Introduction

Abstract

The welfare state is an integral part of western industrialised democracies. While its most obvious function is to provide a safety net for individuals, who are threatened by social risks, it also has far-reaching effects on a multitude of social phenomena, such as the formation of attitudes and behaviours.  This introductory chapter illustrates why the welfare state serves as a popular independent variable and sets out the main problems associated with its operationalisation, leading to the research questions of this book.

The welfare state is an integral part of western industrialised democracies. While its most obvious function is to provide a safety net for individuals, who are threatened by social risks, it also has far-reaching effects on a multitude of social phenomena, which seem only indirectly associated with social policy arrangements at first glance. Examples include the formation of attitudes and behaviours. Such relationships reveal that the welfare state is much more than a conglomerate of social rights: the arrangement of social policies actively shapes e.g. social stratification, incentivises behaviours, and conveys solidarity and justice principles. In this way, it bears the potential to influence almost all areas of social live. Who is covered by policies? How is social security organised and financed? How generous is a welfare system? Does it aim at preserving status differences or does it promote egalitarian principles? The answers to those and similar questions reveal important characteristics, which are quite consequential for individuals covered by a welfare state. Because of this strong tie between social policies, social inequality, and various other social phenomena, the welfare state is an important object of research in a variety of disciplines within the social sciences. Much research is focussed on the welfare state itself, but especially comparative approaches often highlight the consequences of different welfare state arrangements and ask how far different social policies (as independent variables) lead to different outcomes. Such outcomes can be found mainly on two analytical levels: the macro- and micro-level. In this contribution, I focus in particular on outcomes on the micro-level and thus mainly argue within a multilevel framework.¹

There is a great variety of outcomes examined in this regard: some research focusses on direct consequences of social policies like well-being (e.g. Cruz-Martínez, 2017; Schuck & Steiber, 2017) or the consequences of risk exposure (e.g. Angel & Heitzmann, 2015). Other scholars examine phenomena which are influenced in more subtle ways, such as attitude formation (e.g. Jordan, 2013; Eger & Breznau, 2017), political participation (e.g. Schneider & Makszin, 2014), political trust (e.g. Mattila & Rapeli, 2018), and much more. In such cases, an empirical operationalisation of the welfare state or of specific elements of social policy-making is always required. However, some substantial problems arise concerning prevalent operationalisation practices. Essentially, these problems all relate to one key issue: while there is a great number of contributions addressing the measurement of differences between welfare states per se and as a dependent variable, there is a distinct lack of feasible recommendations when it comes to the operationalisation of welfare stateness as an explanatory variable (or independent variable). In light of the great number of studies assuming an effect of welfare policies on other social phenomena, the lack of standardised proceeding we encounter in this context surprises. To this date, there is no systematic test of how such varying proceedings may affect results and their comparability. Similarly, a detailed conceptual discussion on which features of the welfare state are relevant for the explanation of specific outcomes is missing. This contribution intends to fill both gaps. It seeks to unravel the pitfalls and potentials of existing approaches and to propose a conceptual framework, which is intended to guide empirical operationalisations and may help to overcome some of the problematic issues that are identified.

1.1 Aim of the Study and Research Questions

At its core, this contribution takes up a very specific problem within a very broad field. While the treatment of the welfare state as an independent variable is indeed a small issue within the well-established field of comparative welfare state research, it is an important one in light of the great body of work that is affected. In the course of this book, I will argue that the lack of a standardised strategy in this case considerably impairs reliability and comparability of analytical approaches and empirical results, as the concepts behind the operationalisations are not sufficiently clear. This leads to one essential premise: we must be sure of the concepts we want to measure with an operationalisation if we are to measure them correctly.

In a way, this endeavour is related to a growing body of literature on replications in the social sciences and beyond. More specifically, it takes up issues that are raised by scholars who explore the reproducibility (or replicability) of results (Munafò et al., 2017; Breznau et al., 2019) and the transparency of scientific publications using cross-national survey research data (Damian et al., 2019). While those scholars mainly emphasise issues related more generally to the transparency of methodological choices and the rules that have to be followed when it comes to performing empirical analyses and interpreting results, my contribution focusses on one specific example: treating the welfare state as an explanatory factor. Furthermore, I approach issues from a more conceptual point of view. Thus, problematic practices such as p-hacking² and HARKing³ (Munafò et al., 2017: 1–2) are only brushed. Instead, the focus rests on the potential to avoid such sources of bias if an agreement on the process of operationalisation and the specification of underlying concepts is reached.

In order to narrow down the aim and research questions pursued in this book, it has to be discussed why the measurement of welfare states as an independent variable is an important issue and why its operationalisation is not (yet) standardised in this particular case. Starting with a very general observation, many comparative empirical studies dealing with individual-level outcomes implement a multilevel design⁴ in which properties of the welfare state serve as independent variables at the level of countries. As there is no agreed-upon way to operationalise ‘welfare stateness’ as an indicator in such studies, scholars usually borrow instruments from literature, which examines welfare policies as a dependent variable. These instruments include a broad spectrum of approaches such as a variety of different welfare regime typologies, single indicators, and composite measures. Even though the demand for treating the welfare state as an independent variable is high, the literature hitherto lacks comprehensive discussions of the extent to which the various existing measures can actually serve as suitable independent variables and the problems that may be associated with different operationalisations. In order to address these topics, it is necessary to have a brief look at the debate surrounding the general measurement of different welfare states.

From early on, research on the welfare state has sparked lively and critical methodological debates.⁵ More recently, the so-called dependent variable problem receives growing attention (e.g. Clasen & Siegel, 2007). This methodological debate emerged as a by-product of a discussion about welfare state change and retrenchment (e.g. Pierson, 1996). A key problem identified in this debate was the lack of a common understanding of what the object of research—the dependent variable—entails and how it should be measured (Green-Pedersen, 2004). Until today, there is an ongoing discussion based on the repeated observation that different conceptual and operational strategies produce different results (e.g. Kühner, 2007; Bolukbasi & Öktem, 2018).

In light of this existing debate on how to conceptualise and measure features of the welfare state best, one might wonder why we need an additional independent variable perspective instead of simply relying on the insights produced by literature addressing the dependent variable problem. Five arguments speak in favour of such an endeavour. First, there is no thorough account of how different conceptualisations affect explanatory power and informative scope when used as independent variables. Only recently, scholars start to voice concerns because existing measurements are treated as interchangeable options for the operationalisation of welfare policies as dependent as well as independent variables (Bolukbasi & Öktem, 2018). Second, the existing methodological discussions mainly remain on the macro-level. How far the proposed measures can be embedded in macro-micro-analyses remains unclear. Third, the exchange of feasible recommendations between general literature on the welfare state and research, which examines its outcomes, is highly underdeveloped. Systematic comparisons of varying strategies are rare and focus only on consequences of different ways to operationalise within one of the approaches and for singled out dependent variables (e.g. Howell & Rehm, 2009; Bergqvist et al., 2013). Fourth, difficulties in choosing an appropriate independent variable are frequently expressed in the literature and ultimate selections often entail compromises.⁶ Fifth, it has never really been discussed or tested, whether the existing indicators adequately capture theoretically assumed mechanisms in multilevel analyses of the outcomes of welfare policies, even though concerns are voiced sporadically (e.g. Pfau-Effinger, 2005). Exploring in more detail, why features of the welfare state serve as independent variables, how they are empirically operationalised and whether different approaches influence results, should therefore be a helpful contribution to the state of research. The first research question that is pursued in this contribution addresses this issue:

How comparable are the results that emerge from different approaches to operationalising the welfare state as an independent variable?

The second research question is a logical consequence of the first. After the description and critical assessment of the status quo, it constructively focusses on how to potentially improve the inclusion of the welfare state as an explanatory factor:

How can we derive a more standardised, transparent and comparable approach to operationalising the welfare state as an independent variable?

Before exploring in more detail how this contribution approaches these issues, one restricting remark is necessary. Overall, the main aim of this study is not to provide a complete overview and final evaluation of existing and possible conceptualisations and empirical operationalisations of welfare stateness. Instead, it aims at exemplifying a possible approach to standardise how we conceptually and empirically include properties of welfare states in cross-cultural comparative analyses of their micro-level outcomes. In doing so, it takes up various important fields and functions of the welfare state, but it does not claim to be exhaustive.

1.2 Remarks on the Analytical Perspective

This book argues within one very popular analytical framework in cross-cultural research, which is the so-called multilevel analysis. It is thus necessary to clearly distinguish this approach from others within the field of welfare state research. Hence, this part of the introduction elaborates in more detail the analytical perspective that is pursued.

The choice of a methodological approach depends on two key questions: (1) Which analytical levels do analyses incorporate? In other words: does the study aim to explain an outcome on the macro-level (usually countries or regions) or on the micro-level (usually individuals). (2) How many regional cases are analysed? Here, studies comparing few cases in in-depth case studies or a large enough number to explore statistical variance between regional units are differentiated.

Since this book focusses on research, assuming the welfare state has an impact on individuals, I will restrict the following discussion to literature incorporating at least two levels (individuals and countries or regions). Regarding the second question, those research endeavours are emphasised, which explore differences between welfare states in a greater number of countries (such as all member states of the European Union). This means that two kinds of studies are treated only marginally. First, those focussing on only one level (usually the macro level). Examples for this are analyses of the impact of social policies on other policies (e.g. Guo & Gilbert, 2007; den Dulk et al., 2012) or on aggregated micro-level information (e.g. Mano-Negrin, 2004; Schneider & Makszin, 2014). Second, case studies (e.g. Alves, 2015; Oesch, 2015) are of secondary importance as the aim is not to confront a few examples in great detail but to explain variance within a sample of countries in a regression framework.

Testing research questions addressing at least two analytical levels requires a multilevel conceptualisation. Such macro-micro-models often draw on what is sometimes described as a transformed version of Coleman’s (e.g. 1986) ‘boat’ (cf. also Lucas, 2016). Coleman himself distinguishes his approach from a limitation to the macro-level. His main objective is to formulate a theory, which incorporates micro- and macro-level phenomena and outlines transformations and processes between the two levels. Hence, he attempts to overcome limitations, which are especially prominent in macro–macro analyses following the logic of methodological holism. Coleman notes three shortcomings of such approaches: limited variation, limited insight about why a relationship exists, and the immanent premise that social orders exist⁷ (Coleman 1986: 1321–1322).

A graphical representation of his model is given in Figure 1.1. Instead of remaining on the macro-level, Coleman proposes to move down to the individual level, exploring how macro-level phenomena affect individuals on the one hand (type 2 relation) and how individual outcomes in turn shape macro-level processes (type 3 relation) on the other hand. In sum, such methodological individualism offers a framework in which relationships between the individual and the social context can be analysed in more detail than if one remains at only one level.

Figure 1.1

Coleman’s boat. (Figure based on Coleman (1986: 1322))

Even though Coleman was more interested in the opportunity to explore type 3 relations, this effect is usually not of key interest in the kind of multilevel approach addressed in this contribution. Instead, the dependent variable is usually situated on the micro-level. Thus, the last type 3 relation in the Coleman-inspired boat-model—the aggregation of individual outcomes—is excluded going forward. Moreover, what Coleman describes as type 2 relation (the macro-micro effect) is frequently included in hypotheses but rarely modelled empirically as this would often require the exploration of a temporal sequences. Instead, the direct macro–micro link between social context and individual outcome is explored as well as the moderating effect of the social context on type 1 relations. The most common conceptual set up of multilevel explanations for individual outcomes is represented in Figure 1.2.⁸

Figure 1.2

The multilevel model

One may add general problems in causal assumptions, which such multilevel conceptualisations avoid. This especially includes several types of fallacies. So-called ecological fallacies can arise if macro–macro correlations are used to interpret macro-micro or micro-micro phenomena assuming a given association adequately reflects similar processes on the micro-level. An example for such ecological fallacies (e.g. Robinson, 1950) could for instance mean that a correlation between insurance coverage and aggregated health is interpreted as evidence that insured individuals are healthier. The latter could be true—but it does not have to be. In addition, there are also fallacies, which may arise if the context is simply not considered. They occur, if effects, which are dependent on a social context, are interpreted without any reference to it (psychologistic fallacy). The opposite, analysing and interpreting effects on the contextual level alone without including relevant individual level processes, is termed sociologistic fallacy (an overview is given by Diez-Roux, 1998; Loney & Nagelkerke, 2014). Such fallacies are potentially avoided if the presence of two (or more) levels is acknowledged in methodological approaches. Explicitly modelling both levels and the relationship between them represents the most accurate and most commonly used methodological approach to capture the effect of contextual features on outcomes on the level of individuals in cross-cultural comparisons.⁹ For the research endeavour at hand it is therefore especially relevant because it allows to model how and why welfare policies—as a contextual influence—account for differences in observed outcomes between individuals in different countries. Only if variance between individuals and countries can be conceptualised and empirically tested simultaneously, are we able to determine if differences between individuals are actually due to specific features of welfare states.

1.3 Structure of the Book

This contribution addresses various relevant facets of the conceptualisation and operationalisation of welfare state policies as an explanatory factor and gradually approaches associated problems and possible solutions. After this introductory chapter, the second part (chapter 2) deals with fundamental issues, which constitute the groundwork for this contribution by addressing two smaller questions that are prerequisites for the discussion of the actual research questions: how is the welfare state approached as an object of research and why is it an important independent variable? This includes a general discussion of the evolution of comparative research on the welfare state and the debates that are relevant to this day. Such research usually treats the welfare state as a dependent variable. The chapter therefore also addresses general arguments for exploring social policy-making as an explanatory factor. This is pursued by discussing the functions performed by welfare states—either because they represent essential tasks (such as risk aversion) or because they are an unintentional side effect (such as conveying solidarity and justice principles).

The following part (chapter 3) explores the first main research question: how comparable are the results that emerge from different approaches to operationalising the welfare state as an independent variable? This raises various critical issues, which have already been brushed throughout this introductory chapter. Besides a more detailed discussion of these issues, this chapter also explores the implications for research findings, their comparability and transparency. As the conceptual as well as empirical confrontation in this chapter is going to show, there is a need for more standardisation and conceptual work, in order to more reliably include the welfare state as an independent variable in multilevel analyses.

Starting with the fourth chapter, such conceptual work on the welfare state as an explanatory factor commences as the second main research question is explored: how can we derive a more standardised, transparent and comparable approach to operationalising the welfare state as an independent variable? The discussion of this question is based on two steps. In a first step, popular objects of research, which are expected to be shaped by features of welfare states are reviewed. In each case, the focus rests on identifying how and why welfare states are assumed to have an impact on different individual-level outcomes between countries. By pinpointing such explanations in more detail, I deduce distinct conceptualisations of welfare stateness that can be found embedded in hypotheses. As argued in the second step (chapter 5), these distinct conceptualisations can be used not only to narrow down explanations but also to explicitly select empirical measurements. They thus present an intermediary step between theoretical discussions and empirical tests. In the sixth chapter of this contribution, the derived conceptualisations are put to the test in empirical analyses. This is done by applying them to several exemplary dependent variables.

A comprehensive discussion of main findings and open questions, which may spark critical debates and future research on the matter, concludes this contribution in the seventh and final chapter.