Portfolio Allocation under Decentralization∗

Decentralization shapes the way policy authority is shared between the national and regional levels and that, we argue, will have consequences for government formation. In particular, we contend that the allocation of portfolios in regional coalition governments will be affected by the degree of decentralization of each policy. To analyze this relationship we exploit the cross-time, cross-regional, and cross-policy variation of the process of competence devolution to the Spanish Autonomous Communities between 1980 and 2010. We find that, as expected, an increase in the competences managed by the region in a given policy jurisdiction makes the related portfolio more attractive to all coalition parties, although the effect seems to be particularly significant in economically strong and fast-track regions. ∗This work has received the support from the Spanish Ministry of Science and Innovation through the research project “Coalitional Agreements and Government Formation in Spain, 1979-2011” (Ref. CSO201016337). †Department of Constitutional Law and Political Science, Av. Diagonal 684, 08034 Barcelona, Spain.

Political decentralization changes the way policy responsibilities are shared between levels of government. Political parties respond to these processes adapting their strategies to new scenarios. Decentralization encourages parties to concentrate more on their regional interests and strategies (Brzinski 1999) and also affects party organization, programs, and policy-making (Hopkin and Bradbury 2006), among others. And yet, we know little about how decentralization shapes the way parties negotiate over government formation. At some point during this process, coalition governments need to allocate ministerial posts among its members. Will-be partners try to get the most out of this negotiation by seeking 1) the highest number of portfolios possible (quantitative dimension) and/or 2) those they most prefer (qualitative dimension). This article examines the role of decentralization on the latter dimension.
The attractiveness of ministerial posts depends on a number of conditions, including the type of policies under the jurisdiction of the portfolio (Browne and Feste 1975;Budge and Keman 1990;Bäck et al. 2011), the importance of the post (Laver and Hunt 1992;Thies 2001;Warwick and Druckman 2001;Druckman and Warwick 2005;Warwick and Druckman 2006), and the amount of resources managed. The latter has a very particular nature in multi-level polities because policy competences are shared between the national and regional levels. A very relevant post in the national government can have little importance in regional cabinets if the related policy competences are highly centralized. Conversely, if the powers over a certain jurisdiction have been devolved to regional entities, parties at the regional level will find the corresponding portfolios more attractive.
Interestingly, literature on portfolio allocation has not yet considered the potential importance of decentralization. In this article we argue that the incentives to grab a certain portfolio in a regional coalition government change in response to decentralization processes, the key idea being that regional cabinet posts become more attractive the more decentralized the related policy is.
We seek to address this issue by studying the coalition governments formed between 1980 and 2010 in the Spanish regions. In the next section we review what has been said about the qualitative aspect of portfolio allocation in coalition governments and how political decentralization relates to this literature. Next, before explaining the hypotheses of this work, we deal with how decentralization can affect the allocation of portfolios in a multi-level context like the Spanish one. We then describe the data and operationalization of the variables used in the empirical analyses presented in the following section. Finally, the last section reflects on the implications of the main contributions of the article for the study of federalism and suggests some directions for future research.

Policy Sharing and Qualitative Portfolio Allocation
The distribution of government portfolios has a quantitative and a qualitative dimension.
While a great deal of the research on government formation processes has concentrated on the former, the latter has deserved far less attention. But the importance of the qualitative dimension of ministries' allocation in coalition governments is beyond doubt as it completes a well developed research agenda on portfolio allocation with additional relevant information: it is not just the number of cabinet posts that matters for parties, but also the type of ministries they get.
Over the years several studies have addressed this issue from different perspectives. To place them in a common framework, one could start by defining what they share. In our view, a minimal definition of the qualitative dimension of portfolio allocation requires simply that portfolios differ. The quantitative dimension views portfolios merely as numbers since the only thing that matters is how many portfolios each party obtains (every portfolio is valued 'one'). In the qualitative dimension, what matters is which portfolios parties get. If we accept this minimal criterion, two sides of the qualitative dimension follow.
There is first the supply side. Portfolios are different but each one provides the same incentives to all the parties involved in the coalition game. In other words, what makes portfolios different is not party-dependent. Supply side studies have covered issues like the importance of ministerial posts. Laver and Hunt (1992), for instance, proposed a list of portfolios ranked by importance that other authors applied to the empirical analysis of portfolio distribution.
These analyses basically confirmed that parties tend to receive office payoffs -now taking their relative importance into account-proportional to their parliamentary contribution to the government (e.g. Thies 2001;Warwick and Druckman 2001). Later, Druckman and Warwick (2005) published what they considered to be the missing piece: a country-specific numeric index of each portfolio salience, richer than mere rank-orderings, which covered an important number of nations. Again, the empirical application of this index essentially reinforced previous conclusions (Warwick and Druckman 2006).
But there is also a demand side to the qualitative dimension of portfolio allocation. Portfolios can also be regarded as different by those who demand them: i.e. political parties. In other words, what makes portfolios different are the qualities that each party (according to its preferences) attributes to each post. Examples of this can be found already in Browne and Feste (1975), who investigated how parties' ideological ascription affected their choice of certain kinds of portfolios. Later, Budge and Keman (1990) also dealt with how parties' ideological family affected the way they negotiate over portfolios. And in a recent article, Bäck et al. (2011) found that, in Western Europe, a greater policy salience makes parties more likely to obtain control over the portfolios that have the related policy remit.
The implication of Laver and Shepsle's (1996) 'portfolio allocation model' (PA model) for the qualitative allocation of cabinet posts can be considered as a special case of the demand side arguments. The PA model implies that parties controlling the median position on a specific dimension are more likely to be allocated that dimension's portfolio. That means that it depends on party characteristics (i.e. being median) who will get a certain portfolio.
In this model, portfolios do not differ regardless of parties but by virtue of their attributes. This is what the demand side of qualitative portfolio allocation requires.
To sum up, different portfolios can be of different interest for the parties involved in a coalition negotiation because 1) the post may be 'objectively' different than another one (supply side) and/or because 2) the policies under the post's jurisdiction may 'subjectively' mean different things for each individual party (demand side). This hopefully useful division can help organize the arguments behind qualitative portfolio allocation. In this article we present a supply side argument that, to our knowledge, has not been explored hitherto. Apart from its 'nominal' importance, the degree of policy authority that a party will be able to exercise within the ministerial jurisdiction will vitally depend on which administrative level is actually deciding its policies. And that can make some ministerial posts more attractive than others. It is very seldom the case that a single administrative level entirely decides the issues under a policy jurisdiction. In many cases, supranational political entities exist that somehow limit the extent to which parties in national governments can develop their policy agenda. And, within national boundaries, various regional entities share political power with the national administration.
But obviously the intensity of this policy sharing differs across jurisdictions depending on their level of decentralization. Some governments will manage the policies under certain portfolios more autonomously than others where authority is shared between various administrations. How policy authority is shared between levels, this article contends, will shape the incentives of coalition parties to pick one post or the other when it comes the time to allocate portfolios. The general idea is that the more the competences managed by the cabinet post of a given administrative level, the more attractive the control of that post will be.

Regional Portfolio Allocation in a Multi-level Context
The idea of policy sharing and its relationship with the allocation of portfolios is particularly relevant in multi-level countries, where political competences are divided between the national and regional levels. A relatively unimportant portfolio in the national government can have greater importance in a regional cabinet if the region has strong policy competences in the related area. On the other hand, the portfolio of a highly centralized policy will be more attractive for the parties of the national cabinet rather than for those of regional cabinets. While governance in multi-level settings has become a prominent research field in comparative politics (see e.g. Swenden 2002;Pogorelis et al. 2005;Debus 2008;Bäck et al. 2013;Däubler and Debus 2009;Stefuriuc 2009), it has not yet considered its implications for a particular aspect of government formation: the allocation of portfolios. This article argues that greater decentralization will, on average, make the control of regional cabinet posts more attractive. However, the decentralization of competences need not be homogenous across policies. In fact, it is rarely so. Defense or foreign affairs are policy sectors that the national government typically keeps for itself, even in very decentralized countries. In other policies, instead, regional governments can have strong competences. Hence, the effect of decentralization on portfolio allocation should be stronger for certain cabinet posts than for others.
This article takes Spanish regional coalition governments as a testing ground for this idea.
Spain is known to be a considerably decentralized country, where the devolution of policy authority to regional entities has experienced a dramatic increase since democratization.
Yet, this process of decentralization has varied a great deal across regions and across policies.
That, jointly with the formation of many coalition cabinets in the regional level, makes Spain the convenient source of data to test empirically the relationship between decentralization and portfolio allocation. 1 Spanish regions have been increasingly autonomous in policy-making. After the end of the Francoist regime, the approval of the Spanish Constitution of 1978 defined a rather flexible system of political decentralization that through the years evolved to a federal-like territorial distribution of political authority (Montero 2001;Moreno 2002;Hooghe et al. 2010;Sala 2013). In the early 1980s, a total of seventeen regions became Autonomous Communities (hereafter ACs). Today, taking the level of decentralization as a whole, Spain scores high in self-rule (i.e. the degree of authority exercised by the regional governments over their own issues (Hooghe et al. 2008)), and ACs' public spending accounts for approximately a 38 percent of the total. But apart form this overall assessment, there is also considerable cross-time, cross-regional, and cross-policy variation within Spain over the degree of policy authority the ACs can exercise as a result of the negotiations that take place between the central government and each AC for the transfer of competences.
The Spanish Constitution lists two types of competences, ones that are reserved exclusively for the central government (over the regulation of defense, foreign policy, justice, criminal and commercial law, foreign trade, currency, citizenship, customs, and immigration) and others that can be transferred to the ACs (e.g. Aja 2003;Bäck et al. 2013;Chaqués and Palau 2011).
Each AC can claim to exercise authority over the wide range of issues contained in the latter, which are typically assumed in the Statutes of Autonomy of each AC (requiring approval in the regional and national parliaments). Nonetheless, this de jure decentralization is actually confirmed through the approval of Royal Transfer Decrees (Reales Decretos de Traspasos, hereafter RTDs) that make effective the devolution of policy authority over specific issues to ACs. The distribution of political authority across issues may be revised and redefined according to bilateral negotiations between the national and each regional government (Linz and Stepan 2006). The results of these bargaining processes are then expressed in new RTDs. Therefore, even for those ACs that reached the maximum level of formal autonomy in the early 1980s, subsequent negotiations that took place over the years effectively increased their autonomy over a wide range of issues through the approval of the RTDs. The regional asymmetry and policy-specificity of this decentralization process, we argue, should have consequences on the formation of regional governments. In particular, the members of a coalition government are expected to take into account how decentralized each policy is when it is time to allocate policy responsibilities, which will depend on the region and time (and the particular policy). The hypotheses derived from this expectation are explained in greater detail next.

Hypotheses
This article aims at exploiting the sources of variation mentioned above by looking at how the transfer of competences affected the choice over portfolios in the coalition governments that formed in the Spanish ACs between 1980 and 2010. We will first try to identify the portfolios that Spanish parties have found more attractive at the time of allocation. We expect that the greater the size and bargaining power of the party, the more likely they will be to obtain the best portfolios. In this mainly descriptive stage we will simply analyze the impact of parties' strength on portfolio choice with the idea that a positive relationship between the two reflects the attractiveness of the cabinet post. Similarly, we expect to find that certain types of parties (left-wing, conservative, and regionalist) will prioritize the obtention of some portfolios over the others. This is something we will also examine in this first stage.
The latter, though, is not an objective of this study as such. Our key argument is that increasing competences for the regional level in each policy jurisdiction will make the related portfolios more attractive. The logical consequence of this idea is that when a portfolio becomes more appealing parties will use their resources to get it. To test this argument we will assess the impact of the jurisdictional decentralization on the relationship between party strength and portfolio choice, the main hypothesis being:

Hypothesis 1
The greater the decentralization of a policy, the more party strength will determine the choice of the related regional portfolio.
This general hypothesis should apply, in principle, to all Spanish regions. Nonetheless, the main effect may well be conditional on or moderated by some regional characteristics. First, one could expect that the region's economic capacity will shape the attractiveness of taking over the decentralized policies. In economically weaker ACs, party leaders may be reluctant to receive policy-making responsibilities when the capacity to actually decide autonomously is financially jeopardized. Both if parties were motivated by policy influence or office spoils, regional economic capacity will increase the incentives to take the control of decentralized policies because they will find it easier to influence policy and, also, the spoils derived from office that parties are going to distribute will also be more juicy. In other words, in poor regions, policy-making autonomy might be a double-edged sword: having the ability to decide the direction of policies may clash with having the capacity to do so. The hypothesis that follows from this is: Hypothesis 2 The effect of the decentralization of a policy on regional portfolio choice will be stronger in economically capable regions.
Another potential moderator of the main effect can be the overall asymmetry of the Spanish decentralization process. The starting point was clearly different depending on the region.
The Basque Country, Catalonia, Galicia, Andalusia (through the special route of the 151 article of the Spanish Constitution), and also Navarre (through an exceptional constitutional route) accessed the autonomy through a fast-track procedure that granted them a high level of competences already in the early 1980s. According to León (2010), different levels of decentralization in the regions affect the ability of voters to hold governments accountable.
In low decentralized regions, the attribution of responsibilities is not problematic because it is clear that the national government is the one to blame or give credit to. In highly decentralized regions, voters also find it easy to identify the regional government as responsible for the policies whose competences have been devolved to the region. Nonetheless, in medium stages of decentralization, León (2010) argues, accountability is much more problematic.
An implication of this argument is that increasing decentralization should provide parties with different incentives for portfolio control depending on the region. If the region moves from a low decentralization stage to a medium decentralization stage, then regional political parties will have little incentive to choose the decentralized portfolios since the credit deserved for policy-making is unlikely to be profitable. On the other hand, increasing decentralization can make portfolio control particularly attractive in those regions whose starting competence levels were higher. If so, decentralization will make decentralized portfolios more attractive for parties trying to use policy as a means to advance their goals and claim credit for them.
Hence, the third hypothesis conjectures that the effect of decentralization on portfolio choice will vary across regions according to their initial level of decentralization: Hypothesis 3 The effect of the decentralization of a policy on regional portfolio choice will be stronger in regions that started with a higher level of competences.
Read together, the latter hypotheses imply that decentralization will make portfolio control more attractive in those regions that are economically strong and started with high levels of decentralization. Therefore, we will also test these hypotheses simultaneously to see to what extent the effect of the decentralization of a policy on regional portfolio choice will be stronger in economically capable regions that started with a higher level of competences.
Finally, it bears mentioning that the Spanish party system is characterized by the competition of national and regional parties which can be referred to as state-wide parties and non-state-wide parties, respectively (henceforth SWPs and NSWPs). SWPs compete in all districts in all elections whereas NSWPs contest regional and/or polity-wide elections in one or few regions. As Reniu (2011) argues, the difference in their territorial pervasiveness makes their goals diverge. Whereas SWPs are said to value access to regional executive offices (only) instrumentally -that is, as a 'bargaining chip' to negotiate parliamentary support at the national level-, NSWPs are considered to intrinsically care about cabinet posts at the regional level (Reniu 2011, 119). If so, the obtention of specific cabinet posts will matter more for NSWPs, since SWPs will instead give more importance to the implications of their general formation strategy -either oppose, participate in, or offer outside support to the regional government-2 and less about the specific allocation of portfolios. Hence, the effect of increasing regional policy competences will increase the likability of the related portfolios especially for NSWPs rather than for SWPs. The hypothesis we put forward is hence the following: Hypothesis 4 The effect of the decentralization of a policy on regional portfolio choice will be stronger for NSWPs as compared to SWPs.
It is true that one could think of alternative mechanisms relating the type of party, decentralization, and portfolio choice, but in any case what we want to do is to check whether or not the effect of competence devolution on the allocation of cabinet posts is different for SWPs and NSWPs.

Data and Method
Our empirical analyses are run on a recently created dataset on ministerial appointments of the Spanish regional coalition governments from 1980 to 2010. To build this dataset we took several sources of information containing the names, partisan affiliation, and portfolios controlled by all the ministers (consejeros) of the regional cabinets. The main primary sources have been the annals of one of the most important Spanish newspapers (Anuarios El País ), complemented with case-by-case research to complete some missing information on ministers' partisanship. Information on the seat shares of both governmental and opposition parties was added in the dataset in order to calculate party sizes and bargaining power.
Those cabinets participated by two or more parties which ran in joint lists in the previous elections were excluded from the database. Pre-electoral coalitions present a measurement problem since it is very difficult to discern the seats that each separate party has won in elections and therefore it is not possible to estimate their relative strength. Hence, they have been considered as single parties. Finally, following Strøm's (1990) counting rules, the definition of a new government is based on the partisan composition (any change in the cabinet parties is counted as a new government) and on the call of elections (even if the partisan composition of a cabinet is repeated after elections, these are counted as two different governments). Nonetheless, the Anuarios El País contain the information on regional cabinets' composition as of each year's January 1st. As a result, those (few) governments whose start and termination date happened within a calendar year are not taken into account.
Between 1980 and 2010, forty-six regional coalition governments formed in the ACs of Spain, thirty-five of which started immediately after elections and eleven were inter-election. On the other hand, thirty-three of these coalition governments commanded the majority of seats in their legislature, while, perhaps surprisingly, almost a third had a minority status (thirteen specifically). 3 The descriptive statistics in Table 1 show, by AC, the number of coalitions that formed in Spain up to 2010 as well as information on two of their characteristics. 4 Looking at the totals, we can see that the average number of coalition members in the forty-six cabinets analyzed is closer to two than to three (2.28), with a relatively low standard deviation (0.5) and a maximum and minimum of two and four, respectively. As for the number of portfolios that coalitions allocate to their members, the mean is around twelve (11.61), although the inter-governmental variance is considerable, with a minimum of seven and a maximum of seventeen (standard deviation of 2.31).
As said, after the creation of the ACs, regional governments negotiated with the central government the actual transfers of competences. These negotiations were reflected in the approval of many RTDs over the whole democratic period. Data on RTDs are very useful because they capture real decentralization by policy. RTDs transfer roles and services that are necessary for the full exercise of the competences that each AC adopted in their Statutes 4 The ACs where no coalition government has formed were omitted from the table.
of Autonomy. 5 These cross-time and cross-policy data had been codified by Chaqués and Palau (2011) but for four ACs only: Andalusia, Catalonia, Galicia, and the Basque Country.
We revised and completed the data collection procedure for the rest of ACs ourselves looking directly at the titles of all the RTDs approved in Spain since 1978.
Once data on cabinet portfolios and RTDs had been collected, the obvious challenge was to link the information of both. The final objective was to have a measure of how decentralized was the policy authority of each portfolio. For that purpose we started following the coding scheme of the Comparative Agendas Project (CAP), and more specifically the coding rules for the Spanish case. The CAP classifies events such as bills, parliamentary questions, or, in the Spanish case, RTDs, in a taxonomy of nineteen major topics and over 200 subtopics. 6 To classify cabinet portfolios in this coding scheme we took the nineteen more general categories, which in the case of Spain are macroeconomics, civil rights, health, agriculture, labor, education and culture, environment, energy, transportation, home affairs and justice, social policy, urban planning and housing, commerce and industrial policy, defense, science and technology, foreign trade, international affairs, government and public administration, and public lands and water management.
In our own portfolio classification, we ended up with seventeen policy responsibilities over which the cabinet portfolios of the forty-six coalition governments under study had had jurisdiction. 7 We also added three other cabinet responsibilities unrelated to specific policies: the Prime Minister, vice-premierships, and the spokesperson. The final categorization does not really refer to the number of portfolios in each cabinet. Rather, they are policy re-5 RTDs can transfer rights, obligations, goods, staff, financial resources, and documentation (frequently some of these at the same time). To give an idea of what is actually transferred we may look at a particular example. The RTD named Real Decreto 141/1997, de 31 de enero, sobre traspaso de funciones y servicios de la Administración del Estado a la Comunidad Autónoma de Andalucía en materia de provisión de medios personales al servicio de la Administración de Justicia transferred the revision and approval of the composition of groups of justices of the peace (and other related issues) from the central government to the Administration of Justice of Andalusia in 1997. This illustrates the fact that RTDs actually transfer policy responsibilities that the respective regional cabinet portfolios will assume.
6 For more information visit www.comparativeagendas.org. See also Baumgartner et al. (2006). 7 The CPA's nineteen except defense and international affairs, which are typically absent in the cabinets of the regional level. sponsibilities over which parties can exercise their authority through the control of cabinet portfolios. This difference between policy responsibilities and portfolios created some problems such as portfolios with more than one responsibility (e.g. education and science) or one policy responsibility related to more than one portfolio (e.g. home affairs and justice).
The procedure in the former example was to attribute the control of the policy responsibility 'education' and the policy responsibility 'science and technology' to the party obtaining the portfolio 'education and science'. For the latter example, if the portfolios 'home affairs' and 'justice' were assigned to different parties, the portfolio responsibility 'internal affairs and justice' was considered as chosen by both parties. 8 Figure 1 presents the number of competences by portfolio area that were transferred between 1980 and 2010 in Spain. That is, while the main independent variable in the empirical analyses is the cumulative number of competence transfers by portfolio, by region, and by year (see the description below), this figure simply presents a summary of the overall frequencies of competence transfers adding up the numbers of all the regions. It bears mentioning that these numbers do not reflect actual RTDs but the number of transfers by policy. That is, if a given RTD transfers competences both on policy X and policy Y, this RTD counts as two as it actually transfers two competences. The graph shows that the overall number of competences transferred are above the median in the policy areas health, agriculture, labour, education/culture, transportation, social welfare, urban planning and housing, and industry and commerce.
Given the type of data at hand, the methodology used here is the same as in Bäck et al. (2011). Their methodological approach sees the allocation of each post in a coalition cabinet as the unit of analysis and the different parties in government as the discrete choice alternatives. In other words, the observations are each portfolio and the choice variable is a 8 That causes a number of double (even triple) party choices of the same portfolio responsibility. The downside of this procedure is that these portfolios, in practice, fall out of the statistical analyses, while on the positive side we avoid to make ad hoc and potentially biased choices of which is the most important policy (portfolio) related to a portfolio (policy). For more details about the exact procedure contact the author.   dichotomous one that refers to the party having obtained that policy responsibility. 9 Therefore, a conditional logit model is applied to analyze this discrete choice between parties. 10 In this model, the probability that individual i (each cabinet portfolio in our case) chooses alternative j (where j = 1, 2, ..., J are the different coalition parties in our study) is defined by: With this method, the influence of different party characteristics (i.e. the choice-specific variables) can be easily evaluated, as well as the role of different features of the portfolio by interacting them with the choice-specific variables. More specifically, one of the partyspecific variables included in the empirical analyses is parties' strength. We expect that the more the strength of a party, the more likely it will be allocated a post. The two related measurements are the following: • Seat Share Contribution: This variable measures parties' size relative to the rest of coalition partners. It is calculated as the proportion of seats each party contributes to the cabinet resources.
• Bargaining Power Contribution: This variable measures parties' bargaining power contribution to the coalition, relative to the other members of the government. We have chosen the Banzhaf index, defined by the probability that a voter has to change an outcome of a vote where voting rights are not necessarily equally divided among the voters (parties, in our case). To calculate the voting power of each party, all winning coalitions must be listed and then critical parties must be counted (that is, a party who, if it changed its vote, would cause the measure to fail -50 percent being the quota for passage-) (Penrose 1946). A party's power is then measured as the fraction of all swing votes that it could cast. 11 The other party-specific variable identifies the type of party in terms of geographical scope and main ideological family. Its three values are 'zero' for conservative nation-wide parties (the Partido Popular and the Centro Democrático y Social ), 'one' for left-wing/socialdemocratic parties (the Partido Socialista Obrero Español, Izquierda Unida, and its respective regional branches), and 'two' for the myriad of Spanish NSWPs, irrespective of their left-right position. We expect there will be a relationship between the ideologic and territorial profile of a party and their portfolio choice and this is why we consider this variable.
It would certainly be preferable to complement that with a measure of the saliency of each party for each policy in order to test how parties' preference profiles affected portfolio choice (see Bäck et al. 2011) and, perhaps most interestingly, whether portfolio decentralization shaped the effect of policy saliency. Unfortunately, data on policy saliency are not readily available for all Spanish regional parties.
Regarding the portfolio-specific variables, we simply dichotomize the seventeen portfolio categories mentioned above to assess the role of parties' size and power on the choice of specific portfolios. This first step is aimed at identifying those portfolios whose choice is more strongly related to parties' strength and capture the relative attractiveness of each post.
Second, the main portfolio-specific variable refers to how decentralized is the policy area under the jurisdiction of the ministerial post. The extent to which the regional government 11 We made use of the so-called 'Strauss Calculator' (www.mindlessphilosopher.net/weights) to calculate the indices. All the power indices below are normalized between zero and one in a way that the sum of powers of all the parties for which we have data is always equal to one. The empirical results do not depend on whether we choose the Banzhaf or Shapley-Shubik indices.
has autonomous authority to decide the policies under the remit of a given portfolio has been measured through the number of transfers of competences contained in the RTDs of each region that had been made in the respective policy area until the year when the coalition formed. That is, the cumulative number of competences that were transferred to that specific region from 1980 to the formation year in each portfolio jurisdiction. This variable offers cross-time, cross-regional, and cross-policy variation and has been labeled as Portfolio Decentralization. 12 In the next section we present the related estimates of the conditional logit analyses and discuss the results and main findings.

Empirical Analysis
The statistical analyses first explore the determinants of obtaining specific portfolios (Table 2). To do that we interact the portfolio type dichotomous variables with the choicespecific variables (i.e. Party Strength, Left, and NSWP ). 13 The table simply summarizes the estimates of the conditional logit analysis (only statistically significant estimates are emphasized: pos. and neg. for positive and negative coefficient signs, respectively). What stands out the most is that bargaining power -not so much size-positively influences the obtention of two of the policy-unrelated posts. That is, an increase in the negotiation power of a party increases its probability of being able to choose the premiership and the spokesperson post (relative to the economy/finance portfolio, which is the reference category). Interestingly, though, we see that party size (column 1) equally affects the obtention of economy/finance and the PM post whereas, when such a position is created, the spokesperson tends to go to the largest party, even more clearly than the premiership. On the other hand, party size 12 Other measures of political decentralization such as the Regional Authority Index used in many recent studies (e.g. Hooghe et al. 2010;Hooghe and Marks 2013;Amat and Falcó-Gimeno 2013) do not offer the three sources of variation simultaneously. 13 In a conditional logit model the principal components of the case-specific interaction terms are not included because they do not vary across choices (here, the parties in the coalition) and, if included, would fall out of the analysis (see for instance Martin and Stevenson (2001) or Bäck et al. (2011)).
is negatively correlated with the allocation of the vice-premiership, what seems to indicate that this post typically goes to the second partner in the coalition.
It is perhaps surprising that the estimates related to most of the posts in column (1) are negative, which basically reflects the importance of the reference category (Economy/Finance) for larger parties, as well as health, transportation, home affairs/justice, and presidency (i.e. cabinet's office). Read together, columns (1) and (2) seem to suggest that the choices less influenced by party strength are those related to issues like civil-rights and minorities, environment, energy, social welfare, urban planning and housing, and foreign trade. Columns (3) and (4) of Table 2 show that the Spanish regional coalition governments seem to allocate the labor, urban planning and housing, and natural resources posts to left-wing nation-wide parties relative to conservative nation-wide parties, whereas the opposite is true for energy and science and technology posts. As for NSWPs, they seem to be specially interested in being assigned the responsibilities of agriculture, education and culture, transportation, urban planning and housing, and natural resources, and less in energy and science and technology, which seem to be the most preferred by conservative parties. Similarly, the likelihood of obtaining the spokesperson post is lower for NSWPs, what appears to reflect SWPs' concern about letting the communication of the coalition be handled by a NSWP.
The main deductive hypothesis of the article refers to the decentralization of the policies under the remit of coalition cabinet portfolios. Table 3 shows that, indeed, there is a positive relationship between competence devolution to regional authorities and the attractiveness of regional portfolios for the parties involved in a coalition formation process. Strength makes parties more likely to obtain any regional portfolio, but this effect is even stronger the greater the decentralization of the policy under the jurisdiction of the portfolio. The coefficients of the interaction are always positive, although, in the full sample, the effect is statistically significant for the bargaining power specification only. However, when the analysis is restricted to economically capable regions, 14 the effect of decentralization on 14 The GDP per capita of these regions, labeled as 'rich', have consistently been above the Spanish average portfolio choice is stronger -larger coefficients-and statistically significant both interacted with seat share contribution and bargaining power. Regarding the fast-track regions, 15 the effect is again more modest. The coefficients are higher than in the 'All ACs' sample, but still statistical significance is reached in one of the specifications only. Nonetheless, the influence of fast-track regions seem to arise indirectly in models (7) and (8)  and confidence intervals narrower in the subsamples. This is especially so in richer regions and in the more restricted fast-track-rich subsample. Decentralization, therefore, seems to make portfolios more attractive, as parties use their negotiation resources to be allocated more decentralized cabinet posts, yet the effect seems to be particularly important in those regions where parties can take more advantage of having greater responsibilities over policies.
Finally, according to the fourth hypothesis, the effect of decentralization on portfolio choice should be stronger for NSWPs as compared to SWPs. Table 3 shows that the more regional governments have authority over the policies under the jurisdiction of a portfolio, the more NSWPs appear to go for these portfolios. The estimations of the interactions NSWP X Port. Decentr. in columns (1) and (2)    ideas put forward in the article: NSWPs are thought to intrinsically care about the policies under the remit of regional portfolios, whereas SWPs are said to value access to regional cabinet posts for instrumental reasons only. As a result, the effect of greater policy-making autonomy in a regional portfolio is more important for those who care more about regional policy-making per se, namely NSWPs. Interestingly, though, the effect is less clear in fasttrack regions. This piece of evidence seems to suggest that decentralization is important enough for all parties in these specific ACs, while in the rest it is NSWPs who more rapidly respond to the new incentives created by the devolution of competences.

Conclusion
This article represents an effort to place political decentralization at the center of the academic debate over the formation of governments in parliamentary democracies. Needless to say, the decentralization of fiscal and political authority has been a very productive topic for political scientists but, to our knowledge, it had not yet been considered as a determinant for how parties in coalition governments negotiate over ministerial posts. The devolution of competences to regional entities has several political consequences among which there is the fact that (certain) policies can be managed more autonomously by regional cabinets.
That in principle should qualitatively increase the attractiveness of reaching office at the regional level. Yet there is great variation regarding 1) the policies that are decentralized, 2) the timing of this decentralization, and 3) the regions in which decentralization takes place within a country. Four related hypotheses have been put under empirical scrutiny. First, the decentralization of a policy should make powerful parties more likely to obtain the related portfolios. Second, this effect should be stronger in economically capable regions and, third, in regions that started with greater levels of decentralization. And fourth, decentralization should have a greater impact on those parties that are in principle interested intrinsically in regional policy-making, namely NSWPs.
The evidence provided for the Spanish regional coalition governments has revealed the importance of decentralization for the qualitative dimension of portfolio allocation. The statistical analyses have shown that the decentralization of a given policy boosts the effect of party strength on the choice of the related portfolio, but particularly so in economically stronger regions and where the initial level of decentralization was higher. This suggests that, indeed, the devolution of competences to subnational governments makes regional cabinet portfolios more attractive as parties seem to use their power to take charge of 'decentralized posts' in those regions where they may find it financially easier to advance their policy goals through portfolio control and claim political credit for them. In addition, NSWPs seem to have been more attentive to obtain those portfolios whose policy remit is more decentralized than their partners.
The analyses of this article could be extended to other countries where the empirical study of coalition portfolio allocation is more established. As said, decentralization has been largely overlooked as a potentially important variable in these processes. It might be therefore worthwhile to comparatively explore its influence on the distribution of ministerial posts in coalition cabinets as well as on coalition formation more generally.
Also, the issues treated in this article have implications for federalism scholarship that certainly merit more attention than devoted in these pages. Which are the consequences of both federalism and decentralization has been one of the perennial questions in this field.
Apart from other economic and political effects, this article has shown that decentralization can also shape the incentives parties face when it comes the time to form a government and distribute executive responsibilities. A related question that might be worth exploring would be whether or not decentralization and federalism not only affect regional government formation processes as shown in this article, but also the national ones. The flip side of the argument presented here implies that competence devolution to regional entities empties the political responsibilities of the central government, which should shape the motivations of political parties to reach office, especially if they are policy-seeking. That in turn opens various questions that deserve further exploration about how political parties are adapting their strategies to these new contexts. In any case, we hope this article represents a first step toward considering the importance of political decentralization on government formation processes.