Experiencing and supporting institutional regionalization in Belgium: a normative and interpretive policy feedback perspective

Regionalization has been a defining feature of European politics since the 1970s. Previous work has studied political drivers of the movements of competences to the subnational level, including the role of citizens’ preferences. Yet, we still know little about how these new divisions of competences between government levels have impacted the development of public opinion about this division. The article builds on the literature on policy feedback and argues that institutional regionalization may both directly and indirectly affect support for regionalization through normative and interpretive effects. To empirically qualify these expectations, the article uses eight cross sections of the Flemish and Walloon populations in Belgium (1991–2019). This approach explains differences in support for regionalization between citizens that were socialized in different institutional and regional contexts. The analyses show that Walloons who came of age in the context of more institutional regionalization tend to be more supportive of regionalization. In Flanders, in contrast, support for regionalization is most consistently and substantially explained by regional and Belgian identification. However, our analyses show no support for the expectation that coming of age in a more regionalized Belgium is associated with a greater sense of regional identification.


Introduction
In the past decades, territorial rescaling has featured prominently in European polities both in the form of regionalization (Keating 2013;Hooghe et al. 2016) and European integration (Bartolini 2005;Ferrera 2005). Expansive discussions have investigated the drivers of such changes, as well as how territorial restructuring has in turn, over time, reshaped European polities, including policy-making and partypolitical dynamics (Alonso 2012;De Winter and Tursan 2003;Erk 2003a, b;Keating 2013;Swenden and Maddens 2009).
When considering regionalization processes, scholarship on individual attitudinal changes associated with regionalization has started to emerge (Henderson et al. 2014;Dupuy and Van Ingelgom 2014;Dupuy et al. 2021;Cole et al. 2018). A key question is how, once established, the new division of competences between government levels may impact public support for (or opposition to) this division-which in turn may or may not lead to calls for further regionalization. Over time, we could indeed expect to observe a cyclical process where attitudes about regionalization and institutional regionalization affect one another. In this contribution, we study how experiencing regionalization may impact support for certain divisions of competences between the central and the regional level as an indicator of support for and opposition to regionalization. This article thus contributes to the literature that asks why citizens support regionalization.
So far, it has been shown that support for regionalization is related to both regional identification and the perception that one's region is economically better off than other regions within the same state (Henderson et al. 2014;Verhaegen et al. 2018). Moreover, a few studies have suggested that regionalization can have a socializing effect on citizens' attitudes regarding regionalization, but that this effect is moderated by elite discourses politicizing regionalization and putting it in a positive or negative light (Dupuy and Van Ingelgom 2014;Dupuy et al. 2021). These studies provide fruitful ground to account for citizens' support for regionalization. However, our understanding remains limited when it comes to the mechanisms at play. In particular, what is missing is a theoretical and empirical account of how institutional regionalization, socialization and identification are interlinked, and affect support for or opposition to regionalization. This article tackles this limitation.
We show that the literature on policy feedback offers analytical traction to study how experienced institutional regionalization may have impacted citizens' support for it (Campbell 2012;Kumlin and Stadelmann-Steffen 2014). From the policy feedback literature, we can derive two explanations for how institutional regionalization may shape citizens' support for regionalization: through a normative and an interpretive effect 1 (Mettler and Welch 2004;Pierson 1993). They, respectively, pertain 1 Note that the literature on policy feedback points to a third effect: the resource effect. Resource effects stress that institutions allocate resources, to the benefit or expense of different groups of citizens (Pierson, 1993). When institutional change occurs, citizens' position may improve, deteriorate or remain the same. It is expected that policy change will foster support among those who perceive that they benefit from a change in resource allocation, and opposition among citizens who think they lose. The underlying rationale, utilitarian calculation, that lies at the heart of this step has been extensively studied in the case of the EU, but also to some extent in the case of regionalization (e.g. Gabel and Palmer 1995;Henderson et al. 2014;Hooghe and Marks 2005;Verhaegen et al. 2018). Similarly, research on the attitudinal to the fact that because institutions and policies bring about new norms regarding what governments ought to do and look like, and convey meaning and worldviews related to one's position in the political community, they may 'create a new politics' (Schattschneider 1935: 288) to which citizens may adapt their preferences about government. We therefore draw from the scholarship on policy feedback that mostly focuses on single policies, to study the influence of a broader institutional change, composed of a wide range of policies. Following other research (Svallfors 2010;Mettler 2016;Béland and Schlager 2019), we thereby acknowledge that individuals' attitudes are likely to be affected not only by specific policies, but also by the wider institutional environment they are embedded in.
Specifically, this article argues that the experience of living in a regionalized system of government may have an impact on citizens' support for regionalization. The increased salience of the subnational level as the result of institutional regionalization and the related regional policy-making makes it plausible that institutional changes have feedback effects, both directly-through a normative effect-and indirectly-through an interpretive effect. Directly, regional institutions-in the form of formal institutions, policies, or any rules and practices-may exert a socializing effect on citizens' preferences in favour of these institutions through a mechanism of adaptive preferences (Campbell 2012;Svallfors 2010). Feedback effects are in this context explained by a normative mechanism as 'public policies provide citizens with a sense (…) [of] the desirable state of affairs' and what governments ought to look like (Svallfors 2010, p. 120). In line with the literature on political socialization, it is expected that such socializing effects are particularly strong at an early age and that attitudes formed in this period will persist throughout life (Abdelzadeh and Lundberg 2017;Campbell 2006;Jennings and Stoker 2004;Rekker 2018). It is thus expected that citizens will see the division of competences between the regional and central level as they experienced it when they came of age as more 'natural' or evident, and therefore more desirable (Dupuy et al. 2021).
Indirectly, the increased salience of the subnational community is expected to increase the likelihood that citizens will identify as part of a regional community alongside their national community. Particularly when individuals came of age within the context of higher levels of regionalization, the regional community is expected to be seen as more relevant, which would result in a stronger regional identification relative to one's national identification (Oakes 1987;Tajfel 1981). Social identity theory, as frequently studied in the context of EU integration, forms the legacy of authoritarian regimes showed that winners of such regimes-individuals who benefited from socioeconomic redistribution and gained some form of political rights-remain more supportive of the regime, and are less prone to support democracy in the post-autocracy period than the losers (Neundorf et al. 2019). However, this theoretical argument is very difficult to translate when it comes to institutional regionalization in terms of the division of competences as gains or losses are hard to assess empirically. Moreover, the resource effect is policy-specific and therefore can only be studied when the unit of analysis is specific policies rather than institutional change, as this article addresses. Finally, our data do not allow for an exact test of the resource policy feedback effect as the survey questions did not ask respondents whether they think that their financial situation improvement or deterioration is due to regionalization.
Footnote 1 (continued) basis of the interpretive effect (Gabel and Palmer 1995;Hooghe and Marks 2005;Verhaegen et al. 2018). In turn, stronger regional identification is expected to feed back into support for regionalization as citizens prefer a correspondence between the community they feel part of, and the level at which decision-making takes place (Henderson et al. 2014;Verhaegen et al. 2018). In other words, this contribution theorizes how the interpretive effect may introduce indirect effects of institutional regionalization-through identification-on attitudes about regionalization, in addition to the direct normative policy feedback effect.
To empirically qualify these expectations about normative and interpretive policy feedback effects, we study Belgium as a crucial case where the variables under scrutiny are magnified (Lijphart 1971). Belgium underwent a major transformation from a unitary to a federal state (Deschouwer 2012;Dodeigne et al. 2015). This process started with the first state reform in 1970, where the role of linguistic communities and regions was formalized. Incrementally, this further evolved into the constitution of Belgium as a federal state in 1993. Since then, two more state reforms took place that increased the competences of the sub-national levels. As such, the Belgian case provides a setting in which gradual institutional regionalization 2 occurred, allowing for a study of different cohorts that came of age in the context of various intensities of institutional regionalization.
These institutional reforms have had a real impact on people's lives, as the point of reference for policy in the areas of employment, education, culture, housing, social benefit schemes, health care, public transport and so on increasingly has, in a stepwise manner, become people's region or community. This has impacted the mere exposure to references to these groups and, thereby, their relevance to individuals. In particular, the structuring of political parties, public television and radio along community and regional lines illuminates the split of the public debate in the Belgian society. However, the salience and the framing of the institutional regionalization have been radically different in both regions (Sinardet and Morsink 2011;Dupuy et al. 2021). In Flanders, the salience has been high and the framing has relied both on identity and socio-economic discourses. In Wallonia, the salience has been very low. This allows for studying the effects of institutional regionalization in two contexts with the same institutional reforms, but a different salience of the issue.
In addition to these traits, the Belgian case is unique as data are available on citizens' attitudes about the division of competences between the central and subnational levels for a substantial period of this regionalization process (1991-2019).
2 In this contribution, we write about 'regionalization' when referring to the devolution of competences from the central level to the subnational level. In Belgium, the central level is the 'federal level' since the establishment of Belgium as a federal state in 1993. The subnational level consists of both regions and communities, which partly overlap and partly have cross-cutting boundaries. There are three regions: Flanders (58% of the population), Wallonia (32% of the population) and the Brussels-Capital Region (10% of the population), responsible for territorially linked issues such as economics, employment and environment. The three communities are defined by the three main language groups: the Dutch-, Frenchand German-speaking communities. These govern in areas that are linked to individuals such as education, healthcare and justice. As Belgium moved from a unitary to a federal state, this process is commonly referred to as 'federalization' (rather than 'regionalization') in the Belgian public debate.
Cross-sectional data from eight election surveys offer the opportunity to describe the evolution of support for regionalization among representative samples of the Flemish and Walloon populations. Given the absence of a long-term panel study, we cannot follow the trajectories of individuals. Yet, the repeated cross sections do allow for tracing trajectories across cohorts in these populations, while allowing us to study individual-level correlations between experience with institutional regionalization, identification, and support for regionalization at different points in time (Grasso et al. 2017;Svallfors 2010;Tiberj 2017).
The analyses show evidence for direct policy feedback effects through early age socialization processes in the Walloon samples. In Flanders, in contrast, support for regionalization is most consistently and substantially explained by regional and Belgian identification. Thus, this case study suggests that the process of institutional regionalization may lead to support for regionalization, also in regions where regionalization has not been very salient or where regional identification is low (i.c. in Wallonia). It also demonstrates the importance of identification in explaining support for regionalization. However, our analyses show no support for the hypothesis that coming of age in a more regionalized Belgium is associated with a greater sense of regional identity. As such, this is an important result as the experienced intensity of institutional regionalization leads to support for regionalization, but not to a stronger regional identification. Explanations for the development of regional identification should thus be sought elsewhere.
In sum, this study contributes to the literature in three ways. First, theoretically, it further specifies the underlying mechanisms of processes of policy feedback by linking the literature on territorial politics to the literature on political socialization and identity development. Second, it extends empirical work on the effect of institutional reform on attitudes about institutional design by distinguishing between direct (normative) and indirect (interpretive) policy feedback effects. Third, the article sheds new light on the continuing process of regionalization in Belgium and the diverging explanations for support for regionalization in Flanders and Wallonia.
In the remainder of the article, we first elaborate on the theoretical underpinnings of our argument. Second, the choice for Belgium as a crucial case and the data selection are justified. Third, we present the empirical analyses. The final section discusses the implications of the results for the literature on policy feedback, for scholarship on regionalization in general, and Belgium in specific.

Theorizing feedback effects of experiences of institutional regionalization on citizens' support for regionalization
In the most recent and comprehensive study of citizens' attitudes towards regionalization, Henderson et al. (2014) show that identification and utilitarian considerations play a key role in explaining citizens' attitudes about the division of competences. Similarly, in studies on attitudes about the movement of competences towards the EU-level, identification and utilitarian considerations are observed to play a role as well (e.g. Gabel and Palmer 1995;Hooghe and Marks 2005;Verhaegen et al. 2018). In their comparative study of citizenship after the nation state, Henderson et al. show that 'a greater sense of regional identity is positively associated with support for regional policy control ' (2014, 169; see also Guinjoan and Rodon 2014). However, they also report that this relationship is not straightforward. They observed that only when average regional identification in a region is high, individuals' regional identification is significantly related to support for further regionalization. When it is low, there is no significant relationship between regional identification and support for regionalization. 3 Also, dual (regional and national) and exclusive regional identification result in support for different types of regionalization. Individuals with dual identification may support further regionalization, but not in the form of a secession, which individuals with exclusive regional identities are more likely to favour (Serrano 2013;Guibernau 2006).
In the same study, Henderson et al. hypothesize that sociotropic economic evaluations could drive support for regionalization (2014). In particular, they argue that citizens in richer regions are more supportive of regionalization than citizens in poorer regions. The latter tend to favour more authority for the central state. Based on their comparative data, the relationship they report is weak, yet consistent with the hypothesis.
In line with other research (Svallfors 2010;Neundorf et al. 2019;Dupuy et al. 2021), the article argues that a substantial part of the explanation of the variation in citizens' support for regionalization lies elsewhere: in their experiences of regionalization. Research on policy feedback offers a stimulating theoretical framework to understand how citizens' experiences of regionalization could impact their support for it. This literature intends to explain how institutions and policies contribute to shaping citizens' political attitudes and behaviour, and under what conditions such effects may occur (for a review see Campbell 2012; Larsen 2019). The core idea is that institutions and policies are not only outputs of political systems; they also impact ('feed back into') mass publics and can thus be considered as inputs too. Policy feedback studies thereby highlight how institutions and policies contribute to shaping the context through which individuals experience government on a daily basis, and in which they develop their political views and behaviour, more so than through sporadic visits to the voting booth (Rothstein 1998).
It is important to note that most studies focus on a single policy and investigate how specific political behaviours (e.g. voting, demonstrating, or membership to political groups) and attitudes (e.g. interest in politics, sense of self-efficacy) relate to this policy. For example, Andrea Campbell's research shows that social security in the United States has turned American seniors into 'über citizens', mostly due to the politically relevant resources allocated by the policy-time and money for the most part (2003). Yet, as recently highlighted, individuals benefit and loose from various policies and are embedded in a context composed of a wide set of policies and institutions. This combination, rather than just a single policy, shapes their political attitudes and behaviours. Suzanne Mettler (2016) coined the term 'policyscape' that captures this feature. Consequently, some studies have adopted a broader perspective to expand the scope of policy feedback effects under consideration from a single policy to a larger set. Stefan Svallfors (2010) makes a compelling case for this approach in his study of different cohorts of West and East Germans' attitudes towards redistribution after the fall of the Berlin wall. His analysis considers that distinct notions of redistribution and relationships between the state and the market in both parts of the country have shaped Germans' attitudes about redistribution, rather than looking at specific policies and their design. Our article follows in these tracks and considers that institutional reforms have resulted in regional institutions and policies that distinguish themselves from national-level institutions and policies. We are interested in how this shift in the institutional setup of Belgium may have impacted citizens' attitudes toward regionalization.
Specifically, two distinct types of feedback effects are relevant for our study of the effects of institutional regionalization: the normative and the interpretive effect. 1 In this section, we set out how both mechanisms taken together pave the way to a comprehensive theoretical framework to account for how experiences of institutional regionalization impact citizens' support for regionalization. To this end, we link the policy feedback literature to the literature on political socialization and identification.
First, normative policy feedback effects affect citizens' attitudes through processes of adaptive preferences. It is argued that citizens will grow to prefer the policies and institutional setups they are used to (e.g. Campbell 2012;Mettler 2005;Pierson 1993;Soss and Schram 2007). This process ties in closely with what is referred to as political socialization in the literature on the development of political attitudes. According to theories on political socialization, the context in which individuals come of age is crucial in the development of political and societal attitudes (Flanagan 2013). Particularly early and mid-adolescence are observed to be periods in which political attitudes and orientations are open to change, after which they become more stable (Hooghe and Wilkenfeld 2008;Hatemi et al. 2009;Dinas 2013;Abdelzadeh and Lundberg 2017;Flanagan 2004). In this contribution, we study political socialization as the process of acquisition of prevailing norms about the organization of government across levels (Greenstein 1970;Niemi and Hepburn 1995). Linking policy feedback and political socialization approaches, normative policy feedback effects are expected to particularly impact younger citizens who are in their 'impressionable years'. We thus expect that respondents who came of age in a context of more institutional regionalization are more supportive of regionalization (H1). This relationship qualifies as a direct policy feedback effect as it is expected that exposure to a regionalized institutional setup in itself will raise support for regionalization through the increased salience of the subnational level as a locus of policy-making and habit formation.
Second, we argue that political-territorial identification may mediate this relationship, giving rise to indirect interpretive policy feedback effects on support for regionalization. The interpretive effect points out that institutions convey meaning and worldviews related to one's position in the political community (Pierson 1993;Mettler 2005). In the case of institutional regionalization, this entails the increased relevance and salience of the regional community and its perceived significance as a marker in one's political and social position, at the expense of the national community. Research in the field of territorial politics already mentioned this relationship between the institutionalization of regions and the cultivation of regional identification through regional symbols, institutions and policies (Paasi 1986(Paasi , 2003. This perceived societal position and group memberships plays a key role in individuals' social identity development (Brewer 2001;Hammack 2015;Oakes 1987;Tajfel 1981). It is thus expected that identification with the subnational community strengthens when institutional regionalization is more intense (Hechter 2000;Rabushka and Shepsle 1972). Identification with the national community may adapt too, as a consequence of the weakened position of the central government level. As a result, citizens' primary political-territorial identification may shift from national to regional; citizens may even shift from a dual to an exclusive regional identification. The impact of regionalization on identification is, however, expected to be stronger when regionalization takes place at a young age. Similar to other political attitudes, it has been observed in the context of the development of European identification that people who came of age when their country was part of the EU tend to identify more strongly with the EU than people who were already deep into adulthood (Ceka and Sojka 2016).
The second part of the interpretive feedback effect builds on the proposition that citizens prefer congruence between the community they feel part of, and the community by and for which governing takes place (Beetham and Lord 1998;Sigalas 2010). It echoes existing research on the relationships between regional identification and support for institutional regionalization presented above (Henderson et al. 2014). Hence, a positive relationship is expected between regional identification and support for regionalization, bearing in mind Henderson et al.'s qualification (2014) that this relationship is particularly expected when average regional identification in a region is high. In sum, we hypothesize (H2) that citizens who came of age in the context of more institutional regionalization would be more likely to primarily identify with this community (H2a), and as a result be more supportive of moving more competences to the government level that corresponds to that community (H2b). The inverse hypothesis can be made regarding national identification.
In sum, we propose that, along the identity-based and utilitarian explanations, experiences with institutional regionalization affect citizens' support for regionalization directly through changes in public norms, and indirectly through the impact on their identification (see Fig. 1).

Data
The analyses draw on the data of eight consecutive post-electoral surveys conducted in Belgium in 1991, 1995, 1999, 2007(ISPO-PIOP Belgian Federal Election Studies), 2009. More information on the datasets is in Appendix A of ESM. These voter surveys provide insightful information regarding citizens' support for the regionalization of competences. After each election, a representative sample of Flemish and Walloon citizens of voting age (18 years and older) was interviewed about their attitudes about regionalization, political-territorial identification, a broad range of views on political and societal topics, and socio-demographics. As the surveys were carried out in the context of elections, the political debate in this period has been documented, which allows for a contextualized interpretation of the analyses. Moreover, the availability of data at different points in time when regional institutionalization progressed, in both regions, and with respondents from different age cohorts, offers a unique opportunity to analyse the socializing impact of institutional change, i.e. institutional regionalization, on support for regionalization.

Measures
Support for regionalization is measured in the 1991-2007 Belgian election surveys by asking respondents about their preferences about the division of competences between the regional/community levels on the one hand, and the federal level on the other. Respondents were asked the following question: 'The form of government that the country should have is still a matter of discussion. Some think that "Flanders and Wallonia must each be able to decide over everything by themselves." Others think that "Belgium, Flemish and Walloons together, must be able to decide about everything." Where would you place yourself on the scale?' On a scale of 0 to 10, respondents were asked to indicate the level of government in Belgium that should be in charge of decision-making. In our analyses, 0 reflects the opinion that 'Belgium, Flemish and Walloons together, must be able to decide about everything'; 10 refers to the opinion that 'Flanders and Wallonia must each be able to decide over everything by themselves.' In the 2009-2019 surveys, an adapted question wording was used: 'There is currently a lot of debate regarding the proper distribution of powers between the federal and regional levels of government. Some people think that more powers should go to the regions and communities. Other people think that more powers should go to the federal state. Where would you place your opinion on a scale ranging from 0 to 10, where 0 means that Regions and Communities should have all the competences,

H1 -Normative effect (direct)
Experienced intensity institutional regionalization when coming of age Regional identification Support for regionalization The value 5 means that you agree with the current situation.' The latter question asks about preferences relative to the status quo, instead of the desired distribution of competences, as the middle category was labelled as such (even though the extremes of the scale do refer to absolute levels of competences for each level). Because of this adaptation, separate analyses are carried out for each time period. In our analyses, the scale is reversed so that a higher score reflects higher support for regionalization.
The construction of our measure for experienced intensity of institutional regionalization when coming of age builds on the study of Svallfors (2010). In this study, attitudes about state intervention were studied among different cohorts in Eastern and Western Germany. A cohort is a group of individuals that has been exposed to a shared socialization process, within the same political historical context (Svallfors 2010;Yang and Land 2008). Hence, we distinguish between Flemish and Walloon cohorts, defined by the context of institutional regionalization in which they came of age (see Table 1). These cohorts serve as a measure for experienced intensity of institutional regionalization when coming of age. The measure is based on the birth year of respondents, and the timing of the six Belgian state reforms. Following the literature on political socialization, which argues that the most impressionable years are before adulthood, we make cohorts based on the year respondents turned 18. The first cohort includes all respondents that turned 18 before the first state reform in 1970. These respondents were socialized up to adulthood in unitary Belgium and experienced the change in the institutional setup of their country since 1970 after their most impressionable years. The second cohort includes all respondent that were 18 years or older in 1980, when the second state reform took place, but were not yet 18 years old in 1970. This cohort was thus born between 1953 and 1962. Part of their youth political socialization process-during adolescence-took place in a context of institutional regionalization. The third cohort groups respondents born between 1963 and 1970, the fourth cohort groups respondents born between 1971 and 1975, the fifth cohort includes respondents born between 1976 and 1983 (this is the first cohort coming of age when Belgium was a federal state), and the sixth cohort includes all respondents that were born after 1983. Figure 2 presents the mean support for regionalization per region and per cohort. While the explanatory analyses will use all cohorts as displayed in Table 1, Fig. 2 uses a simplified categorization of respondents in cohorts for reasons of presentation. The figure presents a comparison between respondents whose youth socialization took place before the process of regionalization started (1970), and all more recent cohorts. Two main patterns can be observed. First, support for regionalization is significantly higher in Flanders than in Wallonia in six 4 out of eight survey years. Second, we only observe significant differences between the oldest cohort which consists of respondents who reached adulthood before the regionalization process Table 1 Cohorts indicating experienced intensity of institutional regionalization when coming of age a i.e. institutional context in which respondents of a cohort came of age, i.e. in which they turned 18 b i.e. birth years of respondents who were 18 years and older at time of state reform, but were not 18 years old yet at the time of the previous state reform c In some survey years, the most recent cohorts were not yet 18 years old, so not included in the data collection. In the analyses with data from 2009 to 2019, we distinguish between 6 cohorts, in the other analyses there are 5 cohorts.  1962Between 1963and 1970Between 1971and 1975Between 1976and 19831984 Surveys including this cohort c started and younger respondents in Wallonia in 1995 and2007. 5 Elsewhere, we provide an elaborate discussion of these patterns from 1991 to 2007, interpreting them in the context of elite discourses regarding regionalization (Dupuy et al. 2021). In all datasets, political-territorial identification was measured by asking respondents about their primary group of identification. In the analyses, the dummy variable 'regional identification' indicates all respondents who primarily identify with their region or linguistic community, 'Belgian identification' indicates all respondents who primarily identify with Belgium, 'Other' indicates respondents who primarily identify with their province, municipality, Europe, 6 or other entity. As support Responses ranged on a scale from 0 to 10 where a higher score indicates that a respondent prefers more regionalization. Note that the survey question changed in 2009. Results are weighted for age and gender (1991,1995,2003,2007,2019), age, gender and region (1999), age, gender and province (2009,2014). See Appendix B of ESM for the number of respondents per cohort per survey year for regionalization is operationalized with a measure asking respondents about their preferred division of competences between the regional and national level, 'Belgian identification' will be used as reference category in the analyses, so that the results show differences between respondents who primarily identify with their region as compared to respondents who primarily identify as Belgian. Descriptive statistics of all used variables are included in Appendix C of ESM. In addition to the dependent and explanatory variables of interest, the analyses control for the most established alternative explanations and covariates for attitudes about the division of competences between government levels: economic utilitarian calculation, political awareness, and socio-demographics (Henderson et al. 2014;Verhaegen et al. 2018). Including these alternative explanations allows us to draw conclusions about the relationship between the cohorts respondents are part of and their political-territorial identification on the one hand, and their support for regionalization on the other, independent of potential covariation with respondents' socioeconomic position, political awareness and socio-demographics (Svallfors 2010).

Methods
To test our hypotheses, OLS regression models that estimate the relationship between experience with regionalization when coming of age as operationalized by cohorts, and political-territorial identification on the one hand, and support for regionalization on the other, are estimated following the logic of Svallfors (2010). These models allow for testing direct policy feedback effects, while controlling for identification, socio-economic position, political awareness and socio-demographics. Yet, to fully test hypothesis 2 that proposes an indirect policy feedback effect from experience with regionalization through regional identification on support for regionalization, we also need path models which we estimate as the structural component of structural equation modelling (Acock 2013). While these models are sometimes referred to as 'causal models', we do not claim to test causality in this article. Given the absence of a randomized intervention and a control of the level of the independent variables, and in the absence of panel data where respondents' identification and attitudes are measured at different points in time, we cannot rule out alternative explanations or reversed causality. Instead, the set-up of our models is grounded in a set of theoretical expectations about the relationship between identification and attitudes, we include relevant covariates as controls in the models as to prevent omitted variable bias, and the measure for experienced intensity of institutional regionalization is exogenous to regional identification and support for regionalization as respondents do not have agency in deciding about their birth year. Given these limitations to path analyses with the (repeated) cross-sectional data we have at our disposal, OLS regression models are the primary source for testing the hypotheses. We only use the path models where it is required to test the mediation expected in hypothesis 2.
All analyses distinguish between the Flemish and Walloon subsamples, as previous research has indicated that the political discourse about and implications of regionalization are very different in these regions, and that this has impacted political socialization processes (Sinardet and Morsink 2011;Dupuy et al. 2020). Moreover, Henderson et al. (2014) have demonstrated that when average regional identification in a region is high, individuals' regional identification is significantly related to support for further regionalization and that citizens in richer regions are more supportive of regionalization than citizens in poorer regions. Here too, Flanders and Wallonia differ greatly. In a first step, pooled analyses with survey fixed effects are estimated to inquire the overall relationship between the concepts of interest. In a second step, the analyses are presented per survey year. Table 2 presents the pooled OLS regression results for both the Flemish and Walloon subsamples, for the period 1991-2007, and for data collected between 2009 and 2019. 7 Table 3 presents the analyses per survey year.

Analyses
First, we inquire whether we indeed observe a normative policy feedback effect, which proposes that respondents who were socialized in a context of more institutional regionalization from an early age onwards would be more supportive of regionalization (hypothesis 1). Tables 2 and 3 show that the results are very different in both regions. In the pooled model for the Flemish samples from 1991 to 2007, we observe a weak but significant positive relationship between the experienced institutional regionalization when respondents came of age, and support for having more competences on the regional and community levels. If all other variables are kept at their mean, we observe in the estimation of marginal effects that respondents in the most recent cohort tend to score 4.450 on the 10-point scale measuring support for regionalization, compared to 4.280 in the oldest cohort. Hence, both are close to the middle of the scale, preferring a somewhat equal distribution of competences between both levels, yet younger cohorts lean slightly more towards regionalization. However, the relationship is not significant in the individual survey years, nor in the 2009-2019 pooled model. In the Walloon samples, in contrast, we consistently observe that respondents who came of age in a context of more regionalization are more supportive of regionalization (only in 1999 and 2019 no significant relationship is observed). The marginal effects estimated for the pooled analysis for 1991-2007 show that-keeping all other variables at their mean-respondents in the youngest cohort score 4.048 on support for regionalization (close to what is observed in the Flemish sample), while respondents in the oldest cohort clearly prefer more competences on the central, than on the regional level (3.468). In 1991 and 1995 (Table 3), this difference mounts to an entire point difference on the support for regionalization scale. In the 2009-2019 data, the youngest cohort of Walloons has an average support for regionalization of 4.909 (close to the response option that agrees with the current situation of regionalization), the oldest cohort would on average prefer some more competences on the central than on the regional level Table 2 Pooled OLS regressions explaining support for regionalization Data: 1991-2007: ISPO-PIOP Belgian Federal Election Studies, 2009: PartiRep Election Study, 2019: EOS RepResent post-election survey Regression coefficients (b) and Standard Errors (SE) are presented. *p < .05, **p < .01, ***p < .001. Results are weighted for age and gender (1991, 1995, 2003, 2007, 2019), age, gender and region (1999), age, gender and province (2009, 2014 Primary identification (Belgian is ref.  Regression coefficients (b) and Standard Errors (SE) are presented. *p < .05, **p < .01, ***p < .001. Results are weighted for age and gender (1991,1995,2003,2007,2019), age, gender and region (1999), age, gender and province (2009,2014). The Variance Inflation Factors (VIF) tests show that the model does not suffer from multicollinearity Table 3 (continued) .027 .075 .018 .112 .107 (4.505). We thus conclude that hypothesis 1 is confirmed in the Walloon context, but not in the Flemish context. 8 Hence, in a region-Wallonia-where the average regional identity is rather weak and that is poorer, the experience of regionalization plays a key role in explaining citizens support for regionalization. Second, we hypothesized that citizens who came of age in the context of more institutional regionalization would be more likely to primarily identify with this community, relative to the national community (H2a), and as a result be more supportive of attributing competences to the regional government level as it corresponds to the community identified with (H2b). We thus expect an indirect policy feedback effect by institutional regionalization, through political-territorial identification, on support for regionalization. While the association between regional and Belgian identification and support for regional identification is included in the OLS regression models (Tables 2, 3), path models are required to fully test this hypothesis. Path models for the pooled datasets per region are included in Fig. 3a-d. The results of the path analyses per survey year are included in Appendix D of ESM. The analyses in Tables 2 and 3, and Fig. 3a-d show a clear confirmation of part of the expected relationship (Hypothesis 2b): respondents who primarily identify with their region or community, compared to primarily identifying as Belgian, are more supportive of competences on the regional/community level, rather than on the central/Belgian level. This is in line with previous research on attitudes about regionalization and Europeanization that showed that citizens tend to prefer a correspondence between the governance level they identify with, and the level at which decision-making takes place (Verhaegen et al. 2018).
It should be noted, however, that this relationship is stronger and more consistent in Flanders than in the Walloon samples. In the 1991-2007 pooled analysis (Table 2), for instance, we observe that Flemish respondents who primarily identify as Flemish tend to be 2.803 more supportive of regionalization than Flemish respondents who primarily identify as Belgian, all else being equal. When all other variables are kept constant at their mean, the estimated marginal effects show that respondents who primarily identify as Flemish would prefer clearly more competences on the regional level (6.307 on the support for regionalization scale) than other respondents (3.503). In the Walloon sample, this difference is less than half 8 We replicated the analyses with only two cohorts (Appendix E.1 of ESM), making a distinction between respondents that came of age before the first state reform, and respondents that came of age after 1970 (i.e. cohort 1 as in Table 1 vs. all other cohorts). The conclusion holds that, generally, Walloon respondents who came of age in the context of institutional regionalization are more supportive of regionalization than Walloons of the first cohort. In Flanders no significant relationship is observed except for in the pooled 1991-2007 model. Yet, the relationship is significant in fewer survey years in Wallonia when using this alternative operationalization. This suggests that the gradation of depth of regional integration when coming of age matters. A replication of the analyses with a distinction of cohorts based on the 1993 state reform where Belgium became a federal state lead to the same conclusions. Appendix E.2 of ESM presents the analyses when the variable 'experienced institutional regionalization when coming of age', which is an ordinal variable, is entered in the model as a categorical variable through dummy coding. While these models show specific differences between cohorts and the oldest cohort in the data, the conclusions drawn about H1 are very similar. as large (b = 1.095). 9 Also, we observe a significant relationship in Flanders in each survey year, but in Wallonia we do not observe a significant relationship between regional identification and support for regionalization in 2003 and 2014 (Table 3). This difference between the regions corresponds to the higher salience of political discourse in Flanders that connects Flemish identity to calls for independence (Béland and Lecours 2008), and with the study of Henderson et al. (2014) where regional identification is observed to be only related to attitudes about regionalization in contexts where average regional identification is high. 10 Additionally, we observe that respondents who primarily identify with another political-territorial entity (province, municipality, EU etc.) too, tend to be more supportive of regionalization than respondents with a primarily Belgian identification. As the large majority of respondents in this 'other' category primarily identifies with other subnational governance levels, we can interpret this relationship as indicating a preference for governance at lower levels than the central one more generally.
No support is observed for the expectation (Hypothesis 2a) that cohorts that grew up in a more regionalized Belgium are more likely to primarily identify with their region or community rather than as Belgian (Fig. 3a-d). In contrast, we show that respondents from more recent cohorts are less likely to primarily identify with their region or community, and more likely to identify with another entity, than to primarily identify as Belgian. Given this unexpected finding, we also ran the models with an alternative operationalization of regional and Belgian identification for the surveys where this is available (1995 and 2014, see Appendix E.3 of ESM). Here, respondents were asked how strongly they identify with their region and Belgium respectively, rather than as a type of trade-off between both communities (this measurement is in line with research arguing that political-territorial identification is often not hierarchically ordered within individuals, see Bruter 2008, Duchesne andFrognier 2008). 11 The findings are robust for this alternative operationalization: when significantly related to cohort, respondents in younger cohorts tend to identify less strongly with their region and Belgium. However, strength of regional and Belgian identification are respectively positively and negatively related to support for regionalization. Explanations for the development of regional identification, and explanations for younger cohorts' higher tendency to primarily identify with another community (city, village, province or Europe) should thus be sought elsewhere.
Finally, the control variables reveal mixed relationships between respondents' satisfaction with the financial situation of their household, political awareness and gender, and support for regionalization. Regarding satisfaction with the financial situation of one's household, we each time observe either a significant relationship in Flanders, or in Wallonia between 1991 and2007 (Tables 2, 3). In later years, no significant relationship is observed between respondents' socio-economic position and support for regionalization in Flanders. In 2019, a relationship opposite to the one expected is observed in the Walloon sample. The observations are thus highly mixed. Additionally, the 1991-2007 surveys offer an alternative operationalization of socio-economic position: respondents were asked which class they count themselves in (the working class, lower middle class, higher middle class, or upper class). The results presented in Appendix E.4 of ESM are more consistent: in all survey years, and most strongly in the Flemish samples, we find that respondents who count themselves in a higher social class tend to be more supportive of regionalization. This corresponds to the discourses observed among Flemish elites which are both more widespread, and specifically make a link with socio-economic issues when arguing that regionalization is required as to prevent the loss of Flemish wealth to Walloons who 'willingly overuse country wide social insurance benefits' (Béland and Lecours 2008: 166;Dupuy et al. 2021). Further research is required to improve our understanding of the relationship between socio-economic factors and attitudes towards regionalization. Taking the policy feedback literature as a starting point, in particular the resource effect, we may expect to reach better understanding through analysing data on respondents' impression whether they or their region is better or worse of as a result of institutional regionalization, or by evaluating shifts in specific policies for which benefits and losses can be more specifically evaluated (Mettler and Welch 2004;Pierson 1993).  Flanders 1991Flanders -2007 Standardized coefficients from path model estimated with sem in Stata 16. N = 8728; RMSEA = .000, CFI = 1.000. Indirect effect Intensity institutional regionalization when coming of age-Support for regionalization = − .028*. R 2 Support for regionalization = .189; R 2 Regional identification = .037; R 2 Other identification = .016. Controls included for Satisfaction financial situation household, political discussion, following the news, gender, and survey fixed effects. (b) Pooled path model with survey fixed effects- Wallonia 1991Wallonia -2007 Standardized coefficients from path model estimated with sem in Stata 16. N = 5093; RMSEA = .020, CFI = .999. Indirect effect Intensity institutional regionalization when coming of age-Support for regionalization = .005. R 2 Support for regionalization = .065; R 2 Regional identification = .012; R 2 Other identification = .013. Controls included for Satisfaction financial situation household, political discussion, following the news, gender, and survey fixed effects. (c) Pooled path model with survey fixed effects-Flanders 2009-2019. Standardized coefficients from path model estimated with sem in Stata 16. N = 3051; RMSEA = .000, CFI = 1.000. Indirect effect Intensity institutional regionalization when coming of age-Support for regionalization = − .030***. R 2 Support for regionalization = .084; R 2 Regional identification = .035; R 2 Other identification = .013. Controls included for Satisfaction financial situation household, political interest, gender, and survey fixed effects. (d) Pooled path model with survey fixed effects-Wallonia 2009-2019. Standardized coefficients from path model estimated with sem in Stata 16. N = 2970; RMSEA = .000, CFI = 1.000. Indirect effect Intensity institutional regionalization when coming of age-Support for regionalization = − .011**. R 2 Support for regionalization = .047; R 2 Regional identification = .008; R 2 Other identification = .006. Controls included for Satisfaction financial situation household, political interest, gender, and survey fixed effects

Conclusions
This article inquired the relationship between experienced institutional regionalization and citizens' support for such processes, through the lens of policy feedback. The article first extended this theoretical framework by further theorizing the aspect of political socialization, and by specifying indirect policy feedback mechanisms. We argued that when institutional regionalization occurs or progresses, it shapes a different context in which young people grow up. Citizens may adapt their preferences about the division of competences to this new setting either directly through processes of early age socialization (normative policy feedback), or indirectly as political-territorial identification may mediate the relationship (interpretive policy feedback).
The analyses show clear evidence for direct policy feedback effects through early age socialization processes in the Walloon samples, but evidence for this is largely absent in the Flemish case. In Wallonia, cohorts that grew up in a more centralized state tend to support more competences on the central level than younger cohorts. After having lived for up to 49 years in an increasingly regionalized country, the cohort of Walloons that came of age before the regionalization process started still prefers to have more competences on the central level than younger cohorts. The attitudes of younger cohorts of Walloons are closer to their Flemish counterparts. This observation matches our expectations based on the literature on political socialization where it is argued that especially in childhood and adolescence-the impressionable years-attitudes are open to change, and remain more stable afterwards. We expected that older and younger cohorts would maintain a preference for a division of competences that is more similar to the way in which competences were divided (or not divided at all, as for the oldest cohort) between the subnational and central level when they came of age. Among Walloons, we thus observe that when institutional regionalization took place, youth political socialization enforced preferences for this distribution of competences across government levels. People tend to stick to these preferences throughout the life course, as we see in the persistent division between younger and older cohorts.
These findings are in line with differences in the political context in both regions. Discourses on regionalization have been much stronger in Flanders than in Wallonia. In Flanders, both regionalist/separatist and mainstream parties have advocated for more competences on the regional and community levels, based on arguments related to Flemish identity and economic arguments (Béland and Lecours 2008). This is mirrored in the correlates we observe to explain support for regionalization in Flanders. In Wallonia, in contrast, there is much less discussion about regionalization. If anything, Walloon politicians react to the arguments of Flemish politicians, and appeal to country-wide interests and feelings of Belgian unity. Party manifestos rarely refer to regionalization (Sinardet and Morsink 2011). As a result of this weaker salience of regionalization and fewer party-cues regarding the issue in Wallonia, the mechanism of adaptive preferences to the institutional setup that people are used to and that most actively takes place at a younger age becomes visible in Wallonia (Dupuy et al. 2021). In Flanders in contrast, this mechanism is obfuscated because the entire population is exposed to elite discourses and party manifestos that put regionalization centre-stage.
When it comes to the role of identification, our empirical analyses demonstrate that it is not entirely as expected. Flemish support for regionalization is consistently and substantially explained by citizens' sense of regional, relative to Belgian, identification. This is consistent with Henderson et al.'s research (2014) that shows that the role of regional identification in accounting for support for further regionalization depends on a strong average level of regional identification. It also corresponds with the work of Reuchamps et al. (this issue) that observes that, next to making references to efficiency, identity plays a key role in Belgian citizens' discourses about regionalization. However, no support is observed for the expectation that cohorts that grew up in a more regionalized Belgium are more likely to primarily identify with their region or community rather than as Belgian. In fact, when significantly related to cohort, respondents in younger cohorts tend to identify less strongly with their region (and Belgium). Similarly, Hildebrandt and Trudinger (this issue) expected but did not observe a socialization effect of the establishment of Länder in Eastern Germany-with the reunification in 1990-on the regional identification of cohorts that came of age before and after German reunification. While explaining this result is beyond the scope of our study, it is an important result that requires further investigation. How can we explain that experiencing regionalization affects youngest cohorts' attitudes towards regionalization, yet has a reversed impact on regional identification? Bringing European integration into the equation may help solving this puzzle as the youngest cohorts might identify more as European than their older counterparts. Indeed, it has been demonstrated that citizens' European identification and support for European integration are related to their generation (Ceka and Sojka 2016;Down and Wilson 2013). Moreover, we observe in our data that younger cohorts tend to identify more often primarily with an 'other' community, which includes Europe, but also the municipality and province. 12 Taking the multilevel-regional, national and European levels-dimension of identification systematically into account is thus a promising avenue for future research, yet challenging when it comes to data availability and theorization.
These conclusions show how this study contributes to the literature on policy feedback, and to scholarship on regionalization in Belgium and beyond. First, we linked the literature on policy feedback to the literature on (youth) political socialization and identification and specified direct and indirect policy feedback mechanisms. In previous studies, socialization and identification would only be entered in the equation as controls for alternative explanations for attitudes about regionalization. Their relationship with institutional regionalization would not be theorized or inquired. Our results demonstrate of a direct normative policy feedback, but there is no evidence in our data of an indirect interpretive policy feedback when it comes to regionalization in Belgium. While growing up in a more regionalized context strengthens citizens' support for regionalization, experiencing regionalization is not related to higher levels of regional identification.
Second, this contribution extends previous insights on regionalization in Belgium in various ways. Earlier studies mostly analysed attitudes about regionalization and support for nationalist parties by using single cross sections. They drew conclusions about which explanation (typically identity or economic position) is more important to explain attitudes about regionalization, or support for nationalist parties (e.g. Thijssen et al. 2015;Verhaegen et al. 2018). We contribute to this debate that both identity and youth socialization indeed explain part of the puzzle, and we qualify the reach of their explanatory value in the two largest regions in Belgium.
Third, our study suggests more generally that when adaptations are made to the division of competences to various levels of government, we can expect this to affect citizens' preference for this new division of competences through processes of youth socialization. This process would be an explanation for generational differences in attitudes about regionalization and Europeanization, and has important implications for expectations about the effects of generational replacement for future developments of public opinion.
Comparing Flanders and Wallonia, where citizens have experienced the same institutional reform yet have been exposed to very different contexts of elite discourse both in intensity and in content, we show that this context matters. We observe that in the Flemish context with intense elite discourse about regionalization, and a high average level of regional identification, the process of youth socialization is not visible in citizens' support for regionalization. In Wallonia, in contrast, where regionalization is less salient in elites' discourses and where the level of regional identification is rather weak, we do observe socializing effects of institutional regionalization. Hence, if we want to understand the effects of institutional regionalization (or Europeanization) on individuals' attitudes towards the division of competences between government levels in other cases, our study indicates that both identification and socialization may play a role, depending on the particular context in terms of elite discourses and average strength of (regional) identification.
Our findings thus point at relevant aspects of the context of (calls for) changes in the division of competences, which future case studies may look at. Importantly, our case study underlines the importance of also studying cases such as Wallonia, where regionalization is not very salient and regional identification is rather weak. While calls for regional competences and secession in, for instance, Flanders, Scotland and Catalonia rightly draw scholarly attention, our study shows that also in regions such as Thuringia, Andalousia and Limousin too, experiencing institutional regionalization may affect citizens' support for regionalization. This in turn may be consequential for the course institutional reform takes in these countries.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.