Suntory and Toyota International Centres for Economics and Related Disciplines (STICERD) LSE RSS Contact Us YouTube Twitter

With support from a STICERD small grant, this project examines the drivers of social exclusion in EU Member States. The work builds on earlier work on social inequalities in Europe commissioned by Eurofound.

Social exclusion has moved rapidly up the EU agenda in recent years. The Europe 2020 goals include a headline target to lift at least 20 million people out of the risk of poverty and social exclusion. This target follows on from earlier monitoring efforts following the adoption of an initial set of poverty and social exclusion indicators by the European Council in 2001. Foundational work in this area includes Atkinson et al (2002).

Our project builds on these developments. The objective is to develop a robust understanding of the relationship between social exclusion in European Union member states and underlying individual and country level variables. The dataset used for the study is the 2011 European Quality of Life Survey (EQLS). The EQLS is a major cross-country survey which has a rich variety of questions on different areas of quality of life and is repeated every four years. The 2011 survey covered adults in 34 European countries with a total sample size of 35,516. Here, we focus on the 27 countries that were European Union member states in 2011.

The particular focus of the project is on a subjective rather than an objective measure of social exclusion, based on a direct survey question about whether people feel excluded from society. Specifically, the EQLS 2011 asked respondents whether they strongly agree, agree, neither agree or disagree, disagree or strongly disagree with the statement: "I feel left out of society". In this paper, we focus on responses to this question and interpret such responses as providing empirical evidence on "self-reported social exclusion". In focusing on a subjective rather than an objective measure of social exclusion, we aim to contribute to broader discussions about the role that subjective data including data on views, attitudes, happiness, satisfaction, and user feedback can play in the examination of social phenomenon and as a guide to public policy (Dolan & Metcalfe, 2012). Nolan and Whelan (2009) note that improved "non-monetary" indicators - including indicators of how people feel, regard and report on their own situation - have an important role to play in improving knowledge and measurement of poverty and social exclusion. The current paper contributes to literature by building up an evidence base on how people in European countries feel, regard and report on their own social exclusion status.

The project examines the importance of four sets of independent variables as potential "drivers" of (self-reported) social exclusion using multilevel logistic regression analysis. These are: (1) a set of focus "equality characteristics" (gender, age, disability, socio-economic status, citizenship status)[1] ; (2) a broader range set of individual level characteristics including income poverty, material deprivation, education, social support, caring activities, marital status and participation; (3) country level macro variables, specifically average national income per capita (GDP per capita) and national income distribution (gini coefficient); and (4) public policy variables specifically, public expenditure on social protection and health, and variables which provide information about institutions and social arrangements, such as health system and welfare regime.



Findings based on our penultimate model are set out in Figure 1. At the individual level, age, disability, non-EU citizenship, socio-economic status (being unemployment or working in a nonprofessional occupational social group), poverty (income poverty or material deprivation), having a lower of educational achievement, family type (not having a child, widowhood), having poor self-rated general health, being a carer and not participating in social activities were found to be associated with self-reported social exclusion after controls had been introduced into the model. At the country level, GDP per capita, the Gini coefficient, health expenditure per capita and social protection expenditure per capita were found to be not statistically significant and were dropped from the model. Different social institutions and arrangements were, however, found to be important. For instance, Corporatist and Liberal welfare regimes were observed to be associated with a higher probability of self-reported social exclusion, compared with Social Democratic welfare regimes. Looking at the impact of health system, countries characterised by a high proportion of "out-of-pocket" health expenditure within total health expenditure were associated with a higher probability of self-reported social exclusion, compared to countries with a high share of public expenditure in total expenditure, and low shares of private and out of pocket expenditure.

figure 1
Figure1. Click to see larger image (in PDF)

Our final model includes an interaction effect between unemployment status and welfare regime. Based on this model, we conclude that the effect of being unemployed rather than employed on self-reported social exclusion is differentiated by welfare regime; and that the effect of being unemployed rather than employed in Social Democratic, Corporatist and Liberal regimes has a substantial effect on self-reported social exclusion. Being short-term unemployed rather than employed has a particularly strong effect in a Social Democratic / Corporatist welfare regime; whilst being long-term unemployed has a particularly strong effect in Liberal welfare regimes.

Robustness testing was undertaken as part of the project. Bryan and Jenkins (2015) develop a critique of multilevel regression techniques where the number of clusters is relatively low. This critique is applied to a number of cross-country European studies using mixed effects model, and for multilevel logistic regression, the authors recommend a minimum number of clusters as 30. Since only 27 countries were included in our analysis, we also ran a fixed effect model based on individual level variables with country included as a dummy variables. A cluster correction was applied (in order to obtain standard error with minimal 'clustering effect'). Broadly speaking, no appreciable difference in coefficients - in terms of their significance, direction or magnitude - was observed.