Administrative capacity and the territorial effects of EU support to firms: a two-step analysis

ABSTRACT This paper investigates whether territorial characteristics and, in particular, regional administrative capacity influence the effects of European Union (EU) Cohesion Policy support to firms. A novel two-step methodology is applied. First, the effects of Cohesion Policy on employment growth of supported manufacturing firms are estimated separately for the regions of six different EU countries. Second, potential territorial factors influencing these effects are explored using meta-analysis techniques. The empirical results point to a significant relationship between firm-level policy effects and territorial capital, especially mixed-materiality assets, as well as administrative capacity as proxied by citizen engagement and administrative efficiency.


INTRODUCTION
Next to the European Union's (EU) Common Agricultural Policy (CAP), Cohesion Policy (CP) is one of the EU's two most important budget items, accounting for around one-third of the budget in the 2014-20 and 2021-27 programming periods.Given the alignment of EU expenditures with current challenges and priorities of the EU and its member states, a stronger focus on the European added value is becoming more important.Therefore, knowledge about the factors that determine the effective use of the financial assistance provided by the policy becomes even more essential.
The question of whether CP funds are spent effectively has been intensively investigated by scholars and policy evaluators (see Crescenzi & Giua, 2017;Pieńkowski & Berkowitz, 2015;Dall'erba & Fang, 2017, for literature reviews).The results turn out to be rather inconsistent due to different methodological approaches, different time periods, different country or regional samples, and different expenditure axes considered in the studies (Fratesi, 2016).This is not surprising, as the large set of CP measures and financial instruments pursue distinct broad priority themes.For instance, in the 2007-13 programming period covered by this study, these eight broad themes include 74 specific priority themes (Annex IV of Council Regulation (EC) No. 1083No. /2006) to be addressed through numerous measures and an even larger number and variety of projects.
Since 2007, every EU region has been a beneficiary of CP and receives funding from the European Structural and Investment Funds (ESIFs).Regional data provided by the Directorate-General for Regional and Urban Policy (DG REGIO) show that the allocation of the different funds to different thematic target dimensions varies considerably between regions and countries.On top of that, the analysis of individual project data for the 2007-13 programming period reveals strong regional differences in terms of the characteristics of co-funded projects and beneficiaries (Bachtrögler et al., 2019;Darvas et al., 2019).
The regional focus in the implementation of CP, such as the possibility to select appropriate projects based on regional needs and priorities, has been conceived as a strength following the conceptual and practical upheavals of place-based regional policy (Barca, 2009;Barca et al., 2012;Farole et al., 2011;Partridge et al., 2015) and, more recently, Smart Specialisation (Foray, 2015;Gianelle et al., 2019;McCann & Ortega-Argilés, 2015).However, a large stream of the literature on conditioning factors has found that the responsiveness of regions to CP, in terms of the effect on regional gross domestic product (GDP) per capita or employment growth, is highly heterogeneous (Crescenzi et al., 2017;Fratesi & Wishlade, 2017).Territorial, regional or national characteristics, including institutional factors such as administrative capacity, have been identified as crucial for different outcomes in different countries and regions (Milio, 2007;Farole et al., 2011;Bachtler et al., 2014;Rodríguez-Pose & Garcilazo, 2015;Surubaru, 2017;Rodríguez-Pose, 2020;Di Caro & Fratesi, 2022; see also Bachtler et al., 2016, for a review).
Attention to territorial issues in political strategies varies over time and between political parties (Basile, 2016), but the issue of territoriality is relevant for all EU policies (Medeiros, 2017a) and is expected to become even more important in the coming years.Regional disparities have increased in the years leading up to the Covid-19 pandemic (Camagni et al., 2020) and the pandemic had asymmetric effects both in terms of health and the economy (Bourdin et al., 2021;Ascani et al., 2021).Taking this into account, the recently agreed Territorial Agenda 2030 (EU, 2020) provides the framework for action to promote territorial cohesion in Europe.It is based on the shared recognition that development needs and opportunities vary across territories, while at the same time it requires cooperation and coordination between different places, as well as between different levels of government and societal groups, in order to address complex issues and boost the potential impact of policies.
Empirical evidence is needed to learn from the past and ensure that future policy initiatives will be effective, and this paper aims to contribute to this, overcoming some methodological limitations of previous studies.A large number of previous studies have analysed the determinants of CP effects at the regional level, but data available at the NUTS-2 level are usually not detailed enough to allow a comparison of the same policy in different contexts, which requires to distinguish between different types of policies and in particular different types of beneficiaries.In contrast, a micro-level approach allows the analysis of the same policy in different territorial contexts by comparing its effects in different regions, keeping constant the type of projects and beneficiaries, even if this entails the drawback of limiting the analysis to specific types of intervention.
By combining the regional and micro-level approach, this paper investigates the causes of the variation of the firm-level effects of CP across different countries and regions.The hypothesis to be tested is that these differences are related to territorial characteristics of the regions and, in particular, to administrative capacity.Regional administrative capacity may impact the effects of regional policy and consequently on regional economic development in several ways (Fratesi, 2023): with good administrative capacity, local authorities will be able to absorb the funds available; they will be able to design or select promising and successful projects; they will be able to assist beneficiaries in efficacious project implementation; project evaluation will be accurate and allow learning from the past when designing new policies.
In this paper, territorial characteristics are measured in a multidimensional way, following the conceptual framework of territorial capital (Camagni, 2009).With regard to regional administrative capacity, three measures used in previous literature are considered: civic engament (voter turnout in EU elections as a proxy of trust in government); efficiency in meeting the needs of businesses (the number of days needed to register a new company); and spending capacity on EU policies (the absorption rate).The details are provided in the methodological sections of the paper.
From a methodological point of view, the paper is among the first considering the effects of CP at the level of beneficiaries in different countries at the same time, and can therefore compare a standardized estimate of the regional success of CP.In a novel two-step procedure, and for a sample covering the NUTS-2 regions of six different EU member states, a standard micro-econometric approach is combined with methodological features often used in meta-analyses.The first step consists of a difference-in-difference estimation of the average treatment effect of financial assistance on supported manufacturing firms in the 2007-13 programming period.In the second step, these estimates serve as a measure of the success of CP in different regions, whose relationship with relevant regional and national characteristics is explored.To model territorial factors, principal component analysis (PCA) is applied to a set of relevant indicators identified using the concept of territorial capital.In order to learn more about the relevance of local administrative capacity for firms that benefit from CP, in terms of employment growth, this factor enters the empirical analysis separately.
The remainder of this paper is organized as follows.Section 2 provides a literature review on the effects of CP and their determinants.Section 3 illustrates the methodological framework and the multidimensional measurement of territorial determinants and policy typologies.Section 4 explains how territorial characteristics and administrative capacity are measured.The empirical results are presented in Section 5, while Section 6 concludes and provides policy recommendations.

CONTEXT MATTERS: DETERMINANTS OF HETEROGENEOUS REGIONAL POLICY EFFECTS
The extensive literature on the evaluation of CP does not reach consistent conclusions on the policy's success in generating economic growth or employment.While one major reason for the lack of consistency of study results is the use of different methodological approaches and data samples, there seems to be a consensus on the existence of contextual factors that interact with the policy and thereby influence its effect.
Part of the literature focuses on less developed regions and, by exploiting the 75% threshold of EU-average GDP per capita that defines eligibility for the Convergence objective, on average finds a positive effect of Structural Funds on income growth (e.g., Pellegrini et al., 2012).Further research has revealed that this effect turns out higher in regions with better institutional quality and human capital (Becker et al., 2013).
Many evaluation studies considering all EU regions confirm the assumption that the effect of CP strongly depends on the territorial context in which it is implemented.In some analyses (e.g., Mohl & Hagen, 2010), these local conditions have been broadly defined by the overall level of regional development (which is correlated with the intensity of EU support).Empirical results have shown that the return on CP actions is higher in more developed regions.Other studies address the same issue by modelling regional policy settings with specific local conditions (which are likely to be influenced by national characteristics as well) and territorial assets, which are expected to mediate the effect of CP.Ederveen et al. (2006), Rodríguez-Pose and Garcilazo (2015) and Crescenzi et al. (2016) point to the importance of institutional quality as a crucial factor in explaining CP effectiveness.Becker et al. (2013) include human capital and institutional quality in a region as mediating factors to measure the absorptive capacity of regions.Crescenzi and Giua (2020) conclude that the most socio-economically advanced regions are able to profit the most from CP, which is likely to be directly related to their high absorptive capacity.Further local characteristics investigated with respect to their influence on the differential effects of the policy include regional economic structure (Cappelen et al., 2003;Percoco, 2017), population density and regional settlement structure (Gagliardi & Percoco, 2017), the concept of territorial capital and local characteristics (Fratesi & Perucca, 2014, 2019;Sotiriou & Tsiapa 2015), as well as regional public investment and infrastructure stocks (Dall'erba & Fang, 2017).Country-specific characteristics have been shown to matter as well, among them national institutional quality, the degree of economic openness, or whether the country is governed by a federal or centralized government (De Freitas et al., 2003;Ederveen et al., 2006;Kemmerling & Bodestein, 2006).
Another approach that has gained importance in terms of policy relevance and complements previous approaches is territorial impact assessment (TIA).TIA was advocated in the European Spatial Development Perspective (Council of Ministers Responsible for Spatial Planning (CMSP), 1999) as a tool for multidimensional assessment of the potential effects of cohesion policies.This is particularly important as the multidimensional objective of territorial cohesion has replaced the simpler objective of reducing regional disparities in the EU, and has accompanied economic and social cohesion.According to Camagni (2006, p. 136), 'territorial cohesion may be seen as the territorial dimension of sustainability' and comprises three main componentsterritorial efficiency, territorial quality and territorial identity.
Based on these seminal contributions, several papers have developed methodologies to operationalize the TIA and ways to promote territorial cohesion, building on a broader conceptualization of territory as introduced in the TEQUILA model (Camagni, 2006).For instance, Zaucha et al. (2014) show that paying attention to the territorial context improves the effectiveness of policy implementation and therefore propose a set of territorial keys which could enable successful implementation of the EU2020 strategy, illustrating the case of Poland.Medeiros (2017a) examines the territorial impact of CP in Spain across four programming periods, going beyond the standard socio-economic and environmental indicators, which allows him to quantify the positive outcomes of the CP in several dimensions.Further examples of TIA applications can be found in Medeiros (2017b).
This approach allows helpful policy conclusions to be drawn.Medeiros and Rauhut (2020), for example, discuss the role of medium-sized cities as crucial anchors in achieving the policy goal of territorial cohesion and explain the opportunities and challenges of using 'territorial cohesion cities' as 'development centres in lagging regions' to achieve territorial cohesion at the national level.
According to the empirical evidence, the local context is not the only aspect that can promote or hamper the effects of CP.Another aspect concerns the goal of the policy.CP pursues several objectives: reducing economic and social disparities, which entails a focus on less developed regions; increasing regional competitiveness and employment in all European regions; and European territorial cooperation.For all these purposes, a wide range of activities are funded, some of which are aimed at the economic return on investments, while others are more focused on improvements in the social sphere.A pioneering study by Rodríguez-Pose and Fratesi (2004) provided empirical evidence on a differential effect of CP across different axes of intervention.They found that only funding aimed at fostering human capital generates significant and lasting benefits.Similarly, in an analysis focused on Italy, Percoco (2005) found that the best performing regions were those that allocated CP funds according to the hierarchy of marginal productivities in different policy areas.More recently, several studies have examined the concentration of funds on specific priority axes (Crescenzi et al., 2017) and the interaction of CP with the EU Common Agricultural Policy (CAP) or other national policies with heterogeneous regional effects (Crescenzi, 2009).While many studies use the amount of CP commitments or expenditure as explanatory factors for the policy's effect on, for example, regional income growth, Crescenzi et al. (2017) also provide evidence that funds targeted at regional needs seem to increase regional growth to a greater extent.
The two issues discussed above, namely (1) the specific objectives of CP funds spending and (2) in which territorial setting they are invested, are not independent for two reasons.First, the type of policy actions implemented in different regions is related to the territorial characteristics of the implementation environments (Dall'erba et al., 2009;Fratesi & Perucca, 2016).Lagging-behind regions allocate a higher share of funds to infrastructure provision, while more advanced regional economies focus more on issues such as promoting the productive environment and social policies (Fratesi & Perucca, 2016).Second, the mediating effect of territorial conditions in a particular policy environment may not be the same for all types of policy measures.In an empirical study on EU-12 member states, Fratesi and Perucca (2014) have shown that territorial characteristics of regions are likely to foster the effectiveness of CP actions only in certain fields of intervention.
This implies that a thorough assessment of the effects of CP actions and their comparison across regions and countries is difficult because (1) it requires the comparison of the same type of policy measures, which (2) are implemented in similar territorial contexts.Therefore, considering aggregate regional indicators for both CP expenditure (e.g., commitments per capita) and outcomes (e.g., change in GDP or employment) in empirical studies only provides an aggregate, average estimate of the effect of the EU intervention.The underlying heterogeneity resulting from the different effects of the various measures in regions of different types, typically remains hidden.
Using a multi-country firm-level dataset with microlevel information on the distribution of CP funds in the multi-annual financial framework (MFF) 2007-13) Bachtrögler et al. ( 2020) have investigated the firm-level effects of CP in different countries and regions.Considering a sample of seven EU member states, the empirical results indicate that financial support for manufacturing firms has generated additional growth in employment and value added.The effect on firm productivity growth also turns out to be significant, though much smaller.
Although the micro-level analysis takes into account potential heterogeneities in policy actionsas it only allows assessment of support to manufacturing firms for a cross-regional and cross-national comparison of policy effectsthe regional response to the policy, in terms of the average effect of financial assistance on supported firms in a region, varies significantly both across countries and, within countries, between firms in different types of regions.
In this paper we extend the possibilities of using the results of this micro-based approach to further address the issue of heterogeneous policy effects in NUTS-2 regions.Again, in order to keep the type of policy constant, only supported manufacturing firms are considered, assuming that firms of the same industry apply for and make use of CP funding for projects with similar objectives, that is, increasing production or fostering productivity.Few studies have used information on the micro-level distribution of funds within regions to estimate beneficiary-level effects, and they usually did so for only one country (Bondonio & Greenbaum, 2006, for Italy;Bernini & Pellegrini, 2011, for Italy;Benkovskis et al., 2018, for Latvia).In general, their results show either no or negligible effects of CP financing on firm productivity, while having a significant effect on employment growth.
This study explores the reasons for the heterogeneous effects of CP in different regions by looking at regional characteristics in terms of territorial capital and, especially, in terms of administrative capacity.With respect to the literature on the effects of administrative capacity on regional growth, we are thus able to adopt a micro-based approach that allows us to single out only one policy intervention (support to manufacturing firms) which is the most homogeneous in different regions.We are able to exploit the potential of a multi-country dataset and link it to macro-level territorial characteristics of different regions and countries.This also requires the application of a novel methodological procedure that integrates micro-and macro-approaches in a two-stage analysis, as described in the next section.

Estimation of CP effects at the firm level
The aim of the first step of the empirical analysis is to estimate the effect of receiving CP funding on employment growth in supported manufacturing firms located in different regions.This variable is of interest for the assessment of CP because, as discussed above, one of the main difficulties in evaluating EU regional policy (at the regional level) is its multifaceted nature: CP pursues different objectives (such as job creation, competitiveness and social inclusion) through actions in different priority areas (e.g., productive environment, human capital and infrastructure) and by supporting different types of actors (e.g., firms, public institutions, etc.).Consequently, the limitation of a general assessment of the programme at a macro-(regional) level is that this variety of measures is treated as an undifferentiated aggregate.As enhancing employment growth in European regions is a central objective of CP, the additional employment growth induced by the policy in supported manufacturing firms is considered in this analysis.
Specifically, the methodological strategy of this study follows Bachtrögler et al. (2020) by focusing on similar and therefore comparable policy actions and considering only financial support to manufacturing firms (NACE Rev. 2) in the 2007-13 programming period.To identify the supported firms, beneficiary names provided in lists of beneficiaries, published by managing authorities by the second half of 2016 (see Bachtrögler et al., 2019, for a detailed description of the data and its source), have been linked with the AMADEUS business database provided by Bureau van Dijk.This makes it possible to collect information on the industry in which the beneficiaries operate, as well as on other firm characteristics such as the number of employees, value added or fixed assets.In order to form a control group of firms not identified as assisted by CP in 2007-13 and located in the same region, which is necessary to estimate the effect of the policy, these data are needed before and after receiving the support for all manufacturing firms (for which business sheet data are available in the AMADEUS database).Given the quality of the available data on beneficiary and non-beneficiary firms, this analysis is conducted for a sample of countries consisting of the Czech Republic, France, Italy, Portugal, Slovakia and Spain.These countries do not cover the whole of the EU, but they represent countries with different levels of per capita GDP, as well as old (EU-15) and new member states.Moreover, by design, countries consisting of only one region had to be excluded in order to analyse regional aspects.In total, the sample includes more than 125,000 firms, of which 17,421 were treated.As the number of treated manufacturing firms per NUTS-2 region is limited, we consider the NUTS-1 regional level for France. 1  The effect of financial assistance from the CP is estimated by comparing employment growth in supported firms between 2006 (pre-treatment, if not available 2007) and 2014 (if not available 2015 or 2016) with that of non-supported firms in the control group.Employment growth is modelled by the log difference between the postand pre-treatment value in the number of employees. 2 For the treated (supported) manufacturing firms, the outcome in absence of the treatment is not observed.Therefore, propensity score matching (PSM) (Rosenbaum & Rubin, 1983) is applied to create an appropriate control group of manufacturing firms that are similar to the assisted firms and located in the same region.Similarity is modelled based on a set of pre-treatment characteristics measured in 2006 (if available, otherwise in 2007; in linear and quadratic terms) reflecting relevant firm, sector and region level characteristics (Bondonio & Greenbaum, 2006).Firm characteristics include the number of employees (log), firm age (log), fixed assets per employee (log) as a measure of capital intensity in production, and current assets divided by current liabilities measuring liquidity (log).Taking these firm characteristics into account, the probability of being treated, that is, the propensity score, w, is estimated.
Based on the propensity score, treated units are matched with similar untreated units.Difference-indifference estimation is applied to compare the average employment growth of treated and non-treated firms and to estimate the average treatment effect on the treated group (ATT) (equation 1) per region: 3 where T stands for treatment, Y(1) represents the outcome variable (employment growth) for treated firms (T ¼ 1) and Y(0) the one for the control group (T ¼ 0).Standard errors of the ATT are bootstrapped (1000 replications).
To sum it up, this first step of the analysis provides an indicator of the differential effect of CP across regions.

Exploring factors that determine firm-level policy effects
In the second step of the analysis, meta-analysis techniques are applied to investigate whether the differential effects found in the first step are related to regional characteristics.More specifically, first, meta-analysis controls are considered to take into account regional sample sizes in the first-stage analysis, and second, PCA is applied to a large set of territorial characteristics in order to avoid multicollinearity in the analysis of the determinants of the regional effects of CP support to manufacturing firms.
Meta-analyses are not new to regional science.Abreu et al. (2005) use meta-analysis to analyse the large number of studies that examined growth and, in particular, betaconvergence, and look for consistency with respect to the 2% convergence rate, which was first found by Barro andSala-i-Martin (1991, 1992).The authors note that 'apart from information on the effect sizes, it is also desirable for the meta-analysis to take into account the fact that the standard errors of the respective estimates are different, among other things because the sample sizes of the primary studies differ' (Barro and Sala-i-Martin, 1992, p. 369).Hoogstra et al. (2017) conducted a meta-analysis of studies that empirically investigate the question of whether people follow jobs or jobs follow people.Using a multinomial logistic regression estimation, they analyse four types of outcomes based on the significance level of the results of the studies considered.They find substantial differences in effects, with more studies pointing out that jobs follow people but, at the same time, high variability due to the geographical scope of the studies and other characteristics.
Thematically related to this paper, Dall'erba and Fang (2017) perform a meta-analysis of studies on the effect of structural funds on regional growth.They find a large heterogeneity stemming not only from characteristics of the studies, but also from the period of investigation.This suggests that there is a learning effect in more recent studies.Controlling for human capital and institutional quality also yields significantly different results.In terms of methodology, several weighted regression models are used, namely a fixed-effects model, a mixed-effects model and a hierarchical model.The use of different techniques leading to the same results underpins the findings.
Compared with the papers mentioned above, the metaanalysis of the present study has a relevant advantage.All first-stage difference-in-difference estimates are based on the same firm characteristics as well as control and dependent variables, unlike in standard meta-analysis where the coefficients come from different regression models.Hence, the differences in the estimated coefficients depend on the region to which the data correspond.Considering the sample size of each region, and the number of observations in the treatment and control groups, thus allows for a comparative analysis of the ATT across regions.
When interpreting the results of this study, some methodological limitations should be kept in mind.First, even though previous analyses based on firm data usually include only one country, the sample consists of a limited number of countries, a sample of the EU.Still, the countries include richer and poorer countries as well as old and new member states.Furthermore, it has to be noted that the dependent variable of the second step, the ATT, is an estimated coefficient with standard errors.In order to capture the statistical significance of the ATT in the regression of the second step, it is set equal to the estimated coefficient if it is significant and set to zero otherwise.
The two-step procedure applied in this paper allows us to address two important issues in the empirical assessment of the 'net' effect of EU regional policy, that is, to disentangle the type of policy action (in this case, only grants to manufacturing firms are considered) and the territorial context in which it is implemented.
The aim of this paper is therefore to explain the geographical variation of this net effect: Why did assisted manufacturing firms in some regions perform better than assisted firms in other regions when the same type of policy was implemented?We assume that this policy effect depends on territorial characteristics of EU regions as well as on national factors and the way CP funds are distributed within regions.Equation (2) tests whether this hypothesis holds: Policy effects r = f (territorial characteristics r ; administrative capacity r ;policy characteristics r ; national controls r ; meta − analysis controls r ) (2) The dependent variable reflects the estimated average treatment effect of the policy on supported firms (ATT) in a NUTS-2 region 4 which is estimated in the first step.It is set to zero when the effect is not statistically significant considering the 90% confidence interval. 5 As ATTs are not normally distributed across regions (Figure 1), the analysis of their potential determinants is not possible with a standard ordinary least squares (OLS) regression, but requires a general linear model and maximum likelihood estimation.In the baseline estimation, a Poisson regression with robust standard errors is applied, and the model fit is confirmed by a goodness-offit chi-squared test.As a more general approach, although the data is neither characterized by over-dispersion nor excessive zeros, negative binomial regression is applied as a robustness check.The potential issue of collinearity was addressed by using PCA (whose components are uncorrelated by definition) and by inserting the explanatory variables together and separately.
The empirical model to be tested assumes that policy effects are a function of these territorial characteristics and a set of controls.Equation (2) defines five groups of regressors.
The territorial characteristics of each region (r) define the types and combinations of territorial assets of the 62 regions included in the sample, measured in 2007.In the context of this paper, the territorial capital framework (Camagni, 2009) is used, which has already been applied in the context of policy assessment (Fratesi & Perucca, 2019).This approach aims to measure regional sources of economic development and classifies territorial assets according to their degree of materiality and rivalry.It is therefore particularly well suited for studying the relationship between the outcomes of public policies and the characteristics of their implementation environment, as it conveys direct policy implications.Indeed, territorial assets with different levels of materiality and rivalry follow different mechanisms of accumulation and depletion and therefore require different policy interventions.For instance, the proliferation of assets with high levels of materiality and rivalry, such as private investment, requires different incentives and policies than intangible and nonrival resources, such as relational capital. 6Section 4 describes in more detail the approach to measuring these territorial characteristics used in this paper.
Among the characteristics of the regions, administrative capacity and the quality of governance are at the centre of our analysis.Institutional quality itself can be interpreted as a dimension of territorial capital, characterized by both low rivalry and materiality.Therefore, according to Camagni's (2009) framework, it is a public good from which the whole regional community benefits.We assume that, in line with previous literature (Rodríguez-Pose & Garcilazo, 2015) and the mechanisms outlined in the introduction, the effect of CP is enhanced by administrative capacity.To test this hypothesis, this territorial feature is separated from the other elements of territorial capital.Within the literature analysing the relationship between policy effects and institutions, this work stands out in two ways.First, the analysis focuses on a specific type of policy intervention, thus reducing the influence of other (observed and unobservable) determinants of policy outcomes.Second, three dimensions of administrative capacity are considered in this analysis, each capturing a specific mechanism through which institutions can influence the effect of EU regional policy.The first dimension relates to citizen participation and engagement in the political and institutional spheres.Civic engagement, in turn, is positively related to citizens' trust in institutions (Warren et al., 2014).The second dimension concerns the transaction and bargaining costs that countries' bureaucratic systems impose on businesses.A broad literature has shown how these constraints affect firms' investment and financing decisions (Schaller, 1993).From a policy evaluation perspective, especially in the case of productive environment policies such as in the present study, bureaucratic and administrative complexity is expected to negatively affect policy outcomes by imposing higher transaction costs on firms.Finally, the third dimension of administrative capacity relates directly to the efficiency of institutions in implementing CP measures.The successful allocation and use of funds allocated to each region reflects this aspect, that is, the skills and competences of local administrators in managing EU regional policy.
The policy characteristics capture the overall regional distribution of CP funding within a region.As discussed in Section 2, EU regional policy covers a wide range of policy areas, from transport to energy to business support.The policy assessment conducted in the first stage of this analysis focuses on a narrow area of intervention (i.e., support granted to manufacturing firms) but we cannot rule out a priori the possibility that there may be spillover effects from co-financed activities in other areas.Therefore, we consider whether a region is classified as a less developed region and therefore receives funding (a relatively high amount of funding as compared to other regions) under the Convergence objective.As the allocation and management of EU funds varies from country to country, in addition these differences, together with unobserved country characteristics, are accounted for by considering national controls in the form of country dummies.
Since the dependent variable is the outcome of the PSM estimation of the effects of CP at the firm level, some meta-analysis controls are included in equation ( 2).They serve to control for differences in the size of regional samples used to estimate the ATT.The variables considered in this context are the number of treated manufacturing firms and their share in the sum of treated and nontreated firms in the control group.

MULTIDIMENSIONAL MEASUREMENT OF TERRITORIAL DETERMINANTS AND ADMINISTRATIVE CAPACITY
To measure the territorial determinants of policy effects, we rely on a large set of different types of territorial characteristics derived from the territorial capital framework (Camagni, 2009).It is assumed that the effects of CP are not influenced by individual elements, but rather by the simultaneous occurrence of certain combinations of these elements.Therefore, in order to be able to consider a wide range of different aspects of territorial capital and, at the same time, to avoid multicollinearity in the regression analysis, a PCA is performed.Here, the seven elements of territorial capital, as listed in Table 1, are bundled into three significant, uncorrelated components.
Please note that institutions and administrative capacity represent a particular form of the territorial capital of regions.Since this paper focuses mainly on the relationship between administrative capacity and CP effects in a region, this element is treated separately from the other regional characteristics.
Table 1 reports a list of the variables used, their interpretation, the respective type of territorial capital, the description of the empirical measurement and the data source. 7These measurements cover almost all possible combinations of high/intermediate/low levels of materiality and rivalry underlying the typology introduced in the concept of territorial capital, and they are fully consistent with the values used in previous studies. 8 Figure 2 shows the results of the PCA for the seven elements of territorial capital. 9It allows a rather straightforward interpretation of the three components.The first component (component 1) is called IM (Intermediate Materiality), as it is characterized by the highest level of territorial capital elements with an intermediate degree of materiality, that is, agglomeration economies and relational services.The second one (component 2), called high materiality (HM), is characterized by a high endowment of territorial capital with a high degree of materiality, that is, tangible public and private capital as well as collective goods.Finally, the last group (component 3) is called low materiality (LM) and is characterized by relatively large endowments of intangible elements of territorial capital.This classification allows us to identify the different combinations of territorial capital that characterize the regions in our sample.Taken together, the three components account for more than 70% of the variance (see Appendix A in the supplemental data online for more details).Therefore, the three groups identified by the PCA serve as a measure of territorial characteristics in equation ( 2).
Administrative capacity and institutional quality are empirically measured by three variables that capture the different dimensions defined in section 3.2 (Table 1).
The first dimension, that is, citizens' participation and engagement in the political and institutional sphere, is measured by voter turnout in the 2009 elections to the EU Parliament.Voter turnout is often used as an indicator of civic engagement (Putnam, 1993), which is then linked to trust in institutions.The simultaneity of the EU elections allows for a comparison of voting behaviour in different countries and regions.Second, transaction and bargaining costs are proxied by the number of days it takes to open a business.This variable is defined at the country level as it comes from World Bank data (Table 1).Lastly, the effectiveness of institutions in implementing CP is captured by the regional absorption rate, that is, the share of the total amount committed in the EU budget paid to the region that is assumed to be related to government effectiveness (Tosun, 2014;Incaltarau et al., 2020).
Finally, two further independent variables are considered that are indirectly related to the presence and structure of institutions.The first is a dummy variable that equals one if the country's capital is located there and zero otherwise.Apart from urbanization economies (which are controlled for by including population density in the elements of territorial capital) (Table 2), capital cities typically host the high-level administrative and institutional functions.The second variable is the structure of the national political systemmonocentric or polycentricwhich is expected to influence how CP is implemented and whether the implicit objectives it pursues in the country favour growth throughout the country or whether, on the contrary, there is a concentration of resources and economic growth in the centre.Monocentric countries within the sample are the Czech Republic, France and Slovakia.

THE TERRITORIAL FACTORS SHAPING POLICY EFFECTS
The results of the first step stem from the PSM estimation of the effect of CP support on employment growth of manufacturing firms in each region of the sample. 10 These results then form the dependent variable of the second step, where the presence of a positive and significant effect is explained by testing it against a number of variables, as indicated in equation ( 2).The results of the second-step estimations are reported in Table 2. 11  The first group of coefficients reported in Table 2 (model 1) represents the territorial characteristics that were condensed using PCA.The only component that is significantly related to the effect of CP on employment growth in supported manufacturing firms is the first component, which is called IM.This implies that in regions with the highest degree of agglomeration economies and relational private services (Table 1), CP is more successful than elsewhere.The presence of urbanized areas together with the high-level institutional functions typically located in these regions thus appears to favour the effect of CP on businesses.By contrast, the other two components are not statistically significant, which means that neither the most tangible elements of territorial capital (i.e., private investment and public infrastructures) nor the intangible ones (i.e., human and relational capital) seem to enhance the effect of CP support to manufacturing firms.For this policy, results therefore suggest that the most complementary assets are those with mixed materiality.However, it may be that some of the territorial assets that have not been shown to play a significant role in fostering the success of this particular policy intervention, may be relevant in other policy fields.
The third column of Table 2 (model 2) reports the coefficients for the additionally considered control variables derived from the meta-analysis literature.The dependent variable appears to be generally positively and significantly associated with the number of treated firms in the sample.The share of treated among all firms in the sample also appears to be positively related to the effects of the policy at the firm level in general.Model (2) also introduces a set of country dummies to control for the effects of unobserved country-specific characteristics.The reason for introducing this set of dummies is that the differential effect of CP may have been influenced by national processes.
Model (3) in Table 2 adds a dummy variable equal to one for convergence regions as an indicator for the relative amount of CP funding received in the region.The corresponding coefficient is positive and significant in most specifications, meaning that in convergence regions (formerly called 'Objective 1 regions'), where the level of CP funding is relatively high, the effects of CP firm support are higher.This could imply that the supported firms in convergence regions received larger amounts of money, that the firm support was more complementary with other policy measures, or that firms in convergence regions were more in need of support.Although it is not possible to determine the actual cause of this finding, it does have a positive implication because, as Bachtroegler et al. (2020) show, there does not seem to be an obvious trade-off between equity and cohesion: firms in economically weaker regions that need more support are also those that are more positively affected by CP support.
The next columns of Table 2 further consider the variables that capture different aspects of administrative capacity and regional institutions.Initially, these variables are introduced separately (models 4-7) to account for possible multicollinearity.Voter turnout (model 4) is positively associated with the CP effect.In other words, the policy effect is significantly higher in regions with a higher percentage of eligible voters participating in EU elections.This suggests that civic engagement and trust in institutions promote the effects of EU regional policy, which is consistent with previous literature (Crescenzi et al., 2016).While previous literature on the relationship between institutional quality and CP effects has mainly focused on public infrastructure policies, which usually require significant and direct involvement of public bodies, the result reported here is interesting because it suggests that institutions also play a role in policies dealing with Administrative capacity and the territorial effects of EU support to firms: a two-step analysis The second proxy for administrative capacity and institutions, the complexity and efficiency of the bureaucratic apparatus (i.e., the number of days needed to open a business), is introduced in model 5. Since this characteristic is measured at the country level, country dummies are skipped in this regression.The coefficient of this variable is negative and statistically significant, which is consistent with our expectations.The longer the time needed to process the administrative procedures, the smaller the CP effect.
Moreover, it turns out that the third proxy, the regional absorption rate (model 6), is not significantly associated with policy effects.The ability of a region for using the EU funds committed to them does not necessarily lead to higher effects of support for manufacturing businesses.
Furthermore, interestingly, the region where the capital is located appears to experience higher CP effects (model 7).It is worth noting that the component of territorial capital associated with intensive agglomeration economies (component 1, IM) maintains its statistical significance.This suggests that, apart from the urbanization economies emanating from large cities, the functions typically located in capital cities promote the effects of regional policy.
Model 8 includes all the administrative capacity variables as well as the capital city dummy, except for the absorption rate which was not significant.Their association with CP effects is consistent with that presented above, still, the level of significance of the coefficient capturing the days needed to open a business diminishes with the inclusion of the capital city dummy.
As additional control, the next column (model 9) introduces the dummy variable for the monocentric countries that may encounter difficulties in implementing policies designed for a polycentric context.The coefficient of this variable is not statistically significant, suggesting that, at least for the specific type of policy considered in this study, the structure of the country is not directly related to the effect of CP.Since this variable may play a role for the influence of the location of the capital city on regional policy effects, the last column of Table 2 (model 10) shows that the coefficient capturing the relationship with the monocentric structure of the country does not interact significantly with the dummy variable for the capital city.

CONCLUSIONS
This paper has addressed the question of whether EU CP has different effects in different regions and whether the territorial context plays a role.Among the territorial characteristics of places, the paper focused on administrative capacity, as it is expected to improve the effects of policies through higher absorption of funding available, better selection of projects, better implementation and better learning from implementation (Fratesi, 2023).This is also very relevant in view of the 2021-27 programming period, when thematic objectives are streamlined and EU regional policy will focus on five macro-objectives.Unlike in the programming period 2014-20, where there was a separate thematic objective on administrative capacity (TO11), this objective will be embedded as a horizontal priority in the other objectives.
As CP is very complex and involves many objectives that may vary from country to country and from region to region, this paper adopted a novel micro-based approach that allows to single out a specific policy aspect that should be similar across regions by focusing on the effects of CP support on employment growth of supported manufacturing firms.
In this way, it was possible to analyse the determinants of the effects of CP at the territorial level and avoid the simplification of most macro-approaches, which consider the general impact of overall CP on regional growth.At the same time, a novel dataset (Bachtrögler et al., 2019) allows this analysis to be conducted in a multi-country setting for a sample of EU member states, which is new in the literature on the effects of CP at the firm level.
The methodology is new and comprises two steps.The empirical results of the PSM conducted in the first stage show that CP support to manufacturing firms has different effects in different locations.Therefore, controlling for the sample size of the underlying estimates, the second step examined whether territorial capital, policy characteristics, regional and national factors influence the effects of the same policy measure in different regions, focusing on different proxies for administrative capacity.The results first show that territorial capital is a crucial determinant of the policy's effects.In particular, territorial capital assets with mixed materiality (in particular agglomeration economies and relational services) seem to have the highest complementarity with the possibility of more effective business support.
Furthermore, the findings of the paper indicate that regional administrative capacity, proxied by different indicators, is related to the ability to implement policies effectively.It turns out that two indicators tend to be significantly associated with the effects of CP support for manufacturing firms, one measured at the regional level and the other at the national level: voter turnout in EU elections, which is a measure of citizens' trust in national and EU institutions, and the time taken to start a business, which is a measure of the efficiency of administrations in providing their services to businesses.
This suggests that the quality of governance is not only important as a mediator between regional growth and the macroeconomic effects of CP (e.g., Rodríguez-Pose & Garcilazo, 2015), but also when regions implement measures to support local manufacturing firms, as these measures are more effective in a better institutional environment.
Furthermore, the empirical analysis indicates that the effects of CP support on employment growth in supported manufacturing firms are higher in capital regions, where the presence of high-level institutional functions implies complementary assets that contribute to the success of the policy.In addition, the effects of CP support for manufacturing firms are found to be higher, and this is a positive message, for lagging regions where firms are more in need of support, that is, the policy appears to be more effective where there is greater need for it, which implies no trade-off between equity and efficiency.These results, should similar data become available, could be good a starting point for assessing other policy initiatives and countries, also complemented by multidimensional approaches such as TIA.
Two policy implications emerge from this analysis.The first relies on the finding that territorial conditions interact with the policy effects, even when considering a specific policy, as in this case support for manufacturing firms.In particular, a higher level of territorial capital with a medium level of materiality appears to be promoting policy success.This type of territorial capital includes assets, for example, agglomeration economics, whose endowments cannot be enhanced directly by targeted measures and in the short term (as it may be more likely the case for territorial assets with high or low materiality, such as infrastructure or human capital) but that require more complex policy action.
The second policy implication is that administrative capacity matters, also for the success of grants to firms.The reason is that regions with good institutions are likely to be better able to design calls for proposals, identify projects and select firms which are likely to boost employment growth.As a consequence, all the efforts that the EU has made in the past to improve regional administrative capacity are worthwhile and should be maintained, even if this analysis does not allow us to measure the extent to which capacity-building measures have been able to actually improve capacity.Therefore, additional analysis is needed to identify the best ways to improve administrative capacity and institutional quality.
Finally, it should be noted that a purely quantitative approach to measuring the relationship between administrative capacity and policy effects at the firm level has the limitation that it ignores further regional specificities as well as unobservable characteristics, such as skills, experience and competencies of the public agencies and institutions involved in the policy process.Examining these aspects would allow the underpinning of the results discussed above.For this reason, future (qualitative) research should complement the results of quantitative empirical studies such as the present one in order to gain further insights into the administrative characteristics that are primarily responsible for the positive relationship between policy effects and administrative capacity.

Figure 1 .
Figure 1.Distribution of the average treatment effect on the treated firms (ATTs).Note: Estimated ATTs are set to zero if not statistically significant.

Figure 2 .
Figure 2. Principal component analysis (PCA) scores for the territorial characteristics of the regions.

Table 1 .
The territorial characteristics of regions: an empirical measurement.
Note: ISCED, International Standard Classification of Education.