The Impact of Macro-Economic Circumstances and Social Protection Expenditure on Economic Deprivation in 25 European Countries, 2007–2011

Abstract

In this study, we investigate to what extent macro-economic circumstances and social protection expenditure affect economic deprivation. We use three items from round five of the European Social Survey (2010–2011) to construct our latent outcome variable, which we label economic deprivation in the 3 years before 2010–2011. The results of our linear multilevel regression analyses indicate that in countries that perform worse economically, individual experiences of economic deprivation are more prevalent: the stronger the rise in the unemployment rate and the lower a country’s wealth, the more economic deprivation individuals experience. We also find that in countries with high levels of social protection, people experience less economic deprivation as compared to countries with low levels of social protection. In turn, adverse economic conditions in a country temper these positive outcomes of social welfare arrangements. Finally, our study reveals that the strength of the relationship between a low income and economic deprivation strongly varies according to the economic circumstances in a country and the generosity of the welfare state.

Appendix

As stated in the data section, we tested if economic deprivation is equivalently measured across our sample of nations. We started by calculating covariance matrices and mean structures for each country separately by importing the original scores on the items into PRELIS. Using confirmatory factor analysis (CFA) within LISREL 8.80 (Jöreskog and Sörbom 2006), we constructed a single factor, which we labeled ‘economic deprivation in the three years before 2010-2011′ or in short, economic deprivation. Maximum likelihood estimation was used as the procedure to obtain the model parameters. To identify the model, we fixed the factor loading of the item ‘I have had to manage on a lower household income’ to 1. This ensures that the response scale on which this item is measured, is also the scale on which economic deprivation is expressed. We selected this item as our reference variable, because for most people a lower household income represents the most relevant form of economic deprivation.

We then tested our model for three levels of measurement equivalence (i.e., configural, metric and scalar invariance), applying multigroup confirmatory factor analysis (MGCFA), again within LISREL. Following the approach suggested by Vandenberg and Lance (2000), we employed a bottom-up procedure to detect misspecifications in the MGCFA model, because the number of parameters that are potentially misspecified in the most constrained model (i.e., scalar invariance) in a top-down approach could be overwhelming. Therefore, we started by testing for configural invariance, which is the least constrained model.

The standard procedure to assess whether a constrained parameter is misspecified or not, is by looking at the modification index (MI) and the expected parameter change (EPC). As Saris, Satorra and Van der Veld (2009) showed, however, the power of the test may influence the MI. To solve this issue (thus, to take into account the power of the test), we made use of the software package JRule 3.0.4 (Van der Veld et al. 2008). This program computes judgment rules based on LISREL output, indicating for each parameter if it is misspecified, not misspecified or if it depends on the extent to which the parameter will change when it is not constrained.

The results of our configural invariance test indicate that the measurement model holds across all countries. We also evaluated the Root Mean Square Error of Approximation (RMSEA). According to various scholars, a value of the RMSEA of 0.050 or less would indicate a close fit of the model, a value between 0.050 and 0.080 would indicate a reasonable error of approximation and a global model fit greater than 0.100 suggests a poor fit (Arbuckle 2010; Browne and Cudeck 1993; Hu and Bentler 1999). We found a RMSEA of 0.000 (df = 1), indicating a perfect model fit.

For our metric invariance test, we have chosen Poland as our reference country, which has factor loadings close to the average across all counties. A reference country with extreme factor loadings will result in a decreased likelihood of that county being invariant. If we would select a country that is not invariant, it will become harder to compare a large set of countries. For Poland, we estimated the factor loadings, whereas in all other countries the factor loadings are constrained to being invariant. With help of JRule, we evaluated the imposed invariance constraints. Since configural invariance already holds, it makes no sense to look at misspecifications regarding correlated random error terms and correlated unique components. After all, these constrained parameters should have been detected as a misspecification during the configural invariance test. The results of JRule show that we have to inspect some of the EPC’s. There are four factor loadings that would significantly change when estimated, namely λ ₂₁ in Germany and λ ₃₁ in Croatia, Cyprus and Greece. We decided to adjust our model, because the model fit is mediocre: the value of the RMSEA is 0.095. When we released the constraints on these factor loadings, the value of the RMSEA becomes 0.060, almost indicating a close fit.

Given that we already tested for configural and metric invariance, we only evaluated the mean structure of our measurement model in our next step. We already estimated the intercept means of the items of which the factor loadings were released in our metric invariance test. First of all, the results of the scalar invariance test show that the RMSEA is 0.110, which indicates a poor model fit. We decided to adjust our model by looking at the MI in combination with the EPC. Ultimately, we released (in a stepwise manner) the constraints on the intercepts of several indicators in the following countries: ε ₁ of item G8 in Cyprus, ε ₂ of item G9 in the Czech Republic, ε ₂ of item G9 in Estonia and ε ₁ of item G8 and ε ₃ of item G10 in Finland. After adapting these intercepts, the RMSEA of our model is 0.080. Since this implies reasonable error of approximation and because we would like to avoid partial scalar invariance for other countries, we chose to accept the model. Hence, our final model is partially invariant for Croatia, Cyprus, Czech Republic, Estonia, Finland, Germany and Greece. This results in 7 of the 25 countries (28 percent) being partially invariant. We provide an overview of the factor loadings and indicator intercepts of our final measurement model in Table 4.

Table 4 Results measurement model economic deprivation