Challenge and Response of Regional Disparities: Romania in a Comparative Perspective

NUTS is a classification system of the EU regions for statistical data analysis and policy purposes. This classification is the basis for the way structural funds could be accessed (Eurostat 2018). The configuration of NUTS 3 and NUTS 2 for Romania is presented in Fig. 1.

At that time, the counties that had been replaced by Soviet-type artificial regions in the period 1950–1967 were reinstitutionalised (Săgeată 2012). Even if the counties that were established in 1968 were consistently different from the counties during the two world wars, they represented a return to a long-term administrative tradition of the country and the majority of them were potential functional units (Stahl 1969).

The rather high stability of the development hierarchies of counties and an increase in inter-county disparities were also noticed for the period 1990–1994 (Hansen et al. 1996).

The period 2012–2020 is taken here as a key reference due to the fact that it offered the possibility to consider a relatively homogeneous period after the global crisis starting in 2008. Secondly, 2014 marks the beginning of the EU financial period for the allocation and absorption of funds, up to 2021. For causal approaches on understanding different country absorption rates in the EU, we need to consider the value of some causal factors at the beginning of this financial period.

The bivariate regression having only RCI efficiency as predictor for the increase in GDP per capita at NUTS 3 level gives a regression coefficient of only 1.31 compared to 2.11 in the multivariate regression for EU15 in Appendix 1.

The average GDP per capita as a percentage of the EU mean for mostly urban NUTS 3 regions, in the new Member States, was 2.5 times larger than in the mostly rural NUTS 3 regions for the year 2012. The same ratio for the case of the EU15 countries for the same year was lower, at 1.3. Computations from Eurostat data are not presented here.

The LHDI is a measure of development at commune and city level, aggregating indicators on socio-human capital, material capital and health (Sandu 2020b). For the first computations of the index we used data that were available only from the last census in 2011. The LHDI values for 2018 are computed on a slightly different methodology without being obliged to use census data. In spite of slightly different methodologies of computation, the two indices are comparable (ibid.).

OLS regression models having LHDI 2011 and LHDI 2018 as dependent variables. LHDI 2018 is a factor score (multiplied by 100) aggregating indicators referring to material capital of locality, rate of penetration of internet in locality and rate of standardised mortality 2016–2018. LHDI 2011 is also a factor score aggregating education stock, material capital, life expectancy at birth, multiplied by 100. LHDI 2011 and LHDI 2018 are only partially comparable. Their constituent indicators refer to the same forms of community capital but the measures are different (Sandu 2020b). Population of locality as predictor is measured by a log transformation of locality population in 2017 in the equation for LHDI 2018 and population of locality for 2012 in the equation for LHDI 2011.

The percentage is given by what is called in statistical analysis R2 change. Model LHDI 2018b explains 27 per cent of the variation of the dependent variable. Once the urban regions are added in the model LHDI 2018a, the prediction power of the model reaches 48 per cent. The difference between the two measures is what is called R squared change and gives, in this case, a measure of the urban regions belonging to the explanatory power of the model.

The cultural areas from Fig. 3 are, in fact, subregions of historical regions. The identification of these subregions is detailed in Sandu (2020c). Each county is characterised by a profile of human capital, vital capital (mean age, general fertility rate), economic capital (GDP per capita), cultural capital as historical region of location, and degree of urbanisation. The cultural areas of the country as determined for the year 2018 are relevant for the most similar neighbouring counties within the same historical region.

The statement is based on the results of multiple regression computations that are not presented here due to the limited space. Each type of region (historical, NUTS 2 and cultural area) was included in the regression model having the predictors from LHDI 2018b, from Table 2. Resulted R2 are compared to those from the model LHDI 2018b in Table 2 by using R squared change. This value is of 27 per cent in the equation including development regions as predictors and of 26 per cent for the impact of historical regions.

They included EAFRD (the European Agricultural Fund for Rural Development), EMFF (the European Maritime and Fisheries Fund), ESF (the European Social Fund), ERDF (the European Regional Development Fund) and CF (the Cohesion Fund).

The highest rates of absorption are held by Austria, Ireland, Luxembourg and Spain (with rates between 65 and 73 per cent).

The countries that are situated on the regression line or are very close to it could be considered as having an absorption rate that is consistent with their economic development. Luxembourg, for example, is a highly developed country and its performance in absorbing European funds is as expected or in accordance with the general rule presented in Fig. 4. About 28 per cent in the variation of ESIF is explained by GDP per capita as measured in 2019. If one considers GDP in 2013, the corresponding percentage is slightly lower by 21 per cent.

For a comparison of CPI with other corruption measures see Hamilton and Hammer (2018). The correlation between the CPI 2013 and the share of people declaring that in their country corruption is widespread (Eurobarometer 79.1, 2013) is r = -0.72 in the series of EU28. CPI could be considered as a proxy variable for measuring the corruption per se in political and public administration sectors.

All the standardised path coefficients are significant for p = 0.05, excepting the coefficient for the arrow linking new Member States (1 yes, 0 no) with the Corruption Perceptions Index (CPI) 2013 with p = 0.08. All the other possible paths into the diagram that are non-significant from the statistical point of view were omitted. The AMOS model is run on the five variables for the 28 EU states for the specified year. The discussion on the significance levels is conventional as the data set refers to all the EU countries. The key reason of adopting the procedure is that the measures of ESIF are variable in time function of the reference year within the programming period. The model has a good fit to the data: CMIN = .923 and p = .921; GFI = .987; RMSEA = .000. This could be an effect of the reduced number of cases in analysis. An alternative model was run without new Member States as exogenous variable, excluding the causal path from GDP to CPI but with a covariance relation between GDP and CPI. All the other path coefficients in the model remain the same as in the presented model.