The domestic sources of sub-state foreign policymaking: determinants of subnational development cooperation across European regions

ABSTRACT Subnational governments have come to challenge the foreign policymaking monopoly of national governments. An increasingly salient area of sub-state foreign policymaking is international development cooperation, but systematic knowledge about the drivers of such activities is limited. We argue that subnational governments develop capacities for international development cooperation for domestic region-building purposes, geared toward advancing political claims for regional autonomy that can take institutional, political, and cultural representations. To test this argument, we construct a new dataset of 195 politically relevant European regions which maps the autonomous engagement of regions in international development cooperation. Large-N regression analysis establishes that regions are significantly more likely to be engaged in international development cooperation where they have greater constitutional autonomy, cultural-linguistic distinctiveness, and a regionally-based party to advance the regional agenda. These results have important implications for our understanding of sub-state foreign policy actors and their role in the global governance of development.


Introduction
Sub-state governments are ever-more prominent foreign policy actors. They undertake autonomous foreign policies in areas including climate change, migration, and trade and investment, which have long considered to be under the exclusive authority of nation-states. To emphasize the autonomous nature of sub-state foreign policies, scholars have referred to this phenomenon as 'paradiplomacy' (Aldecoa & Keating, 1999;Béland & Lecours, 2005;Kuznetsov, 2014;Schiavon, 2018). Despite in-depth case studies on paradiplomacy (Alexander, 2014;Duran, 2015;Royles, 2017), we lack a systematic understanding of its underlying determinants.
In this article, we cast light on the determinants of sub-state foreign policymaking using the example of international development cooperation. This policy field has become more important in the context of the Sustainable Development Goals (SDGs), which acknowledge the potential for development contributions from diverse actors beyond the nation-state, including sub-state governments, municipalities, and public-private partnerships (Kania, 2021;Valmorbida, 2018;Wingens et al., 2021). We use a new dataset that allows us to systematically track the development cooperation activities of 195 politically relevant European regions. We find that a surprisingly high number of around 70 regions provide development aid, although typical aid budgets are small and regional aid programs are delivered primarily through civil society organizations.
Our new dataset is uniquely positioned to examine why and when substate governments engage in international development cooperation. Theoretically, building on a rich body of case studies on development (Alexander, 2014;Criekemans, 2009;Royles, 2017) and the literature on European regional mobilization (Blatter et al., 2010;Donas & Beyers, 2013;Huwyler et al., 2018;Massetti & Schakel, 2017;Tatham & Thau, 2014), we argue that sub-state governments have incentives to utilize international development cooperation for domestic region-building. Related claims can take many forms and can have institutional, political, and cultural representations. They do not all involve secessionismas a few well-known cases like Catalonia or Scotland would suggestbut more generally the 'region-building' attempts by sub-state governments. 1 Our argument thus emphasizes politics as the most salient driver of sub-state development cooperation. Therefore, we expect subnational development cooperation to be more likely where regions have greater existing constitutional powers for policymaking, higher levels of cultural-linguistic regional distinctiveness, and where there are regional parties as political entrepreneurs that can leverage existing opportunity structures.
We empirically test our expectations by combining our dataset with other regional-level datasets. We find that regions with high levels of de jure policy autonomy and with distinct political-cultural features are significantly more likely to engage in development cooperation. The associated effects are sizeable. For example, increasing the regional authority index by a standard deviation is associated with an increase in the likelihood of sub-state development cooperation by about 40 per cent (p < 0.01). While regions with greater bureaucratic capacity and financial resources also tend to be more likely to engage in development cooperation, we find no evidence that subnational development cooperation is driven by economic concerns and commercial interests. Importantly, our results withstand cross-sectional robustness checks and hold even within time-series cross-section analysis. Alongside supportive qualitative evidence from a number of elite interviews with sub-state government policymakers, these findings bolster our argument about the domestic sources of sub-state foreign policymaking.
Our article connects hitherto separate branches of literature on paradiplomacy, foreign aid, and multi-level governance, besides offering distinct contributions to each of them. First, our article contributes to the burgeoning literature on paradiplomacy (Cornago, 2010;Keating, 1999;Paquin, 2021). Related studies have examined subnational policies in other issue areas, such as environmental governance (Setzer, 2015), trade (Paquin, 2021), and diplomatic relations more generally (Dickson, 2014), but ours is the first to offer systematic evidence on international development. While previous works have heavily drawn on case studies, our articleto our knowledgeis the first comparative large-N study on the determinants of subnational development cooperation. This remedies a major methodological challenge in the case-study literature: Focusing on (a few) positive cases is insufficient for establishing conditions under which regional governments engage in development cooperation.
Second, by introducing subnational governments as autonomous development policymakers, we extend the scope of the foreign aid literature, which focuses on donor countries and their aid allocations (Hoeffler & Outram, 2011;Reinsberg, 2015;Younas, 2008). Closest to our study is work that unpacks donor countries to consider donor bureaucracies as aid providers (Carcelli, 2019). This work overcomes the limited unitary-actor view of government but still neglects subnational governments, failing to capture the political dynamics leading to autonomous subnational aid policies. In addition, while related scholarship broadly distinguishes between donor interest and recipient need, we identify sub-state nation-building as an entirely new motivation for aid giving.
Third, multi-level governance research has long argued that national governments no longer have the monopoly on foreign policymaking (Hooghe & Marks, 2003;Reinsberg & Westerwinter, 2021). While much research focuses on the powers of supranational bodies like the European Union (Carbone, 2007;Michaelowa et al., 2018), our article focuses on subnational governments. Unlike multi-level governance research, which presumes that regions organize at the supranational level to advance their interests (Callanan & Tatham, 2014;Huwyler et al., 2018;Jeffery, 1996;Tatham, 2015), we emphasize the autonomous foreign policy choices of these regions.

The domestic sources of sub-state foreign policymaking
Our argument about why regions (or sub-state or subnational governments) become donors is rooted in domestic politics: it advances their territorial-based claims and related region-building attempts. Our argument pertains to a wide range of claims, from greater autonomy within a quasi-federal polity, regional influence within national-level structures, and full-fledged independence from the central state, which follow similar causal drivers (Sambanis & Milanovic, 2014). As we show below, region-building involves a dual process of fostering a regional identity based on distinct values and distinct policies as well as bolstering legitimacy of an autonomous regional polity at the international level.
Existing research on 'paradiplomacy' 2 explains how the undertaking of foreign policies can help regions establish themselves as (autonomous) political entities. Early paradiplomacy research emphasized the public diplomacy function, whereby regions seek to enhance their reputation abroad. This research focuses on international audiencesincluding diverse actors such as (sub)national governments, international organizations, foreign diasporas, and transnational actorsthat can endow subnational governments with legitimacy by endorsing regional aid activities and interacting with regions as state-like entities, thereby explicitly recognizing them as autonomous political entities (Criekemans, 2009;Huijgh, 2019;Lecours, 2003).
However, paradiplomacy also has a domestic dimension, which involves a process of identity construction (Criekemans, 2009;Paquin, 2018). Pathbreaking work by Béland and Lecours (2005) shows how subnational political elites develop social policies instrumentally to foster a regional identity and, in turn, how sub-state welfare policies bolster such identity from the bottom up through 'territorial mobilization'. Using the examples of Flanders, Québec, and Scotland, they show that sub-state social policy allowed regional governments to foster distinct regional identities, while at the same time using these policies and the subnational identification that they foster to protect regional autonomy within their quasi-federal polity (Béland & Lecours, 2005).
Extending this line of argument, we argue that sub-state governments can further strengthen regional identity through the continuation of social policies at the international level. International development cooperationunlike other foreign policiesis uniquely positioned for this as it is based on the same values as social policy at home. For example, Scottish governments have used foreign aid to instill a national identity based on 'Scottish values' (Interviews 1−3). In the foreword of its current development strategy, the Scottish aid minister asserts that our values continue to provide a voice for humanity, tolerance, compassion, and progressiveness, as part of an interlinked global community … Scotland cannot act with credibility overseas if we are blind to inequality here at home. And our ambitions for a fairer Scotland are undermined without global action to tackle poverty, promote prosperity, and tackle climate change. (Scottish Government, 2020, p. 4) Similarly, a Basque government representative said, 'we can put our local expertise in social policy at the service of international development' (Interview 4). These quotes indicate the strong congruence between domestic social policy and international development policybased in strong solidaritythat is integral to regional identity-building.
Our argument has two implications as to the organizational structures of sub-state development cooperation as well as its substantive policy content. Specifically, we would expect institutionalization of aid activities through the establishment of regional aid agencies because it will make regions look 'like' states and they will be perceived as such (Dingwerth & Pattberg, 2009;Finnemore, 1993;Meyer & Rowan, 1977). In the words of Lecours and Moreno (2003, p. 270), the 'development of an international agency on the part of a regional government is full of symbolic meaning, and it is therefore an attractive strategic option … through paradiplomacy, regions can both behave as nations and present themselves as such'. Development cooperation is highly suited for state-building purposes because of established norms about the form of cooperation through 'aid agencies' (Gulrajani & Swiss, 2019). Furthermore, as aid relationships are bilateral, sub-state actors are directly legitimized as an autonomous foreign policy actor through (implicit or explicit) recognition by recipient states (Criekemans, 2009;Huijgh, 2010;Lecours & Moreno, 2003). 3 Our argument also helps anchor our expectations on the content of regional aid programs. To be useful for regional identity formation, substate aid must be different from its national-level counterpart. Such differences, articulated publicly by regional elites, become an integral part of the collective regional identity (Lecours & Moreno, 2003, pp. 271-72). Examples from our own research on sub-state development policy corroborate this expectation. A regional government representative asserted that their agency 'wants to be more flexible [than the central government]', which would involve 'taking a flexible approach towards stakeholders […] without fixed quotas' and 'taking more risks [to fund] innovative pilot projects' (Interview 6). An interviewee from another regional donor stated that 'central states face significant limits when addressing sensitive issues such as human rights, indigenous rights, and climate change'. In contrast, regional donors could 'cooperate altruistically', freed from 'competing foreign policy agendas' (Interview 4). 4 A Basque government official said their aid program would allow them to 'project the Basque country as a kind actor' [that is] 'doing things in a different way'. Hence, the possibility to present themselves as altruistic helps regions foster a regional identity which in turn sustains the distinct content of regional development policy anchored in 'good global citizenship'.
We claim that region-building attempts provide a parsimonious and general explanation for regional engagement in development cooperation.
They come in various shapes and have institutional, political, and cultural representations (Blatter et al., 2010). Mirroring these dimensions, we identify three facilitating conditions of sub-state foreign policymaking: a permissive constitutional framework, cultural-linguistic distinctiveness of a region, and a regionalist party that can lead the process of 'territorial mobilization'.
First, subnational governments need constitutional powers to undertake autonomous foreign polices (Setzer, 2015). Over the past decades, subnational governments have strengthened their policy-making competencies, following a series of de jure reforms for greater autonomy and greater influence in national policy-making (Hooghe et al., 2016). For example, Flanders has developed capacities for development cooperation following a devolution process that provided the region with the authority to act in all areas internationally in which it can act domestically. Flanders even has the constitutional right to conclude binding international treaties Criekemans, 2010;Paquin, 2021) (Interview 6). Hence, sub-state governments will find it easier to establish autonomous capacities for international development cooperation where they have the legal authority to do so. 5 Hypothesis 1: A region is more likely to engage in development cooperation if it has a high level of de jure regional authority.
Second, we expect those regions with distinctive cultural traditions to be most likely to become aid donors. Nationalism requires an identity based on some common marker like shared language, religion, or ethnicityreal or imagined (Anderson, 1983;Barbour & Carmichael, 2000;Hobsbawm, 1992)that fuel a subsequent process of 'territorial mobilization' (Brass, 1991). As Béland and Lecours (2005, p. 678) put, [c]ultural distinctiveness, particularly in language, can serve as a relatively straightforward criterion for defining the national community, that is, for specifying who should be included and excluded. Furthermore, once the contours of the nation have been determined, culture acts as a powerful reference for national identity because it is full of symbolic meanings that can provide binding ties to a community.
Such straightforward markers of regional distinctiveness make it easier to mobilize politically for distinct sub-state agendas. Past research found that regions are more likely to break apart from a federal state when they are culturally distinct from the federal state (Rode et al., 2018). Cultural distinctiveness also provides incentives for international engagement more generally (Blatter et al., 2010). Québec and Wallonia provide two illustrative cases. As Francophone regions in multi-national states, they engage in cultural diplomacy to protect their cultural distinctiveness (Criekemans, 2010). Regions with their own language often seek support abroad from like-minded communities, especially when federal governments have little sympathy for their concerns, as for example Catalonia (Keating, 1999). International development cooperation can be a useful mechanism underpinning those foreign relations (Interview 5).

Hypothesis 2:
A region is more likely to engage in development cooperation if it has a distinct cultural-linguistic tradition.
Third, there may need to be a regionalist party which can exploit opportunity structures for sub-state foreign policymaking. These parties often lead processes of political-territorial mobilization that build on the presence of regional markers like a distinct cultural-linguistic tradition (Béland & Lecours, 2005;Brass, 1991;Fitjar, 2010;Keating, 1988). Regionalist parties serve to bridge existing cleavages to rally the subnational population around the region-building project and to advance political claims for regional autonomy (Lecours & Moreno, 2003, p. 272). If in power, regional parties can implement policies that strengthen subnational identity and further underscore the distinctiveness of the region vis-à-vis the central state. This should not stop short of pursuing domestic policy objectives. Claiming foreign policymaking turf fulfills an instrumental role in escalating demands for autonomy versus the central state. Scotland provides a case in point. The 1998 Scotland Act devolved power to the Scottish government but formally limited its foreign policymaking authority to EU-related matters (Blatter et al., 2008, p. 472). In 2005, the Scottish government under Labor initiated its development cooperation program, but it was the Scottish National Party which substantively scaled it up and made it distinctively 'Scottish', in the context of a sustained drive toward independence. This evolution of Scottish international development cooperation in a phase of partisan incongruence with the central government chimes with regional and federal studies showing how party incongruence incentivizes the use of bypassing strategies by regional governments (Callanan & Tatham, 2014, p. 203;Huwyler et al., 2018, p. 768).

Hypothesis 3:
A region is more likely to engage in development cooperation if it has a regionalist party.
In sum, we offer a domestic politics explanation for regional development cooperation that establishes three conditions under which regions should engage in this activity. Moreover, our framework anchors our expectations about the form and substance of such activity. Our argument is analytically distinct from approaches that conceive regional foreign policymaking as a response to globalization Keating, 1999;Paquin, 2021). Acknowledging that globalization matters, as well as the diversity of motivations for paradiplomacy (Blatter et al., 2010), we think that paradiplomacy in development cooperation more often than not follows a political logic, ushering in attempts of regional elites to foster regional autonomy.

Research design
In this section, we introduce a new dataset on sub-state governments and their international development cooperation activities. While related work describes the data in much detail (Reinsberg & Dellepiane, 2021), our focus here is to examine the conditions under which subnational governments engage in development cooperation. To our knowledge, this is the first attempt to address this issue using large-N analysis. To do so, we combine the dataset with other pertinent regional-level datasets. 6 The subnational donor governance dataset We collected data covering 195 politically relevant European subnational governments in 23 quasi-federal states. 7 We consider as politically relevant those regions that are members in the Committee of the Regionsan EU advisory body with elected representatives from sub-federal entities. 8 In most EU countries, the political relevant regions are the NUTS-1 regions. In some others, these are the NUTS-2 regions. 9 The supplemental appendix includes a list of all regions in the dataset (Table A1). 10 To construct the dataset, we drew on information published on the official websites of those regions. Where we could not find information on official websites, we triangulated the relevant information from other sources. Where we found no information, we ascertained the region was not engaged in international development cooperation. To ensure reliability, validity, and reproducibility of our data, we double-coded all regions, involving a research assistant and a senior researcher. Coders generally concurred on fundamental decisions such as whether the region had a foreign aid program. Differences occurred where information had to be quantified, such as for numbers of sub-sites, aid programs, and regional staff. By using dummy variables, we enhance the reliability of our coding.
To further enhance the validity of our data, we contacted subnational government representatives, asking for confirmation of our findings. Where contacts were not available through websites, we contacted the relevant Committee of the Regions representatives. To probe the plausibility of our arguments beyond large-N analysis, we also conducted half a dozen openended interviews, with representatives from Bavaria, the Basque Community, Catalonia, Flanders, and Scotland, learning about motivations for aid-giving and governance modalities (Appendix C).

Key variables
Our main dependent variable drawn from the subnational donor governance dataset is AID WEBSITE, indicating whether a region has a website on development cooperation. We believe this is a pertinent measure because regions which undertake autonomous aid programs will likely report on them publicly. Our alternative indicator, AID AGENCY, captures the presence of institutionalized capacity for subnational development cooperation in the region. Similarly, the variable AID PROGRAM captures whether the regional government reports any development program under its responsibility. While these indicators may tap different motivations for development cooperation, they are highly correlated (ρ > 0.85), thus reliably identifying the regions that are engaged in development cooperation. 11 The data reveal that at least 70 regions undertake international development cooperation. This is a surprisingly high number, given the few well-cited cases in the paradiplomacy literature.
In line with our theoretical expectations, we construct the following key predictors. First, we include a measure of de jure regional autonomy, the REGIONAL AUTHORITY INDEX (RAI). The RAI codes systematically the constitutional powers of regional actors, distinguishing between self-rule prerogatives and shared-rule powers (Hooghe et al., 2016). Second, to proxy for cultural-linguistic distinctiveness, we include DISTINCT LANGUAGEindicating whether a majority of the regional population speaks a language (including dialects) that is different from the majority language in the country (Shair-Rosenfield et al., 2021). 12 Culturally-linguistically distinct regions will have greater incentives to strive for autonomy, which may influence their willingness to establish autonomous development cooperation. Third, we include a dummy for REGIONALIST PARTY, which measures whether the region ever had a party with a pro-independence platform (Rode et al., 2018). Admittedly, this is a too narrow proxy for our concept of regional parties, whose aim may not be full-fledged independence from the central state; however, we prefer this indicator due to its extensive coverage. 13 Given the high correlation among all three variables, we would expect some insignificant coefficient estimates. The ultimate assessment of our argument therefore relies on a joint F-test of all three predictors. 14

Control variables
We consecutively include three sets of control variables capturing alternative explanations for paradiplomacy. These control variables draw on previous large-N research at the regional level (Donas & Beyers, 2013;Huwyler et al., 2018;Tatham, 2015;Tatham & Thau, 2014). We also consider control variables that are likely to be particularly relevant in the regional development cooperation context, drawing on previous foreign aid literature (Bermeo, 2017;Fuchs et al., 2014;Younas, 2008).
The first set captures demands arising from intensifying globalization (Aldecoa & Keating, 1999). In line with arguments about commercial motivations for donors (Younas, 2008), international development cooperation may help regions attract investment and promote domestic exporters (Keating, 1999). Given that regional aid programs are small, we would expect them to help tip the balance in favor of enterprises in regions that are already economically competitive. Therefore, we include the REGIONAL COMPETITIVENESS INDEX (Annoni & Dijkstra, 2019). This measure combines indicators on macroeconomic stability, regulatory quality, social services, labor market efficiency, technological readiness, business sophistication, and innovation. In addition, following national-level aid allocation literature, we expect that aid programs will be harder to sustain for donor governments when economic circumstances are dire (Heinrich et al., 2016). Therefore, we expect that economic woes reduce the chances that subnational actors establish aid programs, due to the limited fiscal space for providing aid as well as increased pressures to support domestic welfare. To capture economic hardship, we include the rate of total UNEMPLOYMENT in the region from Eurostat (EC, 2020). Finally, an important local consequence of globalization is migration. For example, some German Länder stated on their websites to have become more engaged in development cooperation following the European migration crisisallegedly to help tackle the root causes of migration overseas. Lacking a region-specific measure of migration inflows, we use NET MIGRATION as percentage of the population from Eurostat (EC, 2020). This measure is imperfect as it is non-directional and includes different types of migration.
The second set captures capacity-based explanations for sub-state foreign policymaking. We expect that regions with higher governance quality, greater resources, and greater wealth are more likely to build up development cooperation programs. Hence, we include the European QUALITY OF GOV-ERNMENT Index (EQI), which uses surveys to measure the quality of institutions at the regional level (Charron et al., 2014). Survey questions are about perceived and experience with corruption, impartiality of services, and quality of public services in several public service sectors. In addition, we draw on Eurostat to measure regional (logged) GDP and (indexed) GDP per capita (EC, 2020).
The third set of measures captures opportunity structures. We include BRUSSELS OFFICE, a dichotomous variable for whether a region has a representation in Brussels. 15 The Brussels dummy has been used in previous research and is commonly understood to proxy for the political clout of regional actors beyond the national level (Callanan & Tatham, 2014;Tatham, 2015;Tatham & Thau, 2014). In our context, we consider Brussels to be a venue of policy diffusion where regions can interact with their peers and learn about development cooperation in other regions. Furthermore, we include a dummy variable for BORDER REGION with a different European country. Our rationale for this variable is that proximity to foreign regions can make regions more aware of international issues. Others have argued that border regions, which are also peripheral regions, have a stronger sense of regional identity (Brigevich, 2012). Finally, we include (logged) population (Blatter et al., 2010;Huwyler et al., 2018;Tatham & Thau, 2014). Larger regions can expect to have more leverage while at the same time their per-capita fixed costs of establishing international development capacities are lower.
As baseline control variables, we always include the FOUNDING YEAR of the regional entity, available from the RAI dataset (Hooghe et al., 2016). This may account for institutional legacies or cohort effects. In addition, we control for whether the region is an ISLAND REGION. While islands may be more likely to develop distinct cultural traditions, they may also have greater incentives to turn their geographic isolation into an advantage by forging transnational links. We discarded potential control variables with limited data availability, such as the political ideology of regional governments, and the number of local NGOs, which may be potential beneficiaries of regional aid programs. Table 1 shows descriptive statistics of all variables in the cross-sectional analysis.

Empirical model
Given the binary nature of our dependent variable, we proceed with probittype models in our main analysis. For robustness tests, we estimate linear probability models and multi-level models. We cluster standard errors on countries, thereby accounting for potential spatial interdependence between any two regions of a given country. While linear probability models would allow us to include country-fixed effects and thereby control for omitted country-level factors, their drawback is that countries without subnational variation in the outcome will drop out from the estimation. This is a considerable drawback given the size of our dataset. A preferred alternative are multi-level random-effect models, which observe the twolevel structure of the data whereby regions are nested in countries (Schmidt-Catran & Fairbrother, 2016). Table 2 shows our main results, using AID WEBSITE as dependent variable. We find a strongly statistically significant positive relationship between all three facilitators and regional aid websites (p < 0.01). In substantive terms, based on the last column, an increase of de jure regional authority at its mean by seven index pointsroughly its standard deviationis related to a change in the predicted probability of an aid website from 35.9 per cent (95%-CI: 19.1%-52.8%) to 76.0 per cent (95%-CI: 63.0%-88.9%). This is a substantial effect given the empirical range of the RAI. 16 The other paradiplomacy variables are similarly important. Having a distinct language increases the predicted probability of an aid website from 40.0 per cent (95%-CI: 17.7%-62.4%) to 60.1 per cent (95%-CI: 37.7%-82.5%). Finally, the presence of a regionalist party increases the likelihood of a regional aid website from 37.7 per cent (95%-CI: 15.2%-60.1%) to 87.4 per cent (95%-CI: 68.5%-99.9%). Alternative explanations do not receive much support in our data. Regional competitiveness does not have the expected positive significant coefficient. The effects of unemployment and net migration are insignificant. Furthermore, we find that regional governance quality is weakly positively associated with the likelihood of an aid website (p < 0.1). While the total output of the region is significantly positively related with an aid website (p < 0.05), the per-capita income level of a regional economy is not significant. Finally, we find that having a Brussels representation is positively related to having an aid website (p < 0.05). The effect of being a border region is not significant. Younger regions tend to be more likely to have an aid website. Our models explain a decent share of the variation, given a pseudo-R 2 of up to 66.1 per cent (Domenich & McFadden, 1975).

Cross-sectional analysis
In the supplemental appendix, we probe the robustness of these findings. First, we consider two alternative dependent variables. Findings are qualitatively similar for AID AGENCY (Table B1) and AID PROGRAM (Table B2). Both the RAI and the regionalist party indicator are statistically significant across both outcomes and different models. Distinct language is statistically significant in three models for aid programs, but not for aid agencies. 17 All three indicators Table 2. Determinants of regional aid website. (1) ( are jointly significant, with F-statistics well beyond conventional levels of statistical significance (p < 0.001). Control variables are relatively similar across alternative outcomes, with some minor exceptions. Specifically, strong quality of governance predicts whether a region has an aid agency, but not aid programs. A similar pattern exists for Brussels office, which is more statistically significantly related to having an aid agency, relative to having an aid program. Second, we consider alternative modeling approaches, given that nonlinear models may be less robust to model misspecification. Across all three dependent variables, we estimate linear-probability models. For AID WEBSITE (Table B3), AID AGENCY (Table B4), and AID PROGRAM (Table B5), the RAI is always positively significant (p < 0.01). Similarly, regionalist party is always significantly positively related to the outcome variables, even if only marginally in two out of nine models. Distinct language is no longer consistently estimated. However, as confirmed by the F-statistics, all three variables jointly continue to be highly significant. Adding countryfixed effects, we find that the RAI is the only remaining robust predictor of AID WEBSITE (Table B6), AID AGENCY (Table B7), and AID PROGRAM (Table  B8). The joint effect of all three key independent variables is strongly significant. Therefore, our findings are not driven by country-level factors. 18 Turning to multi-level random-intercept models, we find that again for AID WEBSITE (Table B12), AID AGENCY (Table B13), and AID PROGRAM (Table  B14), the RAI is positively significant. Distinct language is a consistently positive correlate of both AID WEBSITE and AID PROGRAM, while regionalist party is consistently significant across all outcomes. The joint F-statistics are again very strong.
Third, we probe robustness to different sets of control variables from previous large-N regional studies with an interest in regional foreign policymaking. For example, Huwyler et al. (2018) examine venue-shopping by regional governments, while Donas and Beyers (2013) examine the determinants of transregional associations as a conduit of influence beyond representation in Brussels. In turn, Tatham and Thau (2014) study the correlates of the number of Brussels-based regional diplomats. Tatham (2015) examines success conditions for regional interest representation in Brussels. Using the control sets of these four studies and AID WEBSITE as outcome, our three key predictors remain jointly significant (Table B15). In addition, we verify robustness of our findings using alternative control variables and measurements of predictors. Specifically, we include measures of partisan incongruence between the regional government and the national government (Schakel & Massetti, 2018), party system congruence, and regional distinctiveness based on comparing regional election results and national-level election results of the same regional electorate (Schakel, 2013). Despite this demanding test, our results hold. Our results also hold when using a measure of regionalist parties based on a list approach (Donas & Beyers, 2013). Neither are results driven by processes of regional mobilization, using population density as additional control (Tatham & Thau, 2014), nor by peer diffusion of regional aid agencies (Table B16). Finally, we eliminate two country-level confounding effects. While regions that implement aid on behalf of the national government tend to be more likely to undertake their own development cooperation, our main results are unaffected. Similarly, while regions that have authority over a greater share of policy fields (Schakel, 2010) are less likely to undertake autonomous development cooperation, we observe no change in our main results (Table B17).

Exploiting temporal variation
One limitation of cross-sectional analysis is that it ignores temporal order. Although regional characteristics such as constitutional autonomy, distinct language, and regionalist parties take time to develop and therefore are plausibly exogenous with respect to the establishment of international development cooperation, we would like to test our argument using temporal variation. Hence, we construct a time-series cross-section dataset of all regions in 1949-2019. Each region enters the dataset in the year it was established. The variable AGENCY ESTABLISHED is a binary indicator for the launch of the aid agency, after which the region leaves the dataset. For regions without an agency, the data are right-censored, and the region is carried through until the end of the sample period. As alternative outcome, we consider the establishment of a regional AID PROGRAM where the data does not record establishment of an aid agency. Our independent variables are the time-varying RAI (Hooghe et al., 2016) and the regionalist party dummy, now capturing the annual observations in which such party is active (Rode et al., 2018). Since substantive control variables are unavailable for long timeframes, we only include cubic splines modeling duration dependence, a global time trend, and country-fixed effects, eventually replaced by region-fixed effects for coarser identification.
The results corroborate our earlier cross-section findings. Table 3 shows that a region is significantly more likely to establish an aid agency after increases in regional autonomy, as proxied by the RAI (p < 0.01). This result holds robustly for alternative outcomes. The appendix shows that this result also holds using three-year periods instead of annual data (Table  B18). Both are plausible timeframes for when to expect regional autonomy to translate into greater institutional capacity for development cooperation. There is no consistent effect of regionalist parties on subnational development cooperation. A positively significant effect only holds for the withinregion model and annual data (p < 0.05), but not for the three-year period data. Notes: AID PROGRAM expands the observations by using the stated start dates of any aid programs where information on aid agencies were unavailable. Significance levels: * 10%, **5%, *** 1%.
A glance at the data shows that about ten cases conform to a pattern whereby increases in regional autonomy or the arrival of regionalist parties are followed by the establishment of a regional aid agency. For example, as the Spanish regions gained more autonomy in two steps in 1982 and 1987, many of them established aid agencies, including the Basque Community (1988), Galicia (1992), Aragon (1994), andCatalonia (2003). In Belgium, Flanders increased its RAI (1989) and established an aid agency (2001). In Italy, South Tyrol increased its RAI gradually over 1989-1991, which culminated in the establishment of an aid agency in 1991. 19

Discussion and conclusion
This article examined the determinants of subnational development cooperation as an increasingly important trend in multi-level polities. Using the 'subnational donor governance dataset' covering 195 politically relevant European regions (Reinsberg & Dellepiane, 2021), we found that regions are more likely to undertake development cooperation under circumstances of greater constitutional authority, cultural-linguistic distinctiveness, and where regionalist parties are present. These conditions help regions bolster claims for regional autonomy, by enhancing their international legitimacy and fostering a regional identity. Our findings suggest that region-building is the driving force behind subnational development cooperation. Related claims come in various forms and do not necessarily involve full-blown secessionism from the central state, as a narrow focus on well-known cases would suggest. In fact, our research suggests that intense nation-building around proto-diplomacy momentsor cultural paradiplomacy a la Québec (Paquin, 2018) may be the exception rather than the rule, considering that over 70 regions engage in development cooperation.
We note three limitations of our research before discussing some broader implications and avenues for future research. First, a full account of the motives of regional development cooperation remains beyond the scope of our article. Our work should thus be read in complement to small-N qualitative work. Given our goal of explaining large-N patterns, our conclusions are based on proxy measures, and we do not claim to have established causality, despite careful modeling choices. 20 Neither do we claim that our explanation is exhaustive. Second, given that some key measures are missing outside a European context, our dataset excludes subnational actors from non-European countries, such as Australia, Canada, and Japan. Whether our results hold beyond Europe or for municipalities as another group of subnational actors (Interview 9-10) could be addressed in future research. Third, opportunities remain to further unpack the dynamics of multi-level politics, specifically the practical obstacles to growing a regional aid program, such as public opinion (Brownlee, 2017).
Taking a paradiplomacy perspective, our research provides new insights to the vast literature on aid allocation. Most aid allocation studies have examined how donor countriesconceived as unitary actorsallocate aid budgets to different causes and among different recipients (Hoeffler & Outram, 2011;Reinsberg, 2015;Younas, 2008). In contrast, we perceive aid donors as non-unitary actors, but focus on subnational governments, not national-level bureaucracies as most studies in the 'organizational turn' (Arel-Bundock et al., 2015;Campbell, 2018;Carcelli, 2019). In addition, we extend the list of aid motives (Bermeo, 2017;Clist, 2011;Younas, 2008), emphasizing how development cooperation allows subnational governments to advance their ambitions for greater political autonomy. In line with some national-level research (Gulrajani & Swiss, 2019), our argument emphasizes the highly symbolic nature of becoming an aid donor, which endows regional governments with legitimacy as a 'state' internationally while fostering a common identity domestically based on distinctive aid policies.
Our goal is to consolidate a research program on subnational development cooperation as a dimension of sub-state foreign policy. Motivated by pioneering case studies, our novel large-N analysis has mapped out the scope of the phenomenon and its potential determinants. We now invite scholars to build on our study to engage in 'systematic case selection' (Seawright & Gerring, 2008) or 'nested analysis' (Lieberman, 2005). We also encourage theoretically-informed 'paired comparisons' (Tarrow, 2010), within and across countries. In the spirit of 'working together' (Poteete et al., 2010), such integration of large-N analysis and case study research could provide a deeper understanding of at least two sets of issues.
One is a deeper understanding of motivations. Our study motivates a more nuanced exploration into how different forms of territorial claims (institutional, political, cultural) play out and indeed intersect in shaping substate development policy. 21 A glance at the international development strategies of sub-state governments unveils an increasingly consolidated bifurcationbetween a 'progressive' account based on solidarity and global citizenship and a more 'conservative' account around trade, security, and migration control. These two worlds of development mirror wider political divides. This motivates another puzzle for future research. The political and societal consensus around regional development cooperation is too often taken for granted. Yet, are regional development policies really aligned with public opinion? How strongly does civil society engage? How is policy formation affected by regional party competition? These issues deserve further attention.
Another area of future research pertains to the actual workings of subnational development cooperation. This includes subnational aid allocation and aid effectiveness and how they relate to policies at the national level. For example, a core claim among regional policymakers is that they 'do development differently'a shared belief that challenges dominant accounts around aid effectiveness and aid fragmentation (Kania, 2021). Indeed, regional donors do not face competing foreign policy imperatives. To the extent that national aid remains plagued by competing donor interests, regional policymakers have incentives to uphold more progressive norms of good donorship as a signaling device to their domestic audiences, which can further legitimize their engagement. However, lack of scale and limited capacity are common challenges for regional donors that may undermine aid effectiveness in the traditional sense. There is reason to believe that the shape of territorial politics affects aid effectiveness through its impact on the relations between regional, national, and supranational structures. While lack of systematic data currently prevents related rigorous tests, the expanding roles of subnational development actors will make this an increasingly important endeavor.
Notes potential lever of influence although actual influence differs considerably (Lecours & Moreno, 2003;Loughlin & Seiler, 1999;Tatham, 2015). 9. All information pertaining to aid governance is based on the cutoff date 1 December 2019. As detailed further below, other covariates may be from earlier years, depending on data availability. 10. The OECD/DAC continuously has improved data coverage on 'decentralized development cooperation' and conducted surveys on how regions engage as aid providers (OECD, 2005(OECD, , 2018(OECD, , 2019, but covers fewer regions. 11. We discard aid budgets because the data are patchy and the theoretical relevance of regional donors relates to their approach to aid governance rather than aid volumes. The appendix defines our key outcome variables in greater detail (Table A2). 12. We considered alternative measures of linguistic distinctiveness but found them to have fewer observations. For example, Fitjar (2010) enlists regions with distinct language and historical sovereignty, but country coverage is unclear. In the overlapping sample, the correlation was moderately high (ρ = 0.45). 13. An alternative indicator that we use in robustness checks draws on a list approach combining several sources (Donas & Beyers, 2013). Data capturing the strength of regional parties (not necessarily with separatist ambitions) is available only for 13 European countries (Massetti & Schakel, 2017) but inferences are limited because we do not know whether these parties were in government. 14. Their pairwise correlations are up to ρ = 0.5. Hence, if jointly included, we would expect their estimates to be imprecise, despite being individually significant. 15. Information on regional Brussels offices can be found on http://www.aalep.eu/ regional-offices-brussels (accessed 1 October 2020). 16. The empirical minimum is 1.9 and the maximum is 27. Figure B1 in the appendix plots the simulated probability of an aid website over the range of the RAI. 17. If we are willing to make further assumptions about what different outcomes reveal about donor motivations, we may infer regions with constitutional powers and a regionalist party are even more serious about their commitment to development cooperation because they establish their own agencies and devise their own programs. In contrast, regions that are (only) culturally distinct have increased preferences for an aid program but may lack both authority and a political party acting as institutional entrepreneur to do so. 18. In another robustness check, we verify that our results are not driven by sampling effects by dropping all countries without subnational variation in the dependent variable. This mimics the within-country regressions using fixed effects. For aid website (Table B9), aid agency (Table B10), and aid program (Table B11), we find significantly positive results for all our predictors, except for distinct language with respect to aid agency. The joint effect of these variables is strongly significant. 19. The variation in start years also helps discard the empowerment of European regions in the Maastricht treaty as a common cause. 20. While our choice of dummy variables for analysis is a concession to robustness, we could think of theoretically plausible relationships between the intensity of territorial claims and levels of institutionalization, as well as the direction of territorial claims and forms of institutionalization.
21. In this regard, the intersection between political and economic motivations is an interesting question in its own right.