Database description

The analyses performed in this article rely on the European Survey of Income and Living Conditions (EU-SILC), a survey that contains datasets at individual and household level. Furthermore, data are classified by regions – in particular, EU NUTS-II regions, which correspond to the Spanish Autonomous Communities (CCAA). They are first level political and administrative divisions and each one has its own organic laws known as Statutes of Autonomy, which determine the powers they have. They are also divided into provinces (EU NUTS-III regions).

Specifically, in this article we use the EU-SILC surveys conducted in 2011 and 2019 which provide income variables referred to years 2010 and 2018. The use of these two surveys is particularly relevant since they include an ad-hoc module which incorporates variables regarding individuals’ family background.

Our analysis is focused on the seventeen Spanish Autonomous Communities and the sample is restricted to individuals aged twenty five to fifty nine years because parental background variables are only provided for individuals within this range of age.

The variables we use refer to individuals’ circumstances, i.e. factors that are beyond individuals’ responsibility. For this purpose, we use variables regarding individuals’ family background when they were children. The first is Parental Education, which is constructed using the maximum level of education of both progenitors classified into three categories: low if both progenitors have at maximum a degree of compulsory education, medium, if at least one progenitor has completed secondary education and high, when at least one progenitor has completed tertiary education.

Similarly, we use the variable Parental Occupation, which refers to the occupation of individuals’ progenitor when they were children. This variable is also divided into three categories, according to the ISCO-08 jobs classification. More specifically, these three categories are low skilled when both parents work in elementary occupations (groups eight and nine of ISCO-88 classification), medium skilled when at least one parent works in an occupation comprised in groups five to seven of the classification and high skilled if at least one progenitor works in a high skilled occupation (groups one to four in ISCO-08)Footnote ¹.

We also include the variable Childhood Financial Difficulties, which is a perception-based measure that considers respondent’s feeling about the financial situation of the family when the respondent is around fourteen years old. It is divided into four categories: Bad/very bad, Moderately bad, Moderately good and Good/very good.

The variables used as circumstances correspond to a period prior to the one for which the income of individuals is analysed. These variables correspond to the childhood period of the individuals, since the idea is that they provide an approximation of their social and family environment and that they are circumstances – that is, that they do not depend on the responsibility of the individuals and, clearly, individuals cannot occur to them due to situations that occurred during their childhood.

Finally, we include a variable referred to income, which will be used to measure inequality – we use the equivalised disposable income of households, which is equivalised by considering the structure of each household. Specifically, this variable is constructed according to the equivalence scale used by EU-SILC, which is the Organisation for Economic Cooperation and Development (OECD)-modified scale: $e = 1 + 0.5\left( {{N_{{{14}^ + }}} - 1} \right) + 0.3{N_{{{13}^ - }}}$ where ${N_{{{14}^ + }}}$ is the number of household members who are over fourteen years and ${N_{{{13}^ - }}}$ the number of members who are thirteen or younger.

The equivalised disposable income is considered a good proxy of the available income individuals have. In addition, this variable has been used in other studies with EU-SILC data referred to education and inequalities – such as Brzezinski, Reference Brzezinski2015; Marrero and Rodríguez, Reference Marrero and Rodríguez2012; Palomino et al., Reference Palomino, Marrero and Rodríguez2016; Suárez Álvarez and López Menéndez, Reference Suárez Álvarez and López Menéndez2018b, Reference Suárez Álvarez and López Menéndez2018a.

Table 1 provides a summary of the descriptive statistics at the regional level and for the two years analysed, 2011 and 2019. More specifically, it shows the share of the population within each category of the above-mentioned variables.

Table 1. Main descriptive (part 1)

Table 1. Main descriptive (part 2)

Additionally, Table 1 shows the sample size for each region and year. It can be noticed that in some regions the sample size is not very large, the survey used (EU-SILC) provides weights to make the results representative. Nevertheless, caution is advised when interpreting the results.

Estimating inequality of opportunity

This Section is devoted to estimate income inequality and inequality of opportunity (IOp) and to compute the contribution of the variables used to circumstances to IOp with the aim of showing at what extent family background characteristics and other circumstances are affecting individuals’ performance in terms of income.

For this purpose, in addition to estimating income inequality, we compute inequality of opportunity (IOp henceforth) for the Spanish regions and then we analyse the contribution of the variables used as circumstances to this indicator.

The basis of the study of IOp can be found in Roemer (Reference Roemer1993) and Yitzhaki and Schechtman (Reference Yitzhaki and Schechtman2013); later on Roemer (Reference Roemer1998) formalise this concept and distinguish between circumstances, understood as factors beyond individuals’ control, and efforts, that can be attributed to individuals’ performance and commitment. In this context, the analysis of IOp tries to compute the part of inequality due to the first kind of factors, in which parental background characteristics are included.

To know the effect that parental background characteristics have in inequality and more specifically in inequality of opportunity, we estimate the indices of both sorts of inequalities for each of the seventeen Spanish regions analysed before computing the contribution of these characteristics.

Firstly, to estimate IOp, in addition to previous variables referring to individuals’ family background, we include two more variables regarding individuals’ personal circumstances in order to obtain reliable estimates and reduce the omitted variables bias. These two variables are: Gender, with two categories, female and male and Country of birth with two categories Foreign-born persons and Spanish-born.

To estimate IOp we use the ex ante parametric method (Ferreira and Gignoux, Reference Ferreira and Gignoux2011), although there are many other alternative methodologies (see Ferreira and Peragine, Reference Ferreira and Peragine2015; Ramos and Van de Gaer, Reference Ramos and Van de Gaer2016; Roemer and Trannoy, Reference Roemer and Trannoy2016). We have decided to use the ex-ante parametric method since this is the most widely used procedure in the IOp literature and thus allows comparability with other studies. Furthermore, this method has the additional advantage that it does not require including information referred to the level of effort.

In order to compute IOp we divide individuals into types $T$ , being each type formed by individuals that share the same categories of each circumstantial variable. In our case we have five variables Parental Education, Parental Occupation, Childhood Financial Difficulties, Gender and Country of birth with three, three, four, two and two categories each, thus leading to a total of 144 different types of individuals. These are the variables considered to be beyond individuals’ responsibility and therefore treated as circumstances.

Then, we estimate individual incomes following the expression: $ln{y_i} = {C_i}\beta + {u_i}$ , where ${C_i}$ denotes the different variables used as circumstances and $\beta $ represents the effect these circumstances have on income. Once the equation is estimated we get the fitted values: ${\hat \mu _i} = exp\left( {{C_i}\hat \beta } \right)$ which are a counterfactual distribution of income that depends only on the circumstances. (Tables A1 and A2 in the Appendix section show the results of the regressions).

The inequality of opportunity indices can then be obtained in absolute (IO_pA) and relative (IO_pR) terms quite straightforward: $IO{p_A} = I\left( {{{\hat \mu }_i}} \right)$ and $I{O_{pR}} = {{I\left( {{{\hat \mu }_i}} \right)} \over {I\left( y \right)}}$ .

Table 2 shows the results of both income inequality and IOp in the equivalised disposable income through two indices, the Gini and the GE (0). We use the Gini index due to its easy interpretation and because it is widely used to measure inequality. Generalised Entropy GE (0) is also computed since this is the only measure with a path-independent decomposition (Foster and Shneyerov, Reference Foster and Shneyerov2000) using the arithmetic mean as reference and can be easily used to compute the contribution of the circumstances to IOp.

Table 2. Estimates of inequality

According to the obtained results, North-western regions (Galicia, Asturias, Cantabria and Basque Country) and also two central regions (Madrid and La Rioja) experienced an increase in both income inequality and IOp whilst for the remaining regions we observe a decrease in both inequalities, with the exception of Castile and Leon and Castilla-La Mancha for which we observe an increase.

Likewise, in most regions we can see an increase in relative IOp. That is to say, the share of inequality of opportunity in overall inequality increases; and consequently, factors that are beyond individuals’ responsibility become more important in shaping individuals´ well-being and performance.

To answer our question ‘to what extent are family background characteristics important in shaping inequalities?’, once we get the estimates of IOp we estimate the contribution of the circumstances and pay attention to the amount of IOp due to Parental Education and Parental Occupation. To this end, we decompose IOp using the Shapley value procedure (Shorrocks, Reference Shorrocks2013), and we compute the marginal effects each variable has in IOp in terms of the GE (0) index.

Table 3 shows the average contribution of family background characteristics to IOp, these circumstances are of great relevance in shaping inequalities, the three of them account for about half or more of the total IOp. The exceptions are four regions of the northeast (Navarre, La Rioja, Aragon and Catalonia) where the relative importance of family background characteristics is lower.

Table 3. Average contribution of family background circumstances

As it can be seen in this Section, both inequalities, income inequality and inequality of opportunity, both in levels and evolution, differ between regions. Additionally, the importance of family background characteristics to determine the levels of inequality of opportunity also varies greatly. For this reason, it is relevant to analyse the determinants of these regional differences.

Drivers of inequalities

The aim of this section is to know which are the main drivers of the observed inequalities across regions. We analyse the reasons behind the differences in inequality levels between regions and how these differences can be reduced.

As a first step, we look at the association between inequality levels and several variables. We analyse the correlations between inequality variables and potential determinants of inequality, we include economic, demographic and educational variables. Table 4 shows the pairwise correlations, it can be seen that, as expected both sorts of inequality are positively and strongly correlated, as Figure 1 also illustrates.

Table 4. Pairwise correlations

Significant correlations are in bold

Figure 1. Relationship between Income inequality and inequality of opportunity.

Regarding the economic variables included, income inequality is positively and significantly correlated with the unemployment rate, therefore, those regions with higher rates of unemployment also suffer from higher levels of income inequality. Nevertheless, its effect on IOp is not significant. The other economic variables included are the per capita Gross Domestic Product (GDP), the average wage, the wage of the tenth percentile and the wage of the ninetieth percentile. The GDPpc, the average wage and the wage of the 10^th percentile are negatively correlated with income inequality and positively correlated with IOp, whilst the wage of the ninetieth percentiles is positively correlated with both sorts of inequalities. Nonetheless, these correlations are weak and non-significant.

With regard to demographic variables, we include the share of population over sixty five years and the dependency rate, both variables are negatively correlated with the levels of inequality. Moreover, the correlation between the population shares over sixty five years and the inequality variables is quite strong and significant, as can be also seen in Figure 2. It implies that regions with aged population have less inequalities, possibly because retirement benefits an equalising effect on income. We also include the net inter-regional migration rate but its correlation with inequality is weak and not significant.

Figure 2. Relationship between IOp and the share of population over 65 years.

Educational variables seem to have the highest levels of correlation with inequality variables. As educational variables, we include five variables at regional level. The first of them is the share of individuals with tertiary education, which is negatively correlated with both income inequality and IOp, although this correlation is not significant. Then, the educational drop-out rateFootnote ², which is positively and significantly correlated with income inequality and the success rate of high-secondary school which is negatively and significantly correlated with income inequality.

The two educational variables that show the highest level of correlation with the inequality variables are the average student/teacher ratio and the education expenditure per studentFootnote ³. The students/teacher ratio correlation implies that the higher the average number of students per teacher, the higher the levels of inequality and IOp. Additionally, the correlation between the education expenditure per student and the inequality variables shows that the greater the education expenditure the lower the inequality levels, this relationship is also illustrated by Figure 3. These results show that there is a significant association between the levels of inequality and educational resource endowments, both economic (educational expenditure) and human capital (students/teacher ratio). It is observed that those regions with a greater endowment of educational resource exhibit lower levels of inequality.

Figure 3. Relationship between IOp and the Education expenditure per student.

In addition to the pairwise correlations, we perform several regressions to see if the previously observed associations also entail causal relationships. Table 5 shows the results of the regressions – as expected not all positive and significant pairwise correlations are significant in the regression analysis. Moreover, regression results are not robust when comparing fixed effects and pooled Ordinary Least Squares (OLS).

Table 5. Regressions’ results

Standard errors in parentheses.

***p < 0.01, **p < 0.05, *p < 0.1.

Still, the regression analysis shows that an increase in unemployment rate would cause an increase in the levels of income inequality. Additionally, the expenditure on education is significant to explain IOp, showing that an increase in education expenditure would reduce inequality of opportunity.

Summarising, the analysis carried out in this Section highlights the great relevance of education in reducing inequalities, especially the expenditure on education and the number of teachers.