1 Introduction

Given the financial demands of the EU cohesion policy, it is no surprise that the use and effectiveness of the EU Structural Funds are up for debate and research. The most frequently investigated issue by far is the Funds’ effect on GDP growth or employment. Less common but still well-represented are papers that establish the conditions under which EU transfers produce desired outcomes. Studies looking at what happens when the inflow of EU money is sharply reduced or uncovering unintended side-effects of financial assistance to lagging regions remain scarce.

This paper examines whether the receipt of the EU Structural Funds affects trust or distrust in politicians. Access to EU funding enables local and regional governments to provide otherwise unaffordable goods and services. But it also offers opportunities for misappropriation by political representatives. Implementation and administration of EU co-financed projects may showcase politicians’ management skills as well as reveal their incapacities. By influencing citizens’ perceptions of politicians’ competence, benevolence and integrity, the EU transfers potentially bear upon citizens’ trust or distrust in politicians.

Trust in politicians increases citizens’ willingness to comply with the government’s regulations and policies. Should the inflow of the EU Structural Funds curtail this willingness, chances of the resources being put to effective use would be reduced. If, on the other hand, EU transfers strengthen trust in politicians, then the regions whose funding is about to dry up face not only the challenge of securing new resources for investment but also need to prepare to deal with attenuating trust in politicians and increasingly cumbersome governance that comes along with lower political trust.

When estimating the EU Structural Funds’ effect on trust in politicians, one must take care to avoid biased estimates due to possible mutual causality between trust in politicians and the receipt of transfers. To the extent that citizens’ trust in politicians reflects actual quality of government, poorly-managed regions are likely to exhibit low trust in politicians s well as deprived economic conditions, which would qualify them to receive large amounts of resources from the Funds. A related question is the extent to which any unintended effect of the EU Structural Funds on trust in politicians may be due to their intended effects on economic convergence or institutional improvements.

This study takes advantage of two thresholds, set in terms of a percentage of average EU per-capita GDP, which determine regions’ eligibility for EU funding under the convergence objective or for transitional phasing-out support. It employs a quasi-experimental strategy—known as regression discontinuity design—to identify the causal effect of EU Structural Funds on trust and distrust in politicians. The estimation is performed with data for NUTS 2 regions of the ‘old’ EU-15 Member States and the resources disbursed within the 2007–13 programming period.

2 Overview of related literature

Full consensus has not yet emerged regarding the effects and effectiveness of EU Structural Funds. Nevertheless, a mainstream view appears to be forming around a positive effect on economic outcomes in the recipient regions, typically faster per-capita GDP growth (Becker et al. 2013; Pellegrini et al. 2012; Bachtrögler 2016). The effect, however, appears temporal (Di Cataldo 2017; Barone et al. 2016), conditional upon good institutional quality and sufficient human capital (Becker et al. 2013; Rodríguez-Pose and Garcilazo 2015; Bachtrögler 2016;), and characterised by decreasing marginal benefits (Bachtrögler 2016).

While reducing regional disparities through faster economic growth is an intended effect of the EU Structural Funds, regions that receive large amounts of extra resources from higher levels of government may experience unintended, undesirable side-effects of this support. Examples include worsening of the quality of political institutions (Brollo et al. 2013) and decreased social trust (Accetturo et al. 2014). The common denominator in these cases is the misuse of the incoming funds.

Brollo et al. (2013) show how extra government revenue worsens the functioning of political institutions in Brazil. An increase in federal transfers intensifies the corruption engagement of the incumbent mayor (measured by a random audit program) and lowers the quality of his challengers (measured by years of schooling and private sector occupation). Extra resources create more room for corruption, the possibility of high rents attracts lower-quality challengers, and the two combined allow the incumbent mayor to engage in more corruption.

The authors caution that the Brazilian institutional environment is fragile, and their results need not apply to more institutionally sound settings. Yet, it is not uncommon for the developed countries to intentionally channel resources toward struggling areas (Brollo et al. 2013). This is indeed the case for the EU Structural Funds, which target lagging regions with the aim of helping them catch up with the more prosperous ones.

Accetturo et al. (2014) find that financial support from the EU Structural Funds reduces regional endowments of social trust. They rationalise this effect with a theoretical model in which local governments receive transfers from higher levels and allocate it either to the provision of public goods and services or leave it up for grabs by selfish, uncivic individuals. The incentive for grabbing (at least some of) the transfers is higher when the provision of local public goods is inefficient. Thus, lower social trust is a consequence of the increased incentive for uncivic behaviour provided by the transfers. The unintended effect is softened, in some cases eliminated, by effective provision of public goods.

A more recent and still ongoing stream of research has focused on the capacity of the EU Structural Funds to mitigate Euroscepticism and support for anti-EU political representatives (Bachtrögler and Oberhofer 2018; Borin et al. 2020; Crescenzi et al. 2020).

Borin et al. (2020) find that EU transfers mitigate Eurosceptic attitudes and voter’s support for anti-EU parties in regions that have been long-term recipients of the EU Structural Funds. The resources do not change citizens’ fundamental beliefs about the economic integration the EU represents. Rather, they seem to reduce the likelihood that citizens pick the EU as the target to blame for any socioeconomic discontents they may be experiencing. In their analysis, Borin et al. (2020) explicitly abstract from any economic effects generated by the EU Structural Funds. By contrast, Crescenzi et al. (2020) and Bachtrögler and Oberhofer (2018) use evidence from the Brexit Referendum and the 2017 French presidential election, respectively, to show that EU funding alleviates anti-EU sentiments only when accompanied by improvements in labour market opportunities.

3 Institutional background of the EU Structural Funds

The EU Structural Funds consist of the European Regional Development Fund (ERDF) and the European Social Fund (ESF). The EU Structural Funds, together with the Cohesion Fund (CF)—jointly referred to as ‘the Funds’—are instruments of the EU’s cohesion policy.

3.1 Objectives and regions

Resources from the Funds are allocated under different objectives. These have evolved over time. As of 2007, there are three objectives, each with its specific aim, scope and eligibility rules. Under the convergence objective, transfers are provided to lagging EU NUTS 2 regions to help them catch up to the rest of the Union. These funds are usually spent on road or rail construction and repair, the environment, research and technology (Applica and Ismeri Europa 2016). EU NUTS 2 regions with per-capita GDP below 75% of the EU average qualify for funding under the convergence objective. The Regional and Competitiveness Objective (RCE) targets more developed regions, again at the NUTS 2 level. Its funds are mainly allocated to projects in the area of research, technological development and innovation (Applica and Ismeri Europa 2016). The European Territorial Co-Operation (ETC or Interreg) offers support for implementing joint actions and policy exchanges at the level of NUTS 3 regions (Applica and Ismeri Europa 2016). While the convergence objective is financed from all three Funds—the ERDF, ESF and CF, only the ERDF and ESF contribute to the remaining two objectives (European Commission 2011).

After the list of eligible regions under different objectives is compiled, allocations per member state are calculated (Sapala 2015). The vast majority of these funds are allocated under shared management of the Member States and the European Commission (Sapala 2015). While the EU authorities determine eligibility, Member States have discretion over the funds’ allocation to individual regions under the specified objective. Some countries appear to allocate funds based on need, transferring larger amounts to impoverished areas (Spain, France, Greece, Italy); others seem to pursue higher returns to investment, allocating more funds to wealthier regions (Bulgaria, Czech Republic, Germany, Portugal, Romania); and yet in others (Hungary, Poland, Slovakia), no apparent relationship between per-capita GDP and per-capita transfers exists (Applica and Ismeri Europa 2016).

3.2 The 2007–2013 programming period

When ten new Member States joined the EU in 2004, the average regional per capita GDP in the EU fell. Many regions previously eligible for funding under the convergence objective found themselves above the 75% threshold. Transitional support to so-called phasing-out and phasing-in regions has been introduced in the 2007–2013 programming period to soften the impact of withdrawn convergence funding in regions that lost their eligibility due to the statistical effect of EU enlargement (Gorzelak et al. 2017; European Commission 2016). Phasing-out regions were those which would have fallen below the 75% threshold had it been computed using the EU15 average rather than EU25. Phasing-in regions were former convergence regions that were more developed than the phasing-out regions. Transitional assistance during 2007–2013 offered increased funding to the phasing-out and phasing-in regions. This was considerably less than funding under the convergence objective, yet more than would have been the case had the regions transferred directly into the RCE objective (Gorzelak et al. 2017).

In the 2007–2013 programming period, 84 regions in 18 EU Member States were eligible for convergence funding of 281 billion EUR in total, and 168 regions were covered by the Regional Competitiveness and Employment objective with a total budget of 55 billion EUR (Gorzelak et al. 2017). In addition, there were 15 phasing-out and 15 phasing-in regions.

Expenditure under the 2007–2013 programming period was allowed until the end of 2015. Figure 1 shows the timing of the release of ESIF funds.

Fig. 1
figure 1

Source: Own calculations; data on payments from the ESIF Open Data Portal, population and inflation figures from Eurostat. Note: ‘PP’ stands for programming period

Payments released from the ERDF and ESF (constant prices, reference year 2000).

Full size image

4 Method

This paper investigates the effect of EU Structural Funds on trust in politicians. Such effect may emerge through the Funds’ impact on economic developments or corruption. These matter to citizens’ perceptions of politicians’ competence, benevolence and integrity, which in turn matter to trust in politicians (Tomankova 2019). Borin et al. (2020) note that national and local administrators have an incentive to overstate their roles in the management and disbursement of EU resources at the expense of the EU. Citizens may, therefore, credit or blame any of the multiple levels of government involved in the process for the outcomes of EU funding. Also, the effect could be random in the sense that when EU transfers succeed in alleviating some of the socioeconomic hardships citizens in the lagging regions face, they feel less need to find a scapegoat, be it the EU (Borin et al. 2020), politicians, or the society as a whole (Accetturo et al. 2014). One can conjecture an effect of the EU Structural Funds on trust in politicians, absent any tangible economic or institutional outcomes, simply from the categorisation of regions by the EU authorities. The status of a convergence region or of a phasing-out region as opposed to phasing-in may signal to its citizens that something in the management of their region is not quite right and weaken their trust in politicians.

Because of the multiple possible channels of effect, the aggregate impact of EU Structural Funds on trust in politicians must be examined empirically. The challenge in doing so is dealing with the potential mutual causality between the inflow of EU resources and trust in politicians.

While resources are expected to affect trust in politicians depending on how they are utilised, the opposite direction of effect—from trust in politicians to resources—occurs when the money is assigned to regions in a non-random manner. If—as in Spain, France, Greece, Italy—resources are allocated based on need, poorly managed regions are likely to show less trust in politicians and also receive higher transfers. Conversely, if resources are allocated on the grounds of expected returns—as in Bulgaria, the Czech Republic, Germany, Portugal, and Romania—well-performing regions are likely to show more trust in politicians and receive higher transfers. This reverse causality from trust in politicians to the resources received is further intensified if trust in politicians serves as an enabling factor for institutional and economic improvements. Low trust in politicians may promote uncivic or uncooperative behaviour among citizens, making positive institutional or economic developments difficult and creating the need for transfers. Conversely, high trust in politicians may encourage cooperative behaviour, achieving growth without any extra resources.

The mutual causality between trust in politicians and EU transfers calls for an identification strategy if the causal effect of EU Structural Funds on trust in politicians is to be estimated correctly. This paper employs a quasi-experimental method known as regression discontinuity design (RDD), which is commonly used to evaluate the impact of government interventions. In terms of methodology, it relates closely to the papers by Accetturo et al. (2014) and Borin et al. (2020), who examine the effect of the EU Structural Funds on social trust and Euroscepticism, respectively.

Any policy that assigns subjects to treatment based on a cut-off (threshold) is a potential candidate for RDD, subject to additional requirements that ensure the method’s validity. In the context of EU funds, RDD commonly exploits the 75% per-capita GDP threshold that determines eligibility under the convergence objective. In addition, this paper takes advantage of the threshold between the phasing-out and phasing-in regions. This threshold is specific to the 2007–2013 programming period, and for that reason has been neglected in papers using RDD to analyse effects of the EU Structural Funds. It can only be used when working with data for the 2007–2013 period; whereas using the 75% threshold allows researchers to merge periods and scrutinise long-term effects of the receipt of EU Structural Funds.

If compliance with treatment is less than perfect, i.e. some regions below the threshold do not receive treatment or some regions above the threshold do, a modification of the classic (sharp) RDD, so-called fuzzy RDD is used.

For the RDD to correctly estimate the average causal effect, it is required that the only change that occurs at the threshold is the change in treatment. As long as expected potential outcomes are continuous, it is not necessary that units in the vicinity of the threshold be assigned to treatment as if at random (Cuesta and Imai 2016). The empirical evidence that this assumption holds is provided by checking that no other variables’ jump’ at the threshold; those that do must be included in the regression. However, conducting many such checks increases the likelihood that some discontinuities will be falsely identified by chance. Tested covariates should therefore be substantively important (Cuesta and Imai 2016).

The covariates suspected of ‘jumping’ at the threshold in the context of this study include the quality of government, as poor management may pre-determine regions for low economic performance and hence the receipt of the EU Structural Funds; social trust, for which Accetturo et al. (2014) have identified a discontinuity at the 75% threshold in the 2000–2006 programming period; and a set of economic variables capturing economic outcomes that might result from the receipt of the funds. The last point sets this paper apart from Borin et al. (2020), whose analysis abstracts from the funds’ economic effects, and addresses the concern raised by Bachtrögler and Oberhofer (2018) and supported by the research of Crescenzi et al. (2020) that it is not the EU Structural Funds per se that affect attitudes—or in this case citizens’ trust—but rather the economic improvements they induce. Thus, if some effect of the EU Structural Funds on trust in politicians is observed in the regressions estimated further on, the set of covariates considered as control variables provide for determining whether such effect is driven by a more general change in trust in others, a change in the quality of government or economic circumstances; or whether it is a direct effect of the EU Structural Funds alone.

To avoid bias in the estimated effect due to between-country differences in trust in politicians or previous receipt of EU Structural Funds, sets of country and past treatment dummies are included as control variables. Also, checks for discontinuities in pre-treatment trust in politicians are run to ascertain that any effect identified had not been present prior to treatment.

5 Data

The dataset is compiled at the level of EU NUTS 2 regions. The dependent variable—trust in politicians—is acquired from the European Social Survey (ESS). At the time of writing, eight waves have been published, starting in 2002 and finishing in 2016. Not all states are included in all waves, so averaging across multiple waves increases the number of available observations. In the baseline analysis for the 2007–2013 period, data across the ESS waves from 2010, 2012, 2014 and 2016 are averaged, with 2016 included as a post-treatment observation, as disbursements were allowed until 2015. The year 2008 is not included because in that year the resources received from the previous 2000–2006 programming period still considerably exceeded those disbursed under the analysed 2007–2013 period (Fig. 1).

In the ESS survey respondents are asked to rate on a scale from zero to ten—zero meaning no trust at all and ten indicating complete trust—how much they personally trust politicians. Their responses are used to construct two versions of the dependent variable, the share of people who trust politicians and the share of those who distrust them. The ESS dataset also provides the identification of each respondent’s region. When aggregating the responses into region-level observations, the post-stratification weights provided in the ESS are applied (European Social Survey 2014). The survey question does not aim at politicians at any particular level of government; the survey responses therefore may reflect trust in national-level as well as local-level politicians. Data on pre-treatment trust in politicians and on social trustFootnote 1 are retrieved in a similar fashion.

The ESS data on trust are matched with data on EU NUTS 2 regions’ objectives according to the EU regulations. While eligibility is assigned strictly at the level of NUTS 2, the ESS data is sometimes provided at NUTS 1 or NUTS 3. Data at the level of NUTS 3, i.e. a more detailed breakdown that NUTS 2, are provided for small countries, where the entire member state constitutes a NUTS 2 region (e.g. Estonia, Cyprus, Latvia, Lithuania). In these cases, survey responses are aggregated to the NUTS 2 level. For Germany and the United Kingdom, trust information is provided at the level of NUTS 1, i.e., in less detail than required to match the observations with the eligibility and treatment data. These observations are therefore discarded. Data on the amounts of EU resources are retrieved from the ESIF Open Data Portal.

This paper exploits the European Quality of Government Index (EQI) provided by Charron et al. (2014, 2015, 2019) to control for regional differences in governanance. The index captures how local governments provide their services—whether they do so in an efficient, impartial and uncorrupt manner. The higher the EQI value, the better the region’s governance.

Table 1 presents the variables used in the analysis, along with their brief description and the source from which they originate.

Table 1 Variables’ description
Full size table

The resulting sample contains 125 NUTS 2 regions of 13 Member States which joined the EU before 2004. Figure 2 plots regions’ trust and distrust in politicians (trst-plt-1016 and no-trst-plt-1016) against their per-capita GDPs as a percentage of the EU25 average during 2000–2002 (gdp). The chart is drawn for the entire range of GDP values, with the vertical line indicating the 75% threshold. EQI data on government quality are available for 92 of the 125 observations.

Fig. 2
figure 2

Trust and distrust in politicians in the EU NUTS 2 regions. Note: Plots relate to the variables trst-plt-1016 and no-trst-plt-1016. Y-axes are not aligned

Full size image

Out of the 125 regions in the sample, in the 2007–2013 programming period, nineteen were eligible under the convergence objective, ten were phasing out, eight were phasing in, and the remaining 88 were eligible under the Regional Competitiveness and Employment objective (RCE). Table 2 shows the average annual payment, treatment intensity, trust and distrust in politicians by these groups. Treatment intensity is calculated by dividing the average annual payments in the 2007–2013 programming period by the region’s 2007 GDP. More developed regions tend to have lower average yearly payments and treatment intensity, higher trust and lower distrust in politicians.

Table 2 Average values of selected variables by regions’ groups
Full size table

In the RDD set-up at the 75% threshold (of average regional per-capita GDP in the EU25), regions eligible under the convergence objective are considered treated (variable treat-conv equal to one), and all other regions as non-treated. This definition leads to the estimation of a sharp regression discontinuity. All regions below the 75% threshold are treated, while none above it are.

The cut-off between the phasing-out and phasing-in regions occurs at 75% of the average regional per-capita GDP in the EU15, which corresponds to 82.47% in the EU25. One non-compliant region (Namur, Belgium) appears at this threshold. Because EQI data are lacking for this observation, it is automatically excluded from model estimation whenever eqi enters as a control variable, and the fuzzy RDD becomes sharp. The single non-compliant region is dropped when working with the 82.47% threshold to avoid switching back and forth between fuzzy and sharp designs; hence, to make comparisons of results between specifications convenient.

The common practice is to estimate the discontinuity on a subsample of observations relatively close to the threshold. Table 3 displays the optimal bandwidths used for estimating regressions with trust and distrust in politicians as dependent variables.

Table 3 Optimal bandwidths
Full size table

The regression discontinuity estimation is performed in Stata. While it is in principle possible to estimate polynomials of various orders below and above the threshold, depicting the relationship between GDP and trust or distrust in politicians with complicated curves is not theoretically substantiated. Therefore, the analysis in this paper does not extend beyond the second-degree polynomial. In the sample used, quadratic and linear specifications yield similar estimates, but the former are considerably less robust. For this reason, estimations of linear specifications are presented.

6 Results

Table 4 presents the RDD estimation results at the 75% threshold.

Table 4 RDD estimation results, 75% threshold
Full size table

The statistically insignificant coefficient estimate for treat-conv in 4a suggests that being treated as a convergence region does not affect the share of citizens expressing trust in politicians. In 4b, variables that appear discontinuous or come close to a discontinuity at the 75% threshold within the optimal bandwidth for trst-plt-1016 are included: eqi-1013Footnote 2 (treated regions tend to exhibit worse governance; p-value 0.076) and unempl-1016 (treated regions tend to have lower unemployment; p-value 0.105). Still, no effect of the receipt of convergence funds on trust in politicians is observed.

For every estimation reported, two alternative specifications are performed as robustness checks. The first robustness check considers a shorter period, 2010–2014. In this alternative specification, dependent variables are calculated from the 2010, 2012 and 2014 ESS waves (trst-plt-1014, no-trst-plt-1014), and control variables are adjusted to reflect the shorter period as well (unempl-1014). The alternative period is chosen to contain trust data only from the years in which funds disbursed under the 2007–2013 programming period exceeded those released under the previous and subsequent period (Fig. 1). The second robustness check uses a wider definition of trust (or distrust) in politicians. Whereas the baseline dependent variables trst-plt (no-trst-plt) aggregate responses of eight or higher (two or lower) on a zero-to-ten scale, the wider definition includes an additional point, so citizens responding with a seven or higher are considered as trusting and those reporting a three or lower as distrusting. When included in the robustness specification, the social trust variable is adjusted in a similar fashion. The baseline definition of the dependent variables was chosen as the middle way between a narrow definition of very high trust (or distrust), which is unresponsive to circumstances, and a wider definition, in which respondents who report more intermediate values of trust (distrust) introduce noise to the variables.

The conclusion of no effect from 4a and 4b is robust to re-estimation with the shorter period as well as with the wider trust definition. The p-value for treat-conv never drops below 0.2.

Specifications 4c and 4d estimate the effect of being treated as a convergence region on distrust in politicians; no-trst-plt-1016 is the dependent variable. A mildly statistically significant distrust-reducing effect appears in 4c. This effect, however, may be considered causal only if the variables that share the discontinuity at the 75% threshold controlled for. Since no discontinuity is observed for pre-treatment distrust in politicians (no-trst-plt-08, no-trst-plt-0208), the effect in 4c pertains to the programming period under investigation. No statistically significant discontinuity is observed for any of the variables capturing economic development. However, social distrust and government quality are discontinuous at the 75% threshold within the optimal bandwidth for no-trst-plt-1016. Social distrust appears lower in the convergence regions. When no-trst-ppl-1016 is regressed on the independent variables of 4c, the coefficient estimate for treat-conv is −13.355 (p-value 0.043); and thus exceeds the effect observed in 4c for distrust in politicians in both magnitude and statistical significance. Convergence regions also display poorer governance (eqi-1013 p-value 0.027).

Social distrust (no-trst-ppl-1016) and government quality (eqi-1013) enter as covariates in 4d. Once included, the distrust-decreasing effect of treat-conv initially observed in 4c disappears, and a nearly significant distrust-increasing effect of treat-conv emerges. Each covariate eliminates the effect of treat-conv on distrust in politicians even when included on its own, so the absence of the treatment effect is not due to the drop in observations between 4c and 4d that results from the inclusion of eqi-1013.

Alternative robustness specifications of 4c and 4d provide similar results. While a distrust-reducing effect is observed when 4c is re-estimated for 2010–2014 (treat-conv p-value 0.068) or with the wider definition of trust (treat-conv p-value 0.056), it turns positive or near-zero and insignificant when social distrust and government quality are included as control variables. The main result of the no treatment effect in 4d is also robust to the addition of the unemployment rate, which was the most discontinuous of the economic variables. These alternative specifications confirm the main result of no causal effect of convergence funding on distrust in politicians. However, they speak ambiguously about the variables that pick up the statistical significance associated with treat-conv in 4c. In 4d, both social distrust and government quality relate positively and significantly to distrust in politicians. When unemployment is added or when 4d is estimated with the wider trust definition, social distrust is the only significant variable (p-value 0.017 and 0.046, respectively). Finally, in the 2010–2014 alternative specification, neither social distrust nor government quality appears statistically significant.

Thus, with regard to 4c and 4d, even if a distrust-reducing effect of being treated as a convergence region on distrust in politicians may be observed, it cannot be considered causal. The confounding factors are social distrust and government quality. Treated (convergence) regions exhibit lower political distrust, lower social distrust, and poorer governance.

Table 5 shows the RDD estimation results at the 82.47% threshold. The variable used to capture government quality in the specifications in Table 5 is eqi-sub-1013. At the 82.47% threshold, the modified eqi-sub-1013 variable exhibits a similar discontinuity as the original eqi-1013. Therefore, it can replace eqi-1013 in the specifications in Table 5 and prevent a drop in the number of observations whenever government quality is included as a control variable. The modified variable was not used in the specifications in Table 4 because at the 75% threshold, eqi-1013 and eqi-sub-1013 exhibit different behaviour (eqi-1013 is discontinuous, eqi-sub-1013 is not).

Table 5 RDD estimation results, 82.47% threshold
Full size table

Once more, treatment—in this case being treated as a phasing-out or convergence region—has no effect on the share of citizens expressing trust in politicians, as is apparent from 5a. This continues to hold when government quality, the only covariate observed to be discontinuous at the 82.47% threshold within the optimal bandwidth for trst-plt-1016 is added as a control variable in 5b. The alternative robustness estimations with the 2010–2014 period and the wider trust definition confirm this no-effect result.

Column 5c shows a mildly significant distrust-increasing effect of treat. Previous political distrust (no-trst-plt-08, no-trst-plt-0208) and social distrust (no-trst-ppl-1016) appear continuous at the 82.47% threshold within the no-trst-plt-1016 optimal bandwidth, so the effect is not inherited from previous periods, nor is it driven by distrust in people overall. Economic variables also appear continuous. Once again, however, government quality is discontinuous, with treated regions displaying better governance (eqi-sub-1013 p-value 0.038).

Specification 5d, therefore, includes government quality as a control variable. Being treated as a phasing-out regionFootnote 3 increases the share of citizens expressing distrust in politicians by eight percentage points. Since the effect remains even in the presence of the discontinuous covariate of government quality, it can be considered causal. Interestingly, the distrust-increasing effect occurs despite better governance in the phasing-out regions.

As for robustness checks, the treatment effect in 5c just falls short of statistical significance (treat p-value 0.106) when the 2010–2014 period is used but re-emerges when government quality is added (as in 5d) with a coefficient estimate of 10.322 and p-value 0.021.

The re-estimation with a wider distrust definition tells a similar story. Initially, with no significant effect (treat p-value 0.464), a distrust-increasing effect emerges when government quality is added as a control (treat coefficient estimate 5.734 with p-value 0.072).

Finally, the RDD estimation at the 82.47% threshold is performed with a reduced data sample that contains only phasing-out and phasing-in regions. The reduction limits the noise from increasing heterogeneity of observations more distant from the threshold but may compromise estimates’ external validity, as they need not apply beyond the comparison of phasing-out and phasing-in regions. Nonetheless, the estimation results from this reduced sample closely resemble those reported in Table 5. Being treated as a phasing-out region does not affect the share of citizens expressing trust in politicians, but it does increase the share of those expressing distrust by some nine percentage points (p-value 0.086). This effect can be considered causal and comes close to the distrust-increasing effect of eight percentage points reported in 5d.

7 Discussion

The results show that being classified as a convergence or phasing-out region and receiving substantial support from the EU Structural Funds has no effect on the share of citizens expressing trust in politicians, but it does affect the share of citizens expressing distrust in politicians. This would suggest that the determinants of trust and distrust in politicians differ and distrust is not merely the absence of trust (Van de Walle and Six 2014).

Table 6 summarises the main results regarding distrust in politicians. The treatment effect reported refers to the 95% confidence interval of the estimated effects of treat-conv (at the 75% threshold) and treat (at the 82.47% threshold) on the percentage of citizens expressing distrust in politicians. ‘Observed’ treatment effect relates to specifications 4c and 5c, whereas ‘causal’ treatment effect pertains to 4d and 5d, which include discontinuous variables as controls.

Table 6 Summary of key RDD results on distrust in politicians
Full size table

In the 75% threshold RDD estimation, treated (convergence) regions have, on average, a lower share of politician-distrusting citizens. However, the receipt of convergence funding cannot be presented as the cause of this decrease in distrust. If anything, when accounted for the influence of government quality and social distrust, a causal distrust-increasing effect begins to emerge.

Although the initial observed effect points to a decrease in distrust at the 75% threshold and an increase in distrust in politicians at the 82.47% threshold, the contradiction vanishes when the causal treatment effect is considered. It appears that the receipt of EU Structural Funds causes citizens’ distrust in politicians to rise. The evidence comes chiefly from comparing the phasing-out regions with the more developed phasing-in and RCE regions. However, comparing the convergence regions with those above the 75% threshold does not contradict this finding.

It may, however, be a worthwhile exercise to consider why the observed effect diverges between the two thresholds so much. Since the phasing-out regions switch roles between the two thresholds, being non-treated at the 75% but treated at the 82.47% threshold, the observed decrease in distrust at the lower threshold and increase in distrust at the higher threshold could be a manifestation of the phasing-out regions having particularly high distrust in politicians in comparison with all other regions, be it convergence, phasing-in or RCE. If the phasing-out regions happen to exhibit particularly high distrust in politicians, convergence regions, which are considered treated at the 75% threshold, would show lower distrust in politicians relative to the phasing-out regions (the observed distrust-increasing effect). In comparison, the phasing-out regions would show higher distrust in politicians than the phasing-in and RCE regions (the observed distrust-increasing effect at the 82.47% threshold).

However, the figures in Table 2 do not show that distrust in politicians in the phasing-out regions would be particularly high. So, Table 7 provides a more detailed breakdown and reports average distrust in politicians by regions’ objective and country, since between-country differences not accounted for in Table 2 could mask differences in distrust in politicians between the regions with particular objectives.

Table 7 Average distrust in politicians (no-trst-plt-1016) by country and regions’ objectives
Full size table

Phasing-out regions display very high distrust in politicians relative to all other regions in Greece and Portugal; and relative to the phasing-in and RCE regions in Italy and Belgium. Estimations are re-run with these countries individually excluded from the estimation sample to see whether any particular countries drive the results.

At the 75% threshold, Greece and Portugal are excluded in turn, as these display the largest differences in distrust in politicians between the treated (convergence) regions and phasing-out regions. When Greek regions (three convergence regions, three phasing-out, one phasing-in) are excluded from the estimation sample, the observed effect disappears, while the causal effect remains as estimated in 4d. However, this is because Greece was already excluded from the estimation in 4d due to missing EQI data. Excluding Portugal does not alter the observed nor causal effect. Because the original estimation sample included only one Portuguese region, so there was no influence on the effect estimated at the cut-off.

At the 82.47% threshold, Greece, Portugal, Italy, and Belgium are removed from the estimation in turn. Removing Portugal’s single phasing-out region leaves the observed and causal effects unchanged. Similarly, exclusion of three Belgian RCE regions makes only marginal difference in the estimates. Removing Italy’s three regions—one phasing-in, one phasing-out, one RCE—diminishes the observed effect in 5c (to a positive coefficient estimate for treat 5.641 with p-value 0.124) but maintains the causal effect (treat coefficient estimate 9.411, p-value 0.044). Removing Greek regions—two phasing-out and one phasing-in—has the largest impact. It eliminates the observed effect (treat coefficient estimate 3.428 with p-value 0.238) and well as the causal effect (treat coefficient estimate 4124 with p-value 0.241).

The country exclusion exercise thus shows that the observed effects at both thresholds are due to Italy and Greece, while the causal effect of EU Structural Funds on trust in politicians identified at the 82.47% threshold is driven solely by Greece. Table 8 takes a closer look at the average variables in the Greek regions.

Table 8 Average values of selected variables by Greek regions’ objectives
Full size table

Average distrust in politicians between 2010 and 2016 was up from its 2008 levels in all groups of Greek regions. The increase was the highest in the phasing-in regions (by 28.5 p.p.), followed by the phasing out regions (up by 24.6 p.p.), while the lowest increase occurred in the convergence regions (up by 15.4 p.p.). Distrust in people also rose, roughly by a factor of four in the convergence and phasing-out regions and by a factor of two in the phasing-in regions. Unemployment rose more in the phasing-out regions than in the convergence or phasing-in ones, while the largest drop in per capita GDP occurred in the convergence regions.

Since it is difficult to draw a sensible conclusion from the descriptive evidence on the Greek regions, let us conclude that the causal distrust-increasing effect of EU Structural Funds was driven by a country that faced severe economic challenges during the period under investigation. The effect could only be attributed to the receipt of EU Structural Funds per se, not to their effects on the economy or government quality or general social distrust. In Greece, simply being given EU money during the time of crisis increased distrust in politicians. This appears sensible in light of the research of Borin et al. (2020) on the link between EU Structural Funds and Euroscepticism. The economic crisis and the associated negotiations with the EU authorities have sparked negative sentiments towards the EU across all social groups in Greece (Clements et al. 2014). However, since the inflow of EU money reduces the likelihood that citizens pick the EU as a target to blame (Borin et al. 2020), in the Greek phasing-out regions, which received comparably more EU resources than the phasing-in regions, citizens would have been more likely to pick a scapegoat other than the EU. On the grounds of this paper, the politicians, it seems.

Finally, despite the body of evidence that points to a positive effect of EU Structural Funds on economic variables (Becker et al. 2013; Pellegrini et al. 2012; Bachtrögler 2016), no significant discontinuity was observed throughout the analysis for the per capita GDP growth rate, the employment growth rate, or the unemployment rate. This may be an offshoot of Bachtrögler’s (2016) finding that relative to previous programming periods, the effectiveness of convergence funds deteriorated during the 2007–2013 period analysed in this paper.

8 Conclusion

Well-intended policies sometimes backfire. As the effects of the EU Structural Funds on the economic progress of the less developed regions have been thoroughly researched, the focus is now shifting to outcomes that policymakers may have neglected when designing funding policies. This paper examines the effect of EU Structural Funds on citizens’ trust and distrust in politicians in regions of the ‘old’ EU Member States in the programming period 2007–2013. The two legislative thresholds embedded in the EU Structural Funds architecture permit the identification of causal effects via the regression discontinuity design.

Results show that in the sample under investigation, receiving support from the EU Structural Funds does not affect the percentage of citizens’ expressing trust in politicians, but it does affect the percentage of those expressing distrust. At the threshold that separates the phasing-out and phasing-in regions, there appears to be a distrust-increasing effect of the receipt of the EU Structural Funds, which is not attributable to changes in economic conditions, quality of government or wider social distrust. Receiving EU transfers causes citizens’ distrust in politicians in the phasing-out regions to rise by eight percentage points on average. The effect is driven by Greece, which experienced severe economic turmoil coupled with austerity measures during the period under investigation. At the threshold between convergence regions and the phasing-out regions, the evidence of a causal effect of EU Structural Funds on trust in politicians is inconclusive.