Geographically targeted measures and electoral institutions

Our point of departure is Persson and Tabellini’s (Reference Persson and Tabellini1999, Reference Persson and Tabellini2000) seminal model of competition between two office-seeking parties, which offer a district-specific transfer and a public good to the electorate. These scholars show that, in equilibrium, spending on district-specific transfers is higher and on the public good lower in a polity with three districts than in one with a single district. In the former case, the chance of reelection is maximised if spending is shifted from the public good to transfers in favour of marginal districts.Footnote ² Early evidence has not been encouraging,Footnote ³ but later studies have shown more promise. For instance, Chang (Reference Chang2008) reported slightly higher district-specific spending, close to elections, in countries with a single veto player and single-member electoral districts. Rickard (Reference Rickard2012a) reported higher spending on subsidies in countries with majoritarian or mixed-member majoritarian electoral systems. In the more specific case of subsidies for the manufacturing sector, Rickard (Reference Rickard2012b) found more spending in majoritarian systems when manufacturing employment is geographically concentrated and the opposite when employment is diffuse. Many of these works solely distinguish, with some coding differences, between single-member district and other systems (e.g. Persson and Tabellini Reference Persson and Tabellini2003, Reference Persson and Tabellini2004; Cao et al. Reference Cao, Prakash and Ward2007; Chang Reference Chang2008; cf. Rickard Reference Rickard2012a, Reference Rickard2012b), ignoring the considerable heterogeneity of the latter group. After all, the model of Persson and Tabellini (Reference Persson and Tabellini1999, Reference Persson and Tabellini2000) suggests that district magnitude should drive district-specific spending.

District magnitude, as well as ballot structure, matter because they may “undermine to varying degrees the alignment of […] collective goals [i.e. maximizing party seats in the legislature] and the individual goals of candidates and legislators [i.e. (re)election]” (Shugart Reference Shugart, Valdini and Suominen2013, 818). These features specify how seats are allocated among candidates within parties and affect the incentives of candidates to develop personal reputations distinct from those of their parties (Carey and Shugart Reference Carey and Shugart1995, 417–418; see also Shugart et al. Reference Shugart2005), and usually these reputations are built by securing measures that deliver district-specific benefits.

Carey and Shugart (Reference Carey and Shugart1995) identified three key characteristics of the ballot that – interacting with district magnitude – shape the incentives to build a personal reputation. The first one is ballot control – that is, the degree of control party leaders exercise over the access to their party’s label. If leaders have control over the ballot, such as in closed-list PR systems, the incentives to cultivate personal reputations decrease with the magnitude of the district. Only if there are few seats in a district, with correspondingly few and identifiable candidates, politicians may have incentives to highlight their personal record or characteristics. On the other hand, if leaders do not have control over the ballot, such as in open-list PR systems, candidates of the same party must compete against each other to secure a seat. As the magnitude of the district increases, there are stronger incentives to distinguish oneself from a larger crowd of copartisans.

The second characteristic is vote pooling – that is, whether the votes cast for one candidate of a given party contribute to the number of seats won in the district by the party as a whole as well. The last characteristic is whether voters are allowed to cast only a single vote for a party, multiple votes or a single vote for a candidate. For instance, the electors of the Danish Folketingesst can choose either a candidate or a party, and votes are pooled across the whole party for the assignment of seats to parties. In the Irish single transferable vote system, voters rank candidates and pooling takes place across candidates only. The incentives to cultivate a personal reputation increase with less pooling and multiple- or candidate-level voting, especially if the district magnitude increases because intraparty competition becomes more intense.

Admittedly, evidence for the impact of district magnitude and ballot structure on spending is still limited. For instance, as a qualified support for Persson and Tabellini’s (public good-related) expectation, Edwards and Thames (Reference Edwards and Thames2007) found that education expenditure increases with district magnitude in systems with low incentives to cultivate a personal vote, but it decreases in systems with high incentives. Golden and Picci (Reference Golden and Picci2008) illustrated the importance of the open-list in PR systems by showing how it biases infrastructural investment. Other studies that have investigated the impact of these electoral institutions on spending, such as those of Hallerberg and Marier (Reference Hallerberg and Marier2004) and Rickard (Reference Rickard2012b), do not subject Carey and Shugart’s expectations to testing. Closer to our interest, Franchino and Mainenti (Reference Franchino and Mainenti2013) found that EU governments spend less on state aid when high district magnitude is associated with ballot control, party-based voting and pooling. In these circumstances, fewer measures are also adopted. On the other hand, more measures are implemented when high magnitude is combined with a lack of ballot control.

Beyond spending: electoral institutions and compliance

There is growing evidence that electoral institutions shape public spending, but why would they influence compliance with EU law as well? Couldn’t governments just spend more on these policies, without running afoul of EU provisions? As discussed above, the persistence and patterns of unlawful aid measures are puzzling. Unlawful aid will be most likely detected sooner or later, either by the Commission or by an interested party, as the cost to lodge a complaint is negligible. Even if there is some probability of escaping the Commission’s radar, this cannot explain cross-country variation, unless you implausibly argue that the supranational executive is biased in performing its duties.

In order to understand why governments may be willing to run these risks, we should start from the premise that government survival in parliamentary systems rests on the ongoing support of a party or a coalition of parties in the legislature,Footnote ⁴ and the incentives of legislators may not be aligned with the collective goals of their party. From the government perspective, implementing the measure without notification carries the risk of detection, the embarrassment of recovery and the criticism of incompetence, having invested scarce time and resources to no avail. However, these calculations look quite different from the perspective of the politician representing the constituency benefitting from the measure and whose individual reputation for dealing with a pressing local problem is on the line. The supranational procedure is costly and time consuming; it can take up to 20 months for a decision. Detection occurs several months after the adoption; 30% of the unlawful measures adopted between 2000 and 2012 have been detected more than one year after their implementation. Legislators can meanwhile preserve their reputation, taking credit for the attempted action and eventually blaming the Commission for blocking the measure.

As electoral institutions can shape the severity of the misalignment between the goals of a government party and those of its members, they can also influence the degree to which governments eventually comply with the procedure. Evidence in favour of this expectation has been recently produced by Rickard (Reference Rickard2010), who found that governments elected via majoritarian electoral rules are less likely to comply with the restrictions on narrowly targeted transfers, agreed within the WTO, than governments elected via PR. Even stronger evidence, because noncompliance in this context may have serious personal implications, is offered by Chang and Golden (Reference Chang and Golden2007). These scholars showed that, at high levels of district magnitude, (the perception of) corruption is greater in open-list than in closed-list PR systems. In some circumstances, the opposite is true at low levels of district magnitude.Footnote ⁵ If electoral institutions shape the incentives to undertake illegal activities, such as corruption, for the search of the personal vote, they should also influence compliance with EU law for the sake of securing geographically targeted measures that run afoul of its provisions.

In conclusion, the way electoral institutions influence the propensity for geographically targeted spending, as discussed in the previous section, should shape compliance with EU provisions on state aid control in a similar way. This brings us to a set of expectations. The first one follows Persson and Tabellini (Reference Persson and Tabellini1999, Reference Persson and Tabellini2000) and the work of Rickard (Reference Rickard2010) on compliance. It simply states that the lower the district magnitude, the higher the likelihood of adopting unlawful measures (H1).

The next expectations take seriously Carey and Shugart’s (Reference Carey and Shugart1995) contribution on the interactive effects of electoral rules on the search for the personal vote, following Chang and Golden’s (Reference Chang and Golden2007) work on corruption. Let us focus first on the interaction between ballot control and district magnitude. We should expect that if party leaders have control over the ballot, the lower the district magnitude, the higher the likelihood of adopting unlawful state aid measures. Conversely, if party leaders have no control over the ballot, the higher the district magnitude, the higher the likelihood of adopting unlawful state aid measures (H2). In other words, the marginal effect of district magnitude on the likelihood of adopting unlawful measures should be negative if party leaders have control over the ballot; it should be positive if party leaders have no such control.

Following the advice of Berry et al. (Reference Berry, Golder and Milton2012) to exploit fully the symmetric feature of theories positing interactions, what would, therefore, be the effect of ballot control on the likelihood of adopting unlawful measures? Compared with open-list systems, a closed-list system should unequivocally reduce the incentives to pursue the personal vote, and therefore decrease the propensity to adopt unlawful measures, for any value of district magnitude. However, this effect should be weaker when few seats are available in the district, because politicians in closed-list systems would, nevertheless, be tempted to highlight their personal record when competing against few other candidates (Carey and Shugart Reference Carey and Shugart1995, 431; Shugart Reference Shugart, Valdini and Suominen2013, 819). In other words, the marginal effect of ballot control on the likelihood of adopting unlawful measures should be negative at all values of district magnitude; this effect should be the weakest when district magnitude is at low levels and it should increase in magnitude as district magnitude increases. As we will explain in more detail below, single-member district systems are assumed to have no ballot control, and therefore this expectation is only meaningful for electoral systems with district magnitudes higher than 1.

Let us move on now to the interaction between district magnitude and the other electoral rules. We should expect that, as electoral systems provide for less pooling and for multiple or candidate-level voting (i.e. as these rules induce more search for the personal vote), the higher the district magnitude, the higher the likelihood of adopting unlawful state aid measures (H3). In other words, the marginal effect of district magnitude on the likelihood of adopting unlawful measures should be positive at all values of personal vote-inducing electoral rules. This effect should be the weakest when these rules are at their lowest level (e.g. in the case of pooling and party-level voting) and it should strengthen in magnitude as their value increases (e.g. in the case of no pooling and multiple- or candidate-level voting). The symmetric proposition is straightforward: the marginal effect of personal vote-inducing electoral rules on the likelihood of adopting unlawful measures should be positive at all values of district magnitude; this effect should be the weakest when district magnitude is at its lowest levels and it should strengthen as district magnitude increases. Again, as single-member district electoral systems have exclusively candidate-level voting and no pooling – that is, the highest values in terms of personal vote-inducing electoral rules – this proposition is only meaningful when district magnitude is higher than 1. The following sections subject these expectations to testing.

Unlawful state aid measures

There are several benefits of focussing on the EU policy of state aid control. First, state aid fits neatly the definition of distributive policy provided by Weingast et al. (Reference Weingast, Shepsle and Johnsen1981, 644)Footnote ⁶ as it provides a selective financial benefit to social groups and is geographically concentrated. Information on state aid measures can, therefore, be appropriately used to test hypotheses related to distributive policies, as works using government subsides do (Chang Reference Chang2008; Rickard Reference Rickard2012a). Second, determining noncompliance is straightforward as national governments can implement state aid only upon notification to and approval by the Commission. Third, this procedure has been consolidated over the past few decades (Cini and McGowan Reference Cini and McGowan2008, 175; Aydin and Thomas Reference Aydin and Thomas2012, 533–537; Kassim and Lyons Reference Kassim and Lyons2013, 8–9). Data on EU state aid are now comprehensive in terms of coverage of transactions and economic sectors and are highly comparable across countries as they fall under common EU legal criteria (Franchino and Mainenti Reference Franchino and Mainenti2013).

We used the information on unlawful aid to industry and services available from the official database of the European Commission,Footnote ⁷ covering the period from 2000 to 2012. In other words, we considered all unlawful aid that has been investigated by the Commission following the procedure codified in Council Regulation 659/1999. The dependent variable, Unlawful Measures, is the ratio between the number of unlawful aid measures – those that have been implemented either without notification to the Commission or before the Commission had taken a decision – over the total number of measures adopted by a given member state in a given year.Footnote ⁸

Electoral institutions

We begin with the District Magnitude of an electoral system – an explanatory variable at the centre of our expectations. We have computed the weighted average district magnitude of each electoral system that member states have used between 2000 and 2012. In the case of majoritarian and proportional systems, the value of District Magnitude is equal to the ratio between the number of assembly seats and the number of electoral districts, at the electoral tier at which votes are translated into seats. In the case of proportional multi-tier systems and mixed systems, we first computed the magnitude of each tier by dividing the number of seats assigned or won at a given tier in a given election by the number of electoral districts at that tier. We then summed the resulting tier-level values, weighted by the proportion of tier-level seats over the assembly size.Footnote ⁹

Ballot Control is a dummy variable taking the value of 1 if a large majority of the assembly is elected under a close-list electoral system.Footnote ¹⁰ For the remaining rules on pooling and votes, we use the data collected by Johnson, Wallack and their colleagues (Wallack et al. Reference Wallack, Gaviria, Panizza and Stein2003; Johnson and Wallack Reference Johnson and Wallack2007). We use a syncretic variable Pool-Vote Rules that averages the scores of the “pool” and “vote” indices produced by these scholars (for a similar procedure, see Edwards and Thames Reference Edwards and Thames2007). Higher values indicate that these rules strengthen the incentives to search for a personal vote. We also consider an alternative measure that adds up the scores of these two indices.

Control variables

We control for some additional factors that may influence compliance. First, the political business cycle model suggests that governments have incentives to manipulate the level of unemployment through, for instance, higher public investment in order to maximise the chances of winning the next election (Nordhaus Reference Nordhaus1975, 174). Unlike other contributions (e.g. Aydin Reference Aydin2007; Zahariadis Reference Zahariadis2010), Chang (Reference Chang2008) has recently found that closeness to elections led to slightly higher district-specific spending in some circumstances. An approaching electoral competition may, therefore, make governments more cavalier about complying with EU provisions. The variable Election takes the value of 1 in the year preceding an election in a given country and 0 in the other years. In the election year, the value is the weighted preelection period of the year.Footnote ¹¹

Second, left-wing governments may be more inclined to public investment than right-wing governments, either because of a predisposition to believe in beneficial government intervention in the economy or because of the benefits accruing to key left-wing constituencies, such as manufacturing workers, underemployed and unemployed (Hibbs Reference Hibbs1977). A partisan effect has been found in several studies on compliance (e.g. Falkner et al. Reference Falkner, Treib, Hartlapp and Leiber2005, Reference Falkner, Treib and Holzleithner2008; Mastenbroek and Kaeding Reference Mastenbroek and Kaeding2006; Jensen Reference Jensen2007; Toshkov Reference Toshkov2007, Reference Toshkov2008; Jensen and Spoon Reference Jensen and Spoon2011; Sedelmeier Reference Sedelmeier2012), but corroborating evidence is weak in the specific case of distributive spending (Franchino and Mainenti Reference Franchino and Mainenti2013, 505). We can, nevertheless, concede that greater willingness or pressure to spend of left-wing governments could be translated into a greater propensity to escape the constraints of EU law. The underlying policy dimension is the traditional left-right economic cleavage, pitting market liberals, favouring a small state and low taxes, against interventionists, supporting public spending and a larger government. We, therefore, use the “taxes versus spending” dimension used in the expert surveys on party positions by Laver and Hunt (Reference Laver and Hunt1992) and Benoit and Laver (Reference Benoit and Laver2006). The dimension ranges from 1 for a party that promotes raising taxes to increase public services to 20 for one that promotes cutting public services to cut taxes. Government Preference is the sum of the positions of each government party along this dimension, weighted by its share of ministerial portfolios.

Third, noncompliance may be accidental rather than intentional as the management approach to compliance would suggest (e.g. Chayes and Chayes Reference Chayes and Chayes1995). Some national administrations are less experienced than others in dealing with EU policies and a failure to notify a measure could simply reflect gaps in administrative and legal know-how. Administrative experience has indeed been identified as an important facilitator of compliance (e.g. Berglund et al. Reference Berglund, Gange and Van Waarden2006; Kaeding Reference Kaeding2006; Steunenberg and Rhinard Reference Steunenberg and Rhinard2010; Haverland et al. Reference Haverland, Steunenberg and Van Waarden2011). As experience is generally gained over the years, we control for the length of EU membership of each country in a given year. Years in EU is (the logarithm of) the number of years since joining the EU. This variable is also a measure of the misfit that is likely to exist, especially in new member states, between EU state aid policy principles and preexisting legacies. Finally, we control for the gross domestic product (GDP) per capita to account for different levels of economic development across the EU, which may influence the propensity to spend, and therefore comply. This variable has also been used as proxy for administrative capacity – another facilitator of compliance (e.g. Mbaye Reference Mbaye2001; Falkner et al. Reference Falkner, Treib, Hartlapp and Leiber2005; Berglund et al. Reference Berglund, Gange and Van Waarden2006; Hille and Knill Reference Hille and Knill2006; Kaeding Reference Kaeding2006; Haverland and Romeijn Reference Haverland and Romeijn2007; Perkins and Neumayer Reference Perkins and Neumayer2007; Börzel et al. Reference Börzel, Hofmann, Panke and Sprungk2010, Reference Börzel, Hofmann and Panke2012; König and Mäder Reference König and Mäder2013). Descriptive statistics of the variables we use for the analysis are available in Table A.1.

A closer look at within-country effects of district magnitude

Electoral rules change slowly over time. Descriptive statistics show that the between-units component of their standard deviation is larger than the within-units component. Therefore, our results are driven by differences across countries to a greater extent than by differences within countries. For district magnitude, however, within-country variance makes up an appreciable portion of total variance. The average magnitude has changed in 12 states over this time period, in some cases because of redistricting (Denmark, Germany, Ireland and Poland), in others because of peculiarities of the multi-tier system (Austria, Estonia and Hungary) and in yet others because of broader reforms (Belgium, Bulgaria, Greece, Italy and Romania). We complete our empirical investigation paying particular attention to the within-country effects of district magnitude.

Let $X_{{it}}^{d} =X_{{it}} {\minus}\overline{X} _{i} $ be the deviation of variable X _it from the mean $$\overline{X} _{i} $$ of country i at time t. We use the following specification:

(2) $$\eqalignno{ Y_{{it}} = \,& \alpha _{0} {\plus}\beta _{1} DM_{{it}}^{d} {\plus}\beta _{2} \overline{{DM}} _{i} {\plus}\beta _{3} BC_{{it}} {\plus}\beta _{4} PV_{{it}}^{d} {\plus}\beta _{5} \overline{{PV}} _{i} {\plus}\beta _{6} DM_{{it}}^{d} {\times}BC_{{it}} \cr & {\plus}\beta _{7} \overline{{DM}} _{i} {\times}BC_{{it}} {\plus} \beta _{8} DM_{{it}}^{d} {\times}\overline{{PV}} _{i} {\plus}\beta _{9} \overline{{DM}} _{i} {\times}PV_{{it}}^{d} {\plus}\beta _{{10}} \overline{{DM}} _{i} {\times} \overline{{PV}} _{i} \cr & {\plus}\bibeta \circ {\mib X}_{{\mib it}}^{\mib d} {\plus}\bibeta \circ \overline{\mib X} _{\mib i} {\plus}\beta \overline{Y} _{i} {\plus}\mu _{{it}} $$

Each independent and control variable enters the specification as country mean, in order to account for the between-units effects, and as deviation from the country mean in order to account for the within-units effects. The only exception is Ballot Control because it is an indicator variable. The interactions operationalise H2 and H3.Footnote ¹⁴ The specification also includes the mean proportion $$\overline{Y} _{i} $$ of unlawful measures in country i. The results are displayed in Table 2.

Model 1 provides evidence in support of the first expectation. As the district magnitude increases within countries, governments are less likely to adopt unlawful measures. Moreover, the coefficient of District Magnitude (dev) in Model 1 of Table 2 is close to the respective coefficient in Model 1 of Table 1. Therefore, the substantive effect of an increase in magnitude within countries is equivalent to the effect of an increase when within- and between-units components are pooled. In other words, within-units effects carry a greater proportion of explanatory power than between-units effects.

Figure 5 displays the within-country marginal effects of District Magnitude on the proportion of unlawful measures in the presence and absence of ballot control, whereas Figure 6 displays the same effects for different levels of country mean Pool-Vote Rules.

Figure 5 Marginal effects of a within-country increase in district magnitude, with and without ballot control Note: Left panel Model 2, right panel Model 3, Table 2. The left half of each panel displays the effects in case of no ballot control, the right half in case of ballot control. Pool-Vote Rules (mean) cannot take the minimum value of 0 in case of no ballot control and the maximum value in case of ballot control.

Figure 6 Marginal effects of a within-country increase in district magnitude across Pool-Vote Rules Note: First column Model 2, second column Model 3, Table 2. Top row without ballot control, bottom row with ballot control.

An increase in magnitude lowers the proportion of unlawful measures in closed-list systems, but the impact is not sufficiently different from 0 at the standard level of significance (see also bottom panels in Figure 6). In open-list systems, higher magnitude leads to a higher proportion of unlawful measures when the other rules incentivising the personal vote are at their maximum values. This effect can be seen in Figure 5 and in the upper panels of Figure 6 at the right end of the Pool-Vote Rules axis. The proportion of unlawful measures increases from 11 to 56% with a standard deviation increase in within-country magnitude. The increase in within-country magnitude has a substantially larger effect when Pool-Vote Rules takes the maximum value rather than the minimum value. F-tests for the joint significance of the interaction terms reject the null hypothesis that these effects are identical (p-value⩽0.000).

Within-country effects explain of course a smaller proportion of the variation in our dependent variable, but the results point in the same direction. Consider the Italian electoral reforms in 2005. The switch from a mixed and (mostly) open-list system to a bonus-adjusted closed-list PR system where seats are allocated to parties at the national level is associated with a drop of the proportion of unlawful measures from 12 to 5% per year.