of party j in states at time t9 and I1st is the policy making insti-tutional variables discussed above including,


8 These preferences could be derived from voterpreferences by supposing a subset of voters belongs tothe parties. If Mj denotes the set of members of partyj, then

V (x, χjst, yst) =∑

θj∈Θ,j∈ j

v (x, yst, θ)π(θ;χjst) ; j ∈ {D,R} .where π(θ;χjst) is the density of taste characteristicsin party j in state s at time t.

9 Whether (XDst,XRst) actually constrains ex post pol-icy making is an issue of some debate. In the simpleDownsian model, it completely determines policy expost.

7 These are reduced form preferences based on amodel of how policy affects outcomes that voters careabout.

A

M

Article 1 2/21/03 4:17 PM Page 13

for example, whether the governor has a line-item veto or whether there is a supermajorityrequired for tax increases. The function G(·)is intended to capture, in reduced-form, apotentially complicated policy process suchas a legislative bargaining model or a modelof the separation of powers between theexecutive and the legislature.

In principle, the mapping from institu-tions into policy outcomes induced by (1) canbe studied empirically. To do this, considerthe following empirical model for the kthpolicy in state s at time t of the form:

xkst = αks + βkt + ωkI1st + γkyst+ψk�st+ dkXDst + rkXRst+ηkst, (2)

where αks is a state indicator variable and βktis a year indicator. The focus is on how I1staffects the outcome of interest. For ordinaryleast squares to yield an unbiased estimateof ωk, it is clear that all relevant elements of�st need to be included, as they are likely tobe correlated with I1st and that XDst, XRstmust either be fully observed or be uncorre-lated with I1st. In practice, researchers can-not measure the commitments made by par-ties before elections. Moreover, there arereasons for believing that such commit-ments could be influenced by institutions.For example, the Democrats may adopt aless ambitious tax-and-spend platform ifthey know that such policies cannot be im-plemented in a situation where a super-majority is required for tax increases.

While (2) makes an effort to model thepolicy process (1), many researchers con-sciously omit �st from this equation to esti-mate

xkst = αks + βkt + ωkI1st + γkyst + ηkst

which gives an unbiased estimate of the di-rect effect of I1st on policy outcomes ex postonly if I1st have no impact on �st . Otherwisethe estimate of ωk must be interpreted asthe reduced-form impact of I1st, telling usboth about how policy is affected ex post andhow the electoral outcome is affected by I1st.The latter could happen, for example, if

passing a supermajority requirement for taxincreases made voters more inclined to votefor a Democrat.

3.1.3 Elections

Electoral outcomes are also modeled inreduced-form. Let

(3)P (�;XDst,XRst, cDst, cRst, yst, I2st,Hst)

denote the probability that a particular polit-ical outcome is � , when the platforms of theparties are (XDst, XRst), the candidates’ char-acteristics are (cDst, cRst), the history of policyis Hst and the institutions thought to affectthe electoral process are I2st. As outlinedabove, these institutions include the designand structure of districts, or whether gover-nors face term limits. Relevant candidatecharacteristics may include policy prefer-ences, such as whether the candidate is pro-choice, and may also include elements of thecandidate’s past reputation.

The role of the variable Hst is potentiallyquite important and surfaces, in particular, inmodels of political agency relationships as in-troduced by Barro (1973) and Ferejohn(1986). This variable could also be expandedto include policies in other states. This wouldbe relevant if yardstick comparisons aremade by voters, as in Besley and Case(1995a).

In principle, the mapping described by (3)can be studied empirically. Below, we discusssome studies that have done so. Good exam-ples are provided by those studies that modelthe probability that an incumbent governor isreelected. In this case, � is a discrete variableequal to one if the incumbent is reelected.Let wst be the incumbent’s advantage overthe challenger, with

wst = as + bt + ζI2st + ιyst+ µHst + ρst (4)

where as is a state fixed effect and bt is a yeareffect. Then we could suppose that

(5)�st ={= 10

ifwst ≥ 0otherwise.


if

Article 1 2/21/03 4:17 PM Page 14

This kind of model can help to pinpoint poli-cies for which the governor is held to ac-count, which are those elements of Hst thatinfluence reelection. It can also be used tosee whether reelection rates are dependenton the institutions within a state.

The outcome variable in an election neednot be a discrete variable. For example, thevariable wst could be the share of the votesobtained by one of the parties. In that case,we could estimate the empirical model (4)directly.

3.1.4 Party Strategies

The mapping from policy pronounce-ments and candidate choices into electoralsuccess shapes the parties’ strategies.These can be described both in terms ofsetting party platforms, as in the classicalDownsian approach (Downs 1957), or interms of candidate choices, as in the citi-zen-candidate approach of Martin Osborneand Al Slivinsky (1996) and Besley andCoate (1997). In reduced-form, let theparty preferences induced by (3) be de-noted:10

W (XDst,XRst, cDst, cRst, yst,Ist, χjst,Hst) for j ∈ {D,R} . (6)

The strategic problem facing parties at elec-tion times is to select platforms and candi-dates to maximize these payoffs.11 For thepurposes of taking these relationships to thedata, it would typically be assumed that a

Nash equilibrium exists and is unique.12

The outcomes are now party platforms andcandidate lists.

In principle, this part of the process isamenable to empirical testing when suffi-ciently detailed data on candidate composi-tion and party platforms are available. Thereare clearly interesting issues here, such asthe effect of primary rules on candidatecomposition, and the effect of campaigncontribution restrictions on candidate selec-tion and party policy. To date, however, theliterature has not studied these issues.

3.1.5 Political Outcomes

Given a set of party strategies, the electoralprocess (3) gives rise to a particular realiza-tion of �st . Key measurable components ofthis include the strength of Democratic partyrepresentation in the upper and lowerhouses, and the fraction of women elected.This could be modeled empirically for the kthpolitical outcome, as follows:

�kst = ζks + ξkt + λkIst + φkyst + νkst, (7)

where ζks is a state indicator, ξkt is a yearindicator. The exact choice of the institu-tional vector to be included can be deter-mined pragmatically. As we observed above,this may include institutions that are thoughtto primarily have an impact on ex post policymaking (I1st) as well as those that affect theelection process (I2st).

The reduced-form relationship may, by it-self, be interesting to analyze. It is clear,however, that interpreting the sources of in-fluence in the reduced-form relationship (7)could be quite difficult. For example, a linkobserved between the fraction of women inthe legislature and the rules that governcampaign finance could be working througha change in the candidate composition orthrough a change in the probability thatwomen candidates are elected.


10 These can be thought of as a function of

P (�;XDst,XRst, cDst, cRst, yst, I2st,Hst)andV (G(�,XDst,XRst, I1st, yst), χjst, yst)

over all legislative variables � in state s at time t.11 It is useful to note that this approach encompasses

within it the standard Downsian approach to policycompetition where the choice of candidates is irrele-vant—policy is determined ex ante with full commit-ment. It also includes a Lindbeck and Weibull (1993)type of model. It also encompasses the approach sug-gested in Besley and Coate (1997), (2000a,b) wherecompetition takes place by proposing candidates withdifferent characteristics.

12 It is well known that this is not a trivial matter.Here, it would require sufficient smoothness and con-cavity assumptions on the payoff function and the func-tion that maps policies into number of seats.

Article 1 2/21/03 4:17 PM Page 15

In (3), political outcomes were allowed todepend on political and policy history. In theempirical work to date, however, estimationof (7) has rarely included history variables.In principle, this could be done by estimat-ing

�kst = ζks + ξkt + λkIst + φkyst+ κkHst + νkst. (8)

In practice, Hst can be represented bylagged policy and political control. Issuesarise if these are not strictly exogenous, if forexample there were long lasting shocks topolicy preferences. Short panels also presentproblems, if Hst includes predetermined butnot strictly exogenous variables.

A wide variety of legislative outcomemeasures have been discussed in the exist-ing literature. Part of our task is to look at aconsistent set of outcomes with a commonmethod, and we have chosen to study thefollowing outcomes in this essay:

Legislative outcomes: Characteristics ofstate office holders, including the fractionsof Democrats in the state house and senate;party affiliation of the governor; the fractionof women representatives in the upper andlower houses; the ideology of the state gov-ernment; voter turnout; and the degree ofparty competition.

3.1.6 Policy Outcomes

The model that we have spelled out givesa sense of the rich variety of routes by whichpolicy can be affected by institutions.However, it isolates two main routes: di-rectly, through the ex-post policy process(1), and indirectly, through the ex-ante elec-toral process (3). As we have already dis-cussed, isolating the latter empirically maynot be easy. That the party preferences in-duced by the electoral and policy process,(6), depend on both sets of institutions couldimply that party strategies respond to insti-tutional change. This makes identification ofthe source of influence difficult.

For this reason, a fully reduced-form ap-proach is often chosen, in which no effort is

made to isolate the chains of influence. Inthis case, the model is:

(9)xkst = αks + βkt + ωkIst + γkyst + ηkst

where Ist contains the full institution vector.In fact, most studies that we discuss aremuch more piecemeal in their approach andtend to include only a small set of institu-tional variables in the analysis, typicallythose in I2st.

Again, history is generally overlookedin the estimation of policy equations.This need not be the case, and one couldestimate

xkst = αks + βkt + ωkIst + γkyst+ τkHst + ηkst. (10)

The variable Hst could again include laggedpolicy and political controls, raising similareconometric issues to those that arise inestimating (8).

There is a wide variety of policies that canbe studied, as the literature we discuss be-low makes clear. For our empirical work, wefocus on the following narrower set of policyoutcomes to comprise the xst:

Policy outcomes: Total state income, salesand corporate taxes per capita; state expen-ditures per capita, both in total and disag-gregated to focus on redistributive expendi-tures (such as family assistance and workerscompensation).

3.2 Institutional Change

Institutions change over time. This ismore easily dealt with theoretically than inempirical applications. One possibility is thatelectoral and decision making institutionsare chosen strategically to affect future elec-tion outcomes. This draws a parallel be-tween the literature on institutional changeand strategic policy making as developed,for example, in Tabellini and Alesina (1990)and Persson and Lars Svensson (1989). Insuch models, an incumbent chooses to con-strain future policy makers by changing thelevel of public debt that they inherit. There


Article 1 2/21/03 4:17 PM Page 16

are cases where this Machiavellian theory ofinstitutional change would seem relevant (atleast anecdotally). A good example is thecase of decentralization. In the UnitedKingdom, Tony Blair’s increased decentral-ization will result in a counterweight to thepower of the Conservative party. It may alsomake it less likely that the Conservativeparty will win office in the future. Repub-lican efforts to decentralize welfare policy inthe United States could also be viewedstrategically as an effort to encourage a raceto the bottom with a resultant reduction inwelfare spending.

At the core of any strategic theory of pol-icy making are preferences defined over in-stitutions. In terms of the theoretical ap-proach discussed here, consider

W (yst, Ist, χjst,Hst) =

W (X∗Dst,X∗Rst, c

∗Dst, c

∗Rst, yst, Ist, χjst,Hst)

for j ∈ {D,R}

where the * denotes that we are consideringthe equilibrium values of platforms and can-didate choices which themselves dependupon institutions and other exogenous vari-ables.

Suppose that a party is in office and, by in-curring some costs, could change the institu-tions that affect future payoffs. Then itschoice of optimal institution will be

I∗jst+ arg max = W (yst, Ist, χjst,Hst) .Ist

It is clear that there could be differences inthe different parties’ support for particularinstitutional changes, just as there are differ-ences in their policy preferences. In practice, there are many factors that cre-ate inertia in the reform of institutions.Some of the institutions that we consider re-quire elaborate processes of institutionalchange, including judicial review and consti-tutional amendments. A good example con-cerns the institutions that affect campaign fi-nance in the United States. These have beenaffected by federal legislation and by theway in which the Federal Supreme Court

has interpreted their boundaries vis a vis theright to free speech.

From the standpoint of empirical model-ing, institutions can be modeled in the sameway as policy and legislative outcomes.Consider the following equation:

Iest = αes + βet + γeHst

+ σewst + ωest. (11)

History Hst may well now include the his-tory of Ist and wst includes (potentially time-varying) state level variables thought to ex-plain the specific institutional change inquestion.

Problems arise in estimating the impact ofinstitutions on policy or political outcomeswhen the determinants of institutionalchange are correlated with the error terms inany of equations (2), (7) or (9) above. This ismore likely to be of concern in cross-sectionalstudies than in panel data studies. It may beplausible in many instances to think that thevariables wst are not time varying and hencecould be absorbed in state fixed effects.Good examples are long-run variations in po-litical culture across states. Authors of cross-sectional studies or panel data studies thatomit fixed effects have to assume that thereare no time-invariant omitted regressors thatdrive institutional choice.

3.3 Summary and Outline

To summarize, there are five main theo-retical relationships suggested by the theory:

• The effect of policy-making institutionsI1st on the ex post policy process as repre-sented by equation (2).

• The effect of institutions I2st on the elec-toral process as represented by equation(4).

• The effect of institutions (mainly electoralinstitutions) I2st on political outcomes asrepresented by equation (7).

• The effect of institutions, Ist, on policy asrepresented by equation (9).

• The process determining institutionalchange as represented by equation (11).


Article 1 2/21/03 4:17 PM Page 17

Discussion of the theoretical componentsunderscores the point that the institutionsthat affect the policy process through (1)and those that affect the electoral process(3) may not be easy to separate, once thestrategies employed in the political processare recognized. This has implications for theinterpretation of the results that we discussbelow.

The remainder of the paper is organizedaround looking at specific examples thatstudy one or another of these relationships.In section 5, we study legislative outcomesas a function of institutions along the lines of(7). Section 6 examines (2), with a particularfocus on how �st enters this relationship.Section 7 looks at a variety of empirical mod-els representing (2) and (9). It also considersthe process of electoral accountability.Section 8 discusses studies based on equa-tion (11).

One of the contributions of this paper isto estimate relationships discussed in theliterature. To do so, we have assembled across-state panel data set for the 48 conti-nental U.S. states from 1950–99. Whilenot all variables are available for the en-tire study period, many are. Of the studiesthat we review, many use shorter time pe-riods—often for the obvious reason thatthey were written some time ago. A vari-ety of empirical methods, however, is alsoat large in the literature. Where possible,we identify the relationships that we seekfrom a baseline model that includes stateand year fixed effects.13 We believe thatthis provides credible estimates for manyof the relationships in question.

The above specification assumes thatthere is time series and cross-sectionalvariation in the institutional variables(Ist). In many cases discussed below, re-searchers are interested in studying theimpact of institutions that do not vary

over time. Then, the above model is notestimable. Such studies usually proceedeither in cross-section or by omitting thefixed effects, hoping that xst will capturethe relevant heterogeneity over the cross-sectional units. For this to be a satisfac-tory way of estimating the impact of insti-tutions on outcomes, we need to assume,in the usual way, that ηst is uncorrelatedwith the institution in question. This isoften a very strong assumption.

An alternative way to get mileage out offixed institutions is to derive.(preferablyfrom some underlying theoretical structure)their implications for the response of an out-come to a time-varying regressor. For exam-ple, Besley and Coate (2000a) consider theeffect of electing versus appointing regula-tors on the relationship between cost andprice. If such implications exist, one can ex-ploit the panel nature of the data evenwhere there is only cross-sectional variationin institutions.

The fixed effects, in particular, are cru-cial to dealing with long-lasting unobserv-able differences between states that affectboth institutional rules and fiscal out-comes. It will not surprise the reader thatmany of the (unreported) specificationsthat do not include state fixed effectsyield dramatically different results. Whenfixed effect estimation is not possible—when, for example, the policy in questionvaries by state, but not over time—we es-timate robust standard errors, allowingfor an unspecified form of correlation be-tween observations from the same stateover time (although, below, we discusscases where robust standard errors maynot be appropriate; see section 7.1). Webelieve it important that a commonmethod of estimation be used across thespecifications that we report, to allow theresults to be read as all being cut fromthe same piece of cloth. Differences inmethods of estimation are an importantpotential source of discrepancy betweenresults in the literature.


13 Results presented below are largely robust to esti-mation using state and year effects and robust standarderrors.

Article 1 2/21/03 4:17 PM Page 18

4. Institutions, Politics and Policy in the States

Our study uses data from the 48 continen-tal states for the post-war period. The back-ground against which institutional featureswill be highlighted is the growing impor-tance of state governments over this period,especially in taxing and spending. In 1950,state governments took on average 3 percentof state income in taxes, a percentage thatroughly doubled in fifty years, rising to 5.7percent by 1999. U.S. states had tax rev-enues of $161 per capita on average in 1950,which climbed to $833 per capita in 1999(all measured in real 1982 dollars).14 Thesame story holds true on the expenditureside. State spending rose from an average of6.9 percent of state income in 1950 to 12.4percent in 1999. This constitutes an increasefrom $370 to $1764 per capita in real terms.No single factor led to this dramatic change,just as no convincing mono-causal force hasbeen identified behind Wagner’s law, in thevast literature on this.15 Table 1 presentsdata from the decennial census years, unlessotherwise noted, to provide some sense ofthe pattern in the data over time. Thechange in magnitude of state governmentcan be seen in the fourth panel, which pre-sents average state tax revenues and statespending per capita. In the empirical workthat follows, we analyze annual data from1950 to 1999 where possible. Some of ourdata series begin later. The Center for theAmerican Woman in Politics, for example,publishes data on women legislators onlyfrom 1975 to the present.

Aggregate tax revenues and expendituresin table 1 provide only a crude look at thechanges observed. Within taxes, the moststriking finding is the growth in income andsales taxes. In 1950, seventeen states raisedno income taxes. By 1999, all but six of the

continental states had an income tax. Thepicture is similar for the corporate tax—eighteen states were without one in 1950,true of only three or four states in 1999 (de-pending on how one characterizes the “fran-chise tax” on earned surplus in Texas).Substantial changes also occurred over thisperiod in the distribution of governmentspending. Family assistance per capitatripled between 1960 and 1980 (fourthpanel, table 1). Since that time, it has fallenin real terms, particularly in the second halfof the 1990s. In contrast, state spending onworkers compensation has increased mono-tonically since 1950.

Over this period, there has been a reason-able amount of institutional change. All fiftyU.S. states have broadly similar constitutionswith a bicameral system (Nebraska is an ex-ception here) and an elected governor.However, there is significant variation inelectoral processes between states, with dif-ferences in the way states organize their leg-islative districts, and in the way registeredvoters go about voting for candidates. Statesalso vary in their campaign-finance laws, andvoter registration and party primary rules.Some of these differences are highlighted inthe first panel of table 1. In 1950, sevenSouthern states (roughly 15 percent of statesin our sample) had poll taxes, which re-stricted voting among the poor, and roughly14 percent of state residents were affectedby a literacy test, a device that severely lim-ited the ability of less-educated people orthose whose command of English was poor,to vote. These voting restrictions were elim-inated in the 1960s, but voter registrationcontinued to be restricted by regulations onplace of registration, the timing of registra-tion relative to the next election, and thecontinuation of a place on the voting rolls forthose who missed elections. With the pas-sage of the National Voter Registration Act(NVRA) of 1993 (the “Motor-Voter Act”),voter registration through vehicle registra-tion is currently becoming a reality in all buta handful of states. States that were flexible


14 Unless otherwise stated, dollar values in the em-pirical work that follows will be given in 1982 dollars.

15 Cheryl Holsey and Thomas Borcherding (1997)review the issues with a U.S. focus.

Article 1 2/21/03 4:17 PM Page 19

in allowing registration on polling day or thatrequired no registration (MN, ND, WI, WY,NH, ID) were exempt from the NVRA. Thefraction of states in compliance with NVRA,and those with flexible registration, has in-

creased over time, as can be seen in the firstpanel of table 1.

There has been much variation over timeand between states in the types of primariesparties have run, with the fraction of states


TABLE 1INSTITUTIONS, LEGISLATIVE OUTCOMES AND POLICY OUTCOMES,

U.S. STATES 1950 TO 1999

1950 1960 1970 1980 1990

Institutions: electoral rules Percent of states covered by the following institutions

Poll taxes 15.2 10.9 0 0 0% state population affected by a literacy test 13.5 13.5 1.8 0 0Voter registration through vehicle registration 0 0 0 4.2 22.9Voter registration on polling day or no registration 0 2.1 2.1 6.3 6.3Open primaries 29.2 14.6 14.6 20.8 27.1Restrictions on corporate campaign contributions 62.8 76.2 71.4 65.1 64.6Gubernatorial term limits 40.8 43.8 50.0 54.2 58.3Citizens initiatives na 39.6 41.7 45.8 45.8

Institutions: decision-making rules

Tax and expenditure limitations 0 0 0 2.1 22.9Super-majority requirements na 2.1 6.3 14.6 14.6Gubernatorial line-item veto 81.3 81.3 85.4 85.4 85.4

Legislative outcomes

Fraction Dem in lower house 58.5 69.0 55.4 61.6 59.7Fraction Dem in upper house 54.5 65.7 55.2 64.8 60.2Indicator: Dem governor 60.4 68.8 35.4 62.5 56.3Fraction Female lower house na na na 11.4 18.0Fraction Female upper house na na na 5.5 13.4Voter turnout (presidential election years) 63.6 63.6 62.2 55.5 52.2Party competition in legislature −.092 −.079 −.058 −.054 −.034

Policy outcomes

Total taxes per capita $1982 161.4 249.8 464.4 568.2 734.9Total spending per cap $1982 370.5 534.3 974.1 1200.7 1526.0Family assistance per capita $1982 na 17.2 44.9 50.5 42.8Workers compensation per capita $1982 5.03 7.39 9.17 18.0 34.7

Income, demographic and state controls

Ideology: Citizen COPE score na 49.7 48.7 43.9 49.7State income per capita ($1982) 5554 6725 9058 10636 12985State population (millions) 3.13 3.72 4.21 4.68 5.13Percentage aged 65 and above 8.13 9.18 9.77 11.2 12.7Percentage aged 5 to 17 21.5 25.2 26.3 21.2 18.9

Notes: Poll taxes and literacy test data do not include Nebraska or Minnesota. The first column for corporate cam-paign restrictions presents results for 1952. Tax and expenditure limitations present an indicator for potentiallybinding tax and expenditure limitations. Voter turnout is the turnout for the highest office in the race in that year,divided by the state’s age-eligible voting population, reported here for election years: 1952, 1960, 1968, 1976, 1988.

Article 1 2/21/03 4:17 PM Page 20

running open primaries doubling between1960 and 1990. Another important institu-tional restriction is the ability of governorsand state legislators to stand for reelection.In the 1990s alone, ten states passed termlimits on the number of terms their gover-nors can hold successively and, by the end ofthe 1990s, three-quarters of all the conti-nental U.S. states had some sort of guberna-torial term limit.

Once elected, officials face different con-straints on what they can do. For example,there are differences between states in tax-setting power, and on whether the governorpossesses a line-item veto. There are alsoimportant differences between states in thescope of policy responsibility. For example,some states directly elect certain office hold-ers while others allow the governor or legis-lature to choose them. The responsibilitiesof different elected officials may also vary.

These political institutions may affectoutcomes directly—for example, superma-jority requirements may have a direct effecton tax rates. In other cases, these institu-tions may affect policy choices through theireffect on the character of the legislature.The fractions of seats in the states’ lowerand upper houses that are held byDemocrats, or the number of womenelected to the legislature, may depend uponelectoral rules. The middle panel of table 1shows that the proportion of seats held bywomen rose in the second half of our sam-ple period, and that the proportion of seatsheld by Democrats varied substantiallyfrom decade to decade, with the fraction ofseats in the lower house as high as 70 per-cent in 1960 and as low as 55 percent in1970. To the extent that these swings aredue to macro shocks that affect all states(the Vietnam War, for example), theirmovements will be absorbed in year effectsincluded in all regressions. Contemp-oraneous state-by-year movement in insti-tutional rules and, say, the fraction of seatsheld by Democrats will then be attributedto the institutional change.

An important feature of parties is the wayin which they generate competition for po-litical office. There is a view, perhaps mostfamously articulated by Valdimer Key (1950)that parties are likely to be most effective asa representative mechanism when they arein a truly competitive environment. There isno unanimously agreed method of measur-ing this. Authors have variously used differ-ences in seat or vote shares at the last elec-tion as a means of quantifying the extent ofcompetition between the parties. Table 1uses a very simple measure of party compe-tition equal to –1 times the absolute differ-ence from 0.5 in the fraction of seats held inthe lower house by Democrats times the ab-solute difference from 0.5 in the fraction ofseats held by Democrats in the upper house.In this way, larger (less negative) numbersare associated with more competition—thatis, with a closer balance in seats between thetwo parties.16 Table 1 suggests that partycompetition in state legislatures has risenthrough time.

The final panel of the table contains infor-mation on some of the time-varying state-level controls we will use in the empiricalwork presented below. States became signif-icantly wealthier over this period, with realstate income per capita more than doublingbetween 1950 and 1990. State populationshave become older on average as well, withthe proportion aged 65 and above risingfrom 8.1 percent to 12.7 percent, and theproportion that is school aged falling from21.5 percent to 18.9 percent. It is antici-pated that these income and demographicchanges will move states’ political and policychoices.


16 Our definition of competition is related to, butdistinct from, the fraction of seats in the upper andlower houses held by Democrats. As the fraction heldby Democrats in lower house increases from 0 toward0.5, both fraction Democrat and our competition meas-ure will increase. However, when fraction Democratincreases from 0.5 toward 1.0, the fraction Democratobviously increases while, ceteris paribus, our measureof competition falls.

Article 1 2/21/03 4:17 PM Page 21

While the United States is broadly a two-party system, there are differences in the po-litical complexion of the parties in differentstates. Thus the political scientists Erickson,Wright, and McIver (1989) observe that “theDemocratic party of Mississippi is far moreconservative than the Democratic party ofNew York and perhaps the New YorkRepublican party as well” (p. 731). Severalmeasures of the ideology of the citizens andparties in different states have been devel-oped. Berry et al. (1998) construct measuresof state citizen ideology from widely usedideology ratings of the state’s congressionaldelegation.17 These are the Americans forDemocratic Action (ADA) rating, and theAFL/CIO’s Committee on Political Edu-cation (COPE) rating. Berry et al. assign anideology rating to the citizens of each con-gressional district using a weighted averageof the congressional member’s score and hisor her election opponent’s score, weightingthe scores according to the number of voteseach received—0 denotes the most conser-vative score and 100 the most liberal. Theythen generate a statewide measure by takingthe simple average over all congressionaldistricts. Berry et al. also construct a mea-sure of government ideology, by assigning tothe governor and major party delegations inthe legislature the ratings of the members ofCongress from their party. Table 1 suggests amove toward a more conservative ideologyin the middle years of our period (roughly1980), followed by a return to a slightly moreliberal stance in the 1990s, as measured bythe citizens’ COPE scores.

The literature has long recognized thatdifferences exist in institutions, ideology,and legislative and policy outcomes be-tween states in the Southern region of theUnited States and elsewhere. We take asecond look at our institutional and out-

come variables in table 2, where for 1960and 1990 we present means for the U.S.South and for all other parts of the UnitedStates. In the early part of our sample, it isclear that the South restricted access to thevote: it was the Southern states that reliedon poll taxes and literacy tests in the 1950sand early 1960s. In 1960, Southern stateswere also significantly less likely to holdopen primaries or to allow citizens’ initia-tives. Southern states in all parts of oursample have significantly higher fractionsof Democrats in both the lower and upperhouses of their legislatures. In the earlypart of the period, their governors were sig-nificantly more likely to be Democrats—but were also significantly more conserva-tive. This was true of citizens in southernstates as well, measured using the COPEscore. Voter turnout is significantly lower inthe South (in part as a result of the restric-tions on registration and voting), as is partycompetition in the legislature, perhaps as aresult. Southern states are also less afflu-ent, measured using state income percapita. In the estimation results we presentbelow, we do not generally rely on thelevel-differences between states—becauseour estimation strategy generally includesstate fixed effects. To the extent thatSouthern states have changed their institu-tions and legislative composition over time,these changes help to identify the resultswe present below.

Overall, it is clear that there is interest-ing institutional, political, and economicvariation in the U.S. states, both acrossspace and time. This provides the core ofthe empirical exercise that follows.

5. Institutions and Political Outcomes

In this section, we consider how institu-tions affect political outcomes. The main in-stitutions of interest in this section are thosethat govern voter registration, candidate se-lection (primary rules), legislative redistrict-ing, and campaign finance. We examine


17 There is a large political science literature onmeasuring ideology of citizens and elected officials.William Berry et al. (1998) provide a useful review ofthe various methods used to construct these, as well asnew data.

Article 1 2/21/03 4:17 PM Page 22

their association with measurable politicaloutcomes—such as party control, politicalcompetition and ideology. From a theoreti-cal point of view, these exercises can best bethought of as getting an empirical handle on

(3), while bearing in mind that the institu-tional effect may be working through thechoice of candidates cjst (j ∈ {D,R}) aswell as through the process by which a givencandidate is elected.


TABLE 2REGIONAL DIFFERENCES IN INSTITUTIONS,

LEGISLATIVE OUTCOMES AND POLICY OUTCOMES

South Non-South South Non-South

Institutions: electoral rules 1960 1990

Poll taxes 31.3 0* 0 0% state population affected by a literacy test 33.6 2.8* 0 0Voter registration through vehicle registration 0 0 12.5 28.1Voter registration on polling day or no registration 0 3.1 0 9.4Open primaries 0 21.9* 12.5 28.1Restrictions on corporate campaign contributions 71.4 78.6 75.0 59.4Citizens initiatives 12.5 53.1* 18.8 59.4*

Institutions: decision-making rules

Tax and expenditure limitations 0 0 12.5 28.1Super-majority requirements 6.3 0 31.3 6.3*Gubernatorial line-item veto 93.8 75.0 93.8 81.3

Legislative outcomes

Fraction Dem in lower house 93.4 60.0* 71.7 53.6*Fraction Dem in upper house 91.8 51.9* 76.5 51.8*Indicator: Dem governor 87.5 59.4* 56.3 56.3Fraction Female lower house na na 11.3 21.4*Fraction Female upper house na na 9.6 15.3*Voter turnout (presidential election years) 47.1 71.8* 46.5 55.1*Party competition in legislature −.190 −.020* −.068 −.015*

Policy outcomes

Total taxes per capita $1982 251.1 249.1 691.5 756.6Total spending per cap $1982 500.1 551.4 1388.4 1594.7*Family assistance per capita $1982 16.1 17.8 28.9 49.7*Workers compensation per capita $1982 2.89 9.64 19.3 42.3

Income, demographic and state controls

Ideology: Citizen COPE score 32.1 58.5* 47.2 51.0State income per capita ($1982) 5687 7243* 12028 13463*State population (millions) 3.40 3.87 5.30 5.05Percentage population aged 65 and above 8.43 9.55* 12.8 12.6Percentage population aged 5 to 17 26.2 24.7* 19.2 18.8

Notes: An asterisk (*) notes that the difference between the South and the Non-South is significant at a 5 percentlevel. States in the South are AL, AR, DE, FL, GA, KY, LA, MD, MS, NC, OK, SC, TN, TX, VA, and WV. Tax andexpenditure limitations present an indicator for potentially binding tax and expenditure limitations. Voter turnoutis the turnout for the highest office in the race in that year, divided by the state’s age-eligible voting population,reported here for election years: 1960 and 1988.

Article 1 2/21/03 4:17 PM Page 23

5.1 Voting

If voting is an effective mechanism for de-termining policy outcomes, then it requiresthat individuals show up at the polls to ex-press their views. This has two parts. First,whether individuals have registered to voteand second, whether registered voters turnout on election day. As we noted in section 4,there are important institutional variationsin the way in which states organize voterregistration. There are good reasons to ex-pect that these will affect registration andhence turnout.

While some political scientists viewturnout as a good in itself—signaling thehealth and legitimacy of a democracy—it isless obvious whether turnout affects policyoutcomes. This depends on whether thereare important sources of bias in turnout, inthe sense that low turnout favors one partyover another. It is widely known thatturnout varies across different groups in so-ciety,18 with richer and better educated cit-izens more likely to vote. There is a long-standing view in political science thatbiases in turnout bias policy choices—themain expectation being that larger turnoutamong low-income groups will be corre-lated both with better election perfor-mance by the Democrats and with more re-distributive policies, as the income of thedecisive voter falls. This mechanism is keyto some recent papers, such as RolandBenabou (2000).

Kim Quaile Hill, Jan Leighley, and AngelaHinton-Andersson (1995) look at whetherincreased voting changes policy outcomes,by increasing the representation of low-income interests. Using panel data from1978–90, they find that turnout among

lower-class voters is positively associatedwith increased welfare spending. This is ro-bust to year effects, although they do not in-clude state effects. Using turnout as a re-gressor is problematic, however, since it islikely to be determined by some of the sameunobservable factors that drive policy. Forexample, it is possible that some informationabout the economy is revealed, informationthat leads both to higher desired welfarespending and to more people turning out tovote. Timothy Feddersen and WolfgangPesendorfer (2000) provide a particular ex-ample of this.

A promising approach to estimating theimpact of turnout on policy choice is to findinstitutional change that affects registrationand turnout independently of the demandfor policy. In recent U.S. history there arethree reforms that have affected the proba-bility that citizens will register to vote. First,in the early part of the century, many stateswere forced by the federal government toextend the franchise to women. Second,changes to voting rights laws in the 1960sand 1970s led to greater registration by mi-nority voters. More recently, states have en-couraged voter registration under so-called“motor-voter” laws that tie vehicle licenseregistration to voter registration. For all ofthese institutional changes, a strong casecan be made that the changes were drivenby a desire to protect citizens’ rights, andnot by a desire to change legislative out-comes.

John Lott and Lawrence Kenny (1999)consider the effect of extending the fran-chise to women, using data from 1870 to1940. They exploit the fact that some statesgave women the vote before it was federallymandated in 1920, which generates state-to-state variation in the timing of women’s suf-frage. They find that women’s participationincreased the size of government. While thefinding is interesting, they offer no theoreti-cal explanation, but suggest that it may bedue to the different policy priorities of menand women vis a vis child-rearing issues. It


18 The classic study in political science is RaymondWolfinger and Steven Rosenstone (1980). They, and ahost of subsequent researchers, estimate the effects ofvoter registration laws using micro data with turnout asthe outcome variable. They then interact these lawswith socioeconomic information about individuals,finding that registration laws have the most bite forlow-education groups.

Article 1 2/21/03 4:17 PM Page 24

could also be because women are poorerand specialize in nonmarket skills.19

Thomas Husted and Kenny (1997) con-sider the effect of the Federal voting acts ofthe 1960s and 1970s, which struck down lit-eracy requirements and poll taxes. As wediscuss in greater detail below, they find sig-nificant correlations with welfare spending.Earlier work by John Filer, Kenny, andRebecca Morton (1991) found these actshad a significant impact on turnout.

The ease with which voters may registerhas also been studied as a source of differ-ence in voter turnout (Benjamin Highton1997 reviews the political science literature).One of the key institutional variations thathas been studied is whether voters can regis-ter on election day. Suggestively, Highton,using data from 1980 and 1992, finds thatturnout is 10 percent higher in NorthDakota (where no voter registration is re-quired) and other states with election-dayregistration. The biggest difference inturnout appears among those with low levelsof education.

Arguably, studying these institutionalchanges provides a more satisfactory basisfor assessing whether who votes affects pol-icy outcomes than does simply controllingfor turnout in a reduced-form regression.Moreover, there is both cross-sectional andtime-series variation to exploit, as somestates changed their laws without promptingfrom the Federal government.

Table 3 examines the impact of a varietyof institutions on voter turnout. Specifically,we examine whether literacy tests and polltaxes, voter registration procedures, citizens’initiatives, and restrictions on corporatecampaign financing affect voter turnout,while controlling for year effects and a num-ber of time-varying state-level variables thatmay affect turnout (state income per capita

and income squared, state population andpopulation squared, the proportion of thepopulation aged 65 and above, and the pro-portion aged five to seventeen). Turnout isdefined here as the number of votes cast forthe highest office holding an election in thatyear divided by the total voting age popula-tion in the state. The impact of literacy testsand poll taxes on turnout is immediately ap-parent in column 1 of table 3. Consistentwith the results of Filer, Kenny, and Morton(1991, 1993), voter turnout was roughly 15percentage points lower on average in stateswhen poll taxes were in place. Because wecontrol for state fixed effects, the impact ofpoll taxes is identified using turnout infor-mation from states that ever had a poll tax,by calculating differences (state-by-state) inturnout, before and after the poll taxes werelifted.

Table 3 also shows that turnout is posi-tively correlated with having a constitutionthat permits citizens’ initiatives: turnout ison average 3 percentage points higher instates where initiatives are allowed by law.This is consistent with the often articulatedview among political scientists that directdemocracy promotes political participation(see, for example, David Butler and AustinRanney 1978). However, there is very littletime-series variation in our initiative vari-able, and this regression does not includefixed effects. Hence, the result could be dueto some state-specific omitted variable suchas “political culture” that drives initiativesand voter turnout.

States with the least cumbersome voterregistration—either allowing registration onpolling day, or not requiring registration atall—have voter turnout that is roughly 2 per-centage points higher on average (althoughthe effect is only marginally significant whenall controls are included in column 5 and notsignificant in column 3). We find no impactof motor-voter registration on voterturnout—the effects are small and never sig-nificantly different from zero. Perhaps statesthat turned to motor-voter registration early


19 Lena Edlund and Rohini Pande (2002) look athow changing family circumstances in the UnitedStates have changed the voting patterns by gender inthe post-war period.

Article 1 2/21/03 4:17 PM Page 25

on were those where it was most difficult toencourage people to vote.

Voter turnout is significantly correlatedwith restrictions on corporate campaigncontributions. Controlling for state fixed ef-fects, in those states that adopted restric-tions on corporate campaign contributions,turnout was 2 percentage points higher (apoint we return to below). These results arerobust to estimating the impact of poll taxes,literacy tests, voter registration and corpo-rate campaign finance requirements simul-taneously (column 5), where these institu-tional rules are jointly highly significant(F-test = 66.09, p-value = .0000).

That institutional rules may also affect thepolitical composition of the legislature isclear from table 4, where we regress threeoutcome measures—the fraction of the seats

in the states’ upper houses held byDemocrats, whether the governor is aDemocrat, and measures of party competi-tion—on indicators that the state has a polltax, a literacy test, voter registration throughvehicle registration, day-of-polling registra-tion, and an indicator for restrictions on cam-paign contributions. We control for state andyear effects and the same time-varying state-level controls introduced above. In table 4,all controls have been lagged one period, torepresent conditions in the state in the yearin which these office-holders were elected.

Table 4 shows that literacy tests protectthe seats of Democrats in those Southernstates that had these restrictions on voting.The effects of poll taxes and literacy tests areidentified from the timing of the changes instate laws relative to the timing of changes


TABLE 3DEPENDENT VARIABLE:

ELECTION TURNOUT OF AGE-ELIGIBLE VOTERS

(1) (2) (3) (4) (5)

Poll tax −.140 −.157(.010) (.013)

Literacy test −.117 — — — −.138(.011) (.012)

Citizen initiatives — .033 — — —(.014)

Indicator: voter registration through — — .003 — .004vehicle agency (.008) (.007)

Indicator: voter registration possible on — — .017 — .025polling day or no registration necessary (.013) (.014)

Indicator: restriction on corporate — — — .021 .018campaign contributions (.006) (.005)

State fixed effects included? Yes No Yes Yes Yes

Years over which regression run even years even years even years even years even years1950–98 1960–98 1950–98 1952–98 1952–98

Number of observations 1174 958 1198 1060 1038

Notes: Standard errors in parentheses. All regressions control for year effects, and include controls for the propor-tion of population aged 65 and above; the proportion of population aged 5 to 17; state income per capita in $1982and income per capita squared; state population and population squared. Omitted voter registration category incolumns 3 and 5 is “conventional” registration. We do not include state fixed effect in column 2 because only fourstates changed whether they allowed initiatives over the period 1960–98. In column 2, we estimate robust stan-dard errors, and allow for correlation in the unobservables from the same state. Campaign finance data are cur-rently not available for 1950.

Article 1 2/21/03 4:17 PM Page 26

observed in the composition of the legisla-tures. The relationship between the two canbe seen most clearly in figure 2, which dis-plays the fraction of seats in state upperhouses held by Democrats in four of the fivestates that had literacy tests in the 1950s andearly 1960s. These tests were eliminatedwith the 1965 Voting Rights Act, a point intime marked in figure 2 by the vertical

line.20 In three of these states (LA, MS, SC),100 percent of seats in the state upper housewere held by Democrats until 1965. Startingin 1966, Democratic control began to erode.The fourth state (VA) begins with a slightly


TABLE 4POLITICAL INSTITUTIONS AND REPRESENTATION

Fraction Democrat in Party Competition in Indicator: DemocraticState Upper House Legislature Govenor

Poll tax .032 .031 — −.025 .004 — .148 .046 —(.028) (.030) (.009) (.010) (.130) (.138)

Literacy test .081 .082 — −.022 .001 — .006 −.101 —(.025) (.027) (.008) (.009) (.116) (.124)

Indicator: voter registration through −.015 −.015 — −.002 −.003 — .048 .052 —vehicle agency (.015) (.015) (.005) (.005) (.069) (.069)

Indicator: voter registration possible on .056 .056 — −.039 −.043 — .007 .025 —polling day or no registration necessary (.029) (.029) (.010) (.010) (.136) (.136)

Indicator: restriction on corporate .021 .020 — .010 .007 — .038 .053 —campaign contributions (.011) (.011) (.004) (.004) (.049) (.049)

F-test: institutional variables 4.47 4.10 — 8.36 5.05 — 0.48 0.50 —(p-value in parentheses) (.0005) (.0011) (.0000) (.0001) (.7913) (.7772)

Voter turnout — −.002 — — .178 — — −.693 —(.067) (.022) (.312)

IV estimation: Voter turnout — — −.309 — — .169 — — −.340(.116) (.038) (.531)

F-test: — — 3.79 — — 5.88 — — 0.54(.0021) (.0000) (.7467)

Number of observations 1027 1025 1025 1027 1025 1025 1040 1038 1038

Notes: Standard errors in parentheses. All regressions run over odd-years from 1953 to 1999. All regressions con-trol for year and state fixed effects, and include controls for the proportion of population aged 65 and above; theproportion of population aged 5 to 17; state income per capita in $1982 and income squared; and state populationand population squared. Omitted voter registration category is “conventional” registration. All control variables arelagged one year, to reflect the conditions in place at the time of the election. Results in column 3 are for instru-mental variables estimation, where voter turnout is instrumented on the institutional rules that appear in columns1 and 2. The F-test in column 3 compares the fit of the regression using the predicted value to that in column 1,where the institutional rules are allowed to enter in an unrestricted fashion. Results in column 6 report an analo-gous comparison for party competition in the legislature. We reject that these instituational rules are affectingfraction Democrats and party competition solely through their effect on voter turnout.

20 The fifth state, Georgia, looks very much likethese four, but is not shown in order to make it easier tofollow the changes across the states over time.

Article 1 2/21/03 4:17 PM Page 27

lower fraction Democrat (95 percent), andDemocratic control erodes from this lowerinitial level. That this is not simply pickingup the general movement away from theDemocratic party in the South can be seenby comparing results for these states withthose for states that did not have literacytests. Erosion of Democratic control in thelatter does not match the pattern seen in fig-ure 2. In Kentucky and Tennessee, for ex-ample, erosion began in the 1950s, while inAlabama and Arkansas it did not begin untilthe 1980s.21

Poll taxes and literacy tests kept turnoutlow and the fraction of Democrats in thelegislature high. This negative associationbetween turnout and the fraction ofDemocrats elected may be special to theSouth in the 1960s. These restrictionsstopped minority voters from going to the

polls in Southern states, where protest votesor votes for change would be votes for theRepublican party. Thus, once enfranchised,minority voters in the South lessened theDemocrat’s stranglehold on the state houses.Elsewhere, institutions that increasedturnout appear to act as the literature sug-gests, with increased turnout giving way toan increase in Democratic representation.Day-of-polling registration has a positive ef-fect on turnout (see table 3), and is associ-ated with an increase in the fraction of seatsheld by Democrats. Restrictions on corpo-rate campaign contributions are also associ-ated with both higher turnout and with ahigher fraction of the states’ upper housesheld by Democrats: on average, if a statepasses restrictions on corporate contribu-tions, the share of the state upper house thatis held by Democrats increases by 2.1 per-centage points. Jointly, these institutionalrules are significant correlates of house


21 Details are available from the authors on request.

1

.9

.8

.7

.6

.5

20001990

literacy tests abolished

Virginia

South Carolina

Mississippi

Louisiana

19801970

year

196519601950

Figure 2. The Abolition of Literacy Tests, and the Compositionof State Upper Houses

Article 1 2/21/03 4:17 PM Page 28

composition (F-test = 4.47, p-value = .0005).(Results are quite similar when the propor-tion of Democrats in the states’ lower housesis regressed on these same controls.)

The institutional rules we have identifiedhere also have significant effects on partycompetition in the legislature (columns 4and 5). Whether rules that systematically fa-vor one party strengthen competition in thelegislature depends upon the initial legisla-tive composition. Thus poll taxes, literacytests, and day-of-polling registration weakencompetition, in general improving the posi-tion of Democrats in legislatures whereDemocrats are strong. In contrast, restric-tions on corporate campaign contributionsincrease competition by improving the posi-tion of Democrats in states where Demo-crats are less likely to hold majorities in thestate houses.22

While these institutional rules have a largeand significant effect on the state legisla-tures, they appear to play little role in theelection of a governor. None of the institu-tional controls is significant by itself, nor arethey jointly significant. This may indicate thegreater relative importance of idiosyncraticpersonality factors in gubernatorial elections.

As we have seen above, turnout is signifi-cantly correlated with these election rulesbut, interestingly, the impact of institutionalrules on the party composition of the legisla-ture does not work through voter turnout.When we include a control for turnout (col-umn 2), it has no significant effect on thefraction of the upper house held by Demo-crats, and has no effect on the coefficients onthe electoral rules. (Neither is the turnoutvariable significant when the institutional

rules are omitted as controls.) Where turnoutappears to have a significant effect on thelegislature is in the degree of party competi-tion (column 5). In those years in whichturnout is large, the legislature becomesmore evenly divided between Republicansand Democrats. The insignificant associationof turnout with the fraction of seats held byDemocrats, given the significant associationwith competition in the legislature, meritsfurther discussion. Turnout may keep thefraction of Democrats close to 0.5, whichhere is registered as increasing competition.However, this does not imply a monotonicrelationship between turnout and the frac-tion of seats going to Democrats. If the frac-tion held by Democrats is above 0.5, in-creased turnout appears to bring the fractiondown. Alternatively, when the fraction is lessthan 0.5, turnout pulls the fraction up.

For each dependent variable, we can testwhether institutional rules are affecting leg-islative composition solely through their ef-fect on turnout. In the third column, wepresent results for instrumental variables es-timation, where voter turnout is instru-mented on the institutional rules that appearin columns 1 and 2. The F-test in column 3compares the fit of the regression that usesthe predicted value of turnout in column 3to that in column 1, where the institutionalrules are allowed to enter in an unrestrictedfashion. Results in column 6 report an analo-gous comparison for party competition inthe legislature. We reject that these institu-tional rules are affecting the fraction Dem-ocrat and party competition solely throughtheir effect on voter turnout.

Overall, results in this section provide evi-dence that institutional variables have realeffects on legislative composition and com-petition, only part of which can be explainedby their impact on voter turnout.

5.2 Primaries

Institutional rules may also affect who isselected to run for office. The main vehiclefor candidate selection is the primary


22 For example, Maine limited corporate campaigncontributions beginning in mid-1970s. Prior to that,Democrats were a minority in the state upper house.Averaged over all years from 1950 to 1975, Democratsheld 29 percent of seats in the upper house. After theimposition of corporate restrictions, averaged over allyears from 1976 to 2000, Democrats held 54 percent ofseats in the upper house. Regression results in table 4are based on this sort of calculation (while allowing foryear effects and state-level time varying controls).

Article 1 2/21/03 4:17 PM Page 29

process, and there are important differencesin primary rules across states. Models ofelectoral competition that include primariestypically divide the action into three stages.At stage one voters affiliate themselves witha particular party. At stage two parties nomi-nate some candidates to run for office fromamong the subset that choose to stand. Atstage three voters choose between partynominees.

States vary in the primary rules that theyuse to determine candidate choices. Thereare basically two institutional variants. Inopen primaries, participants do not need todeclare their party allegiance in advance ofthe primary.23 In closed primaries, participa-tion is limited to voters who have declared aparty affiliation a sufficient time prior to theprimary.24

Elisabeth Gerber and Morton (1998a) ar-gue that more open primary arrangementswill likely lead to more moderate candidatesbeing chosen. Closed primaries increase theinfluence of party elites, who are more likelyto have strong ideological preferences. Byfacilitating crossover voting (i.e., for a partywith whom one does not identify), open pri-maries will tend to lead to more moderatecandidates being selected (although Gerberand Morton 1998a observe that things areless clear-cut if crossover voting has a strate-gic element). They cite a significant numberof political science studies suggesting thatcrossover voting is more common in moreopen primaries. To look at this empirically,they examine the ideological stances of win-ners in congressional races. They use pastvoting records in presidential races as a con-trol for the ideology of the districts, and ar-gue that the evidence is consistent with their

hypothesis that more open primaries spawnless extreme candidates.25

We examine the impact of primary ruleson political outcomes in table 5. The firsttwo columns provide evidence on the rela-tionship between open primaries and voterturnout for the highest office holding anelection in that year. Data are available forthe even years between 1950 and 1998. Wefind that open primaries are positively re-lated to voter turnout, increasing turnout byabout 1 to 2 percentage points. When con-trols are added for other institutions relatedto turnout, open primaries are a significantdeterminant of turnout (column 2).

Open primaries are not significantly re-lated to the party-composition of the legisla-ture (columns 3 and 4). However, open pri-maries are negatively correlated with thefraction of women elected to state lowerhouses (columns 5 and 6). Again, becausestate fixed effects are included in each spec-ification, this effect is identified usingchanges in the election of women legislatorswithin states that changed their primaryrules during the period 1977 to 1999, whichoccurred for half of all U.S. states. This find-ing could signify that party elites are morekeen on women’s representation than arevoters at large, but it clearly requires morethorough investigation. It also raises the is-sue of what motivated changes in primaryrules in the states that implemented them—a theme that we return to in section 8 below.

In the spirit of the discussion of Gerberand Morton (1998a), we also test whetheropen primaries reduce the ideology gap be-tween citizens and their elected officials. Todo so, we first generate measures of this gapusing data from Berry et al. (1998). We findthat open primaries are associated with re-ductions in the absolute differences be-tween the ideologies of citizens and electedofficials, measured using their COPE scores


23 Within this general category are blanket prima-ries, where voters may participate in all parties’ prima-ries. In nonpartisan primaries, voters choose amongcandidates without declaring any party allegiance.

24 Primaries are semi-closed if new registrants canregister for a party on the day of the primary or if inde-pendents are allowed to participate.

25 Gerber and Morton (1998b) examine the effect ofprimary rules on how inclusive the parties are likely tobe in forming electoral coalitions.

Article 1 2/21/03 4:17 PM Page 30

(columns 7 and 8). In line with the findingsof Gerber and Morton (1998a), this suggeststhat open primaries may have a systematiceffect on political representation, perhapsbeing indicative of greater empowerment ofless ideologically motivated voters.

5.3 Campaign Finance

Campaign finance reform has been a ma-jor political issue of late, given a generalpopular concern about the level of politicalspending. It is estimated, for example, that

more than $3 billion was spent on politicalcampaigns in the year 2000 elections.26

Current campaign financing rules raisemany difficult issues, including the possibil-ity that public officials may become be-holden to special interests, and that thesums of money necessary to launch a cam-paign may discourage able challengers, tothe benefit of incumbents. Steven Levitt(1994) takes a more sanguine view, at least


26 Public Campaign, www.publicampaign.org.

TABLE 5THE IMPACT OF PRIMARY RULES ON TURNOUT, IDEOLOGY AND PARTY COMPETITION

Dependent Variable:

AbsoluteFraction Fraction difference

Democrats in women in (citizen-state lower state lower government

Turnout house house COPE score)

Indicator: open primaries .011 .015 .001 −.001 −.015 −.014 −3.47 −3.41(.007) (.007) (.012) (.013) (.007) (.007) (2.30) (2.44)

Poll tax — −.155 — .014 — — — —(.134) (.025)

Literacy test — −.137 — .045 — — — —(.012) (.022)

Indicator: voter registration through — .010 — .021 — .009 — −2.42vehicle agency (.009) (.017) (.007) (2.25)

Indicator: voter registration possible on — .020 — .039 — −.056 — 2.95polling day or no registration necessary (.016) (.029) (.017) (3.72)

Indicator: restriction on corporate — .018 — .020 — .015 — 1.89campaign contribution (.005) (.010) (.006) (1.43)

even years odd years odd yearsYears over which regression run 1950–1990, 1951–1991, 1975–1991, even years

1996, 1998 1997, 1999 1997, 1999 1960–1990

Number of observations 1099 942 1067 934 525 498 768 709

Notes: Standard errors in parentheses. All regressions control for year and state effects, and include controls forthe proportion of population aged 65 and above; the proportion of population aged 5 to 17; state income percapita in $1982 and income squared; and state population and population squared. Omitted voter registration cat-egory in columns 3 and 5 is “conventional” registration. No registration was necessary in North Dakota from 1951to 1998, and we have added that state to “registration possible on polling day.” For regressions in columns 3–6, allcontrol variables have been lagged one period, to reflect the conditions in place at the time of the election.

Article 1 2/21/03 4:17 PM Page 31

for congressional elections, suggesting thatthe most careful recent studies have notfound spending by challengers to be moreeffective than that of incumbents and that,for this reason, placing limits on spendingwill do little to alter the incumbency advan-tage. In addition, Levitt (1995) reviews workthat questions whether political action com-mittees (PACs) have much of an effect onelected officials’ behavior once in office. Heargues that if spending by incumbents is lesseffective, then caps on election spendingwill be anti-competitive, whereas if spendingby incumbents and challengers has equal ef-fects then caps will tend to be pro-competi-tive. The history of campaign finance re-form, details of the current laws, andinnovations in the states is provided inAnthony Corrado et al. (1997).

Lott (2000) looks at the recent history ofcampaign spending in the U.S. states, andfinds that gubernatorial real per-capita cam-paign spending rose by 58 percent between1982 and 1990 in those states that held elec-tions in 1982, 1986 and 1990; and by 62 per-cent in those that held elections every fouryears from 1980 to 1992. Using data on statehouse and senate races for sixteen statesover various periods for which data wereavailable, and gubernatorial races for the pe-riod from 1977–94, and controlling for bothstate fixed effects and year effects, Lott findsthat state per-capita expenditure is a closepredictor of the level of campaign spending.In addition, open seats for governor attracthigher levels of campaign spending.

There has been a great deal of variationacross states and over time in the adminis-tration and enforcement of campaign fi-nance laws. Robert Huckshorn (1985) dis-cusses variation across states in four types offinance regulations: spending limits by can-didate; contribution limits by different typesof supporters (corporations, unions, individ-uals); disclosure rules; and monitoring andenforcement mechanisms to stop illegal orunethical expenditures. Differences acrossstates creates a source of institutional varia-

tion that may have political ramifications. Aswell as creating an interesting source of in-stitutional variation for political testing, thetheoretical consequences of these institu-tional variations are of interest. They may,for example, affect the kinds of candidateswho are selected, as in Coate (2001). Bychanging the provision of information, cam-paign finance can also affect voter attach-ment to particular parties or candidates.27

While there is a large body of work examin-ing the impact of campaign spending in fed-eral elections, explorations of the empirical ef-fects of campaign finance laws at the statelevel are extremely sparse. Lott (2000) reportsthat limits on donations and outright banshave no statistically significant effects on cam-paign spending in his regressions. ThomasStratmann and Francisco Aparicio-Castillo(2001) use the competitiveness of state houseraces as their outcome variable to examine theempirical consequences of campaign financelaws, in state fixed-effect models using data on45 states for the period 1980 to 1999.Focusing on regulations that affect individualcontributions, they find that contribution reg-ulations hold down incumbents’ vote shares.

Our empirical results focus on limitsplaced on corporate campaign financing andtheir effects on turnout, Democratic control,and women’s representation in the legisla-ture. Corporate financing may affect a can-didate’s popularity, if advertising engendersgood will. Alternatively, corporate financingmay affect which candidates choose to standfor office, and which are chosen by theirparties to run. Our results, presented in thebottom row of table 5, suggest that when astate restricts corporate contributions,turnout is higher, and the fraction of thelower house held by Democrats and bywomen is significantly higher. This gives cre-dence to the conventional view that thecomposition of the groups giving money topolitical campaigns can affect political


27 For an overview campaign finance reform seeMorton and Charles Cameron (1992).

Article 1 2/21/03 4:17 PM Page 32

representation. It is clear that there is scopefor expanding and refining this sort of analy-sis in future work. There is also the need toconsider in more detail what forces areshaping the propensity to change existingcampaign finance laws.

Public funding of political parties is per-haps the main alternative to relying on pri-vate contributions. The states may also pro-vide a context for studying this through thecheck-off available on some state tax returns,in which tax filers can allocate some smallamount of their returns to public funding ofcampaigns. This phenomenon seems to havebeen studied only to a limited degree. RuthJones (1981) examines the impact of thecross-state variation in schemes for publicfunding of political parties during the 1970s,mainly through tax check-off schemes. Shefinds that around 20 percent of the popula-tion participated in the tax check-offschemes introduced. The data suggest that,in terms of financing, the Democrats are dis-proportionate beneficiaries. However, shedoes not conduct formal empirical testing ofthe effect of state funding on outcomes.

5.4 Redistricting

Another important institutional variationacross states that can have a significant im-pact on political (and hence policy) out-comes is the relationship between votingand political control. It is common to char-acterize an idealized political system as onein which the proportion of seats held is pro-portional to vote share. Although the norma-tive foundations of this are unclear, it is auseful benchmark. In practice, majoritariansystems like the United States do not yieldproportionality and, for many years, the so-called cube law was purported to character-ize the relationship observed.28 The rela-

tionship between seats and votes is consid-ered to be extremely important in theprocess of designing districts.

Legislative redistricting has been impor-tant in increasing the number of African-Americans who hold office, through the cre-ation of so-called “majority-minority”districts. Tim Sass and Bobby Pittman(2000) look at the link between electionstructure and black representation in theSouth using data between 1970 and 1996.Their data are for city council elections,where they explain the fraction of blackselected as a function of the percentage ofblacks in the population and variables repre-senting the electoral system. They estimatethis in levels and first differences, findingsome evidence that the move to district elec-tions had an effect on black representationearly in the sample period.

Andrea Gelman and Gary King (1994) usedata on state legislative districts from 1968 to1988 to examine whether redistricting affectselectoral responsiveness—the degree towhich the party composition of the legislatureresponds to voter preferences—and partisanbias—the fairness with which state-cumula-tive votes for a party translate into seats in thestate legislature.29 They find that redistrictingincreases electoral responsiveness, largely byinducing uncertainty in the electoral process.Gelman and King, however, find that whenone party controls the redistricting map, thisinduces partisan bias. On average, the con-trolling party gains 6 percent of the seats thatwould have gone to the other party, had itcontrolled the redistricting. They concludethat “even though redistricting makes theelectoral system substantially fairer over-all than if there were no redistricting, the


28 See Maurice Kendall and Alan Stuart (1950). Thisis represented by

s

1− s=

(v

1− v

)3

where s is share of seats and v is share of votes.

29 They propose a generalization of the cube law re-ferred to in the previous footnote where

s

1− s= β

(v

1− v

)ρ

where β measures the bias and ρ measures the respon-siveness.

Article 1 2/21/03 4:17 PM Page 33

difference between Democratic and Rep-ublican control over the drawing of districtmaps is still one that politicians are rightfullyconcerned about” (p. 553).

We test the extent to which a unified partycontrol influences party competitionthrough redistricting, by analyzing the leg-islative composition of the 48 continentalU.S. states between 1952 and 1995. Afterdata on the U.S. population have been re-leased by the Census Bureau following thedecennial census, states go to work to reap-portion legislative districts. In all but a hand-ful of states (AR, HI, ID, NJ, and WA), re-districting begins with the legislature andthe governor. If power is divided, agreementis often difficult to reach, and redistrictingcan end up in the courts. We create an indi-cator that a party controls redistricting, us-ing information on whether it has unifiedcontrol of both houses of the legislature andthe governor’s office in the decennial censusyear.30 We then test whether the change inthe number of seats held by the Democraticparty in the legislature following redistrict-ing (decennial year + 2) is significantly cor-related with whether the Democrats orRepublicans controlled the state’s redistrict-ing following the census.

In table 6, we regress the change in thefraction of seats held by the Democrats be-tween all years (t) and years (t − 2) on indi-cators that the Democratic party held uni-fied control, and this Democratic controlindicator interacted with an indicator thatthis election year is immediately after redis-tricting. We include analogous variables forRepublican control (thus leaving mixed con-trol as the omitted category). We find evi-dence of the same partisan bias discussed byGelman and King (1994). Controlling forstate and year effects and time-varying state-

level variables, we find that when theDemocrats controlled redistricting, theyprotected Democratic seats in the lowerhouse of the legislature following redistrict-ing, and that the opposite held whenRepublicans controlled redistricting. Weneed to be careful in our wording here: onaverage, if the Democrats held unified con-trol in a state in year t − 2, then in year t theDemocrats are likely to lose seats, with theopposite holding true for Democrats whenRepublicans hold unified control. This maybe a regression to the mean phenomenon—the political pendulum swings and thenswings back—but is not special to redistrictingyears. What is special to the post-redistrictingyears is that the Democrats, when in control,can stop the pendulum from swinging backas far as it otherwise would. To take an ex-ample from table 6: column 3 presents re-sults on the change in the number of seatsheld by the Democrats in the state lowerhouse. If Democrats had unified control inyear t − 2, we would expect them to losefour to five seats in year t on average. If yeart is just after a redistricting overseen by theDemocrats, that loss is cut by 3 to 4 seats.

The results in table 6 suggest that thereis potential for using redistricting as a wayof trying to deal with the potential endo-geneity of political control. Redistrictingfollowing a census may lead to changes inlegislative composition that are simply dueto the redrawing of district boundaries,and not to voters’ underlying preferences.This is an additional area ripe for furtherinvestigation.

5.5 Overview

There is little doubt that variations in in-stitutions affecting voting, primaries, redis-tricting, and campaign finance have signifi-cant consequences for political outcomes.Voter turnout, party representation, the de-gree of competition in the legislature, andthe distance between the ideology of votersand their representatives are all correlatedwith political institutions. For these to have


30 We recognize that this will not yield a perfectmeasure of where redistricting is partisan, due to courtchallenges and legislative impasses among other rea-sons (see Gelman and King 1994 on this point).However, it is likely to be a useful instrument in pre-dicting when partisan redistricting will occur.

Article 1 2/21/03 4:17 PM Page 34

real effects on policy, however, it must bethe case that these political outcomes arealso drivers of policy. We pick up this threadin the next section, where we review what isknown about the extent to which thesemeasures of legislative outcomes influencepolicy choice.

As an intermediate step, we discusswhether the institutions that we have inves-tigated in this section appear to have any re-duced-form effects on policy outcomes, as inequation (9). There is some existing evi-dence that they do. For example, Hustedand Kenny (1997) consider whether institu-tions had an impact on welfare spending, us-

ing data from a panel of states over the pe-riod 1950–88. Their results are consistentwith the idea that the income of the “aver-age” voter was lowered by the extension ofthe franchise and that this led to higher wel-fare spending. This effect is essentially iden-tified by the time-series variation inSouthern states, where most of the restric-tions obtained prior to changes in federallaw.

Racial bias in turnout has been investi-gated in Benjamin Radcliff and Martin Saiz(1995), who examine the impact of voterturnout over the period 1979 to 1992 onwelfare spending and on policy liberalism


TABLE 6REDISTRICTING AND LEGISLATIVE COMPOSITION

Dependent Variable:

Change in Change inChange in Change in number of number of

Fraction Dem Fraction Dem Dems lower Dems upperlower house upper house house house

Indicator: Post-redistricting and Dems held .036 .008 3.81 0.24unified legislature and governor during (.010) (.010) (1.16) (0.40)redistricting (year t − 2)

Indicator: Post-redistricting and Reps held −.018 −.031 −2.06 −.097unified legislature and governor seat during (.013) (.013) (1.50) (0.52)redistricting (year t − 2)

F-test: joint significance of redistricting with 10.37 4.16 9.20 2.65power variables (p-value) (.0000) (.0157) (.0000) (.0712)

Indicator: Democrats held unified legislature −.047 −.039 −4.73 −1.45and governor seat (year t − 2) (.006) (.006) (0.70) (0.24)

Indicator: Republicans held unified legislature .055 .056 5.81 2.03and governor seat (year t − 2) (.007) (.007) (0.83) (0.29)

Indicator: post redistricting −.015 .033 −0.39 0.94(.035) (.036) (4.04) (1.40)

Years over which regression run all years all years all years all years1952–1995 1952–1995 1952–1955 1952–1995

Number of observations 2024 2024 2024 2024

Notes: Standard errors in parentheses. All regressions control for year and state effects, and include controls forstate income per capita in $1982 and income squared; and state population and population squared. Nebraska isremoved from the analysis, because it has a unicameral, non-partisan legislature. Observations for Minnesota arepresent only from 1973 on.

Article 1 2/21/03 4:17 PM Page 35

measures (à la Wright, Erikson, and McIver1987). They focus on welfare spending inthe 26 U.S. states whose populations weremore than 5 percent black in 1990. Con-structing measures of bias in turnout as theratio of black to white turnout, they find, inpanel data estimation that includes statefixed effects, that welfare spending is lowerin states with higher black turnout. This isbroadly consistent with the Husted andKenny findings. Radcliff and Saiz also find anegative association between cross-sectionalmeasures of policy liberalism and theirmeasure of turnout bias. While interesting,these findings leave open the concern thatturnout is endogenous.

Table 7 looks at the extent to which insti-tutional rules are associated with state taxes,overall spending, and transfers. Here, weregress taxes and spending on institutionalrules, together with the time-varying statelevel variables discussed above, controllingfor state fixed effects and year effects.

In table 7, we find that open primaries areassociated with a reduction in total taxes andspending of roughly $20 per capita. Open pri-maries also appear to have distributional ef-fects on state spending: total transfers are sig-nificantly higher when office holders areelected under open primaries and, in particu-lar, family assistance per capita spending issignificantly higher. In contrast, the generos-ity of the state’s workers’ compensation pro-gram, measured here as total workers com-pensation payments per capita, are negativelyand significantly correlated with open pri-maries. Less costly voter registration—through motor-voter rules, or through day-of-polling registration—is generally associatedwith higher taxes, higher spending, and largerfamily assistance and workers’ compensationpayments. Restrictions on corporate contri-butions are associated with lower overalltaxes, but higher transfers in the form of fam-ily assistance. For all of these fiscal outcomes,our institutional rules are jointly significant.

In section 7.6, we extend this discussionto see whether there is evidence that these

policy correlations are capturing the re-duced-form effect of institutions on politicalrepresentation. As we have already shown,these institutions are correlated withturnout and representation, providing somereason to think that these aspects of the pol-icy process are important to underlying pol-icy outcomes. The intermediate step inthis—the effect of political representationon policy—is the subject of the next section.

6. Political Representation and PolicyOutcomes

This section investigates the link betweenpolitical representation—measured chieflyby party identity and political competition—and measurable policy outcomes. For thesake of consistency, we maintain a small coreset of outcomes for the empirical work pre-sented, focussing on total taxes per capita,total state spending per capita, family assis-tance spending per capita and workers com-pensation spending per capita.

6.1 Party Representation

At the heart of the democratic ideal is thenotion that government policy should beguided by citizens’ preferences. In represen-tative democracies, public opinion affectspolicy only indirectly by influencing theidentity of elected decision makers. Theseoften represent different parties, and thedifferent ideologies of the parties are oftenargued to be the core choices on offer. A keyissue is how opinion finds its way into thepolicy process, and a key role is given toelections in shaping policies that are repre-sentative of widely held opinion. This, ofcourse, requires that people vote, the out-come of which is limited by the choices peo-ple have available to them—involving gener-ally two candidates (one from each party) inthe case of state legislatures. (As we discussbelow, some states permit more directopportunities to influence policy throughcitizens’ initiatives.)

How parties represent preferences is acentral area of research in political science.


Article 1 2/21/03 4:17 PM Page 36

Moreover, many of the key approaches tomodeling representative democracy hingeon assumptions about parties’ strategies andmotivations. While parties are frequentlycharacterized as ideologically based organi-zations with distinct agendas, there remainsan important empirical question aboutwhether party control really does delivermeasurable policy differences and whetherparticular policies appear to be more re-sponsive to party identity. In light of its cen-trality, it is not surprising that a large litera-ture has developed that attempts to gaugehow the process of representation works

empirically, and whether party controlmakes a difference in determining policyoutcomes.31 Finding that parties do not mat-ter would deal a blow to the stereotypicalcharacterization of party differences thatmost commentators take for granted. Thereis a huge literature in political science usingcross-country and within-country evidence


TABLE 7REDUCED FORM IMPACT OF INSTITUTIONAL RULES ON STATE TAXES AND SPENDING PER CAPITA

Dependent Variable:

TotalTotal Total workers

government transfer Total family compensationTotal taxes spending per payments assistance paymentsper capita capita per capita per capita per capita

Open primaries −19.25 −18.24 31.8 2.73 −6.31(8.03) (13.7) (10.9) (1.37) (1.80)

Indicator: voter registration through 35.76 9.78 52.8 9.41 6.98vehicle agency (7.10) (11.5) (10.1) (1.27) (1.66)

Indicator: voter registration possible on 120.38 114.7 0.77 20.5 1.64polling day or no registration necessary (13.41) (22.2) (18.5) (2.34) (3.08)

Indicator: restriction on corporate −16.00 8.35 34.1 5.73 −0.62campaign contributions (5.33) (8.54) (7.57) (0.95) (1.25)

F-test: joint significance institutional 28.46 7.55 13.87 35.94 7.83variables (p-value in parentheses) (.0000) (.0000) (.0000) (.0000) (.0000)

Years over which regression run All years All years All years All years All years1958, 1958, 1958, 1958, 1958,

1960–97 1960–96 1960–98 1960–98 1960–98

Number of observations 1822 1781 1877 1877 1877

Notes: Standard errors in parentheses. All dependent variables are in 1982 dollars. All regressions control for yearand state effects, and include controls for the proportion of population aged 65 and above; the proportion of popu-lation aged 5 to 17; state income per capita in $1982 and income squared; and state population and populationsquared. Omitted voter registration category is “conventional” registration. No registration was necessary in NorthDakota from 1951 to 1998, and we have added that state to “registration possible on polling day.” Rules governingregistration and voting have been lagged one or two periods, to reflect the conditions in place at the time of theelection.

31 It should be borne in mind that this leads to exclu-sion of Nebraska, since it holds nonpartisan electionsand, in early years, Minnesota, since its parties werenot comparable to Democratic and Republican partiesin other states.

Article 1 2/21/03 4:17 PM Page 37

on whether parties matter for policy. The lit-erature that has looked at the federal level inthe United States is far from conclusive. Byexploiting both cross-sectional and time-series variation, the U.S. states are a muchmore promising testing ground.

In principle, it is straightforward to inves-tigate this issue. Returning to equation (2)above, party control and party “competitive-ness” can be thought of as part of the vector�st . However, before estimating (2) withsuch variables on the right-hand side, thereare a number of econometric and measure-ment issues to be resolved.

At any point in time, the controlling par-ties may vary—both between the two housesof the legislature and vis a vis the governor’soffice—raising the interesting question ofhow to characterize the political complexionof the state. Policy making may well differbetween unified and divided control of thelegislature and the executive. James Alt andRobert Lowry (1994) consider the bargain-ing game that will ensue and its conse-quences. For the budgetary process, thegovernor typically has the power to propose,with the legislature having the power toamend or even throw out the budget.Although their main focus is on the federallevel, the whole question of how dividedgovernment affects policy outcomes is alsotackled by Alesina and Howard Rosenthal(1994). From the point of view of voters, di-viding control may provide a means of get-ting an outcome that is intermediate be-tween the policy outcomes offered by eitherparty under unified control. Once the possi-bility of divided government is admitted,one needs to be careful about how to mea-sure party control in a given state, which au-thors have approached in a number of ways.

Beyond the questions of measurement liethe issues of endogeneity and omitted vari-able bias. Equation (3) represents the effectof party strategic choices (platforms andcandidate choices) on election outcomes,with party choices taken as fixed. These aredetermined in equilibrium, however, and

the empirical analysis needs to control for allfactors that affect election outcomes and thestrategic choices made by parties (such asparty platforms) if those factors are them-selves determinants of the policies chosen in(1). For example, a recession in a state couldlead to more Democrats being elected in (3)and a larger demand for transfers in (1).Again this may render problematic the inter-pretation of the effect of including measuresof party composition of the state houses in�st when estimating (2).

Similarly, if parties choose platforms tosuit the electorate within a state, but thesecampaign promises cannot be measured,then in effect the variables Θst and χjst enterthe error term in (2). But from (3), we ex-pect these variables to affect the composi-tion of the legislature. Hence �st is corre-lated with the error in (2), biasing inferencesabout the effect of parties on policy out-comes. If voters’ and party members’ tastes,Θst and χjst, are largely stable over time, theinclusion of state fixed effects may be ade-quate to deal with this problem. Otherwise,it is important to see what happens to esti-mation results when parties’ ideologiesand/or public opinion are added as controlsin (2).

One to another, papers in the literaturecome to very different conclusions on thesignificance of party control. This is perhapsnot surprising, given that the studies use dif-ferent estimation strategies, focus on differ-ent outcome variables over varying periodsof time, and use different measures of partycontrol while including different state spe-cific controls. Much of the early literaturefinds little evidence that party control mat-ters—Richard Winters (1976) provides asurvey. Winters (1976) finds little effect ofparty controls (party of the governor, per-centage Democrat in state lower house, anda competition index) on benefits and taxesborne by low income groups using pooledtime-series cross-sectional data for 1961 and1965. Thomas Dye (1984) runs a series ofstate specific time-series regressions for the


Article 1 2/21/03 4:17 PM Page 38

period 1950–80 that examine the effect of achange in party control on state welfarespending. Allowing for divided control of thestate legislatures and controlling for state in-come, he finds an effect of party control infewer than half the states. Robert Plotnickand Winters (1985) look at the empirical de-termination of AFDC benefits using a five-equation structural model that treats partycontrol as endogenous (identified by sup-posing that voter preferences affect policyonly through their effects on party out-comes), and also find little role for partycontrol. James Garand (1988) considers anumber of explanations for the growth ofstate government in the post war period,among them the idea that party control (byDemocrats) is associated with faster growthof government. He runs separate time seriesregressions for each state during the period1945–84, in which the dependent variable isthe level of state government spending as aproportion of total state income, and partycontrols include the party of the governor,the party controlling the legislature andthese two interacted. He finds little evi-dence for the importance of parties. ThomasGilligan and John Matsusaka (1995) find lit-tle effect of party control (number of seats inlower and upper houses, control over legisla-tures and the governor’s chair, competitive-ness of legislative elections) on state andlocal general expenditures.

In contrast, recent work finds support forparty control. Brian Knight (2000) finds thatcontrol of both houses by the Democratsleads to significantly higher tax rates relativeto state GDP, and that control of bothhouses by the Republicans has the oppositeeffect. Colleen Grogan (1994) finds thatparty control (Republican control of bothhouses and the governor’s chair) matters forMedicaid spending in random effects mod-els, for a biannual panel of states between1979 and 1989. Besley and Case (1995b)find that Democratic and Republican gover-nors respond differently to term limits, withDemocrats more inclined to raise taxes and

spending when working under a term limit.Diane Lim Rogers and John Rogers (2000)find that Democrat control in the house isassociated with larger government, meas-ured in either revenue or expenditure terms.They also find that divided government is animportant moderating device, with a Dem-ocratic governor and a Democratic houseleading to larger government than aRepublican governor with a Republicanhouse.

Divided government is also consideredby Alt and Lowry (1994), who are motivatedby a theoretical analysis in which states withdivided governments find it more difficultto respond to a shock, with the end resultbeing an increase in deficit finance. Usingdata from 1968–87, they run a two-equationsystem for taxes and expenditure in separateanalyses for groups of states, treating theSouth separately. Overall, they find deficitsmore likely under divided government, andgreater Democrat representation associatedwith higher public spending. Alt and Lowry(2000) use the difference in public spendingfrom year-to-year as their dependent vari-able and study the impact of divided gov-ernment for 33 states over the period1952–95. They find significant differencesbetween parties in their adjustment to fiscalimbalance during their first two years in of-fice. In cases of divided control, they find asmaller shift in the policy direction of thegovernor’s party.

Wes Clarke (1998) considers the impact ofdivided government on the degree of conflictbetween the executive and legislature by ex-amining the difference between gubernato-rial recommendations and actual appropria-tions for twenty states from 1985–94 (6027observations). A larger percentage change tothe governor’s recommendation is taken asevidence of greater conflict. Estimating statefixed effect models, he finds that a unifiedlegislature and opposition governor yieldsgreater conflict and that a greater ideologicalspread and party system liberalism leads togreater conflict.


Article 1 2/21/03 4:17 PM Page 39

We present a set of unified results on theimpact of party control in table 8. Here, weestimate the impact of control on a set ofoutcomes (taxes and expenditures), using aconsistent set of controls (state income, pop-ulation, and demographic variables), in afixed effect framework for the time period1950–98. We find, controlling for state fixedeffects and year effects and time-varyingstate level controls, that the higher is thefraction of seats held by Democrats in thestates’ lower house, all else equal, the higherare state taxes and spending per capita. Onaverage, and all else held equal, a ten per-centage point increase in the fraction ofseats held by Democrats in both the lowerand upper houses is associated with an in-crease in overall state spending per capita of$10 in 1982 dollars. Roughly a third of thatincrease is attributable to higher spendingon family assistance. F-tests in row 3 of table8 suggest that the fractions of Democrats inthe lower and upper houses jointly signifi-cantly affect total taxes and spending percapita, and the distribution of spending infavor of family assistance.

Table 8 also shows that Democratic con-trol of both the lower and upper houses ofthe legislature is associated with significantlyhigher taxes (roughly $13 per capita) and aredistribution of state spending in favor offamily assistance: overall spending per capitadoes not change significantly, but spendingon family assistance per capita increases byalmost $4 in 1982 dollars. This is consistentwith the idea that the exact structure of po-litical control is important and that dividedcontrol of the legislature does provide acheck on policy. We find little evidence ofDemocratic governors spending more over-all. However, Democratic governors on av-erage increase the generosity of state work-ers’ compensation programs, increasingspending on the program by over $2 percapita. We also find that greater party com-petition in the legislature is associated withsignificantly lower taxes, and significantlylower spending on workers’ compensation.

These indicators of party control and legisla-tive composition are jointly highly significantin our fiscal policy regressions. (F-tests aregiven at the bottom of each regression col-umn.)

One reason why we find such large andsignificant effects when other researchers donot is that our sample is much larger.Gilligan and Matsusaka use a similarmethodology to that used here, but havedata only at five-year intervals between 1960and 1990. Some of the research discussedabove analyzed states one at a time. Thislimits the number of observations to fewerthan fifty per state. When we analyze statesone at a time, regressing total taxes percapita on state income per capita, state pop-ulation, a year trend and the party controlvariables presented in table 8, we find theparty control variables jointly significant in33 of 47 states (Nebraska is excluded).Researchers writing in the mid-1970s by ne-cessity worked with samples half the size wehave available. When we restrict our analysisto the years from 1950 to 1979, we find partycontrol variables jointly significant in only 21of 46 states (Nebraska and Minnesota areexcluded).

6.1.1 Controlling for Ideology and PartyHeterogeneity

Erikson, Wright, and McIver (1989) arguethat the failure of much of the literature tofind a link between party control and policyoutcomes is due to inadequately controllingfor the ideological differences in partystances across states (a problem somewhatameliorated by the inclusion of state fixedeffects in many regressions). They see thelink as being from liberal public opinion toliberal party elites to liberal policy. Thus, instates with more liberal electorates, bothparties will be more liberal and the effect ofparty control is muted if public opinion isomitted when studying this relationship. Ingeneral, they find that the relationship be-tween opinion liberalism and Democraticcontrol is quite weak. They operationalize


Article 1 2/21/03 4:17 PM Page 40

the study of party elite liberalism by lookingat the conservatism of congressional candi-dates, local party chairmen, national conven-tion delegates and state legislators. Publicopinion is measured using CBS/New YorkTimes surveys from the period 1976–82.Using a range of measures of policy liberal-ism in key areas, such as AFDC and EqualRights Amendment ratification, they find anegative correlation between Democrats’policy liberalism and Democratic strength inthe legislature after controlling for liberal-ism in public opinion. While provocative,

their empirical models use only small num-bers of observations, and cannot includestate fixed effects.32

Robert Brown (1995) disaggregates partysupport among different sub-groups to re-flect the different cleavages between the par-ties that dominate in different states. He usespolling data to show that there are distinctdifferences in partisan support among socio-


TABLE 8LEGISLATIVE COMPOSITION AND POLICY CHOICE

Dependent Variable:

Total Family WorkersTotal taxes spending assistance compensationper capita per capita per capita per capita

Fraction Democrat in state lower house 78.71 101.38 28.78 −2.09(19.79) (33.33) (4.28) (4.56)

Fraction Democrat in state upper house 10.49 2.49 9.03 9.42(18.64) (31.50) (3.87) (4.29)

F-test: Coefficient on 4.15 3.06 8.30 2.23fraction upper house = fraction lower house (.0417) (.0802) (.0040) (.1359)

Indicator: Democrats control both lower 12.68 −1.99 3.88 −2.15and upper house (5.51) (9.36) (1.10) (1.26)

Indicator: Dem governor −5.79 4.56 −0.78 2.17(3.20) (5.39) (0.64) (0.73)

Party competition in legislature −101.13 29.72 4.40 −53.0(41.37) (70.10) (9.74) (9.39)

F-test: joint significance of party variables 11.92 3.43 31.19 10.98(p-value) (.0000) (.0043) (.0000) (.0000)

Years over which regression run all years all years all years all years1950–58 1950–58 1958, 1950–581960–97 1960–96 1960–98 1960–98


Notes: Standard errors in parentheses. All regressions control for year and state effects, and include controls forthe proportion of population aged 65 and above; the proportion of population aged 5 to 17; state income percapita in $1982 and income squared; and state population and population squared. Nebraska is removed from theanalysis, because it has a unicameral, non-partisan legislature. Observations for Minnesota are present only from1973 on.

32 Charles Barrilleaux (2000) provides further dis-cussion of these results.

Article 1 2/21/03 4:17 PM Page 41

demographic groups across states. Parties arethen characterized according to their domi-nant party cleavage. He suggests a three-wayclassification of states—a southern group ofstates where the Democrats are predomi-nately supported by black, low-income, andrural groups, a New Deal group withDemocratic support based on Catholic, low-income, union and female groups, and a Post-New Deal group with Democratic supportbased on black, Catholic, urban, union, low-income, female, and Jewish groups. He thenregresses welfare spending between 1976and 1985 on party control interacted with thetype of cleavage identified. However, he doesnot include state or year fixed effects. This ex-ercise suggests that Democratic party controldoes matter. However, it matters less in thePost-New Deal and Southern party cleavagestates.

Wright, Erikson, and McIver (1987) areamong those who have considered the directlinks between public opinion and policy out-comes. They find significant correlation be-tween state opinion and state policy whenusing their measure of ideology based on theCBS/New York Times surveys (above) to es-timate the cross-sectional relationship be-tween ideology and state policy choices for47 U.S. states.

We examine the relationship betweenstate policy and state opinion in table 9,where we present regression results on thedeterminants of state taxes and spending,controlling for state and year effects andtime-varying state-level controls. Our mea-sure of public opinion is the Berry et al.(1998) state citizens’ COPE score, which hasa range from 1.43 (South Carolina in 1965)to 96.99 (Massachusetts in 1988) and a me-dian of 50.83 (Minnesota in 1980). Thisanalysis is limited by the fact that the Berryet al. measure is available only for the yearsfrom 1960 to 1993. Nonetheless, we con-tinue to find that Democratic party variablesare significantly correlated with state taxesand spending. In addition, we find that hav-ing a liberal state citizenry, as measured

using the COPE score, is significantly corre-lated with state taxes, state spending overalland, in particular, state spending on familyassistance and workers compensation. Moreliberal states spend more on family assis-tance, and less on workers compensation.This is consistent with the idea that ideologyis important—possibly for the reasons iden-tified by Erikson, Wright, and McIver(1989).

Overall, the results in table 9 suggest thatthere is still information in the party identi-ties even after controlling for ideology. It isclear, however, that party identity and ide-ology measure somewhat different things indifferent states, depending on their politi-cal culture, history, and policy priorities.

6.1.2 Intensity of Party Competition

There are good theoretical reasons to be-lieve that a party’s behavior may depend onthe intensity of the challenge it faces fromthe other party in the state. Incentives to tar-get particular groups depend on how easilythis can swing an election, as in the theoreti-cal models of Lindbeck and Weibull (1993)or Dixit and Londregan (1998).

Whether these effects are important inpractice is an empirical matter, and U.S.states provide an interesting context inwhich to study it. Operationalizing this,however, requires some way of measuringinter-party competition. Measures of com-petition tend to be functions of vote and seatshares in the upper and lower houses.Information about the share of the vote re-ceived by the governor may also be relevant,as well as whether control of the legislatureis unified or divided. Thomas Holbrook andEmily van Dunk (1993) review some of thealternatives that have been used.

The most common measure of competi-tion used in the political science literature isthe Ranney index. This is generated by aver-aging together the proportion of seats wonby Democrats in the state house and senateelections along with the Democratic per-centage in the gubernatorial election, and


Article 1 2/21/03 4:17 PM Page 42

the percentage of time that the governorshipand the state legislature were controlled bythe Democratic party. This is readily com-puted using state-level data. The measurethat we use here is more limited, beingbased only on the share of seats held by eachparty in the upper and lower houses of thestate legislature. Other measures can bebased on more disaggregated data, such asthose used by Holbrook and van Dunk

(1993), which rely on the percentage of thevotes won by the winning candidate and thewinning candidate’s margin of victory ineach district.

Rogers and Rogers (2000) examinewhether party competition in gubernatorialraces, measured as the percentage of thevotes won by the current governor in themost recent election, is related to growth inthe size of government. They acknowledge


TABLE 9IDEOLOGY AND POLICY CHOICE

Dependent Variable:

Total Family WorkersTotal taxes spending assistance compensationper capita per capita per capita per capita

State citizens’ COPE score .400 .545 .201 −.119(.188) (.3310) (.038) (.050)

Fraction Democrat in state lower house 53.15 87.95 26.8 9.17(22.12) (39.39) (4.56) (5.97)

Fraction Democrat in state upper house 49.02 47.05 12.6 −1.46(19.90) (35.41) (4.10) (5.37)

F-test: Coefficient on 0.01 0.43 3.82 1.25fraction upper house = lower house (.9064) (.5137) (.0508) (.2629)

Indicator: Democrats control both lower 1.08 −11.74 3.00 −1.88and upper house (5.69) (10.16) (1.18) (1.54)

Indicator: Dem governor −1.01 −2.35 −.314 1.21(3.23) (5.74) (.664) (0.87)

Party competition in legislature 71.28 149.05 −5.78 −76.8(54.75) (96.78) (11.2) (14.7)

F-test: all political variables included 5.53 2.33 27.01 8.62(.0000) (.04040) (.0000) (.0000)



Notes: Standard errors in parentheses. The F-tests in row 8 are for the joint significance of fraction Democrat inlower house, fraction Democrat in upper house, an indicator that the Democrats control both houses, and indica-tor for Democratic governor, and our measure of party competition in the legislature. All regressions control foryear and state effects, and include controls for the proportion of population aged 65 and above; the proportion ofpopulation aged 5 to 17; state income per capita in $1982 and income squared; and state population and popula-tion squared. Nebraska is removed from the analysis, because it has a unicameral, non-partisan legislature.Observations for Minnesota are present only from 1973 on. All dollar-denominated variables are in 1982 dollars.

Article 1 2/21/03 4:17 PM Page 43

that there is no necessary theoretical linkbetween government size and intensity ofcompetition—it seems just as likely thatthere would be tax cuts as expenditure in-creases—and they do not put forward atheoretical framework to analyze this. Intheir study, government size is measured inboth revenue and expenditure terms. Usingyear effects and state effects and panel datafrom 1950–90, they find no positive link be-tween government size and political com-petition. The coefficient is either not signif-icantly different from zero or negative andsignificant—although the latter, wherefound, is sensitive to the sample periodchosen.

Table 8 presents results on the impact ofcompetition, using a measure based on dis-tance from 0.5 in the fraction of seats heldby Democrats in the two houses. We findsignificant effects of party competition ontotal taxes per capita and for workers com-pensation, both of which are significantlylower, the greater is party competition.These results require more investigation.Intense competition may change the policypriorities of the elected representatives, orthe policy stances that the party adopts, orthe types of candidates that it fields. In addi-tion, competition may influence legislativecoalition formation and bargaining in thelegislature. More research is needed to un-derstand the mechanisms at work behindour results.

Competitiveness measures based on polit-ical outcomes are easy to compute fromavailable data. However, they are likely to bejointly determined with the policies chosenby state governments. Suppose that a partyin a particular state has little chance of beingelected if it tries to follow policy platformsand candidate strategies that appeal to theparty faithful (perhaps best proxied by χst).This may induce parties to strategicallychange Xjst and cjst to please the voters, as cap-tured by equation (3). But then the outcomebased measures of competition and policieschosen are jointly determined.

Measures of competition, based either onthe number of seats held by the parties oron the vote shares for candidates within dis-tricts, may be poor measures of underlyingvoter preferences in favor of a particularparty. As discussed above, parties maychange their platforms or candidates to ap-peal to voters. In addition, a dominant partymay use redistricting to provide the partywith small margins of victory in the largestnumber of districts possible. Measurementerror in the competition variables used willbias estimates of the impact of competitionon policy choice. More work is needed here;these concerns have not been addressed inthe existing empirical literature, and it ishard to know whether they impose seriousdifficulties in practical applications. Itseems likely that progress could be made ifdata on Θst and χjst could be brought intothe analysis.

6.2 Non-Party Identity of Representatives

Legislators are characterized by morethan just their party label. Of particular issuehas been the extent of involvement in thepolitical process by women and minorities,not just as voters, but also as legislators. Forthis to have a fundamental effect on politicalrepresentation, it must be the case that citi-zens find it necessary to delegate authorityto citizens of particular kinds to further theirpolicy ends.

In principle, the effects of legislatureidentity on policy outcomes can be studiedby including this information in �st whenstudying the impact of composition on out-comes. In theoretical terms a process oflegislative bargaining in which policy pref-erences of representatives matter wouldsuggest that policy stances of representa-tives should matter (see, for example,Daniel Diermeier and Antonio Merlo2000). However, the literature investigat-ing these issues in cross-state data is quitethin.

The presence of women in state legisla-tures appears to have had a significant effect


Article 1 2/21/03 4:17 PM Page 44

on state policy. The Center for the AmericanWoman and Politics finds that women on av-erage give higher priority to policies relatedto children, families, and health care. SueThomas (1994), in a study of twelve state leg-islatures, finds that women spend more timeand effort on bills related to family issues.Case (1998), using U.S. state panel data from1978–91, documents the extent to which astate’s child-support enforcement policiestightened as the number of women legislatorsin the state grew. Controlling for state fixedeffects and year effects, and time-varyingstate-level economic variables, Case findsthat the number of women legislators is sig-nificantly correlated with the passage of sev-eral child support laws. Besley and Case(2000), using state panel data from 1975–88,find that the fraction of women in state upperand lower houses are highly significant pre-dictors of state workers compensation policy.

We examine whether women in statelegislatures have a significant effect on pol-icy making in table 10. We focus on a num-ber of policy variables that researchers haveposited are sensitive to the gender composi-tion of the legislature (Thomas 1994), in par-ticular on family assistance per capita, andpolicies related to child-support enforce-ment—here, withholding of child supportimmediately after the nonresident parentbecomes delinquent in payments, and theability to establish paternity to the child’seighteenth birthday. We find, controlling forstate and year effects and time-varying state-level controls, with or without controls forthe ideology of state citizenry, that women inthe legislature apply pressure to increasefamily assistance and to strengthen child-support laws. These results add to the grow-ing evidence that the identity of representa-tives affects policy choice.

Given the fairly rich biographical infor-mation now becoming available on electedrepresentatives, there is scope for furtherwork that looks at how characteristics of leg-islators affect policies. We anticipate growthin this area of research in the future.

7. Policy and Institutions

In this section we look at the effect of po-litical institutions on policy outcomes di-rectly. For the most part, we interpret theseas the direct policy consequences of an insti-tution—for example, when possession of theline item veto increases the bargainingpower of the state governor. There may alsobe important indirect effects—as when in-stitutions change the ideology, composition,and/or control of the state legislature alongthe lines discussed in the previous section.This latter possibility and its implications arediscussed in section 7.6.

7.1 Direct Democracy

At the present time, 23 U.S. states have aprovision for some form of direct democ-racy, typically through an initiative processwhereby citizens can place ballot proposi-tions, which are voted on subsequently.There is now widespread interest in thistype of institution and its effects on the pol-icy process.33

Our partition of institutions in section 3allowed for initiatives to appear both as partof I1st in equation (1) and in I2st in equation(3). The first of these is motivated by themodel of Gerber (1996), which consideredhow, given a set of policy preferences in alegislature, the availability of the initiativecould lead to a change in the equilibriumpolicy. Moreover, the legislature may makesuch a change preemptively if legislators an-ticipate the possibility of an initiative atsome later date. Hence, the possibility of ini-tiatives forces greater agreement betweenvoter preferences and policy outcomes, as-suming that representatives elected to thelegislature have views that are out of stepwith the citizens at large.

Similar conclusions follow from the theo-retical analysis of Besley and Coate (2000b)


33 See David Magleby (1984) and Shawn Bowler,Todd Donavan, and Caroline Tolbert (1998) for back-ground discussion.

Article 1 2/21/03 4:17 PM Page 45

but for quite different reasons. They de-velop a model in which initiatives affectelectoral outcomes. This also motivateswhy the availability of citizens’ initiativesappears in equation (3). They argue initia-tives have an impact via issue unbundling.In general elections, many issues are de-cided at once, which may result in issuesthat are not electorally salient being dis-torted away from what a majority desire.34

Initiatives allow such issues to be unbun-dled from the rest as voters can have a di-rect say on them. Besley and Coate (2000b)show that this can change the probabilitydistribution of a range of policy outcomesand the composition of candidates who arechosen to run.

Both of these theoretical approaches, aswell as many popular discussions of initia-tives, imply that citizens’ initiatives are adevice for bringing policy into line withpublic opinion.35 Gerber (1996) and Besleyand Coate (2000b) both argue that the ef-fect of an initiative can be felt even if aninitiative is not actually called. For empiri-cal purposes, this suggests that the actualconduct of initiatives in states that havethem need not be a very good indicator oftheir influence. Hence, it does make senseto study the availability, rather than use, of


TABLE 10WOMEN’S LEGISLATIVE REPRESENTATION AND POLICY CHOICE

Dependent Variable:

Child support: Child support:Family assistance per immediate withholding paternity establishment

capita ($1982) upon delinquency to age 18

Fraction female state lower house .025 .038 −.053 −.321 −.369 −.555(.008) (.009) (.237) (.311) (.236) (.310)

Fraction female state upper house −.006 −.011 .712 1.06 .311 .883(.006) (.008) (.188) (0.26) (.187) (.255)

F-test joint significance female 4.85 8.96 7.21 8.62 2.26 6.71representation (p-value) (.0000) (.0000) (.0000) (.0002) (.1044) (.0000)

State citizens’ COPE score — .000 — −.000 — .001(.0000) (.001) (.001)

Years over which regression run all years all years all years all years all years all years1975–98 1975–93 1975–97 1975–93 1975–97 1975–93

Number of observations 1152 912 1104 912 1104 912

Notes: Standard errors in parentheses. All regressions control for year and state effects, and include controls forthe proportion of population aged 65 and above; the proportion of population aged 5 to 17; state income percapita in $1982 and income squared; and state population and population squared. All dollar-denominated variables are in 1982 dollars.

34 A large body of empirical evidence from politicalscience supports the lack of congruence of policy andvoter preferences on a variety of issues—Besley andCoate (2000b) provide references.

35 This is not universally believed. Some, such asDavid Broder (2000), are concerned that voters areeasily influenced by slick advertising campaigns, inwhich case the initiative process in the United Statesmay actually enhance the power of special interests.From a welfare point of view, there is also the concernthat initiatives will lead to minorities being unfairly tar-geted and with citizens being forced to choose onissues on which they are ill-informed.

Article 1 2/21/03 4:17 PM Page 46

the initiative when looking for its policyimpact.36

One strand of empirical literature on ini-tiatives has used data from U.S. states to testwhether public opinion and policy outcomesare closer together in initiative states. Forexample, drawing on the work of Wright,Erikson and McIver (1987), EdwardLascher, Michael Hagen, and Steve Rochlin(1996) and John Camobreco (1998) investi-gate whether the link between aggregatemeasures of policy outcomes and publicopinion is closer when states allow citizens’initiatives. They find no significant effect.37

With respect to specific policy issues,Gerber (1999) finds that policy outcomes onthe death penalty and abortion regulationare closer to public opinion in states thatpermit citizens’ initiatives, even thoughthese policies are not directly determinedvia initiatives. Gerber uses cross-sectionalstate variation from the 1990s and comparesstances on an array of policies. She finds sig-nificant differences (at the 10 percent level)for personal income taxes (initiative stateslower); highway, natural resources and hos-pital spending (initiative states higher in allcases); and the implementation of “threestrikes” legislation (initiative states lower).Gerber looks in greater detail at the deathpenalty and parental consent laws for abor-tion, using public opinion data to estimatemedian voter preferences. With cross-sectional data for 1988, 1990, and 1992, she

runs a logistic regression that interactswhether a state has an initiative with publicopinion, and finds that states with initiativesmirror public opinion more closely.

Other studies have focused more onwhether initiatives are devices for reducingthe size of government. There is a wide-spread belief that agency problems lead togovernment that is too large, and that gov-ernment can be reigned in by initiatives.This is the essence of the theoretical ap-proach of Arthur Denzau, Robert Mackay,and Caroly Weaver (1981).

In this spirit, Jeffrey Zax (1989) investi-gates how access to initiatives affects stateexpenditures per capita, in a cross-section offifty states for 1980. Contrary to the ideathat initiatives promote smaller government,he finds that state spending is significantlyhigher in states that permit direct statutoryinitiatives. Paul Farnham (1990) estimatesthe cross-sectional effect of citizens’ initia-tives and referenda using data on 735 com-munities, taking the log of community ex-penditures as the dependent variable,finding little or no evidence that access tothe initiative is important.

Matsusaka (1995) is the most thoroughcross-sectional study of the effect of initia-tives on cross-state taxing and spending. Heregresses government expenditures and rev-enues in 49 states (Alaska is excluded) on anumber of control variables for a panel ofstates sampled over a thirty-year period atfive-year intervals from 1960 to 1990. He in-cludes year effects, but not state fixed ef-fects, since the presence of initiatives islargely fixed within states over time. Hefinds a strong negative effect of access to theinitiative on expenditures. He also findssome evidence that the effect is strongestwhen the number of signatures required toget an initiative on the ballot is low. This is intune with the idea that initiatives may have arole to play in reducing agency problemsthat result in the state being too large.However, the cross-sectional nature of theidentification strategy raises the issue of


36 There are a number of other theoretical contribu-tions on initiatives. These include Matsusaka (1992),which discusses what kinds of issues will be decided byinitiative (as opposed to elected representatives). Heargues that controversial issues that are not too techni-cal will tend to be tackled via initiative—largely to al-low legislators to avoid making decisions on controver-sial issues. Technical issues are often too complicatedto be settled via initiative. Denzau, Mackay, andWeaver (1981) make a similar point concerning theability of initiatives to constrain agenda-setting politi-cians with non-majoritarian preferences. Matsusakaand McCarty (1999) emphasize the possibility thatholding an initiative can reveal information to legisla-tors about policy preferences.

37 Matsusaka (2000) provides a persuasive critique oftheir methodology.

Article 1 2/21/03 4:17 PM Page 47

whether some other state characteristic, cor-related with initiatives, is driving the results.Dale Bails and Margie Tieslau (2000) pre-sent somewhat similar results running a ran-dom effects panel data regression of total ex-penditures on initiatives for the period1969–94. They also find a negative and sig-nificant effect of a state permitting aninitiative.38

This variety of findings makes it interest-ing to take a fresh look at the issue. Wepresent results on the impact of citizen ini-tiatives on state spending and revenues intable 11. We regress state total taxes, stateincome taxes, total government spending,and spending on family assistance pro-grams on an indicator that the state allowscitizen initiatives, with controls for year ef-fects and for state income per capita and in-come squared, state population and popu-lation squared, the proportion of stateresidents aged 65 and older, and the pro-portion aged five to seventeen. LikeMatsusaka (1995), we do not control forstate fixed effects, because only four states(FL, IL, MS, and WY) changed their policyon initiatives during the 38-year periodfrom 1960 to 1997 inclusive. In table 11 wepresent three estimates of the impact of ini-tiatives on state fiscal outcomes. In the firstpanel, we estimate OLS regressions withrobust standard errors, allowing for an un-specified pattern of correlation betweenthe unobservables from the same state overthis period. The estimation procedure con-structs a 38-by-38 variance-covariance ma-trix for each state, which may lead to a lackof precision in the standard error estimates.Using this estimation procedure, we findno significant effect of citizens’ initiativeson state revenues or state spending. In thesecond panel, we estimate the impact ofinitiatives using a random effects specifica-tion. Here the estimates suggest that state

tax revenues per capita—and income taxrevenues in particular—are significantlylower in states that have citizens’ initiatives.When we use a between-state estimator inpanel three, in which regressions are runon state means, the between-state esti-mates, like the robust estimates in panelone, suggest that the standard errors arelarge—and too large to say conclusivelythat state initiatives have held down statetaxes and spending. Note that if nothingchanged within the state from year to year,and we estimated OLS regressions using 38years’ worth of data from each state, thestandard errors would fall by a factor of6 :√

38 = 6.16.Estimating the effect ofstate initiatives using only the state meansignores information that can make our esti-mates more precise, and is therefore lessthan ideal. However, the fact that the ob-servations from a given state each repli-cates roughly the same information everyyear makes it difficult to interpret thesefindings. In no specification presented intable 11 are initiatives significantly corre-lated with total government spending percapita, or with family assistance spending.

Using a random effects specification, wefind a negative correlation between initia-tives and taxes. However, results presentedin table 11 illustrate the great difficulties inproviding reliable estimates when institu-tions vary only across states, and not withinstates over time. Panel data are only oflimited value in this context.

7.2 Electoral Accountability

An important issue is how voters holdpoliticians to account for their perform-ance while in office. Folk wisdom suggeststhat deterioration in economic perform-ance, and tax increases, in particular, arenot conducive to electoral success. Thisclaim was made forcefully by SamPeltzman (1992) in his assessment of gu-bernatorial electoral chances. He arguesthat permanent increases in the size of


38 In related evidence, Werner Pommerehne (1990)shows that Swiss cantons that use the initiative havesmaller state governments.

Article 1 2/21/03 4:17 PM Page 48

government are punished, with spendingbeing the “primary bad.”39

In section 3.1.3 above, we discussed howempirical models of elections can incorpo-rate this possibility, through the addition ofhistory Hst into estimating equations. Froma theoretical point of view, these claims arebest justified in political agency modelswhere there is private information about anincumbent’s type (representing his compe-tence or his willingness to consume rents atthe citizens’ expense) or there is uncer-

tainty about the true state of public fi-nances. Models along these lines were firstdeveloped by Barro (1973) and Ferejohn(1986). It is straightforward to see how theycan generate an aversion to tax increases, ifthe latter are correlated with greater in-competence or greater likelihood of rent-seeking behavior.

A key feature of these models is that vot-ers will condition their voting decision onincumbent behavior, either to curb moralhazard problems or else to sort in politicianswith desirable characteristics. Hence, wewould expect to see voters punish indicatorsof poor effort, and reward the opposite.Electoral accountability is in large measure


TABLE 11CITIZENS’ INITIATIVES AND STATE POLICY CHOICES

Dependent Variable:

Total Government FamilyTotal taxes income taxes spending Assistanceper capita per capita per capita per capita

OLS with robust standard errorsIndicator: State Allows Citizens’ Initiatives −30.78 −34.02 −35.00 −.995

(30.22) (32.96) (51.34) (4.69)

Random effects models

Indicator: State Allows Citizens’ Initiatives −38.40 −51.98 20.57 −1.76(11.82) (9.62) (19.25) (2.11)



Regression on state means

Indicator: State Allows Citizens’ Initiatives −28.50 −45.83 −74.42 −1.14(36.11) (37.20) (58.83) (5.26)


Notes: Standard errors in parentheses. All regressions include year indicators and controls for the proportion ofpopulation aged 65 and above; the proportion of population aged 5 to 17; state income per capita in $1982 and in-come squared; and state population and population squared. We do not include state fixed effects because only 4states changed whether they allowed initiatives over the period 1960 to 1998. (These were: FL 1972, IL 1971, MS1992, and WY 1968). For all regressions in panel one, we estimate robust standard errors, and allow for an unspec-ified pattern of correlation in the unobservables from the same state. Panel two allows for state random effects.Panel three estimates between state regressions on state means.

39 There is a long-standing tradition of studying thedeterminants of retrospective voting in U.S. nationalelections. See, for example, Morris Fiorina (1981).

Article 1 2/21/03 4:17 PM Page 49

about reputation formation, which can bebuilt around the historical record of partiesand governors. Besley and Case (1995b) ar-gue that whether governors are working un-der term limits provides a test of agencymodels, as the governors’ time horizons varyexogenously with term limits. We discusstheir results in greater detail below.

Relative performance may also be an im-portant determinant of electoral success.Besley and Case (1995a) develop a model inwhich voters use cross-state comparisons ofpolicy outcomes to evaluate office holders,which in turn generates yardstick competi-tion between incumbents. In a world whereneighboring jurisdictions face correlatedshocks, information about one’s own statecan be gleaned through observation of thepolicy decisions taken by another. In such aworld voters would be rational to use rela-tive performance comparisons. This will, inturn, make tax setting decisions acrossneighboring states interdependent. In termsof electoral accountability, we should expectto see incumbent success correlated withneighboring states’ policies and economicconditions, as well as one’s own.40

The idea that voters respond to changes inthe economy is widely documented in a vari-ety of empirical contexts—Peter Nannestadand Martin Paldam (1994) review the litera-ture. For the U.S. states, this means linkingvoting outcomes to general indicators ofstate-level economic health. John Chubb(1988) considers the determinants of stateelections as a function of the performance ofthe state economy and other factors usingdata from 1940–82. He finds very little evi-dence that changes in state income levels af-fect election outcomes. These results areconsistent with the findings of James Adams

and Kenny (1989) and Peltzman (1987) whoboth show that whether a government is re-elected is not correlated with economicgrowth in the state in question. However,voters may not view state income per capitaas being strongly affected by unobservableincumbent effort or type. This line of rea-soning ties into the large literature on politi-cal budget cycles where the most persuasivetheoretical contributions, such as KennethRogoff (1990), use electoral accountabilitybased on incomplete information as thefoundation. Cross-country evidence, such asthat developed in Alesina and NourielRoubini (1992), finds only limited evidencein support of the existence of such cycles.

Table 12 looks at this in the context of theU.S. states, where we focus exclusively on thestates that have a four year electoral cycle forthe sake of easy comparison. The data do notshow any marked political business cycle in ei-ther state income or unemployment. This isconsistent with Chubb’s observation that theseare not key indicators for which incumbentsare held accountable at the state level.Arguably, the openness of the U.S. statesmakes it very difficult for individual state gov-ernments to have much impact on state levelunemployment. Seen in this light, the insignif-icant effects observed are not surprising.

A more promising approach is to look forevidence of electoral accountability as a func-tion of policy variables. In principle, tests forwhether electoral success depends on pastpolicy choices are straightforward. As dis-cussed in section 3.1.3 above, for governors,electoral success can be represented as a dis-crete variable measuring whether the gover-nor is reelected, as in Besley and Case(1995a), or as the percentage of the vote go-ing to the dominant party, as in Lowry, Alt,and Karen Ferree (1998). The latter also con-tend that the ruling party will be held togreater account when there is unified ratherthan divided control of state offices.Furthermore, accountability should be morepronounced in gubernatorial elections, wherea single agent can be blamed, rather than in


40 This idea can be used to underpin the observationthat states appear to make interdependent fiscal deci-sions in a variety of contexts as suggested by Case,James Hines and Harvey Rosen (1993). Jack Walker(1969) applied this idea to the diffusion of policy inno-vation across states—see also Frances Stokes Berry andBerry (1992) and Koleman Strumpf (2000).

Article 1 2/21/03 4:17 PM Page 50

legislative elections where blame is harder toattribute.

Lowry, Alt, and Ferree (1998) find thatthere are fewer votes for incumbents who ex-perience a shock when there is unified ratherthan divided government, and that the effectis larger in gubernatorial elections. In theirstudy, voters also respond to the differencebetween state income growth and nationalincome growth. Susan Kone and Winters(1993), in pooled time series and cross-sectional regressions from 1957–85 that donot include year or state fixed effects, findthat Democratic governors are punished forputting up taxes. Besley and Case (1995a), ina model that includes year and state fixed ef-fects, find that a governor is more likely to bedefeated if he puts up taxes, but is morelikely to win if his geographic neighbors do.

Richard Niemi, Harold Stanley, andRonald Vogel (1995) also test for the impor-tance of tax increases on gubernatorial elec-tions using individual data from exit polls for34 states in 1986. They model the probability

that a respondent voted Republican in a par-ticular state as a function of respondent,state, and national variables. They allow theeffect of the state-level economic and policyvariables to vary according to whether the in-cumbent governor was a Democrat or aRepublican. Consistent with the results ofBesley and Case (1995a), they find that sup-port for the incumbent party falls when taxesare increased, and that state-level incomeappears to be an important determinant ofvoting decisions.

Justin Wolfers (2002) also considers the na-ture of gubernatorial electoral accountabilityin U.S. states. He shows that events beyondthe control of a governor (specifically oilprices) appear to be correlated with whetheror not the governor is reelected. He inter-prets this as irrational behavior by voters.41


TABLE 12POLITICAL BUSINESS CYCLES

Dependent Variable:

State income per capita State unemployment rate

Indicator: gubernatorial election in t + 1 −36.80 −5.41 16.48 −.122 .011 .012(197.53) (149.95) (54.73) (.145) (.154) (.113)

Indicator: gubernatorial election in t + 2 161.98 48.17 25.95 −.050 −.088 −.081(196.34) (120.60) (44.05) (.145) (.124) (.090)

Indicator: gubernatorial election in t + 3 108.00 26.44 7.07 .136 .061 .048(197.20) (149.74) (54.69) (.145) (.154) (.112)

F-test: joint significance of election 0.44 0.08 0.13 1.15 0.35 .051variables (p-value) (.7246) (.9710) (.9417) (.3270) (.7865) (.6756)

Year effects? no yes yes no yes yes

State effects? no no yes no no yes


Notes: Standard errors in parentheses.

41 If the management skills of a governor are more inevidence when there are good times rather than bad,then this would be consistent with rational updating onthe part of voters.

Article 1 2/21/03 4:17 PM Page 51

Overall, these results leave little doubtthat the electoral process does hold policymakers to account for the policies chosenduring their tenure. This cements the linkbetween economic and political outcomes ina way that is consistent with theories basedon political agency. The literature, however,has not looked in much detail at how institu-tions affect the process by which voters holdpoliticians to account. An exception is thework by Stratmann and Aparicio-Castillo(2001), who show that incumbent defeat inhouse elections is more likely when cam-paign restrictions are in place.

7.3 The Scope of Elective Office

While every U.S. state has an elected leg-islature (in almost all cases two) and a di-rectly elected chief executive, there aremarked differences in the extent to whichother kinds of state officers are directlyelected. Good examples of institutional vari-ation are public utility commissioners, whoare elected in eleven states, high courtjudges, who are elected in 23 states, and in-surance commissioners, who are elected intwelve states. In states that appoint officerholders, it is typically the state’s governorwho is key (although confirmation maysometimes be required in the legislature).

In terms of the theoretical structure laidout in section 3, institutions affecting whichpositions are elected or appointed may beconsidered part of either I1st or I2st. Havinga more independent judiciary (if indeed thisis achieved by appointment of judges) could,for example, affect judicial oversight of legis-lation, and hence constrain policy making ina legislature. In that case, such rules wouldenter as part of I1st. If �st is expanded to in-clude the characteristics of a range of stateoffice-holders, such as judges and regula-tors, then we might expect the pattern ofselection to enter as part of I2st in (3).

The latter possibility can be motivated bythe theoretical approach of Besley andCoate (2000a). They argue that direct elec-tion of office holders is a force for increasing

the salience of that dimension of policy. Inthe context of regulators, we should typicallyexpect this to move policy in a pro-consumerdirection. General elections in which regula-tory issues are bundled with other “bigger”policy issues enhance the power of specialinterests, party elites and single issue voters.Similar arguments would also apply to theelection of the judiciary.

If there is a move towards pro-consumerpolicies, then this may have important dy-namic implications. For example, in electric-ity, network investments have long lives andmay benefit from long-run regulatory com-mitments which may be harder to achieve ifregulators favor consumer interests.Similarly, judges may reach more short termpopular judgements if they face reelec-tion.42 To some degree, such incentive prob-lems can be mitigated by having longerterms in office. There is, however, a possibletrade-off between the degree of accounta-bility and the ability to commit.

Whether elected regulators behave differ-ently from those that are appointed has beenwidely studied in the empirical literature. Anumber of studies, including Berry (1979),Kenneth Costello (1984), Mark Crain andRobert McCormick (1984), Peter Harris andMalcolm Navarro (1983), Navarro (1982),and Walter Primeaux and Patrick Mann(1986), have looked at the evidence fromdifferent perspectives.43 Some of these con-tributions have looked at rate setting, whileothers have looked at broader indicators ofthe state’s regulatory climate. Costello(1984) reviews the evidence, and concludesthat the appointment or election of public-utility commissioners makes little difference


42 Eric Maskin and Jean Tirole (2001) offer an in-sightful model of the pros and cons of judicial inde-pendence. In their model, elected policy makers maybe inclined to “pander” to voters—do what is popularrather than what they know to be right.

43 The large empirical literature on the effects ofregulation in U.S. states begins with the seminal contri-bution of George Stigler and Claire Friedland (1962),and is expertly reviewed in Paul Joskow and NancyRose (1989).

Article 1 2/21/03 4:17 PM Page 52

to the average rate payer. The data and timeperiods used differ across studies, however,as do the set of controls and institutionalmeasures. More worrisome for convincingempirical testing is the fact that the litera-ture is predominantly cross-sectional, withparticular researchers choosing selectedyears and available controls to report theirfindings.

The more recent literature seems to bemore encouraging to the view that electingregulators makes a different. John Formby,Banamber Mishra, and Paul Thistle (1995)use data from 1979–83 on a selection ofinvestor-owned utilities and find that elec-tion of public utility commissioners has anegative effect on bond ratings, consistentwith a squeeze on margins due to morepro-consumer choices. Joseph Fields, LindaKlein, and James Sfiridis (1997) also findthat elected commissioners from the insur-ance industry are more pro-consumer—themarket value of life insurance companies do-ing business in California declined sharplyfollowing the passage of Proposition 103,which changed the method of selection ofthe insurance commissioner from appoint-ment to election. Using data from 1985,Susan Smart (1994) reports that telephonerates are lower in states that elect theirpublic utility commissioners.

Besley and Coate (2000a) exploit paneldata to look at electricity regulation. Theyfind a robust negative correlation betweenelecting regulators and the price tariff facedby consumers. However, this begs the ques-tion of whether the decision to elect or ap-point regulators is simply correlated withimportant unobservable differences be-tween states. They suggest a test, motivatedby theory, that prices should respond less tocost shocks when states elect their regula-tors, if this results in more pro-consumerregulation. They test this prediction usingdata on electricity prices from a panel ofU.S. states and find strong support for theidea that direct elections produce more pro-consumer regulators. One important feature

of this test is that, even though the institu-tion may be relatively fixed over time, thecomparative static refers to a variable thatvaries over states and time (the shock to thecost of producing electricity).44 Hence, theprediction can be tested even with the inclu-sion of state fixed effects.

There is a significant body of work on thedifference between elected and appointedjudges. In line with the theoretical discus-sion above, there is evidence that appointedjudges are more independent than electedjudges. As we discuss further below,Henning Bohn and Robert Inman (1996)find that whether a constitutional restrictionon deficit finance is effective depends onwhether the court that has to enforce the re-striction is elected or appointed. Restric-tions with appointed courts do not appeareffective in their data. Andrew Hanssen(2000) tests the idea that election leads togreater judicial independence by looking atstaffing levels in three budgetary agenciesthat are subject to judicial review: publicutility commissions, insurance commissionsand education bureaucracies. He argues thatthe kind of defensive activity that more inde-pendent judiciaries engage in will result inthem having more staff. Using cross-sectionaldata for 1983, he shows that states withelected judges have significantly smaller bu-reaucracies controlling for a number ofother observables. Hanssen (1999) looks atwhether states that elect their judges havemore or less litigation activity, arguing thatthis may reflect the degree of uncertainty inthe operation of courts. Using data from allfifty states, he tests whether there are signif-icantly more public-utility disputes(1978–83), and High Court and Trial Courtfilings (1985–94) in states that elect theirjudges. The main finding, identified fromcross-sectional differences, after controllingfor a number of economic and demographicvariables, is that appointing states have


44Cross-sectional variation comes from the differingproduction structures across states.

Article 1 2/21/03 4:17 PM Page 53

significantly higher rates of judicial activityin public-utility disputes and High Courtfilings, but not in Trial Court filings. Thereare also more public-utility commissionrulings in states that appoint their utilitycommissioners.

7.4 Term Limits

U.S. states also vary in whether they holdelected officials to term limits. Some kind ofterm limit is observed for governors inroughly half of the states, and a key issue iswhether we should expect such limits to af-fect policy outcomes. Besley and Case(1995b) consider a framework in which termlimits change the incentives of politicians tobuild a reputation. They base this on anagency model with imperfect information.When a term limit binds, there may be atemptation to play the end game, resultingin significant changes in policy.

One way to examine term limits is cross-sectionally, looking at the permanent differ-ences in policies in those states that havesuch limits. This is the approach in earlywork on the topic by Mark Crain and RobertTollison (1977, 1993), and Mark Crain andLisa Oakley (1995). Crain and Tollison(1977) make the interesting and importantpoint that if political office is a productiveasset, one used to produce political out-comes, then candidates for the office shouldbe willing to pay more for the opportunity toserve in states with longer terms, and instates without term limits. They find, usingcross-sectional data for races run in 1970,that challengers spend less money whenrunning for two two-year terms than dothose running for one four-year term. In ad-dition, challengers spend less in states withterm limits. Crain and Oakley (1995) exam-ine whether states that allow governors tosucceed themselves indefinitely have differ-ent public capital stocks and flows than dostates where governors are restricted bysome sort of term limit. They find, usingdata from the 1980s, and controlling for a

number of state institutions, that the stock ofstate government capital per capita, thechange in the stock, and the percentagechange in the stock, are all lower in stateswithout term limits. Bails and Tieslau (2000)argue that term limits should lower the rateof growth of spending, by making policymakers more responsive to citizens’ prefer-ences. They test for this using a random ef-fects model for the period 1969–94, andconfirm a negative coefficient on state ex-penditures. All of these results raise theusual issue of whether such limits are merelyproxying for omitted state level characteris-tics. State income per capita and state popu-lation are significantly lower in states withterm limits, to name but two important dif-ferences between states with and withoutterm limits (Besley and Case 1995b, tableIII, p. 778).

A second strand of the term limits litera-ture uses data on the behavior of represen-tatives in the U.S. Congress to predict howstate representatives facing term limitswould be expected to behave. Lott andStephen Bronars (1993) analyze congres-sional voting data from 1975–90, and findno significant change in voting patterns in arepresentative’s last term in office. It is farfrom clear, however, that congressionalrepresentatives who announce they arestepping aside provide an adequate pictureof the behavior of state governors who arebound by law not to run again for reelec-tion. A provocative paper on potential end-games in the U.S. Congress is provided byJohn McArthur and Stephen Marks (1988),who observe Congressional behavior in alame duck session of Congress: in post-election sessions, members who have notbeen reelected are at times called upon tovote on legislation before the swearing in ofthe new Congress. McArthur and Marksobserve that lame duck representativeswere significantly more likely in 1982 tovote against automobile domestic contentlegislation than were members who werereturning.


Article 1 2/21/03 4:17 PM Page 54

Besley and Case (1995b) identify the ef-fect of a term limit from the difference be-tween first and second terms in office for in-cumbents who face term limits. Controllingfor state fixed effects and year effects, andusing annual data from the 48 continentalU.S. states from 1950–86, they find that avariety of policy measures are affected byterm limits. Specifically, state taxes andspending are higher in the second termwhen term limits bind in states that havethem. Such limits tend to induce a fiscal cy-cle with states having lower taxes and spend-ing in the first gubernatorial term comparedto the second.

John List and Daniel Sturm (2001) applya similar methodology to cross-state varia-tion in environmental policy. Using data forthe period 1960–99, they find that governorsin their last term in office are significantlymore likely to spend resources on environ-mental protection. However, this term limiteffect is muted in states where a larger frac-tion of citizens belong to environmental or-ganizations. They also show that their termlimit effect varies according to the margin ofvictory in the gubernatorial race—withterm-limit effects being attenuated whenthe margin of victory is larger.

We update the results from Besley andCase (1995b) using data from 1950–97, andpresent the results in table 13. Controllingfor state and year effects and time-varyingstate-level variables, we continue to findthat in those years in which an incumbentgovernor cannot stand for reelection be-cause of a term limit, on average statespending per capita is significantly higher(roughly $15 per person higher in 1982 dol-lars). However, our earlier finding that taxesare higher when the governor is a lame duckdoes not hold in this longer panel. Totaltaxes per capita are lower, although not sig-nificantly so, on average in those states inwhich the current governor cannot stand forreelection.

We examine this further in figure 3, inwhich we display graphically the effect of

having a lame duck, year by year, from 1950to 1997, controlling for the same time-vary-ing variables used in the regressions pre-sented in table 13. We find that, over time,the impact of having an incumbent who can-not stand for reelection has changed frombeing on average positive and significant, inthe first half of the period, to being on aver-age negative and significant (and much morevariable year-to-year) in the second half. Wecan offer no simple explanation for this pat-tern. It continues to emerge when we con-trol for other time-varying state level vari-ables that we think may influencetaxes—such as the need for a super majorityto change taxes, the fraction Democrats inthe lower house of the legislature, theamount of competition between the partiesin the legislature, whether Democrats con-trol both houses of the legislature, and thelevel of grants from the federal government.We parameterize the changing pattern inthe impact of having a lame duck governorin regression results presented in the secondand fourth columns of table 13. We see that,for both taxes and state spending, governorsin the first half of our time period spent andtaxed more when they could not stand forreelection but that, over time, this phenom-enon has changed. It seems likely that someomitted variable is responsible for thechange in behavior observed for governorsworking under a term limit. This is an arearipe for future research.

The literature to date mostly treats termlimits as exogenous. However, there havebeen some changes in such limits over time.The major trend in the second half of thetwentieth century was the abandonment ofone-term limits, which survive today only inVirginia.45 An interesting unanswered ques-tion is whether this can be rationalized as awelfare improving change for voters, given


45 Adams and Kenny (1986) consider the optimalrole of term limits and correlate the presence of termlimits with the length of term in office. On the whole,term limits are more prevalent where lengths arelonger.

Article 1 2/21/03 4:17 PM Page 55

the incentive effects of term limits that havebeen uncovered.

7.5 Budgetary Institutions

Central to much of the public choice liter-ature is the idea that a fiscal constitutionshould limit the policy choices of electedrepresentatives—see, for example, Brennanand Buchanan (1985). In this section westudy a variety of budgetary institutions thataffect budgetary procedures, mostly in thedirection of greater fiscal conservatism.Stricter budgetary institutions are generallymotivated by the notion that governmentstend to be too large relative to voters’wishes. This view can be motivated either bythe Leviathan model of Brennan andBuchanan (1980) or a more sophisticatedagency view as in Barro (1973) and Ferejohn(1986). Daphne Kenyon and Karen Benker(1984) characterize the proliferation of suchmeasures as part of a broader “Tax Revolt” inthe 1970s, with California’s Proposition 13leading the way. It is interesting to note thata number of the restrictive budgetary insti-tutions that are now observed were imposedby citizens’ initiatives, fuelling the notion of

a popular rebellion against governmentprofligacy.

In terms of the theoretical structure ofsection 3, budgetary institutions are bestthought of as belonging to the vector I1st inequation (1). However, they would belong toI2st if they changed the kinds of representa-tives who were elected. The six main institu-tional categories that we discuss here are asfollows.

Tax and Expenditure Limitations. Thesefall into three broad categories: (i) indexedlimits on the growth of revenues or expendi-tures, for example, to the population growthrate; (ii) requirements that voters approveall new taxes; and (iii) supermajority re-quirements that require anywhere betweenthree-fifths and three-quarters of the legisla-ture to approve tax increases. There are 24states with indexed limits, thirteen allow anoverride with a supermajority vote, and fiverequire a simple majority if the governor hasdeclared a state of emergency. Kim Rueben(1997) gives a useful overview of the historyand content limitations in category (i) above.Half the states with such limitations restrictthe growth in state expenditures to the


TABLE 13BINDING TERM LIMITS AND POLICY CHOICE

Dependent Variable:

Total taxes per capita Total spending per capita

Indicator: Incumbent governor −6.40 1216.73 14.80 1968.81cannot stand for reelection (4.28) (514.32) (6.73) (820.91)

Indicator: Incumbent cannot — −0.619 — −0.990stand for reelection × year (0.260) (0.416)

Years over which regression all years all years all years all yearsrun 1950–97 1950–97 1950–96 1950–96


Notes: Standard errors in parentheses. All regressions control for year and state effects, and include controls forthe proportion of population aged 65 and above; the proportion of population aged 5 to 17; state income percapita in $1982, income squared and cubed; and state population, population squared and cubed. All dollar-denominated variables are in 1982 dollars.

Article 1 2/21/03 4:17 PM Page 56

growth rate in personal income averagedover some previous period. Five others re-strict the size of appropriations to a specifiedpercentage of state income, while four oth-ers restrict growth to an index of populationgrowth and inflation. Three other states re-strict the absolute expenditure growth rate.Spending on capital projects is excluded, asare federally funded projects. Half of thelimits in place are constitutional, with the re-mainder being statutory. Most tax and ex-penditure limitations were introduced in the1970s, which many believe reflected a gen-eral disillusionment with government and aview that spending was out of tune withwhat a majority of voters preferred.46

Balanced Budget Requirements andLimited Carry-over Provision. The Reagan

era of budget deficits ushered in a period ofheightened concern about the causes andconsequences of budget deficits. This fil-tered into state politics too, with manystates passing measures restricting the abil-ity of elected officials to use deficit finance.Indeed, all states, with the exception ofVermont, have some measure of this form.There are, however, differences in the wayin which the laws are structured (see JamesPoterba 1997). The weakest form of restric-tion requires that the governor submit a bal-anced budget (44 states). This need not leadto balanced budgets being passed by thelegislature (37 states have a rule requiringthis). Stricter still is a prohibition on carry-ing forward a deficit, which has been passedby 24 states. (See also Bohn and Inman1996, table 2.)

Restrictions on Issuing State Debt.Restrictions on raising public debt come in


46 Helen Ladd and Nicolaus Tideman (1981) providebackground discussion.

1950

68

0

−71

1960 1970 1980

year

1990

Figure 3. The Changing Impact of Term Limits on Total Taxes Per Capita

Article 1 2/21/03 4:17 PM Page 57

four main categories (see D. RoderickKiewert and Kristin Szakaly 1996, table 1).At one extreme are outright prohibitions onguaranteed long-term debt. Even if there isno prohibition, some states require referen-dum approval. The weakest restrictions arerevenue-based limitations or some kind ofsupermajority voting requirement in the leg-islature.

Line-Item Veto for the State Governor.This allows the state governor to strike out aparticular unwanted budgetary item ratherthan vetoing the budget as a whole. Thisshould, in theory, enhance the bargainingpower of a governor in the budgetaryprocess. We expect this to be most impor-tant when the preferences of the legislatureand the governor diverge.

Budgetary Transparency. States’ budget-ary procedures differ on a number of dimen-sions including the extent to which they runtheir budget accounting on a single year ormulti-year basis, whether they use generallyaccepted accounting standards, and whetherthey use publicly available performancemeasures. Alt, David Dreyer Lassen, andDavid Skilling (2001) construct an index forthe U.S. states based on a variety of indica-tors from the National Association of StateBudget Officers and the National Con-ference of State Legislatures. This is in-tended to capture the overall transparencyin a state’s budgetary process.

Budgetary Cohesion. States vary in the ex-tent to which the budgetary process is frag-mented across a variety of committeesRoughly half the states divide spending re-sponsibility over a number of committeesrather than having a centralized appropria-tions committee. There is also variation inthe extent to which states have separate tax-ing and spending committees.

The effectiveness of budgetary institu-tions in restraining government is moot.Some regulations are difficult to enforce,and government agents are thought to be re-sourceful in crafting ways around others.Poterba (1997) points to how the nominal

timing of tax changes can be altered to com-ply with many rules. There are also issuesabout the willingness of external agencies toenforce them.

The early literature on tax and expenditurelimitations fueled such skepticism by findingonly weak responses to their introduction.For example Burton Abrams and WilliamDougan (1986) could find no effect of theline-item veto or borrowing limits, but didfind a marginally significant negative effectof tax and expenditure limitations. However,because their results rely solely on cross-sectional variation, it is not clear how muchweight we should place on them, for the rea-sons we have already outlined. Bails andTieslau (2000) run random effect models ofstate spending with both supermajority re-quirements and expenditure limits on theright-hand side, for the period 1969–94, andfind a negative and significant effect of ex-penditure limits and supermajority require-ments.47 In contrast, Rueben (1997) findsthat binding state tax and expenditure limitshave no effect in either OLS cross-sectionalor fixed effects specifications. RonaldShadbegian (1996) considers the effect of taxand expenditure limitations on the growth ofgovernment expenditures in a panel of statesover the period 1972–87, allowing tax and ex-penditure limitations to have both a level ef-fect and an interaction effect with state in-come. Consonant with Rueben (1997), hefinds no level effect. However, the interac-tion effect is always significant. This is in-triguing, although there is no obvious inter-pretation of this result from a theoreticalpoint of view. Supermajority requirements inthe legislature are studied in Brian Knight(2000). He finds that supermajority require-ments reduce taxes as a proportion of stateincome by between 1.7 percent and 3.6 per-cent in various specifications.


47 In fact the latter is interacted with balancedbudget amendments which makes it difficult to assesswhether the supermajority requirements is significantwhen it is entered alone. However, results below sug-gest that it probably is.

Article 1 2/21/03 4:17 PM Page 58

Rueben (1997) and Knight (2000) bothtake seriously the idea that tax and expendi-ture limitations and/or supermajority re-quirements are determined endogenously.Many are fairly recent, and it seems unlikelythat their introduction is exogenous to thehistory of fiscal policy in the state in ques-tion. Both authors suggest that the availabil-ity of direct legislation is an instrument forthe budgetary institution, and estimate two-stage least-squares models that use this.Given that citizens’ initiatives have oftenbeen used to constrain taxing and spendingpower, this is a promising avenue. Rueben(1997) finds tax revenues as a percentage ofpersonal income are approximately 2 per-cent lower in states with tax and expenditurelimits in her two-stage least-squares specifi-cation. This contrasts with her ordinary leastsquares estimates in which she finds no sig-nificant effect. Because she is using a fixedinstitution as an instrument, however, she isunable to employ state fixed effects in thetwo-stage least-squares case. In addition todirect legislation, Knight (2000) uses thelegislative vote required to pass a constitu-tional amendment and the sessions requiredto consider an amendment as instruments.He finds that permitting direct legislation ispositively correlated with having superma-jority requirements and that having a largernumber of sessions is negatively correlatedwith it. His two-stage least-squares analysiscontinues to find a negative effect of super-majority requirements on taxes. As well asbeing a useful robustness check, bothRueben (1997) and Knight (2000) help toisolate the mechanism behind Matsusaka(1995)’s finding that there is a negative cor-relation between public spending and theavailability of direct legislation.

Poterba (1994a) also finds that budgetaryinstitutions are important. Using state-leveldata from the National Association of StateBudget Officers for the period 1988–92, hefinds that no-deficit carryover rules and taxand expenditure limitations affect the short-run patterns of taxes and spending in re-

sponse to shocks (calculated as differencesbetween actual and budgeted outcomes).48

Another important budgetary rule whichdiffers across U.S. states concerns the base-line against which spending plans are formu-lated. Mark Crain and Nicole V. Crain(1998) look at the effect of the differentbudget base lines across U.S. states on thegrowth of government spending. In the pe-riod that they look at (1980–90), 34 statesuse the previous year as the baseline whilesixteen use a current services baseline. Theyfind that growth rates in state spending inthe period 1980–90 are significantly higherin states that use the current services base-line.

Poterba (1997) looks in detail at rules thatrestrict the ability of states to run budgetdeficits. He points out that there are goodreasons, a priori, to be suspicious of the effi-cacy of budget rules. This is due to the rela-tive ease with which accounting procedurescan be modified and the nominal timing oftaxes and expenditures can be changed.Moreover, in many states they apply only topart of the government budget (the “generalfund” budget). Poterba’s review of the exist-ing literature leads him to conclude that fis-cal institutions do matter, but that studies todate provide little guidance on whetherstricter anti-deficit rules will reduce spend-ing or taxes.

Bohn and Inman (1996) examine the ef-fect of budgetary rules on deficits usingpanel data for 47 states over the period1970–91. They model general fund surplusas a function of economic and political con-trol variables and the state’s budgetary insti-tutions, using fixed and random effects mod-els. Because the former cannot identify theeffects of budgetary institutions that do notvary in the cross-section, they regress thefixed effects on budgetary institutions. Theirmain finding is that end of year balance


48 Poterba also provides evidence that divided stategovernments respond more slowly to unexpecteddeficits.

Article 1 2/21/03 4:17 PM Page 59

requirements enforced as constitutional (notstatutory) constraints by an elected, ratherthan politically appointed supreme court, af-fect states’ general fund surpluses. They alsofind that states with the strictest budgetaryinstitution—no-deficit carry forward—andthe line-item veto have lower deficits.However, beginning of year requirements orend-of-year requirements enforced by a po-litically appointed supreme court have noimpact on general fund deficit behavior.They also find that states requiring refer-enda ahead of debt issues also have lowerdeficits. In contrast to other studies, how-ever, they do not find an effect of dividedgovernment.

A number of studies have looked at theimpact of restrictions on debt finance.Beverly Bunch (1991) considers the rela-tionship between such limits and the num-ber of state authorities and their scope.Although she finds some relationship be-tween the two, the cross-sectional approachdoes not make such findings reliable, andthe theoretical link is far from clear. Jurgenvon Hagen (1991) considers the cross-sectional impact of debt limitations on statedebt in 1985, and finds little effect on totaldebt. Kiewert and Szakaly (1996) look at theeffect of state debt limitations on debt inU.S. states using data from 1960–90. Theyuse a random effects model and find thatstates that prohibit guaranteed debt and re-quire a referendum for approval had lessguaranteed debt than those that required asupermajority or those with revenue-basedlimitations.

There is a large literature that looks at theeffect of budgetary institutions on municipalbond markets, in particular bond ratings.This is not surprising in light of the above re-sults—Poterba (1994b) shows that there is aclose link between tax policy and bondyields. One of the earliest quantitative stud-ies, Morris Goldstein and Geoffrey Woglom(1992), relates the interest rate on generalobligation debt to the strictness of anti-deficit provisions as measured by the

Advisory Council on Inter-governmentalRelations for the period 1982–90. They findthat restrictive fiscal limits are associatedwith lower interest rates.

Poterba and Rueben (1999) examine thelink between fiscal institutions and spreadson bond yields, using the Chubb yieldspreads data, which are available as a panelfor forty states from 1973–95. They showthat states with binding tax limitations have a15 to 20 basis points higher yield on theirstate bonds. This makes sense, if it makes itmore costly for such states to raise revenuesto repay their debts. Expenditure limitationshave the opposite effect. Lowry and Alt(2001) describe a mechanism via which lim-its on carrying forward deficits improve theability of investors to extract informationfrom noisy signals. They find support fortheir views using panel data on yield spreadsusing data from 1973–95. Poterba andRueben (2001) use data from the NationalAssociation of State Budget Officers be-tween 1988 and 1997 to investigate whetherstates with restrictive fiscal institutions usedeficits less as a smoothing device againstshocks to the fiscal system. Unexpecteddeficits are shown to have a particularlylarge effect in raising the bond yields ofstates with tax limits.

Douglas Holtz-Eakin (1988) considers theimpact of the line-item veto on budgetaryoutcomes, using a variety of measures ofstate expenditures and tax revenues as thedependent variables from 1965–83, and arange of economic and political variables ascontrols. As well as looking at cross-sectionaldifferences, Holtz-Eakin identifies the ef-fect in a fixed effects model by interactingwhether a state permits a line-item veto withthe nature of political control, in particularwith whether there is divided control of thelegislature. The cross-sectional results findno effect of the line-item veto, consonantwith Abrams and Dougan (1986). However,the fixed effects results suggest that whengovernment power is divided between twoparties—one controlling the executive and


Article 1 2/21/03 4:17 PM Page 60

the other the legislature—then having theline-item veto reduces spending and raisestaxes, leading to smaller deficits. The line-item veto appears to have most impact whenthere is a preference conflict. In contrast toHoltz-Eakin’s cross-sectional results, Bohnand Inman (1996) find that mean statedeficits are somewhat lower in states wherethe governor has a line-item veto.

Alt, Dreyer Lassen, and Skilling (2001)correlate their budgetary transparency in-dex, discussed above, with the size of gov-ernment. The exercise is motivated by thetheoretical approach of Ferejohn (1999)who argues that greater transparency willlikely lead to voters being happier withlarger levels of government spending. In across section using data from 1986–95, theyfind in favor of this hypothesis, with greatertransparency being positively correlatedwith higher state spending per capita. Theyalso find that transparency is positively cor-related with the popularity of the governor.This work is important in bringing more spe-cific aspects of the agency view to the data.Given current discussions about the impor-tance of transparency in government, this islikely to be an important area in the future.

Mark Crain and Timothy Murris (1995)consider how the fragmentation of the budg-etary process affects taxing and spending inthe states. The fact that the institutions arefixed in the cross-section does not permit apanel data analysis. They develop a cross-sectional analysis based on the period1982–88, and find that states with a central-ized spending committee spend 8 percentless on average than those without. Rev-enues are higher by 28 percent, on average,in states with a combined revenue andspending committee.

We present results on the association be-tween tax and expenditure limitations andstate taxes and spending in table 14.Controlling for state and year effects, andtime-varying state-level variables, we findthat non-binding limits (those that are advi-sory or require only a simple legislative

majority to change) are not significantly cor-related with either taxes per capita or spend-ing per capita. However, potentially bindingtax or expenditure limits are positively (i.e.,perversely) correlated with taxes. This againhighlights the problems associated withquantifying the impact of institutions on pol-icy outcomes when the institutions maychange in response to policy choices—suchas taxes and spending that citizens considerinappropriately high. We return to the issuesassociated with endogenous institutions insection 8.

Estimates of the impact of supermajorityrules and the line-item veto are also pre-sented in table 14. Like much of the litera-ture, we find a large, negative and significanteffect of super-majority rules on total taxescollected per capita. On average, and withall else held equal, state taxes per capita areroughly $46 lower in the years after the statepasses a supermajority requirement. Thisamounts to roughly 8 percent of state taxesduring the period over which the estimationis run (1960–97). The supermajority re-quirement has a similarly large, negative ef-fect on state spending (columns 6) but its ef-fect is not significant when other rules areincluded in the regression (column 8) oftable 14.

In this period, only two states changedtheir rules on allowing the governor a lineitem veto: Iowa and Washington both movedto allow line item vetoes in 1969. Thus, theindicator that the governor is allowed a lineitem veto is identified only off of the differ-ence in these two states before and after itspassage (and is for that reason not particu-larly meaningful.) More interesting is theHoltz-Eakin interaction term of the line-item veto with divided government. We find,in contrast to Holtz-Eakin, that in a dividedgovernment the line-item veto reduces bothtaxes and spending. However, given that thecoefficient on taxes is smaller in absoluteterms than that on spending, the finding isstill consistent with Holtz-Eakin’s findingthat deficits are lower in situations where


Article 1 2/21/03 4:17 PM Page 61

there is divided government and a line-itemveto.

Overall, the evidence that we have pre-sented, alongside that from the existing liter-ature, supports an emerging consensus thatgovernment behavior can be influenced bybudgetary rules. However, the exact form ofthe rules is important.

7.6 Indirect Effects of Institutional Rules

The theoretical discussion suggested a pos-sible distinction between institutional rulesthat affect the policy process ex post andthose that primarily have an effect on elec-toral outcomes. Institutions such as open pri-maries, voter registration laws and restric-tions on corporate campaign contributionsseem most likely to be examples of the insti-

tutions I2st which enter equation (3). Thesemay in turn affect the equilibrium strategieschosen by parties. We would then expect theimpact of these variables on policy to be me-diated via their effect on election outcomessuch as party competition or Democratic con-trol of the legislature. We already know fromsection 5.5 that there are policy effects ofthese institutions. Table 15 explores whethertheir effect works via �st .

Column 1 of table 15 shows that these in-stitutions are highly significantly correlatedwith total taxes per capita. Specifically, openprimaries and corporate restrictions are neg-atively and significantly correlated with taxes,while less-costly voter registration is posi-tively and significantly correlated with taxes.The F-statistic of their joint significance is


TABLE 14INCUMBENT DISCRETION AND POLICY CHOICE

Dependent Variable:

Total taxes per capita Total spending per capita

Indicator: Non-binding tax or 4.79 — — −16.28 −10.07 — — −44.71expenditure limitation (7.61) (7.87) (12.17) (13.45)

Indicator: Potentially binding tax 24.13 — — 38.84 10.37 — — 41.45or expenditure limitation (9.09) (9.41) (14.58) (16.07)

Supermajority needed to — −46.12 — −52.28 — −46.27 — −8.61increase taxes (8.51) (10.16) (13.94) (17.15)

Indicator: Governor has a line — — −25.99 −37.73 — — 3.01 4.00item veto (13.82) (16.03) (23.34) (27.42)

Indicator: Governor’s party is — — 10.37 9.07 — — 32.44 33.15not that of the united majority (8.12) (8.36) (13.71) (14.29)party in the legislature

Line item veto × divided — — −25.34 −21.33 — — −45.47 −44.14government (8.83) (9.14) (14.91) (15.63)

Years over which regression all years all years all years all years all years all years all years all yearsrun 1960–97 1960–97 1950–91 1960–91 1960–96 1960–96 1950–91 1960–91

Number of observations 1817 1817 1961 1529 1776 1776 1968 1536

Notes: Standard errors in parentheses. All regressions control for year and state effects, and include controls forthe proportion of population aged 65 and above; the proportion of population aged 5 to 17; state income percapita in $1982, income squared; and state population and population squared. All dollar-denominated variablesare in 1982 dollars. “Non-binding” tax and expenditure limitations are those that are either advisory or requireonly a simple legislative majority to amend or overrule.

Article 1 2/21/03 4:17 PM Page 62

large ( F = 20.22, p-value = 0.0000). It is dif-ficult to make a case that any of these institu-tional rules have a direct effect on taxes,which leads us to ask whether we might un-cover the chain through which they do affectpolicy.

Column 2 of table 15 presents two-stage-least-squares (2SLS) estimates of the impactof party competition in the state legislatureon total taxes, in which our measure of partycompetition is instrumented on the institu-tional rules. We find that, instrumented onopen primaries, registration rules and corpo-rate restrictions, party competition is nega-tively and significantly correlated with taxes.The results in column 2 suggest that a onestandard deviation increase in competitive-ness (an increase in our measure of 0.07points) is associated with a reduction in totaltaxes per capita of $240, measured in$1982.49 This is roughly a one standard devi-ation decrease in taxes over this period (aone standard deviation decrease wouldamount to $230). Moreover, we can com-pare the fit of two regressions—that in col-umn 1, where the institutional rules are al-lowed to enter in an unrestricted fashion,and that in which the institutional rules en-ter only through their effect on predictedparty competition. The F-test at the bottomof column 2 answers the question ofwhether the fit of the regression is signifi-cantly worse if we force the institutionalrules to enter only through their effect onparty competition. It is not: F = 0.76, p-value= .5483, and we cannot reject that theserules are affecting total taxes solely throughtheir effect on party competition. (Note thatthis does not prove that party competition isthe mechanism at work—but it cannot berejected by the data.) This suggests thatthese variables can be interpreted as beingpart of the vector I2st in equation (3).

We repeat this exercise in column 3, hereasking whether the institutional rules mightbe working through Democratic control.

The 2SLS estimate of Democratic control ispositive and significant. Unlike the resultsfor party competition, however, the fit of theregression in which the institutional rulesare entering through the indicator ofDemocratic control is significantly worsethan that when the institutional variables areunrestricted (F = 13.57, p-value = 0.0000).When both predicted values are entered inthe same regression, as in column 4, it isclear that party competition is the favoredmechanism through which these institu-tional rules are affecting total taxes.

We present this discussion solely as an il-lustration. We believe, however, that moregenerally this sort of exercise may prove use-ful when analyzing whether and how the leg-islative political landscape is affecting policychoice. For many analyses, the institutionalrules will provide a source of exogenous vari-ation in political control that can be used toanalyze how policies change in response tothe character of the legislature. In all ofthese analyses, it will be important to askhow the institutional rules have been set—atopic we turn to in the next section. Moresignificantly, this exercise shows that there isscope for imposing some greater theoreticalstructure on the problem to analyze the ef-fects of institutions on outcomes and to testthat structure empirically.

8. Endogenous Institutions

The common assumption in the vast ma-jority of cross-state studies is that the institu-tions are exogenous. This is more or lessplausible depending on the time frame andissue in question. Endogeneity is really onlypart of a broader set of concerns about thepossibility that institutions are correlatedwith omitted right hand side variables, andone hope is that state fixed effects and time-varying state-level variables will capturestate differences that determine both insti-tutions and policy. To better understandwhether and when it is appropriate to takeinstitutions as exogenous, we provide in this


49 That is, –3434.63 × 0.07 = –240.42.

Article 1 2/21/03 4:17 PM Page 63

section a look at some of the systematicdeterminants of institutional rules.

Even if institutions do not change overtime, there is no guarantee that it is legiti-mate to treat them as exogenous. Moreover,as we observed in our study of citizens’ ini-tiatives, it may be difficult to control forsources of unobserved heterogeneity withstate fixed effects when institutions are fixedover time. Thus, it remains difficult to dis-tinguish between a genuine institutional ef-

fect and the possibility that tastes for citi-zens’ initiatives and taxation are correlated.The main hope here is that some kind ofcomparative static with respect to some ex-ogenously changing variable can be identi-fied and tested. For example, Besley andCoate (2000a) used the comparative staticwith respect to fuel prices to gauge the influ-ence of elected versus appointed regulatorseven where the latter institution did notchange.


TABLE 15INSTITUTIONAL RULES AND LEGISLATIVE CONTROL

Dependent Variable: Total taxes per capita

Open primaries −19.37 — — —(7.04)

Indicator: voter registration through 32.02 — — —vehicle agency (6.86)

Indicator: voter registration possible on 100.68 — — —polling day or no registration necessary (15.92)

Indicator: restriction on corporate −18.10 — — —campaign contributions (4.95)

F-test: joint significance of institutional 20.22 — — —variables (p-value in parentheses) (.0000)

IV: Party competition in legislature — −3434.63 — −3376.73(776.53) (928.31)

IV: Democrats control both lower — — 165.27 8.44and upper house (39.08) (76.38)

F-test (see notes to table) — 0.764 13.57 0.753(p-value in parentheses) (.5483) (.0000) (.5561)

Years over which regression run All years 1950–1958, and 1960–1996Number of observations 1925 1925 1925 1925

Notes: Standard errors in parentheses. All regressions control for year and state effects, and include controls forthe proportion of population aged 65 and above; the proportion of population aged 5 to 17; state income percapita in $1982 and income squared; state population and population squared. All dollar-denominated variablesare in 1982 dollars. In an auxiliary regression, we regress total taxes on all other right side variables and the pre-dicted value of party competition in the legislature, where we use open primaries, voter registration through vehi-cle agency, voter registration on polling day, and restrictions on corporate contributions as instruments. The F-testin column 2 compares the fit of the regression using the predicted value to that in column 1, where the institu-tional rules are allowed to enter in an unrestricted fashion. Results in column 3 report an analogous comparisonwhen an indicator that Democrats control both houses is instrumented using the institutional rules. We cannot reject that these institutional rules are affecting total taxes solely through their effect on party competition in thelegislature.

Article 1 2/21/03 4:17 PM Page 64

The charge of endogeneity is not by itselfvery meaningful. The key empirical issue isto identify possible sources of correlationbetween institutional variables and the errorterm in an equation which has either xjst or�jst on the right hand side. The most diffi-cult cases are those where there are goodreasons to believe that there are shocks tothe policy in question that drive the demandfor institutional reform—as when states thathave a history of deficits implement somekind of balanced budget rule. In each in-stance, the possible fix (if one exists) must bethought out on a case-by-case basis. It is un-likely that there is any kind of panacea forthese problems.

Progress is most likely to come throughan underlying theoretical account of whatmotivates institutional change. The largeliterature on strategic use of long-run poli-cies is relevant here. The early literature—Persson and Svensson (1989) and Tabelliniand Alesina (1990) focused on the strategicuse of debt to constrain the flexibility of fu-ture incumbents. A general account of thistype of argument and its relationship to thenotion of political failure is in Besley andCoate (1998). Institutional reform is cer-tainly one way of trying to influence futurepolitical outcomes. Our results on redis-tricting in table 6 provide a particular an-gle on this showing that the composition ofthe legislature at the time of redistrictinghad an impact on subsequent patterns ofrepresentation.

In some contexts, the theory may give riseto a natural instrument for the institution inquestion. One of the main cases where thisis an issue is the case of tax and expenditurelimitations that were adopted comparativelyrecently and may have been favored in statesthat have a history of high spending. Reuben(1997) and Knight (2000) both use citizens’initiatives as instruments for tax and expen-diture limitations and supermajority re-quirements respectively. While ingenious,there are a couple of drawbacks. First, forthe reasons that we discussed in our treat-

ment of citizens’ initiatives above, fixed ef-fects cannot be included and this may be arather important omission. Second, in in-strumenting one institution using another,we must assume that the institution used asan instrument is not itself correlated withthe unobservables in the policy equation.

As discussed above, Stratmann andAparicio-Castillo (2001) also consider thepossibility that campaign finance laws areendogenously chosen to advantage the in-cumbents. They use state education as an in-strument, finding that states with more edu-cated voters tend to have a lower restrictionon individual campaign limits.

In one of the few studies to examine di-rectly the endogeneity of institutions,Hanssen (2001) considers the strategic useof appointment rules for state court judges.He argues that incumbents without firmcontrol on the reins of power are likely toprefer more independent judiciaries giventhat they might not be in office in the futureand perceive the judiciary as a check on fu-ture incumbents’ behavior. However, a partythat is expected to remain in office will notbenefit from such countervailing powers.Hanssen considers the 27 changes in judicialselection rules over the period from 1950 to1990. Modeling the timing of the switches asa function of the prevailing political condi-tions, he finds that firmer single party con-trol is found to be associated with less inde-pendently accountable judiciaries.

The extent to which institutions respondto the political composition of the legisla-ture, and to state demographic change, is ex-plored in table 16. Specifically, we askwhether current institutional rules on pri-maries, on voter registration, and on corpo-rate campaign restrictions, are significantlycorrelated with measures of past politicallegislative control and with past demo-graphic and economic variables. In the firstcolumn for each institutional rule, weregress the institution on lagged values ofthe proportion of Democrats in the statelower house in years t − 4, t − 6, t − 8 and t − 10;


Article 1 2/21/03 4:17 PM Page 65

the proportion of Democrats in the state upperhouse in years t − 4, t − 6, t − 8 and t − 10;indicators that the Democrats controlledboth houses in t − 4, t − 6, t − 8 and t − 10;and our measure of party competition in thelegislature in t − 4, t − 6, t − 8 and t − 10.(Our results are robust to the inclusion/exclusion of lags at t − 4 and t − 12.) In all ofthese regressions, we control for the sametime-varying state-level variables describedabove, along with state fixed effects and yeareffects. We find—both for open primariesand for constraints on corporate contribu-tions—that past legislative variables are sig-nificant determinants of current state policy.In the second column for each rule, weregress the institution on lagged values ofstate population, state income per capita,

the proportion of the population aged 65and above, and the proportion aged five toseventeen in years t − 4, t − 6, t − 8, and t − 10.We find, for open primaries, motor-voterregistration, and corporate campaign restric-tions, that these past economic and demo-graphic variables are significant determi-nants of current rules. Jointly, the pasteconomic and legislative variables are signif-icant for open primaries, for motor-voterregistration and for corporate restrictions.

In studying the impact of a particular in-stitutional rules on policy outcomes, then,we must ask whether the determinants ofthe institutional rule are thought to have in-dependent effects on current policy. For ex-ample, past legislatures may affect both theinstitutional rule (as we find that they do for


TABLE 16THE LONG-RUN IMPACT OF LEGISLATIVE CONTROL ON STATE INSTITUTIONS

Indicator: VoterIndicator: Open Registration through

Primaries Vehicle Registration

F-test: lagged legislative 2.19 — — 0.69 — —variables (p-value) (.0194) (.7903)

F-test: lagged — 5.55 — — 2.77 —demographic variables (.0000) (.0034)(p-value)

F-test: lags in both — — 6.12 — — 3.96legislative and (.0000) (.0000)demographic variables


Years even years 1960–90, even years 1960–981996, 1998

Notes: All regressions control for year and state effects, and include controls for the proportion of population aged65 and above; the proportion of population aged 5 to 17; state income per capita in $1982 and income squared;state population and population squared. The F-test for lagged legislative variables is a test for the joint signifi-cance of the following variables: the proportion of the lower house held by Democrats in years t − 4, t − 6, t − 8and t − 10; the proportion of the state upper house held by Democrats in years t − 4, t − 6, t − 8, and t − 10; indi-cators that the Democrats controlled both houses in t − 4, t − 6, t − 8, and t − 10; and our measure of party

Article 1 2/21/03 4:17 PM Page 66

corporate contributions) and may also tiethe hands of the current legislature (for ex-ample, by taking on debt). As always, thismust be addressed on a case-by-case basis.

9. Concluding Remarks

U.S. states are an important laboratory forempirical investigation of policy processes.Much encouragement can be taken from thevariety and richness of the existing body ofcross-state literature on the policy impact ofpolitical institutions. There can be littledoubt that the structure of political repre-sentation, the terms on which elections arefought, and the rules governing the policyprocess, all influence policy outcomes.Moreover the directions in which these in-fluences go are forming an increasingly ro-

bust body of knowledge as the literature inthis area matures. Above all, this is encour-aging for the future theoretical work on po-litical economy. For, without access to dataand testing, there is little to anchor theoreti-cal progress.

Apart from the detailed insights availablein individual studies, the following five gen-eral lessons emerge. First, processes for theselection of candidates and for gaining ac-cess to the vote are strongly correlated withpatterns of political representation. Thisshould undermine the cynical view that elec-tions are a veil that affect very little. Thismessage is further reinforced by the secondrobust observation—that party control andidentity matter. Early attempts to quantifythis achieved mixed results. However, themore recent literature and findings shown


TABLE 16 (Cont.)

Indicator: Day-of- Indicator: Restrictions Election Registration or on Corporate

No Registration Contributions

F-test: lagged legislative 0.34 — — 1.70 — —variables (p-value) (.9893) (.0811)

F-test: lagged — 0.48 — — 1.81 —demographic variables (.9447) (.0593)(p-value)

F-test: lags in both — — 0.40 — — 2.28legislative and (.9961) (.0053)demographic variables


Years even years 1960–98 even years 1960–98

competition in the legislature in t − 4, t − 6, t − 8, and t − 10. The F-test for lagged demographic variables is a testfor the joint significance of the following variables: state income in periods t − 4, t − 6, t − 8, and t − 10; state pop-ulation in periods t − 4, t − 6, t − 8, and t − 10; proportion of population aged 65 and above in t − 4, t − 6, t − 8,and t − 10; and proportion aged 5 to 17 in t − 4, t − 6, t − 8, and t − 10. Results reported are were estimated usingrobust standard errors.

Article 1 2/21/03 4:17 PM Page 67

here leave little doubt on this score. Ourthird robust finding is that elections serve akey role in achieving accountability—officeholders who do not face reelection behavedifferently from those that do. Fourth, wewould highlight the finding that efforts toempower or disempower policy makersthrough restrictions on taxes and expendi-tures are strongly correlated with differ-ences in fiscal behavior. This suggests thatthe original Public Choice agenda, whichcentered on the debate of how to redesignthe fiscal constitution, is not without sub-stance when experiments in this area aretried. Finally, the data suggest in a variety ofcontexts that the there are empirical differ-ences between the determinants and conse-quences of elections for governors and thelegislature. There is scope for making thesemore precise and developing their theoreti-cal implications in more detail.

While our focus here is exclusively on theUnited States, we recognize the link be-tween the progress in empirical politicaleconomy being made on U.S. data and thatmade elsewhere. Of particular importance isthe recent cross-country work by Perssonand Tabellini (1999, 2002) and Gian-MariaMilesi-Ferretti, Roberto Perotti, andMassimo Rostagno (2002). They have beenbuilding data sets that model the differencesbetween political institutions across coun-tries. They have focused particularly on dif-ferent models of the separation of powersand the extent to which systems are closer toproportional representation.

Cross-country data is also being used tostudy the effect of budgetary institutions.For example, Alesina, Ricardo Hausman,Rudolf Hommes and Ernesto Stein (1999)study this for Latin America. It is clear thatin all such work similar issues to those dis-cussed here, such as identifying chains of in-fluence and possible sources of endogeneity(particularly sources of unobserved hetero-geneity) are relevant. Overall, this strand ofresearch is complementary in its ambitionand scope to that discussed here. Moreover,

there are sources of variation that can bestudied transnationally, but not using U.S.states—variations in the electoral systembeing a key example.

There is also parallel work at a subna-tional level in other federal systems that ex-hibit important institutional variations. Agood example is the work on citizens’ initia-tives in Switzerland by Pommerehne (1990)and Lars Feld and Matsusaka (1999). Pande(2001) exploits differences in rules that man-date representation of disadvantaged mi-norities (scheduled castes and scheduledtribes) in India. She finds that the transferprograms favoring the disadvantaged groupsexpand (other things being equal) when rep-resentation increases. Subnational work ofthis type is an important part of the agendato deepen our empirical understanding ofthe policy process.

Viewed as applied economics, the empiri-cal study of policy processes presents chal-lenges. Ideally, the empirical agenda givesdirection to theoretical modeling. At a gen-eral level, the median voter model, theworkhorse of so much political economymodeling for more than a generation, re-ceives little empirical support. It is indirectlyrefuted by two of the key findings from theliterature. First, that institutions appear toaffect the link between citizens’ preferencesand policy. Second, the identity of represen-tatives, such as their gender and party affilia-tion, matters. The Downsian model, takenliterally, encourages us to believe that insti-tutions and parties are irrelevant—all thatpoliticians care about is seeking out the pre-ferred outcome of the median voter. It hasbeen known for some time that this is prob-lematic theoretically. It also appears that itdoes a poor job at fitting the data.

But what kind of theory is likely to work?First, it is clear a key role of parties with pol-icy preferences, mediating between policyand preferences, is a promising component.Attention needs also to be paid to frameworksin which multiple issues are determined—thedata suggest that different policy issues


Article 1 2/21/03 4:17 PM Page 68

respond to institutions in heterogeneous ways.The predictions of agency theoretic models inwhich voters learn about their incumbentsover time also appear to have some empiricalrelevance in understanding the data.

Overall, it is doubtful that any grand uni-fying theory will emerge. It is more realisticto expect piecemeal theoretical analyses thatyield tight empirical predictions for a con-crete institutional variation. However, estab-lishing empirical regularities for their ownsake also has a role to play in stimulatingfurther theoretical work.

REFERENCES

Abrams, Burton A. and William R. Dougan. 1986. “TheEffects of Constitutional Restraints on GovernmentSpending,” Public Choice 49:2, pp. 101–16.

Adams, James D. and Lawrence W. Kenny. 1986. “Op-timal Tenure of Elected Public Officials,” J. LawEcon. 29:2, pp. 303–28.

———. 1989. “The Retention of State Governors,”Public Choice 62:1, pp. 1–13.

Alesina, Alberto. 1988. “Credibility and Policy Conver-gence in a Two-Party System with Rational Voters,”Amer. Econ. Rev. 78:4, pp. 796–805.

Alesina, Alberto and Howard Rosenthal. 1994. Parti-san Politics, Divided Government and the Economy.NY: Cambridge U. Press.

Alesina, Alberto; Ricardo Hausmann, Rudolf Hommesand Ernesto Stein. 1999. “Budget Institutions andFiscal Performance in Latin America,” J. Devel.Econ. 59:2, pp. 253–73.

Alesina, Alberto and Nouriel Roubini. 1992. “PoliticalCycles in OECD Economies,” Rev. Econ. Stud.59:4, pp. 663–88.

Alt, James E. and Robert C. Lowry. 1994. “DividedGovernment, Fiscal Institutions and BudgetDeficits: Evidence from the States,” Amer. Polit. Sci.Rev. 88:4, pp. 811–28.

———. 2000. “A Dynamic Model of State BudgetOutcomes Under Divided Partisan Government,” J.Politics 62:4, pp. 1035–69.

Alt, James E.; David Dreyer Lassen and David Skilling.2001. “Fiscal Transparency, Gubernatorial Popular-ity, and the Scale of Government: Evidence from theStates,” unpub. manuscript, Harvard U. and U.Copenhagen.

Bails, Dale, and Margie A. Tieslau. 2000. “The Impactof Fiscal Constitutions on State and Local Expendi-tures,” Cato J. 20: 2, pp. 255–77.

Barrilleaux, Charles. 2000. “Party Strength, PartyChange and Policy-Making in American States,”Party Politics 6:1, pp. 61–73.

Barro, Robert. 1973. “The Control of Politicians: AnEconomic Model,” Public Choice 14, pp. 19–42.

Becker, Gary S. 1985. “Public Policies, PressureGroups and Dead Weight Costs,” J. Public Econ.28:3, pp. 329–47.

Benabou, Roland. 2000. “Unequal Societies: IncomeDistribution and the Social Contract,” Amer. Econ.Rev. 90:1, pp. 96–129.

Berry, Francis Stokes and William D. Berry. 1992. “TaxInnovation in the States: Capitalizing on PoliticalOpportunity,” Amer. J. Polit. Sci. 36:3, pp. 715–42.

Berry, William D. 1979. “Utility Regulation in theStates: The Policy Effects of Professionalism andSalience to the Consumer,” Amer. J. Polit. Sci. 23:2,pp. 263–77.

Berry, William D.; Evan J. Ringquist, Richard C. Ford-ing and Russell L. Hanson. 1998. “Measuring Citi-zen and Government Ideology in the AmericanStates, 1960–93,” Amer. J. Polit. Sci. 42:1, pp.327–48.

Besley, Timothy and Anne Case. 1995a. “IncumbentBehavior: Vote Seeking, Tax Setting and YardstickCompetition,” Amer. Econ. Rev. 85:1, pp. 25–45.

———. 1995b. “Does Political Accountability AffectEconomic Policy Choices? Evidence from Guberna-torial Term Limits,” Quart. J. Econ. 110:3, pp.769–98.

———. 2000. “Unnatural Experiments? Estimatingthe Incidence of Endogenous Policies,” Econ. J.110:467, pp. F672–94.

Besley, Timothy and Stephen Coate. 1997. “An Eco-nomic Model of Representative Democracy,” Quart.J. Econ. 112:1, pp. 85–114.

———. 1998. “Sources of Inefficiency in a Represen-tative Democracy: A Dynamic Analysis,” Amer.Econ. Rev. 88:1, pp. 139–56.

———. 2000a. “Elected versus Appointed Regulators:Theory and Evidence,” NBER work. paper 7579.

———. 2000b. “Issue Unbundling via Citizens’ Initia-tives,” NBER work. paper 8036.

———. Forthcoming. “On the Public Choice Critiqueof Welfare Economics,” Public Choice.

Black, Duncan. 1987. The Theory of Committees andElections. London: Kluwer Academic.

Bohn, Henning and Robert P. Inman. 1996. “BalancedBudget Rules and Public Deficits: Evidence fromthe U.S. States,” NBER work. paper, 5533.

Bowler, Shaun; Todd Donavan and Caroline J. Tol-bert, eds. 1998. Citizens as Legislators: DirectDemocracy in the United States. Columbus: OhioState U. Press.

Brennan, Geoffrey and James M. Buchanan. 1980. ThePower to Tax: Analytical Foundations of the FiscalConstitution. Cambridge: Cambridge U. Press.

Brennan, Geoffrey and Alan Hamlin. 1998. “Constitu-tional Economics,” in The New Palgrave Dictionaryof Economics and the Law. Peter Newman, ed. Lon-don: MacMillan, pp. 401–409.

Broder, David S. 2000. Democracy Derailed: InitiativeCampaigns and the Power of Money. NY: Harcourt.

Brown, Robert D. 1995. “Party Cleavages and WelfareEffort in the American States,” Amer. Polit. Sci. Rev.89:1, pp. 23–33.

Buchanan, James M. 1967. Public Finance in Demo-cratic Process. Chapel Hill: U. North Carolina Press.

———. 1987. “Constitutional Economics,” in The NewPalgrave Dictionary of Economics. John Eatwell,Murray Milgate and Peter Newman, eds. London:Macmillan.


Article 1 2/21/03 4:17 PM Page 69

Bunch, Beverly S. 1991. “The Effect of ConstitutionalDebt Limits on State Governments’ Use of PublicAuthorities,” Public Choice 68:1–3, pp. 57–69.

Butler, David and Austin Ranney, eds. 1978. Referen-dums: A Comparative Study of Practice and Theory.Washington, DC: American Enterprise Institute.

Carmines, Edward G. 1994. “Political Issues, PartyAlignments, Spatial Models and the Post-New DealParty System,” in New Perspectives on AmericanPolitics. Dodd and Jillson, eds. Washington DC: CQPress, pp. 77–97.

Case, Anne. 1998. “The Effects of Stronger Child Sup-port Enforcement of Non-marital Fertility,” in Fa-thers Under Fire: The Revolution in Child SupportEnforcement. Irwin Garfinkel, Sara McLanahan,Daniel Meyer and Judith Seltzer, eds. NY: RussellSage, pp. 191–215.

Case, Anne C.; Harvey S. Rosen and James R. Hines Jr.1993. “Budget Spillovers and Fiscal Policy Interde-pendence: Evidence from the States,” J. PublicEcon. 52:3, pp. 285–307.

Center for the American Woman and Politics. 1991.The Impact of Women in Public Office, Findings at aGlance. New Brunswick: Rutgers U.

Chubb, John E. 1988. “Institutions, The Economy, andthe Dynamics of States Elections,” Amer. Polit. Sci.Rev. 82:1, pp. 133–54.

Clarke, Wes. 1998. “Divided Government and BudgetConflict in the U.S. States,” Leg. Stud. Quart. 23:1,pp. 5–22.

Coate, Stephen. 2001. “Political Competition withCampaign Contributions and Informative Advertis-ing,” NBER work. paper 8693.

Corrado, Anthony; Thomas E. Mann, Daniel R. Ortiz,Trevor Potter and Frank J. Sorauf, eds. 1997. Cam-paign Finance Reform: A Sourcebook. Washington,DC: Brookings Institution.

Camobreco, John F. 1998. “Preferences, Fiscal Policiesand the Initiative Process,” J. Politics 60, pp. 819–29.

Costello, Kenneth W. 1984. “Electing Regulators: TheCase of Public Utility Commissioners,” Yale J. Regu-lation 2, pp. 83–105.

Crain, W. Mark and Nicole V. Crain. 1998. “Fiscal Con-sequences of Budget Baselines,” J. Public Econ.67:3, pp. 421–36.

Crain, W. Mark and Robert E. McCormick. 1984.“Regulators as an Interest Group,” in The Theory ofPublic Choice II. James M. Buchanan and Robert D.Tollison, eds. Ann Arbor: U. Michigan Press, pp.287–304.

Crain, W. Mark and Timothy J. Muris. 1995. “Legisla-tive Organization of Fiscal Policy,” J. Law Econ.38:2, pp. 311–33.

Crain, W. Mark and Lisa K. Oakley. 1995. “The Politicsof Infrastructure,” J. Law Econ. 38:1, pp. 1–17.

Crain, W. Mark and Robert D. Tollison. 1977. “Attenu-ated Property Rights and the Market for Gover-nors,” J. Law Econ. 20:1, pp. 205–11.

———. 1993. “Time Inconsistency and Fiscal Policy:Empirical Analysis of U.S. States, 1969–89,” J. Pub-lic Econ. 51:2, pp. 153–59.

Denzau, Arthur T.; Robert J. Mackay and CarolynWeaver. 1981. “On the Initiative-Referendum Op-tion and the Control of Monopoly Government,” in

Tax and Expenditure Limitations. Helen F. Laddand T. Nicolaus Tideman, eds. Washington, DC: Ur-ban Institute.

Diermeier, Daniel and Antonio Merlo. 2000. “Govern-ment Turnover in Parliamentary Democracies,” J.Econ. Theory 94:1, pp. 46–79.

Dixit, Avinash. 1996. The Making of Economic Policy:A Transaction-Cost Politics Perspective. Cambridge:MIT Press.

Dixit, Avinash, and John Londregan. 1998. “Ideology,Tactics, and Efficiency in Redistributive Politics,”Quart. J. Econ. 113:2, pp. 497–529.

Downs, Anthony. 1957. An Economic Theory ofDemocracy. NY: Harper Collins.

Dye, Thomas R. 1984. “Party and Policy in the States,”J. Politics 46:4, pp. 1097–116.

Edlund, Leena and Rohini Pande. 2002. “Why HaveWomen Become Left-Wing? The Political GenderGap and the Decline in Marriage,” Quart. J. Econ.117:3, pp. 917–61.

Erikson, Robert S.; Gerald C. Wright Jr. and John P.McIver. 1989. “Political Parties, Public Opinion, andState Policy in the United States,” Amer. Polit. Sci.Rev. 83:3, pp. 729–50.

Farnham, Paul G. 1990. “The Impact of Citizen Influ-ence on Local Government Expenditure,” PublicChoice 64:3, pp. 201–12.

Feddersen, Timothy and Wolfgang Pesendorfer. 1999.“Abstention in Elections with Asymmetric Informa-tion and Diverse Preferences,” Amer. Polit. Sci. Rev.93:2, pp. 381–98.

Feld, Lars P. and John G. Matsusaka. 1999. “The Polit-ical Economy of Tax Structure: Some Panel Evi-dence for Swiss Cantons,” manuscript, U. St. Gallenand U. Southern California.

Ferejohn, John. 1986. “Incumbent Performance andElectoral Control,” Public Choice 50, pp. 5–25.

———. 1999. “Accountability and Authority: Toward aTheory of Political Accountability,” in Democracy,Accountability and Representation. Adam Prze-worski, Susan C. Stokes and Bernard Manin, eds.NY: Cambridge U. Press, pp. 131–53.

Fields, Joseph A.; Linda S. Klein and James M. Sfiridis.1997. “A Market Based Evaluation of the Electionversus Appointment of Regulatory Commissioners,”Public Choice 92:3–4, pp. 337–51.

Filer John E.; Lawrence W. Kenny and Rebecca B.Morton. 1993. “Redistribution, Income, and Vot-ing,” Amer. J. Polit. Sci. 37:1, pp. 63–87.

Filer John E.; Lawrence W. Kenny and Rebecca B.Morton. 1991. “Voting Laws, Educational Policies,and Minority Turnout,” J. Law Econ. 34:2, pp.371–93.

Fiorina, Morris P. 1981. Retrospective Voting in Amer-ican National Elections. New Haven: Yale U. Press.

Formby, John P.; Banamber Mishra and Paul D. This-tle. 1995. “Public Utility Regulation and Bond Rat-ings,” Public Choice 84:1–2, pp. 119–36.

Garand, James C. 1988. “Explaining GovernmentGrowth in the U.S. States,” Amer. Polit. Sci. Rev.82:3, pp. 837–49.

Gelman, Andrew and Gary King. 1994. “EnhancingDemocracy Through Legislative Redistricting,”Amer. Polit. Sci. Rev. 88:3, pp. 541–59.


Article 1 2/21/03 4:17 PM Page 70

Gerber, Elisabeth R. 1996. “Legislative Response tothe Threat of Popular Initiatives,” Amer. J. Polit. Sci.40:1, pp. 99–128.

Gerber, Elisabeth R. 1999. The Populist Paradox: In-terest Group Influence and the Promise of DirectLegislation. Princeton: Princeton U. Press.

Gerber, Elisabeth R. and Rebecca B. Morton. 1998a.“Primary Election Systems and Representation,” J.Law, Econ. Org. 14:2, pp. 304–24.

———. 1998b. “Electoral Institutions and Party Com-petition: The Effects of Nomination Procedures onElectoral Coalition Formation,” manuscript, UCLAand U. Iowa.

Gilligan, Thomas W. and John G. Matsusaka. 1995.“Deviations from Constituent Interests: The Role ofLegislative Structure and Political Parties in theStates,” Econ. Inquiry 33:3, pp. 383–401.

Goldstein, Morris and Geoffrey Woglom. 1992. “Market-based Fiscal Discipline in Monetary Unions: Evidencefrom the U.S. Municipal Bond Market,” in Establish-ing a Central Bank: Issues in Europe and Lessons fromthe U.S. M.B. Canzoneri, V. Grilli and P.R. Masson,eds. NY: Cambridge U. Press, pp. 228–70.

Green, Donald P. and Ian Shapiro. 1994. Pathologies ofRational Choice Theory: A Critique of Applicationsin Political Science. New Haven: Yale U. Press.

Grogan, Colleen M. 1994. “Political-Economic FactorsInfluencing State Medicaid Policy,” Polit. ResearchQuart. 47:3, pp. 589–623.

Grossman, Gene M. and Elhanan Helpman. 1996.“Electoral Competition and Special Interest Poli-tics,” Rev. Econ. Stud. 63:2, pp. 265–86.

Hanssen, F. Andrew. 1999. “The Effect of Judicial In-stitutions on Uncertainty and the Rate of Litigation:The Election versus Appointment of State Judges,”J. Legal Stud. 28:1, pp. 205–32.

———. 2000. “Independent Courts and Administra-tive Agencies: An Empirical Analysis of the States,”J. Law, Econ. Org. 16:2, pp. 534–71.

———. 2001. “Is there a Politically Optimal Level ofJudicial Independence?” Stanford Law School, JohnM. Olin Work. Paper 218.

Harris, Malcolm C. and Peter Navarro. 1983. DoesElecting Public Utility Commissioners Bring LowerElectric Rates?” Public Utilities Fortnightly 112, pp.23–88.

Highton, Benjamin. 1997. “Easy Registration andVoter Turnout,” J. Politics 59:2, pp. 565–75.

Hill, Kim Quaile; Jan E. Leighley and Angela Hinton-Anderson. 1995. “Lower-Class Mobilization and Pol-icy Linkage in the U.S. States,” Amer. J. Polit. Sci.39:1, pp. 75–86.

Holbrook, Thomas M. and Emily van Dunk. 1993.“Electoral Competition in the American States,”Amer. Polit. Sci. Rev. 87:4, pp. 955–62.

Holsey, Cheryl M. and Thomas E. Borcherding. 1997.“Why Does Government’s Share of National IncomeGrow? An Assessment of the Recent Literature onthe U.S. Experience,” in Perspectives on PublicChoice: A Handbook. Dennis C. Mueller, ed. Cam-bridge: Cambridge U. Press, pp. 562–90.

Holtz-Eakin, Douglas. 1988. “The Line Item Veto andPublic Sector Budgets: Evidence from the States,” J.Public Econ. 36:3, pp. 269–92.

Huckshorn, Robert J.. 1985. “Who Gave it? Who GotIt? The Enforcement of Campaign Finance Laws inthe States,” J. Politics 47:3, pp. 773–89.

Husted, Thomas A. and Lawrence W. Kenny. 1997.“The Effect of the Expansion of the Voting Fran-chise on the Size of Government,” J. Polit. Econ.105:1, pp. 54–82.

Jones, Ruth S. 1981. “State Public Campaign Finance:Implications for Partisan Politics,” Amer. J. Polit. Sci.25:2, pp. 342–61.

Joskow, Paul L. and Nancy L. Rose. 1989. “The Effectsof Economic Regulation,” in Handbook of IndustrialOrganization, vol. 2. Richard Schmalensee andRobert Willig, eds. Amsterdam: North Holland, pp.1449–506.

Kendall, Maurice G. and Alan Stewart. 1950. “TheLaw of the Cubic Proportion in Election Results,”Brit. J. Sociology 1, pp. 183–96.

Kenyon, Daphne and Karen Benker. 1984. “Fiscal Dis-cipline: Lessons from the State Experience,” Nat.Tax J. 37, pp. 433–46.

Key, Valdimer O., Jr. with Alexander Heard. 1950.Southern Politics in State and Nation. NY: A.A.Knopf.

Kiewiet, D. Roderick and Kristin Szakaly. 1996. “Con-stitutional Limitations on Borrowing: An Analysis ofState Bond Indebtedness,” J. Law, Econ. Org. 12:1,pp. 62–97.

Knight, Brian G. 2000. “Supermajority Voting Require-ments for Tax Increases: Evidence from the States,”J. Public Econ. 76, pp. 41–67.

Kone, Susan L. and Richard F. Winters. 1993. “Taxesand Voting: Electoral Redistribution in the Ameri-can States,” J. Politics 55:1, pp. 22–40.

Ladd, Helen F. and Nicolaus Tideman, eds. 1981. Taxand Expenditure Limitations. Washington, DC: Ur-ban Institute.

Laffont, Jean-Jacques. 1999. Incentives and PoliticalEconomy. Oxford: Oxford U. Press.

Lascher, Edward L., Jr; Michael G. Hagen and StevenA. Rochlin. 1996. “Gun Behind the Door? Ballot Ini-tiatives, State Policies and Public Opinion,” J. Poli-tics 58:3, pp. 760–75.

Levitt, Steven D. 1994. “Using Repeat Challengers toEstimate the Effect of Campaign Spending on Elec-tion Outcomes in the U.S. House,” J. Polit. Econ.102:4, pp. 777–98.

———. 1995. “Policy Watch: Congressional Cam-paign Finance Reform,” J. Econ. Perspect. 9:1, pp.183–93.

Lijphart, Arend. 1999. Patterns of Democracy: Gov-ernment Forms and Performance in Thirty-sixCountries. New Haven: Yale U. Press.

Lindbeck, Assar and Jorgen Weibull. 1993. “A Modelof Political Equilibrium in a Representative Democ-racy,” J. Public Econ. 51:2, pp. 195–209.

List, John and Daniel Sturm. 2002. “Politics and Envi-ronmental Policy: Theory and Evidence from U.S.States,” unpub. manuscript: U. Maryland and U.Munich.

Lott, John R., Jr. 2000. “A Simple Explanation for WhyCampaign Expenditures Are Increasing: The Gov-ernment Is Getting Bigger,” J. Law Econ. 43:2, pp.359–93.


Article 1 2/21/03 4:17 PM Page 71

Lott, John R., Jr.; and Stephen G. Bronars. 1993. “TimeSeries Evidence on Shirking in the U.S. House ofRepresentatives,” Public Choice 76, pp. 125–49.

Lott, John R., Jr. and Lawrence W. Kenny. 1999. “DidWomen’s Suffrage Change the Size and Scope ofGovernment?” J. Polit. Econ. 107:6, pp. 1163–98.

Lowry, Robert C.; James E. Alt and Karen E. Ferree.1998. “Fiscal Policy Outcomes and Electoral Ac-countability in American States,” Amer. Polit. Sci.Rev. 92:4, pp. 759–74.

Lowry, Robert C. and James E. Alt. 2001. “A VisibleHand? Bond Markets, Political Parties, BalancedBudget Laws, and State Government Debt,” Econ.Politics 13:1, pp. 49–72.

Magleby, David B. 1984. Direct Legislation: Voting onBallot Propositions in the United States. Baltimore:Johns Hopkins U. Press.

Maskin, Eric and Jean Tirole. 2001. “The Politician andthe Judge,” unpub. manuscript, U. Toulouse and In-stitute for Advanced Study, Princeton.

Matsusaka, John G. 1992. “Economics of Direct Legis-lation,” Quart. J. Econ. 107:2, pp. 541–71.

———. 1995. “Fiscal Effects of Voter Initiative: Evi-dence from the Last 30 Years,” J. Polit. Econ. 103:3,pp. 587–623.

———. 2000. “Problems with a Methodology Used toEvaluate the Voter Initiative,” unpub. manuscript,U. Southern California.

Matsusaka, John G. and Nolan M. McCarty. 1999. “Po-litical Resource Allocation: Benefits and Costs ofVoter Initiatives,” manuscript, Columbia U.

Milesi-Ferretti; Gian Maria, Roberto Perrotti andMassimo Rostag 2002. “Electoral Systems and Pub-lic Spending,” Quart. J. Econ. 117:2, pp. 609–58.

McArthur, John and Stephen V. Marks. 1988. “Con-stituent Interest vs. Legislator Ideology: The Role ofPolitical Opportunity Cost,” Econ. Inquiry 26:3, pp.461–70.

Morton, Rebecca B. and Charles Cameron. 1992.“Elections and the Theory of Campaign Contribu-tions: A Survey and Critical Analysis,” Econ. Politics4, pp. 79–108.

Nannestad, Peter and Martin Paldam. 1994. “The VP-Function: A Survey of the Literature on Vote andPopularity Functions after 25 Years,” Public Choice79:3–4, pp. 213–45.

Navarro, Peter. 1982. “Public Utility Commission Reg-ulation: Performance, Determinants and EnergyPolicy Impacts,” Energy J. 3:2, pp. 119–39.

Niemi, Richard G.; Harold W. Stanley and Ronald J.Vogel. 1995. “State Economies and State Taxes: DoVoters Hold Governors Accountable?” Amer. J. Polit.Sci. 39:4, pp. 936–57.

North, Douglass. 1990. Institutions, InstitutionalChange and Economic Performance. Cambridge:Cambridge U. Press.

Osborne, Martin J. and Al Slivinski. 1996. “A Model ofPolitical Competition with Citizen Candidates,”Quart. J. Econ. 111:1, pp. 65–96.

Pande, Rohini. 2001. “Minority Representation andPolicy Choices: The Significance of Legislator Iden-tity,” manuscript, Columbia U.

Peltzman, Sam. 1976. “Toward a More General Theoryof Regulation,” J. Law Econ. 19:2, pp. 211–40.

———. 1987. “Economic Conditions and Gubernato-rial Elections,” Amer. Econ. Rev. Papers Proceedings77:2, pp. 293–97.

———. 1992. “Voters as Fiscal Conservatives,” Quart.J. Econ. 107:2, pp. 327–61.

Persson, Torsten and Lars E.O. Svensson. 1989. “Whya Stubborn Conservative Would Run a Deficit: Pol-icy with Time-Inconsistent Preferences,” Quart. J.Econ. 104:2, pp. 325–45.

Persson, Torsten and Guido Tabellini. 1999. “The Sizeand Scope of Government: Comparative Politicswith Rational Politicians,” Europ. Econ. Rev. 43:4–6,pp. 699–735.

———. 2000. Political Economics: Explaining Eco-nomic Policy. Cambridge: MIT Press.

———. 2002. Economic Policy in RepresentativeDemocracies. Cambridge, MIT Press.

Plotnick, Robert D. and Richard F. Winters. 1985. “APolitico-Economic Theory of Income Redistribu-tion,” Amer. Polit. Sci. Rev. 79:2, pp. 458–73.

Pommerehne, Werner W. 1990. “The Empirical Rele-vance of Comparative Institutional Analysis,” Europ.Econ. Rev. 34:2–3, pp. 458–69.

Poterba, James M.. 1994a. “State Responses to FiscalCrises: The Effects of Budgetary Institutions andPolitics,” J. Polit. Econ. 102:4, pp. 799–821.

———.. 1994b. “Tax Policy Changes and the Pricing ofTax-Exempt Bonds,” in Handbook of MunicipalBonds. Susan C. Heide, Robert A. Klein and JessLederman, eds. Chicago: Probus.

———. 1997. “Do Budget Rules Work?” in Fiscal Pol-icy Lessons from Economic Research. Alan J. Auer-bach, ed. Cambridge: MIT Press, pp. 53–86.

Poterba, James M. and Kim Rueben. 1999. “State In-stitutions and the U.S. Municipal Bond Market,” inFiscal Institutions and Fiscal Performance. James M.Poterba and Jurgen von Hagen, eds. Chicago: U.Chicago Press.

———. 2001. “Fiscal News, State Budget Rules, andTax-Exempt Bond Yields,” J. Urban Econ. 50, pp.537–62.

Primeaux, Walter and Patrick Mann. 1986. “RegulatorSelection Methods and Electricity Prices,” LandEcon. 63:1, pp. 1–13.

Radcliff, Benjamin and Martin Saiz. 1995. “Race,Turnout, and Public Policy in the American States,”Polit. Research Quart. 48:3, pp. 775–94.

Roemer, John E. 2001. Political Competition: Theoryand Applications. Cambridge: Harvard U. Press.

Rogers, Diane Lim and John H. Rogers. 2000. “Politi-cal Competition and State Government Size: DoTighter Elections Produce Looser Budgets?” PublicChoice 105:1, pp. 1–21.

Rogoff, Kenneth. 1990. “Equilibrium Political BudgetCycles,” Amer. Econ. Rev. 80:1, pp. 21–36.

Romer, Thomas and Howard Rosenthal. 1978. “Politi-cal Resource Allocation, Controlled Agendas, andthe Status Quo,” Public Choice 33:4, pp. 27–43.

Rueben, Kim. 1997. “Tax Limitations and GovernmentGrowth: The Effect of State Tax and ExpenditureLimits on State and Local Government,” manu-script, Cal. Institute Public Policy.

Sass, Tim R. and Bobby J. Pittman Jr. 2000. “TheChanging Impact of Electoral Structure on Black


Article 1 2/21/03 4:17 PM Page 72

Representation in the South, 1970–96,” PublicChoice 104:3–4, pp. 369–88.

Shadbegian, Ronald J. 1996. “Do Tax and ExpenditureLimitations Affect the Size and Growth of StateGovernment?” Contemporary Econ. Pol. 14:1, pp.22–35.

Shepsle, Kenneth A. and Barry R. Weingast. 1981.“Structure-Induced Equilibrium and LegislativeChoice,” Public Choice 37:3, pp. 503–19.

Stigler, George J. and Claire Friedland. 1962. “WhatCan Regulators Regulate? The Case of Electricity,”J. Law Econ. 5, pp. 1–16.

Stratmann, Thomas and Francisco J. Aparicio-Castillo.2002. “Competition Policy for Elections: Do Cam-paign Contribution Limits Matter?” unpub. manu-script, George Mason U.

Smart, Susan R. 1994. “The Consequences of Appoint-ment Methods and Party Control for Telecommuni-cations Pricing,” J. Econ. Manage. Strategy 3:2, pp.301–23.

Strumpf, Koleman. 2000. “Does Government Decen-tralization Increase Policy Innovation?” J. PublicEcon. Theory, forthcoming.

Tabellini, Guido and Alberto Alesina. 1990. “Voting onthe Budget Deficit,” Amer. Econ. Rev. 80:1, pp. 37–49.

Von Hagen, Jurgen. 1991. “A Note on the EmpiricalEffectiveness of Formal Fiscal Restraints” J. PublicEcon. 44:2, pp. 199–210.

Thomas, Sue. 1994. How Women Legislate. NY: Ox-ford U. Press.

Walker, Jack L. 1969. “The Diffusion of Innovationamong the American States,” Amer. Polit. Sci. Rev.63:3, pp. 880–99.

Winters, Richard. 1976. “Party Control and PolicyChange,” Amer. J. Polit. Sci. 20:4, pp. 597–636.

Wolfers, Justin. 2002. “Are Voters Rational? Evidencefrom Gubernatorial Elections,” Stanford BusinessSchool work. paper 1730.

Wolfinger, Raymond E. and Steven J. Rosenstone.1980. Who Votes? New Haven: Yale U. Press.

Wright, Gerald C., Jr.; Robert S. Erikson and John P.McIver. 1987. “Public Opinion and Policy Liberal-ism in the American States,” Amer. J. Polit. Sci. 31:4,pp. 980–1001.

Zax, Jeffrey S. 1989. “Initiatives and Government Ex-penditures,” Public Choice 63:3, pp. 267–77.


Article 1 2/21/03 4:17 PM Page 73

Political Institutions and Policy Choices: Evidence from...

Documents

Transcript of Political Institutions and Policy Choices: Evidence from...