Welfare Stateness as an Explanatory Variable: Empirical Illustration

Kunißen, Katharina

doi:10.1007/978-3-658-39422-6_6

Katharina Kunißen⁵

Part of the book series: Sozialstrukturanalyse ((SSA))

1317 Accesses

Abstract

While the previous chapters explored ways to conceptualise the welfare state as an independent variable, the following part will deal with the practical applications of the proposed framework. For this purpose, poverty risk and attitudes towards the welfare state serve as exemplary dependent variables.

You have full access to this open access chapter, Download chapter PDF

While the previous chapters explored ways to conceptualise the welfare state as an independent variable, the following part will deal with the practical applications of the proposed framework. Several objectives are pursued. First, I am going to illustrate how embedding research questions and hypotheses in the discussed systematised concepts of welfare stateness may be used in order to choose conceptual as well as empirical approaches for a given research objective. Second, I use the following exemplary analyses to determine—as far as possible—the validity of the proposed measurement. Third, I hope to contribute to the state of research in two subject areas where the welfare state is considered an important explanatory factor. In this case, welfare attitudes and the risk of poverty serve as exemplary dependent variables. These two topics tie in with the previously discussed strands of literature that have proven to be important and popular in the relevant literature. Furthermore, they represent both elementary perspectives that were identified in the previous chapters: while explanations for differing poverty levels conceptualise the welfare state as an institution influencing individuals top-down, individual attitudes are explained using both perspectives. Thus, they are expected to result from bottom-up perceptions of welfare stateness and to be shaped by welfare state arrangements top-down at the same time. They should therefore serve as adequate examples for the differences between the two perspectives and—in a more fundamental sense—provide a test of their usefulness. Even though these two examples only cover singled-out issues and it seems an overstatement to expect an exhaustive test of the proposed framework from the following analyses, I expect to be able to test the possibilites and limits of my proposal, at least in part. This refers to the proposed selection framework in general and the concrete recommendations for the operationalisation of specific systematised concepts of welfare stateness in particular (cf. chapter 5).

The chapter presents analyses of the two exemplary topics that follow all steps usually taken when working on a research question—albeit in some passages in a slightly shortened way suitable for this demonstration because many general theoretical and conceptual premises for both topics were already discussed previously. Apart from that, hypotheses, mechanisms, conceptualisation, and operationalisation follow the proposed path and are tested in analyses using standard methodological approaches for the analysis of multilevel comparative data.

6.1 Remarks on the Following Analyses

Some remarks have to be made before starting with the analysis of the first exemplary dependent variable. This concerns general premises of the following analyses. In both cases, I chose to rely on the newest available data—when it comes to macro- as well as micro-level. Furthermore, I made sure that both analyses rest on data collected during a similar period—even though they stem from different datasets. This is intended to allow the best possible comparison of the applicability of indicators as it ensures that the country-level data on welfare stateness are identical—even though the individual survey data differ. Since the dependent variables require different methodological approaches, I will elaborate on the method of analysis applied in the individual chapters for each outcome. For now, I will only introduce the different data sources used in the analyses as well as the indicators of welfare stateness that will be tested.

6.1.1 Data Sources for Individual-Level Survey Data

The following analyses are based mainly on two data sources for individual-level data. The selection follows the objective to model both exemplary research questions as adequately as possible. This means that recent data, covering most countries in the relevant sample^{Footnote 1} and allowing a satisfying operationalisation of the dependent variables, are selected.

The micro-level data in the analyses of the risk of poverty (section 6.2) stem from the European Union Statistics on Income and Living Conditions (EU-SILC) provided by the European Commission. EU-SILC is the main reference for comparative information on income distribution and social exclusion in the 28 European member states^{Footnote 2} and 5 additional countries (Republic of Macedonia, Iceland, Turkey, Norway, and Switzerland) (European Commission 2017: 13). The analysis focusses on the EU-27 countries as well as Norway and Switzerland. In order to maximise comparability with the other analyses, data collected in 2016 were chosen. The dataset has a variety of advantages, which justify why it was preferred over other comparative survey data including similar information on income such as the European Social Survey (ESS) or the International Social Survey Programme (ISSP) used in chapter 3. First, it covers an exhaustive sample of all member states of the European Union. The number of observed countries is thus much higher and does not rely on an arbitrary selection, which makes it more meaningful compared to other datasets. Second, it offers high-quality data as it is an important source of information and policy-making within the EU. Third, the information on personal and household income—which is normally affected by non-response or other biases—is quite comprehensive. Since poverty is determined based on income, the high coverage of income information in EU-SILC is advantageous.^{Footnote 3}

In the case of the analysis of welfare attitudes (section 6.3), the eighth wave of the European Social Survey (ESS 2016) is very fitting. This dataset was collected in 2016 and early 2017. It covers 20 countries from the relevant sample^{Footnote 4} and includes a comprehensive module on attitudes towards the welfare state, covering various facets of the subject, including welfare chauvinism, opinions about state responsibility for the provision of social services and benefits, and general attitudes about the welfare state, its purpose, and effectiveness. While the relatively small number of relevant countries is surely a downside of the dataset—especially for applying multilevel methods—the differentiated measurement of welfare attitudes is a great advantage.

In both analyses, I am going to refer sporadically to additional analyses that were performed in instances where I felt that it was necessary to explore the robustness of results. Furthermore, in the case of welfare attitudes, results are compared to those obtained in chapter 3 using earlier data from the ESS and ISSP.^{Footnote 5}

6.1.2 Data Sources for Macro-Level Data and Country Sample

Several sources are used in order to obtain data on welfare stateness and other features of the observed countries. Following the criterion of availability and the agenda of increasing the reproducibility of results, only publicly available data are chosen. The datasets correspond to those used in previous analyses in this book (cf. chapter 3) and are among those discussed at the end of the previous chapter.

The majority of indicators stem from the Social Insurance Entitlement Dataset (SIED). This source, which represents the continuation of the SCIP (Korpi & Palme 2008), was referred to in detail throughout this book. In short, it offers very comprehensive information on social rights in 34 countries, including all member states of the European Union (with the exception of Croatia) and some adjacent countries. While my previous analyses relied on the 2005 version of this dataset, the following analyses will be based on the newest available version at this time, which was published in early 2019 and includes data from 2015 (SPIN 2019a).

Additional information is obtained from Eurostat and in particular the European system of integrated social protection statistics (ESSPROS) dataset (cf. European Commission 2016) and the Organisation for Economic Co-operation and Development (OECD 2019). These sources offer detailed information on social expenditure, as well as additional variables on the level of countries such as gross domestic product (GDP) and unemployment rates. Again, data from 2015 were chosen to match the indicators of welfare stateness best.

Due to the missing data on many indicators of welfare stateness in Croatia, it is excluded from the analysis. Apart from that, the macro-level data allow the analysis of a very comprehensive sample of countries consisting of a full coverage of the former EU-27 countries, Switzerland, and Norway as adjacent countries with strong political and economic ties to the European Union.

6.1.3 Indicators of Welfare Stateness

The following analyses will take up and compare a wide bundle of indicators. While the operationalisations chosen in each example will follow theoretical and conceptual premises as outlined in the conceptual framework, I will introduce the full set of indicators at this point already to illustrate the range of available information.

Furthermore—in the sense of comparability—the selection of indicators is guided by their previous application as independent variable. I hope that the following analyses not only lead to new insights but also measure up to existing evidence that has been generated using similar or even identical indicators.

As mentioned when introducing the datasets, I rely on information from 2015 in case of all macro-level variables. Thus, the individual-level survey data (which stems mostly from 2016) is delayed by one year. This is done intentionally, as it seems unlikely that policies manifest in individual reality instantly. Rather, it is to be expected that the outcomes of a specific policy reach the individual with some delay. It may be debatable whether one year is too short an interval or whether the average value for several previous years should be chosen to ensure that no outliers disturb the information. Such considerations seem useful, but they will not be at the core of my analyses, as I focus foremost on the indicators themselves.

Overall, welfare stateness is covered by five blocks of indicators, which stem from the above-mentioned sources and should allow to model at least the majority of systematised concepts. In the following, the distribution of the chosen indicators is briefly introduced. For a comprehensive list and additional information, the appendix can be consulted.^{Footnote 6}

The first block of indicators includes net replacement rates (NRR) in four scenarios of need: unemployment, old age, sickness, and accident. It would have been possible to add an income replacement rate for parental leave as well (cf. Saraceno & Keck 2011). However, as the chosen dependent variables are more closely related to the old risks, I remain within the more traditional areas of social policy-making for the time being. In all four cases, the replacement rates stem from the Social Insurance Entitlement Dataset (SIED) and represent combined information. The net replacement rate in case of unemployment refers to the average production worker and incorporates the mean benefit obtained by a single person a) during the first week and b) after 26 weeks of receipt. A single person is chosen instead of another type of model family because a family may receive other benefits that interfere with the clarity of the measure (for a similar reasoning cf. Ferrarini et al. 2014b: 658). The net replacement rates for those unable to work because of sickness or accident are based on the same information. Since pensions do not vary depending on how long they are obtained, the net replacement rate again refers to the average production worker and is calculated for a single person but without the combination of different times of benefit receipt.

A descriptive overview of the differences between the social policy areas and the 29 countries in the analysis is provided in Figure 6.1. It reveals strong variation—not only between policy areas but also between countries. Overall, replacement rates in case of sickness or accident tend to be higher in most countries than replacement rates for unemployed or retired citizens. In addition, the Anglo-Saxon countries (Ireland and the United Kingdom) are clearly among the less generous ones, while the more generous countries—such as Luxembourg and Portugal—do not seem to reveal a clear pattern. Since the aim of this contribution is not to cluster countries, this is not consequential for my analyses. The strong heterogeneity in replacement rates remains remarkable, as it seems to support the argument that welfare state typologies may unduly reduce the existing complexity of the matter.

As argued before, net replacement rates are often considered a measure of benefit generosity. As such, they may be relevant for the operationalisation of several systematised concepts. By signalling how much an average worker is entitled to, they seem suitable for operationalising the Responsive Welfare State (cf. section 5.3.1). However, it is also plausible to expect that the generosity of potential benefits one is entitled to is a very salient feature of social policy-making. As previously argued (cf. section 5.3.4) this would mean that net replacement rates could also be suitable to operationalise the Assessed Welfare State. Lastly, it can also be argued that comprehensive benefits signal an underlying welfare culture and thus fit the concept of the Normative Welfare State. It is not clear at this point, which of the concepts can best be operationalised by NRRs, and it needs to be observed if the results obtained in the exemplary analyses contribute to shedding light on the issue.

The second block of indicators addresses insurance coverage. As before, the four old risks are differentiated. Both, accident insurance and unemployment insurance, refer to social policy instruments that are clearly linked to employment. Therefore, both indicators refer to coverage as a proportion of the labour force. In contrast, since referring to insurances needed by every citizen, pension and sickness insurance coverage is measured as the insurance coverage as a proportion of the population.^{Footnote 7} An overview of insurance coverage is provided in Figure 6.2.

Overall, insurance coverage—regardless of the insurance scheme—is above 50 percent in most cases (exceptions can be found in Spain, Italy, Greece, and Romania). Since insurance coverage is very important for how well citizens in a country are potentially protected against risks, this indicator seems to suit the operationalisation of the Responsive Welfare State. As mentioned before (cf. section 5.3.3), it may also be worth discussing, whether insurance coverage also signals universal access. If so, it may also be a suitable operationalisation of the Normative Welfare State.

The third block of variables addresses the contribution period required to be eligible for benefit receipt. The contribution period indicates how quickly welfare states step in when needed and how small the barriers to obtaining benefits are. As such, they are closely related to the Responsive Welfare State. In addition, one could argue that the contribution period signals the strictness of eligibility criteria and thus the equality and universality of benefit access. In this sense, it may also fit the Normative Welfare State. Like before, different policy fields are differentiated. However, the contribution period required to be eligible for accident insurance is excluded from the analysis, as it has almost no variation.^{Footnote 8} For the remaining benefit areas—unemployment, pension, and sickness—the contribution required is measured in weeks (cf. Figure 6.3). In several cases, countries exhibit very similar regulations. For instance, in almost half of the examined countries, there is immediate eligibility for sickness benefits. Similarly, countries seem to agree on how long contributions must be made in order to qualify for unemployment and pension benefits: in most cases, benefits can be obtained after one year of contributions. Only a few outliers (Lithuania and Slovakia) require a longer contribution period—and that is only in the case of unemployment benefits. Because of the uneven distribution of the variables, they are turned into dichotomous variables. This results in indicators capturing, whether the required contribution is zero in the case of sickness and lower than the median for unemployment and pensions.^{Footnote 9}

Another indicator that is closely linked to social rights and thus to the concept of the Responsive Welfare State is the duration of benefit receipt. In this fourth block of indicators, too, I distinguish between different types of benefits, but omit the duration of pensions, as there is no time limit on their receipt. Like in the case of the contribution period, benefit duration is measured in weeks. There is considerable variation in how long benefits can be obtained across schemes and countries.

Overall, the duration of unemployment benefits is the shortest with a median value of 51 weeks, while the median duration of accident benefits is 104 weeks (cf. Figure 6.4). In several countries, benefits can be obtained for 10 years or longer (especially in the case of accident insurance). However, values higher than that were all fixed at an upper threshold of 520 weeks in the SIED data. As a result of the extreme range of the variable and its uneven distribution with most benefit schemes being obtainable for up to 2 years and just a few exceptions with substantially longer benefit duration, the variables were again recoded into dichotomous indicators. The resulting variables capture whether a country offers benefit duration, which is higher than the median.

All of the above-mentioned indicators were taken from the SIED Dataset (SPIN 2019a). In addition, the fifth and last block of variables includes four indicators capturing social expenditure on Eurostat data (ESSPROS and European Commission 2018). Analogously to the other indicators, the different areas of social policy-making are distinguished by including spending in the fields of old age, health, and unemployment (cf. Figure 6.5).

Furthermore, total social expenditure is examined, which incorporates the sum of spending in the three above-mentioned areas and additional policy fields, which are of secondary importance for my analyses, such as child and family benefits and survivor benefits (European Commission 2016). While these four indicators represent the overall effort, an additional spending indicator is added to capture the commitment to labour market activation. More specifically, it captures expenditure in the field of training, employment incentives, supported employment and rehabilitation, direct job creation, and start-up incentives. All expenditure measures represent spending as percentage of the gross domestic product.

Throughout this book, concerns about using social expenditure have been voiced. Such indicators were described as unintelligible and fuzzy. For two reasons they are still included in the following analyses. First, they are by far the most popularly chosen single indicator in the literature. As such, it seems commendable to use them as a reference and to compare them to other strategies of operationalisation. Second, knowledge about expenditure may be biased, but it still represents a better known feature of welfare stateness compared to other alternatives. As such, indicators of expenditure may offer an operationalisation of the Assessed Welfare State. Third, since comprehensive operationalisations of social investment and active labour market policies are scarce, spending on ALMP is among the only available options of capturing the Enabling Welfare State.

Turning to the types of measurement validity discussed in the previous chapter, the only one that can be tested at this point is the convergent (or discriminant) validity. According to this criterion, indictors for the same systematised concept should correlate. A glance at the correlation matrix does reveal a mixed picture.^{Footnote 11} While policy areas within one type of indicator tend to be correlated, there is also evidence for an association of different indicators within the same policy field (e.g. unemployment expenditure and unemployment replacement rates). Since there is, however, considerable overlap in possible indicators for different conceptualisations, clearly delineating groups of indicators based on correlations is impossible. This overlap has implications for the entire endeavour of this chapter, which will be discussed shortly.

The introduced five blocks of variables (cf. Table 6.1) were selected with the aim of best fulfilling the three criteria set out in the previous chapter. They are comparable in that they originate from transparent sources and for the most part were already used in the relevant literature.^{Footnote 12} They are available, as information is easily and freely accessible for scientific use. Lastly, they at least approach clarity as, apart from social spending (and to some extent also net replacement rates), the selected variables can be assigned to one or a few systematised concepts. In addition, they capture clearly distinguishable elements of social policy-making and differentiate the various areas within the old risks.

Table 6.1 Indicators of welfare stateness—summary

Full size table

However, the chosen selection also entails limitations. As briefly mentioned, information on the whole area of child and family policies is not taken into account. The only indicator including such benefits and services is the total social expenditure. By omitting measures in the areas of child and family policies, I rely on a selection that matches the chosen exemplary dependent variables. Another restriction is that only one operationalisation of active labour market policies and social investment is included, which means that the entire concept of the Enabling Welfare State is only scarcely covered in the following analyses. For several reasons, both restrictions seem acceptable for the purpose of this project. On the one hand, the analyses are reduced to the old risks, as they are easiest to compare with previous studies. Moreover, the availability of indicators, especially for activating measures (ALMP and social investment), is very low and the discussion on how to operationalise these measures is still ongoing and comparatively new (as repeatedly noted in the previous chapters). This by no means implies that the policies that were left out are incompatible or of subordinate importance. Rather, they may provide valuable insights depending on the chosen explanandum. However, for the two dependent variables this chapter focusses on, the old risks promise to be sufficiently fitting for an illustration of the proposed framework.

While these limitations do not directly affect the results and the way in which the goal of this book is achieved, there is one issue that could have serious implications: there is considerable overlap when it comes to which indicators are suitable for which conceptualisation. This even happens between the two main perspectives (top-down and bottom-up). As a result, a clear-cut empirical distinction between the concepts is impossible. This is very regrettable, as the main advantage of using systematised conceptualisations is supposed to be their resulting in distinct operationalisations—at least if one envisions an ideal empirical test of the framework. However, arguing based on the existing indicators of welfare stateness (the ones used here and potential others), this ideal clear allocation of the one perfect indicator (or set of indicators) to only one concept is not feasible. If indicators are selected based on criteria such as the ones chosen in this chapter, some amount of overlap has to be taken into account—especially during this first application of the framework. Furthermore, the very nature of some of the conceptualisations—in particular the Responsive Welfare State—is to span different policy fields. For that reason, they simply cannot be reduced to one facet of policy-making.

In order to maximise the output from the following trial, several aspects are considered. First, the results are linked to the theoretical assumptions, which should help to assign the indicators and the results they produce to a specific conception. Second, if overlap cannot be resolved, it will be discussed if indicators should be ruled out as potential operationalisations. Third, it is acknowledged that some overlap is actually theoretically plausible, as—for instance—indicators representing responsiveness may be the same ones as those responsible for cost-benefit considerations. The fact that they can be interpreted in two ways may actually bear important additional insights. These and more issues will be discussed at the end of each analysis and in the discussion at the end of this chapter.

6.2 The Welfare State and the Risk of Poverty

Since responding to needs and risks is among the most basic functions of welfare states, the literature often asks whether social policies are responsible for differences in individual poverty risk across countries (cf. section 4.2.2). For this reason, the risk of poverty was selected as an exemplary topic for the following exemplary application of the proposed framework. In this section of the chapter, I will illustrate all proposed steps that lead from hypotheses and mechanisms to conceptualisations and lastly distinct operationalisations of welfare stateness.

6.2.1 Conceptualisation of the Welfare State

The previous chapter concluded with a list of questions that should guide the theoretical conceptualisation and empirical operationalisations of welfare stateness. The first question (which perspective on the welfare state is chosen?) can be answered quickly when explaining cross-national differences in the risk of poverty. Here, only the top-down perspective is relevant, since individual perception and evaluation of social policies in the sense of the bottom-up perspective are inconsequential. This is reflected in the mechanisms outlined in the literature and the hypotheses deduced from them. Turning to the second question (which mechanisms are addressed?) several mechanisms are relevant. As one of the main objectives of social policy is to protect individuals from risks and provide support though benefits and services when needed, security is a prominent mechanism that explains how and why welfare states achieve different levels of poverty. This happens through the redistribution of resources, where targeted and universal premises represent opposed logics (cf. the paradox of redistribution). In chapter 4, this was summarised in the following explanation:

Explanation 2.1: Comprehensive social rights (including benefit coverage, generosity, and eligibility criteria) and redistributive budget decrease poverty because these actions secure against risks by redistributing resources to those who require them.

Phrased as hypotheses and distinguishing between the institutional perspective (social rights) and the expenditure perspective (redistributive budget) this means:

Hypothesis Pov1: The more comprehensive the social rights, the lower the prevalence of poverty in a country.
Hypothesis Pov2: The higher the redistributive budget, the lower the prevalence of poverty in a country.

In addition, moderating effects were discussed. Thus, security not only has a direct impact on the risk of poverty but also moderates the effects of individual determinants such as low educational attainment or unemployment. Such moderating effects are more pronounced, yet chapter 4 addresses additional mediated effects in which the welfare state shapes those individual determinants (social stratification), which in turn are responsible for the risk of poverty.^{Footnote 13} Since the latter is of secondary importance, the moderating effects of social policies are highlighted in the following exemplary analyses:

Explanation 2.2: Comprehensive social rights (including benefit coverage, generosity, and eligibility criteria) and redistributive budget decrease poverty because these actions secure those who are especially vulnerable.
Hypothesis Pov3: The more comprehensive the social rights, the lower the risk emanating from social determinants of poverty.
Hypothesis Pov4: The higher the redistributive budget, the lower the risk emanating from social determinants of poverty.

Lastly, activation is emphasised. Even though activating policies may have a direct effect on risks, the more relevant influence is a moderating one. The last explanation and hypothesis thus capture the potential to lower risk by equipping individuals in vulnerable situations with tools needed to avoid being at risk of poverty. This is again expected to manifest especially as a moderating effect:

Explanation 2.4: Activating policies aiming at ending periods of vulnerability (such as unemployment) decreases poverty because these actions increase labour market participation through activation.
Hypothesis Pov5: The higher the effort directed towards activating policies, the lower the risk emanating from social determinants of poverty.

At this point, it is irrelevant whether these five hypotheses fully cover all theoretical assumptions in the literature, as they cover a sufficiently broad spectrum of the topic to allow for the discussion of several systematised concepts that might be relevant to the empirical investigation.

The third question (which concepts of welfare stateness are addressed?) is of particular importance. Based on the assumed mechanisms, two systematised concepts of welfare stateness are highlighted. The Responsive Welfare State captures those explanations emphasising security, social stratification, and redistribution, while the Enabling Welfare State captures the idea of activation and incentive. The other systematised concepts—the Normative and Assessed Welfare State—are not relevant in this case.

Two more questions have to be addressed before turning from theoretical conceptualisation to empirical operationalisation: which policy fields are relevant and which temporal perspective has to be chosen? Since the risk of poverty emanates from very different situations, a restriction to specific policy fields is not necessary. Even though the labour market is closely tied to poverty, other fields such as health and pension policies are equally important depending on which social groups (e.g. with regards to employment, educational status, or age) are highlighted in an analysis. In terms of the temporal perspective, it seems plausible that the risk of poverty is determined at all times by the prevailing contextual influences. The conceptualisation and operationalisation of the welfare state can therefore be based on a reference period similar to the survey data.

6.2.2 Operationalisation of the Welfare State

The previous section of this chapter reveals that two conceptualisations of welfare stateness are especially important when exploring the impact of social policies on the risk of poverty. The aim of this contribution is to explore how well such different conceptualisations can be operationalised empirically and whether this improves the informative value of results and discussions on the one hand and the transparency and comparability of operational approaches on the other hand. Following the recommendations in chapter 2, the two systematised concepts of welfare stateness should be grasped by the following sets of indicators:

The Responsive Welfare State should be operationalised using indicators of social rights (e.g. benefit generosity, coverage, eligibility criteria).

The Enabling Welfare State should be operationalised using indicators of efforts to activate and incentivise (e.g. active labour market policies).

At the beginning of this sixth chapter (cf. 6.1.3), specific indicators for each systematised concept were introduced. Following those recommendations, all but one set of indicators were discussed as potentially suitable for the operationalisation of the Responsive Welfare State. Only social expenditure was not added to the list of potential operationalisations, because of its lack of clarity. However, as spending continues to be widely used (e.g. capturing the redistributive budget) and spending on active labour market policy is the only available operationalisation of the Enabling Welfare State, this set of indicators is still included. The following analysis should help to find out which of these operationalisations depicts the conceptual premises more accurately and tests the hypotheses more precisely.

6.2.3 Additional Variables and Analytical Strategy

While poverty is the dependent variable in the following analyses, several independent variables are also added at the individual level. The inclusion of such micro-level determinants is particularly important as the hypotheses suggest that social policies not only directly influence poverty by reducing or increasing risk but also act as moderators. In this sense, the welfare state is expected to mitigate the risk posed by situations in which the individual is vulnerable—such as unemployment or low education. In the following, the operationalisation of the dependent and independent variables is briefly outlined as well as the analytical strategy.

Dependent variable

The dependent variable—being at risk of poverty—can be operationalised in various ways. In the literature, such different approaches have in common that they are usually based on information about the disposable household income, which is then equivalised in order to control for different household sizes and compositions. However, they differ when it comes to details. For instance, when constructing equivalised income, some authors rely on the so-called “old” OECD scale, while others choose the modified one or divide income by the square root of household members (e.g. Haughton & Khandker 2009; UNECE 2017). Furthermore, different cut-off points for determining whether a person is at risk of poverty are used. While some define it as having less than 50 percent of a country’s median equivalised disposable income (e.g. Brady et al. 2017), others use 60 percent as a threshold (e.g. Polin & Raitano 2014). For my analyses, I will rely on the latter version, as it corresponds best to official approaches. This measure of poverty (using the modified OECD scale) is already included as a variable in the EU-SILC dataset.

Independent variables

Individual characteristics are needed to model moderated effects and to serve as control variables in those models that include a cross-level interaction term. Therefore, several socio-economic indicators are added as control variables.^{Footnote 14} These control variables include employment status as a predicator of economic well-being. Employment status is derived based on a variable on the self-defined economic status. The recoded variable captures whether a person is unemployed, employed, self-employed, retired, or inactive for other reasons such as being in school, military service, disabled, or fulfilling domestic or care responsibilities. In addition to indicators of economic status, age, sex, and education are included as sociodemographic control variables. Since EU-SILC only provides detailed information on year and month of the birth of respondents until 1930 and all older respondents are treated as if they were born in 1930, using age as a continuous variable seems inadvisable. Instead, seven age cohorts are constructed starting with respondents aged 29 and younger, continuing in 10-year steps and ending with respondents aged 80 and older. Sex is included as a binary predictor with women being the reference category.^{Footnote 15} Lastly, information on the highest level of attained education is provided in the dataset following the ISCED-97 classification. This information is recoded into three groups of educational attainment. The first encompasses individuals who attained primary or lower secondary education or less (ISCED 0, 1, & 2). The second group includes individuals with (upper) secondary or post-secondary non-tertiary education (ISCED 3 & 4) and the highest group possesses tertiary education (ISCED 5 & 6).

Analytical strategy

The following analyses are based on a number of multilevel logistic regression models. This procedure deviates from the linear approach used in the third chapter, because the dependent variable (poverty) is dichotomous. The main idea behind the method is still the same. Following the approach described in chapter 1 from a conceptual point of view and later in chapter 3 in the case of (pseudo-)continuous dependent variables, the main advantage of multilevel modelling is that it is able to estimate effects of individual- and country-level effects simultaneously. In the case of dichotomous variables, the outcome $y_{ij}$ for an individual $i$ in a country $j$ is either zero or one. Thus, the analytical approach focusses on the probability ($P_{ij}$) of the occurrence of an outcome (for a detailed description cp. Snijders & Bosker 2012: 293–295). In the full multilevel logistic regression model, the logarithmised probability of poverty occurring $logit\left( {P_{ij} } \right)$ is based on the average probability $\gamma_{0}$ (much like a grand mean in linear models), the effects of all individual-level variables $x_{ij}$, country-level variables $z_{0j}$, and a random group-dependent deviation in the intercept $u_{0j}$. Since moderating effects of social policies are assumed to exist from a theoretical point of view, there is also the need to explore random slopes. Thus, the impact of at least one independent variable on the individual level is assumed to have different slopes in different countries, which adds the random effect $u_{1j} x_{1j}$. It is furthermore assumed that features of the welfare state can at least partly explain those different slopes, which means that a cross-level interaction effect is included ($\gamma_{1} x_{1ij} z_{10j}$). This results in the following Random-Intercept-Random-Slope-model:

$$\begin{aligned} logit\left( {P\left( {\text{risk of poverty}} \right)_{ij} } \right) & = \gamma_{00} + \mathop \sum \limits_{h = 1}^{r} \gamma_{h} x_{hij} + \mathop \sum \limits_{l = 1}^{r} \gamma_{l} z_{l0j} \\ & + \gamma_{1} x_{1ij} z_{10j} + u_{0j} + u_{1j} x_{1ij} \\ \end{aligned}$$

Following the advice given by Heisig and Schaeffer (2019), the slope of the lower level variable in the cross-level interaction is explicitly defined as random in all models.^{Footnote 16} ^{Footnote 17}

Turning to the analysed population sample in the EU-SILC data, three different strategies can be found in cross-cultural analyses of poverty. The first focusses on in-work poverty and thus reduces the sample to employed individuals (e.g. Lohmann 2009; Halleröd et al. 2015). The second highlights poverty among the working-age population (e.g. Saltkjel & Malmberg-Heimonen 2017) and the third strategy explores poverty among the entire population (e.g. Watson et al. 2018). In this contribution, I focus on the second option—the working-age population (16–64)—for several reasons: first, reducing the sample to the “working poor” (e.g. Andress & Lohmann 2008) would exclude the effect of unemployment and thus leave out key situations of risk in which welfare states potentially intervene. Secondly, old-age poverty is excluded because it is based on different determinants than the poverty of those who could potentially participate in the labour market (for a review, see Kwan & Walsh 2018). Moreover, poverty among the retired population is presently much less pronounced than among the working-age population (Watson et al. 2018: 8–9). There is reason to believe that this will change in the future as pensions are expected to continue to decrease (Kwan & Walsh 2018: 1–2) and the threat of old-age poverty to rise (e.g. Ebbinghaus 2015). However, at the time of this analysis, poverty among the older population is still significantly lower. There is thus reason to suspect, that analysing the entire population would interfere with results.^{Footnote 18} Still, pension policies are included but they are only expected to be relevant as indicators of overall responsiveness of welfare systems. Apart from that, all other operationalisations of welfare stateness refer to the decommodifying effects of social policies. Besides this focus on the working-age population, the sample is further reduced to those cases, which have valid responses in all relevant variables in the analysis.

Lastly, weighting of data can be necessary. Several weights are provided in the EU-SILC data. They correct for non-response patterns, data shortcomings and adjust to household and population distribution in the target population. Depending on the unit of analysis, weights are given at household level and at person level (European Commission 2017: 33–45). Since the following analyses focus on the cross-sectional information for selected respondents, a weight for this specific population is chosen in descriptive analyses. In the multivariate analyses, weighted and unweighted analyses were tested. The unweighted results will be presented for three reasons: adjusting for the differences in the size of European populations (1) strongly reduces the presence of smaller countries, (2) severely complicated the estimation procedure, and (3) did not lead to noteworthy differences in the results.

6.2.4 Results of Bi- and Multivariate Analyses

It does not come as a surprise that poverty varies between the European countries and the share of population at risk of poverty by country (cf. Figure 6.6) corresponds to the numbers reported by the EU.^{Footnote 19}

In addition to this visual confirmation of the existence of variation, the intraclass correlation coefficient (ICC) is an important indicator of how much of the variance between individuals can be attributed to the contextual level—in this case, the country.^{Footnote 20} The resulting ICC of roughly six percent may appear low at first sight, however, it is in line with the numbers obtained in comparable analyses of poverty in Europe (e.g. Lohmann 2009; Saltkjel & Malmberg-Heimonen 2017). Before turning to the question of whether welfare states explain some of this variance between countries, and whether differentiating between conceptualisations of welfare stateness helps to shed some light on the results, the micro-level determinants of poverty are briefly examined (cf. Figure 6.7).^{Footnote 21}

The risk of poverty is distributed unequally among the population. Age exhibits a nonlinear effect. It increases the risk of poverty among those aged under 50 but decreases the risk among the highest age group compared to the youngest cohort. Furthermore, men do not exhibit a different risk than women, while lower educational attainment strongly increases the risk of poverty. The same effect can be found for all other employment status compared to regular employment. Unemployment increases the risk severely. This result—that poverty manifests especially among disadvantaged social groups—is in line with previous findings using similar data and a similar country-sample (e.g. Lohmann 2011; Ingensiep 2016; Brady et al. 2017).

In the following part of this chapter, the empirical application of distinct conceptualisations of welfare stateness will be explored. For this purpose, only the effects of social policy indicators will be reported. Still, all models include the micro-level determinants of poverty discussed above. Furthermore, all coefficients will be reported in two versions: with and without macro-level control variables. Those control variables are the GDP and the unemployment rate. While the results within sets of indicators are reported in one figure, they were all analysed in separate models.^{Footnote 22}

All five sets of indicators are tested: net replacement rates, benefit coverage, contribution period, benefit duration, and expenditure. As explained before, they are expected to represent operationalisations of two different analytical perspectives on the welfare state. The first is the Responsive Welfare State. According to the hypotheses, generous social rights compensate for deficits and combat poverty more effectively. The most relevant mechanism underlying this assumption is the provision of security. As discussed previously, measuring the Responsive Welfare State through social rights means that all sets of indicators represent potentially relevant operationalisations. The second relevant conceptualisation is the Enabling Welfare State. Here, activation was highlighted as an important mechanism. However, the operationalisation offers less alternatives. Therefore, only spending on active labour market policies represents a distinct operationalisation of this systematised concept.

In order to achieve a lean presentation of results, the analyses are reported in pairs of sets of indicators. Within the Responsive Welfare State, the first two sets—replacement rates and insurance coverage—relate to generosity of benefits.

The results reported in Figure 6.8 reveal that net replacement rates do not appear to reduce the risk of poverty significantly in my analyses. This partly contradicts the notion that generous benefits should decrease poverty (cf. section 4.2.2) and will be explored in the discussion, which succeeds this short description of main results. Insurance coverage, on the other hand, appears to reduce the risk of poverty in all policy fields with the exception of accident insurance. One may argue that especially the effect of pension insurance coverage is somewhat implausible considering that the analyses conducted are based on the working-age population. However, as the variables are expected to indicate the responsiveness of a welfare state in general, pension coverage may be a general characteristic of the welfare state, that affects the working-age population as much as the retired population.^{Footnote 23}

The second set of indicators representing the Responsive Welfare State includes the contribution period and the duration of benefit receipt (Figure 6.9). While it seems highly plausible that such criteria of eligibility and the temporal comprehensiveness of benefits and services reduce risks, neither of the variables appears to reduce poverty systematically in this analysis. I will discuss why this might be the case in more detail later. For now, suffice it to say that both sets of indicators do not appear to contribute to an explanation for differing risks of poverty between individuals in different European countries—at least not if their impact is examined separately.^{Footnote 24}

As a last set of indicators, several measures of social expenditure are included. It was argued repeatedly throughout this book, that social expenditure is a potentially ambiguous indicator. Its lack of clarity renders it an undesirable candidate for a clear and comprehensive measurement of any of the theorised conceptualisations of welfare stateness. They are, however, frequently referred to in the relevant literature (cf. section 4.2.2) and as such were even considered with their own hypothesis (Pov2). Indeed, all expenditure items produce negative results—with the exception of unemployment expenditure, which is negative but insignificantly so (cf. Figure 6.10). Thus, the redistributive budget (or “welfare effort”) reduces the risk of poverty in a country—however, it is not possible to attribute this effect clearly to one of the systematised conceptualisations of welfare stateness. It seems worth discussing whether other effects may be underlying the impact of expenditure on poverty. Furthermore, it is curious that of all policy fields, unemployment expenditure is the one with an insignificant impact on poverty—after all, individual unemployment is a strong predictor of being at risk of poverty. While unemployment expenditure tends to decrease poverty, this effect does not appear to be as systematic as the impact of other measures of social expenditure. I return to these issues later in the more detailed discussion of findings.

In contrast to the four expenditure indicators tied to more classical policy fields, one of the indicators is expected to capture a distinct conceptualisation: expenditure on active labour market policies is the only available indicator for measuring the Enabling Welfare State in this contribution. ALMP should reduce poverty by enabling individuals to re-enter the labour market or continue to participate even if they are in situations of vulnerability—such as unemployment or low educational attainment. Indeed, expenditure on ALMP decreases the risk of poverty significantly. This is generally in line with expectations. However, the mechanism of activation should be especially relevant when it comes to a moderating influence of welfare states on risks. Thus, the next step is to explore, whether ALMP spending actually captures the assumed reduction of risk emanating from situations of vulnerability.

For this purpose, several cross-level interaction effects are tested. If the effect of ALMP actually captures the assumed causality behind the Enabling Welfare State, it should significantly reduce the risk of poverty among vulnerable individuals—in particular, when they are unemployed or have a low educational status. Thus, cross-level interactions between ALMP spending on the national level and unemployment and low educational status (ISCED levels 0–2) on the individual level are tested (cf. Figure 6.11). In both cases, the risk of poverty is higher in those two vulnerable groups, but—in line with the expectation—ALMP spending reduces the risk more strongly in these groups than in the reference groups.^{Footnote 25}

When explaining different poverty risks between countries, the Enabling Welfare State is not the only conceptualisation assumed to be a relevant candidate for such a moderating effect. As discussed, especially the mechanism of providing security is considered to not only directly influence risks, but also mitigate the risk posed by vulnerable situations. Hence, the Responsive Welfare State can also manifest as a moderator here. Again, this is tested using cross-level interactions. While all sets of indicators with the exception of social expenditure were introduced as plausible candidates for this systematised conceptualisation, I only tested those combinations empirically that are theoretically plausible. Considering the sample is restricted to respondents in working-age, indicators representing pension policies are excluded from the reported interaction models.^{Footnote 26} Furthermore, since individual health is not included in these models either, sickness and accident policies are equally implausible candidates for an interaction. This leaves especially the unemployment policy indicators.

As Figure 6.12 reveals, unemployment insurance coverage tends to lower the risk in both groups, but there is not much evidence for a moderating effect. In case of education, a slightly steeper reduction of risk among individuals with low educational attainment appears (even though this effect is not statistically significant). In case of individual unemployment, however, no notable moderating effect can be detected.

The same result can be found for the interaction effects of net replacement rates, benefit duration, and length of the contribution period required to qualify for unemployment benefits (all included only in the appendix).^{Footnote 27} All three indicators of welfare stateness produced no significant effect by themselves and the cross-level interactions reveal no noteworthy pattern.

The only other—albeit insignificant—hint towards a moderating effect of unemployment policies is revealed by unemployment expenditure (cf. Figure 6.13). It shows a tendency to decrease risk, especially among those with low educational attainment and tends to lower risk more effectively among the unemployed. This is curious because unemployment expenditure does not produce a significant effect by itself (cf. Figure 6.10). Again, this should inspire caution when using indicators of social expenditure. Spending in the field of unemployment should decrease poverty considerably if one follows theoretical expectations. The fact that this is not the case in this empirical test underscores the concerns about whether expenditure correctly measures a generous redistributive budget.

Overall, the preceding analyses generated several insights regarding the link between the welfare state and the risk of poverty. Social expenditure and insurance coverage tend to decrease the risk of poverty, while benefit generosity, contribution period, and duration of benefit receipt all produced insignificant results. Furthermore, activating policies appear to mitigate risks emanating from vulnerable situations as expected. So does unemployment expenditure. Roughly, these results are in line with existing literature (cf. section 4.2.2), although some of the insignificant results—especially in case of the net replacement rates and unemployment expenditure are unexpected.

6.2.5 Summary and Discussion

The results of the preceding analyses generated insights regarding the five hypotheses derived from three explanations for differing levels of poverty in the literature. Benefit generosity did not appear to have an effect on the risk of poverty, which partly rebuts Hypothesis Pov1 (The more comprehensive the social rights, the lower the prevalence of poverty in a country). In contrast, benefit coverage did reduce risk, which can be interpreted as a partial confirmation of this hypothesis. Similarly, the weak effects of unemployment expenditure as the main and moderating effect partly contradict Pov2 (The higher the redistributive budget, the lower the prevalence of poverty in a country), while the significant effects of all other spending indicators tend to confirm it. However, there is not much support for Pov4 (The higher the redistributive budget, the lower the risk emanating from social determinants of poverty). Furthermore, the evidence for a moderating effect of social rights (Pov3: The more comprehensive social rights, the lower the risk emanating from social determinants of poverty) is weak at best. Lastly, activating policies leads to the expected reduction of risk among vulnerable individuals and therefore can be seen as evidence supporting Pov5 (The higher the effort directed towards activating policies, lower the risk emanating from social determinants of poverty).

While these are interesting results, which add to research on the matter and are partly controversial, the interest of this contribution does not rest on the verification of hypotheses. Instead, it explores whether or not conceptualising and operationalising welfare stateness following the proposed framework, helps to achieve a more standardised, transparent, and comparable process. The following discussion is therefore divided into two steps. First, I discuss how well the systematised concepts could be applied to the object of research. Second, the empirical measurement and the results are critically discussed.

Turning to the first step, two mechanisms were highlighted in the explanations and hypotheses guiding this exemplary empirical test. Those mechanisms are provision of security and activation. Among those two, security is assumed to influence the risk of poverty directly (accounting for a lower prevalence of poverty) and as a moderator (reducing the risk emanating from situations of vulnerability). Activation, on the other hand, is only expected to moderate the impact of being at risk. While it may also signal a particularly involved welfare state in a direct effect, it would do so only because it serves as a proxy for a more general perspective not pursued in this contribution.

I argued that underlying these two mechanisms are two systematised concepts of welfare stateness. The Responsive Welfare State combines those mechanisms and hypothetical effects where the welfare state is assumed to directly reduce the risk of poverty by providing security (e.g. through income replacement and insurance coverage). Moreover, the Responsive Welfare State is in line with the explanations for poverty where social policy is expected to have a moderating effect. The second relevant systematised concept is the Enabling Welfare State, which is mainly assumed to shape the risk of poverty as a moderator. Since these two conceptualisations represent very different perspectives on the welfare state, distinguishing between the two appears to be very helpful. Furthermore, applying this kind of differentiation not only helps to conceptualise and operationalise more systematically but to interpret and differentiate results. Overall, embedding the framework into the research process seems quite unproblematic and will help to standardise the structure of argumentations.

The second step of this discussion is more complex as it entails discussing how successfully the concepts were operationalised in order to explain the exemplary outcome. The essential question is how success can be determined. A confirmed hypothesis might partly help to assess this in the sense of nomological validity. However, this alone is somewhat tautological (cf. section 5.2.2). Since the initial selection of indicators was already guided by the premise that they should fit the nature of the systematised concept they were selected for, the effects they produce in analyses do not change this initial assessment. Such difficulty of determining measurement validity will be discussed in more detail at the end of the chapter (cf. 6.4). For now, I will focus on the results and their interpretation.

The Responsive Welfare State embodies what it is at the heart of the welfare state: securing against risks and meeting needs. As such, it incorporates the primary perspective we choose when we examine the link between social policies and poverty. In this exemplary analysis, I chose several indicators of social rights, which capture the institutional set-up of welfare states: eligibility criteria (contribution period), generosity (duration of receipt, replacement rate), and insurance coverage. These elements of social policy-making should embody responsiveness as they signal how quickly, how easily and how comprehensively security is provided. In theory, they should therefore all account for cross-national variations in the level of poverty (direct effect) and moderate the consequence of vulnerability. However, as the results show, this can only be observed for insurance coverage and in this case only when it comes to a direct effect. There is only a small tendency for coverage to benefit particularly those with low educational attainment, but it is not statistically significant. Various factors could explain this result. First, it could of course mean that the chosen indicators are not suitable. Perhaps, other indicators capture the essence of responsiveness better than the ones selected. Second, it could also mean that the selected operationalisation or source of the indicators is not suitable or that some countries distort the results (such as the CEE countries). Third, I cannot rule out that the impact of those mechanisms tied to the Responsive Welfare State is not as clear or as strong as expected. Since only a comparatively small part of variance can be attributed to the country level (ICC = 5.5%), differences in the risk of poverty between European countries are perhaps not that pronounced. All of these aspects require further attention in the future. Nevertheless, speaking very strictly, the indicators in this analysis were chosen following clear criteria. Thus, the fact that some of them fail to produce significant effects does not yet mean that they are unsuitable candidates for the Responsive Welfare State. Instead, they should be tested again—in other analyses of outcomes that are associated with this specific perspective on welfare stateness. Furthermore, the result that social expenditure does decrease poverty should only encourage further efforts to flesh out the nature of distinct conceptualisations of welfare stateness. Since we do not know what exactly is responsible for the observed effects of spending—redistributive budget, comprehensive benefits, etc.—this should discourage from using expenditure indicators if specific mechanisms and conceptualisations are tested. Since, however, there is a notable effect of expenditure and welfare effort (or redistributive budget) highlighted prominently in the relevant literature, it is only logical to try to figure out why.

In contrast to those mixed findings, the operationalisation of the Enabling Welfare State proved successful—at least in terms of confirming the hypothesis. Both—a direct and a moderating influence of ALMP spending on the risk of poverty were found. Since this indicator was reduced to spending on very specific and explicitly activating measures (training, employment incentive, supported employment and rehabilitation, direct job creation, and start-up incentives), it is more clear-cut than the heavily criticised other measures of spending. Still, it is only one indicator and it is highly recommendable to try other operationalisation of social investment in the future.

Some additional remarks have to be made about the preceding analyses. While I focused especially on the impact of macro-level indicators of welfare stateness and thus on the question if social policies account for some of the variances in the individual risk of poverty between countries, it predicts poverty with some limitations. A stronger focus on explaining poverty instead of exploring the impact of welfare stateness would be more informative if a longitudinal instead of a cross-sectional strategy was implemented. This especially refers to one particular aspect of the Responsive Welfare State: the moderating effect of securing individuals once they enter vulnerable situations. Thus, my cross-sectional analyses reveal how far welfare states potentially reduce (or increase) poverty among the unemployed or low educated but they do not tell us if welfare states lower the risk of poverty at the moment of entering unemployment.^{Footnote 28} Similarly, a comprehensive analysis of the Enabling Welfare State would also spell out the paths in more detail: is a reduction of risk among the low educated and unemployed actually due to their active participation in ALMP measures? These things have to be kept in mind when interpreting the causality behind the security and activation mechanism in the cross-level interactions that have been performed.

Further restrictions relate to the test of the introduced hypotheses and in particular to the first one (the more comprehensive the social rights, the lower the prevalence of poverty in a country). Here, the direct effect of generous social rights may be included, but its link to redistribution—as expected based on the redistribution paradox—was not explicitly tested. Neither was the support for redistribution, which may serve as an intermediary factor in this case (cf. section 2.2.2). A comprehensive test of these different paths, leading to a comprehensive explanation of poverty is not the main objective of this contribution. Therefore, the results should be discussed in terms of policy measures, but should not be understood as detailed contributions to the literature on poverty and the redistribution paradox.

Overall, the findings of this first test of the proposed framework are mixed. Using distinct conceptualisations of welfare stateness as an analytical tool that guides the operationalisation is certainly helpful. This is not just the case for the selection of indicators but also for the interpretation of results. Furthermore, using single indicators forces to ask whether we actually test the mechanisms we assume. Again, this seems very advisable. Still, the selection should be expanded in the future, tested on other dependent variables, and be critically discussed. The following section takes up at least the latter two recommendations, as it presents a second exemplary application of the framework using attitudes towards the welfare state as a dependent variable.

6.3 The Welfare State and Welfare State Attitudes

Analysing how far welfare state policies influence attitudes towards the welfare state is an evident undertaking. As such, it is also a very popular one. The literature on the matter was summarised in brief in chapter 4 and it revealed a great diversity in theoretical and conceptual as well as empirical approaches and results. The state of research will not be repeated at this point, but it is important to recapitulate the main explanations for how and why social policies are assumed to shape attitude formation in this case.

6.3.1 Conceptualisation of the Welfare State

Like in the previous example, the first question to be addressed is: which perspective on the welfare state is chosen? As was argued during the literature review in chapter 4, bottom-up and top-down perspectives on the welfare state can both be found in hypotheses about how it influences individual attitude formation. Thus, it is possible to ask how the welfare state influences the individual as well as how the individual perceives the welfare state and to what end. This is reflected in the mechanisms highlighted in the literature and the hypotheses that can be deduced from them. This relates to the second question that should guide conceptual considerations: which mechanisms are addressed? Within the top-down perspective, the welfare state is assumed to shape attitudes by conveying solidarity and justice principles through the mechanisms of socialisation. In addition, it is also argued that responsiveness—as represented by the comprehensiveness of the provision of security—leads to political support and positive attitudes towards the welfare state. The following explanations and corresponding hypotheses sum up these assumptions:

Explanation 3.1: Egalitarianism, universalism, and comprehensive social rights (including benefit coverage, generosity, and eligibility criteria) lead to positive attitudes towards the welfare state because individuals are socialised corresponding to such principles (top-down perspective).
Hypothesis Att1: The higher the emphasis on egalitarianism and universalism, the more positive the attitudes towards the welfare state.
Explanation 3.2: Comprehensive social rights (including benefit coverage, generosity, and eligibility criteria) lead to positive attitudes towards the welfare state because they provide security and thus increase political support (top-down perspective).
Hypothesis Att2: The more comprehensive social rights, the more positive the attitudes towards the welfare state.

Turning to the bottom-up perspective the focal point is the individual perception of welfare stateness and the perceived (potential or actual) individual benefit. Here, two principal explanations were deduced from the literature, which highlight the mechanisms of evaluation and self-interest. Underlying both explanations is the premise that social policies have to be known to a certain degree in order to be included in the formation of attitudes. The corresponding hypothesis highlights support for welfare states stemming from the assessment of performance in general (Att3) or specific considerations of individual costs and benefits (Att4) that are tied to social policies. This emphasises individual perception over responsiveness. However, it comes with the restriction that individual perception, which is assumed as a bridging hypothesis, cannot easily be included in empirical models. In essence, the link between welfare state and attitudes in both cases is therefore similar to Hypothesis Att2.

Explanation 3.3: Perceived fairness and good performance lead to more positive attitudes towards the welfare state because individuals evaluate these actions positively (bottom-up perspective).
Hypothesis Att3: The more comprehensive those social rights that are perceived, the more positive the attitudes towards the welfare state.
Explanation 3.4: Expected personal benefits from policy-making lead to positive attitudes towards the welfare state because this is in line with self-interest (bottom-up perspective).
Hypothesis Att4: The more comprehensive those social rights that are perceived to be beneficial for an individual, the more positive the attitudes towards the welfare state.

Again, it is irrelevant whether these four hypotheses exhaustively cover all theoretical assumptions in the literature. Instead, they serve as a minimal consensus on how and why welfare states may shape attitudes. For this contribution, it is important that they reveal different perspectives, highlight different mechanisms and help to answer the third question: which concepts of welfare stateness are addressed? When exploring this question, we have to take into account that the impact of social policies on attitude formation can be approached from the top-down as well as the bottom-up perspective. In accordance with the proposed framework, the chosen perspective and highlighted mechanisms guide the conceptualisation. Within the top-down perspective, two mechanisms are highlighted in particular: provision of security and socialisation. One could argue that redistribution and social stratification may provide explanations for differing attitudes as well—they are, however, of less importance as their influence is mostly covered by security and socialisation.^{Footnote 29} Thus, the way welfare states redistribute and shape stratification might influence solidarity and justice principles—but this happens in the sense covered by socialisation. Similarly, redistribution and stratification are also integral parts of responsiveness. Again, their influence can be seen as embedded in the security function. Since the way in which security is provided (redistribution) and its impact on social stratification is not explicitly emphasised in the tested hypotheses, the security mechanism seems to be the most relevant one. Thus, when the security mechanism is emphasised in examining the relationship between welfare stateness and welfare attitudes, the welfare state is conceptualised as the Responsive Welfare State. If socialisation is highlighted, the Normative Welfare State is at the core of the analysis.

Turning to the bottom-up perspective, evaluation and self-interest are tied to the concept of the Assessed Welfare State. As previously argued, this conceptualisation emphasises the individual perspective on the welfare state, its performance and potential individual benefits gained from it. Regardless of whether the process leading to the formation of a certain attitude is guided by self-interest or other considerations, it is always based on an individual assessment and is therefore summarised in the same systematised concept.

None of the three concepts is a better fit than the others and nothing speaks against conceptualising, operationalising, and testing all three. However, the distinction is of additional informative value and a means of achieving more targeted and comparable measurements, as the concepts underlying the welfare state are more clearly distinguished.

Two more questions have to be addressed before turning from theoretical conceptualisation to empirical operationalisation: which policy fields are relevant and which temporal perspective has to be chosen. Regarding policy fields, there is no reason to limit the selection to just one area. The Responsive Welfare State manifests regardless of the policy field and—contrary to the analyses performed in chapter 3—welfare attitudes will not be measured in just one specific policy area in the following analyses. It could be argued that there is reason to believe that some issues are more salient than others (e.g. unemployment, cf. section 4.2.3) and might therefore be more fitting for the Assessed Welfare State. However, since this cannot be generalised to all examined countries with certainty, reducing the analysis to only one policy field seems inadvisable. Furthermore, distinguishing between as many issues as possible could bear important insights.

Grasping the temporal perspective is a bit more difficult. Regardless of the perspective, attitudes can be formed with reference to the current state of social policies. However, it might be more than just the status quo that is responsible for attitude formation. This especially relates to the Normative Welfare State, which potentially shapes individual attitude formation over a long period and varies between individuals depending on their age. Likewise, the Assessed Welfare State may be affected by a certain delay, as it is unlikely that all citizens are up to date on political issues—especially in areas, such as accident policy, which may be far removed from individual reality. Such considerations are important and are—to the best of my knowledge—not prominent in the relevant literature.^{Footnote 30} Still, in my analyses, the selected datasets and indicators (cf. section 6.1.3) only cover features of welfare states in 2015 and therefore only allow a short-term perspective. Since the primary objective is to test and compare the different indicators and assess their fit with the systematised concepts, this restriction seems acceptable. The issue does however deserve much more attention in the future.

6.3.2 Operationalisation of the Welfare State

The previous discussion reveals three conceptualisations of welfare stateness, which help to explain different attitudes towards the welfare state. More specifically, they narrow down particular characteristics of social policy-making, which offer starting points for targeted operationalisations of these characteristics. In the previous chapter (cf. chapter 5), the following recommendations were introduced:

The Responsive Welfare State should be operationalised using indicators of social rights (e.g. benefit generosity, coverage, eligibility criteria).

The Normative Welfare State should be operationalised using indicators linked to principles that guide the provision of social services, such as universalism or egalitarianism (e.g. coverage, eligibility criteria).

Since they both refer to indicators stemming from the social rights perspective, there is some overlap in the operationalisation of these two conceptualisations within the top-down perspective. This is important to note and makes it difficult to distinguish between the two.

In case of the Assessed Welfare State, the expectations include that salient features of welfare stateness and those representing individual benefits represent the most fitting indicators.

This recommendation remains a bit more abstract than in the case of the other two concepts. Benefit generosity is expected to be more salient than other features and there is reason to expect that some policy areas are better known than others (e.g. unemployment policies might be more salient than accident insurance policies).

Consulting the list of indicators described at the beginning of this chapter (cf. 6.1.3) shows that almost all sets of variables promise an explanatory contribution as they relate to at least one of the three relevant conceptualisations of welfare stateness. Social expenditure is the only exception. It is included only because it can be assumed to be somewhat salient—albeit knowledge about actual expenditure may be biased (cf. chapter 4). The following analyses might therefore reveal important insights that can be added to the conclusions drawn from the analyses in the previous section of this chapter.

6.3.3 Additional Variables and Analytical Strategy

Several dependent and independent variables are included to test the conceptualisation of welfare stateness. The following part describes the operationalisation of these variables as well as the analytical strategy for testing the welfare state’s impact on attitudes towards the welfare state.

Dependent variables

Welfare attitudes are measured through various items in the following analysis. While chapter 3 relied on government responsibility alone—arguably a common indicator in the field—the ESS 2016 offers a more nuanced operationalisation of such attitudes. As noted previously (section 4.2.3), the operationalisation of welfare attitudes is far from consistent and surrounded by debate (e.g. Svallfors 2012b). As pointed out earlier, the preferred amount of government responsibility presents a conceptually different issue than more general welfare attitudes or other related phenomena such as welfare chauvinism.

In the following analysis, I take a more differentiated perspective on the matter than before and capture different components of welfare attitudes. Differentiating between distinct perspectives aids in determining, how the different welfare state indicators relate to different manifestations of welfare attitudes. The European Social Survey 2016 includes a comprehensive module on welfare attitudes called “Welfare Attitudes in a Changing Europe”, which suits this purpose well. It includes a battery consisting of six items measuring general attitudes towards the welfare state. Here, respondents indicate how much they agree with the following statement: social benefits (1) place a great strain on the economy, (2) cost businesses too much in taxes, (3) make people lazy, (4) make people less willing to care for one another, (5) prevent widespread poverty, and (6) lead to a more equal society. A principal components analysis reveals that two dimensions underlie these statements. The first includes the four negative attitudes towards social benefits, the second the two positive ones. The items are bundled in two mean indices, which are labelled welfare state scepticism and welfare state support.^{Footnote 31}

While preferred role of government (government responsibility for welfare provision) was covered by six items in the 2008 version of the ESS used in chapter 3, the 2016 wave only includes three of those items. In this case, respondents were asked to indicate how much responsibility they think governments should have for ensuring (1) a reasonable standard of living for the old, (2) the unemployed and (3) providing childcare services for working parents. The three items are combined in a mean-index.^{Footnote 32}

The last of the four dependent variables is welfare chauvinism and it is covered by one item asking respondents how long immigrants should have to wait until they are eligible for social benefits and services—ranging from immediately on arrival to never. An overview of these four dependent variables and univariate descriptive statistics are provided in the appendix.

Independent variables

The analyses include several independent variables, which address individual- as well as country-level features. Even though the focus rests on the explanatory contribution of different aspects of welfare stateness, this is explored in models representing comprehensive analyses of welfare attitudes best. Therefore, I added several socioeconomic and sociodemographic variables on the level of individuals. This includes information on respondents’ age (also age squared in order to account for non-linear effects), sex, educational level, employment status and financial insecurity. Most of these variables and their operationalisation correspond to those used previously in the analyses of the risk of poverty. Even though they stem from different data sources, the operationalisation of the different response categories follows the previous proceeding as accurately as possible. The only noteworthy difference lies in the operationalisation of poverty. While EU-SILC allowed to actually model whether a person falls under the poverty threshold, the ESS only includes broad income categories with many missing values and no correction for household size and composition. Therefore, a proxy is included: subjective financial insecurity (finding it very difficult to live with present income). In addition to the individual-level variables, the gross domestic product (GDP) and unemployment rate are included on the level of countries. This selection is guided by established literature on the matter, and was discussed in more detail in my previous analyses (cf. chapter 3 and section 6.2).^{Footnote 33}

Analytical strategy

In accordance with the theoretical conceptualisation and analogous to the analyses performed previously (cf. chapter 3) and the majority of similar research, multilevel analyses are performed (MLA). More specifically, since all four dependent variables are surveyed using a common response scale, hierarchical linear models are applied. It can be argued that dichotomising the items and performing logistic or linear probability analyses would be a viable option as well. In previous research, comparisons of both strategies did not produce gravely deviating results if the ordinal scale was maintained (in case of the responsibility items cf. Brady & Finnigan 2014: 25). Furthermore, it could be argued that since the original response scales were ordinal, multilevel ordered logistic models may be more appropriate than linear ones. Since, however, the initial scales were further broken apart and differentiated by constructing composite indices, the ordinal nature of the original items is softened up (with the exception of welfare chauvinism). Still, to avoid bias, ordered logistic models were tested in addition to the linear multilevel models. They did not lead to differing conclusions.

The advantages of multilevel linear models were already introduced in an earlier chapter of this book (cf. chapter 3). Like before, the analyses are realised in several successive models. Since it is plausible to expect that GDP and unemployment rate are not independent of features of the welfare state, they are added stepwise. The resulting regression equation for the full model predicting attitudes among individuals $i$ in country $j$ including all variables at the individual ($x_{ij}$) and country level ($z_{j}$), is identical to the one in chapter 3:

$$y\left( {{\text{attitude}}} \right)_{ij} = \gamma_{00} + \mathop \sum \limits_{h = 1}^{r} \gamma_{h} x_{hij} + \mathop \sum \limits_{l = 1}^{r} \gamma_{l} z_{l0j} + u_{0j} + e_{ij}$$

The explanatory contribution and fit of the models is determined based on various indicators. This includes tracing several information criteria such as AIC, BIC and Loglikelihood. Concerning changes in variance, the linear estimation allows obtaining information on R-squared in addition to the ICC. Like in chapter 3, Bryk and Raudenbush’s (2012) R-squared is obtained for the individual and the country level.^{Footnote 34}

6.3.4 Results of Bi- and Multivariate Analyses

A first glance at the distribution of the dependent variables (Figure 6.14) reveals noteworthy variations between countries. The most positive general attitudes towards the welfare state (welfare state support) are observed in Norway, Austria and Belgium, while Hungary, Lithuania and Estonia range at the lower end of welfare support. Interestingly, this does not seem to correspond directly to the index of more sceptical attitudes towards the welfare state confirming its status as an independent latent dimension. Here, Portugal, France and the UK exhibit the highest average values. Welfare provision is especially seen as government responsibility in Lithuania, Italy, Portugal and Spain while the Netherlands, the UK and Switzerland exhibit the lowest preference for government responsibility for the provision of welfare benefits and services.

Lastly, welfare chauvinism is strongest in Hungary, Czech Republic and Lithuania and lowest in Spain, Sweden and Portugal. Overall, there seems to be a tendency for less positive and generous attitudes in CEE countries and the UK. However, there is no obvious pattern for the rest of the European regions.

While this descriptive information already reveals some variation, it is important to determine how much variance can actually be attributed to differences between countries. Like before, the ICC is used for this purpose. Here, welfare chauvinism, preferred government responsibility and general attitudes towards the welfare state exhibit very similar ICC values. In all three cases, between eight and nine percent of variance can be attributed to the country-level. While these figures are not overwhelmingly great, they still reveal a noteworthy amount of variance that can potentially be explained by differences in welfare stateness. Considering that attitude formation is at its core an intra-individual process, a tenth of variance actually seems to be quite a lot and it corresponds to previous research and my own analyses of 2008 data (cf. chapter 3). In contrast to these three variables, scepticism towards the welfare state produces an ICC of only four percent. The potential to explain differences through country-specific features is thus considerably lower for this dependent variable. Nevertheless, the ICC is high enough in all four cases to justify performing multilevel analyses.

The next step entails exploring how much of the observed variance can be explained by indicators of welfare stateness and—more importantly—whether the indicators produce results, which are in line with the described expectations. While the focus of the analyses rests on the macro-level, a short glance at the effects produced by key micro-level variables is still important to make sure that main premises are fulfilled.^{Footnote 35} In order to provide a short overview of the individual-level determinants of the different welfare attitudes, Figure 6.15 summarises the result of the purely individual-level models. They reveal that general support for welfare states decreases with age (a), while age increases chauvinism (d) and demand for government responsibility (c) and has no significant impact on scepticism (b). Sex shapes at least some of the attitudes with male respondents being less in favour of strong government responsibility for welfare provision and more chauvinistic when it comes to extending benefits to immigrants. Furthermore, high education (tertiary, ISCED 5) leads to more positive general attitudes towards the welfare state (a), less scepticism (b), and less chauvinism (d) compared to the lower educational groups. The only exception is preferred government responsibility (c), where the lowest educational group is in favour of more responsibility. In contrast, being employed appears to lower support and increase scepticism. Lastly, respondents expressing difficulties when it comes to living on present income show less general support for welfare states, more chauvinism, but also a stronger preference for government responsibility and less welfare state scepticism.

Overall, the individual-level results reveal some interesting insights about the different dependent variables, which are noteworthy even though the focus of this contribution rests on the country-level. First, the patterns of effects for the four dependent variables are quite different—even between the two pairs of positive (support, responsibility) and negative (scepticism, chauvinism) attitudes. This supports the approach of distinguishing different facets of welfare attitudes. For instance, the fact that high education fosters general support for welfare states but decreases the preference for government responsibility seems worth more systematic exploration in future research. Similarly, it should be explored further why while those who contribute (the employed) overall tend to have more negative attitudes towards the welfare state than those who benefit (unemployed or retired individuals), this mainly manifests in scepticism and preferred role of government, not in chauvinism and only partly in general support. Second, the individual-level indicators only explain a small share of variance. The highest explanatory contribution is achieved in the case of chauvinism (roughly two percent).^{Footnote 36} Thus, the control variables may account for different socioeconomic and sociodemographic backgrounds of respondents but do not deliver comprehensive explanations for varying attitudes between individuals. While my findings do not contradict the state of research for the selected individual-level characteristics and the respective dependent variables (e.g. Häusermann et al. 2016; Eger & Breznau 2017; Kölln 2018), other explanatory factors should be included if a comprehensive explanation of micro-level processes is targeted.

After exploring individual-level determinants of welfare attitudes, the following section focusses on indicators of welfare stateness. Like in the analysis of the risk of poverty, the coefficients are reported in two versions: before and after controlling for GDP and unemployment rate.

Net replacement rates (NRR) represent the first set of indicators. According to the conceptual expectations, they should relate to various systematised concepts of welfare stateness: the Responsive, Normative and Assessed Welfare State. Since all three conceptualisations are assumed to explain differences in attitudes, net replacement rates should be strong indicators—albeit being too undifferentiated to be attributed to either one of the three concepts. The results reported in Figure 6.16 support the expectation that NRRs indeed deliver an explanation for varying attitudes. Respondents in countries with more generous benefit replacement exhibit more positive general attitudes towards the welfare state (a), are in favour of more government responsibility (b) and are less sceptical. In case of unemployment replacement rates, they also tend to be less chauvinistic. There are some exceptions—for instance, the NRR in case of pensions does not appear to predict welfare state scepticism and chauvinism. Furthermore, it loses significance for the explanation of preferred government responsibility once the control variables are included, while it is only significant as a predictor for welfare state support with controls. The latter happens in several instances, which confirms the importance of controlling for GDP and unemployment rate, which are clearly cofounded with welfare stateness. Overall, the majority of indicators show a clear relationship between benefit generosity and attitudes, which is supported by the post-estimates.^{Footnote 37} When evaluating these results and their fit with the three potentially relevant conceptualisations, there seems to be support for using this indicator, since generosity—in most policy fields—fosters support for the welfare state. However, whether this is due to responsiveness, the socialisation of respondents or their perception and evaluation cannot be determined at this point.

While NRRs may suit several conceptualisations of welfare stateness, the contribution period required to qualify for a benefit, was selected as an indicator fitting only two conceptualisations: the Responsive Welfare State and partly also the Normative Welfare State. Turning to the results, the picture seems to be a bit ambivalent (cf. Figure 6.17). While being eligible for sickness benefits without having to fulfil any contribution period tends to increase positive attitudes (a) and decrease scepticism (b), it also increases chauvinism (d). In contrast, a short contribution period required to be eligible for pension benefits, tends to lower support and increases scepticism (d), while the contribution period is insignificant in the case of unemployment benefits. There seems to be no clear pattern and without exploring in detail why the contribution period in the case of sickness tends to foster positive attitudes and chauvinism at the same time, it is not possible to evaluate whether the indicator adequately captures either the Normative or the Responsive Welfare State.

The next set of indicators – benefit duration—was proposed as mainly providing a measurement of the Responsive Welfare State. As such, it is assumed to shape attitudes only as so far as factual performance automatically generates support. As the results (Figure 6.18) show, benefit duration only explains varying welfare attitudes between countries in some policy fields. The only significant effects are produced by unemployment benefit duration, which increases general support (a) and decreases chauvinism (d) and by accident benefit duration, which increases scepticism (c). While the effects of unemployment benefit duration are in so far plausible, as unemployment policies are assumed to be comparatively salient features, the effect of accident benefit is counterintuitive and may signal some other underlying or confounding effect, which cannot be explored in more detail at this point. In addition, as salience is no prerequisite of the Responsive Welfare State, it is possible that the observed effect of unemployment benefit duration also expresses another confounding effect. Overall, the results seem to suggest that the Responsive Welfare State is of secondary importance for the explanation of attitude formation—if (and this has to be analysed very critically later on) the selected indicators allow an adequate operationalisation of this concept.

Contrary to benefit duration, insurance coverage was selected as an indicator of universalism of benefits, which may not only represent the Responsive Welfare State, but also the Normative Welfare State. However, the indicator produces barely any significant effects—especially not after controlling for GDP and unemployment rate (Figure 6.19). At first glance, insurance coverage therefore does not seem to be a fitting indicator for either conceptualisation, which is supported by a neglectable share of explained variance.^{Footnote 38} Again, one could also critically add that it may signal a subordinated relevance of the conceptualisations in case of welfare attitudes.

The last set of indicators consists of several measures of spending (Figure 6.20). Again, the disadvantages of this operationalisation have to be kept in mind. As argued before, expenditure was included in this analysis because it is by far the most commonly used among the single indicators. Furthermore, there is some reason to believe that it is among the more salient features of social policy-making and as such may be considered when trying to capture the Assessed Welfare State. Even though perhaps salient, it may still be biased by a lack of accurate knowledge about actual spending (as discussed in section 4.2.3). A first glance at the relevant coefficients reveals several insights. First, social expenditure increases a positive general attitude towards the welfare state (a). This happens regardless of the addressed policy field and remains a robust finding even after controlling for GDP and unemployment rate. Second, neither of the expenditure measures delivers significant explanations for variations in welfare state scepticism and preferred role of government—even though there seems to be a small (but insignificant) tendency to support more government responsibility if health care spending is low. Interestingly, health expenditure and overall expenditure also decrease welfare state chauvinism—this time significantly so. Thus, high spending in those fields does not appear to activate a need to secure benefits against outsiders, but instead fosters generous attitudes. This could be interpreted in the sense of the Normative Welfare State. However, this has to happen in combination with other indicators, more closely tied to this conceptualisation and suffering from less fuzziness.

Lastly, ALMP spending, which was only included for the sake of exhaustiveness, does not produce significant effects. Since there is no reason to assume the Enabling Welfare State shapes attitudes, this finding is in line with the expectations.

To sum up the main findings: benefit generosity leads to more positive attitudes towards the welfare state in most models. A similar tendency is observed regarding social expenditure, although not in the case of scepticism and only partly for government responsibility. Benefit coverage in contrast, does not appear to notably shape attitudes. Lastly, the duration of benefits only fosters positive attitudes in the case of unemployment duration and contribution period—as an indicator of eligibility criteria—exhibits a tendency to increase scepticism in case of contribution to pension schemes and chauvinism in the field of sickness polices. The question how well the distinction between different conceptualisations of welfare stateness was embedded in this analysis and how accurately the different perspectives on the welfare state were modelled will be addressed in the following section.

6.3.5 Summary and Discussion

Do social policies explain differing attitudes towards the welfare state between countries? The results obtained in this chapter support the same assumptions as the literature on the matter (cf. section 4.2) and as my previous analyses (cf. chapter 3). Indeed, welfare policies account for cross-national variation and very broadly speaking, comprehensive and generous social policies foster positive views towards the welfare state. This finding can be integrated in the state of research and even the effects pointing in other directions (such as the fact that some pension and sickness policies increase scepticism and chauvinism) could be interpreted in line with existing argumentations. In fact, one can interpret this result as support for all hypotheses stated previously. However, there seems to be more support for those hypotheses highlighting the bottom-up perspective (Hypothesis Att3 and Hypothesis Att4) since indicators that should capture the top-down perspective perticularly well, mostly produced ambiguous, unexpected, or insignificant results.

However, the aim of this contribution is to offer more than that. By conceptualising and testing narrower perspectives on welfare states, the results offer additional insights and help to differentiate to a higher extent. Based on the argument that adequately modelling the different hypotheses in the literature requires not only adequate measurement of the dependent variable but also of the independent variable, it introduces a more nuanced view on welfare states. In contrast to the previous analysis of the relationship between social policies and the risk of poverty, this approach is put to the test more extensively in the case of welfare attitudes because here the top-down and the bottom-up perspective applies. Potentials and limitations of the proposed approach should therefore be revealed especially in this exemplary case. The usefulness of the analytical framework is assessed in two steps. First, I discuss the applicability of the systematised concepts to the object of research. Second, I discuss the empirical measurement, bearing in mind validity criteria.

Three systematised concepts were applied in order to capture different perspectives on welfare stateness behind mechanisms and hypotheses. The first conceptualisation is the Responsive Welfare State, which underlies hypotheses assuming that factual performance creates general support for welfare states (independently of individual perception). The second concept is the Normative Welfare State. This conceptualisation can be found if hypotheses focus on the way in which welfare states convey solidarity and justice principles and socialise citizens. Finally, the Assessed Welfare State can be seen as a counter perspective to the Responsive Welfare State as it manifests in individuals’ perspective on welfare states and their assessment of policy-making. Especially these two closely related conceptualisations reveal the potential usefulness of distinguishing different facets of welfare stateness early in the research process. Even if they capture very similar things, the Responsive Welfare State may perform “well” without individuals explicitly noticing why, while the Assessed Welfare State requires a certain amount of knowledge about policy-making. Discussing this difference may already enrich debates and the attempt to operationalise the difference empirically may contribute even more insights.

Evaluating how well this empirical operationalisation of the concepts worked, is however again difficult. And,—as discussed before—it is complicated even more by the fact that the indicators selected for this empirical test can be suitable for more than one conceptualisation.

The operationalisation of the Responsive Welfare State is the same as in the analysis of poverty. Indicators are attributed to this conceptualisation if they signal how comprehensively security is provided (benefit coverage, replacement rates, duration of receipt) and how quickly benefits can be obtained (contribution period). The fact that these indicators for the most part fail to deliver clear results is conspicuous. It could signal that responsiveness is not adequately captured through these indicators. However, it could also mean that the mechanisms tied to the Responsive Welfare State are not as influential for the explanation of attitudes as expected. It is difficult to discuss whether this might be because another conceptualisation such as the Assessed Welfare State (as a counter perspective) is actually more suitable for the research subject. Rejecting an entire conceptualisation requires more detailed analyses and more attempts to capture the Responsive Welfare State—perhaps from another perspective. If a more clear-cut operationalisation of the Responsive Welfare States does not succeed, this finding may bear insights that go beyond matters of operationalisation. If responsiveness does not shape attitudes as much as expected, hypotheses that highlight mechanisms tied to the provision of security might require a critical examination. This renders the exploration of nomological validity all the more difficult.

Findings are similarly unsatisfying when it comes to the measurement of the Normative Welfare State. Here, benefit coverage and benefit generosity were highlighted as operationalisations. Benefit coverage does not contribute much to the explanation of different attitudes. Even though replacement rates overall produce results in line with expectations, this cannot be taken as clear confirmation for their usefulness as operationalisations of the Normative Welfare State because they may also suit other conceptualisations—especially the Assessed Welfare State. Turning to the latter, indicators were selected if they are assumed to be salient and represent either individually perceived performance or potential benefits. Out of the sets of indicators, this referred to net replacement rates and social expenditure.

The aim of this project is not to assess which perspective on the welfare state is theoretically more appropriate, but how to measure different conceptualisations more accurately. Still, at this point one wonders, whether attitude formation is actually related as much to a top-down perspective as to a bottom-up perspective. All conceptualisations but the Assessed Welfare State were only partially confirmed empirically. Still, the indicators chosen to operationalise the Assessed Welfare State—social expenditure and net replacement rates—are not only used in a very general way, but also lack clarity. I will thus refrain from detailed assessments of theoretical premises; exploring much more, whether we can actually assume that the welfare state shapes attitudes independently of individual evaluations still seems highly valuable.

Again, there are restrictions to the performed analyses and thus limitations when it comes to their interpretation. One is the mentioned incapability of promoting one conceptualisation over another based on the results. Even more serious, however, is the fact that formulating substantial recommendations for the empirical operationalisation of each of the concepts is almost impossible because similar indicators could suit all conceptualisations. Thus, there is only a tendency to say that at least benefit generosity (as measured through replacement rates) is more closely related to the Assessed Welfare State because other more specific indicators of either the Responsive or the Normative Welfare State fail to produce results that are in line with the hypotheses. However, this definitely requires more research.

Another restriction is that actually testing the Normative Welfare State and the mechanism of socialisation is difficult, as it requires detailed data on the contextual and the individual level. It is a strong postulate to say attitudes are formed because respondents grow up under certain political conditions—or at least live under them long enough. Strictly speaking, this would require longitudinal or retrospective data for respondents and information about social polices during their formative years. Overall, the Normative Welfare State is perhaps the most difficult concept to grasp.

A final limitation is that the hypotheses described all involve a bridging assumption that has not been tested in detail. Thus, assuming the welfare state shapes attitude formation, because it raises support, or that individual preferences entail weighting up social policies with one’s own values is full of prerequisites. And for the sake of completeness, I must also point out that the analyses carried out cannot do justice to the complex contributions on policy feedback and the interdependency between public opinion and policy-making (Breznau 2017, 2018). Like in the analysis of the risk of poverty (cf. section 6.2), this exemplary test of the analytical framework was intended to compare operationalisations, not to contribute substantially to the complex literature on welfare attitudes.

There are however also important insights to be gained for the objective of this book. In particular, the distinction between the bottom-up and the top-down perspective on the welfare state is fruitful in this exemplary analysis—not necessarily in terms of operationalisation, but in terms of discussing theoretical premises and interpreting results. Furthermore, the results at least hint towards confirming the expectation that unemployment policies constitute especially salient features of welfare stateness and therefore shape attitude formation—in the sense of the Assessed Welfare State—the most.

Even though this is not the focal point of this book, it should also be noted that my results support the need for a more differentiated perspective on welfare attitudes. The different measurements offered in the 2016 ESS show potential in capturing various facets of the issue. I would like to encourage such endeavours based on my results.

6.4 Concluding Remarks

Aiming at operationalising distinct conceptualisations of welfare stateness bears great potential to improve the transparency and comparability of measurements. It does so for several reasons. It demands that we look more closely at the mechanisms behind theoretical premises, justify the choice of indicators, discuss the scope of the results they offer, and it guides the interpretation of results. The conceptualisations are a tool not unlike others used in scientific endeavours when it comes to the conceptualisation of complex phenomena. However, as argued in the second chapter of this book, welfare states—as very complex arrangements of policy-making—have so far mainly received attention when it comes to their operationalisation as dependent variable and it does require a different perspective when their features are included as an independent variable. The analytical framework proposed in this book and tested in this chapter aims at filling this gap. And it succeeded in doing so at least partially by presenting a guideline for important analytical steps and also delivering some more specific insights—such as the fact that distinguishing between those operationalisations tied to assessment and those representing the mode and effectiveness in which welfare states function, does indeed narrow down operational choices. This refers to the indicators themselves (benefit coverage seems to be a promising indicator of responsiveness and benefit generosity appears closely tied to individual assessments) but also to analytical strategies (oftentimes moderating effects are more adequate than direct ones—especially in case of responsiveness).

While evidence for the usefulness of distinguishing conceptualisations of welfare stateness exists, discussing the validity of the proposed operationalisations from a broader perspective is difficult. Here, several things have to be noted critically. First, only a selection of possible indicators was tested. As argued in the beginning of this chapter, I prioritised indicators used previously in the literature that fulfil the criteria of clarity, comparability, and availability best. However, it is not possible to rule out that other indicators that were not tested are potentially more suitable. Thus, in the sense of content validity, it is not possible to determine with confidence that the selection perfectly fits the concepts. Second, most chosen indicators fit more than one conceptualisation of welfare stateness and therefore judging, which fit is better or worse should be done with caution. Again, this restricts the discussion of content validity. Third, the initial introduction of the indicators of welfare stateness already revealed that correlations among potential candidates for a similar systematised concept are difficult to interpret. This hampers determining convergent validity. Lastly, concluding that those indicators are suitable (in the sense of construct validity) that confirm established hypotheses is potentially tautological. As discussed in section 5.2.2, operationalisations validated through hypotheses cannot be used to test the very same hypotheses.

All of these issues were foreseeable and they were mentioned before in this book. They are the reason why more data sources were discussed than tested, why more than one dependent variable was selected, and why the discussion of selection criteria and measurement validity criteria pervades throughout almost all sections of this book. Even though my analyses only shed light on some of the possible candidates for distinct operationalisations, I believe that following these and similar guidelines to test and interpret operationalisations of welfare stateness in follow-up research has great potential to improve measurement even more. The findings in this chapter should be used to deduce more indicators fitting the different conceptualisations. Potential additions include transfer share^{Footnote 39} for the Responsive Welfare State, private-public mix of benefit financing for the Assessed Welfare State, and many more.

Furthermore, the temporal perspective deserves much more attention—especially when it comes to the operationalisation of the Normative Welfare State. If the focus rests on the socialising potential of social policies, the set-up of welfare states during the formative years of respondents might bear valuable insights.

Regardless of these restrictions, this chapter demonstrated that taking a step back to conceptualise the chosen analytical perspective on the welfare state before turning to its empirical operationalisation is quite fruitful. It can be integrated easily into argumentations, it helps to frame hypotheses, justify operationalisations, and interpret results. At a minimum, it forces to spell out the assumed mechanisms in much more detail. Thus, even though only a part of the proposed indicators proved to contribute to explanations for the selected exemplary outcomes, the potential to standardise the selection by explicitly committing to a systematised conceptualisation is great.

Notes

1.
Analogously to chapter 3, EU-27 plus Norway and Switzerland are chosen. Croatia is left aside in the analyses because macro-data availability for the relevant period is unfortunately limited.
2.
This includes the United Kingdom, which was still part of the European Union in 2019 when these analyses were conducted.
3.
This high coverage is not only due to very thorough sampling and surveying but also to statistical imputation performed prior to the data release (cf. European Commission 2017: 55–56).
4.
18 EU member states plus Norway and Switzerland.
5.
Those analyses were performed with data from 2008 (cf. chapter 3). I chose not to update the analyses when new macro-level data was published in 2019, because I believe that comparing the periods allows to draw some interesting additional insights and will be valuable to determine the robustness of some of the findings over time.
6.
Cf. Table A6.1–1.
7.
There is one important addition necessary, which relates to Greece. In the SIED data, there are no up-to-date information on health and accident insurance coverage in Greece. This may be due to the fact that the Greek healthcare system was in some turmoil in the wake of the financial crisis. As a result of the crisis, there has been a loss of health insurance coverage—especially among long-term unemployed and self-employed citizens (OECD 2015: 10). This was met with several attempts to reform the healthcare system (WHO 2019, 2018). Since insurance coverage in 2015 is thus most likely distorted by a temporary sharp increase in the number of uninsured individuals, health and accident insurance coverage is missing in this case.
8.
Apart from Malta (1 week) and Estonia (2 weeks), none of the countries has any waiting period for being eligible for accident insurance.
9.
More information can be found in the appendix.
10.
Since the data on social expenditure stem from another dataset than the information about spending on active labour market policies, some overlap between unemployment spending and labour market spending may exist. Interpreting the stacked graph as a sum of expenditure might thus be biased. Furthermore, data on ALMP spending in Switzerland are missing.
11.
Due to the low number of countries (N = 29), strength of effects is interpreted as a tendency even if statistical significance is not given (cf. Table A6.1–2).
12.
Benefit duration and contribution period have been used as descriptive indicators (Kuitto 2018) and in composite measures and typologies (Esping-Andersen 1990), but—to the best of my knowledge—they are not commonly used as single indicators.
13.
This was captured in Explanation 2.3: Comprehensive universal social rights decrease poverty, because these actions shape patterns of social stratifications in a way that increases equal opportunities, thereby lowering risks among those who are especially vulnerable.
14.
A summary of all independent variables used in the analysis of the EU-SILC dataset is provided in Table A6.2–1 in the appendix.
15.
Since there were no missing values or additional categories in the original variable, there was no need to distinguish a third (or diverse) sex.
16.
Since cross-level interactions are included, it can furthermore be advisable to use grand-mean centring to reduce multicollinearity (Dedrick et al. 2009). However, since there are no continuous micro-level variables in the models and centring of macro-level variables is not necessary, centring is not applied.
17.
A side note to support the caution advised by Heisig and Schaeffer (2019): if the slope is not explicitly set to random, the significances of the lower level variable and the interaction effect itself are severely overestimated.
18.
Still, additional analyses of the full population sample were conducted. They will only be discussed if they offer other or complementary findings.
19.
Deviations from official reports (cf. European Commission 2019) are due to the listwise deletion of missing values.
20.
The reported ICC stems from the Random-Intercept-Only-Model before predictors are added.
21.
All analyses in this chapter are visualised in coefplots (Jann 2014), which allow a comparatively easy interpretation of results. More comprehensive tables of all estimated models and information on the model fit are provided in the appendix.
22.
The individual models are reported in the appendix to this chapter.
23.
Tentative analyses of the full sample reveal comparable effect directions. However, they have a much greater error. This supports the argument that poverty in old age follows a different logic than poverty among the working-age population.
24.
Both indicators are frequently used for the construction of composite indices and typologies (most prominently by Esping-Andersen 1990). However, they have—to the best of my knowledge—not been used as single indicators in the literature so far.
25.
The effects only brush statistical significance (p < 0.10). In the light of the small number of macro-level cases, they are still interpreted as indicating a moderating effect.
26.
As discussed before, the only reason pension policies are not omitted all together is that they may signal overall responsiveness.
27.
Cf. Figures A6.2–1, A6.2–2, and A6.2–3 in the appendix.
28.
Obviously, low educational attainment is not effected by this.
29.
They would be of more importance in analyses of attitudes towards specific modes of redistribution or if hypotheses assumed a specific way of stratifying societies is responsible for attitude formation. Both research objectives are plausible but represent very specific examples, which are not pursued further in the following analysis.
30.
Perhaps, with the exception of studies using welfare state typologies based on historical trajectories—although the broader time horizon is seldom a clearly stated objective.
31.
Cronbach’s alpha is 0.73 for welfare state scepticism and 0.67 for welfare state support.
32.
Cronbach’s alpha is satisfactory (0.68).
33.
For more information and univariate descriptive statistics cf. Table A6.3–1 in the appendix.
34.
In order to avoid bias stemming from design issues or the under- or oversampling of population groups, the ESS offers several weights (ESS 2014). In descriptive analyses (in the appendix), a combination of population and design weight is chosen. All descriptive statistics in the appendix use weighted data; the multivariate analyses are calculated with and without active weights. Since there is no difference in effect direction and significance, the unweighted version is reported.
35.
A comprehensive documentation of all estimated models and the effects of further control variables is provided in the appendix.
36.
Information on explained variance and other fitness indicators is provided in the full regression tables in the appendix.
37.
For instance, Bryk and Raudenbush’s R-squared indicates a value between .1 and .3 for all significant indicators (before adding control variables). These and more fitness indicators can be found in the corresponding regression tables in the appendix.
38.
Cf. tables in the appendix.
39.
Meaning the share of individuals who are entitled to benefits but do not receive them (e.g. Brady & Bostic 2015).

Author information

Authors and Affiliations

Mainz, Germany
Katharina Kunißen

Authors

Katharina Kunißen
View author publications
You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Katharina Kunißen .

6.1 Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary file1 (ZIP 2657 kb)

Rights and permissions

Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.

Reprints and permissions

Copyright information

About this chapter

Cite this chapter

Kunißen, K. (2023). Welfare Stateness as an Explanatory Variable: Empirical Illustration. In: The Independent Variable Problem. Sozialstrukturanalyse. Springer VS, Wiesbaden. https://doi.org/10.1007/978-3-658-39422-6_6

Download citation

DOI: https://doi.org/10.1007/978-3-658-39422-6_6
Published: 13 November 2022
Publisher Name: Springer VS, Wiesbaden
Print ISBN: 978-3-658-39421-9
Online ISBN: 978-3-658-39422-6
eBook Packages: Social Science and Law (German Language)

Publish with us

Policies and ethics