The impact of the EU "development of Eastern Poland" operational programme on individuals' well-being

(1)

Page | 1

The impact of the EU “Development of Eastern Poland”

Operational Programme on individuals’

well-being.

-Master Thesis Economics- August 2017 University of Amsterdam

Agata Makowska MSc. Economics (Public Policy)

Prof. Dr. E.J.S. Plug Supervisor

(2)

Page | 2

Statement of Originality

This document is written by student Agata Makowska who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Page | 3

Abstract

Using the Social Diagnosis data collected by the Polish Council of Social Monitoring, this thesis explores the effect of the European Union’s “Development of Eastern Poland” Operational Programme (DEPOP) on the development gap between the less advanced Eastern and prosperous Western Poland. In particular, a Difference-in-Differences technique is applied to determine whether additional grants from the European Regional Development Fund (ERDF) had the expected positive impact on individuals’ material prosperity and their subjective perception of life satisfaction. Firstly, I find that the EU programme has, as would be expected, a positive impact on regional government revenues. With this knowledge, I hypothesise that these increased regional government revenues are a reasonable proxy for public investment. Public investment is shown to be positively related to individual well-being. Therefore, the DEPOP should increase well-being. My results show, however, that the introduction of the programme did not have a significant effect on any of the indices measuring the well-being of regions’ inhabitants. I argue, that these findings cast doubt on the effectiveness of the DEPOP and its attempt to encourage economic convergence between Eastern and Western Poland.

(4)

Page | 4

1. Introduction

One of the main budgetary expenses of the European Union (EU) are the transfers to the poorest regions within the member states. These transfers aim to accelerate their economic convergence with more affluent regions (Mohl & Hagen, 2010). The foundation of the European Regional Development Fund (ERDF) in 1975 prioritized the problem of the regional development disparities. Since then the ERDF contributes significant financial resources to stimulate the

economic growth and public investments for the regions with the lowest output in the EU1.

With the subsequent admissions of new country members, the share of the EU central budget

spent on the ERDF increased from around 7% for the years 1975-1988 to 36% during the period between the years 2007 and 2013 (Becker et. al, 2012). Hence, the increased magnitude of

transfers means that accurate evaluation of their impact has become more crucial (Dall'erba &

Le Gallo, 2008).

This thesis focuses on the problem of regional imbalances within the Republic of Poland and evaluates the efficiency of the EU’s Development of Eastern Poland Operational Programme (DEPOP). The DEPOP was introduced between the years 2007 and 2013 to stimulate development in the five poorest regions of Eastern Poland, as defined by those with the lowest GDP per capita in the EU in 2005. Due to cooperation between the ERDF and the Government of Republic of Poland, the grant totalled 3.68 billion euros. It was expected that this additional financial support from the EU, substantially exceeding the beneficiary regions’ yearly revenues, would also improve the welfare and life satisfaction of region inhabitants.

In my study I measure the success of the DEPOP by examining the programme’s effect on well-being of inhabitants in Eastern part of Poland. I use the individual-level panel data obtained from the Social Diagnosis2 research conducted every two years on the territory of Poland. I distinguish two types of well-being: material (related to the consumption possibilities), and non-material (indicating subjects’ subjective quality of life). The former is reflected by individuals’ yearly nominal income and its expected growth in the next two years, whereas the latter refers to their own perception of happiness. Due to the availability of numerous waves of questionnaires, I adjust the sample so that it contains the observations for two years before and four years after the introduction of the DEPOP in 2007. Moreover, I divide the sample into a subset of a Treatment Group for five Eastern regions receiving the grant and a Control Group

1_{See more: Regulation (EC) No 1080/2006 of the European Parliament and of the Council of 5 July 2006 on the} European Regional Development Fund and repealing Regulation (EC) No 1783/1999

(6)

Page | 6 consisting of the observations for the remaining voivodeships3. As a consequence of the above setup, each of the empirical models exploits the effect of the DEPOP using the Difference-in-Differences framework.

The empirical strategy is combined of two stages. First, I evaluate whether the financial aid granted by the European Union leads to an increase in beneficiary regions’ revenues. I find that the introduced policy has a positive and significant influence on the level of financial resources available to Eastern regions’ local governments. Next, I focus my analysis on the evaluation of the effect of the DEPOP on well-being indicators for Eastern regions’ inhabitants. I run the regressions for the full set of outcome variables reflecting individuals’ well-being. Surprisingly, despite the observed effect of the DEPOP at the regional level, a similar and significant relationship is not found for the dependent variables aggregated at the micro level. What is more, the estimated coefficients are robust to a different sample selection and an introduction of additional control variables.

This thesis adds to the literature in numerous ways. It evaluates the impact of the “Development of Eastern Poland” Operational Programme econometrically rather than qualitatively as observed in the past literature. It also provides the analysis of the problem at both a macro and micro level and uses a dataset collected by individual questionnaires. These questionnaires offer detailed information not only on respondents’ economic and demographic characteristics but also on their mindsets and opinions. Contrary to the official final report of the DEPOP, this thesis provides the novel evidence slightly undermining its overall success. However, the microeconomic perspective in this analysis means that a comparison between this paper and the official final report is difficult. The main objective of the conducted research is to broaden the base of evidence needed for future improvements in the field of economic development of EU country members and Poland, in particular.

The remainder of the thesis proceeds as follows. Section 2 provides a brief description of the Programme and reasons for its implementation. Section 3 reviews findings from the previous literature. Section 4 presents the dataset, summary statistics for the main variables and the applied methodology. Section 5 follows with the presentation of empirical results and

3_A_{voivodeship is an English name for a top-level Polish province (pol. województwo). The Republic of Poland}

is divided into 16 voivodeships. According to the EU regional classification, a voivodeship is classified as NUTS2 region. (Central Statistical Office of the Republic of Poland, 2013)

(7)

Page | 7 robustness checks. Section 6 discusses potential interpretation of the findings. Section 7 closes the thesis with main conclusions.

2. Development of Eastern Poland Operational Programme (DEPOP)

2.1. Gap between Western and Eastern Poland

When analysing the socio-economic development of Poland, dividing the country into two parts is inevitable (Gorzelak, 2007). We can clearly distinguish the richest Western part from the poorest regions in the Eastern part. This distinction, which is still applicable today, has its foundations in the historical process of the partition of Poland in the 19th_{century (Krasowska,}

2013). By signing a partitioning convention in 1797, Austria, Russia and Prussia took over the country for the following 123 years. Polish markets under occupation of three different countries were becoming less and less cohesive. Gradually, the Polish lands belonging to the invading powers started to resemble their domestic economies due to the introduction of local laws and different monetary, credit and customs systems. The Prussian territory located in the Western part of Poland was significantly more advanced economically than the southern Austrian and eastern Russian partitions. The main source of the dominance of this region were the accelerated industrialization and the expansion of the railway infrastructure network initiated by Prussians. On the contrary, economies of Austrian and Russian partitions, mostly based on the traditional agriculture methods, lacked the stimulus required for the intensive growth and innovations.4

The disproportions between the East and West established two centuries ago, together with the change of the country’s borders in the 20th century, led to the existence of a persistent gap in the development of these two regions. Attempts to reverse this divergence in development between the regions, have had limited success. These limitations are contributable to a number of factors. The most important factor, according to the Polish Ministry of Regional Development (2013), is the structural regress of Eastern Poland. With the largest population share working in the primary sector in the country, inefficient restructuring of main cities (especially those specialised in heavy industries, located in Swietokrzyskie and Podkarpackie5₎

and no involvement of foreign capital, the inequality between the so-called “Poland A” and “Poland B” eventuated.

4_{Further reference to the history of Poland can be found in the paper of Krasowska (2013).}

5_{Swietokrzyskie and Podkarpackie are the regions (voivodeships) of Poland. Figure A in Appendix presents a map} of Poland with the names of Polish voviodeships.

(8)

Page | 8 Furthermore, the admission of Poland to the European Union (EU) in 2004 highlighted the existence of the historically conditioned disparity in the development between the Eastern and Western Poland. Figure I reports the GDP per capita in EUR for sixteen Polish NUTS2 regions in 2004. The mean GDP per capita in Eastern Poland amounted to 80% of the average calculated for the whole country. Moreover, it was approximately two times smaller than the value of this indicator for the richest voivodeship, Mazowieckie (17 000 GDP per capita). The ranking of the least developed regions was as follows6: Lubelskie (69.4%), Podkarpackie (69.8%), Podlaskie (74.7%), Swietokrzyskie (77.4%) and Warminsko-Mazurskie (77.7%). Low levels of GDP per capita ranked those regions not only on the bottom in Poland but also in Europe, being below 40% of the average for the EU. Given this economic disparity between Eastern Poland and the EU, a quick recovery exercise was required for this under-developed Eastern region.

Figure I

GDP per capita in EUR for NUTS2 Polish regions in 2004

Note: Highlighted regions are the beneficiaries of the DEPOP. Figure A in Appendix shows a map

of Poland with the names of NUTS2 regions (voivodeships).

2.2. Introduction to DEPOP

6_{The value in parentheses is equal to the percentage of the average GDP per capita for Poland in 2004.}

17 000 9 900 8 600 8 000 10 400 8 300 8 900 8 200 10 000 11 400 12 100 11 100 10 200 1 0 0 0 0

(9)

Page | 9 Due to the notable differences in GDP per capita between the five poorest regions in Eastern Poland and the Western region of the country in 2005, the Polish government decided to introduce a special funding programme to address this problem. In 2007 the European Commission accepted this initiative and committed the ERDF to subsidize Lubelskie, Podkarpackie, Podlaskie, Swietokrzyskie and Warminsko-Mazurskie voivodeships with the amount of 2.38 billion euros during the years 2007-2013. In addition, financing to the amount of 1.3 billion euros was also granted by the Government of the Republic of Poland7. Therefore, the total budget of the DEPOP amounted to 3.68 billion euros. Principally, the Programme’s funds were assigned to large public investments crucial for regional development. Figure II shows the proportion between the additional funds from the DEPOP and the sum of yearly nominal revenues of the subsidized regions (more specifically regions’ local governments).

Figure II

Comparison of the yearly nominal local governments’ revenues of the five beneficiary regions with the yearly amount of subsidy from the DEPOP for years 2007-2013.

Figure II shows that the grant received from the DEPOP was significantly greater than the own revenues of the recipient regions each year. In fact, the grant constituted, on average, more than 170% of the local governments’ yearly revenues during the period of the programme. For that reason, the financial aid was expected to substantially improve the state of the under-developed parts of Poland and their inhabitants.

The main goal of the Programme was to accelerate the pace of the sustainable socio-economic development of Eastern Poland and to inhibit the proceeding marginalization of the region. This

7_{The grant was earmarked by the Polish government from the financial grant received by Poland from the} ERDF. 0 50 100 150 200 250 300 350 400 2007 2008 2009 2010 2011 2012 2013 Millio n eu ro s

(10)

Page | 10 was the answer to the challenges of the EU’s Lisbon Strategy and the Community Strategic Guidelines on Cohesion for the years 2007-20138. Both strategies aimed to increase the competitiveness of the Polish economy based on knowledge and innovation as well as to improve the level of social, economic and spatial integrity within the country. In order to meet the primary objective of the DEPOP, the financing plan was divided into six priority axis: Modern Economy, Information Society Infrastructure, Regional Growth Centres, Transport Infrastructure, Sustainable Tourism and Technical Assistance. Hence, the additional financial stimulus from the EU was intended to advance different features of beneficiary local economies: enhance business innovations; increase investments in infrastructure, higher education and communication; create new jobs; improve the image of the region and attract foreign direct investment (FDI) in this part of the country. Therefore, having a cumulatively positive impact on the well-being of the recipient regions’ inhabitants.

3. Literature Review

In this section, I summarize findings from the past literature related to the assessment of public policies using individual well-being indicators as outcome variables. Moreover, to the best of my knowledge, there are no evaluations for the EU regional policies at the microeconomic level of regions’ inhabitants. For that reason, I present the papers exploiting quantitatively their efficiency measured for macroeconomic indicators. Additionally, I show the results from the official ex-post final report on the DEPOP, prepared due to the cooperation of the ERDF and the Polish government.

3.1. Impact of policy intervention on individual well-being

In order to improve the measurement of the effect of any policy intervention, one can expand the traditional output macro-analysis to the material and subjective well-being indicators of the affected population (Diener, 2006). Frey and Stutzer (2010) examine the benefits from attaining good quality, subjective well-being data for the policy evaluation research. They find that the impact of a certain policy on individual happiness can be associated with the problem of maximisation of a social welfare function. Economists simply want to have an additional insight in the process of public decision making, which from its nature aims to improve the life of the society. Nevertheless, the authors stress the fact that measuring the role of a policy in

8_{See more about the EU Lisbon Strategy at}

http://www.consilium.europa.eu/en/uedocs/cms_data/docs/pressdata/en/ec/00100-r1.en0.htm; see more about the Community Strategic Guidelines on Cohesion for the years 2007-2013 at

(11)

Page | 11 maximising the aggregated happiness should not substitute the analysis of the effect on traditional socio-economic indicators like GDP. The issue related to the usage of reported well-being as a policy assessment measure is also discussed in the paper of Layard (2006). The author claims that the theoretical background of public economics and policy evaluation should be adjusted for the achievements of modern psychology. Under its current form, the theory fails to correctly explain the implications of policy interventions due to missing subjective information about economic agents. To prove this, Layard shows that across the same time span, the level of happiness of inhabitants of Western European countries has remained the same, even though their income has significantly increased. This finding, known as the “Easterlin paradox”9_{, contradicts the economic theory stating that a higher income increases the}

utility of an individual10. Due to this, showing the impact of a policy using only objective measures might lead to incorrect results. Therefore, it is recommended to control for the potential bias by adding the analysis for subjective indicators.

In particular, the data on individual well-being can be useful when measuring the effects of policies dedicated to adjusting the differences in regional development. Pittau, Zelli and Gelman (2010) look at past research on the effect of EU regional development programmes on both objective and subjective well-being outcomes. According to the authors, the majority of the analyses of the EU incorrectly concentrate on the correlation between the microeconomic indicators for the country’s population with the macroeconomic indicators aggregated at a national level. They argue, that the macroeconomic variables on a regional level are more suitable as they can explain more variation of the well-being variables for regions’ inhabitants. Using the multilevel models11 for the cross-sectional Eurobarometer data12, they find that the changes in the regional GDP per capita and unemployment rate13 have a significant impact on the subjective well-being (income and happiness) reported by the regions’ inhabitants.

3.2. Evaluation of EU regional policies

9_{See more in Easterlin (1974)}

10_{Layard in his study assumes that the subjective level of individual’s happiness is a proxy for individual utility.} 11_{In multilevel models, due to the availability of the data at the individual and more aggregated levels, the outcome} is determined by the combination of between-level characteristics (Pittau, Zelli & Gelman, 2010)

12_{Dataset published by the European Commission, gathered via the “Eurobarometer” survey. Authors use a subset} for respondents living on the territory of 70 NUTS1 regions, reporting their answers during the period between the years 1992-2002.

13_{The change in the regional GDP per capita or unemployment can be treated as a proxy for the impact of a} regional development policy.

(12)

Page | 12 The problem of a quantitative evaluation of regional policies implemented by the European Union has been widely discussed in the past and recent literature without a clear consensus concerning the impact of these programmes (Mohl & Hagen, 2010). Pellegrini et al. (2012) examine the effect of the EU Cohesive Policy by estimating a regression for regional data collected for years from 1994 to 2006. In their study, they apply a discontinuity design for a deterministic comparison group method. To distinguish the least developed regions, they use a threshold of 75% of the EU average level of GDP per capita and compare the group of regions just below and just above it. They conclude that the evaluated policy helped in stimulating the growth of GDP in the treated regions.

On the contrary, the paper of Dall’erba and Le Gallo (2008), finds that the grant from the ERDF had no significant impact on the observed regional convergence. In their study they use a dataset for 145 regions in the EU for the time span of 1989-1999 to estimate the effect of the financial aid from the EU fund on the beneficiary regions. Their results are robust, even when controlling for potential spill-overs and endogeneity by examining the spatial inter-region relations. The paper of Rodriguez-Pose and Fratesi (2004) presents the general result indicating no impact of a similar grant from the ERDF. More specifically, the authors evaluate the impact of the grant for different axes receiving additional financial resources from the ERDF within the Objective 114. They find that, no matter how big was the amount of the grant allocated to a given axis, it had no significant impact on the regions’ growth in the long-run. Only for the grants for the primary sector, the effect on growth was significant, however only in the short-run. Therefore, basing on the previous econometric research, it is difficult to make a general conclusion about the efficiency of the EU regional policies in the economic convergence of less developed regions.

3.3. Ex-post evaluation of DEPOP

In the official ex-post evaluation of the DEPOP published by the Polish Ministry of Regional Development and the ERDF (2017) the findings were that the Programme was successfully implemented and met the ex-ante goals. The final report was prepared in accordance with the guideline for the closure of the ERDF operational programmes approved by the European Commission15_{. According to the report, the DEPOP helped to create 20.6 thousand new jobs}

and to increase the GDP of the macro-region “Eastern Poland” by 2.5% since the year 2006.

14_{Objective 1 is the area of focus of the EU, supporting the economic growth of the least developed regions in} the EU.

(13)

Page | 13 The main part of the assessment of the programme focused on the realization of the goals defined for the six priority axes. The level of expenditures planned for axes II, V and VI (Information Society Infrastructure, Sustainable Tourism, Technical Assistance) did not reach the level of 100% of the allocated funding, whereas the expenditures for axes I, III and IV (Modern Economy, Regional Growth Centres, Transport Infrastructure) significantly exceeded their allocated funding. Nevertheless, due to the application of the rule of elasticity16, it was possible to keep the overall allocation of the funds at the 100% level. Students and R&D institutions were the main final beneficiaries of the DEPOP, receiving the total grant of 1289 million euro.

The key research method in evaluating the success of the DEPOP presented in the report was the analysis of the three “horizontal indicators”. The first two indicators: the net number of created jobs and the impact of the EU financial aid on the GDP, reached the level forecasted in the ex-ante DEPOP evaluation17_{. The third examined statistic, the gross number of created jobs}

(jobs created directly because of the implementation of the DEPOP) was equal to 3149 new posts and represented 126% of its target ex-ante value. These results led to a conclusion, that the development goal of accelerating the pace of socio-economic development of Eastern Poland has been achieved. However, being critical of this analysis, it does not control for external influences on these indicators, for example, GDP increases are unlikely due solely to the DEPOP. Moreover, a more general examination of macroeconomic indicators (GDP and the unemployment rate) in the same report arrived at the inference that the regions in Eastern Poland are still behind the remaining part of Poland in terms of their level of economic development.

4. Data and empirical strategy

4.1. Social Diagnosis Data

In my research I use a panel dataset for years between 2000 and 2015 obtained from the Social Diagnosis conducted on the territory of Poland by the Main Council of the Polish Statistical Association. The main aim of the project is to investigate the quality of life experienced by Poles and how it corresponds to their self-perception. The analysis published by the Social

16_{The rule of elasticity allows to certify to the European Commission expenditures exceeding the limit set for one} axis with the simultaneous decreased level of expenditures for another axis.

17_{See more on the ex-ante DEPOP evaluation at}

http://ec.europa.eu/regional_policy/sources/docgener/evaluation/library/poland/0611_eastpoland_development_s um_pl.pdf

(14)

Page | 14 Diagnosis provides an insight into the changes in the state of Polish society starting from the beginning of the 21st century up to the most current period.

The publicly available datasets are gathered with the use of two separate questionnaires for households and their individual members. Both questionnaires are carefully prepared by experts in the fields of psychology, economics, sociology, statistics and demography due to the interdisciplinary character of the study. The first sample was collected in the year 2000, followed by the second wave in 2003. Subsequent waves took place every two year following 2003. The collected data consists of information on households’ and individuals’ attitudes, mind-sets and behaviours as well as economic and demographic characteristics.

For my analysis I use a dataset aggregated for individual members of Polish households. Initially it consists of 84 478 observations and 3 520 variables for years 2000-2015. In order to obtain a balanced panel, I exclude the records for individuals who did not participate in all of the waves of the program. This step significantly reduces the size of my sample. Next, to address the missing data problem I decide to limit my panel to years 2003-2013. Thus, I keep the observations for two waves before the introduction of the program (years 2003 and 2005) and four waves overlapping with the period of the program (years 2007, 2009, 2011 and 2013). Moreover, to prevent the non-representativeness of the sampling, I drop the observations for people below the legal working age of 16 years old on the day of the first wave of study. My final dataset consists of 4584 observations in total, representing information on 764 respondents participating in 6 waves of the Social Diagnosis study.

4.2. Main variables

4.2.1 Dependent variables

In order to proxy respondents’ material well-being, I use the variables showing their present nominal income and their expected income in 2 years’ time, expressed in Polish Zloty [PLN]. The intuitive choice of individual income for an indicator of material wealth is in accordance with a list of well-being indicators published by OECD (2011). Moreover, in order to evaluate a potential impact of the DEPOP on future income growth, I follow the methodology of Jappellia and Pistaferrib (2000) and use the subjective forecast of individual income as a proxy. For the sake of a more intuitive interpretation and to ensure the linearity, in my models I use a logarithm of individual income and the expected growth rate of individual income18. Thus, an

18_{Expected growth rate of individual income = ln(Individual income expected in 2 years’ time) – ln(Individual} income)

(15)

Page | 15 effect of a marginal change of an independent variable can be read as a percentage change in the dependent variable.

The main variable of interest in the second part of my study is the individual perception of well-being. In the questionnaires for the years 2003-2013 this question was repeated in every questionnaire19:

How do you feel about your life as a whole?

By answering the above question, respondents subjectively assess their level of happiness at the time of the interview. Each life satisfaction rating is then assigned a value where the lower the number, the more satisfied the individual. The most important issues related to the measurement of happiness can be found in the paper of Easterlin (1974). For an unbiased estimation we need the outcome variable to be presented in a cardinal scale (Winkelmann et. al, 1995). As each individual has its own perception of the imposed range of values, making comparisons between individual responses becomes more difficult. For instance, two people could identify as “happy”, however, one could choose “1” to rate their life and the alternate could choose “2”. The existence of this issue suggests that the dependent variable is derived from an ordinal scale. Following the approach of Winkelmann et. al (1995), the potential bias created from this issue can be solved by the usage of a fixed effects model, as will be described in detail in section 4.3.

Moreover, before running the main regressions, I estimate a separate “first stage” model to see if the additional subsidy from the DEPOP indeed increases the revenues of beneficiary voivodeships. For the purpose of this evaluation, I create a new dependent variable measuring local government revenues20 and again, transform it logarithmically.

4.2.2 Independent variables

As in my analysis I apply the Difference-in-Differences method, I use the Social Diagnosis data to create two binary variables; Treatment indicating the region receiving the subsidy (Warminsko-Mazurskie, Podlaskie, Lubelskie, Podkarpackie, Swietokrzyskie), and Year indicating the period between 2007 and 2013 and corresponding to when additional financing from the DEPOP was received. The set of observations for which Treatment is equal to 1 will

19_{Some of the questions were not consistent across all questionnaires.}

20_{Local government revenues are equal to the own revenue of each region in EUR, adjusted for the transfers} from the EU.

(16)

Page | 16 be called a “Treatment Group” hereafter, whereas the remaining observations will create a “Control Group”.

Table I provides summary statistics for both dependent and independent variables used in the study. For the sake of comparison, values for the Treatment and Control groups are shown separately.

The mean logarithm of local government revenue is higher for the Control Group than for the Treatment Group for the period before and after the introduction of the programme. On average the relative gap between the own revenues of Western and Eastern regions is equal to 156% before and 4.6% after 200721. Clearly, the significant decrease in the revenue gap over time is related to the receipt of additional funds from the DEPOP by Eastern regions. The range of variable values is narrower for the Treatment Group, with higher minimum and lower maximum values for both periods. When looking at the log of individuals’ incomes we observe a similar pattern: inhabitants of Western Poland earn more both before and after 2007. The average gap in individuals’ income is equal to 15.8% before and 16.1% after 2007, therefore it increases during the implementation of the programme. Inhabitants of Eastern regions have a higher growth rate of expected income before the DEPOP’s implementation compared to their Western neighbours. The expected growth rate of income is higher by 1.8% for the Treatment Group before and by 0.5% for the Control Group after the year 200722. Moreover, the expected growth rate of income falls for both groups in the years 2007-2013. Nevertheless, a higher standard deviation and a wider range indicate a greater variation in the income of individuals living in the West. Finally, the mean value of the happiness index for the Treatment Group, exceeds the equivalent value from the Control Group by around 0.2 for both periods, implying that on average individuals living in Western regions are slightly happier. The standard deviation is nearly identical across the two groups. From the last two rows we can conclude that 32% of the sample (1 500 observations) can be classified as the Treatment Group and 68% of the sample creates the Control Group (3 084 observations). Furthermore, while 66.6% of the dataset (3 056 observations) comes from the questionnaires collected after the introduction of the programme in 2007, 33.3% represents the preceding period (1 528 observations).

21_{Before 2007: (𝑒}0.939_{− 1) ∗ 100% = 156%; After 2007: (𝑒}0.045_{− 1) ∗ 100% = 4.6%;}

22_{Before 2007: Control Group: (𝑒}0.263_{− 1) ∗ 100% = 30%; Treatment Group: (𝑒}0.276_{− 1) ∗ 100% = 31.8%;}

(17)

Page | 17 Overall, there are clear discrepancies in the level of dependent variables between the Control and Treatment Group, suggesting the quality of life in the West was persistently higher. Further insight into these differences is presented in Figures III, IV, V, VI showing the distribution of mean dependent variables for East and West before and after the introduction of the grant from the EU. Looking at the log of individual income and expected growth rate of income before 2007, we observe parallel trends for both the Control and Treatment Group. Similarly, the distribution functions of the variable log of local government revenue have corresponding slopes for both groups before 2005. For happiness, the slopes between the pre-treatment years 2003-2005 clearly differ. These findings are important for the research methodology used in this thesis, and are explained in detail in the next section. The clear advantage experienced by the West, observed in the log of individual income and level of happiness, remains present throughout the duration of the DEPOP. In addition, in the post-programme period the slope of the expected growth rate of income distribution function changes in the opposite direction. Interestingly, after the year 2009 the average log of local government revenue starts to improve for Eastern regions and to deteriorate for their Western neighbours. Therefore, during the period

Table I Summary Statistics

Control Group Treatment Group

Mean SD Min Max Mean SD Min Max

Dependent variables – before 2007

Log local gov.

revenue 3.630 1.502 1.090 6.547 2.691 1.184 1.114 4.178 Log individual income 9.327 0.547 7.090 11.184 9.180 0.531 7.090 11.002 Expected growth rate of individual income 0.263 0.396 -2.590 2.302 0.276 0.396 -1.686 3.401 Happiness 2.950 0.989 1 7 3.184 1.072 1 7

Dependent variables – after 2007

Log local gov.

revenue 5.317 0.784 3.999 6.954 5.272 0.223 4.716 5.522 Log individual income 9.712 0.562 7.212 11.695 9.562 0.534 7.783 11.695 Expected growth rate of individual income 0.191 0.291 -1.386 2.708 0.187 0.297 -1.945 2.708 Happiness 2.856 0.905 1 7 3.010 0.957 1 7 Full sample

Mean SD Min Max

Independent variables

Treatment 0.327 0.469 0 1

Year 0.666 0.471 0 1

Notes. Log local government revenue is a natural logarithm of region’s own revenue (earned due to the collection of property, income and sales taxes, transfers, charges and other sources); for the Treatment Group the value of own revenue is increased by the amount of subsidy received from the DEPOP.

(18)

Page | 18 of the DEPOP’s financing, the regional revenues’ growth rate stabilizes for both groups, keeping revenues constant at almost the same level.

Figure III Figure IV

Figure V Figure VI

Figures III, IV, V, VI

Distribution of the average values of Log of individual income, Expected growth rate of individual

income, Happiness, Log of local government revenue for years 2003-2013.

Note: The dashed vertical line indicates the year of implementation of the DEPOP. The horizontal

axis shows the timeline for waves of Social Diagnosis study. The vertical axis presents the average values of the main dependent variables.

4.3. Empirical Strategy

4.3.1 Difference-in-Differences framework

The setup of the DEPOP together with the availability of the individual-level panel data encourage the use of a Difference-in-Differences design (DiD). Furthermore, the DiD estimation method is particularly suitable for quantitative policy evaluation as it allows the user to infer causal effects on subjects affected and unaffected by the programme (Lechner, Rodriguez-Planas & Kranz, 2015).

9 .2 9 .4 9 .6 9 .8 10 Ave ra g e l o g o f in d ivi d u a l in co me 2003 2005 2007 2009 2011 2013 YEAR WEST EAST

Average log individual income

.1 5 .2 .2 5 .3 Ave ra g e e xp e ct e d g ro w th ra te o f ye a rl y in co me 2003 2005 2007 2009 2011 2013 YEAR WEST EAST

Average expected growth rate of income

2 .8 2 .9 3 3 .1 3 .2 3 .3 Ave ra g e H a p p in e ss 2003 2005 2007 2009 2011 2013 YEAR WEST EAST Average Happiness 1 2 3 4 5 Ave ra g e l o g l o ca l g o v. re v. 2003 2005 2007 2009 2011 2013 YEAR WEST EAST

(19)

Page | 19 In my thesis I apply the standard DiD regression framework, as I can divide my sample into subsets with respect to beneficiary regions and period of intervention, by means of the binary variables introduced in the previous section: Treatment and Year. The first subset, a Treatment Group, consists of observations for individuals living in the Eastern regions of Poland receiving the subsidy between the years 2007 and 2013. The second subset, a Control Group, contains data collected from the inhabitants of the remaining area of Poland, not exposed to the treatment during both pre- and post-2007 periods. Thus, the subsets derived in a more formal notation are:

a) 𝑌𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡,𝑖𝑠𝑡: outcome variable for an individual i from region s in year t receiving the

treatment;

b) 𝑌_{𝐶𝑜𝑛𝑡𝑟𝑜𝑙,𝑖𝑠𝑡}: outcome variable for an individual i from region s in year t not receiving the treatment.

Moreover, we assume that in the absence of treatment the expected outcome conditional on s and t, is equal to the sum of the time-invariant fixed effect for region s (𝐴_𝑠) and the region-invariant time fixed effect for year t (𝐵_𝑡):

(1) 𝐸[𝑌_{𝐶𝑜𝑛𝑡𝑟𝑜𝑙,𝑖𝑠𝑡}|𝑠, 𝑡] = 𝐴_𝑠+ 𝐵_𝑡

In the presence of treatment, the expected outcome is increased by additional parameter 𝛽: (2) 𝐸[𝑌𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡,𝑖𝑠𝑡|𝑠, 𝑡] = 𝐴𝑠+ 𝐵𝑡+ 𝛽

Using this setting, we can find the formula for the observed outcome variable of the following form:

(3) Y_𝑖𝑠𝑡 = 𝐴_𝑠 + 𝐵_𝑡+ 𝛽𝐷_𝑠𝑡+ 𝜀_𝑖𝑠𝑡

Where 𝑌_𝑖𝑠𝑡 is the outcome variable for individual i from region s in year t, 𝐷𝑠𝑡 is the binary

indicator equal to 1 for the Treatment Group during the period of intervention and 𝜀_𝑖𝑠𝑡 is the unobserved error term23_{. It is important to note that the observed outcome does not depend on}

individual fixed effects. Because each individual i in the sample panel does not change the region of residence across the time and the observed treatment is aggregated on the regional level, adding individual fixed effects to the model would only lead to the absorption of the

23_{In the standard DiD framework we can also find a vector of individual control variables 𝑋}

𝑖𝑠𝑡, however as it is

(20)

Page | 20 treatment indicator 𝐴_𝑠, keeping all the remaining estimated coefficients unchanged (Pischke, 2005).

The main parameter of interest, the treatment effect DiD, is denoted as 𝛽 in the Equation (3). Equation (4) presents the general formula for 𝛽 in the whole population:

(4) 𝛽 = 𝐸(𝑌_{𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡,𝑖𝑠𝑡}− 𝑌_{𝐶𝑜𝑛𝑡𝑟𝑜𝑙,𝑖𝑠𝑡}|𝑠, 𝑡)

Using the population sample we can find the DiD estimator 𝛽̂ according to Equation (5): (5) 𝛽̂ = (Y̅_{𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡,2}− Y̅_{𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡,1}) − (Y̅_{𝐶𝑜𝑛𝑡𝑟𝑜𝑙,2}− Y̅_{𝐶𝑜𝑛𝑡𝑟𝑜𝑙,1})

Where 𝑌̅ is the mean outcome variable, for either the Treatment or the Control Group in the

post-treatment (indexed as 2) and pre-treatment (indexed as 1) periods.

To estimate the DiD estimator 𝛽̂ for the sample represented by the collected dataset, I run an

OLS regression model of a following form:

(6) Y_𝑖𝑠𝑡 = 𝛼 + 𝛿𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡_𝑠+ 𝛾𝑌𝑒𝑎𝑟_𝑡+ 𝛽(𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡_𝑠∗ 𝑌𝑒𝑎𝑟_𝑡) + 𝜀_𝑖𝑠𝑡

Where the DiD coefficient of interest 𝛽 stands next to the interaction term (𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠∗ 𝑌𝑒𝑎𝑟𝑡)

and indicates the average treatment effect. The parameter 𝛿 reflects the difference in the outcome variable for the Treatment Group compared to the Control Group. The coefficient 𝛾 indicates the difference in the outcome variable for the post-treatment period compared to the pre-treatment period.

Looking at Equation (5) we can see that the DiD estimator measures the difference in the

average outcome simultaneously across the two dimensions, pre- and post-treatment period; and Treatment and Control Groups. The introduction of the double differentiation eliminates the potential bias caused by factors other than the DEPOP, which could have an impact on both groups. By applying the formula for the DiD estimator we can isolate the effect of the evaluated policy from these external factors.

4.3.2 Parallel trend assumption

In order to ensure the viability of DiD, several assumptions need to be met. First of all, we can estimate the Difference-in-Differences coefficient using the Ordinary Least Squares (OLS)

estimator (Lechner, Rodriguez-Panas & Kranz, 2015). Thus, all of the requirements applicable

for the standard OLS technique are valid likewise for the DiD. Nevertheless, there remains an

(21)

Page | 21 Treatment and the Control Group (Dimick & Ryan, 2014). Equation (7) presents the formal

notation of this assumption, where the error term stands for all the determinants of outcomes not included in the model.

(7) 𝑐𝑜𝑣(𝜀𝑖𝑠𝑡,𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠∗ 𝑌𝑒𝑎𝑟𝑡) = 0

In order to compute the DiD estimator we must assume that the absence of external treatment factors would lead to similar changes over time for both groups, therefore maintaining the parallel trend. Figure VII helps to explain this idea graphically.

Figure VII

Illustration of the common trend assumption

Note: The CG line represents the distribution of the outcome variable for the Control Group. The

dashed line shows the hypothetical counterfactual distribution of the outcome variable for the Treatment Group. The solid TG line presents the observed distribution of the outcome variable for the Treatment Group.

The solid lines illustrate the observed distribution of the outcome variable for the Control and the Treatment Group before and after the treatment. It is worth noting that the starting value for the outcome variable can differ between the groups, however, ideally the trend in the pre-treatment distribution should be identical. Accordingly, due to the initial differences, we cannot infer that the changes in the after-treatment’s outcome are solely the consequence of the application of treatment. The dashed blue line, with a slope corresponding to the slope observed for the Control Group, shows the counterfactual distribution of the dependent variable for the Treatment Group. In the DiD estimation, we look at the difference that would exist between the two groups without the implementation of the policy, in other words a difference between the implicit constant parallel trends. The estimated treatment effect is equal to the difference between the observed and counterfactual outcome for the Treatment Group in the post-treatment period.

(22)

Page | 22 The parallel trend assumption is impossible to verify as we cannot observe the future outcome for the Treatment Group without the impact of the evaluated policy. However, a common practice is to demonstrate that the trends for the period before the treatment are the same. In section 4.2, I compared the trends for the Control and Treatment Groups before and after 2007. The graphical analysis of the distribution of outcome variables leads to the conclusion that pre-treatment trends for both subgroups were similar for the material well-being indicators and the logarithm of local government revenue, whereas they differed for the happiness variable. In order to validate that the trends for the Treatment and Control Groups are parallel, I perform additional independent t-tests to prove statistical similarity between their mean growth rates before the treatment. Table II reports the p-values testing the null hypothesis that there is no difference in the mean growth rate between the Control and Treatment Groups.

At the standard 5% level of significance, average change of Log individual income and

Expected growth rate of individual income are statistically similar for both groups. Furthermore,

the mean differences for Happiness are statistically similar if we assume a lower significance level of 1%. Mean change for the variable Log local government revenue is shown to be statistically different between the groups. This finding could justify the appropriateness of the application of the DiD method for the assessment of the DEPOP effect on subjects’ material and non-material welfare. Unfortunately, I cannot determine statistically significant similarity between the mean trends for the Log local government revenue variable.

4.3.3 Clustered and robust standard errors

According to Bertrand, Duflo & Mullainathan (2004) the setup of Difference-in-Differences models makes them prone to the problem of serial correlation, resulting in inconsistent and underestimated standard errors. In other words, the models’ errors can be correlated within the subgroups of time series for a given entity. As a consequence, without controlling for the serial

Table II

Paired t-test of mean changes in outcome variables

Control vs. Treatment Group

Log local gov. revenue 0.0000

Log individual income 0.5041

Expected growth rate of individual income 0.8688

Happiness 0.0215

Notes: Table presents p-values for the paired mean comparison t-test between the Treatment and the

(23)

Page | 23 correlation, we can obtain misleadingly high t-statistics, leading to inaccurate statistical interference regarding the significance of estimated coefficients.

The authors list three of the most important sources of serial correlation for this type of regressions, all of which apply to the dataset and specifications found in this thesis. Firstly, estimated models use extensive time series; in the given study the time dimension is divided into six periods, rather than the two basic pre- and post-treatment periods. Secondly, the outcome variables in DiD models are usually significantly serially correlated. As the indicators of material and non-material well-being are measured for one individual over time, their correlation between different years is highly likely. The last reason for the potential existence of serial correlation is the low variance of the main regressor 𝐷𝑠𝑡 within the subset of each

region.

Due to the presence of the above characteristics, I run DiD regressions with clustered standard errors for each region s. As a result, models’ errors can be correlated within the group of observations for each region and are assumed to be uncorrelated between different regions. The division into clusters is made for regions instead of individuals, since clustering on a gradually superior aggregation level is considered better practice (Cameron & Miller, 2015). Finally, I use robust standard errors to avoid issues related to the potential heteroskedasticity of error terms.

4.3.4 DiD limitations

When using the Difference-in-Differences method for the DEPOP evaluation it is important to know its potential limitations. Firstly, there is potential for bias in the measurement of the impact of the EU’s financial aid. The systematic changes observed in the outcome may be thoroughly different for the Treatment Group, however, this may not be a the impact of the programme but rather some external factors influencing only the Treatment Group and having no impact on the Control Group (Imbens & Wooldridge, 2007). The range of possible determinants causing these variations is wide. For example, an increase in the perceived level of happiness for individuals living in the East could be related to particularly stable family relationships or the lower risk of contracting cancer, which is specific to this region (Kowalski, 2013), as opposed to the financial aid from the EU.

The second source of concern is related to the initial differences between the Control and the Treatment Group. For the selected sample, all of the outcome variables reach different pre-treatment values for the two groups. Ideally, for the DiD analysis, these two subsamples should

(24)

Page | 24 match one another. There can be many unobserved characteristics contributing to the observed differences between the subsets, which cannot be controlled for and could lead to the omitted variable bias. Although this drawback does not prevent the usage of the DiD model, it still can be the source of the imprecise estimation of coefficients.

5. Main results

This section presents and interprets the results of the main Difference-in-Differences estimations conducted in the study. Moreover, to ensure the structural validity of evaluated models, I conduct robustness tests.

5.1. Impact of DEPOP on regions’ incomes

As the starting point of my analysis I examine whether the additional funds from the EU received by the Eastern region’s local governments had, in fact, increased their revenues. Hence, I define a relationship between the logarithm of local government revenue and the received treatment:

(1) ln (𝐿𝑜𝑐𝑎𝑙 𝑔𝑜𝑣𝑒𝑟𝑛𝑚𝑒𝑛𝑡 𝑟𝑒𝑣𝑒𝑛𝑢𝑒)_𝑖𝑠𝑡 = 𝛼₁+ 𝛿₁𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡_𝑠+ 𝛾₁𝑌𝑒𝑎𝑟_𝑡+ 𝛽₁(𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡_𝑠∗ 𝑌𝑒𝑎𝑟_𝑡) + 𝜀_𝑖𝑠𝑡24

The regression presented in Equation (1) is estimated in order to support the main research hypothesis, that the DEPOP had a positive impact on individuals’ well-being. In order to explain the logic behind this “first stage” model, I need to make two assumptions. First, I loosely refer to the local government revenue as a proxy for regions’ public investments. This assumption is supported by the fact that during the period between the years 2003 and 2013, the share of the Polish regions’ revenues on public investment was substantial and ranged from 35% to 47% (Śmiechowicz, 2014). Second, in accordance with economic theory, we assume that the well-being of the society is highly correlated with the supply of public goods and services, i.e. public investments. More formally, increased regional public investment should contribute to the increased utility of its inhabitants (Dada, 2015). Furthermore, as the increase in local government’s revenues is non-distortionary, it has no negative impact on the utility of taxpayers (Dahlby & Ferede, 2012). Combining these two assumptions, we can expect that if the DEPOP

24_{In the estimated model, the local government’s revenue is assigned to each individual i in order to keep the} structure of individual-level panel data. In other words, for all the individuals living in region s, values of the outcome variable vary for different years but not for individuals.

(25)

Page | 25 increases local government revenue, it should also improve the material and non-material well-being of inhabitants of the Eastern parts of Poland via the effect of public investments.

Table III examines the estimation result for the first regression.

All of the coefficients are significant at the 1% level of significance. Most importantly, the DiD estimator 𝛽1 (Diff-in-Diff in the table) is positive and indicates that the DEPOP has significantly

increased the revenues of regional governments in Eastern Poland. This finding is in line with expectations; a substantial money influx from the EU received by the beneficiary regions should lead to a rise in their yearly revenues. The coefficient standing next to the variable Year, equal to 1.69, indicates that between the years 2007-2013 the local governments’ revenues for the Control Group were on average higher by 169%25 compared with the preceding period. Furthermore, being affiliated with the Treatment Group implies having on average lower local government revenues by 93.9%, compared with the Control Group during the pre-treatment period.

The estimated regression constitutes a sufficient baseline, supporting the rationality behind the main research dilemma. As local governments increase their financial resources by benefiting from the grant, we can anticipate that the additional funding might also have a positive impact on inhabitants’ prosperity and happiness.

5.2. Impact of DEPOP on individuals’ well-being

25_{I estimate a log-level model; Therefore, with a one unit change of the regressor, the dependent variable} changes by 𝛽 ∗ 100%.

Table III

Log local government revenue

Year 1.690 (0.040)** Treatment -0.939 (0.259)** Diff-in-Diff 0.894 (0.079)** Constant 3.630 (0.242)** R2 _0.50 N 4,584

Notes: Robust standard errors are clustered at the regional level and are shown in parentheses; * p<0.05; ** p<0.01

(26)

Page | 26 Next, I focus my analysis on the three regression models accounting for the impact of the EU’s programme on different measures of individual’s material and non-material well-being: their present and expected income and level of happiness. Equations (4)-(6) show the formulas for the estimated regressions:

(2) ln (𝐼𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙 𝑖𝑛𝑐𝑜𝑚𝑒)𝑖𝑠𝑡 = 𝛼2+ 𝛿2𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠+ 𝛾2𝑌𝑒𝑎𝑟𝑡+ 𝛽2(𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠 ∗ 𝑌𝑒𝑎𝑟𝑡) + 𝜀𝑖𝑠𝑡 (3) ln (𝐸𝑥𝑝. 𝑔𝑟𝑜𝑤𝑡ℎ 𝑟𝑎𝑡𝑒 𝑜𝑓 𝑖𝑛𝑑. 𝑖𝑛𝑐𝑜𝑚𝑒)_𝑖𝑠𝑡 = 𝛼₃+ 𝛿₃𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡_𝑠+ 𝛾₃𝑌𝑒𝑎𝑟_𝑡+ 𝛽₃(𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡_𝑠∗ 𝑌𝑒𝑎𝑟_𝑡) + 𝜀_𝑖𝑠𝑡 (4) ln (𝐻𝑎𝑝𝑝𝑖𝑛𝑒𝑠𝑠)_𝑖𝑠𝑡 = 𝛼₄+ 𝛿₄𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡_𝑠+ 𝛾₄𝑌𝑒𝑎𝑟_𝑡+ 𝛽₄(𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡_𝑠∗ 𝑌𝑒𝑎𝑟_𝑡) + 𝜀_𝑖𝑠𝑡

Table IV contains the aggregated results of estimations for all three models.

Surprisingly, we find that none of the DiD parameters (𝛽2, 𝛽3, 𝛽4) is statistically significant.

Therefore, we have to reject the main research hypothesis that the DEPOP had a positive and significant impact on Eastern regions’ inhabitants’ well-being. What is more, the magnitude and the sign of average treatment effects for each model, if they had been significant, would have indicated a very small and negative impact of the DEPOP on each well-being indicator. From a purely hypothetical point of view, this would suggest that the introduction of the DEPOP led to a deterioration of people’s material prosperity and weakened their expectations about their future incomes. Conversely, a negative DiD coefficient for the model presented in column (III) would indicate that the evaluation program had a positive effect on subjects’ perception of happiness, due to the design of the dependent’s variable scale. Nevertheless, as

Table IV

Log individual income

Expected growth rate of individual income Happiness Year 0.392 -0.071 -0.094 (0.015)** (0.011)** (0.024)** Treatment -0.147 0.013 0.234 (0.062)* (0.027) (0.063)** Diff-in-Diff -0.010 -0.018 -0.080 (0.031) (0.024) (0.057) Constant 9.328 0.263 2.950 (0.039)** (0.017)** (0.046)** R2 _0.11 _0.01 _0.01 N 4,584 4,584 4,584

Notes: Robust standard errors are clustered at the regional level and are shown in parentheses; * p<0.05; **

(27)

Page | 27 the main coefficient of interest is insignificant for each outcome variable, I conclude that the EU programme did not improve the quality of life for Poles living in Eastern voivodeships.

5.3. Robustness checks

In order to assess the structural validity of the core regressions, I perform additional robustness checks. The objective of these tests is to explore whether the main estimated coefficients have a tendency to vary for different sample selection or model specification. Ideally, the applied modifications should not significantly change either the magnitude or the sign of the models’ parameters. Nevertheless, possibly small variations in coefficients are highly desirable, as only with robust results causal interference is permitted (Lu & White, 2014).

Consequently, I repeat the estimations of the three main regressions26_{for well-being indicators,}

each time using different subsamples. First, I generate two subsets conditional on individuals’ age. The subsample Young groups people below the age of 45, whereas Old those above 45 years old. Next, I repeat the exercise and divide the sample depending on subject’s gender. In the last robustness check I expand the model by adding covariates for basic characteristics of each individual: age, gender and a dummy indicating whether the person lives in a city.

Tables V, VI and VII show the coefficients estimated with respect to the above criteria for the set of three regressions.

26_{Due to the absence of variables for regions’ characteristics in the individual-level panel dataset, I cannot} perform similar robustness checks for the first stage regression.

Table V

Log individual income Expected growth rate of

individual income

Happiness

Young Old Young Old Young Old

Year 0.535 0.372 -0.115 -0.028 -0.053 -0.137 (0.037)** (0.012)** (0.031)** (0.014) (0.060) (0.032)** Treatment -0.057 -0.178 -0.057 0.046 0.200 0.233 (0.067) (0.063)* (0.081) (0.021)* (0.132) (0.090)* Diff-in-Diff -0.182 0.035 0.039 -0.048 -0.175 -0.056 (0.045)** (0.042) _(0.087) (0.014)** (0.074)* (0.072) Constant 9.361 9.313 0.418 0.196 2.800 3.014 (0.039)** (0.044)** (0.023)** (0.019)** (0.057)** (0.061)** R2 _0.15 _0.11 _0.02 _0.01 _0.01 _0.01 N 944 3,640 944 3,554 944 3,554

Notes: The sub-column Young tabulates the estimation results for the subsample of individuals below 45 years old.

The sub-column Old tabulates the estimation results for the subsample of individuals above 45 years old.Regressions estimated in columns (IV) and (VI) use unbalanced panels, due to missing observations for the variable Old. Robust standard errors are clustered at the regional level and are shown in parentheses; * p<0.05; ** p<0.01

(28)

Page | 28 When comparing the estimated coefficients for the first robustness check reported in Table V, we can see that the parameters presented in column (II) are more similar to their equivalents in the main regression tabulated in Table IV, compared with those presented in column (I). However, the negative DiD coefficient becomes statistically significant at the 1% significance level for the subsample of younger generation. The analogous situation can be observed for the regression of Expected growth rate of individual income, where in column (IV) of Table V, the negative DiD coefficient for the older generation is also shown to be significant. The findings for the third model match those observed for material well-being indicators. Again, the magnitudes and signs of coefficients closely correspond to the main model’s parameters, while the negative DiD parameter for younger individuals becomes significant.

Table VI

individual income

Happiness

Men Women Men Women Men Women

Year 0.417 0.371 -0.073 -0.070 -0.150 -0.046 (0.023)** (0.022)** (0.010)** (0.017)** (0.024)** (0.029) Treatment -0.269 -0.041 0.039 -0.005 0.281 0.192 (0.078)** (0.057) (0.041) (0.034) (0.084)** (0.085)* Diff-in-Diff 0.009 -0.024 -0.032 -0.006 -0.034 -0.119 (0.051) (0.030) (0.038) (0.027) (0.074) (0.052)* Constant 9.465 9.210 0.289 0.240 2.899 2.995 (0.045)** (0.040)** (0.019)** (0.018)** (0.048)** (0.055)** R2 0.14 0.11 0.01 0.01 0.02 0.01 N 2,094 2,490 2,094 2,490 2,094 2,490

Note: The sub-column Men tabulates the estimation results for the subsample of men. The sub-column Women

tabulates the estimation results for the subsample of women. Robust standard errors are clustered at the regional level and are shown in parentheses; * p<0.05; ** p<0.01.

Table VII

individual income Happiness Year 0.418 -0.039 -0.127 (0.014)** (0.012)** (0.030)** Treatment -0.117 0.017 0.216 (0.046)* (0.030) (0.058)** Diff-in-Diff Age Gender City -0.010 -0.018 -0.080 (0.031) (0.024) (0.057) -0.004 -0.005 0.005 (0.001)** (0.000)** (0.003)* 0.216 0.037 -0.105 (0.040)** (0.012)** (0.043)* 0.321 -0.044 -0.148 (0.035)** (0.013)** (0.069)* Constant 9.358 0.536 2.755 (0.041)** (0.021)** (0.127)** R2 _0.22 _0.07 _0.03 N 4,584 4,584 4,584

(29)

Page | 29 The patterns in coefficient changes found for the first robustness check are also visible for samples divided according to individual’s gender, presented in Table VI. This time, however, the negative DiD coefficient becomes significant only for the regression estimated for

Happiness, at a high significance level of 10%. What is more, when controlling for the

covariates in Table VII, the coefficients do not differ significantly either, keeping the DiD average treatment effect insignificant for all the models.

The main conclusion, which can be derived from the robustness checks, is that the examined models pass the tests for structural validity. Firstly, there is no clear pattern for the varying significance of DiD parameters, which suggest that heterogeneity within the sample is not the source of the potential bias. Furthermore, the magnitudes of coefficients remain within the initial range, with an average deviation from the main regression’s parameters of 0.1 units.

6. Interpretation of the results

Although in section 5.1, it has been proved that the DEPOP significantly increased the revenues of regions’ local governments, all three regressions presented in section 5.2 indicate no impact of the programme on the outcome of main interest: individuals’ well-being. The lack of a positive effect of the programme at the micro level is surprising, as a substantial rise in local governments’ revenues would be expected to improve the quality of life of their inhabitants via large public investments in Eastern Poland. As stated before, from the theoretical point of view, ameliorated public capital should stimulate household welfare and/or life accomplishment.

What are the possible explanations for the absence of the positive effect of the DEPOP on well-being of inhabitants of Eastern Poland? Is it possible that the special regional policy of the EU failed to achieve its main development goal? To answer the above questions I first analyse the ex-post values of the main macroeconomic indicator used as a reference before the introduction of the DEPOP: GDP per capita for each region. Figure VIII provides the graphical presentation of GDP per capita for each voivodeship in Poland in 2015.

(30)

Page | 30

Figure VIII

GDP per capita in EUR for NUTS2 Polish regions in 2015

Note: Highlighted regions are the beneficiaries of the DEPOP.

Two remarks follow from this figure. First, since the year 2004, the GDP per capita in 2015 has on average increased by around 80% in the whole country. Second, the relative gap between the beneficiaries of the DEPOP and the remaining parts of Poland remained unchanged. These findings suggest that the evaluated regional policy did not constitute a sufficient tool for the alleviation of the development gap between Eastern and Western Poland. In other words, that the DEPOP failed to fully achieve its main goal. Another possible interpretation is that the pace of economic growth observed for Western regions was significantly higher compared with the Eastern part of Poland and the money influx from the EU could only keep the disproportion constant. Combining the performance of regional GDP per capita over time with the results obtained for the microeconomic data, builds a base of evidence in favour of the first interpretation. Referring to the main research assumption, a successful implementation of the programme should have a positive impact on well-being of inhabitants of Eastern Poland, even with the observed faster economic growth in Western regions. This conclusion is highly surprising, since as has been summarized in the Literature Review section, the process of the DEPOP’s implementation and execution was evaluated positively by the Polish government and the ERDF. Nevertheless, as raised previously in section 3.3, we can be sceptical when looking at some of the figures presented in the official evaluation reports due to the unsatisfying quality of the evaluated data and problems with the isolation of the EU’s policy effect on economy (Bachtler & Wren, 2007).

31 600 16 100 14 100 13 600 18 500 14 100 14 300 14 000 17 800 22 100 21 500 19 000 16 800 1 6 5 0 0

The impact of the EU "development of Eastern Poland" operational programme on individuals' well-being