Coalition Government, Legislative Institutions, andPublic Policy in Parliamentary Democracies

Lanny W. Martin Bocconi UniversityGeorg Vanberg Duke University

Abstract: Most democracies are governed by coalitions, comprising multiple political parties with conflicting policy positions.The prevalence of these governments poses a significant question: Which parties’ electoral commitments are ultimatelyreflected in government policy? Recent theories have challenged our understanding of multiparty government, arguing thatthe relative influence of coalition parties depends crucially on institutional context. Specifically, where institutions allowcredible enforcement of bargains, policy should reflect a compromise among all governing parties; where such institutionsare absent, the preferences of parties controlling the relevant ministries should prevail. Critically, empirical work has thusfar failed to provide direct evidence for this conditional relationship. Analyzing changes in social protection policies in 15parliamentary democracies, we provide the first systematic evidence that the strength of legislative institutions significantlyshapes the relative policy influence of coalition parties. Our findings have implications for our understanding of coalitiongovernment, policymaking, and electoral responsiveness.

Replication Materials: The data and materials required to verify the computational reproducibility of the results,procedures and analyses in this article are available on the American Journal of Political Science Dataverse within theHarvard Dataverse Network, at: https://doi.org/10.7910/DVN/YU84D7.

I n 2010, for the first time since World War II, a coali-tion government assumed power in the United King-dom. Upon taking office, the Conservatives and Lib-

eral Democrats issued a coalition agreement that sketchedpolicy compromises across 31 issues. And yet, by the nextgeneral election in 2015, the prevailing perception wasthat the Liberal Democrats had not been able to exer-cise much influence, and that the government’s policies—despite the coalition agreement—had largely been reflec-tive of Conservative priorities.1 At least in part, theseperceptions contributed to the Liberal Democrats’ disas-trous electoral performance in that election. In contrast,

Lanny W. Martin is Professor, Department of Social and Political Sciences, Dondena Centre for Research on Social Dynamics and PublicPolicy, Bocconi University, Via Roentgen, 1, 20136 Milan, Italy (lanny.martin@unibocconi.it). Georg Vanberg is Professor, Department ofPolitical Science, Duke University, 140 Science Drive, Box 90204, Durham, NC 27708 (georg.vanberg@duke.edu).

We wish to thank seminar participants at Bocconi University, Duke University, the Hertie School of Governance, Humboldt University,Princeton University, Rice University, Tel Aviv University, the University of Cologne, the University of Illinois at Urbana-Champaign, andYale University for helpful comments on previous versions of this article. We also thank Lyle Scruggs for the information and helpful advicehe provided on the Comparative Welfare Entitlements Dataset.

1See https://www.theguardian.com/politics/2015/apr/15/how-much-of-the-liberal-democrats-2010-election-manifesto-was-implementedfor a Guardian article that assesses the record of the Liberal Democrats under the coalition. The coalition agreement is available athttps://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/78977/coalition_programme_for_government.pdf.

2For an assessment of the record of the Red-Green coalition by a major German newspaper, see https://www.faz.net/aktuell/politik/regierungsbilanz-sieben-jahre-rot-gruen-1260372.html. The coalition agreement is available at https://www.spd.de/fileadmin/Dokumente/Beschluesse/Bundesparteitag/koalitionsvertrag_bundesparteitag_bonn_1998.pdf.

between 1998 and 2005, Germany was ruled by a coali-tion of Social Democrats and Greens. Like the LiberalDemocrats and Conservatives in the U.K., these partiesdisagreed on key issues (particularly energy and the en-vironment). The partners arrived at compromises, whichwere made public in a coalition agreement. Unlike in theU.K., however, the coalition’s subsequent policy choiceslargely reflected their agreement. The policy record of thegovernment clearly bore the imprint of both partners.2

The contrasting experiences of these governmentspose an obvious question. Most democratic governmentsaround the world are coalitions, comprising multiple

American Journal of Political Science, Vol. 00, No. 0, xxxx 2019, Pp. 1–16

C©2019, Midwest Political Science Association DOI: 10.1111/ajps.12453

1

2 LANNY W. MARTIN AND GEORG VANBERG

parties that have conflicting positions on at least someissues. Given such conflicts, whose commitments are ul-timately reflected in what these governments do? Thisquestion is of clear significance. For party elites involvedin coalition negotiations, expectations regarding the poli-cies likely to result from alternative governments shouldshape their preferences over which cabinet to choose. Forvoters, knowing which parties they should hold respon-sible for policy outcomes, or understanding the implica-tions of a particular vote choice, requires understandingthe relative influence of coalition parties in policymaking.

The existing literature offers two competing perspec-tives on this puzzle that correspond, roughly speaking,to the experiences of the coalitions discussed above. One,forcefully expressed in Laver and Shepsle’s (1996) seminal“portfolio allocation” model, stresses the dominant role ofcabinet ministers. It leads to the expectation that policy inany area primarily reflects the commitments of the partythat occupies the relevant ministry. Given that the LiberalDemocrats controlled only a small number of portfolios(and not the most prominent ones), this perspective sug-gests that the coalition’s policy should be dominated bythe Conservatives. In contrast, other scholars have arguedthat coalitions adopt policies that represent a compromiseamong their members (e.g., Becher 2010; Goodhart 2013;Powell 2000). Larger parties may have more influence onthis compromise, but policy will reflect the positions ofall parties in government. Clearly, the experience of theRed–Green coalition more closely resembles this account.

Recent developments in the coalition literature com-plicate these answers—and potentially reconcile the con-trasting experiences of our two coalitions. While the twoperspectives just outlined are “institutions-free,” the keycontribution of these advances is to argue that the rela-tive influence of parties’ commitments is conditional onthe institutional framework within which policy is made(Martin and Vanberg 2005, 2011; Thies 2001). Institu-tional context is critical because effective implementationof compromise agreements requires that parties have theability to monitor and—if necessary—to challenge the ac-tions of cabinet ministers, who play a central role in draft-ing and implementing legislative proposals. Where suchinstitutions are present, one would expect compromiseagreements to prevail. Where such institutional resourcesare not available, it is difficult to control the discretionministers enjoy. As a result, the partisan preferences ofministerial parties are likely to be reflected in policy withintheir jurisdiction. In short, institutional context is centralto multiparty governance.

Although these developments are reshaping the coali-tion literature, critical gaps remain. Most seriously, schol-ars have produced only indirect evidence in support of

the theoretical conjectures. For example, several stud-ies focus on the appointment of junior ministers andparliamentary committee chairs (Carroll and Cox 2012;Lipsmeyer and Pierce 2011; Thies 2001). Appointmentpatterns to these offices suggest that parties view themas checks on “hostile” ministers. But there is no directevidence that they are effective in this task. Other work hasstudied legislative review of ministerial proposals. For ex-ample, in an earlier study (Martin and Vanberg 2011), wedemonstrate that when legislative “policing” institutionsare strong, ministerial bills are more heavily scrutinizedand amended when they concern issues that divide thecoalition (see also Pedrazzani and Zucchini 2013). How-ever, greater legislative activity does not necessarily implythat coalition partners effectively constrain ministerialdrift. As Goodhart (2013, 207) points out, “The existenceof monitoring and control mechanisms . . . does not itselfprove the case against ministerial discretion . . . . [These]mechanisms may act to control policy deviations, butimperfectly, so that ministerial discretion co-exists withactive attempts to contain it.”

In other words, a critical shortcoming in the coalitionpolitics literature is that recent theoretical advances,which have recast our understanding of the policymakingprocess, have not yet been matched with empirical con-tributions that provide direct evidence that institutionalcontext conditions the policies that multiparty govern-ments adopt. This shortcoming can largely be attributedto the fact that direct evidence requires detailed measuresof policy outcomes that can be reliably tied to the compet-ing preferences of parties—a significant challenge that hasinstead steered scholars toward more indirect approaches.

In this study, we seek to narrow this gap. To do so, wetake advantage of an important new data set on welfarestate entitlements (Scruggs, Jahn, and Kuitto 2014) thatallows us to examine changes in two areas of socialprotection—unemployment benefits programs, andshort-term disability and sickness benefits programs—in15 parliamentary democracies over a period of roughly40 years. These data provide a detailed picture of policyoutcomes in two highly salient policy areas. Focusingon a particularly important set of institutions in par-liamentary democracies—namely, legislative committeesystems—we provide systematic evidence that differencesin legislative institutions lead to considerable differencesin the relative impact of ministerial and cabinet partieson policy outcomes. Our results demonstrate that in leg-islative environments that privilege ministerial proposals,and make it difficult for coalition partners to monitorand enforce compromise agreements, policy choices aredriven by the commitments of parties that control rele-vant ministries. In contrast, in legislative environments

COALITIONS, LEGISLATURES, AND PUBLIC POLICY 3

that “level the playing field” by reducing the informa-tional advantage of ministers and providing effectiveopportunities for challenging and amending ministerialproposals, policies reflect a compromise position amongall coalition partners. As we discuss in the conclusion,these findings have significant implications for ourunderstanding of coalition governance, electoral respon-siveness, and democratic representation.

Institutional Context and PolicyResponsiveness

Cabinets occupy a key policy position in parliamentarysystems since major legislative initiatives can only moveforward with their approval. At the same time, given thescope and complexity of public policy, formulating effec-tive legislative proposals requires technical expertise andinformation, and no cabinet minister is in a position tomaster the required knowledge in all areas of governmen-tal activity. As a result, policymaking involves a signif-icant amount of delegation to ministers who have pri-mary responsibility for drafting policy proposals withintheir jurisdiction (working, of course, through the min-isterial bureaucracy they oversee). Such delegation, inturn, implies that ministers enjoy—at least potentially—considerable discretion to shape policy within their juris-diction (Laver and Shepsle 1996).

Under single-party government, this privileged po-sition of ministers may be (relatively) unproblematic, atleast in the absence of significant intraparty heterogene-ity.3 But for multiparty governments, it creates challenges.A central feature that distinguishes coalitions from single-party governments is that parties in a coalition mustgovern jointly, but they are held to account separatelyat election time. This creates tensions. Each party facesincentives to play to its target audiences by steering policyin directions favored by its supporters (or at least to beperceived to be doing so, an important aspect to which wereturn below). The privileged position of cabinet minis-ters implies that parties have the opportunity to do so.Cabinet ministers can use their position to pull policywithin their jurisdiction toward positions preferred bytheir party.

Laver and Shepsle’s (1996) model is built around thislogic. The lynchpin for their argument is the assump-tion that in negotiating a coalition, party leaders should

3The coalition literature has typically treated parties as unitary ac-tors, both as a theoretical simplification and because of the scarcityof data on intraparty preferences. We adopt this approach here,primarily because the data we rely on are only available at theparty level.

expect that each party will take advantage of ministe-rial discretion to pursue the party’s preferred policy inthose jurisdictions under its control. This implies thatincentive-compatible coalition policies are located at theintersections of the dimension-by-dimension ideal pointsof parties. To illustrate, consider a two-dimensional policyspace and two parties, A and B . Parties have convex pref-erences, as depicted in Figure 1. Suppose the parties arecontemplating a coalition in which B receives the port-folio for Dimension 1, and A receives the portfolio forDimension 2. Which policy outcome should the partiesexpect from this portfolio allocation?

Because preferences are inseparable, the policy pre-ferred by each party in a dimension depends on the pol-icy choice in the other dimension; these “induced idealpoints” are indicated by the solid lines through eachparty’s ideal point.4 Where these two lines intersect (thepoint labeled 1B 2A), each minister implements the pol-icy she most prefers in her jurisdiction, given the policyadopted by the other minister. Put differently, the aggre-gate policy outcome 1B 2A is incentive-compatible. How-ever, the policy that results from ministerial autonomy isinefficient. In Figure 1, policies in the overlap area of theindifference curves through 1B 2A, such as policy X , arepreferred by both parties to ministerial autonomy.

Although parties prefer a compromise like X to min-isterial autonomy, implementing it poses a challenge. Un-like 1B 2A, a compromise at X is not (directly) incentive-compatible. To achieve X , Minister A must pursue policyx2, and Minister B must pursue policy x1. But doing sois not in their immediate interests. As indicated by thearrows in the two figures, both ministers would benefitby pursuing a different policy. Moreover, the privilegedposition of ministers provides them with the opportunityto do precisely this. Of course, this is Laver and Shep-sle’s point—implementing compromise agreements con-stitutes a credible commitment problem.

One function of institutions is to resolve problemsof credible commitment (North 1993). This insight hasrecently been brought to bear on coalition governance byseveral scholars who have assumed that, in certain cases,parties may have such institutions available. Because com-promise policies are typically preferable to the results ofministerial autonomy, parties have incentives to make theenforcement of compromises possible. To do this, partiesmust be in position to scrutinize ministerial proposals, todevelop alternatives, and to push for these in the policy

4In their original formulation, Laver and Shepsle (1996) rely onEuclidean preferences. We reformulate their argument with moregeneral convex preferences to highlight that the ministerial au-tonomy argument, and qualifications of it, do not depend on arestrictive assumption about preferences.

4 LANNY W. MARTIN AND GEORG VANBERG

FIGURE 1 Government Policy under Ministerial Autonomy

process. Where policymaking institutions enable them toengage in these actions, coalitions may not be condemnedto tolerating ministerial autonomy, but can instead im-plement genuine compromises.

In an important study, Thies (2001) points to the roleplayed by junior ministers (JMs) in this regard. JMs, whoare political appointees directly below a cabinet minister,are in a position to gather information on proposals be-ing developed in a department. As a result, appointing“shadow” JMs in ministries controlled by their partnersmay allow parties to “keep tabs” on hostile ministers. Theappointment patterns of JMs provide clear support forthis argument (Lipsmeyer and Pierce 2011; Thies 2001).Similarly, other scholars have highlighted the role of cabi-net committees and inner cabinets that allow scrutiny anddebate of draft bills (Andeweg and Timmermans 2008;Muller and Strøm 2000).

These institutions represent significant opportunitiesto rein in ministers. But they face two limitations. One is aresource constraint. Intra-cabinet working groups cannotbe used as an oversight tool for most legislation. Giventhe demands on ministers, such groups are limited to par-ticularly salient initiatives. Similarly, small parties cannotappoint junior ministers to all partner ministries becausethey are only entitled to a limited number of such posts.Moreover, junior ministers have access to information,but they may not be in a position to formulate alternativepolicies or to force ministers to change draft initiatives.

A second constraint is political. These institutionsoperate prior to the public “unveiling” of a bill. This is

significant because the introduction of a legislative pro-posal provides a unique opportunity for position tak-ing by ministers (and their parties), often achieved inpress releases, news conferences, and interviews. Even ifparties expect a legislative proposal to be amended, theycan point to the bill as introduced as an indication of theparty’s “true” intentions. This implies that parties are notindifferent as to when in the policy process ministerialdrift is corrected. All things being equal, parties prefersuch changes to occur after a bill has been introduced,and credit-claiming points have been scored (Martin andVanberg 2005, 2011).

These constraints do not apply to the legislative pro-cess. All major policy initiatives must be presented toparliament, where ministerial proposals can be amended.Moreover, legislative consideration of a bill commencesafter its introduction—implying that ministers have hadan opportunity to engage in position taking. This sug-gests that parliaments may play a central role in enforcingcoalition compromises. But the parliamentary arena canonly be effective for this purpose if its internal organiza-tion allows parties to reduce the informational advantageof ministers, to develop alternatives to ministerial pro-posals, and to push for these to be implemented. Scholarshave identified a number of institutions that matter inthis context (Harfst and Schnapp 2003; Martin and Van-berg 2011; Mattson and Strøm 1995). One set of featuresrelates to the ability to scrutinize legislation:

(1) Number of legislative committees: The largerthe number of committees, the greater the

COALITIONS, LEGISLATURES, AND PUBLIC POLICY 5

opportunities for specialization and develop-ment of policy expertise.

(2) Correspondence of committees to cabinet min-istries: Committees that are organized along thejurisdictional lines of ministries enhance theability to develop relevant policy expertise.

(3) Size of legislative committees: Large committeesare less effective because they tend to becomeunwieldy and encourage free riding.

(4) Binding plenary debate before committee stage:Binding plenary debate before committee delib-erations curtails opportunities for scrutiny andchange.

(5) Right to compel witnesses and documents: Com-mittees with rights to compel testimony and doc-uments (including from senior civil servants andministers) have greater ability to scrutinize leg-islation and develop alternatives.

(6) Urgency procedure: If ministers have the right tocurtail legislative debate and shorten the legisla-tive timetable, effective scrutiny is more difficult.

Two additional features relate to the ability to push effec-tively for changes to ministerial bills:

(7) Rewrite authority: Committees that can rewritedraft bills (i.e., their version of a bill becomes thefloor agenda) are in a more powerful position topush for change than committees that can onlysponsor amendments.

(8) Guillotine procedure: If ministers have the rightto reject amendments or force an up-or-downvote on a bill, the ability to challenge ministerialproposals is reduced.

We are now in a position to outline the key implica-tion of this argument. Coalition governments must con-front the tensions introduced by the need to delegatepolicymaking authority to ministers. Pursuing coalitioncompromises—which parties have strong incentives todo—requires an ability to rein in ministerial drift. Leg-islative environments that feature “strong” internal insti-tutions (i.e., those that make scrutiny and change easier toaccomplish) allow coalition partners to do so. As a result,policy will not be driven exclusively by the preferences ofministerial parties, but will reflect a compromise amongthe positions of all coalition members. In contrast, in“weak” legislative environments, it is difficult to enforcecompromise agreements. As a consequence, policy willreflect the position of the party controlling the relevantministry. This argument therefore implies the following:

Conjecture. As divisions increase between the preferencesof the party of the minister and the preferences of the gov-

ernment as a whole, policy changes will be more sensitiveto the preferences of the minister’s party when legislativeinstitutions are weak, but more sensitive to the preferencesof the government as a whole when legislative institutionsare strong.

In short, institutional context—and in particular, the na-ture of legislative policing institutions—are critical forcoalition policymaking. As a result, policy outcomes de-pend on the interplay of partisan preferences and institu-tional structure.

Party Commitments, LegislativePolicing, and Social Protection

Policies

As we argued earlier, rigorous empirical evaluation of theargument that coalition policymaking is conditioned byinstitutional context has been hampered by the difficultyof obtaining reliable data on policy outcomes. As aconsequence, scholars have relied on indirect, circum-stantial evidence that focuses on legislative activity orappointment patterns of junior ministers or committeechairs. We seek to overcome this limitation and providedirect evidence that legislative institutions conditionpolicymaking under multiparty governments. We do soby examining a new data set—the Comparative WelfareEntitlements Dataset 2 (CWED2)—that documentschanges in social insurance schemes in several Europeandemocracies (Scruggs, Jahn, and Kuitto 2014). Inparticular, we examine changes in two types of socialprotection programs: those dealing with unemploymentinsurance benefits and those dealing with short-termdisability and sickness insurance benefits. Reforms tothese programs have long been politically contentious inadvanced industrial democracies, and increasingly so inthe era of welfare state retrenchment (Korpi and Palme2003). They are also of broad economic significance:They have implications for the national budget, for thelevel of spending in the economy, for the profitabilityof firms, and for the quality of life for vulnerablegroups in the population. As a result, we suspect thatthe policymaking process surrounding these programsprovides a useful testing ground for our argument.5

5Naturally, it would be desirable to expand this analysis to other pol-icy areas. Currently, the only other policy area for which similarlydetailed data are available is pensions (Scruggs, Jahn, and Kuitto2014). However, reform efforts in this policy area—perhaps due tothe sheer size of the aging population in developed countries—tendto be very slow-moving, typically spanning several successive gov-ernments. For example, it is not unusual for pensions reform to takeapproximately 20 years to phase in completely (Kalisch and Aman

6 LANNY W. MARTIN AND GEORG VANBERG

Our analysis encompasses the following 15 coun-tries and time periods: Austria (1973–2006), Belgium(1977–2007), Denmark (1974–2010), Finland (1973–2007), France (1977–2010), Germany (1973–2010),Great Britain (1973–2010), Greece (1983–2004), Ire-land (1973–2010), Italy (1973–2008), the Netherlands(1977–2006), Norway (1973–2005), Portugal (1983–2010), Spain (1985–2010), and Sweden (1973–2010). To-gether, the number of country-years (the units of analysis)in our sample is 498 for unemployment insurance and 496for sickness insurance.6

The CWED2 contains annual data on five character-istics associated with each type of insurance: (1) the aver-age benefit replacement rate, (2) the maximum numberof weeks of benefit duration, (3) the number of weeks ofemployment needed to qualify for benefits, (4) the num-ber of days an unemployed/sick person has to wait untilbenefits start, and (5) the share of the labor force coveredby benefits. For each insurance scheme, these compo-nents are aggregated into an index—an UnemploymentGenerosity Index and a Sickness Generosity Index—both ofwhich we present in Figure 2.7 As the figure shows, thereis substantial variation in both of the generosity indicesacross countries and over time. Although the indices tendto move together, they often do not—their correlationin our sample is moderate, slightly less than 0.60. More-over, the figure shows that, within countries, there areclear differences in the average level of generosity acrossthe two programs, with sickness benefits typically beingmore generous than unemployment benefits.8

To evaluate our expectations, we need to assess (1)whether this variation primarily reflects the policy po-

1998), which obviously makes it difficult to trace changes back toany single government. In contrast, changes to unemployment andsickness benefits tend to be more immediate, rather than graduallyphased in over several years, and thus are easier to attribute to thegovernment that initiated them.

6Our sample was chosen on the basis of data availability. The dataset we use on legislative policing institutions covers only thesecountries plus Luxembourg (a country for which we do not haveCWED2 data). The beginning date of each time period is deter-mined by the start date of the CWED2; the end date is determinedby the last election covered by the 2011 release of party manifestosby the Manifestos Project. There are two country-years for whichwe have information on unemployment benefits but not sicknessbenefits: Finland (1971) and France (1975).

7Each index for a given country-year is calculated as follows: Cover-age ∗ {2 ∗ z[Average benefit replacement rate] + z[ln(Benefit du-ration weeks)] − z[ln(Benefit qualification weeks)] − z[Waitingdays] + 12.5}, where z is the z-score of the associated componentnormalized on its cross-national, cross-temporal mean and stan-dard deviation. The start dates shown in the figure precede the startdate for the analysis because our models include lags.

8Because the indices are each standardized in terms of their respec-tive components (see note 7), they can be directly compared.

sitions of parties that have ministerial responsibility forthese programs or a compromise among the positionsof all parties in government, and (2) whether the rel-ative influence of ministers and their partners dependson the strength of legislative policing institutions.9 Tomeasure party policy positions, we make use of the Man-ifestos Project (MARPOR) data. While some caveats ap-ply (see, e.g., Lowe, Benoit, Mikhaylov, and Laver 2011),these data are the most comprehensive available in termsof countries, elections, and parties covered. Because theyare coded from parties’ election manifestos, the data arealso dynamic, reflecting shifts in party positions over time.

The raw manifestos data consist of the coding of man-ifesto (quasi-)sentences into 56 categories. Using thesedata to construct estimates of party positions on a givenpolicy dimension requires making choices about whichcategories to focus on, and how to aggregate them intoa substantively meaningful scale. Because our two policyprograms deal not only with issues of social protectionbut also (considering their impact on the profitability andglobal competitiveness of firms) with economic and in-dustrial policy more generally, parties’ positions in theseareas are likely to reflect concerns about the protection ofvulnerable groups as well as concerns about regulation ofthe labor market and economic growth. Indeed, changesin these programs are often embedded in broader eco-nomic reforms. Consider, for example, the Agenda 2010project implemented by the Social Democratic/Greengovernment in Germany in 2003, which focused on la-bor market reforms to reduce unemployment (including,among other things, reducing the duration of unemploy-ment benefits and shifting the costs of sick pay insurancefrom employers to employees after a certain length ofillness). In introducing this reform package, ChancellorGerhard Schroeder explicitly invoked both economic andsocial welfare considerations: “Our Agenda 2010 containssweeping structural reforms. By the end of this decade,these reforms will return Germany to a leadership posi-tion in economic prosperity and employment. In this way,these reforms will ensure intergenerational fairness andstrengthen the foundations of our communities” (Speechbefore the Bundestag, March 14, 2003). Because unem-ployment and sickness benefits schemes implicate bothsocial and economic concerns, we construct our LaborInsurance Benefits preference scale, defined a priori, usinga set of 10 manifestos categories relating specifically to

9We identified the ministries responsible for the two social protec-tion programs based on information we gathered from a variety ofsources, including historical descriptions provided by the ministriesthemselves, parliamentary documents detailing their competences,and the bills they introduced in the legislature. See Appendix A inthe supporting information (SI) for a list of these ministries.

COALITIONS, LEGISLATURES, AND PUBLIC POLICY 7

FIGURE 2 Social Insurance Generosity, by Country and Year

party statements on the welfare state, social protection,and economic regulation.10 Parties that pledge, on net,to lower regulations on business and cut back welfarestate benefits receive positive values, whereas parties thatpledge, on net, to increase market regulation and expandwelfare state benefits receive negative values.

For each government in our sample, we assign to theminister responsible for unemployment (sickness) policythe position of his or her party on our economic scale.Our first independent variable of interest, Conservatism

10Specifically, to capture “rightist” party positions (i.e., oppositionto greater benefits for the unemployed or sick/disabled), we usemanifesto categories 401, 402, 414, 505, and 702; to capture “leftist”party positions (i.e., support for greater benefits), we use categories403, 412, 503, 504, and 701. We first performed confirmatory factoranalyses, shown in S1 Appendix B, to ensure that the five items onthe right load together and that the five items on the left loadtogether. We then used the factor scores derived from that analysisto create relative weights for each category, since the factor loadingsindicate that some categories are more important than others inexplaining variance in either opposition to benefits (on the right) orsupport of benefits (on the left). We constructed our scale by usingthe log-odds ratio method recommended by Lowe et al. (2011), asalso detailed in S1 Appendix B.

of the Minister, is measured as the position of the relevantminister aggregated over a country-year.11 We also requirean estimate of the position of the government as a wholeon our scale. As in earlier work, we calculate the govern-ment compromise position as the seat-weighted averageposition of all parties in government (Martin and Vanberg2014). Again, the key conjecture we want to evaluate isthat, as divisions increase between the preferences of theparty of the minister and the preferences of the govern-ment as a whole, the institutional setting within whichpolicy is made becomes critical. Testing this conjecturerequires a measure that (1) separates the position of theminister from the position of the government as a wholeand (2) identifies situations of conflict between the rele-vant minister and the other coalition partners. In other

11Aggregation is necessary because the CWED2 generosity indicesare coded on a calendar-year basis, but there may be more than oneminister (and government) in office in a given year, and the datado not allow us to observe when over the course of the year changesto benefits were made. We weight the minister’s position by theproportion of the year in which the minister served, and then sumthe weighted measures across all ministers serving that year.

8 LANNY W. MARTIN AND GEORG VANBERG

words, we require a measure of the divergence betweenthe minister and the government compromise position.This independent variable, which we call Relative Gov-ernment Conservatism, is calculated as the position of thecompromise minus the position of the relevant minis-ter. A positive value for this variable indicates that thegovernment is to the right of the minister, whereas a neg-ative value indicates that the government is to the left ofthe minister.

In line with our argument, we interact this variable(which is also aggregated over a country-year) with a mea-sure of the Policing Strength of the legislature. In the pre-vious section, we described eight institutional features oflegislatures that serve to strengthen, or weaken, the abilityof coalition partners to effectively scrutinize and amendministerial initiatives. In previous work, we developed anindex of policing strength derived from a factor analy-sis of those institutions (Martin and Vanberg 2011). Oneshortcoming of this measure, for our present purposes,is that it is static (as it is based on averaged values of theeight institutional features over a period of several years).Accordingly, we now extend our data by gathering infor-mation on these (changing) features across time for the15 legislatures in our study. This has allowed us to createa dynamic measure of legislative policing strength, whichwe display in Figure 3.12

We arrange the countries in the figure in descendingorder of their average legislative policing strength, andfor presentational purposes, we divide them into threegroups: strong, moderate, and weak policing.13 Notably,our ranking corresponds closely to the ranking shown inour earlier work; for example, the same set of countriesare in the top five and in the bottom four, albeit withslight movements between countries at the top. But wecan also clearly see that policing strength has not been en-tirely static in most countries over time. The majority ofchanges have been due to modifications in the number oflegislative committees, the size of those committees, andthe degree to which they correspond to government min-istries; other features of policing strength have been ratherstable.14 Moreover, most changes have been gradual (usu-

12In SI Appendix C, we present the results of a new factor analysisof the eight institutional variables (based on information from ourset of country-years) that we use to construct the dynamic policingindex. We code all variables in terms of their values at the beginningof the legislative term. If one term ends in the same year that anotherterm begins, we use a time-weighted average value (rounded forinteger variables).

13Note that the policing strength scale differs across the three subsetsof graphs. In the analysis below, we use the individual country-yearscores, not the country-level averages.

14The major exception is France, which passed a constitutionalamendment in 2008 (going into effect in 2009) that gave par-

ally trending toward greater policing strength), but therehave been a few notable exceptions: the Netherlands in themid-1990s, when new standing orders were adopted thatdrastically reduced the number of committees (bringingpolicing strength down to a moderate level); Belgium inthe late 1970s, when big cuts were also made to the num-ber of committees; and Ireland in the mid-1990s, when anew system of permanent committees was finally put intoplace after several years of reform efforts.

Analysis and Findings

We focus on two dependent variables in our analysis:yearly change in a country’s level of generosity in unem-ployment benefits and yearly change in its level of generos-ity in short-term disability and sickness benefits. Giventhe longitudinal structure of our data, we use time-seriescross-section (TSCS) methods, which allow for estima-tion of the short-term and long-term effects of the covari-ates (all of which are measured prior to the beginning-of-year CWED2 measures of generosity). Thus, we canassess how our key theoretical variables—the economicconservatism of the minister and the conservatism of thegovernment relative to the minister (conditional on thestrength of legislative policing institutions)—translate toimmediate changes in the two social insurance schemesand to changes in subsequent years.

To isolate the effects of these variables, we considerthree macroeconomic indicators per model that are po-tential confounders, in that they may impact both theprovision of benefits and the economic positions takenby government parties. Specifically, in both models, wecontrol for two “supply side” factors relating to the capac-ity of governments to expand benefits: growth in GDP percapita and growth in the budget deficit. With regard to“demand side” factors, in our unemployment insurancemodel, we control for the (log) rate of unemployment.And in our sickness insurance model, we control for theshare of total yearly expenditure on health care that is fi-nanced by the government, as several studies have shownthat greater out-of-pocket spending by individuals onhealth care decreases their use of preventive services (see,e.g., Rezayatmand, Pavlova, and Groot 2013), which leadsto increases in illness-related employee absenteeism andthus the need for paid sick leave. As with our theoreticalvariables, we account for the short-term and long-termeffects of these variables.

liamentary committees rewrite authority (in addition to increas-ing the number of permanent committees from six to eight, anddecreasing their size).

COALITIONS, LEGISLATURES, AND PUBLIC POLICY 9

FIGURE 3 Policing Strength of 15 European Parliaments over Time

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

(10)(9)(8)(7)(6)

(11) (12) (13) (14) (15)

Note: Country rankings are based on the average policing strength score over the set of sample years. The averagescores are as follows: (Strong) Germany (0.79), Netherlands (0.76), Austria (0.76), Sweden (0.71), Denmark (0.63);(Moderate) Finland (0.43), Italy (0.42), Norway (0.41), Belgium (0.41), Portugal (0.30); (Weak) Spain (0.23), Greece(–0.50), France (–1.05), Ireland (–1.61), Great Britain (–2.42). Note that the policing strength scale differs across thethree subsets of graphs.

We also include lags of the dependent variable in ourmodels—specifically, a 1-year lag of the level of unem-ployment (sickness) generosity and a 1-year lag of thechange in generosity.15 As with the macroeconomic indi-cators, our rationale is that we wish to purge any sourcesof possible spurious association between our key theo-retical variables and policy change. We suspect that theelectoral stances taken by government parties on ben-efits depend in part on the generosity of the currentsystem. Moreover, the previous level of unemployment(sickness) generosity and the previous change in gen-erosity are likely to have a direct impact on the currentextent of changes. Governments inherit existing schemesthat set public expectations for what benefits will be inthe future. Even if governments would like to make sud-den, dramatic changes, doing so would potentially createuncertainty about future policy shifts, which could havedisruptive effects on the spending patterns and planningabilities of households and firms. Thus, given the incen-tives of governments to reduce uncertainty, we expect tosee considerable inertia in these areas.16

15Including the latter variable, when the former is included, isequivalent to including a 2-year lag of the level of generosity. Fol-lowing Hendry’s (1995) advice for choosing an appropriate lagstructure, we first estimated a model with multiple lags and then“tested down” (via t-tests) to determine how many lags to include.

16Some scholars have cautioned against the inclusion of laggeddependent variables, which may lead to biased coefficient estimates

Another issue we must consider concerns the pos-sibility of unobserved heterogeneity at the country levelin the provision of benefits. As Figure 2 makes clear, dif-ferences exist in the average level of generosity acrosscountries, which may be due to factors (e.g., a country’shistory or “culture”) that cannot easily be incorporatedinto an empirical model. If these unobserved factors alsoaffect the economic positions taken by political parties,we must be concerned about the possibility of a spuri-ous relationship between our theoretical variables andpolicy change.

The standard approach to dealing with unit-specificheterogeneity in TSCS data is to allow for country-specificintercepts (i.e., unit fixed effects). This is equivalent to“de-meaning” the dependent variable and all covariates.Although this approach eliminates the possibility of omit-ted variable bias arising from time-invariant factors, itintroduces an additional methodological complicationwhen a lagged dependent variable is included in themodel. Namely, the (implicit) de-meaning of covariatesinduces a correlation between the lagged dependent vari-able and the contemporaneous value of the error term,potentially producing biased and inconsistent parame-ter estimates (Nickell 1981). Fortunately, scholars have

in the presence of residual autocorrelation. In our specification,however, the level of residual autocorrelation is insignificant, largelyeliminating this problem (Keele and Kelly 2006).

10 LANNY W. MARTIN AND GEORG VANBERG

proposed several instrumental variable estimators to ef-fectively purge this correlation from the model. We usethe system estimator developed by Arellano and Bover(1995) and Blundell and Bond (1998), which has provento be very successful in both minimizing bias and maxi-mizing efficiency (for an excellent overview of this classof models, see Wawro 2002).

In Table 1, we present the results of our dynamic panelestimation for both dependent variables.17 The estimatesallow us to evaluate both the short-term and long-termeffects of the covariates. We begin by considering the esti-mated short-term effects.18 The coefficient estimates forthe theoretical variables must be interpreted with cautionsince multiple interactions are present in the model. Inparticular, recall that the coalition compromise positionon economic policy is expressed relative to the positionof the minister, and this relative conservatism variable isitself interacted with the policing strength index. Thus,the negative coefficient estimate for the variable, � Con-servatism of the Minister, reflects the expected decrease inunemployment (sickness) insurance generosity from oneyear to the next when the minister moves by one unitin a conservative direction over that year and when thechange in the relative conservatism of the government inthat year is equal to zero. This might happen, for example,in successive years of single-party government (assumingsuch a government is ideologically homogeneous) or insuccessive years of coalition government in which all gov-ernment parties share the same policy position. Similarly,the negative coefficient estimate for � Relative Conser-vatism of Government reflects the expected annual de-

17Post-estimation, we tested for second-order autocorrelation inthe errors, since the moment conditions in the dynamic panelestimator are only valid if serial correlation is not present. Fortu-nately, using the Arellano-Bond test, we were not able to reject thenull of no AR(2) correlation in either model (p > .30, p > .80,respectively). We also conducted a Sargan test to assess the va-lidity of the overidentifying restrictions in the models, where thenull hypothesis is that all instruments are uncorrelated with thecontemporaneous error term. We were not able to reject the null(p > .50, p > .60, respectively), leading us to conclude that the in-struments satisfy the moment conditions of the generalized methodof moments (GMM) estimator.

18Coefficient estimates from TSCS models can be presented in twoequivalent forms: the autoregressive distributed lag (ARDL) form(resulting from a regression of the level of the dependent variableon level-form covariates and their lags) or the error correctionmodel (ECM) form (resulting from a regression of the first differ-ence of the dependent variable on first-differenced covariates andtheir level-form lags). We prefer the ECM form because the in-formation needed to calculate long-run effects, which we discussin more detail below, is readily accessible from the table of coeffi-cients (De Boef and Keele 2008). Thus, in Table 1, we perform aBardsen transformation of the ARDL estimates from the dynamicpanel model to produce ECM estimates. Due to space limitations,the country-specific intercepts are not shown.

crease in unemployment (sickness) insurance generositywhen the government as a whole moves one unit furtherto the right of the minister in that year and when this oc-curs in legislatures for which the policing strength indexis equal to zero (which does not occur in our sample ofcountry-years).

Because these interactions are difficult to interpretdirectly, we provide visual illustrations of the substantiveeffects of our key variables. We begin with the short-termeffect of the relative conservatism of the government. Weexplore this effect by considering a hypothetical scenarioin which the position of the relevant minister is held fixedfrom one year to the next, but the relative conservatism ofthe government increases by a standard deviation (about0.60 points on our economic scale). Such a scenario mightoccur, for example, if the party of the minister leaves onegovernment coalition and enters a new one with moreconservative partners. We examine the impact of such ahypothetical change, for both of our social insurance pro-grams, under two different sets of legislative institutions:those present in Germany, where legislative policing in-stitutions are strong, and those present in France, wherethese institutions are weak.19

After calculating these predicted conditional effectsfor each policy area, along with their standard errors, wegenerated a distribution of 1,000 predicted conditionaleffects, for each set of policing institutions, on unem-ployment and sickness generosity.20 We display these dis-tributions in Figure 4. The curves in the figure correspondto kernel density estimates of the frequency of the corre-sponding predicted values on the x-axis. The solid part ofeach curve covers the range of predicted values from the5th to 95th percentiles of the distribution (i.e., this part ofthe curve corresponds to a 95% confidence interval); thedotted part of each curve covers values falling outside thisinterval. Finally, the median value of each distribution isindicated by a dashed vertical line.

Note first that, for both policy areas, whether (and towhat extent) a shift in the position of the government asa whole makes a difference to policy outcomes dependssignificantly on the strength of legislative policing insti-tutions. Beginning with unemployment policy, we seethat, under the weak institutions characteristic of France,an increase in relative government conservatism has nostatistically discernible impact on change in benefit gen-erosity (i.e., the confidence interval of the distribution of

19In this example, we use the average policing strength score foreach country, shown in the note beneath Figure 3.

20Specifically, each set of simulated effects was drawn from a nor-mal distribution, with the mean set to the appropriate calculatedconditional effect and the standard deviation set to the standarderror of this effect.

COALITIONS, LEGISLATURES, AND PUBLIC POLICY 11

TABLE 1 Effect of Minister and Government Policy Preferences on Change in Social ProtectionProgram Generosity, Conditional on Legislative Policing Institutions

Unemployment SicknessExplanatory Variables Insurance Insurance

� Conservatism of the Minister −0.081∗∗∗ −0.053∗∗

(0.028) (0.026)� Relative Conservatism of Government −0.097∗ −0.046

(0.056) (0.048)� Relative Conservatism of Government × Policing Strength −0.097∗∗∗ −0.097∗∗∗

(0.035) (0.031)� Policing Strength 0.205 0.137

(0.281) (0.261)� GDP per Capita Growth 0.079∗ 0.091∗∗

(0.047) (0.039)� Deficit Growth −0.021∗∗ −0.008

(0.010) (0.010)� Log Unemployment Rate [� Public Health Share]† −0.020 0.010

(0.146) (0.010)Conservatism of the Minister (t − 1) 0.009 −0.037∗∗

(0.021) (0.018)Relative Conservatism of Government (t − 1) −0.027 −0.036

(0.042) (0.038)Relative Conservatism of Government × Policing Strength (t − 1) −0.021 0.015

(0.030) (0.026)Policing Strength (t − 1) −0.011 0.027

(0.126) (0.118)GDP per Capita Growth (t − 1) 0.101∗ 0.130∗∗∗

(0.061) (0.045)Deficit Growth (t − 1) −0.029∗∗ −0.016

(0.015) (0.014)Log Unemployment Rate (t − 1) [Public Health Share (t − 1)]† −0.104∗∗ 0.005

(0.047) (0.005)Unemployment (Sickness) Generosity (t − 1) −0.097∗∗∗ −0.103∗∗∗

(0.018) (0.019)� Unemployment (Sickness) Generosity (t − 1) 0.093∗∗∗ 0.054∗

(0.031) (0.030)

Note: Bardsen-transformed coefficient estimates from Arellano-Bover Blundell-Bond linear dynamic panel model with country fixed effects(GMM standard errors in parentheses) are shown. Estimates for the country-specific intercepts are not shown. N = 498 country-years(unemployment) and 496 country-years (sickness).†Estimate is for the log unemployment rate variable in the unemployment insurance model and for the public health share variable in thesickness insurance model.∗p <. 10, ∗∗p < .05, ∗∗∗p < .01.

predicted effects clearly encompasses zero). This is sim-ilarly the case for sickness benefits policy. Conversely,under the strong institutions characteristic of Germany,this hypothetical rightward shift in the economic prefer-ences of the government is predicted to lead to a clear re-duction in both unemployment and sickness protection.

To better grasp the substantive significance of thisfinding, we unpack how such a predicted change understrong institutions impacts the most important compo-

nents of social insurance generosity. For brevity, we focushere on unemployment benefit generosity, specifically onthe median predicted change in this index under stronginstitutions—a decrease of approximately 0.10 points.We calculate changes in the components of this indexusing the component values for Germany in 2005, the lastyear for which we have CWED2 data on the componentsfor all countries in our sample. Such a decrease in theunemployment generosity index translates to one of the

12 LANNY W. MARTIN AND GEORG VANBERG

FIGURE 4 Effect of Increase in Relative Conservatism of Government onUnemployment and Sickness Insurance Generosity, Conditional onLegislative Policing Institutions

Note: Kernel density estimates of the effect of a standard deviation increase in Relative CoalitionConservatism on benefit generosity (N = 1,000 simulated predictions per legislative policing regime)are displayed. The solid part of each curve encompasses the range of simulated predicted effects fromthe 5th to 95th percentiles of the distribution.

following changes in the index components (holding allother component values constant):

� a reduction of about 2.8 percentage points in thebenefit replacement rate, or

� an increase of over 6 months in time spent onthe job before an individual can qualify for un-employment benefits, or

� a reduction of 0.78 percentage points in the shareof the labor force covered by unemployment in-surance.

In short, we see that, in legislative environments wherecoalition partners can effectively scrutinize (and, if neces-sary, correct) the actions of a cabinet minister, changes inpublic policy will reflect shifts in their preferences even ifthe preferences of the minister remain virtually the same.

We now turn to investigating the short-term impactof the conservatism of the minister. We explore this effectby considering a new hypothetical scenario. In this case,we assume that there exists a two-party coalition in which,during the first year in office, the relevant ministry is

controlled by the leftist party in the coalition, but that, atthe beginning of the second year, a reshuffle occurs thattransfers control of the ministry to the rightist party.21 Wefurther assume that the parties are of equal size, so that thecoalition compromise is located exactly midway betweenthem and is unaffected by the reshuffle. We have arguedthat ministers are more likely to “get their way” whenlegislative institutions are weak. We evaluate this claim bycomparing the effect of the hypothetical reshuffle underthe same set of institutions considered in Figure 4: thestrong legislative institutions of Germany and the weaklegislative institutions of France.

As shown in Figure 5, the differences across the twosets of legislative institutions are as stark as they were inthe previous figure. In particular, under the weak policinginstitutions in France, unemployment and sickness ben-efits are predicted to become markedly less generous afterthe more conservative coalition partner takes over the

21In this example, we set the policy position of the leftist party at the25th percentile value of the Conservatism of the Minister variableand the position of the rightist party at the 75th percentile value.

COALITIONS, LEGISLATURES, AND PUBLIC POLICY 13

FIGURE 5 Effect of Increase in Conservatism of the Minister on Unemploymentand Sickness Insurance Generosity, Conditional on LegislativePolicing Institutions

Note: Kernel density estimates of the effect of an increase in Conservatism of Minister, fromits 25th to 75th percentile values, on benefit generosity (N = 1,000 simulated predictions perlegislative policing regime). The solid part of each curve encompasses the range of simulatedpredicted effects from the 5th to 95th percentiles of the distribution.

ministry. Under German institutions, however, the hypo-thetical reshuffle is expected to have no discernible impacton generosity for either social insurance program. Simi-lar to what we did earlier, we can examine the substantiveimpact of the shift in the preferences of the minister byevaluating the median predicted change in terms of theindividual components of the generosity index. In thiscase, we focus on the median prediction under Frenchinstitutions for unemployment benefit generosity, whichis a reduction of approximately 0.12 points. Based on thecomponent values for France in 2005, this decrease trans-lates to one of the following changes in the index compo-nents (holding all other component values constant):

� a reduction of more than 2.8 percentage pointsin the benefit replacement rate, or

� an increase of 4 months in time spent on the jobbefore an individual can qualify for unemploy-ment benefits, or

� a reduction of over 0.90 percentage points in theshare of the labor force covered by unemploy-ment insurance.

In short, these findings clearly indicate that when legisla-tive institutions are weak and strengthen ministers at the

expense of their coalition partners, policy outcomes areresponsive to the preferences of parties that control therelevant ministry.

We can also assess how quickly changes in thecovariates affect unemployment and sickness benefits. Toassess long-term effects of our theoretical variables, wecalculate the long-run multiplier (LRM) for the laggedConservatism of the Minister and the lagged RelativeConservatism of Government variables (under all values oflegislative policing strength). We find that, for both mod-els, the LRM for each variable is statistically insignificantat the 5% level.22 This implies that both unemploymentpolicy and sickness policy are responsive to changes inthe preferences of government actors in the short term,rather than spread out over several years. Moreover, thesmall negative coefficient on the error correction rate,

22However, in the sickness insurance model, the LRM for the laggedConservatism of the Minister variable is significant at p < .10. TheLRM for a covariate (expressed in ECM form) is computed by di-viding the negative of its lagged coefficient by the coefficient on thelagged dependent variable. Because the LRM is a ratio, its variance

is approximated as ( 1b2 )Var(a) + ( a2

b4 )Var(b) − 2( ab3 )Cov(a, b),

where a represents the coefficient on the lagged covariate and brepresents the coefficient on the lagged dependent variable.

14 LANNY W. MARTIN AND GEORG VANBERG

Unemployment (Sickness) Insurance Generosity (t − 1),implies that once policy is “shocked” by a change ingovernment or ministerial party preferences, the rate ofreturn to its long-run equilibrium is very slow.

These findings provide clear support for our centralargument, but there is a final issue we need to consider.Specifically, it is possible that parties can use other institu-tions, particularly at the cabinet level, to rein in ministerialdrift, and that the presence of these institutions may becorrelated with the presence of legislative institutions. Ifthat is the case, then we may be attributing an impact tolegislative mechanisms that is due instead to mechanismsinside the executive. Exploring this possibility empiri-cally is tricky: Many executive institutions highlightedby scholars as “conflict management” devices—such ascabinet committees and inner cabinets—are, at least inprinciple, available to all governments because they areinformal and can be created, reshuffled, or disbanded atany time. Obviously, this provides no empirical leverageto analyze their effects. Moreover, scholars who study cab-inet conflict management devices (see, e.g., Andeweg andTimmermans 2008) have recorded the presence of thesemechanisms only post hoc—that is, these features haveonly been coded as “existing” if governments relied onthem (publicly) to resolve conflicts. Even then, it is of-ten not clear which cabinet ministers were members ofwhich groups.

Fortunately, this is not true for the cabinet-levelinstitution that has received the most attention fromresearchers—junior ministers (Lipsmeyer and Pierce2011; Thies 2001). Unlike other executive institutions,the existence, and identity, of a junior minister (JM) isgenerally known at the beginning of a government’s life(when cabinet-level positions are assigned). As we notedearlier, the appointment of a “shadow” JM by a coalitionpartner may allow it to “keep tabs” on a minister. Toassess whether coalition partners are able to use theJM post in this way and, if so, whether this alters ourfindings regarding the importance of legislative policinginstitutions, we add two new covariates to the analysis:a dichotomous variable indicating whether a shadowJM is present in the relevant ministry and a positionalvariable measuring the conservatism of the shadow JMon our scale. If this cabinet-level institution matters,we should see a decrease in unemployment (sickness)generosity (holding constant the position of the minister)as the party possessing the shadow JM post is moreeconomically conservative. However, our findings (see SIAppendix D) reveal no such effect: The policy positionof the shadow JM has no discernible impact on changein unemployment or sickness benefits. This finding isconsistent with our earlier theoretical discussion, which

suggests that policing institutions at the executive levelare less likely to be effective than those present inside thelegislature.

Conclusion

Motivated by a concern for the quality of representationin parliamentary democracies, scholars have increasinglydevoted attention to understanding the policymakingprocess under coalition governments. What positions dothese governments take? Which parties’ priorities andpreferences are reflected in policy choices? A prominenttheme in this literature concerns the tensions resultingfrom the delegation problems multiparty governmentsconfront. Parties that govern jointly must pursue com-mon policies, but they often have different preferencesover them. In addition, cabinets must delegate discretionin policy development and implementation to cabinetministers, who have access to departmental resourcesthat provide requisite policy expertise. Consequently,multiparty governments must deal with the fact thatthough the coalition as a whole may want to pursue apolicy compromise, the need to delegate raises the riskthat such compromises are undermined by “ministerialdrift.” Coalition governance thus constitutes a problemof credible commitment.

Over the past decade, scholars have advancedthe argument that whether coalitions can resolve thiscommitment problem depends on institutional context(Martin and Vanberg 2005, 2011; Thies 2001). In theabsence of institutions that allow coalition partners toscrutinize ministerial draft bills and to develop alternativepolicies, it is difficult to control ministerial discretion. Asa result, policy will reflect the preferences of the party withcontrol of the relevant ministry. But where policymakinginstitutions provide parties with the ability to “keep tabs”on their partners, compromise agreements are viable. Insuch contexts, the argument goes, coalition policy willreflect the influence of all government parties. A numberof institutions could conceivably be employed for thispurpose. But as we have argued, legislative institutionsare particularly important for practical and politicalreasons.

This line of scholarship has reshaped our understand-ing of coalition governance, highlighting the importanceof the institutional environment. Yet, there has also been asignificant limitation. Empirical work has lagged behindtheoretical developments, and scholars have thus far pro-vided only circumstantial evidence for the key claim thatcoalition policymaking depends on institutional context.The key contribution of this study has been to overcome

COALITIONS, LEGISLATURES, AND PUBLIC POLICY 15

this limitation. We provide direct evidence that coalitionpolicies are shaped significantly by the nature of legislativepolicing institutions.

Our findings have significant implications. Most ob-viously, they imply that whose preferences are reflectedin policy under coalition government depends criticallyon broader institutional features. In systems that enableministers to dominate the legislative process, policy tendsto reflect the positions of ministerial parties. In contrast,when legislative institutions enable effective scrutiny andamendment of ministers’ proposals, policy tends to re-flect a compromise among coalition partners. Put differ-ently, the same government—in terms of its party compo-sition and portfolio allocation—will make different policychoices in different institutional settings, and these policychoices will be favored by different sets of citizens.

These findings bear on other aspects of coalition gov-ernance and parliamentary politics. Consider the impli-cations for the importance of portfolio allocation. Howministries are distributed among parties is of intrinsic in-terest to politicians and receives considerable media andscholarly attention. Yet, our results highlight that whetherportfolio allocations are significant for policy outputs—and thus for the quality of representation that citizenspresumably care about—varies across institutional set-tings. In some circumstances, control of a ministry byone party rather than another is likely to have impor-tant implications for policy. In other settings, it may notmake a substantial difference: Policy will be jointly deter-mined by coalition partners, regardless of who controlsthe relevant ministry.

The fact that policy outcomes are conditioned byinstitutional context also has ramifications for the coali-tions that are likely to emerge from government formationattempts. Expectations about the policies a governmentwill pursue determine, in part, how attractive alternativecoalitions are for party leaders. Our contribution sug-gests that because institutional context shapes the abilityof coalitions to reliably enforce compromise agreements,and thus policy outcomes, it should also play a role incoalition formation. For example, in systems with stronglegislative institutions, party elites should pay close atten-tion to negotiating a mutually acceptable agreement sincethis compromise is likely to be realized in government pol-icy. Coalitions that include “preference outliers” are notnecessarily problematic because party leaders know thatministers can be effectively controlled. From a policyperspective, the distribution of portfolios across partiesbecomes secondary. In contrast, in systems with weaklegislative institutions, party leaders must approach pol-icy negotiations with a more skeptical attitude, given

the greater influence of cabinet ministers. Compromiseagreements are more difficult to sustain, and coalitionscontaining preference outliers are more likely to posedifficulties. How portfolios are allocated across partiestherefore becomes a primary issue. Thus, we expect thatin systems with strong legislatures, parties will be morewilling to allocate important ministries to relative prefer-ence outliers, whereas parties in weak systems will preferto award these ministries primarily to partners with cen-trally located preferences.

Finally, our results speak not only to the behaviorof party elites or coalition policymaking. They also haveimplications for citizens. In deciding how to cast theirvotes, citizens must attribute responsibility to individualcoalition parties. Our evidence suggests that in doing so,voters should be sensitive to the institutional details oftheir polity. In systems with weak legislative institutions,direct responsibility for policy largely rests with the partycontrolling relevant portfolios; in strong systems, minis-terial parties are effectively constrained by their coalitionpartners. In such systems, voters have less reason to con-cern themselves with the precise details of portfolio allo-cation, whereas such minutiae are far more relevant forattributing blame or forecasting policy in weak systems.

Of course, while these results represent a clear stepforward, considerable scope for future work remains.Given the available data, our analysis has focused on twopolicy areas—unemployment insurance and short-termdisability and sickness insurance—that provide anappropriate testing ground for our argument. Thesepolicies are salient, touch on a number of contemporarypolitical concerns, and are amenable to reforms by singlegovernments (rather than by successive governmentsover several years). Moreover, there are strong theoreticalreasons to believe that our findings translate to other pol-icy domains. The results demonstrate that in at least twocentral policy areas, parties in coalition take advantage ofstrong legislative institutions to constrain the influenceof cabinet ministers. But the institutions that allowthem to do so are general, and not specific to the policydomain we have examined. Given the strong incentivesof parties to enforce coalition compromise agreements,it would therefore be surprising if parties chose not totake advantage of the same opportunities in other policyareas. Nevertheless, it is obviously desirable to expandthis approach to other areas in future work. Similarly, wehave examined only one set of institutions, but coalitiongovernments may rely on other mechanisms to rein inministerial drift. Expanding the institutional scope ofthe analysis thus represents another clear avenue forprogress.

16 LANNY W. MARTIN AND GEORG VANBERG

References

Andeweg, Rudy B., and Arco Timmermans. 2008. “ConflictManagement in Coalition Government.” In Cabinets andCoalition Bargaining: The Democratic Life Cycle in WesternEurope, ed. Kaare Strøm, Wolfgang C. Muller, and TorbjornBergman. Oxford: Oxford University Press, 269–300.

Arellano, Manuel, and Olympia Bover. 1995. “Another Look atthe Instrumental Variable Estimation of Error-ComponentsModels.” Journal of Econometrics 68(1): 29–51.

Becher, Michael. 2010. “Constraining Ministerial Power: TheImpact of Veto Players on Labor Market Reforms in Indus-trial Democracies, 1973–2000.” Comparative Political Studies43(1): 33–60.

Blundell, Richard, and Stephen Bond. 1998. “Initial Conditionsand Moment Restrictions in Dynamic Panel Data Models.”Journal of Econometrics 87(1): 115–43.

Carroll, Royce, and Gary W. Cox. 2012. “Shadowing Ministers:Monitoring Partners in Coalition Governments.” Compara-tive Political Studies 45(2): 220–36.

De Boef, Suzanna, and Luke Keele. 2008. “Taking Time Seri-ously.” American Journal of Political Science 52(1): 184–200.

Goodhart, Lucy. 2013. “Who Decides? Coalition Governanceand Ministerial Discretion.” Quarterly Journal of PoliticalScience 8(3): 205–37.

Harfst, Philipp, and Kai-Uwe Schnapp. 2003. “Instru-mente Parlamentarischer Kontrolle der Exekutive in West-lichen Demokratien.” https://www.ssoar.info/ssoar/handle/document/11154.

Hendry, David F. 1995. Dynamic Econometrics. Oxford: OxfordUniversity Press.

Kalisch, David W., and Tetsuya Aman. 1998. “RetirementIncome Systems: The Reform Process across OECDCountries.” Ageing Working Papers 3.4, Social Pol-icy Division, OECD. http://www.oecd.org/pensions/public-pensions/2429570.pdf.

Keele, Luke, and Nathan J. Kelly. 2006. “Dynamic Models forDynamic Theories: The Ins and Outs of Lagged DependentVariables.” Political Analysis 14(2): 186–205.

Korpi, Walter, and Joakim Palme. 2003. “New Politics and ClassPolitics in the Context of Austerity and Globalization: Wel-fare State Regress in 18 Countries, 1975–1995.” AmericanPolitical Science Review 97(3): 425–46.

Laver, Michael, and Kenneth A. Shepsle. 1996. Making andBreaking Governments: Cabinets and Legislatures in Par-liamentary Democracies. Cambridge: Cambridge UniversityPress.

Lipsmeyer, Christine, and Heather Nicole Pierce. 2011. “TheEyes That Bind: Junior Ministers as Oversight Mechanismsin Coalition Governments.” Journal of Politics 73(4): 1152–64.

Lowe, Will, Kenneth Benoit, Slava Mikhaylov, and MichaelLaver. 2011. “Scaling Policy Preferences from Coded Po-litical Texts.” Legislative Studies Quarterly 36(1): 123–55.

Martin, Lanny W., and Georg Vanberg. 2005. “Coalition Policy-making and Legislative Review.” American Political ScienceReview 99(1): 93–106.

Martin, Lanny W., and Georg Vanberg. 2011. Parliaments andCoalitions: The Role of Legislative Institutions in MultipartyGovernance. Oxford: Oxford University Press.

Martin, Lanny W., and Georg Vanberg. 2014. “Parties andPolicymaking in Multiparty Governments: The Legisla-tive Median, Ministerial Autonomy, and the CoalitionCompromise.” American Journal of Political Science 58(4):979–96.

Mattson, Ingvar, and Kaare Strøm. 1995. “Parliamentary Com-mittees.” In Parliaments and Majority Rule in Western Europe,ed. Herbert Doring. New York: St. Martin’s Press, 249–307.

Muller, Wolfgang C., and Kaare Strøm, eds,. 2000. CoalitionGovernments in Western Europe. Oxford: Oxford UniversityPress.

Nickell, Stephen. 1981. “Biases in Dynamic Models with FixedEffects.” Econometrica 49(6): 1417–26.

North, Douglass C. 1993. “Institutions and Credible Commit-ment.” Journal of Institutional and Theoretical Economics149(1): 11–23.

Pedrazzani, Andrea, and Francesco Zucchini. 2013. “Horsesand Hippos: Why Italian Government Bills Change in theLegislative Arena, 1987–2006.” European Journal of PoliticalResearch 52(5): 687–714.

Powell, G. Bingham, Jr. 2000. Elections as Instruments of Democ-racy: Majoritarian and Proportional Visions. New Haven, CT:Yale University Press.

Rezayatmand, Reza, Milena Pavlova, and Wim Groot. 2013.“The Impact of Out-of-Pocket Payments on Preven-tion and Health-Related Lifestyle: A Systematic Litera-ture Review.” European Journal of Public Health 23(1):74–79.

Scruggs, Lyle, Detlef Jahn, and Kati Kuitto. 2014. “Compar-ative Welfare Entitlements Data Set 2, Version 2014-03.”http://cwed2.org/.

Thies, Michael F. 2001. “Keeping Tabs on Partners: The Logicof Delegation in Coalition Governments.” American Journalof Political Science 45(3): 580–98.

Wawro, Gregory. 2002. “Estimating Dynamic Panel Data Mod-els in Political Science.” Political Analysis 10(1): 25–47.

Supporting Information

Additional supporting information may be found onlinein the Supporting Information section at the end of thearticle.

Appendix A: Ministries Primarily Responsible for Unem-ployment and Sickness Insurance ProgramsAppendix B: Construction and Validity of the Labor In-surance Benefits ScaleAppendix C: Factor Analysis of Components of the Leg-islative Policing Strength IndexAppendix D: Effects of Cabinet-Level Institutions