The impact of decentralisation

At its most general, the impact of decentralisation (our shorthand for vertical power sharing) on government size is conceptualised as the effect of a specific set of “interpersonal, formal or informal rules and norms” (Schmidt Reference Schmidt1993, 378, Reference Schmidt2000, 28) on political action (cf. March and Olsen Reference March and Olsen1989; Hall and Taylor Reference Hall and Taylor1996; Peters Reference Peters2011). This effect is commonly hypothesised to operate through three causal mechanisms that all relate to different aspects of decentralisation: competition, local autonomy and veto-players.

First, competition among lower-level units in terms of taxation, and service provision is thought to dampen the size of the overall state, as public entities would only raise and provide the absolute minimum of both to attract wealthy residents (Tiebout Reference Tiebout1956, 418; Besley and Case Reference Besley and Case1995; Oates Reference Obinger1999, 1122; Alesina and Spolaore Reference Alesina and Spolaore2003, 137; Treisman Reference Treisman2007, 58). Such is the famous “Leviathan hypothesis” (Brennan and Buchanan Reference Brennan and Buchanan1980, 216), which rests on several assumptions, namely complete information, unhindered or at least not too costly a resident mobility, and individuals’ rational desires of neither wanting to pay for nor demand more than absolutely necessary (cf. Tiebout Reference Tiebout1956, 419).

Given that our subsequent empirical analysis uses the 26 Swiss cantons as a comparative template, confidence in the validity of these assumptions is higher than in a cross-national analysis (see also Monogan Reference Monogan2013; Wasserfallen Reference Wasserfallen2014). The average Swiss canton has 310,000 inhabitants and spans 1,600 km² [Bundesamt für Statistik (Federal Office for Statistics) (BFS) 2015]; therefore, complete information and mobility are more likely. Also, moving in our case not only means staying in the same country, but also in the same canton, the level where several important powers are exercised (e.g. police, education, health and environment – thus there are no costs in terms of adjusting to new systems by staying within the same canton) as Switzerland is one of the most federal countries in the world (Linder Reference Linder, Zürcher and Bolliger2012; Füglister and Wasserfallen Reference Füglister and Wasserfallen2014). Finally, the existence of fiscal equivalence in terms of a convergence of decisionmakers, taxpayers and service recipients (Schaltegger and Feld Reference Schaltegger and Gorgas2003) further enhances the logic according to which “voting with the feet” (Tiebout Reference Tiebout1956) indeed leads to service provision matching tax yield.

As the Leviathan hypothesis is concerned primarily with overall government size, it only makes sense to test for the effect of this aspect of decentralisation on total public spending, which is both local and central spending combined (cf. Rodden Reference Rodden2003a, 709). Greve (Reference Greve2012, 7) equally underlines how this competitive logic of federalism would serve to “discipline governments” tout court. Hence, a first hypothesis reads as follows:

A second argument why decentralisation would contribute to smaller governments is that much of the overall state activity is “hidden” at lower levels – that is, decided, financed and carried out by subsystem entities at their own discretion. But for decentralisation to lower “central decision costs” (Greve Reference Greve2012, 6; emphasis added), local governments must have sufficient legal autonomy to actually deliver the required public services. This is an aspect that pertains not so much to competition or political influence but rather to “self-rule” (Elazar Reference Elazar1987; Hooghe et al. Reference Hooghe, Marks and Schakel2010).

That distinction between fiscal and legal autonomy (or between policy- and polity-decentralisation, see below) is often overlooked but has been made before. Watts, for example, distinguishes between the “the scope of jurisdiction exercised by each level of government, and the degree of autonomy or freedom from control by other levels of government with which a particular government performs the tasks assigned to it” (Reference Watts2008, 65–66; original emphasis). To determine the latter, he assesses the “formal allocation by the constitution of legislative powers to each level of government” as well as “the extent to which each field of jurisdiction is exclusively assigned to one level of government, concurrent or shared” (Watts Reference Watts2008, 66). Rodden equally cautions that “it is difficult to know what to make of expenditure decentralization data without additional data on the regulatory framework for subnational finance” (Reference Rodden2004, 484), such as what type of taxes can be raised or how much local discretion there is in determining the tax base (cf. Ebel and Yilmaz Reference Ebel and Yilmaz2002, 4–5). Such rules are usually fixed in the constitution, although political practice and/or legal adjudication thereof might change over time (Gibson Reference Gibson2004, 2; Greve Reference Greve2012, 8). The testable assumption arising from this is that, given local autonomy, a central government can afford to do less since lower-level entities will both provide a safeguard for assuring a minimal service provision as well as act as the first entry points for citizen demands. We thus hypothesise that:

A corollary from this is that, through increased proximity of decisionmakers to service beneficiaries, also the monitoring and sanctioning abilities of taxpayers are strengthened; thus, not only central but also local governments will spend less – and total government size decreases as in H1. However, as Rodden (Reference Rodden2003a, 701) speculates, it might well be that vested interests operate even better at the local level and/or that citizens are more demanding precisely because of better oversight abilities (cf. Oates Reference Oates1985). In both scenarios, polity-decentralisation would lead to more local spending. Hence:

Third, there is the already mentioned political aspect of de- or rather noncentralisation. The argument here is that the existence of noncentral loci of decisionmaking provides for a check on policy change and, through that, functions to curb excessive expenditure (Brennan and Buchanan Reference Brennan and Buchanan1980, 26–28; Obinger Reference O’Dwyer and Ziblatt1998, 46; Good et al. Reference Good, Hurst, Willener and Sager2012, 455). As veto-players (Tsebelis Reference Tsebelis2000), local governments may block attempts by the centre to encroach upon their policy areas by centralising functions otherwise provided by them and/or through the acquisition of new powers (Schmidt Reference Schmidt1998, 223; Braun Reference Braun2000b, 50–51; Vatter and Freitag Reference Vatter2002, 59–60; Freitag and Vatter Reference Freitag and Vatter2008, 275). Schmidt (1996, 177) also provides evidence that “countermajoritarian constraints […] have stopped or reversed the trend towards big government” (cf. also Samuels and Mainwaring Reference Samuels and Mainwaring2004, 86–88). But this means that, to have an effect on policymaking, decentralisation must not only capture expenditure and revenue discretion (the policy dimension) or constitutional autonomy (the polity dimension), but also actual local political influence at higher levels (Braun Reference Braun2000b, 36) – that is, the ability to block or initiate policy change.

Most often this aspect of territorial politics is captured by the notion of “shared rule”, which measures the extent and way in which regions codetermine national decisionmaking (cf. Rodden Reference Rodden, Eskeland and Litvack2006, 38; Hooghe et al. Reference Hooghe, Marks and Schakel2010). However, we prefer the term “political decentralisation” because it better conveys both the nature of central-local relations (political) and the direction of influence (bottom-up) (cf. Riker Reference Riker1964, 10). Thus, subnational governments codetermine central decisionmaking using different channels – for example, through representatives in central political organs, such as elected senators or the appointed delegates of minister-presidents (Rodden Reference Rodden2003b, 165). Alternatively, in the absence of upper chambers, noncentral entities might also resort to bargaining directly with the federal government (Bird and Tassonyi Reference Bird and Tassonyi2003, 94), act through political parties (Riker Reference Riker1964, 137ff.) or both (Samuels and Mainwaring Reference Samuels and Mainwaring2004, 88–90). The point here is that the more powerful these territorial veto-players, the more successfully they can object to enlarging the scope of public activity. Hence, a third hypothesis reads as follows:

However, it may also happen that lower-level entities use their influence to shift public costs upwards and/or force the centre to take on new responsibilities, thus increasing the size of the central government. Rodden (Reference Rodden, Eskeland and Litvack2006, 5, 41) argues along similar lines when emphasising central-local bargaining dynamics and possible solutions to vertical coordination problems. Thus, “local governments, working on behalf of resident taxpayers, may shift the production costs of local services onto nonresidents through federally funded transfers” (Inman Reference Inman2003, 36) that increase central spending. This very much resembles the so-called “flypaper effect” (Rodden Reference Rodden, Eskeland and Litvack2006, 78; Freitag and Vatter Reference Freitag and Vatter2008, 276) but in an opposite direction – that is, bottom-up instead of top-down. Specific examples involve the Brazilian governors “forc[ing] the central government to assume their debts”, in the early 1990s (Samuels and Mainwaring Reference Samuels and Mainwaring2004, 106), or the positive effect of legislative overrepresentation on a state’s share in federal funds in Argentina and Mexico (Diaz-Cayeros Reference Diaz-Cayeros2004, 315; Gibson et al. Reference Gibson, Calvo and Falleti2004, 181).

In other words, giving noncentral politicians a direct say over central policymaking will enable them to have the most expensive policies centralised or, in more technical terms, to “externaliz[e] the costs to others, turning public revenue into a ‘common pool’ that is overfished by provincial governments” (Rodden Reference Rodden, Eskeland and Litvack2006, 6; cf. Freitag and Vatter Reference Freitag and Vatter2008). Hence, our final hypothesis on the effect of decentralisation reads as follows:

Details on how these three different dimensions of decentralisation are measured are provided in the Operationalisation section, below, and in the Appendix. Table 1 summarises our hypotheses. Note that, in principle, interactions between the three dimensions are very well imaginable too. For example, local governments might need to possess a minimum degree of constitutional self-rule for shared rule to be operating efficiently. However, the point of our three-dimensional measurement strategy is precisely to disentangle the mere availability of resources from the power to decide on their use (self-rule dimension: policy and polity) as well as from political influence at the centre (shared rule dimension: politics). But the point about possible interaction effects will be taken up in the concluding section. We next turn to rivalling explanations.

Rivalling explanations

There are several rivalling explanations that could explain government size better than decentralisation. The first is direct democracy: as an opportunity structure with relatively low entry costs, it offers a veto instrument of a particular kind – namely, one for societal groups sufficiently well organised to collect the required number of signatures to initiate or block policy change (Wagschal and Obinger Reference Wagschal2000, 469; see also Wagschal Reference Wagschal1997, 226). But as with political decentralisation above, opening up the space of political decisions to the nonelite (i.e. not necessarily elected politicians) and the nonpolitical (moral, economic, etc.) elite could lead in both directions – that is less or more public intervention than would otherwise be the case (Freitag et al. Reference Freitag, Vatter and Müller2003, 355; Linder Reference Linder, Zürcher and Bolliger2012, 287). It all depends on the purpose and strength of these organised interests (cf. Funk and Gathmann Reference Funk and Gathmann2011, 1258). However, because from the point of view of the people’s final decisions no strategic points are to be scored in direct-democratic votes, their vote will tend to be longer term than that of politicians who want to be reelected in a few years (cf. Eichenberger Reference Eichenberger1999). Moreover, direct democracy regularly practised makes for better informed citizens, raising the bar beyond which a majority of them are convinced that policy innovation is needed (Eichenberger Reference Eichenberger1999, 268; Feld and Kirchgässner Reference Feld and Kirchgässner2000; Kirchgässner Reference Kirchgässner2000). Finally, knowing the threat of a direct-democratic veto to exist, governments will become more cautious as regards the extent of change proposed, all the more so as the default option, the status quo, is always better known and thus inherently favoured by a generally risk-averse demos (Samuelson and Zeckhauser Reference Samuelson and Zeckhauser1988; Funk and Gathmann Reference Funk and Gathmann2013).

Another institutional variable is hard budget constraints. These refer to collectively binding rules on the extent of public expenditure growth and the associated risks of a bailout (Rodden et al. Reference Rodden2003, 4) and are thus institutions par excellence (Schmidt Reference Schmidt1993, 379). Such constraints tie further spending to a corresponding surplus in generated revenue, a favourable debt-per-gross domestic product (GDP) ratio (e.g. Maastricht’s Growth and Stability Pact), and/or satisfactory economic performance in general. Switzerland and most of its cantons have chosen yet another way to ensure balanced budgets, using an instrument called “debt break” (Schuldenbremse), whereby the government is authorised to run deficits during recessions and to run surpluses during booms. Over the whole business cycle, however, it has to ensure that the budget is balanced. Thus, starting with a given level of debt, the debt should not have increased after the completion of a full cycle (Müller Reference Müller2004, 2).

The debt break, in other words, represents a specific kind of self-imposed budget constraint (cf. Rodden et al. Reference Rodden2003, 23). The idea to apply this rule to Swiss policymaking dates back to 1919, when Canton St. Gall first introduced it into its legislation (Stalder and Röhrs Reference Stalder and Röhrs2005, 12; Kirchgässner Reference Kirchgässner2010, 8). Over the 1990s and 2000s, several other cantons followed suit (BAK Basel 2012), but variations on the theme exist in terms of both the constraints imposed and the sanctions to be applied in case of rule violation (Stalder and Röhrs Reference Stalder and Röhrs2005, 3; BAK Basel 2012, 20–28). Building on a substantive body of prior evidence relating debt breaks to lower budget deficits (Feld and Kirchgässner Reference Feld and Kirchgässner2000, Reference Feld and Kirchgässner2008; Schaltegger Reference Schaltegger2002; Krogstrup and Wälti Reference Krogstrup and Wälti2008; Chatagny Reference Chatagny2013; Lüchinger and Schaltegger Reference Lüchinger and Schaltegger2013; Yerly Reference Yerly2013; see also Burret and Feld Reference Burret and Feld2014 for an overview), we would expect that stricter debt break rules lead to lower cantonal expenditures.

Finally, we include the share of voters for cantonal government parties into our empirical analysis to account for collusion. According to Lijphart (Reference Lijphart2012), consensual decision-making procedures encourage the magnitude of state intervention as minority interests have to be considered (Vatter and Freitag Reference Vatter2002, 58; Baskaran Reference Baskaran2013). The more inclusive a policy-making process, the more distributive policies are pursued for which the cost bearers are less obvious (Braun Reference Braun2000a, 13; Schniewind et al. Reference Schniewind, Freitag and Vatter2009). Thus, increased government spending might simply be a reflection of a broad governing coalition.

Operationalisation

Our dependent variable is cantonal, local and total (cantonal+local) public expenditures, measured on a per capita basis to facilitate comparability. However, per capita spending has increased in all the 26 cantons between 1990 and 2009; thus, instead of estimating absolute levels of annual per capita spending for each canton we subtract the mean of all cantons’ per capita spending for each year. In other words, we estimate the deviation from the mean cantonal per capita public spending to control for time-dependent error terms (cf. Stadelmann-Steffen and Bühlmann Reference Stadelmann-Steffen and Bühlmann2008, 36–37).Footnote ²

Turning to our key independent variables, policy-decentralisation is measured using fiscal, personnel and administrative decentralisation within every Swiss canton, understood in turn as the extent to which local governments raise and administer public money (cf. Fiechter Reference Fiechter2010; Rühli Reference Rühli2012). However, full centralisation in one area (e.g. tax-raising capacity) can easily be offset by decentralisation in another (e.g. personnel), which is to say that simply averaging their values would not render an accurate picture. In Goertz’s (Reference Goertz2006, 115) terms, therefore, all three components are necessary, and together they are jointly sufficient conditions for a canton to be decentralised in its policy dimension. We therefore multiply general revenue decentralisation with administrative (the share of local from total public expenditures for administration only) and personnel decentralisation (the share of local staff and local staff salaries from their respective total numbers; cf. Treisman Reference Treisman2002, 13; Chhibber and Kollman Reference Chhibber and Kollman2004, 234).Footnote ³

Polity-decentralisation is defined by the extent of freedom guaranteed by cantonal constitutions (Giacometti Reference Giacometti1941) and expert perceptions of the actual realisation thereof (Ladner et al. Reference Ladner, Steiner, Horber-Papazian, Fiechter, Jacot-Descombes and Kaiser2013). This takes into account possible discrepancies between “rules-in-form” and “rules-in-use” (Rothstein Reference Rothstein1996; cf. Rodden Reference Rodden2004, 492). In practice, we average the standardised values of the Giacometti index (cantonal constitutions are either centralised, decentralised or balanced; Giacometti Reference Giacometti1941) and the results of the local government secretary surveys [Gemeindeschreiberbefragung (GSB)] of 1994, 2005 and 2009 (cf. Ladner et al. Reference Ladner, Steiner, Horber-Papazian, Fiechter, Jacot-Descombes and Kaiser2013).Footnote ⁴ Averaging is possible because the two subdimensions are “substitutable” (Goertz Reference Goertz2006, 108).

Finally, politics-decentralisation captures the degree to which political decisionmaking is decentralised (i.e. local) rather than centralised (i.e. cantonal). There are seven indicators that are assessed here (cf. Mueller Reference Mueller2011, Reference Mueller2014, Reference Mueller2015):

1. 1. Cantonal political party organisation measures the local influence over candidate selection for cantonal parliamentary elections, from purely local discretion to cantonal delegate assemblies without any attachment to local politics.

2. 2. Regionalism assesses the degree to which regional assemblies and/or prefects exist in a canton – that is, whether there are additional noncentral loci situated between cantonal and local governments.Footnote ⁵

3. 3. Territorial quotas take into account the fact that electoral competition for the cantonal executive and/or the legislative branch might be restricted using fixed quotas, such as those for the Bernese Jura region (guaranteed one out of five government seats).

4. 4. Electoral system organisation measures the territorial congruence between local governments and the electoral districts used for cantonal parliamentary elections.

5. 5. The direct representation of mayors in cantonal parliaments is assessed using the self-declarations of Members of Cantonal Parliaments.

6. 6. The organisational strength of local government organisations captures the existence, cohesiveness and public presence of Local Government Associations (LGAs).

7. 7. Finally, the existence of direct-democratic instruments for local governments measures the extent to which local governments qua municipalities can veto a cantonal bill and/or initiate cantonal constitutional change.

All these indicators have in common the fact that they – at least potentially – bring local interests to bear on central decisionmaking (cf. Tarrow Reference Tarrow1977; Page Reference Page1991; Rodden Reference Rodden2004; Stepan Reference Stepan2004). To arrive at a single measure of politics-decentralisation, we rely on the results of a factor analysis of these seven indicators that searches for a single factor only (see Table A.1, in the Appendix). A reliability test of policy-, polity- and politics-decentralisation thus constructed reveals a sufficiently large commonality; therefore, to arrive at a single measure of overall decentralisation, we have calculated their arithmetic mean.Footnote ⁶ The conceptual structure of decentralisation so defined is visualised in Figure 1, while summary statistics and an empirical distribution of the mean values across the whole period are presented in the Appendix.

Figure 1 Conceptual structure of cantonal decentralisation. LGA=Local Government Association

To measure direct democracy, we use Stutzer’s (Reference Stutzer1999) index as updated by Schaub and Dlabac (Reference Schaub and Dlabac2012). It is composed of the mean values of four dimensions, each coded from 1 meaning few direct-democratic rights to 6 equalling extended direct-democratic rights. For debt breaks, we rely on Feld and Kirchgässner’s (Reference Feld and Kirchgässner2008) ordinal variable on the strictness of cantonal debt breaks (0 equals no debt break, 3 indicates the strictest debt break). The strictest debt breaks tie expenditure directly to budget planning, foresee no exceptions, and provide for sanctions in case of nonobedience. For each of these elements missing, strictness is downgraded to 2 or 1, while 0 signifies the absence of a debt break altogether. For the years from 1990 to 2005, we use the coding by Lüchinger and Schaltegger (Reference Lüchinger and Schaltegger2013, 789–790, 804) and Stalder and Röhrs (Reference Stalder and Röhrs2005, 28–30), for 2006–2007 that by Chatagny (Reference Chatagny2013, 34), and for 2008–2009 we have calculated the corresponding cantonal values ourselves based on information from the Année Politique Suisse (2009). The resulting measure does not significantly correlate with any other indicator in our dataset. Finally, as an indicator of the size of the governing coalition we use the summed share of voters for parties in a cantonal government (cf. Vatter and Freitag Reference Vatter2002, 63; data source for our purposes: BFS 2015).

As further control variables we shall use various socio-demographic, economic, cultural and structural indicators. To capture modernisation and market failure (Wagner Reference Wagner1958 [1883]; Verner Reference Verner1979), we assess urbanisation and unemployment (cf. Schmidt Reference Schmidt2000, 23; Schaltegger Reference Schaltegger2001, 4; Kellermann Reference Kellermann2007, 48). To measure those aspects of political culture potentially related to more demand for state intervention (Davis and Robinson Reference Davis and Robinson1999; Schmidt Reference Schmidt2000, 30; Loughlin Reference Loughlin2001), we assess the share of Catholics and German-speakers (Kriesi et al. Reference Kriesi, Wernli, Sciarini and Gianni1996; Stadler Reference Stadler1996; von der Weid et al. Reference Von der Weid, Bernhard and Jeanneret2002, 63–65; Zürcher Reference Zürcher2006; Linder et al. Reference Linder2008). To assess party competition and power resources (Schmidt Reference Schmidt1996, Reference Schmidt2000, 25–27), we measure the strength of left-wing parties and trade unions (cf. Hibbs Reference Hibbs1977; Schmidt Reference Schmidt1996; Wagschal Reference Wagschal and Obinger2005, 38), because to (re)distribute across social strata is politically desirable for them and their electorate or members. To assess mobility and demographic structure, we measure the share of residents older than 65 years, the share of pupils in secondary education, the share of social benefit recipients and real median income (cf. Funk and Gathmann Reference Funk and Gathmann2011, 1260). Finally, to control for the impact of changing macroeconomic conditions (Schmidt Reference Schmidt1996, 167), we measure the performance of a canton using total federal corporate tax yield per canton, divided by that canton’s population, for each year of our analysis.Footnote ⁷

For unemployment and urbanisation, the share of Catholics and French-speakers, the strength of left-wing parties in cantonal parliaments (rather than in cantonal governments, as it is the overall strength of parties and not so much the number of government seats that the theory highlights), socio-demographics and federal tax yield, we rely on data from the BFS (2015). To measure the strength of trade unions, we rely once more on data by Schaub and Dlabac (Reference Schaub and Dlabac2012) (cf. Vatter and Freitag Reference Vatter2002, 63), whereas income data are gathered from federal income tax statistics (Schaltegger and Gorgas Reference Schaltegger and Feld2011). Further details on each variable, its measurement, sources and summary statistics are listed in the Appendix.

Method

To test which of the aforementioned explanations and specifications best match the empirical reality of the Swiss cantons as 26 unit-independent cases, we estimate time-series cross-sectional (TSCS) models as our units are canton-years. We have checked that our time series is stationary using the Augmented Dickey-Fuller Unit Root Test. Models are estimated using the R package “panelAR”. The package estimates linear models on panel data structures in the presence of AR(1)-type autocorrelation that are addressed via a two-step Prais-Winsten feasible generalised least squares procedure, allowing for common correlation coefficients across all panels (Kashin Reference Kashin2014), and panel-corrected standard errors (PCSEs) that are robust to both heteroskedasticity and contemporaneous correlation across panels. Such PCSEs allow for more valid significance estimations. Note that this method of estimating is rather conservative; hence, if significant correlations are obtained, these can be accepted with even more confidence than if another method had been chosen.Footnote ⁸