By the end of October 2009, 26 of 27 EU countries had ratified the Treaty of Lisbon. Supporters and opponents of the treaty closely observed the ratification process in the Czech Republic and in particular the behaviour of two actors. On 3 November the Constitutional Court ruled that the treaty accorded with national constitutional demands, leading the EU-sceptic head of state, Vaclav Klaus, to give up his resistance and finally ratify the treaty. Thus, a fundamental political decision affecting almost 500 million EU citizens ultimately depended on two actors outside of parliament and cabinet, who were not directly elected even by the Czech people. We can easily find other examples in which such actors make far-reaching decisions or influence political competition. Central banks autonomously set interest rates that critically affect financial markets and economic policy. Constitutional courts can declare government proposals unconstitutional and frustrate a government's reform agenda, as for example in the case of the socialist cabinet in France during the early 1980s.Footnote 2 Audit institutions uncovering financial misconduct by the government can affect electoral competition, as in the case of the Canadian sponsorship scandal in 2004 that ultimately led to the defeat of the Liberal Party in the 2006 elections.Footnote 3
These few examples testify to the importance of officeholders in what I call ‘constraint institutions’ (CIs) outside cabinet and parliament. Of course, this is not a new discovery. A broad literature in comparative politics and political economy has discussed CIs using concepts such as checks and balances, consensus democracy, veto players and delegation to non-majoritarian institutions, to name just a few. Leaving aside many relevant differences for now, these concepts share a common core claim: political actors who are not tied into the democratic chain of delegationFootnote 4 can affect policy outputs and constrain the parliamentary majority and the cabinet it supports.
To what extent and under what conditions is this constraining effect likely to occur? Most studies assume that the level of constraint depends on the institutional powers of CIs. By contrast, the preferences of external officeholders and thus their incentives to use these powers are commonly ignored in empirical research, even though theoretical work based on the spatial model – most prominently Tsebelis's ‘Veto Player Theory’ – emphasizes the critical importance of actor preferences.Footnote 5
This article conceptualizes constraint by external officeholders as a two-dimensional phenomenon that includes both the ability to constrain (the powers dimension) and an officeholder's willingness to do so based on his or her preferences (the incentives dimension). The article develops a strategy for measuring the incentives dimension in cross-national studies of different CIs based on the selection method of external officeholders. I provide empirical measures for both dimensions for officeholders in constitutional courts, central banks, audit institutions and ombudsmen across 25 European democracies. Thus, the study goes beyond the frequent limitation of examining only CIs with formal decision or veto power and also discusses offices whose constraining effects derive from their provision of information to political actors and the public. My analysis demonstrates that powers and incentives are largely independent in empirical terms. Assessments of the constraint to be expected from these actors will thus be biased if we focus solely on the powers dimension. The two-dimensional framework proposed here may explain some puzzling findings in the literature; it leads to new research questions; and generates measures that can be used as dependent and independent variables in future research.
THE EFFECTS OF EXTERNAL CONSTRAINT INSTITUTIONS
I use ‘constraint institution’ as a generic term for different political offices outside the parliamentary chain of delegation that can directly or indirectly constrain the cabinet as the key policymaker in parliamentary democracies. The most prominent type of CIs are veto players such as presidents with veto powers or constitutional courts.Footnote 6 Two other types of CIs are also relevant.Footnote 7 First, ‘dictators’, most prominently central banks, can autonomously change the status quo in their area of competence. Second, ‘powerful players’ without veto powers, such as auditors and ombudsmen, can influence the payoffs and thus indirectly the behaviour of other actors, for example by making influential public statements or providing new information.
CIs are dealt with in at least two different strands of literature. First, scholars have studied the effects of specific CIs on political processes and policy outputs. The ‘judicialization of politics’ literature analyses how judges, especially in constitutional courts, constrain the freedom of action of elected politicians, both by affecting policy outputs and by changing the interaction of political actors and the sort of arguments used in political discourse.Footnote 8 Political economy scholars have identified a massive increase in the power and independence of central banks over the last 20 years with systematic (albeit sometimes conditional) effects on inflation and other policy outputs.Footnote 9 Finally, independent regulatory agencies have been created in various sectors such as telecommunication, electricity, food safety and competition.Footnote 10
Second, authors have used CIs alongside other variables to characterize the overall nature of a political system and to study the effects of types of democracy on policy outputs. Lijphart's path-breaking two-dimensional conceptualization of majoritarian and consensus democracy includes variables on constitutional courts and central banks. Regarding outputs, Lijphart finds almost no significant effects for his second dimension, which includes constitutional courts and central banks. The exception is lower inflation in consensus democracies – which can be tied to strong central banks.Footnote 11 Tsebelis mentions constitutional courts as institutional veto players but does not analyse their effects on law-making because they are allegedly absorbed as a result of their ideological position in the core of other veto players.Footnote 12 The principal–agent (PA)-based model of representative democracy as a chain of delegation introduced by Kaare Strøm and colleagues discusses various CIs, also those without direct veto power, as elements of a ‘constrained parliamentarism’. Footnote 13 Finally, information on different CIs is included in various veto point indices.Footnote 14
Which factors determine whether and to what extent CIs constrain the political majority? While some scholars assume that the mere existence of a CI has a constraining effect,Footnote 15 most studies rely on measures of CIs’ institutional powers.Footnote 16 These analyses capture an important dimension of constraint; however, from an actor-centric perspective they remain incomplete. Institutionally powerful external officeholders will not automatically constrain the cabinet. Instead, this will only be the case if they have both the institutional powers and the incentives to use them. As Tsebelis points out with regard to veto players, an actor sharing the preferences of another or located in the unanimity core of other veto players does not provide any additional constraint – it is absorbed.Footnote 17 This logic applies to CIs in general.
Thus the key theoretical argument of this article states that constraint by external officeholders consists of two independent and equally important dimensions: a powers dimension and an incentives dimension. The incentives dimension, however, has been largely ignored in the empirical literature.Footnote 18 This neglect is problematic, both theoretically and empirically. Theoretically it ignores the fundamental insight of the rational-choice institutionalist framework that behaviour depends on institutional rules and preferences. Empirically, estimates of the constraining effects of external officeholders based on institutional powers alone are biased if – as the empirical analysis will show to be the case – the two dimensions are not strongly positively correlated. Under these conditions, we overestimate the constraining effect of institutionally strong officeholders, some of whom have no incentive to use their powers.
The main reason for the empirical neglect of the incentives dimension is the difficulty of measuring it, particularly in broad comparative studies. In the next section I suggest a proximate measurement approach based on the selection methods of external officeholders.
THE INCENTIVES DIMENSION: THE SELECTION OF EXTERNAL OFFICEHOLDERS
Assessing the Preferences of External Officeholders and their Incentives to Constrain the Cabinet
External officeholders only have incentives to use their institutional powers to constrain the cabinet if their preferences diverge from those of the latter. Assessing these incentives thus presupposes measuring the preferences of external officeholders. I discuss three ways in which this could be done and explain why I rely on the third strategy, based on institutional variables.
First, we could infer preferences from officeholders’ statements and actions and from assessments of their ideological positions by outside observers. This strategy has been applied successfully to individual CIs in the USA.Footnote 19 However, this approach is not feasible for broad comparative studies dealing with a large number of offices and countries, where even suitable raw data are lacking. Second, we could rely on an officeholder's party membership, assume homogeneous preferences within parties, and use a preference measure for his or her party. This approach has been applied to presidentsFootnote 20 and constitutional judges.Footnote 21 However, it fails when officeholders in CIs are nonpartisan, either by choice or because of legal requirements, and presupposes systematic data on partisan ties which are frequently missing.
The third approach employs institutional variables as proxy measure for the likelihood of divergent preferences. It is most promising for broad comparative studies because it can circumvent these problems. I suggest relying on the method of selecting officeholders because the selecting actor (in PA terminology the ‘principal’) can systematically affect the preferences of the external officeholder he chooses (the ‘agent’).Footnote 22 According to the ‘ally principle’, principals will generally choose agents with preferences equal or at least very similar to their own.Footnote 23 Thus, the selection method allows conclusions about the preferences of chosen officeholders.
As officeholders in CIs differ in important respects from classic agents, I have to address three potential problems with my measurement approach. First, many officeholders in CIs cannot be recalled, and some are explicitly not accountable to their selectors once in office. However, my argument is based on the principal's ability to select agents with similar preferences ex ante which gives the agent incentives to act in line with the principal's wishes even without ex post control mechanisms. There are good reasons to expect this sort of ex ante influence. First, many candidates for positions in CIs have previously held other state or party offices and have passed intensive scrutiny in hearings prior to their selection, allowing well-informed selectors to pick officeholders with ‘suitable’ preferences (resulting in a small danger of adverse selection). Second, empirical studies indicate that the preferences of selecting actors significantly predict constitutional judges’ voting records, despite the absence of ex post controls (indicating that moral hazard is not a pervasive problem).Footnote 24 As constitutional courts are often considered the most independent CIs, we can plausibly expect similar or even stronger effects for external officeholders with closer ties to their selectors.
Second, does the ally principle, which is well-established for bureaucratic agents, translate to the appointment of officeholders in CIs? For ‘dictators’ the principle is directly applicable because authority is delegated to one distinct (collective) actor – for example, a central bank – which mirrors the scenario of selecting a bureaucratic agent. For additional veto players (e.g. constitutional judges) the answer is more complex because there are situations in which the principal (e.g. the parliamentary majority) could be better off if she selects an agent with preferences different from her own in order to balance deviations brought about by other agents such as the cabinet. However, this balancing strategy can easily backfire because even slight changes in the ideological distances between actors and the status quo can give the additional veto player incentives to block reforms that the principal and the other agents would prefer to the status quo. Thus, selecting external officeholders with divergent preferences is risky. By contrast, applying the ally principle is a safe strategy that can make use of agents’ expertise and at the same time avoid additional agency loss.Footnote 25 Finally, selectors should also apply this principle when choosing ‘powerful players’ charged with reducing informational asymmetries. According to a prominent model of political learning, information provided by an agent will only convince the principal if the agent is considered persuasive, possesses the necessary knowledge and has incentives to reveal it truthfully.Footnote 26 Successful learning is more likely with an agent who holds similar preferences because such an agent profits directly if the principal follows his advice and thus has incentives to reveal his knowledge truthfully instead of strategically distorting or withholding information.
Third, the selection process approximates the preferences of an external officeholder relative to those of the cabinet at the time of his or her selection, but incentives to constrain the cabinet can change with modifications in government composition. The terms of many officeholders in CIs exceed a regular inter-election period. Thus, intervening elections increase the chances of divergent preferences because they can – but do not have to – cause alternation in cabinet. In the absence of specific information on the election outcome, I interpret the ratio of the length of an external officeholder's term and the inter-election period as an institutional estimate of the uncertainty associated with my measurement of the incentives dimension.
Selection Methods for External Officeholders and the Incentives to Constrain Index (ICI)
We can distinguish three pertinent methods for selecting external officeholders: appointment by the cabinet, election by parliament or mixed assemblies including members of parliament, and appointment by the head of state.Footnote 27 Unfortunately, the third method does not allow systematic conclusions because it produces very different outcomes depending on how the preferences of the head of state and the cabinet are related. Furthermore, it is often unclear to what degree appointments by the head of state are de facto influenced by the cabinet. Thus, this selection method is not used for measuring the incentives dimension.
The appointment of CIs by cabinet actors resembles delegation from the cabinet to the bureaucracy. The situations differ in the amount of ex post control available to the cabinet; however, this does not affect the ex ante influence via the selection process. Thus we can expect cabinet actors to choose officeholders with preferences similar to their own who consequently have few incentives to constrain the cabinet after assuming office.
Elections by parliament are more complex because institutional variation in the electoral process systematically influences the likelihood that external officeholders have deviating preferences.Footnote 28 I discuss four variables affecting this likelihood: (1) the admissible number of candidates; (2) the nominating actors; (3) the majority requirement; and (4) the voting method. I combine these variables in an index that will serve as a measure of the incentives dimension. As always, index construction requires potentially controversial assumptions because theoretical arguments only predict the direction of an effect but not its metric size nor the way in which the effects of different variables interact. I justify all assumptions theoretically and show that alternative decisions lead to very similar index values (see notes 33–9). The substantive findings presented below are robust towards alternative decisions as well. Detailed results of these robustness tests and the values of the individual variables are available from the author.
In a first step, I assess the freedom of parliament to implement its collective preference. Based on the admissible number of candidates, I distinguish between the ‘approval model’, which has only one candidate, and the ‘competitive election model’ with (potentially) competing candidates. In the approval model, the agenda setter power of the nominating actor can lead to outcomes far removed from the parliament's collective ideal point as long as a majority prefers the candidate to the status quo.Footnote 29 In the competitive election model, on the other hand, the median voter logic is applicable. Accordingly, we expect outcomes towards the centre of the relevant policy dimension and thus close to the collective preference of parliament, understood as the position of the median deputy.Footnote 30 However, the nominator can still pull the outcome towards his or her ideal point if s/he has a monopoly on nominations but is required to propose more than one candidate, if the right to nominate is limited to a small number of actors (e.g. to parties of a certain size) or if parties suffer reputational costs if they nominate candidates far from their ideal points.Footnote 31
Given the power of the agenda-setter, the influence of parliamentary preferences on the election outcome depends on who may nominate candidates. I distinguish between nominations by parliamentary actors (including political parties as organizations), cabinet actors and third parties, most commonly the head of state. Parliamentary actors are least constrained if they can nominate candidates themselves. On the other hand, parliamentary freedom to choose can be severely limited by cabinet nominations because the preferences of the majority in parliament and the cabinet can diverge, especially with regard to CIs that are charged with controlling the latter. In the case of conflict over such an election, the cabinet is in a good position to prevail because it can link the election to policy decisions or even a vote of confidence, forcing parliamentary actors to accept a candidate they would reject in a stand-alone vote.Footnote 32 The constraint on parliament emerging from nominations by third parties is harder to assess because we lack systematic knowledge about the preferences of these third parties. However, they certainly lack the cabinet's institutional means for making parliamentary actors bow to their wishes. Thus, I assume that third-party nominations constrain parliament more than own nominations but less than cabinet nominations.
Combining these two variables leads to six models of parliamentary elections. We need some additional assumptions to assign numerical index scores to these models. First, I treat the nominator as the more important variable because the agenda-setter advantage is present even in the competitive elections model for monopoly nominations by the cabinet and third parties. Thus, the two models with parliamentary nomination have the highest scores, followed by the models with third-party nomination and the ones with cabinet nomination. These three pairs are ordered internally by the admissible number of candidates.Footnote 33 Second, I assume equal distances between the six models, which is a customary fall-back position in the absence of strong theoretical arguments for unequal distances that would require largely arbitrary ad hoc assumptions about specific distances. Third, autonomous cabinet appointments are given a score of 0 because parliament is not involved and such officeholders are likely to share the preferences of the cabinet. Assigning index values from 0 to 1 based on these assumptions leads to the scores in Table 1.
Table 1 The Freedom of Parliament to Implement its Collective Preference in Elections

Notes: AM = approval model; CEM = competitive election model.
Index scores in parentheses. Index scores = 0 if officeholder is appointed by the cabinet without parliamentary involvement.
Measuring the freedom of parliament to implement its collective preference, however, is not sufficient because parliamentary actors differ widely in their preferences, especially in parliamentary democracies characterized by the dualism between cabinet and opposition parties. To assess the likelihood that parliaments elect officeholders with incentives to constrain the cabinet, we have to analyse whether election rules give influence on the outcome to parliamentary actors who do not share the cabinet's preferences. Two such actors come to mind: opposition parties and minorities within cabinet parties.Footnote 34 Opposition influence depends on the majority requirement. Opposition parties are least influential under simple majority rule that often allows even a minority cabinet to elect its candidate without opposition support. Under absolute majority (i.e. 50 per cent plus one of all members of parliament), minority cabinets must seek votes from non-cabinet parties, whereas majority cabinets can muster the required votes by themselves. Under qualified majority, even majority cabinets usually need opposition support. The influence of intra-party minorities depends on the voting method. Secret voting limits the party leadership's monitoring ability and enhances the influence of intra-party minorities.Footnote 35 However, the impact of the voting method should be smaller compared to the other variables because intra-party minorities usually see their party as a necessary, and in the long run promising, vehicle for pursuing their goals.Footnote 36
I combine these two variables with the six models of elections discussed above to arrive at a comprehensive index. First, I multiply the scores in Table 1 with the factors 1, 1.5 and 2 for elections with simple, absolute and qualified majority requirements.Footnote 37 I use a multiplicative link because the effects of the electoral models and the majority requirement theoretically reinforce each other: that is, the constraint on the cabinet increases disproportionately if parliaments are free to implement their preferences and opposition parties influence the outcome.Footnote 38 In a second step, a value of 0.33 is added for secret elections. The impact of this element on the overall index is relatively small, reflecting the theoretically lower importance of the voting method.Footnote 39
The resulting Incentives to Constrain Index (ICI) ranges from 0 to 2.33 and captures the institutionally mediated likelihood that elections in parliament lead to the selection of external officeholders with preferences deviating from those of the cabinet. The next section uses this index to provide the first systematic measure of the incentives dimension across four CIs in 25 European democracies.
THE POWERS AND INCENTIVES DIMENSIONS FOR FOUR EXTERNAL CONSTRAINT INSTITUTIONS IN 25 EUROPEAN DEMOCRACIES
Case Selection, Data Sources and Coding Decisions
My theoretical argument pertains to stable parliamentary democracies. The empirical analysis covers the 24 parliamentary EU countries at the time of data collection in January 2006 (all current EU member states except Bulgaria, Romania and the presidential system of Cyprus) plus Norway. This selection of countries ensures compatibility with existing comparative studies and expands their geographic reach by including the new EU member states in Eastern Europe.
I analyse four different CIs: constitutional judges, heads of central banks, heads of audit institutions and ombudsmen.Footnote 40 The four CIs, all of which are mentioned in Strøm's account of constraint institutions,Footnote 41 have roughly equivalent functions across countries. At the same time, they vary in their function and political impact. Constitutional judges can be understood as veto players whereas central banks are dictators within their jurisdictions.Footnote 42
Heads of audit institutions and ombudsmen are ‘powerful players’ who can help parliamentary actors overcome or at least decrease informational asymmetries and can thus indirectly influence political competition. A few more examples help underscore the relevance of these CIs. The electoral influence of the Canadian auditor general in 2006 has already been discussed. Its German counterpart dealt a heavy blow to the cabinet when its highly publicized annual report for 2009 denied any leeway for tax cuts – a core electoral promise of the current cabinet.Footnote 43 Ombudsmen are discussed even less in the literature even though they exist in most European countries.Footnote 44 They are usually charged with investigating complaints of administrative misconduct and offer recommendations to parliaments and/or governments.Footnote 45 Even though these recommendations are not legally binding, a recent study of the Dutch case shows that almost all of them are followed, leading to substantial policy changes in areas such as asylum procedures, the taxation of student grants and the supply of public information on criminal cases.Footnote 46 In Austria, a weekly television programme in which the ombudsmen presented exemplary cases was watched by up to 1 million people and created a great deal of publicity and public pressure on the politicians responsible for those areas.Footnote 47 At the EU level, the European Ombudsman created in 1992 has established himself as an important actor at the junction between parliamentary and judicial control of the Commission.Footnote 48 Thus, audit institutions and ombudsmen can affect political processes and policy outputs even in the absence of formal veto or decision power and deserve systematic attention as CIs.
The institutional data used in the analysis were collected from the relevant legal sources (constitutions, parliamentary rules of procedure and specific laws on the respective offices) and from personal communication with all 25 parliaments.Footnote 49 They reflect the rules as of 1 January 2006. Three general coding decisions should be mentioned. First, I use the rules for the last ballot because actors should evaluate the strategic situation from this endpoint, even though many elections succeed on earlier ballots. Second, when different selection methods are used for officeholders in collegial institutions (especially constitutional courts), I weigh the index value by the proportion of officeholders elected by parliament. The resulting ICI score is a conservative estimate because it implicitly assumes that members selected by other methods share the preferences of the cabinet. Third, when officeholders are elected by mixed assemblies consisting of deputies and other actors such as members of a second chamber or subnational parliaments, I weigh the index value by the proportion of deputies in the assembly. Again, this procedure may underestimate the incentives to constrain if these other members have preferences different from those of the cabinet.
Incentives to Constrain the Cabinet
Table 2 provides descriptive information on how officeholders in the four CIs are selected in our 25 democracies. For cabinet appointments and elections in parliament, the ICI allows conclusions about their incentives to constrain the cabinet. I do not discuss officeholders appointed by third parties because we cannot assess their preferences based on the selection method alone. The second column for each CI relates the term length to the maximal inter-election period, providing an uncertainty estimate for the ICI's ability to capture incentives of external officeholders over time.
Table 2 Incentives of Officeholders in Four CIs to Constrain the Cabinet in 25 European Democracies

Notes: ICI = Incentives to Constrain Index; TP = appointment by third party; – = office does not exist; n.a. = not applicable (uncertainty measure not applicable for third-party appointments). Italics indicate no conclusions possible about deviant preferences.
* Uncertainty of ICI: the first score indicates the term length of the external officeholder in years; 99 indicates a selection for life or until a fixed retirement age. The second number indicates the length of the regular inter-election period in years.
Table 2 shows clear differences in how officeholders in different CIs are chosen. Parliaments play an important role in electing constitutional judges, even though in many cases they elect only some of them. Although central banks are often portrayed as independent of the cabinet, their heads are often appointed by the latter. Whatever other rules may promote independence after taking office, the selection procedures provide many cabinets with influence over the preferences of central bankers. By contrast, almost all heads of audit institutions and ombudsmen are elected by parliament, consistent with their frequent characterization as parliamentary support offices.
There is substantial empirical variation in the ICI. The highest possible value indicating a competitive election between several candidates nominated by parliament requiring a qualified majority on a secret vote is observed for the election of ombudsmen in Spain and Portugal, and in a few other cases that obtain lower overall ICI scores due to the weighting described above. On the other hand, the head of the Slovak central bank and the British comptroller and auditor general are nominated by the cabinet and approved in an open vote requiring a simple majority. In this situation, the selection of officeholders with deviant preferences is only slightly more likely than in cases of a direct appointment by the cabinet (ICI = 0; 19 cases). Many different index values are observed between these extremes.
The mean ICI scores vary between offices. Ombudsmen reach the highest average, followed by constitutional judges. The latter's score is decreased by the fact that parliaments often elect only some judges, albeit often by qualified majority in a secret vote. The mean value for heads of audit institutions is slightly lower than for constitutional judges, but the difference is small and the medians are equal for both distributions. Finally, heads of central banks are much less likely to hold deviating preferences than all other officeholders covered in the analysis. Due to the small number of cases, only the difference between heads of central banks and the other CIs is statistically significant.
The data on term length also display clear office differences. Constitutional judges have by far the longest terms; the median value of nine years is more than twice as long as the median inter-election period.Footnote 50 Thus alternations in cabinet composition are quite likely to change their incentives to constrain the cabinet over time. These changes are unpredictable because the outcomes of future elections and potential coalition bargaining are unknown. This uncertainty may be another reason, along with qualified majority requirements, why positions on constitutional courts are often allocated according to an informal quota that includes most established parties. Heads of audit institutions are also often chosen for terms considerably longer than an inter-election period. By contrast, the term length of heads of central banks and particularly ombudsmen are more frequently equal to or only slightly longer than the regular inter-election period. Overall, the ICI is not a perfect measure over time. Nonetheless, it captures the incentives dimension at the time of selection and may do so for a long period because elections do not necessarily lead to cabinet turnover. Empirical data even show that in 15 Western European democracies during the period 1950–2000 only about one in five elections led to wholesale cabinet alternation, i.e. a new cabinet being formed solely by former opposition parties.Footnote 51
Institutional Powers to Constrain the Cabinet
Established indicators are available to measure institutional powers as the second dimension of constraint for three of our four CIs, albeit only for subsets of the 25 countries. The power of constitutional courts is captured by the ‘judicial powers score’, an additive index based on six variables. Footnote 52 I use an index developed by Grilli and colleagues to measure the institutional power of central banks.Footnote 53 The index values reflect the situation in 2003 for the Western European countries, the Czech Republic and Poland, and in 1999 for the remaining Eastern European cases. Thus it captures most institutional changes induced by the Economic and Monetary Union (EMU).Footnote 54 The institutional powers of ombudsmen are measured with an index of 11 investigative competencies which is only available for the sample of countries used by Lijphart.Footnote 55
I use information from a recent survey by the British National Audit Office to construct a new index of the institutional powers of audit institutions.Footnote 56 First, the survey distinguishes between a priori audit and the following three types of ex post audit: judicial audits examine the records of individual persons responsible for transactions, financial audits report on state accounts, and performance audits deal with the efficiency and effectiveness of expenditures. Second, audit institutions are stronger the more means they have to attract attention from the public and political actors. The survey includes three such means: reports to the plenary of parliament containing specific audit results, reports to parliamentary committees and statements such as annual reports dealing with the institution's work in more general terms. My institutional powers index awards one point for each type of report and ex post audit and three points for a priori audits as these are a particularly powerful weapon because they can be used at a stage when cabinet action can still be stopped in the legislative or administrative process. Table 3 displays the scores of the four CIs on the powers dimension.

Figure 1 The Relationship Between the Powers and Incentives Dimensions
Table 3 Institutional Powers of Four CIs in 25 European Democracies

Notes: – = office does not exist; n.d. = office exists but no data available.
The data sources are given in the text body.
The abbreviations for country names, which are also used in Figure Figure 1, follow ISO code 3166.
Constraint by External Officeholders as an Empirically Two-dimensional Phenomenon
This section empirically confirms the theoretical expectation that formal powers of CI and the incentives of officeholders to use them are largely independent dimensions. > Figure 1 displays the relationship between the powers and incentives dimensions measured by the various powers indices and the ICI respectively. The figure contains separate graphs for the four CIs along with linear best-fit lines and their 95 per cent confidence intervals.Footnote 57 The length of the x-axis shows the theoretical range of the institutional powers index.Footnote 58 The unstandardized regression coefficient, its p-value based on two-sided tests and the sample size are displayed below each plot.Footnote 59
The data show that the two dimensions are significantly related for only two of the four CIs. For central banks, I find a moderately strong and statistically significant negative relationship indicating that more powerful heads of central banks actually have lower incentives to use their powers. Thus, some of the constraint expected from strong institutional powers is unlikely to materialize once we introduce actors’ preferences into the picture. Institutionally more powerful ombudsmen, on the other hand, also tend to have more incentives to use their powers. However, this relationship becomes much weaker and statistically insignificant when the powerless British ombudsman is excluded (b = 1.15; p = 0.267). For constitutional judges, the two dimensions are completely independent, with a regression coefficient very close to 0. Finally, the substantively weak and statistically insignificant negative relationship for heads of audit institutions indicates that more powerful officeholders might have somewhat weaker incentives to use their powers. These findings are robust towards other assumptions in constructing the ICI; detailed robustness tests are available from the author. Overall, there is substantial empirical support for the hypothesis that institutional powers and incentives for their use are independent or at most weakly related dimensions for CIs in European democracies. Thus, neglecting one of these dimensions is not only theoretically deficient but will also lead to empirically biased conclusions regarding the constraint we can expect from these offices.
The neglect of the incentives dimension may explain some puzzling findings in the literature. First, many scholars have detected differences between the formal and factual power of central banks. While political economists discuss a number of causes for this divergence,Footnote 60 an additional reason may be the relative neglect of the preferences of central bankers in these analyses. A recent study by Adolph points in this direction by showing that the partisan composition of cabinets affects the preferences of central bankers, which in turn affect policy outputs.Footnote 61 Second, the types-of-democracy literature has produced mixed findings on the effects of CIs on policy outputs.Footnote 62 These differences can in part be explained by conceptual differences, especially the degree to which several institutions are aggregated into summary measures. In addition, however, these studies rely solely on institutional powers and again neglect the preferences of external officeholders such as constitutional judges and central bankers. The results of this article show that this focus is clearly not sufficient for assessing the effects of CIs.
CONCLUSION
The institutional powers of external officeholders and their incentives to use them are theoretically and empirically distinct dimensions. Both have to be studied in order to arrive at an unbiased estimate of the degree to which CIs can be expected to constrain the cabinet. This article identified the selection method of external officeholders as a feasible proxy for measuring the thus-far neglected incentives dimension in broad comparative studies and proposed the Incentives to Constrain Index (ICI) as a quantitative measure. The empirical analysis of four external constraint institutions in 25 European democracies yields three central findings. First, selection methods differ substantially across offices and countries. On average, deviating preferences are most likely for ombudsmen, somewhat less likely for constitutional judges and heads of audit institutions, and fairly unlikely for heads of central banks. Second, many officeholders in CIs are linked to the parliamentary majority and the cabinet via their selection. Such CIs are less ‘external’ than is often assumed and may lose their constraining effect as a result of the unifying bond of partisanship. Third, the incentives of external officeholders to constrain the cabinet are not strongly related to their institutional powers. I find moderate relationships between the powers and incentives dimensions for ombudsmen (positive) and heads of central banks (negative). For constitutional judges and audit institutions, the two dimensions are virtually unrelated. Thus, analyses relying solely on institutional powers will most probably produce biased findings.
My findings open up at least two interesting perspectives for future research that could use the ICI developed in this article as independent and dependent variable, respectively. First, my theoretical argument yields testable hypotheses about the behaviour of external officeholders towards the cabinet: external officeholders with high ICI scores should use their powers more forcefully than officeholders with lower scores; longer terms of external officeholders relative to the inter-election period should weaken this relationship. Second, the observed variation in selection methods for CIs is a puzzle for institutional design research. Why do some institutional designers devise selection methods that induce strong incentives to constrain the cabinet while others do not? The observed office differences may be explicable in broadly functional terms, even though the strong role of cabinets in selecting heads of central banks contradicts the standard argument that independent central banks are a means for political majorities to commit credibly to long-term monetary policy goals such as low inflation. Differences between countries, on the other hand, could be tied to the characteristics of the competitive context – such as polarization, fragmentation and cabinet format – that affect the utility that political actors derive from different institutional rules.Footnote 63 Pursuing this institutional design perspective systematically using the selection methods of CIs and the ICI as a dependent variable could contribute to a more dynamic view of how constraint institutions and ultimately types of democracy affect and are affected by political processes and policy outputs.