Heterogeneity of government social spending in European Union countries

Abstract

Social expenditure plays an important role in European Union (EU) countries. It improves the lives of citizens whose welfare is endangered due to poverty or illness. However, social expenditure represents a considerable share of the budgets of EU member states. Despite evident similarities in their levels of development, EU countries show apparent differences in social expenditure levels. Therefore, this work aims to determine the similarities and differences between EU countries in this regard. The analysis uses clustering methods, such as hierarchical cluster analysis and the k-means, to divide countries into homogeneous groups. The research demonstrates significant differences between EU countries in the years 2008–2018, which resulted in a low number of objects (countries) in the identified groups. In the case of 6 out of 28 countries, it was not possible to assign them to any group. The research proves that EU countries should take more care when organising their social policy, taking into consideration cultural and social factors.

Introduction

The government social expenditure is part of social policy and comprises a considerable share of the EU member states’ budgets. These funds are dedicated to diverse social security programmes. The main priorities of government social expenditure focus on combating social exclusion and reducing income inequality [7, p. 18]. However, it has been pointed out that the amount of government expenditure does not achieve the expected results in reducing the inequality gap. Findings from these studies prove that the same improvements in the social situation could be achieved using better policy rather than higher volumes of spendings [2, 3, 29, 48].

Many researchers have pointed out that the social policy in the EU is uncoordinated and difficult to understand [11, p. 23], even though social policy was included in the Lisbon’s Strategy launched in 2000, and its descendant, the Europe 2020 strategy, launched in 2010. The goal set in the Europe 2020 strategy for smart, sustainable and inclusive growth was to ensure an innovative, sustainable and inclusive economy in the EU. The strategy presented a set of indicators to evaluate the progress of development. All countries were obliged to attain their particular targets, which was intended to boost the economy and society of the whole EU. One of the three priorities was ‘Inclusive growth: fostering a high-employment economy delivering social and territorial cohesion’ [18, p. 3]. The aim was to reduce the number of people at risk of poverty or social exclusion by over 20 million. Other targets related to the remaining two priorities included: increasing the employment rate of the population aged 20–64, reducing the dropout rate in early education and increasing the share of the population aged 30–34 having completed tertiary education. Unfortunately, these indicators still have not reached their targets, with the indicator of poverty in the EU staying above the mark [17]. One reason for this situation is the heterogeneity of EU societies [10, 36] and the differences in social policy across EU countries. The diversity of EU economies and their social politics limits the development of EU countries [33]. Yet, the full integration of social politics is not the proper solution, as the heterogeneity of countries must also be taken into consideration [26]. That is why there is a need to recognise these differences.

EU countries are highly developed, and they have complex public schemes of social securities, often supported by private programmes. However, there are certain differences in social policies that negatively influence social cohesion [16]. Some of these differences are observed in the governments’ social expenditure, in the structure of this expenditure and in social policies’ effectiveness [8, 15]. The reasons for the diversity in the EU countries’ economies are attributed to cultural and language factors [27, 30]. Chen [14] also adds that cultural factors are subject to change due to globalisation. He argues that the development of tourism and migration for jobs brings about changes in the values and principles that consumers follow.

Institutional factors are another aspect influencing the differentiation of countries. These factors include the level of development of financial markets, the accessibility of specific financial instruments [8, 9, 40] and such variables as the functioning of tax and pension schemes. The historical background discussed by Fessler and Schürz [23] is of no less importance. The more developed domestic financial markets are in the degree to which they allow households access to diverse financial instruments, the greater the knowledge of these instruments is in society and the higher preference to use them. The development of financial markets and historical circumstances are believed to be the main reasons for differences between Western European and Central and Eastern European (CEE) countries [47]. The same historical circumstances influence the lower propensity to save but, at the same time, a higher tendency to demand social aid among the CEE countries’ citizens. Social policy is also connected with the countries’ welfare model. Antonelli and De Bonis [6] identify four such models among the original EU countries: the Scandinavian model (Denmark, Finland and Sweden), the continental model (Austria, Belgium, France, Germany and Luxembourg), the Anglo-Saxon model (Ireland and the United Kingdom) and the Mediterranean model (Italy, Spain, Greece and Portugal). Multiple studies focus on a synthetic measure of the efficiency of social expenditure [15, 34, 44, 50]. However, studies focussed exclusively on the structure of the government social expenditure are lacking.

It is assumed that social security contributions paid by persons receiving income are the primary source of financing this expenditure. In practice, the budgets of social security schemes are often not balanced, and they have to be funded from other sources of budget revenues. Additionally, it was proven that there is some negative impact of social expenditure within social security schemes on private saving and employment of women [38].

In the hypothesis of the life cycle [4], social security contributions, mainly retirement pensions, are viewed as pseudo-tax-reducing current revenues, which negatively affect private finances. Across the whole economy, their impact depends on the population’s age structure, and hence, on the relation between the amount of the contributions paid and the amount of securities paid out [20, p. 5].

Empirical analyses carried out in the United States pointed out that the emergence of social securities schemes played a significant role in reducing aggregated savings and capital resources in the US economy. Experts explained this phenomenon in two ways: from the perspective of the elderly and the youth [38, p. 316]. Among older people, access to pension schemes, which give entitlement to some income, considerably increased early [12, p. 372]. Regarding younger persons, access to social security funds within social security schemes makes them less willing to accumulate financial resources towards future pensions, which has an adverse effect on private savings. The final conclusion drawn from the long-term analyses indicates the benefits guaranteed by social security systems have a negative impact on household savings [21].

The introduction of family and child benefits was supposed to lessen the work–family conflict [22] and improve women’s economic activity. Its aims were twofold: financial support for poor children to enhance their economic situation and financial support to young families to raise fertility rates in countries with a high demographic dependency ratio. These aims are crucial given ageing societies in EU countries. Most EU countries have implemented some form of child benefits [13]. However, it is observed that after such implementation, the employment rate of women with children often decreases [42].

At the onset, social securities schemes were subsidised by employers or state budgets. At present, a number of countries seek solutions that would support non-profitable social securities schemes. Especially given ageing societies and longer life expectancy, the relation of revenues from contributions to social expenditure continues to deteriorate. Additionally, private persons covered by social securities schemes do not take actions towards accumulating savings they could benefit from during their retirement. This issue refers, in particular, to persons with high earnings [28]. Thus, the problem is how to respond to the needs of society demanding social support and at the same time keep national public spending under control and, most often, reduce it [5].

Therefore, this work aims to determine the similarities and differences between EU countries in the structure of their social expenditure. Understanding these similarities and differences may help better accommodate various social policies to fight inequality and poverty among EU citizens.

The paper is structured as follows. “Methods” section presents the data and methodology used. “Results and discussion” section gives the results and the discussion, and “Conclusions” section presents the conclusions.

Methods

The data analysed in this study came from the Eurostat database, General government expenditure by function (COFOG [gov_10a_exp]), and cover the years 2008–2018. For the purpose of international comparison, the data used are presented as a percentage of the gross domestic product (GDP). The EU statistics divide social spending into nine groups: “Sickness and disability”, “Old age”, “Survivors”, “Family and children”, “Unemployment”, “Housing” and three others hereafter summed up: “R&D”; “social protection” and “social exclusion Spending on “Sickness and disability” is the healthcare system’s expenditure on support for the disabled and their caretakers. The “Old age” group represents spending on pension and retirements schemes. The spendings on “Survivors” are pensions paid by the government to widows, widowers and orphans of persons entitled to social support who have not acquired these rights themselves. “Family and children” is a group of expenditures consisting of childbirth benefits, statutory maternity pay, supplementary feeding of children and other family members and family allowance, among others. “Unemployment” refers to expenditures on unemployment benefits and programmes such as vocational training and retraining, while “Housing” refers to expenditures on housing benefits and building and maintaining public apartments.

The following indicators of distribution were calculated to analyse differences in government social spending between EU countries: coefficient of variation (V), skewness (A_s) and excess kurtosis. The coefficient of skewness, A_s, measures the skewness of the distribution of a variable in relation to its average. A positive value of this coefficient, so-called right-skewness, indicates that more than 50% of observed objects have a lower value than the variable’s average [45, p. 26].

The analysis of the similarities and dissimilarities of objects (countries) was completed using clustering methods. Grouping, also referred to as clustering or cluster analysis, is a technique of data mining. Similar data are combined into homogeneous groups (clusters) without prior definition or knowledge of these groups’ characteristics [1, p. 16]. The main aim of clustering is to identify a structure of unmarked sets of data by objectively grouping them into homogeneous groups while seeking maximum similarity of internal objects within groups and the highest external diversity between groups [51, p. 1857]. The classification methods are not connected with the determination of reasons for similarities and differences identified by using these same methods [31, p. 64].

When analysing clusters, researchers most often apply at least two methods to assign objects to groups. This study used methods based on partitions, including k-means clustering and hierarchical clustering [51, p. 1857]. When the taxonomy of data is in the form of time series, additional solutions may be utilised. Of these, the simplest one used in this study was the analysis of each examined period separately in order to then make the definitive determination of the number of groups and assignment of the objects by comparing the results [37, p. 100].

The hierarchical cluster analysis belongs to the category of agglomerative methods. However, a target number of groups is not assumed a priori [24, p. 579]. Instead, the focus is on the distances between objects and groups that can be measured with different methods [39, p. 373]. In particular, Ward’s minimum variance criterion is a recommended method [43, p. 277]. In the k-means clustering method, a certain number of subgroups into which analysed objects are to be divided is determined in advance [41, 46, p. 95].

One of the tests assessing the correctness of a certain selected number of groups or clusters is based on the pseudo-F statistic [19, 25]. In this test, the statistic F’s value from the variance analysis for a given number of clusters is compared with the statistic F computed for a smaller or bigger number of clusters. Linked with the increase in the number of clusters, a noticeable increase in the value of the statistic F suggests that a higher number of them should be taken. The silhouette index can also be used to assess the selection of a certain number of clusters [35, p. 88]. It allows comparative assessment of the internal content of clusters and their separability [49, p. 420]. This measurement takes values from  − 1 to 1, where a value above 0.5 indicates a good structure and above 0.7, a very good structure of the created groups.

The countries were assigned to clusters using the hierarchical cluster analysis and the results of the Ward method and the k-means clustering, and the findings from particular years were compared. Consequently, it was possible to establish which countries always fell into the same cluster within the whole analysed period. A set of such countries is referred to as a group, and they are marked with the letters A–F.

Finally, the ranking of predictors was completed using the C&RT model as a measure to calculate the importance of predictors. All calculations were run using STATISTICA software.

Results and discussion

The average value of social expenditure in GDP percentage points ranged from 11.77% in Malta to 25.35% in France (Table 1). In nine out of 28 countries, this value exceeded 20%. A minimum value lower than 10% was observed for Ireland, Lithuania and Malta, while a maximum value higher than 20% was found in France, Belgium, Germany, Austria, Italy, Greece, Netherlands, Finland, Luxembourg, Slovenia and Portugal.

Table 1 Social expenditure of the government sector in EU countries in 2008–2018

The highest variability was observed for Ireland (V = 22.97%) and the second highest in Lithuania (V = 13.34%). The lowest variability was noted for Austria (V = 2.68%), closely followed by Germany, Poland and Belgium (V < 3.0%). The coefficient of variation exceeded 10% in Hungary, Latvia, and Malta, while its value was below 5% in the Czech Republic, Denmark, France, Italy, Luxembourg, the Netherlands, Slovakia, Sweden and the UK.

(Piłatowska [45, p. 26]) Right-skewness was calculated for seven countries. In Poland, A_s was exceptionally low, indicating that government social spending during the examined period was almost symmetrically distributed around the average. For Luxembourg and Sweden, A_s = 0.38, demonstrating that the distribution of government social spending during this time period was also close to symmetrical. However, in Estonia, Germany, Lithuania and Romania, the dispersion of data was very high, with A_s exceeding 1.0, except in Romania where A_s = 0.92. The left side asymmetry of variables was even stronger than on the right side. In Belgium, France and Slovakia, the left-skewness was higher than 2.0, and in nine other countries, it was higher than 1.0. These results indicate that EU countries are a non-homogeneous entity; therefore, for further more in-depth analysis, they should be divided into smaller groups.

The excess kurtosis in nine countries was lower than 0, presenting a platykurtic curve. Thus, the values of government social spending were more dispersed than in normal distribution. The lowest result (− 1.8) was in Ireland. The results for the remaining 15 countries again showed very high variability across countries. Leptokurtic kurtosis revealed that the distribution of values around the average was lower than in a normal distribution. In the case of an extreme positive excess kurtosis, the majority of the values are located in the tails of the distribution rather than around the mean. That was the case in many EU countries. The most extreme positive excess kurtosis, with a coefficient higher than 5.0, occurred in the same countries as the highest left-skewness. Overall, these results demonstrate that the government social spending in relation to GDP cannot be analysed with parametric methods.

As mentioned above, government social spending is used to support different needs in society. It may finance pension systems, people suffering from sickness or disability, the unemployed or the homeless. Table 2 presents the division of EU social spending into seven groups referred to as: “Sickness and disability, “Old age”, “Survivors” (mostly widows and widowers of persons entitled to social support who have not acquired these rights themselves), “Family and children”, “Unemployment”, “Housing” and “Other”, referring to social exclusion and social protection n.e.c.

Table 2 Average government social spending in EU countries in 2008–2018 by category (% of GDP)

The main type of social spending was on “Old age”, which mostly represents financial support to pension schemes. Its value varied between 4.73% of the GDP in Ireland and 14.53% of the GDP in Greece. The minimum share of spending on “Old age” in the total social expenditure was 15.09% in Ireland in 2018, and the maximum share was 74.53% in Greece in 2010.

The maximum average spending on helping the sick and disabled was found in the Netherlands. The Dutch government spent the equivalent of 4.59% of its GDP yearly on this type of social support during the analysed period of ten years. The highest average yearly expenditures on family and children as well as unemployment were in Denmark at 4.8% and 2.94% of the GDP, respectively, and it was the highest on housing in the UK at 1.28% of the GDP. The category of social spending called “Survivors” is very specific, and in some countries, it does not exist. Its functioning depends on the construction of social security schemes, the age and gender structure of the society and the economic activity of women. The amount of yearly social spending on survivors ranged from 0.0% of the GDP in Luxembourg and Denmark up to 2.64% of the GDP in Italy and 2.24% of the GDP in Spain.

The cluster analysis indicated noticeable differences between the EU countries in terms of the value of social expenditure in relation to the GDP achieved in these countries. This was confirmed, in part, by selection tests of different numbers of clusters indicating that the optimum number was five. The ANOVA aggregated results for five clusters in particular years also proved cross-group differentiation, which accounted for 329.17 in 2010 and up to 526.78 in 2017, with differentiation within groups ranging from 9.85 in 2010 to 24.3 in 2013 (p < 0.00001).

After assigning countries to clusters using hierarchical cluster analysis by means of both the Ward method and k-means clustering, the findings from particular years were compared. The results of both methods were similar. Consequently, it was possible to establish which countries within the whole analysed period always fell into the same cluster. A set of such countries was referred to as a group. It was determined that six such groups could be identified. In Table 3, they are marked with the letters A-F. Results indicated that, over the eleven analysed years, some countries did not belong to another cluster only once, compared to the other countries from the group. They were designated as subgroups B1, C1 and D. For example, Austria differs from the countries in B group only in 2013 and as a result is marked B1. The high number of groups, the biggest of which consists of three countries, and the group with a subgroup of four countries, confirms the considerable differentiation across countries based on the analysed criteria.

Table 3 Clusters established with the k-means method and groups of EU countries by similar social expenditure value

In many cases, similar groups were made up of neighbouring countries, as in Austria with Germany, Spain with Portugal, Ireland with the UK, and Belgium with France. Similarities were also observed between countries that are geographically quite dispersed, such as Romania and Lithuania or Denmark with Slovakia. In the case of neighbouring countries, similarities in terms of social expenditure share in GDP might be explained by cultural and historical factors. They follow the pattern of countries with similar welfare models pointed out in the Introduction. Neighbouring societies may expect their governments to implement social solutions similar to those adopted by their neighbours. It can be more challenging to explain the similarities between the countries that are far from each other. The reasons should instead be found in the economic and social policy of those countries.

Dividing government social spending into the seven types listed above, it was even more challenging to group EU countries into homogeneous clusters. The indicators of quality of clustering confirmed that the optimal number of clusters was two, and yet, the silhouette index ranged between 0.32 and 0.4, indicating low similarity of objects within clusters. The results of the analysis of predictors’ importance in influencing variability with the C&RT model are presented in Table 4. The analysis was run for each year separately.

Table 4 Ranking of predictors of heterogeneity

As can be observed, spending on the “Unemployment”, “Survivors” as well as the “Sickness and disability “ had the strongest influence on the variability of government social spending across EU countries. The case of the “Survivors” category should be considered with caution, as in six out of 10 years, ANOVA indicated this predictor was statistically insignificant in differentiating the countries. The least influence on variability was noted in spending on “Old age”, although this variable was considered a significant predictor. However, the similarity lies in the fact that in each EU country, this category has the biggest and most stable share of social spending, mostly due to their ageing societies [32].

Following the methodology for grouping and clustering used in Table 3, as well as the above analysis of the quality of the clusters, the EU countries were divided into two homogeneous clusters using seven categories of social expenditure as predictors of variability. After running the analysis for each year separately, a comparison of the results indicated one large group of EU countries that is relatively homogeneous in terms of volume and structure of social spending. The countries making up this homogeneous group are Austria, Croatia, Cyprus, Estonia, Finland, France, Hungary, Latvia, Malta, the Netherlands, Portugal, Romania, Slovakia, Spain and Sweden. Results also indicated a smaller group consisting of Denmark, Germany, Poland and Slovenia, though there was some variability within this group in 2015. The similarity of Belgium, the Czech Republic, Ireland, Italy, Lithuania, Luxembourg and the UK to other countries was changing, making it impossible to allocate them into homogeneous groups.

Conclusions

This study of the volume of social expenditure points to the high differentiation of EU countries in this respect. Some similarities were observed only for small groups of countries whose inter-group differentiation was high. The results obtained help create some perspective on problems that such a situation might create for EU countries and pose further research questions.

The variability across EU countries in terms of social spending on unemployment may result from different economic factors. Surprisingly, high unemployment rates can be observed in countries recording high demand in the workforce. With higher social benefits, the unemployed tend to spend more time on thorough job market recognisance, enabling them to seek better-tailored jobs. Consequently, job applicants appear to be more productive once they are hired. This, in turn, brings about higher earnings and lower social spending in the future.

Regarding survivors, exceptionally high differentiation indicates a need for further research their problems, particularly in the case of Italy and Malta.

As pointed out previously, high social expenditure may have a negative effect on the level of domestic savings as well as the economic activities of the population. However, the realisation of a number of the EU schemes has been connected with bearing the cost of this sort of expenditure, mainly within policies to combat poverty and social inequalities.

High differentiation of social expenditure in relation to GDP in individual countries is not conductive to support running a common policy within the EU. It also induces the population to make demands aimed at forcing their authorities to extend social support similar to that in the countries with the most developed schemes. This makes it challenging to pursue an efficient economic policy and, additionally, increases the likelihood of social unrest. It must be emphasised that the goal of social integration is not realised through common social policy but through the improvement of the wellbeing of EU citizens with a focus on reducing inequality. Therefore, understanding the differences serves to help identify sources of social problems and adjust different policies accordingly.