Authoritarianism, democracy and de/centralization in federations: what connections?

ABSTRACT What is the impact of democracy/authoritarianism regime change on de/centralization in federations? Based on the annual coding of three politico-institutional aspects, 22 policy fields, and five fiscal categories, this article maps de/centralization in Argentina, Brazil, Mexico, Nigeria and Pakistan from the establishment of their respective federal orders to 2020. It shows that de/centralization varies greatly across its different dimensions as well as between systems, with centralization being the dominant long-term trend but with significant exceptions, notably Pakistan. Regime change plays a major role in de/centralization but not always in line with the usual expectation that authoritarian regimes centralize and democratic ones decentralize. Other factors that cut across the authoritarianism/democracy divide, notably ideological orientations, have substantial impacts on de/centralization. By investigating long-run patterns of de/centralization in federations that have experienced democracy/authoritarianism regime change, the article sheds light on how federalism operates beyond consolidated democracies.


Introduction
Federations vary along several lines, but as Bowman and Krause (2003, 302) argue, the 'vertical distribution of power is of fundamental importance to the study of federalism'.For a long time, though, we lacked detailed measures of such distribution (i.e. of de/centralization). 1 While Dardanelli et al. (2019a;2019b) addressed this lack in relation to continuously democratic federations, systems that experienced regime change between democracy and authoritarianism and vice versa had been understudied from this perspective.This special issue explores de/centralization in Argentina, Brazil, Mexico, Nigeria and Pakistan since the establishment of their respective federal orders and seeks to account for the findings.The five cases were selected on the basis of (a) being long-established (i.e.having been a federation for more than 50 years) and (b) having experienced regime change between democracy and authoritarianism and vice versa.By examining detailed patterns of decentralization in these federations, we address the long-standing question of the relationship between regime type and federalism.Does federalism require democracy?Do authoritarian regimes centralize while democratic regimes decentralize?
Bringing together the findings of the other articles in this special issue, we show here that the impact of regime change varies greatly across both dimensions of federalism and systems.In the remainder of this article, we outline our theorization of the drivers of de/centralization (section 2), describe the conceptual and methodological framework employed to measure de/centralization (section 3), report the results of our measurement across cases (section 4), assess them against our theoretical expectations (section 5), and compare them to the patterns observed among continuously democratic federations (section 6).Section 7 briefly introduces the other articles in this special issue and section 8 concludes our discussion.
Theorizing de/centralization in federations across regime change centralization at the outset) and factors shaping changes in the distribution over time (i.e.dynamic de/centralization).
Two main conditions shape static de/centralization at the outset: the historical period in which a federation came into being and whether the federation was born of a 'coming together' or a 'holding together' process.Given that the scope of government grew significantly between the 'nightwatchman state' of the nineteenth century and welfare states of the second half of the twentieth century, we can expect older federationsin our sample, Mexico, Argentina and Brazilto have been less centralized at birth than those established after World War II, such as Nigeria and Pakistan.So-called 'coming together' federations (Stepan 1999, 22-23), born out of a 'federal bargain' among previously independent or quasi-independent units (Riker 1964, 12-16), should also have started from a lower level of centralization than 'holding together' federations, which evolved from unitary political systems.In our sample, Argentina falls into the coming together category whereas Brazil, Nigeria and Pakistan match more closely the holding together type.Mexico is not easily categorized in these terms but is closer to the holding together type (Olmeda 2023, 4).Given the degree of interaction between the two factors, we thus expect Argentina to have been the most decentralized at birth, Mexico and Brazil to display a medium level of centralization, and Nigeria and Pakistan to have been the most centralized.Federations that were less centralized at the outset can then be expected to have undergone dynamic centralization, particularly in their formative stage, as part of a state building process in which key planks of the new political system such as a federal judiciary or unified armed forces are established.
Shifting to factors shaping dynamic de/centralization, we can group them into different domains and different stages of the causal process.Given the all-encompassing nature of a regime, we discuss regime change as one of these factors but also how it can be expected to interact with the other factors.
At the more distal stage of the process, structural factors of a socio-economic and socio-cultural nature are important.In the socio-economic domain, the broad process of modernizationunder which are subsumed aspects such as technological change, increased mobility, and market integrationhas been argued theoretically (e.g.Beer 1973) and shown empirically (Dardanelli et al. 2019b, 12) to have fostered policy centralization in most federations.We anticipate a similar effect in the federations examined here, operating under both democracy and authoritarianism.After World War II, globalization might have further contributed to centralization, given the scope for the central government to encroach upon the autonomy of the constituent units through international agreements (e.g.Lazar, Telford, and Watts 2003, 4) but may also have spurred greater involvement of the constituent units in international relations, thus producing a degree of decentralization in the field of external affairs.As democracies are more exposed to globalization than authoritarian regimes (Gygli et al. 2019) we can expect the effects of globalization to be higher under the former.
In the socio-cultural domain, in a coming-together federation we generally expect citizens' primary identification with the federation to rise and their primary identification with the constituent units to decline over time (e.g.Riker 1964, 103-110).Other things being equal, rising identification with the federation likely facilitated centralization in the long run under both democracy and authoritarianism.The evolution of citizens' expectations about the role of government, especially rising demands for uniform welfare services countrywide, is also likely to have fuelled centralization in all federations (e.g.Birch 1955), most clearly under democracy but also under authoritarianism.Ethnic fractionalization, especially if distributed along territorial lines and reflected in the design of the federation, is likely to have restrained centralization as territorially concentrated ethnic groups can be assumed to want to maximize the autonomy of their constituent units (McGarry and O'Leary 2009).This can also be expected to be most evident in a democratic setting but not altogether absent under authoritarianism.We thus expect a general increase in centralization to have been fuelled by socio-cultural factors but to a lesser extent in more ethnically diverse federations such as Nigeria and Pakistan compared to Argentina, Brazil and Mexico.
These relatively slow-moving trends are likely to have been deepened or amplified by short-term shocks such as wars and economic crises (e.g.Wheare 1946, 254).We expect these trends and shocks to have led to changes in attitudes toward the vertical distribution of powers in a federation principally among the general public, organized interests, and the media, but also at the elite levelbroadly favouring power accretion at the centre under both democracy and authoritarianism.Centralizing steps should thus be more likely in periods of security or economic crises.
Pressures toward centralization, however, will have been mediated at the more proximate stage of the causal process by political and institutional variables that reinforced or weakened them.Chief among them is the nature of the political regime.In light of the traditional association between authoritarianism and centralization, we generally expect centralizing steps to be more likely to occur under authoritarian than democratic rule, but the degree and form of centralization observed is likely to vary depending on the nature of the authoritarian regime.Cheibub, Gandhi, and Vreeland (2010) distinguish between civilian and military regimes, while Geddes, Wright, and Frantz (2014) subdivide the civilian category into party and personalist regimes.Svolik (2012, 26-39), however, argues that these typologies have several limitations.He proposes a classification along four variables: involvement of the military, 2 restrictions on parties, legislative selection, and executive selection.The latter three variables can arguably be seen as sub-dimensions of what we can call power concentration, so that we can speak of regimes with higher/lower degrees of involvement of the military in politics and of power concentration.Going beyond their former categorical classification, Geddes, Wright, and Frantz (2018, 61-94, 190-201) argue that personalization of rule, which they treat as a form of power concentration, varies within both civilian and military regimes, as well as sometimes within the same regime over time, and has far-reaching consequences on the nature of the regime.
In light of the above, two key variables define the nature of the regime: (1) involvement of the military and (2) power concentration.Given that military regimes are less likely than civilian regimes to have elected institutions at the centre (Gandhi 2008, 93-94), and that power concentration can be expected to have an effect vis-à-vis the constituent units too, we hypothesize that a high level of either of these variables be associated with centralization.Regimes that score highly on bothsuch as 1983-98 Nigeriashould thus be the most likely to engage in centralization, especially in its politico-institutional form.Regimes with no military involvement and low power concentration, such as Mexico, however, may be more likely to retain elected institutions but engage in manipulation of elections.Other things being equal, regimes with high military involvement and/or high power concentration are also more likely to engage in policy and/or fiscal centralization.
Beyond the nature of the regime, de/centralization trends are likely to be shaped by ideological and other institutional variables.For ideology, we focus on the economic dimension of the left/right distinction (Herre 2023) which has also been referred to as developmentalism versus neo-liberalism.Parties of the left are generally seen to favour centralization while parties of the right resist it (e.g.Bowman and Krause 2003, 310;Döring and Schnellenbach 2011, 92-94).Likewise, decentralization has often been associated with neo-liberalism (Eaton 2004, 11;Falleti 2010, 6-8;Gibson 2004, 10;Montero and Samuels 2004, 13-15).Hence, centralizing steps are more likely to occur under left-leaning or developmental rulers, both democratic and authoritarian, and federations having experienced longer periods of left-wing rule should display higher centralization, particularly in the policy and fiscal spheres.
As regards other institutional factors, a large number of constituent units has been considered as facilitating centralization in a democracy (e.g.Simeon 1972, 38-39;Watts 2008, 71-72) and could be expected to have a similar effect under authoritarianism too.On this basis, other things being equal, Argentina, Brazil and Mexico should have experienced greater centralization than Pakistan, while Nigeria should have experienced progressively greater centralization as the number of its constituent units increased over time.

Conceptualizing and measuring de/centralization in federations
Our conceptualization draws on the framework developed by Dardanelli et al. (2019a, 7-13), 3 which, as mentioned above, rests on the basic distinction between static and dynamic de/centralization. 4Given the presence of regime change in the cases analysed in this special issue, and its likely impact beyond the policy and fiscal dimensions, we have conceptualized and measured a third dimension of de/centralization: politico-institutional autonomy. 5 Politico-institutional autonomy is the degree to which constituent units have discretion to decide their own constitutional and institutional set up, free from central-government interference.We divide it into three subdimensions.The first, labelled constitutional autonomy, is the degree to which a constituent unit has discretion over its own constitutional set-up.The second, or institutional autonomy, is whether the institution(s) governing a constituent unit are elected by the citizens of the unit, and hence can be deemed to reflect their preferences, or are appointed by the central government, and can be assumed to reflect the latter's preferences.The third subdimension, which we call electoral manipulation, concerns the degree to which electoral processes in the constituent units are subject to manipulation by the central government so that their outcomes reflect the preferences of the central government more than the preferences of the electors in the unit in question.
In the policy dimension, we distinguish between legislative and administrative autonomy.Legislative autonomy relates to a constituent unit's control of primary legislative powers in a policy field.Administrative autonomy concerns the degree to which a constituent unit implements central government, as well as its own, legislation.We measured legislative and administrative autonomy in 22 policy fields: agriculture; citizenship and immigration; culture; currency and money supply; defense; economic activity; pretertiary education; tertiary education; elections and voting; employment relations; environmental protection; external affairs; finance and securities; health care; language; civil law; criminal law; law enforcement; media; natural resources; social welfare; and transport. 6 In the fiscal dimension, we measured autonomy in five categories.The first is the degree to which a constituent unit directly controls its own revenues, defined as the proportion of own-source revenues out of its total revenues.The greater the proportion of own-source revenues, the more fiscally autonomous a constituent unit is (e.g.Watts 2008, 104).The second sub-dimension relates to the restrictions a constituent unit faces in raising own-source revenues.These include restrictions applied to a source of revenue otherwise controlled by a constituent unit and outright exclusion from a particular revenue source.The third sub-dimension is the degree to which fiscal transfers from the central government to a constituent unit come with strings attached (Oates 1972, 65;Watts 2008, 106-108;Blöchliger 2013, 25).This can be defined as the proportion of conditional grants out of total revenues.The less dependent on conditional grants a constituent unit is, the higher its fiscal autonomy.The fourth sub-dimension concerns the scope and stringency of the conditions attached to the central government's grants.Wideranging or highly stringent conditions constrain more than limited or loose ones.Hence, the more limited their scope and/or the lower their stringency, the more autonomous a constituent unit is in allocating the funds it receives from the central government.The fifth sub-dimension relates to the degree of freedom a constituent unit has in raising revenue through borrowing.
Codes were assigned by the authors of the articles in this special issue and checked with relevant experts to ensure accuracy.Codes were reviewed and adjusted in several meetings held between all the authors to ensure crosscountry reliability.Coding aimed to reflect the de facto situation at the time, rather than merely the de jure order.
Tables A1 and A2 in the online supplemental file outline the scales we employed to measure these dimensions and categories of autonomy.The online supplemental files attached to the other articles in this special issue detail the codes assigned across all categories (politico-institutional, policy, and fiscal), indicate the sources of the codes, and outline the justification for each coding decision.
As regards dynamic de/centralization, drawing on Dardanelli et al. (2019a, 10-13), we are interested in the following properties: direction, magnitude, tempo, form and instruments.Direction refers to whether change occurs in a centralizing or decentralizing direction.Changes shifting our measures from a higher to a lower value signal a reduction in the autonomy of the constituent units and thus constitute centralization.Changes entailing a shift from a lower to a higher value indicate an increase in constituent-unit autonomy and therefore denote decentralization.Magnitude is the size of the change that occurred over a given time period.Tempo refers to the temporal patterns in which change occurs.It can be further subdivided into frequency, pace, timing and sequence.Frequency is the number of instances through which change occurs.Pace is a combination of frequency and magnitude as change can occur in many small steps or a few large ones.Timing relates to when change occurs in the lifespan of a federation.Sequence is the temporal order in which change occurs, such as for instance fiscal centralization preceding policy centralization or vice versa.Form refers to the dimension in which change occurs, for example fiscal as opposed to policy decentralization.Lastly, instruments refer to the 'tools' through which change occurs, such as constitutional amendments, legislation, and court rulings.
Drawing on the literature on authoritarianism and hybrid regimes (Geddes, Wright, and Frantz 2018;Kropp 2019;Schedler 2002) we have identified three additional instruments: abuse of emergency powers, disregard of constitutional provisions, and electoral manipulation.In federations granting the federal government the power to intervene in the constituent units during alleged emergencies, such as in Argentina and Pakistan, the abuse of such power can be a significant centralizing instrument.As regards constitutional provisions, authoritarian rulers can be expected to take actions that contravene the federal constitution while formally maintaining it in force. 7Such actions may include disregarding the rules governing the constituent units' participation in amending the federal constitution so that constitutional change is imposed on them.This is thus an instrument through which dynamic centralization occurs through de facto constitutional change.If the institutional and policy autonomy of the constituent units is formally maintained, central rulers are still likely to try to exercise control through the manipulation of constituent unit elections, hence our project sought to measure its effects.

Patterns of de/centralization
Static de/centralization at the outset Static de/centralization at the outset varied considerably across the five federations.Here we need to bear in mind that the outset date differed greatly across the sample, from 1824, 1862 and 1891, respectively, for Mexico, Argentina and Brazil to 1954 and 1956, respectively, for Nigeria and Pakistan. 8In the politico-institutional sphere, there was a sharp contrast between Argentina and Mexico, with high constitutional and institutional autonomy, and Brazil, Nigeria and Pakistan, with low autonomy on both dimensions. 9The degree of manipulation of constituent unit elections paints a different picture: manipulation was significant in Argentina and Pakistan but absent in the other three cases (Figure 1 and Table A3 in the online supplemental file).
There was less variation in the policy sphere.On a scale ranging from 1 = exclusively federal to 7 = exclusively constituent units, Argentina, the most decentralized, had a mean of 4.79 in the legislative autonomy of its constituent units and 5.65 in administrative autonomy, against 2.47 and 3.16, respectively, for Brazil, the most centralized (Figure 2 and Tables A4 and A5 in the online supplemental file).
In the fiscal realm, there was a sharp contrast between Argentina and Brazil, whose constituent units had high autonomy on all available indicators, and Nigeria and, especially, Pakistan, with much lower levels of autonomy.Mexico stood in between, with a high proportion of own-sources revenues and few restrictions on the latter, but only a medium level of borrowing autonomy (Figure 3 and Table A6 in the online supplemental).

Frequency of change
The frequency of change presents a highly contrasted pattern.Constituent units' autonomy to decide their own constitutional set up was highly stable.By contrast, institutional autonomy changed relatively frequently, albeit with a significant difference between Nigeria and Pakistan, with a higher rate of change, and the three Latin American federations, with a lower rate (Figures 4 and 5).
In the legislative dimension of the policy sphere, there was a stark contrast between Pakistan and the other cases whereas in administration there was  much less variation.In the fiscal sphere, change was most frequent in Pakistan and least frequent in Brazil; and most frequent in the proportion of own-source revenues and least frequent in the degree of conditionality (Table A16 in the online supplemental file).

Direction and magnitude of change
The cumulative direction of change varied greatly across dimensions and federations.In the politico-institutional sphere, Brazil and Nigeria had a mix of  centralization and decentralization, while Pakistan experienced decentralization in two out of three categories, and Argentina in one; Mexico recorded no net change (Figures 4 and 5 and Table A7 in the online supplemental file). 10In the policy sphere, centralization was preponderant.In terms of legislation, all five federations became more centralized over time although the magnitude ranged from −2.38 (mean across policy areas) in Argentina to −0.36 in Pakistan (Figure 6 and Table A8 in the online supplemental file).
We find largely similar patterns in policy administration.All federations became more centralized over time, with the highest magnitude in Argentina and lowest in Pakistan (Figure 7 and Table A9 in the online supplemental file).
Centralization was also generally dominant in the fiscal sphere, especially in Argentina and Mexico.Pakistan is to an extent an outlier; it became more centralized in terms of own-source revenues but less so as regards restrictions on own-source revenues and borrowing autonomy (Figure 8 and Table A10 in the online supplemental file).

Pace, timing, and sequence of change
The pace of change also differed substantially between dimensions and federations.In the politico-institutional sphere, there was a sharp contrast between constitutional and institutional autonomy.While constitutional autonomy was only altered infrequently and to a modest extent, changes in institutional autonomy were mostly deep and sudden, typically from 7 to 1 and vice-versa (Figures 4 and 5 above).This is because they tended to coincide with regime change, when elected institutions in the constituent units were abolished or restored.The pace of change in the policy realm was generally much more gradual.Policy legislation in Pakistan was the main exception where the imposition/lifting of martial law brought about relatively sharp discontinuities.Policy administration, however, followed the same gradual pattern in Pakistan as in the other federations, as provincial bureaucracies remained in operation (Adeney and Boni 2022, 15).Legislation in the field of elections and voting was the other main exception, where regime change brought about major shifts in most federations.
The timing of change was specific to each federation.While change in individual fields was mostly gradual and distributed along the entire life span in all cases, we can nonetheless observe clear critical junctures in aggregate terms at certain points in time.These tended to coincide with crucial turning points in each federation's historical evolution, such as the Mexican revolution in the 1910s, the onset of the Estado Novo in Brazil in the 1930s, and the rise and fall of Juan Perón in Argentina in the 1940s and 1950s, as well as the various timings of democratization in the late twentieth and early twenty-first centuries in all of them.Dynamic de/centralization does not seem to have proceeded in clearly discernible sequences.As noted above, the evolution of institutional autonomy and, to a lesser extent, electoral manipulation followed a highly volatile pattern linked to regime change and often disconnected from the trajectory of policy and fiscal de/centralization.Regarding the latter two, patterns appear to be highly contextual to each case.In Argentina policy (both legislative and administrative) centralization preceded fiscal centralization in the early period but in the second half of the twentieth century fiscal centralization deepened while policies became less centralized.In Brazil, fiscal centralization preceded centralization in policy legislation but, from about 1940 onwards, it largely mirrored the trajectory of policy legislation and administration.In Mexico, fiscal and policy centralization proceeded largely in lockstep between 1870 and 1970, but in the late twentieth century trajectories diverged: fiscal centralization deepened while policies underwent mild decentralization.In Nigeria and Pakistan, policy and fiscal de/centralization appear to have followed largely the same trend.

Instruments of change
There was much less diversity across cases as to the instruments of dynamic de/ centralization.Two are clearly dominant: constitutional change and the enacting or repealing of legislation by either the central government or the constituent units.All five federations experienced extensive constitutional change, including the adoption of several entirely new constitutions as well as amendments to existing charters in between, with far-reaching consequences for the autonomy of the constituent units.With the exception of Argentina, this was the most important instrument of de/centralization.The enacting/repealing of legislation was a close second in terms of importanceand first in Argentina and was closely connected to constitutional change in several respects.In a straightforward way, constitutional provisions often required implementing laws and decrees in order to take effect.In a more problematic way, from the perspective of a distinction between constitutional provisions and ordinary legislation, legislation was often enacted, particularly during periods of authoritarian rule, with a dubious constitutional basis, for example in Brazil, or even, as in Nigeria, in a wholesale bypassing of the (suspended) constitution.
The frequency of constitutional change made the judicial interpretation of constitutional provisions a less prominent instrument of change though in some casesparticularly in Argentina, Nigeria and Pakistancourt rulings also played a significant role.The use of emergency powers or martial law took centre stage in authoritarian periods in Nigeria and Pakistan but less so in the Latin American cases.In Argentina, however, the politicized use of federal interventionsa form of emergency powerswas prominent for a long period and caused sharp asymmetries between the provinces targeted by intervention and the others.Given the different nature of electoral manipulation, changes in this category took place through distinct instruments such as banning of candidates or parties, controlling the media, exercising various forms of coercion, and so forth. 11

Static de/centralization in 2020
Static de/centralization across our cases presented a contrasted picture in 2020.If in some dimensions the five federations had grown more alike, in others significant variation was still present.This is particularly so in the politico-institutional sphere.While the institutional autonomy of the constituent units was at, or near, the maximum level in all cases, major differences persisted in constitutional autonomy and electoral manipulation.Argentina and Mexico cluster together at one end with high constitutional autonomy and low manipulation while Nigeria and Pakistan are at the opposite end with low constitutional autonomy and still significant manipulation.Brazil occupies an intermediate position, with low constitutional autonomy but the absence of manipulation (Figure 9 and Table A11 in the online supplemental file).
Less variation can be observed in the policy sphere, compared with the politico-institutional indicators as well as the situation at the outset.Pakistan was the most decentralized in policy terms, both legislatively and administratively, whereas Brazil's and Mexico's states had, respectively, the lowest legislative and administrative autonomy (Figure 10 and Tables A12 and A13 in the online supplemental file).
Fiscal autonomy was generally higher than legislative policy autonomy but varied significantly across federations as regards the proportion of ownsource revenues and in borrowing autonomy.Restrictions on own-source revenues and the proportion of conditional grants were the most similar across cases (Figure 11 and Table A14 in the online supplemental file).

What explains these patterns?
Here we seek to account for the patterns described above, in light of the theoretical framework presented in section 2. Regime change played a major role but not always in line with the theoretical expectations.We discuss the other hypotheses first before turning to regime change.

Static de/centralization at the outset
Patterns of static de/centralization at the outset were broadly aligned with the theoretical expectations about the role of the historical period in which each federation was born and about the 'coming/holding together' formation process, although this applies more to politico-institutional and fiscal de/centralization than to policy de/centralization.Argentina, the second oldest federation in our sample and the product of a federal bargain, was generally the most decentralized.The only exception was its level of electoral manipulation, which was higher than Mexico's.Nigeria and Pakistan, both newer and holding-together federations, were the most centralized in politico-institutional and fiscal terms.Patterns in the policy dimension were less clear and lend more support to the coming/holding together hypothesis than to the 'age' hypothesis.Argentina was significantly more decentralized than the older Mexico, while Nigeria and Pakistan were not much more centralized than Mexico and actually less centralized than Brazil, a much older federation.In line with expectations, given their lower centralization at the outset, was that Argentina and Mexico experienced the largest magnitude of dynamic policy and fiscal centralization.

Socio-economic trends
Although we lack comparable measures across countries and over time that would allow us to test it directly, the hypothesis that modernization would lead to wide-ranging policy centralization is consistent with our findings.All federations became more centralized in policy termsboth legislatively and administrativelyas well as to a large extent fiscally too.Those federations, such as Argentina and Mexico, where modernization had the greatest scope to produce an effectbecause they came into being earlier and were more decentralized at the outsetwere also those that underwent the deepest centralization.Given their shorter life spans, Nigeria and Pakistan are weaker tests for this hypothesis, but do not contradict it.The partial exception of Pakistan as regards fiscal de/centralization appears to be related to democratization (Adeney 2012) rather than to modernization having had a different effect.The role played by modernization in the policy and fiscal arenas is indirectly confirmed by the fact that the trajectory of politico-institutional autonomy, where we did not expect modernization to have a bearing, was very different.
Results yielded no support for the hypothesis linking globalization and de/ centralization.All five federations became less centralized in policy terms between 1970 and 2020 despite being considerably exposed to globalization. 12Fiscally, the federation that experienced the greatest centralization, Argentina, also had the lowest increase in its exposure to globalization, whereas Brazil, which had the second largest increase in exposure, became less centralized.Globalization does appear to have had its hypothesized effect, however, in the field of external relations in Argentina, Brazil and Mexico, where the constituent units acquired greater legislative and/or administrative autonomy after 1970.

Socio-cultural trends
Socio-cultural trends seem to have operated broadly in line with expectations.Although we do not have relevant data, it is likely that in the older federations -Mexico, Argentina and Brazilcitizens' identification with the federation grew over time.Together with their mono-national character and high cultural homogeneity, 13 this is likely to have facilitated deep policy centralization in Argentina and Mexico and further centralization, from an already high starting level, in Brazil.In Pakistan, by contrast, strong regional identities played a significant role, notably via the agency of political parties, in the decentralization brought about by the 18th amendment to the constitution (Adeney and Boni 2022, 19).Hence, Pakistan experienced less centralization overall than Argentina or Mexico and was the least centralized federation in our sample in 2020.The hypothesis is less well supported, however, in the case of Nigeria, which experienced considerable centralization despite its multi-ethnicsome would say even multi-national (Suberu 2022, 19) character.Patterns of identification are likely to have interacted with other factors; we discuss some of these interactions below.

Economic and security shocks
Economic and security shocks played an important role in de/centralization dynamics in all cases, albeit not always in the hypothesized direction.On one hand, the major economic crisis triggered by the 1929 Wall Street crash spurred centralization in Argentina and Brazil, particularly in the fiscal sphere.On the other hand, economic difficulties facilitated neo-liberal administrative decentralization in Argentina and Mexico in the late twentieth century, while in Brazil, over the same period, the overcoming of hyperinflation (i.e. a major improvement in economic conditions) paved the way for fiscal centralization (Schlegel 2022, 27). 14Severe economic and security shocks also caused at times temporary decentralization by leading to a breakdown of the established order.Hence, for instance, during the Mexican revolution the federal government lost control of the armed forces to a plethora of state-based militias whereas the 2001 corralito crisis in Argentina witnessed the virtual collapse of the country's monetary order and the issuing of parallel currencies by the provinces.
Security shocks were particularly prominent in Nigeria and Pakistan, where they operated broadly in line with the theoretical expectations.In both countries, the wars fought to prevent the secession of Biafra (successfully) and Bangladesh (unsuccessfully) created a context facilitating policy and fiscal centralization.In Pakistan, moreover, the latent conflict with India arguably helped sustain static centralization for a long time, not least by strengthening the role of the military (Suberu 2022, 21;Adeney and Boni 2022, 21-22).In all five federations, economic and security shocks often led to regime change so they indirectly paved the way for the major impact these regime changes had on de/centralization, as we discuss below.

Collective attitudes
Collective attitudes operated, as expected, as 'transmission belts' between broad trends at the societal level, on one side, and political action on the other.Given the weaknessor absenceof democracy for long periods, consequential attitudes in these federations have primarily been those of the elites and/or interest groups.We have less evidence regarding citizens' attitudes as a whole and the extent to which these played a role in de/centralization steps (e.g. in the 'neo-liberal decentralization' in Latin America in the latter part of the twentieth century).Likewise, we lack evidence regarding the expectation that demands for uniform welfare services would fuel centralization where the provision of such services was not already rather centralized at the outset.In general, as remarked above, it is possible that their mono-national character facilitated dynamic policy centralization in Argentina and Mexico, or helped sustain the original high level of centralization in Brazil.

Ideology
The hypothesized role of ideology appears to be broadly confirmed.Economic interventionism and developmentalism were associated with centralization in many cases, especially in Brazil, Mexico and Nigeria, while neoliberalism heavily coloured the policy of, primarily administrative, decentralization in the Latin American federations, as already noted.

Number of constituent units
There is some support for the hypothesis that federations with a smaller number of constituent units would centralize less in that Pakistan, with the fewest units among our sample, also experienced the lowest magnitude of dynamic policy centralization and had the lowest level of static policy centralization in 2020.In Nigeria, however, the steepest centralization occurred in the early phase of military rule, when the number of constituent units was still comparatively modest (12 and later 19), and has not grown in parallel with their subsequent proliferation to 36 states (Suberu 2022, 10).The similar number of units in Argentina, Brazil and Mexico does not help us account for the different magnitude of dynamic policy centralization they experienced, although their static scores for 2020 are not dramatically apart.

Regime change
We now turn to the factor that has been central to this project.Before delving into our analysis, we need to acknowledge that assessing the impact of regime change on de/centralization dynamics faces significant challenges.First, there is disagreement among specialists as to how regimes should be classified: many are classified as democratic by some authors but authoritarian by others. 15For our purposes, we focus on the regimes consistently classified as authoritarian in the five main datasets in the field. 16Second, classification schemes vary in their coverage.Only one (Boix, Miller, and Rosato 2013), for instance, covers the nineteenth century and only one (Coppedge et al. 2022) extends to 2020.Therefore, for the purpose of assessing the impact of regime change, we focus on the post-1945 period, which is covered by all five datasets until at least 2008.Third, authoritarian regimes vary greatly in their duration, from one year, 1962 in Argentina, to over 50 years, from 1945 (though actually 1917) to 1995 (or 1999) in Mexico, with consequently major differences in terms of the time regimes have had to leave their mark.Fourth, there is limited diversity among regimes in our sample in terms of type, the military one being dominant (albeit split between Svolik's [2012] categories of corporate and personal), although there is more variation in terms of power concentration. 17Fifth, centralization being a cumulative process within a finite dimension, prior patterns condition to an extent subsequent ones.If, for instance, there is high dynamic centralization in a policy field in period 1, there is less scope for further centralization in period 2. Applying the above criteria, and disregarding 1962 in Argentina, we focus on 10 authoritarian regimes across the five federations (Table A15 in the online supplemental file) and assess the impact they had on de/centralization by comparing the year before the onset of the regime and the last year of the regime, plus in some cases intermediate timepoints.
The general expectation that regime change is a key factor shaping de/ centralization dynamics is confirmed.Based on the magnitude of change associated with it, regime change is the single most important factor in this respect.Its impact, however, varies greatly across dimensions, regimes, federations, and over time and is far from the simplistic associations authoritarianism = centralization and democracy = decentralization. 18Hence, we should avoid considering authoritarian federations as ipso facto façade federations: federalism can remain alive even under authoritarianism, thus confirming the findings of previous research pointing in that direction.The first observation to make is that authoritarianism had a much deeper impact on politico-institutional autonomy than on policy and, even less, fiscal autonomy.Most authoritarian regimes disbanded elected institutions in the constituent units, though some, in Brazil and Pakistan, suppressed them at first but restored them later while in Mexico elected institutions remained operational throughout.Where elected institutions were maintained, central manipulation typically increased, though, again, not invariably: manipulation, for instance, slowly declined in Mexico between 1945 and 1995.The impact of regime change on policy autonomy was much more subdued, especially as regards administration.On the legislative side, Pakistan is an exception, where the imposition of martial law by authoritarian rulers created sharp discontinuities around regime transitions.Even there, though, such discontinuities are deeper if we consider the initial impact of the authoritarian transition and much less pronounced if we look at the entire lifespan of the regime, as both the 1977-87 regime and the 1999-2007 one decentralized in their final phase.Likewise, in Brazil, the 1964-84 regime had a more centralizing effect in its initial phase and a much shallower one across the entire period.In Mexico, as already noted, policy legislation became less centralized whereas Argentina bucked the trend even more clearly.The Argentine regimes of 1966-72 and 1976-82 had a neutral or mildly decentralizing effect on policy legislation while the 1955-57 regime was clearly decentralizing.The case that most closely matches the theoretical expectation is the 1966-78 regime in Nigeria, which had a deeply centralizing effect on policy legislation and administration.The second Nigerian military regime, from 1983-98, though, had a more limited centralizing effect.Outside Pakistan, the impact on policy administration generally mirrored that on legislation, with the notable exception of the 1976-82 regime in Argentina, which decentralized significantly more in administration than legislation.Regime change had much less impact in the fiscal sphere, most cases recording no change in the key F1 category (proportion of own-source revenues).The two main exceptions were 1957-71 Pakistan, where F1 deepened by two points on our scale and Mexico, which witnessed a decline of five points between 1945 and 1995.In the latter case, though, given the regime's long duration, it is difficult to disentangle the effect of authoritarianism from that of other factors, also in the light of the fact that democratization after 1995 did not trigger fiscal decentralization.
How does this compare to our hypotheses regarding the impact of regime change?The general expectation of centralization being more likely under authoritarian regimes and decentralization under democratic regimes is only partly supported.As we have seen, many authoritarian regimes engaged in various forms of decentralization while plenty of centralization occurred under democratic rule.There is mixed evidence on the effect of regime type.The fact that only one authoritarian regime in our sample was civilian limits our ability to assess the hypothesis that military regimes are likely to centralize more than civilian ones.Moreover, there is high variation among the military regimes in terms of the magnitude of centralization that occurred under their watch and Svolik's (2012) distinction between personal and corporate forms of military rule does not seem to account for it.Power concentration, as measured by Svolik (2012), appears to be a more significant factor, particularly as regards politico-institutional de/centralization, but only up to a point.On one hand, the regimes that centralized the most -1966-78 Nigeria and 1958-71 Pakistan as well as the initial phases of 1977-87 and 1999-2007 Pakistanwere also those displaying, apart from 1999-2007 Pakistan, the highest degree of power concentration whereas the regime with the lowest degree of power concentration, Mexico, was the least centralizing in the politico-institutional sphere and the second most decentralizing in the policy sphere.Brazil's regime, which had the second lowest degree of power concentration was also among those that centralized the least across the three dimensions.On the other hand, Mexico did centralize deeply in the fiscal sphere, while 1955-57 Argentina, with a fairly high degree of power concentration, centralized deeply in politico-institutional terms but was the most decentralizing in policy terms and 1966-72 Argentina, with the second highest degree of power concentration, experienced virtually no policy centralization.Power concentration, as measured by Geddes, Wright, and Frantz's (2018) personalization, does not appear to account for it either.The most decentralizing authoritarian regime, 1955-57 Argentina, had a higher personalization score than 1966-78 Nigeria, one of the most centralizing, while variation in personalization over the course of the same regime is not associated with corresponding shifts in de/centralization (Table A15 in the online supplemental file).
The hypothesis regarding the role of ideology finds greater support.The decentralizing effect of the 1976-82 regime in Argentina had a clear neoliberal flavour and so did the corresponding phase in Mexico, as Falleti (2010) noted.While classifying it as neo-liberal would be avant la lettre, 1955-57 Argentina could be described as a laissez-faire regime engaging in decentralization to restore the status quo ante after the deeply interventionist, and centralizing, Peronist rule (Moscovich and Lacroix Eussler 2023, 15). 19 By contrast, the military regimes in Nigeria, the first one especially, as well as the early phase of the 1964-84 Brazilian regime displayed a developmental orientation, in which centralizing control over economic activity and related policy fields featured prominently.Table A16 in the online supplemental file summarizes our assessment of these hypotheses.

Comparison with continuously democratic federations
How do the experiences of Argentina, Brazil, Mexico, Nigeria and Pakistan compare to those of Australia, Canada, Germany, India, Switzerland and the United States 20 analysed by Dardanelli et al. (2019b)?The first point is that the trajectory of de/centralization in the politico-institutional sphere, which has proven to be the most sensitive to regime change, appears to be the starkest difference between the two sets of federations.While divergence in terms of constitutional autonomy and electoral manipulation remained significant in 2020, as regards institutional autonomy, patterns have converged considerably over time towards the classic model of elected legislatures and executives.Differences are much less stark, though still noticeable, in the policy sphere.The centralization experienced by Argentina and Mexico, for instance, broadly mirrors that of the United States and Australia.The federations among those analysed here that have experienced less dynamic centralization, such as Brazil, started from a much higher level of static centralization.A significant difference is that the regime-change federations have witnessed more clearly defined cycles of centralization and decentralization, in contrast to the largely linear centralization trajectory that characterized most of the continuously democratic cases.Variation in static de/centralization in 2020 across all 11 cases was only marginally higher than among the continuously democratic alone.Regime-change federations were generally more centralized but Pakistan is an exception, as it appears to have been less centralized than all continuously democratic casesincluding, intriguingly, Indiabar Canada in terms of legislation.Legislative-administrative deviation was generally lower than the high levels of Switzerland and Germany but broadly in line with the other continuously democratic cases.On the fiscal side, regime-change federations have generally experienced more dynamic centralization than their continuously democratic counterparts and were more centralized in 2020, although Brazil appears to be an exception. 21Regime-change federations, with the partial exception of Mexico, made less use of conditional grants compared to the United States or Australia.As regards the instruments of centralization, the most prominent difference is the importance of constitutional change, which was much more prominent in the cases analysed here compared to most continuously democratic federations.Lastly, and leaving aside, of course, regime change, the drivers of de/centralization dynamics seem to have operated mostly in the same direction as observed for the continuously democratic cases.The most significant difference is arguably the effect of secession threats.Whereas Quebec's secession threat has been seen as an important determinant of the lack of dynamic centralization in Canada, the secession threats posed by Biafra and Bangladesh appear to have fuelled centralization in Nigeria and Pakistan.This is perhaps because in the latter two cases, unlike in Canada, the threat triggered violent conflict, hence it also became a security shock.

This special issue
In their article on Argentina, Moscovich and Lacroix Eussler (2023) find an overall centralization process in the policy and, especially, the fiscal sphere whereas provincial autonomy actually increased over time, mainly as a result of the decline in the politicized use of federal interventions.This cumulative process masks, however, distinct cycles of centralization and decentralization, particularly in policy legislation and administration.In contrast to expectations, as regards policies, authoritarian periods were often associated with decentralization.The conditions most clearly associated with centralization seem to have been presidential strength combined with economic expansion.
Schlegel ( 2022) also finds very different patterns in Brazil depending on the dimension of de/centralization one considers.While the dynamics of politico-institutional autonomy were closely associated with regime change, policy autonomy followed a distinct centripetal trajectory occasionally punctuated by decentralization episodes.Fiscal autonomy, by contrast, bucked the trend and, with the exception of borrowing autonomy, remained broadly stable or increased over time.He finds ideological orientations to have been an important factor in shaping de/centralization dynamics.The embrace of developmentalism helped sustain or deepen centralization for a long period while movements in the opposite direction were facilitated by the belief that democracy goes hand in hand with decentralization.
Regime change as an important determinant of de/centralization dynamics is confirmed by Olmeda's (2023) analysis of Mexico.While the country, in common with most other cases examined in this special issue, underwent an overall process of policy and, especially, fiscal centralization, the timing and direction of change were heavily influenced by regime change.Centralization was prevalent during the long years of both the Porfirista regime in the nineteenth century and the PRI regime in the twentieth whereas democratization in the late twentieth century coincided with a partial reversal of centralization.The ideological orientation of the ruling elites, though, also emerges as a major contributing factor.Developmentalism provided an ideological underpinning to centralization whereas decentralization was heavily coloured by neo-liberalism.Suberu (2022) shows that the prevailing view of Nigeria having undergone a deep centralization process under the military governments that ruled it from 1966-98 needs to be nuanced.Changes by the military to Nigeria's territorial, fiscal, and constitutional institutions undoubtedly contributed significantly to centralization.But Suberu argues that centripetal forces were inherent in the holding-together character of the federation and the country's ethnic fragmentation, and were reinforced by a developmental ideology common to both democratic and authoritarian rulers.The country's economic dependence on natural resources and the fact that these were always under central control further exacerbated such forces.
In a similar vein, Adeney and Boni (2022) find an overall high level of stability of de/centralization in Pakistan, particularly in the policy and fiscal spheres.The strong influence of the military and the underlying tensions with India are the common threads running through the federation's history.They do detect, however, a significant role for regime change and, more subtly, for variation within authoritarianism.Direct military rule typically resulted in the abolition of elected provincial institutions and the imposition of martial law whereas military presidents often engaged in (mild) decentralization.The more significant decentralization engendered by the 18th amendment to the constitution coincided with a weakening of military influence and steps towards fuller democratization (although weakening as of 2023).

Conclusion
Federalism has long been associated with democracy; yet many federations have been governed by non-democratic rulers for extended periods of time.If we are to advance our understanding of how federalism operates, we need to cast our research net beyond the continuously democratic federations.Here we took a step in that direction by measuring and analysing de/ centralization in Argentina, Brazil, Mexico, Nigeria and Pakistan since the establishment of their respective federal orders, focusing in particular on the impact of regime change from democracy to authoritarianism and vice versa.This article has outlined how we approached the task in conceptual, methodological and theoretical terms and has sought to account for the patterns emerging from the data.The central finding is that regime change is a crucial determinant of de/centralization dynamics but its effects are much more complex than the assumptions that authoritarianism = centralization and democracy = decentralization.While this broadly confirms that federalism tends to survive under authoritarianism, it underscores the need to pay attention to variation across dimensions and fields.Only by being sensitive to these nuances can we advance our understanding of the relationship between regime type and federalism.We hope the analytical framework and the data presented in this special issue open promising avenues for further research across the universe of federal systems.Future research could explore more in depth the determinants of some of the patterns we have uncovered, in particular those that suggest different dynamics in democratic and authoritarian settings such as the role of ethnic fractionalization and secession threats and the legacy of the coming/holding together birth of a federation for its subsequent trajectory.

Notes
1.For a brief review of the literature, see Dardanelli et al. (2019a, 2-5).2. Based on four categories: no involvement (labelled 'civilian'), indirect, and direct, the latter subdivided into corporate and personal.3. The framework was developed specifically to capture the dynamics of de/centralization in federations, which typically occur in the policy sphere.These are not as effectively captured by the Regional Authority Index (Hooghe et al. 2016), which is heavily weighted toward institutional aspects and is also limited to the post-1950 period.4. For the sake of comparability across federations, we focus on the relationship between the central government and the constituent units, leaving aside the local level, which in some cases, such as Brazil and Nigeria, is considered a third tier of the federation. 5.The constituent units' autonomy to decide their own constitutional and institutional set up was not measured by Dardanelli et al. (2019a, 23) on the ground that it is unlikely to experience much change over time in continuously democratic federations.6.These policy fields were selected based on Riker (1964, 49-84), Watts (2008, 194-198) and the UN (2015) and OECD (2015,(194)(195) classifications of the functions of government.7. On authoritarian governments, military ones especially, having lower levels of constitutional compliance, see Gutmann, Metelska-Szaniawska, and Voigt (2023, 18).8. Pakistan became independent in 1947 but adopted its first indigenous constitution only in 1956.9.In Argentina, though, high formal institutional autonomy coexisted with widespread use of federal interventions (Moscovich and Lacroix Eussler 2023, 7).Brazil's low institutional autonomy was due to the fact that, in the initial period after the transition to a federal order, governors and assemblies were not elected.With the elections that took place in 1892, the states acquired full institutional autonomy (Schlegel 2022, 8).10.The decentralization experienced by Brazil in institutional autonomy is an artefact of the delay in establishing directly elected state institutions (see note 9).If we compare 1892 (instead of 1891) and 2020, there was no change in institutional autonomy.11.See the other articles in this special issue for details of each case.12.As measured by the growth in the KOF overall index between 1970 and 2019 (Dreher 2006;Gygli et al. 2019).13.On Argentina, Brazil and Mexico being culturally homogenous across constituent units, see Schwartz (2009, 139) and Minkov and Hofstede (2012, 149-150).

Figure 2 .
Figure 2. Static policy de/centralization at the outset.Note: mean across 22 policy areas.

Figure 5 .
Figure5.Dynamic 'real' institutional de/centralization.Note: to give a synoptic view of the real degree of institutional autonomy, we have multiplied the institutional autonomy score by the manipulation score and divided the product by 7.