The Popular Response to the Ageing Crisis

A Time-Series Cross-Sectional Analysis of the Effects of Demographic Ageing on Individuals’ Support for Welfare State Policy in 13 Advanced Democracies (1996-2016)

Oskar Pettersson

Supervisor: Marcus Österman

Master’s Thesis in Political Science Department of Government Uppsala University Autumn 2020

Page intentionally left blank.

Abstract This thesis examines the relationship between demographic ageing, as captured by temporal, within-country variation in the ratio of elderly to the working-age population – the dependency ratio – on citizens’ support for the welfare state. The research problem is vitally relevant considering the worsening demographic structure of advanced democracies, a process that is having considerable ramifications on the possibilities of financing comprehensive welfare states. Using a time-series cross-sectional design, and building on representative survey data from 13 advanced democracies, the thesis specifically assesses the relationship between the dependency ratio, and individual spending preferences towards 1) the welfare state as a whole, as captured by an additive index, 2) education policy, and 3) old-age benefits. It also assesses whether demographic ageing exacerbates attitude differences between age groups, thereby scrutinising some assumptions made previously on the issue of intergenerational cleavages. The thesis uncovers no significant relationship between the dependency ratio and general support for the welfare state. However, the dependency ratio is shown be positively correlated with citizens’ support for education policy, while being instead potentially negatively correlated with support for old-age benefits. The differences between these two policies, in terms of their enjoyed support, are important considering the presumed shift in welfare state priorities towards what is commonly called social investment. Indeed, they indicate that there may be popular support for the type of reform strategies whose purpose is to invest in tomorrow’s diminishing workforce, whereas the support for more compensatory old-age policies may instead be weakening. There are also signs that the positive effect on the support for education policy is lower among older individuals. This evidence is quite interesting considering the previous expectations of deepening intergenerational cleavages as a consequence of demographic ageing, but the weak indications of this development in previous empirical research.

Keywords: demographic ageing, dependency ratio, welfare state attitudes, spending preferences, social investment, TSCS, pseudo-panel, cross-level interaction, intergenerational cleavages

Word count: 19 213

Acknowledgements I would like to thank Marcus Österman, researcher at the Department of Government, for kindly supervising me during the writing of this master’s thesis in political science. Marcus’ feedback and advice, especially with regards to methods and statistics, have been greatly helpful to me as a thesis student, and I have learned plenty of useful things along the way. While I have admittedly been slow to pick up on your suggestions in certain areas, Marcus, I believe that the end-product turned out quite well in the end. I would also like to thank my fellow student Klara Hvarfner for her kind read-throughs, and her valuable feedback on various parts of the paper. Lastly, I would like to compliment Anders Lindbom, professor at the Institute for Housing and Urban Research, for a number of insights regarding welfare state research that I have gained as a thesis student and as a research assistant – insights that have been quite useful in the process of developing this thesis’ theoretical framework, reflecting on different empirical indicators, and more.

Oskar Pettersson, 28th December 2020

Table of Content

1. Introduction ...... 1 1.1 Problem, Purpose and Research Question ...... 3 2. Background ...... 4 2.1. The Demographic Hangover ...... 4 2.2. Changing Risks, Changing Welfare States ...... 6 3. Theoretical Framework ...... 8 3.1. The Case for Analysing Individual Attitudes ...... 8 3.2. Previous Research ...... 9 3.3. Theory and Hypotheses ...... 12 4. Research Design ...... 18 4.1. A Time-Series Cross-Sectional Approach ...... 18 4.2. Data and Operationalisations ...... 21 4.5. Estimation Strategy ...... 31 4.6. Descriptive Statistics ...... 33 5. Results and Analysis ...... 34 5.1. Demographic Ageing and General Support for the Welfare State ...... 34 5.2. Demographic Ageing and Support for Specific Welfare Policies ...... 36 5.3. Exploratory Analyses and Robustness Tests ...... 42 6. Conclusions ...... 46 7. References ...... 50 Appendix A: Questionnaire Excerpts and Descriptive Graphs ...... 57 Appendix B: Results for Supplementary Variables ...... 65 Appendix C: Results using Continuous Age Variable ...... 71 Appendix D: Results within Income Sub-samples ...... 75 Appendix E: Results with Control for Linear Time-trend ...... 81 Appendix F: Results with Lagged Dependency Ratio (5 years) ...... 84

Appendix G: Stata Syntax ...... 87

Tables and Figures

Table 1: Pairwise correlations between dependent variables ...... 23

Table 2: Dependency ratio at survey years ...... 28 Table 3: Descriptive statistics for key variables ...... 33 Table 4: Effect of dependency ratio on general spending preferences ...... 35 Table 5: Effect of dependency ratio on spending preferences for education ...... 37

Table 6: Effect of dependency ratio on spending preferences for old-age benefits ...... 41

Figure 1: Spending preferences towards welfare state policies at survey years...... 24

Figure 2: Dependency ratio in sample countries, 1960-2016 ...... 27

Figure 3: Predicted spending preferences for education for different values of DR ...... 38

Figure 4: Marginal effects of dependency ratio on spending preferences for education ..... 39

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 1

1. Introduction

“Crisis we may call a situation with three characteristics: an existing state of affairs has become untenable, or is rapidly becoming so; a new more stable and tenable pattern or arrangement is not known or does not appear practicable […]; and the search for such a viable and practicable pattern, its establishment in practice and the decisions relating to this task all must be carried out under great pressure of time.”

(Karl Deutsch, 1981, p. 331)

It is certainly no understatement to say that it has been commonplace for welfare state scholars to argue that the welfare states of advanced democracies are, and will be, fundamentally in a state of hardship, distress, or even worse – crisis. Such accounts abound in the literature, and admittedly not without good reason. The favourable circumstances under which modern welfare states emerged – exceptional levels of economic growth, high employment, and rising wages – are indeed things mostly belonging to the past. Adding to these “exogenous forces” (Esping-Andersen, 1990, p. 6) are also the developments of increased globalisation and migration – the implications of which have been intensely debated. Regardless of the severity of their diagnosis, most would agree that the welfare state of the “golden age” has at least given way to something new. The answer to what this “new” actually represents, is arguably still elusive. Is it so, for instance, that the welfare state now comes in silver, as argued by Taylor-Gooby (2002)? Or, is it perhaps more accurate to say that it comes in a shade of grey? Indeed, there is another major development to add to the list: the rapidly on-going ageing of populations. With a few exceptions (see e.g., Castles, 2004; Gusmano & Okma, 2018; Scott, 2018), most would point to the process of demographic ageing – whereby the proportion of elderly to the working-age population increases, and the relative number of welfare state financiers accordingly decreases – as a key contributor to the on-going hardship of contemporary welfare states (Jaeger & Kvist, 2003). At the same time, it is somewhat difficult to claim that we are currently standing at a cliff-edge with regards to its effects. The process of demographic ageing is structural in the most accurate sense of the term, and one can arguably not state a point at which existing systems become unsustainable because of a worsening demographic structure. Still, there are those who have persevered at precisely this. In probably one of the gloomiest publications of recent years, Kotlikoff & Burns (2005; also 2012) see the welfare states of the United States and others, “going critical” (p. 134) within a few decades (at the time of writing this thesis, probably closer to one). Sinn & Uebelmesser (2002), likewise, writes of an “impending demographic crisis in Germany” (p. 153) that has

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 2 to be resolved before 2016 – after which the demographic structure will make impossible any serious reform attempts. Speaking more generally, Myles (2002), too, argues that “the cost of maintaining the status quo will escalate substantially” due to demographic ageing, and Meier & Werding (2010) sees its effects on the welfare state as “unambiguous—and all in all rather unfavourable” (p. 655). Accordingly, intense policy discussions have been had in various policy communities about how countries can face the grand challenge of demographic ageing, and against this background, it is hard not to get the impression that demographic ageing poses a structural crisis that deserves our immediate attention – although we may want to be careful about pinpointing exact years of no return. These types of matters tend to be more complex than that. Empirical inquiries into the macro-level consequences of demographic ageing are by no means new or lacking (Pampel, 1994; Razin et al., 2002; Castles, 2004; Disney, 2007; Tepe & Vanhuysse, 2009). But, surprisingly, there has been little scholarly interest into how exactly this structural process impacts on individual attitudes towards the welfare state. What do we citizens make of our current welfare state arrangements, given that our populations are rapidly growing older, and as the financial pressure on these arrangements mounts? Do citizens become more inclined to support them – to fill in the gaps that are left by a shrinking working-age population – or do they instead start to see them as having become essentially untenable? Based on existing schools of thought, we can make good cases for both. By the same token, we should also consider the possibility that people’s welfare preferences do not change in a uniform fashion, but in different directions depending on the specific policy in question. Indeed, the popular response to the worsening demographic structure, in terms of welfare state preferences, may very well be of a multidimensional nature. Beyond this, it is very relevant to ask whether citizens’ positions in the life cycle matters for what evaluations they make. During the time of writing, we are living through a pandemic with a severity not seen since the Spanish flu. Naturally, the Covid-19 pandemic has hit our elderly hard. Interestingly, though, as this pandemic has rolled on, societal discussions about whose interests should ultimately govern in these precarious times have become common. Should the young really have to adjust their living patterns as to protect the old and frail? And should the lives of healthy working-age people really have to be put on hold just for the sake of the older generations? Pointed questions like these, although quite specific to the present circumstances, arguably speak rather well to the overall notion of intergenerational cleavages – cleavages in which individuals mobilise politically on the basis of their generation, or more simply put, their age. Indeed, the idea of a deepening intergenerational conflict has garnered attention in the literature in relation to the fact that populations are ageing, and the fact that publicly financed welfare services tend to benefit the elderly more than other age groups (Kohli, 2006; 2015). The purpose of this thesis is to also make an empirical

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 3 contribution with regards to this issue – the importance of which can only be expected to increase. As for the pandemic, it is possible that whatever intergenerational conflicts existed prior to it will be intensified in its aftermath, given that we now seem to be facing a global recession that will have severe implications for ageing countries’ fiscal realities. While this cannot be said for sure at this point, this thesis may perhaps be able to provide some insight into what we can expect from these cleavages as we move forward. The issues that we have brought up here have been framed in terms of citizens’ attitudes and preferences. True enough, it is ultimately the attitudes and preferences of us citizens that constitute the governing force in society (Brooks & Manza, 2008) – and it is precisely these dimensions of the ageing crisis that will be central to this thesis.

1.1 Problem, Purpose and Research Question

The ambition of this master’s thesis is to provide a thorough and systematic assessment of the impact of demographic ageing on the popular legitimacy of contemporary welfare states. Demographic ageing stands at the centre of the many discussions about the future of our welfare states, and, as has been mentioned, plenty is known about its suggested effects on various macro-level indicators, such as spending (Tepe & Vanhuysse, 2009). What we know much less about, then, is exactly what the impact of this process is on individual attitudes towards the welfare state. This represents a research avenue within the welfare state attitudes literature that has been overlooked so far, and I concur with Naumann (2017) in that we simply “do not know how people react to increased reform pressures [such as] ” (p. 266). Hence, there lies something of an empirical puzzle buried within this rather unknown territory, and by helping to fill the empirical gap that yet exists regarding these questions, this thesis will make for a very timely and relevant contribution to the welfare state attitudes literature. Against this background, we can here formulate this thesis’ first, and overarching research question, namely how does demographic ageing affect individuals’ general support for the welfare state? Beyond this general question, we will also assess whether demographic ageing has divergent effects on individuals’ support for different areas of the welfare state. Do people make the same evaluation of policies like health care and old-age benefits, as they do of education, for instance? As will be explained, attitudinal divergences between these policies could have important empirical and theoretical value considering the transformation of welfare state priorities that many would argue has taken place in recent years – most notably towards social investment (Hemerijck, 2018). By assessing this, we add to the small but emerging literature on the relationship between structural, macro-level factors, and citizens’ attitudes towards social investment policies (e.g., Busemeyer & Garritzmann, 2019). Lastly, we will also assess whether the intergenerational cleavage, as described by Kohli (2006; 2015), is actually a salient

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 4 feature of contemporary welfare attitudes towards the welfare state. Specifically, we will assess whether individual age moderates the impact of demographic ageing on individuals’ support for the welfare state. This last question is particularly interesting considering the controversy in which the notion of potent intergenerational conflicts has been couched as being either far too naïve, or even “alarmist” (Lindh et al. 2005, p. 472; cf. Gusmano & Okma, 2018), or as something that will most likely be a defining feature of future welfare state politics (Kotlikoff & Burns, 2005). By performing this study, we heed the call about the need for more systematic inquiries into the “structure of welfare state attitudes” (Kohli, 2006, p. 475) in relation to demographic ageing. And as we will see, the design behind the coming analyses will allow this thesis to make a substantial contribution to the literature by being, to the author’s knowledge, the first temporally designed comparative study made on this important topic.

***

This thesis proceeds as follows. In section 2, we acquaint ourselves more with the matter of demographic ageing, as well as the changing nature of contemporary welfare states. This forms a valuable background to the theoretical framework, which includes our hypotheses, that is presented in section 3. In section 4, we review the choices made regarding the thesis’ research design. The overall approach is explained, the relevant data sources are described, key concepts are operationalised, and key strengths and weaknesses of the design are discussed. In section 5, the main empirical analyses of the thesis are presented, and discussed in relation to the hypotheses. In section 6, lastly, we summarise the accomplishments of this thesis, and discuss the options for future research in relation to the present topic.

2. Background

This section provides a background regarding the process that is central to this thesis, i.e., demographic ageing. We discuss essentially how we ended up in the present situation, and whether there are any readily available solutions. We also map out the changing role of society itself, and how this relates to changes within contemporary welfare states. This background will be valuable in relation to the theoretical issues that we will raise in section 3.

2.1. The Demographic Hangover

The most immediate cause for our current situation can essentially be said to be the industrialisation of the Western world. With industrialisation came prosperity, and while this prosperity was by no means spread equally, living standards of populations quickly increased

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 5

– a process that was accelerated by the expansions of modern welfare states. A key dimension of this development was, as put by Myles (2002), the “democratization of retirement” (p. 113), whereby governments assumed a greater responsibility to care for the old – who could no longer sustain themselves through labour. True enough, life expectancies clearly increased, at the same time as infant mortality decreased, and families became less dependent on having multiple children. Fertility rates dropped – causing some to fear , writing as early as in the 1930’s (Myrdal & Myrdal, 1934). A key consequence was that costs related to an increasing number of elderlies could be offset by decreasing costs for the young. The demographic structure could thus remain quite favourable, giving rise to a situation that has been referred to as a demographic gift (Lindh et al., 2005). Such a structure does not last forever, however. Indeed, the demographic structure in today’s advanced democracies can rather be described as constituting a demographic hangover (Lindh et al., 2005). This development is most straightforwardly illustrated by the change in the relative sizes of populations’ age segments. And so, if we examine the old-age dependency ratio (i.e., the proportion of individuals aged 65 or older to those between 15 and 64) as measured across OECD countries, this ratio increased from 13,9 percent in 1950 to 27,9 percent in 2015. According to projections, it will reach 53,4 percent by 2050 (OECD, 2019). These figures tell the tale of how the large, past generations have begun to retire, but where there are no similarly large cohorts available to fill the gaps that are being left. The result is a significant deficit in the size of working-age populations relative to the elderly populations, and this is something that will have considerable consequences for economic growth, and in turn, the opportunities to finance comprehensive welfare states. Given that we view this development as problematic1, what can be done to alleviate it? The two most proximate causes are high life expectancy, and low fertility. Since we do not want people to live shorter lives, the natural solution would be to increase fertility rates – something that has been pursued to somewhat different degrees in different countries. Apart from these, another frequently discussed solution has been increased migration. True, migration represents a way of increasing the tax-base with which the welfare state can be financed (Freeman, 1986; Lindh et al. 2005). Those who migrate are usually young, and so the dependency ratio may decrease via immigration. Currently, some evidence suggests a positive relationship between demographic ageing, and labour migration openness (Lutz, 2020), and there is also preliminary evidence of it being associated with a more positive immigration outlook within European populations (Irmen et al., 2019). In practice, however, the migration solution hinges on, for instance, that migrants are able to properly enter the domestic labour market. In some countries, this has proven to be easier said than done.

1 To be sure, some would today critique this notion from an ecological perspective.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 6

Other solutions are aimed at increasing the productivity of the existing workforce. Easily enough, the best solution might be for the existing working-age population to work more years, and retire later. The OECD, for instance, sees this as “crucial to maintaining pension adequacy and financial stability” (2019, p. 16), but also acknowledges the wide-spread unpopularity of such a proposal – an unpopularity that has been well illustrated in the literature (Hofäcker, 2015). To conclude, there are some potential remedies to demographic ageing. But it is arguably quite uncertain whether they can be implemented in a sufficient scale (cf. Taylor-Gooby et al., 2017), and as concluded by Kohli (2006, p. 466), “the demographic numbers are such that the issues will remain critical”. True, the incentives for incumbent policy makers to implement potentially useful solutions may be low, and long-term issues like demographic ageing tend to receive little attention in the public debate, compared to more pressing day- to-day issues. One could therefore argue that demographic ageing has all the components of a problem that there is little incentive to forcefully start dealing with within an election cycle. Hence, this process will probably become an even more pervasive feature as we go forward, and we might very well end up with a situation where it is too late to implement any of the “good” solutions that are currently being proposed. This is a key motivation behind assessing the impact of demographic ageing on individual attitudes towards the welfare state, since it may enlighten us regarding whether the types of reforms that are instead aimed at the welfare states themselvescould possibly be accepted by citizens, or whether the public veto will effectively block any attempts at restructuring existing welfare states .

2.2. Changing Risks, Changing Welfare States

The development towards a worsening demographic structure coincides, and interestingly so, with what we may call a fundamental transformation in the structure of life-course risks. On this topic, Esping-Andersen (1996) refers to how the society which the welfare state was built around has changed in some important ways. Notably, time has seen the decline of the male-breadwinner model, where men were chiefly responsible for labouring within the free market economy, and where social policy was designed mainly with the purpose of insuring male workers against income loss. The subsequent entry of women into the labour force, the rise of single-parent households, increasing professionalisation, increasing individualisation, and thus a life cycle less resembling that of old industrial society, illustrate how the welfare state has become less compatible with its surrounding environment. Taylor-Gooby (2004) cordially sums up this transformation in terms of old social risks giving way to new ones, following our transition into “post-industrial society” (p. 3).

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 7

Crucially, and partly as a response to this transformation of life-course risks, there has also been a turn within welfare state politics towards what is most commonly called social investment (Midgley, 1999; Esping-Andersen et al., 2002; Morel et al., 2011). Some go so far as to call it a paradigm shift (Hemerijck, 2018). What is essentially argued is that welfare states have transformed themselves, or should transform themselves (Hemerijck, 2013; Lindh, 2011), from being geared mainly towards compensating individuals for the vagaries of the free market economy – as in the old industrial economy – to being geared more towards ensuring that individuals are themselves equipped to deal with these vagaries. In practice, this means providing individuals with the education necessary to compete in the labour market, as well as taking active measures to put individuals back into work once they do become unemployed – as embodied by so-called active labour market policies (Bonoli, 2010). Providing the (young) individual with the knowledge and the skills that are à jour with the more specialised and knowledge-intensive labour market of today, makes the individual more capable – the argument goes – of staying within the labour market in the long run, and being therefore less likely to require compensation from the welfare state. Expressed in more theoretical terms, this suggests a shift from a “protectivist” welfare state to a “productivist” welfare state – or, indeed, from decommodification (Esping- Andersen, 1990) to recommodification (de la Porte & Jacobsson, 2011). And as a response to those who would see an inherent trade-off between the welfare state and economic development, a welfare state that could prove to have some serious productive potential may therefore dissolve some of the tension between the twin goals of social sustainability (in lack of a better term), and economic development. Indeed, the case for the welfare state might be strengthened if it can be proven to have the ability to yield positive returns for the economy (Midgley, 1999). The crucial point here is that, given finite public resources, and given the magnitude of current compensatory welfare state commitments, an increased social investment-focus must inevitably mean some sort of re-allocation of resources from mature and entrenched policy areas. True enough, an increased focus on the young – which an increased focus on education naturally implies – may in practice mean a reduced focus on the old, which is a group that has traditionally been one of the most important constituencies of welfare states, and perhaps the most obvious, and normatively deserving recipients of compensatory welfare policies. This, of course, leads us to wonder whether this development is truly in line with citizens’ attitudes – given the rather considerable development towards ageing populations.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 8

3. Theoretical Framework

This section presents and discusses the building blocks that together form the theoretical foundation of this paper. First, we discuss the theoretical importance of analysing individual welfare state attitudes. Second, we review the previous inquiries into the current topic, with the purpose of providing a general entry point to the relevant theoretical questions. Third and last, we present this paper’s theoretical argument – which is then also distilled into delimited hypotheses that can be tested empirically.

3.1. The Case for Analysing Individual Attitudes

The purpose of this thesis is to assess the effects of demographic ageing on individuals’ support for the welfare state, and this means assessing individual attitudes. As political scientists, we assume that the public plays a vital role in shaping the future of our contemporary democracies, and that governments are in fact responsive to the will of the public. We would consequently say that the empirical relationships that we find on the level of individual attitudes have causal bearing on the macro-level developments that we usually also want to speak to – an argument in line with that of Brooks & Manza (2008, p. 6) on the independent role of public opinion in shaping welfare states:

“Ignoring the policy preferences of national publics deprives scholars of a fuller understanding of the political foundations and trajectories of contemporary welfare states.”

While Brooks & Manza (2006a; 2006b; 2008) do not primarily focus on individual attitudes as the dependent variable, their argument – as well as that of other scholars within the policy responsiveness-field within political science (e.g., Page & Shapiro 1983; Stimson et al., 1995; Soroka & Wlezien, 2010) – is still relevant, and it serves to emphasise just why studies like this one are relevant to future-oriented discussions about welfare state politics. And going beyond the causal aspect, assessing individual attitudes is still crucial in order to assess the sheer political legitimacy of current welfare state policies (Chung et al., 2018). Analysing attitudes is also relevant considering the difficulty of measuring macro-level changes in welfare states. There are many contributions where welfare state size, for instance, is operationalised in terms of social expenditures (Wilensky, 1975), and it is by analysing expenditures that Pierson (1994; 1996; 2001; see also Alber, 1988), for instance, concludes that welfare states have remained largely intact, despite a period of strong retrenchment pressures. Other authors, however, have criticised this, arguing for example that expenditures may change (or rather not change) for reasons other than policy changes (for an overview,

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 9 see Green-Pedersen, 2004). It is also criticised on more conceptual grounds. Esping- Andersen (1990, p. 37), following authors like Titmuss (1958), argues that “[e]xpenditures are epiphenomenal to the theoretical substance of welfare states”, and it is along these lines that Korpi & Palme (2003) use e.g., benefit eligibility data to illustrate a rather marked welfare state retrenchment. Similarly, Allan & Scruggs’ (e.g., 2004), analysis of benefit replacement rates also shows signs of welfare state retrenchment. This speaks fundamentally to what has been termed the dependent variable-problem within the welfare states literature (Green- Pedersen 2004; Clasen & Siegel, 2007), according to which there is no clear consensus on how one best measures welfare state change. Because of these conceptual difficulties on the macro level, we can be even more certain that analysing attitudes represents a very important avenue of research. And while we recognise the now considerable policy feedback-literature, which stresses the ability of welfare states to influence individual attitudes – thereby reversing the causal arrow between attitudes and welfare states (e.g., Larsen, 2008; Jordan, 2013) – we may still expect individual attitudes to be able to give us important clues with regards to the shape of future macro-level developments in welfare states.

3.2. Previous Research

Much of the existing work on the role of ageing in relation to individual support for the welfare state has built upon the assumption that demographic ageing may exacerbate already existing intergenerational cleavages within populations. Many authors have thus followed – implicitly or explicitly – in the footsteps of the gloomy accounts about the societal effects of demographic ageing (e.g., Kotlikoff & Burns, 2005), and thus forcefully stressing the power of self-interest based on individual age.

The role of individual age Accordingly, the main focus has seemingly not been on the explanatory role of demographic ageing as a contextual factor, but rather that of individual age. The idea has been roughly that, since populations are growing older, evidence of divergent attitudes within different age-segments within populations should indicate an emerging or intensifying intergenerational conflict (Street & Cossman, 2006; Svallfors 2008; Busemeyer et al. 2009; Sorensen 2012). Arguably, then, these contributions therefore only analyse one part of the relationship, since they essentially take some very important macro-micro relationships for granted. Reviewing this literature is useful, though, since it provides a good background for this thesis’ research objectives. The empirical strategy of these papers has often been to study the individual support for welfare policies geared either towards the elderly, or the young, and this strategy is one that we too will pursue later on.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 10

One often-quoted example is Busemeyer et al. (2009), who assess the relationship between individual age, and spending preferences towards health care, unemployment benefits, education, and , using cross-sectional data. The authors find indications of age being negatively related to spending preferences for education, as well as unemployment benefits. Results for health care and pensions are more ambiguous. It is thus not clear that age is correlated with a higher support for pension spending. These results, however, are supported as well as supplemented by Sørensen (2012), who, by using the same data source, shows age to be negatively correlated with spending preferences towards education, and positively correlated with spending preferences towards health care and pensions – thus illustrating the two-edged influence of individual age on support for different policies. In a study of Swedish welfare state attitudes, moreover, Svallfors (2008) shows how younger individuals are more supportive of family-centred policies than elderly-centred policies. Fittingly, the exact reverse is true for older individuals. These results clearly echo those previously mentioned, and they come from a country with relatively little in terms of elderly bias within its welfare state (Birnbaum et al. 2017), and this can be said to provide a least-likely case for salient age differences regarding redistributive issues – since the young should be more appeased with the welfare state, compared to the young in a country with a stronger old-age bias. Weighed together, these studies illustrate that there might well exist an intergenerational cleavage within welfare state opinion (see also de Mello et al., 2016, unpublished). Interestingly, though, contrasting results can be observed in Street & Cossman’s (2006) analysis of spending preferences towards education, health care, and old-age benefits, among Americans. Their analysis shows how young individuals support spending on education more than on old-age benefits, but this is while the elderly expresses nearly the exact same preferences. These results are particularly interesting since it is usually within the US that predictions about an intergenerational conflict has been the strongest (Kotlikoff & Burns, 2005) – partly because of the strong old-age bias within the American welfare state – and so these results add some very critical ambiguity to this literature regarding the role of individual age.

The role of demographic ageing Currently, only a few papers have examined the relationship in full, i.e., that between country- level demographic ageing, individual age, and individual support for the welfare policy. Adding to this, we may note that they provide conflicting evidence. Emery (2012), using ESS data from 2008, shows first how demographic ageing (as captured by the dependency ratio) is negatively correlated with individual support for the elderly. This is taken to be indicative of a “support fatigue” (p. 20) within European populations with regards to old-age policies.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 11

Second, individual age is shown to be positively correlated with support for the elderly – further supporting the notion that one’s social policy preferences change with age. Third, and with bearing on our own predictions, Emery tests for a cross-level interaction between demographic ageing and individual age. There is no significant evidence of such an interaction, however, which is to say that demographic ageing cannot be shown to exacerbate age differences in terms of support for the elderly. These results provide a good first look into these important relationships, although it would have been useful had Emery been able to provide comparisons with youth-oriented policies (such as those related to education). Adding a small but to Emery’s findings, however, Svallfors et al. (2012) illustrate that, when using the same attitudes data (but a differently specified dependency ratio), the dependency ratio is associated with smaller age differences (as indicated by so-called K-values) in attitudes towards a large set of policies, including old-age benefits and education. Contrary to their own expectations, “ageing populations are not followed by increasing age conflicts around the welfare state” (p. 177). We may note that Svallfors et al. (2012) do not study the support for specific policies per se – but their results make for a bit more uncertainty with regards to what we should expect in terms of the dynamic between demographic ageing and individual age. In a more recent paper, Hess et al. (2017) use Eurobarometer data from 2009 to, as they helpfully point out, “scrutinize the claim that the strength of the [intergenerational] conflict will increase as [the] population ages” (p. 12). This really illuminates the issue with the first round of contributions that only focused on the role of individual age. Accordingly, the authors assess how demographic ageing and individual age is related to individual spending preferences with regards to supporting the elderly, as opposed to education for the young. Here, then, we do get to compare results on the opposite ends of the policy spectrum. In line with previous studies, individual age is shown to be positively correlated with support for the elderly, while negatively correlated with support for education – no news there. And echoing Emery (2012), the relationship between the dependency ratio, and individuals’ support, cannot be shown to be moderated by individual age. Thus, the combined indications from the previous cross-country analyses are that intergenerational cleavages are not stronger in older societies. It also worth mentioning the survey experiment by Naumann (2017), fielded in Germany, which assesses the causal effect of information about demographic ageing, on attitudes towards working-life and pensions. This experiment shows how respondents’ opposition to increasing the retirement age is reduced, whereas their opposition to cutting back pensions is increased. Unfortunately, this study does not assess the interaction between treatments and individual age, but the results do at least suggest that the preferred solution among individuals is not to cut back on existing benefits, but working longer in order to maintain them. This

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 12 study has a clear advantage because of its experimental design, which obviates many of the risks related to omitted variables bias that usually accompany attitude analyses, and the results can be seen as refuting those of Emery (2012). Still, they are only immediately generalisable to the German population, and so it is difficult to draw any far-reaching conclusions based on this experiment alone. To summarise, there has certainly been some work done on the effects of individual age, but few have specifically modelled the effect of demographic ageing as a contextual factor, and how it interacts with individual age. Indeed, it has often been assumed that demographic ageing will exacerbate existing but perhaps yet latent intergenerational cleavages (Svallfors et al., 2012, p. 164). The studies on individual age suggest that age does have a potential to shape attitudes towards welfare policies in line with an intergenerational cleavage, although some studies (e.g., Street & Cossman, 2006) add some uncertainty to this conclusion. The small literature that does include demographic ageing as a country-level variable, though, provides a mixed bag. On one hand, demographic ageing has been shown to reduce support for the elderly (Emery, 2012), but on the other, it has been shown to increase opposition to reduced old-age benefits (Naumann, 2017). Additionally, there is currently little to suggest that demographic ageing exacerbates intergenerational cleavages. But, and with significant bearing on the study undertaken in this paper, the authors that have assessed the role of demographic ageing with cross-national data, have not done so with data that allow for analyses of within-country variation over time. We will discuss this more later, but we should acknowledge that these analyses may contain some country-level endogeneity that they cannot account for. Indeed, as far as non-experimental cross-country analyses go, this potential endogeneity can only be remedied by including a temporal component. We should thus not neglect the possibility of a salient cross-level dynamic between demographic ageing and individual age based on the previous literature – there may yet be interesting empirical insights to be made in this regard.

3.3. Theory and Hypotheses

It is appropriate to begin this section by properly establishing whether, and how demographic ageing should matter for individual-level attitudes towards the welfare state. For starters, considering the incremental nature of the process in question, there is a risk that people do not really take notice of the transformation that has occurred within the demographic structure. We know that citizens tend not to be as politically knowledgeable as typically assumed in classic democratic theory (e.g., Dahl, 1989), but we do know that factors such as a country’s past economic performance, for instance, can very well affect political attitudes (Kramer, 1971). There should therefore be sufficient reason to expect demographic ageing to have the potential to be a salient issue for people, and to be able to influence welfare

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 13 attitudes. Having emphasised this, we should highlight the importance of viewing demographic ageing as fundamentally implying a pressure to reform existing welfare institutions (cf. Naumann, 2017; Goerres et al., 2020). If fewer are to support the many, we either need to do more in terms of welfare, or we need to do less – and it is this crucial question that citizens have to grapple with against the background of demographic ageing. As asked by Naumann (2017, p. 266): “Do people ask for a stronger welfare state to be protected against hard times, or do they accept retrenchment when faced with budgetary constraints? In terms of general support for the welfare state – which is the first issue that we will tackle – this reform pressure is then something that we can see as taking two possible main routes – each of which we can put forward quite strong cases for. In the first scenario, demographic ageing leads individuals to decrease their general support for the welfare state. This would partly be based on the idea that the system as such must be made sustainable given the growing number of individuals who will be using some parts of it, and the dwindling number of working-age people who will be financing it. In other words, individuals may see that the future tax-burden on the few who do work will be too high, at the same time as the welfare state in its totality may become hollowed-out. Becoming hollowed-out, the welfare state may therefore not be sufficiently deserving of their support. Indeed – viewed from a rational-choice standpoint – why support something that one would get very little out of? A theory that may be valuable to draw on here is the Paradox of redistribution, as laid out in a seminal piece by Korpi & Palme (1998), which stipulates that to maintain a wide pro- welfare coalition among citizens – one that includes the valuable middle classes – the welfare state must be generous and inclusive enough to keep high-income earners content with supporting it (cf. Esping-Andersen 1990, p. 48). The middle classes do generally not favour an extensive welfare state (Gelissen, 2000), and so they must be appeased with generous benefits or insurances. This theory is not perfectly transferable to this context, since it is more about fostering cross-class support for welfare policy via the design of welfare institutions. But in this case, we could probably expect demographic ageing to lead to a welfare state with a de facto stronger old-age profile, which could in turn lead to a hollowing- out of areas geared towards younger and middle-aged people. This expectation can be summed up in terms of the median voter theorem, as developed by economists like Black (1948), Downs (1957), and Meltzer & Richard (1981), and is concisely summarised by Taylor-Gooby et al. (2017, p. 205), in that the “expanding needs of older people as populations age, backed up by the political clout of this group, result in a tendency to favour spending on older generations against younger ones”. Pension spending, for instance, may increase as a consequence of actual policy changes (e.g., increased benefit generosity), or automatically, following the increased demand that follows from higher numbers of elderlies (Galasso &

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 14

Profeta, 2007; Tepe & Vanhuysse, 2009). Given the costliness of areas such as pensions, the overall price tag on the welfare state can also be expected to increase. In the future, then, we could thus in fact be looking at a twin-development towards a more expensive, but simultaneously more hollowed-out welfare state – at least with regards to areas that are not geared specifically towards the elderly. And this is something that could lead the average individual to become less incentivised to support it – and younger ones in particular – since some areas of the welfare state can be expected to become less generous, and, crucially, since the relative number of financiers will diminish. We can thus put forward arguments both from public interest (welfare state sustainability) and self-interest (higher taxes combined with a hollowed-out welfare state), and if we tie this back to the Paradox of redistribution, we might in other words be inclined to fear an erosion of cross-generation, or cross-age support for the welfare state. In sum, this would have us expecting demographic ageing to lead to a general withdrawal from the welfare state. Contrasting the above scenario, however, demographic ageing might instead lead to an increased support for the welfare state. Here, individuals increase their support with the idea of ensuring that existing benefits – no matter whether it is old-age benefits, health care, or education – should be preserved, despite the decreasing number of effective taxpayers. This expectation builds on the idea that individuals are inherently loss-averse regarding the benefits that they enjoy (or expect to enjoy, as in the case of old-age benefits), wanting little else than to keep things as they are. Embedded in this argument are assumptions based in new institutionalist theory, which stipulate that welfare states, in its granting of various rights on the basis of social citizenship (Marshall, 1950), tend to create their own constituencies, which will react negatively to attempts to diminish said rights, despite retrenchment- encouraging fiscal realities (Pierson, 1994). We could also put this in terms of mature welfare states having gradually imposed their own “logic of appropriateness” (March & Olsen, 2004), making the withdrawal of social rights to some extent unthinkable – perhaps especially in times of increasing hardship. In this scenario, then, we expect little in terms of individuals withdrawing from the welfare state. In fact, individuals might instead be expected to increase their support for it, since demographic ageing may simply instil the sense that society must do more in terms of welfare, and not less. This last argument could perhaps be expected to follow mostly from altruistic values and perceptions, but one should emphasise that this entire scenario can be motivated from self-interest as well, just as the withdrawal scenario. What we have arrived at here, then, is essentially the core of the research problem: given that demographic ageing increases, and given that people start perceiving the current welfare state as becoming more and more untenable, where do they go in terms of support? Considering the few results on this, no expectations are given. Emery (2012) saw indications of so-called “support fatigue” (p. 20), while Naumann (2017) found rather the opposite –

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 15 but these studies only assessed support for old-age benefits, and our scope here in terms of general support for the welfare state is wider than that. In formulating our first hypothesis, we can perhaps draw some insights from other research fields, though. Economic crises, such as the 2008 stock-market crash (which also come with serious pressures to reform existing systems), have been shown to increase support for economic redistribution within Europe (Olivera, 2014), while decreasing it in the United States (Fisman et al., 2015). The available cross-national evidence is arguably more relevant for us, although redistribution preferences may not be perfectly analogous to welfare state support (see also section 4.3). Simultaneously, other authors have found short-term economic crises to be associated with a higher electoral support for the right (Lindgren & Vernby, 2016). This clearly contrasts the above results, of course, but we may simultaneously note that long-term crises have historically been shown to garner support for the left (Lindvall, 2014; 2017), and these long-term crises are arguably more similar to demographic ageing. In sum, then, existing findings may encourage us to put our expectations regarding the effects of demographic ageing more firmly within the second scenario, and we can formalise this expectation in hypothesis 1.

Hypothesis 1: Demographic ageing leads to an increase in general support for the welfare state.

This discussion has so far been couched in terms of general support for the welfare state, which is a frame that we do well to consider. But we should also be interested in analysing support for the welfare state’s constituent areas, and importantly, this will tie the development towards a social investment welfare state into our framework. Crucially, it is not difficult to see how the two goals of (a) sustained or enhanced economic development, which is thought to follow social investment, and (b) a welfare state that can support ageing populations, are intrinsically interlinked. As put by Esping-Andersen & Sarasa (2002), “[t]he well-being of tomorrow’s elderly will depend very much on the welfare of tomorrow’s labour force” (p. 6; see Lindh, 2011 for a similar normative argument). Clearly, a skilled and productive workforce is a prerequisite for a country to be able to generate the surplus needed to finance a comprehensive welfare state in a time of demographic ageing. However, this does not preclude that these two goals have the potential to spur distributive conflict. As we have explained, demographic ageing may bring forth a welfare politics that favours mainly the elderly, while crowding out the young. As one may have guessed, it is against this background that authors have feared deepening redistributive conflicts regarding different welfare policies. And somewhat gloomily, then, it is especially regarding education policy on the one hand, and old-age policies on the other – which are meant to be interlinked in a social investment strategy (Esping-Andersen & Sarasa, 2002) – that authors have expected attitudes to diverge the most. Hence, there arises quite an

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 16 interesting question about whether social investment, as it were, actually enjoys any sort of public legitimacy in the face of demographic ageing, or whether demographic ageing does in fact cause elderly-oriented policies to crowd out social investment-oriented policies such as education, even on the attitudinal level – thereby strengthening the more traditional, or compensatory view of the welfare state. This would very much be in line with established theories about the high degree of welfare deservingness of the elderly (van Oorschot 2000; 2006), which illustrate the strong intergenerational ties that bind us together up and over generations. The idea about “doing more, not less” might thus be particularly pronounced in relation to elderly-oriented policies. There is also the matter of there being a clear contractual nature to old-age benefits, in that (nearly) everyone becomes old at some point. There may thus be strong incentives for people in general to ensure that the intergenerational contract (Lindh et al., 2005) is preserved, so as to be able to enjoy sufficiently favourable benefits when their time comes. On the other hand, one could also imagine that people instead perceive demographic ageing as signalling that continuing to support the elderly to the same extent is especially unsustainable. And perhaps it is instead the young that should be invested in, in order for the system to be sustainable in the long run? Busemeyer & Garritzmann (2019) recently showed how economic globalisation, as measured in terms of countries’ trade openness, is positively related to individuals’ spending preferences towards education, while not being so for unemployment benefits. The authors argue that individuals may simply perceive “investments in education [as] more effective in dealing with the challenges of globalisation in the long term compared to social transfers” (p. 444). In this paper, we are of course not studying the effects of globalisation, but, like demographic ageing, globalisation represents a similar structural factor that many believe has significant effects on welfare state politics (e.g., Burgoon 2001; Swank 2002; Walter, 2010). Moreover, Garritzmann et al. (2018) also use recent data to show how social investment policies like education actually stand out as being very popular within populations, as compared to more compensatory social policies. So, these results can perhaps also be instructive in our context, in that demographic ageing could instil a strong sense that we need to invest in the diminishing cohorts that will serve the future economy – and not least via education. So, while it would perhaps be more intuitive to expect elderly-oriented policies to enjoy higher support following demographic ageing, youth-oriented social investment policies might turn out to be quite popular. Additionally, of course, we have also noted that demographic ageing has been associated with lower support for the elderly in previous comparative studies (Emery, 2012). Apart from these two scenarios, there is also the possibility that there are no noteworthy differences between how individuals evaluate these policies against the background of demographic ageing. The response might simply be one

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 17 and the same: more or less. In this case, however, we will entertain the possibility that demographic ageing does in fact increase individual support for social investment, as captured by increased support for education policy, while instead doing the opposite for compensation – as captured by support for old-age benefits.

Hypothesis 2: Demographic ageing leads to an increased support for education, and a decreased support for old-age benefits.

So far, this discussion has been framed in terms of net effects across populations. True, it is not impossible that most individuals respond to demographic ageing with one and the same answer (more or less old-age benefits, for example). We need not necessarily be the self-interested utility-maximisers that we are commonly made out to be in various attitudinal analyses. The notion of citizens caring about what is in the public interest – or indeed the common good – has gained prominence within the political science literature thanks to for instance Lewin (1991), and we may also refer to contributions like that of Gonthier (2017; cf. Page & Shapiro, 1992), which stress how “social classes and income groups tend to move in tandem with regard to their economic evaluations or to their support for redistribution, public spending and state intervention (p. 94). Still – we should consider that the state of affairs on the macro level may interact with characteristics on the individual level, to create more complex relationships. Naturally, and following in the footsteps of the few previous contributions (Emery, 2012; Hess et al. 2017), we will consider an individual’s own age as a key factor. Theoretically, we may draw on the argument put forward by Kohli (2006, p. 456, see also 2015), a frequent writer on these issues, that the “old” class cleavage regarding the issue of public resource distribution has to some extent given way to an equivalent intergenerational cleavage. Previously, Kohli argues, the intergenerational cleavage would be ameliorated by things such as private resource flows between generations (cf. Lindh et al., 2005), but now, the weakening of the demographic structure might dramatically be changing the preconditions for these types of ameliorating factors. Indeed, and as put by Svallfors et al. (2012, p. 164), “as the groups of elderly grow in relation to the working population […] tensions between age groups would grow in relation to welfare policy priorities”. The empirical question, then, is whether the effect of demographic ageing on support for education, say, is in fact more positive among the young than the old, and vice versa for old-age benefits. In contrast to the previous contributions that have focused solely on individual age (e.g., Busemeyer et al., 2009), this would actually highlight the strength of the intergenerational cleavage, where one’s own age governs how one’s welfare state attitudes are affected by demographic ageing. While the true complexity of these relationships between generations has been well-highlighted in recent qualitative

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 18 work (Prinzen, 2014; 2017), in our third and last hypothesis we will posit that the spending preferences of young individuals are more positively affected than those of the old, for areas that favour them (education). Conversely, we expect that the spending preferences of older individuals are more positively affected than those of the young, regarding areas that favour them (old-age benefits). This would illustrate that demographic ageing causes different age- segments to double down in terms of support for “their” policies. This would then in extension be suggestive of an exacerbated intergenerational cleavage – and by the same token, a weakened intergenerational support coalition – with regards to how public resources should be allocated within the realm of the welfare state.

Hypothesis 3: Demographic ageing affects the support for education more positively among the young than the old, while affecting the support for old-age benefits more positively among the old than the young.

While this third hypothesis will be aimed first and foremostly at the specific policy areas of education and old-age benefits, we implicitly also expect that the effect of demographic ageing on general spending preferences will be more positive among older individuals, compared to young individuals. This would follow the expectation that a general withdrawal from the welfare state is more probable among those who will be the main financiers of all welfare state arrangements, whereas the welfare state as such may be perceived as far more popular among older individuals.

4. Research Design

Having presented our theoretical framework and our hypotheses, this fourth section presents the paper’s overall research design – including the research approach, the sources of data, operationalisations of key variables, and, lastly, the estimation strategy that will be pursued. As we know, every decent paper is open and transparent about its most crucial strengths and weaknesses. There is always more than one credible way of approaching a research problem, and it is vital that one discusses thoroughly the reasons for the choices one makes. This, in turn, makes for a clear and transparent disposition.

4.1. A Time-Series Cross-Sectional Approach

This study uses a time-series cross-sectional (TSCS) design to test its hypotheses. Within this quantitative-comparative approach, we make comparisons across a larger number of observations to distinguish statistical associations between variables (Pennings, 2016). “TSCS”, in turn, refers to a design where cross-sectional data from multiple points in time

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 19 have been pooled together, thus adding a temporal dimension to the data structure.2 This approach is valuable in that it ameliorates one often-encountered problem when analysing the effects of country-contextual factors, namely that the number of countries is naturally limited – which may cause problems for statistical analysis, and especially when using large sets of covariates (Stegmueller, 2013). By pooling multiple cross-sections together, we effectively extend the number of cases on the country level, and thus achieve a sample that is more suitable for statistical analysis. As hinted at earlier, moreover, adding a temporal dimension to our data also allows us to account for things that may vary between countries – and that may thus introduce country- contextual bias into our analyses – but that are more or less time-invariant. Examples of these factors include rather static political institutions, and so forth. Analysing welfare state attitudes, we could think of welfare state institutions as an especially very important factor to want to account for. But controlling for things like political legacies, historical cumulative left-wing incumbency, or various cultural factors, might also be desirable. In practice, we control for these kinds of factors by incorporating country fixed effects into our models, and this marks one important step towards being able to assess the within-country variation of a contextual variable as it develops over time, and analyse how it is associated with individual- level variables. Indeed, using a within-country estimator (instead of a between-country estimator) in a fixed-effects model eliminates important sources of endogeneity (Giesselmann & Schmidt-Catran, 2019). Additionally, we may also include controls for any time-related factors that may affect our units of analysis equally (i.e., they are context- invariant) at some point in time. This would be to ensure data-comparability over time. Common examples of factors that researchers account for via these time fixed effects include time-specific crises such as that of the 2008 global financial crisis, but in our case, we in effect control for any factors that are related to the years at which the surveys were fielded, and that affect all of our observations equally.

Using TSCS for Causal Inference The purpose of this thesis, then, is to assess the causal impact of the country-contextual factor of demographic ageing, on individual attitudes, and to then make empirical generalisations across adjacent cases in space and time. Hence, it is worthwhile discussing exactly what causal inference is all about, and where our TSCS design stands on this matter. For this purpose, we may reiterate the so-called Rubin Causal Model (RCM), after Rubin (1978). According to the RCM, the causal effect of demographic ageing on individual welfare

2 This can also be referred to as setting up a “pseudo-panel”, where the cross-sections that we pool together are indeed repeated, but independent from each other.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 20

state attitudes (τi) would equal the difference between an individual i’s attitudes when exposed to a high level of demographic ageing (Yi1 ), and when exposed to a low level (Yi0 ) of demographic ageing, according to

τi = Yi1 – Yi0

As we may be aware, however, we can never observe both these potential outcomes at the same time (Gerber & Green 2008). So, what we would want to do is compare different individuals from different countries (or, in our case, country-years) with varying levels of demographic ageing, and assess whether attitudes tend to look one way in one context, and one way in another – and thereby arriving at what we would want to call the causal effect of demographic ageing. Doing this, however, we are inevitably exposed to the risk of there being factors other than demographic ageing that causes differences in attitudes – we are exposed to the risk of endogeneity. The way that researchers deal with endogeneity varies across designs. In experiments, we would solve the problem by assigning treatment randomly, but it is hard to see how we could randomly assign individuals to contexts with high or low levels of demographic ageing. One could perhaps try to instrument for demographic ageing using a factor that is perfectly or as good as exogenous, but such an approach comes with its own challenges. In this study, then, the best option is arguably an observational design across space and time, where we avoid endogeneity not via randomisation ex ante, but via statistical controls ex post. If we succeed in this, and arrive at a model that accounts for the most important factors that could distort the relationship between demographic ageing and individual attitudes – within countries, and over time – we can claim to have uncovered its effect, conditional on holding these factors constant. This is to say that we may have met the so-called conditional independence assumption, or the CIA (Angrist & Pischke 2008, p. 39). We achieve this partly by using a fixed-effects model, but as we will show shortly, it will also be necessary to include covariates that vary over time. And while mentioning this, it is worth stressing that no matter how rigorous we are in this regard, we will inevitably risk neglecting the confounding effects of some variable (as captured by the error term, ε, in the coming regression models). Additionally, we are not in a position to ensure that our independent variable actually precedes our dependent variables in the causal chain – an important criterion of any causal model. The TSCS therefore has its issues with regards to making causal inferences, and it is by no means equal to an experimental design. But all things considered, it is arguably the most logical approach to our research problem, and it represents a significant step forward compared to previous research.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 21

4.2. Data and Operationalisations

Our hypotheses require data on both the individual level, and the country level. Data have thus been collected on both levels, which makes for a hierarchical data structure where individual-level data are nested within individuals’ country of residence at the time of the data collection, as well as their country overall. For the source of individual attitudes data, the choice has fallen upon the International Social Survey Programme’s (ISSP) “Role of Government” modules. The ISSP is, as the organisation describes itself, “a cross-national collaboration programme conducting annual surveys on diverse topics relevant to social sciences” (ISSP, 2020), and it provides us with comparable, representative cross-country attitudes data that should be suitable for our purposes. Indeed, the “Role of Government” modules feature a diverse set of measurements of individual welfare policy preferences, as well as individual sociodemographic factors. These factors include the “usual suspects” in comparative political attitudes research: gender, age, income and years of education, but also other potentially interesting factors. We will be using the three latest “Role of Government” rounds – from 1996, 2006 and 2016 – and we will therefore be able to analyse high-quality, representative survey data for roughly 50 000 individuals from a total of 13 advanced democracies. These democracies include Australia, Denmark, Finland, France, Germany, Great Britain, Japan, New Zealand, Norway, Spain, Sweden, Switzerland, and the United States. We may note that Denmark and Finland were not part of the 1996 survey, and respondents from these two countries therefore enter into the dataset in 2006. The perhaps modest number of 13 countries follows from the fact that not all countries in a given round were also included in the previous and following rounds. The number of countries that were included in each consecutive round was therefore lower. Still, these are all countries that are facing the challenge of demographic ageing, and that can be seen as having some significance in global policy development. This sample of countries should thus be relevant in terms of empirical generalisations, as well as future-oriented discussions based on these. While on this topic, we may note why the choice did not fall upon the European Social Survey (ESS), for instance, which would be a strong contender in terms of baseline survey quality. Simply put, in relation to our research objective, the ESS provides an insufficient coverage of welfare attitudes in their two welfare-oriented rounds (2008, 2016), and as its name implies, it focuses only on European populations. There are thus clear advantages to using the ISSP, since we can access variables that are sufficiently differentiated, as well as consistently measured over a longer time frame. The coverage of non-European populations also makes for a contribution in relation to the previous inquiries on the subject. True enough, the ageing challenge is not only present in Europe.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 22

As for the country-level data, we will be drawing mostly upon data from the OECD, the World Bank, and the Penn World Table – sources that all provide state-of-the-art data, and that are frequently used in comparative country-level analyses. These data have mainly been obtained from the Comparative Welfare States Data Set (Brady et al., 2020), but more recent demographic figures were obtained from the Comparative Political Dataset (Armingeon et al., 2020). We will mention the specific source of a given country-level variable in the coming section on independent variables.

Dependent variables For our main dependent variables, we will be analysing individuals’ spending preferences towards the welfare state. This will allow us to distinguish differences in the effort that individuals believe should be put into the welfare state, and the policies that it consists of. In the ISSP Role of Government survey, then, respondents are accordingly asked about their preferred levels of government spending on welfare state policies (see figure A1 in Appendix A for an excerpt). The instruction reads:

“Listed below are areas of government spending. Please show whether you would like to see more or less government spending in each area. Remember that if you say ‘much more’, it might require a tax increase to pay for it”.

The respondents’ preferred levels of spending are then measured on ordinal Likert-scales, ranging from 1 to 5. Having inverted the original response scale, the lowest alternative (1) indicates a preference for much less spending on the policy in question, whereas the highest (5) indicates a preference for much more spending. The middle alternative (3) indicates that the respondent is content with current levels of spending. Following convention, we treat this scale as having been measured on an interval level (Teorell & Svensson, 2007, p. 109), as to allow for statistical analysis. In our empirical analyses, we will assess four core areas of the welfare state in total: (1) health care, (2) education, (3) unemployment benefits, and (4) old- age benefits. Table 1 provides a simple correlation matrix, and shows the general picture of how our respondents evaluate these policies. There, we see that the four variables are positively correlated, although perhaps not as strongly as one may have expected, and this, as we shall see, will have implications for the later empirical analysis.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 23

Table 1: Pairwise correlations between dependent variables Variables (1) (2) (3) (4) (1) Health care 1.000 (2) Education 0.369 1.000 (3) Unemployment benefits 0.237 0.197 1.000 (4) Old-age benefits 0.399 0.237 0.386 1.000

Hypothesis 1 will be tested using a spending preferences index, where we combine the respondents’ responses regarding each of these policies into a general welfare state spending preference. Testing hypotheses 2 and 3, we will instead focus on spending preferences for specific policies – education and old-age benefits in particular. The index-approach that we take when testing hypothesis 1 is frequently used within the political attitudes-literature, and our index can be seen as an attempt to acquire a measurement of an average welfare state preference, as well as a measurement of the general underlying concept of “welfare state support”. It is thus similar to, but not perfectly analogous to the kind of indices that are constructed to measure concepts that are hard to measure on their own, as is common within fields like political psychology. Using such indices, we would be seeking high internal consistency between a number of variables, as indicated by e.g., a high Cronbach’s alpha. This is not the sole objective in this context, however, since we are in practice assessing a summary of our respondents’ spending preferences towards specific policy areas. Out of curiosity, though, when testing the internal consistency of our own index, we receive a Cronbach’s alpha of 0,63. This implies that the index might in fact be said to be capturing an underlying concept, although this ‘alpha’ is probably not sufficiently high compared to established standards (Cortina, 1993). On a more technical note, after having been constructed, the index has been divided by four so as to have the same scale as its component variables (although the index inherently allows for more variation). Having established our choice in variables, it should be valuable to assess how our respondents’ spending preferences towards these variables have developed over the relevant time-period. We can use figure 1 to distinguish a rather clear stability between 1996 and 2016, with a slight peak in 2006 (see figures A3-A7 in Appendix A for developments within countries). Notably, none of the policy areas that are captured by the variables are on average met with a preference for less spending across this time period. This lends empirical support to the proposition that a prominent feature of modern welfare politics is, indeed, high and stable levels of welfare state support (Pierson 2001; Brooks & Manza, 2008). However, there are still some differences between policies. Unemployment benefits receive the lowest support of the four, being the only area in which respondents are, on average, “only” content with current levels of spending (echoing theories of “welfare deservingness” (van Oorschot,

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 24

2000; 2006). Conversely, health care stands out as being the most popular, although education policy is a close runner-up as we reach 2016.

4

3.8

3.6

3.4

3.2

Averagespending preferences 3

1995 2000 2005 2010 2015 Year

Index Health care Education Unemployment benefits Old-age benefits

Figure 1: Spending preferences towards welfare state policies at survey years

A key question that we must ask ourselves in a section such as this is whether spending preferences is indeed an accurate operationalisation of individual support for the welfare state. We may begin by noting that indicators such as ours have been critiqued along the lines that it may be unclear whether wanting “much more spending”, for instance, has the same connotation in Sweden as it does in, say, the United States, or Japan. This objection is certainly valid to some extent. It may also be unclear whether an indifferent spending preference does indeed indicate indifference, or if it instead represents a type of non-answer (Goerres & Prinzen, 2012). Overall, the author recognises these sorts of critiques, and encourages an on-going discussion on what can be done to improve comparative attitudes research in the future.3 In this thesis, the idea is that these indicators can still be very valuable in terms of highlighting interesting relationships in terms of preferred policy effort on the aggregate level, with contextual differences between countries being of somewhat less importance.

3 See Garritzmann et al. (2018) for a good example of an attempt at improving welfare attitudes indicators, specifically in relation to social investment policies.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 25

Continuing on this point about valid operationalisations, there has been (as was alluded to earlier) a debate regarding the conceptualisation and measuring of welfare state change. Authors like Korpi & Palme (2003) argue that macro-level spending indicators obscures important developments within the welfare state. Although I agree, I do not believe that this problem translates to individual spending preferences. We are dealing with very complex developments within equally complex institutions, and it is not reasonable to expect the average citizen to discriminate between spending and benefit generosity in ways that one might need to do within conceptual debates. Studies made in Sweden – a country with a famously comprehensive welfare state – show how citizens know rather little about the generosity or eligibility conditions of common insurances, benefits, and services (Lindbom, 2011), and this echoes past studies about the low levels of political knowledge among citizens (e.g., Bartels, 1996). One may also note that the above discussion is of little relevance for areas like education, where spending may in fact be a rather adequate indicator (see Jensen, 2011). Using spending preferences as measured by the ISSP is also advantageous since respondents are explicitly told that governments might have to increase taxes in order to finance higher spending. This should provide a valuable dampening effect on otherwise too generous answers, since respondents are reminded that there is no such thing as a free lunch. Using these sorts of spending preferences, we also acknowledge empirically that the welfare state is no monolith, but indeed a combination of policies that may be perceived differently by the public. To be sure, this is not always the road taken within the literature. There has for some time now been a rather strong focus on single indicators, like redistribution preferences (e.g., Dallinger, 2010 on the role of structural indicators; Finseraas, 2009 on income inequality; Burgoon, 2014 on immigration). This is an indicator with clear limitations. Asking whether governments should reduce income differences is quite different from asking whether they should secure “some basic modicum of welfare for its citizens” (Esping-Andersen, 1990, p. 36), or, say, ensure adequate health care for all (cf. Jordan, 2013, p. 135). And even if redistribution preferences do tap into welfare state support in a sufficiently valid way, they effectively gloss over all the potential discrepancies between different policy areas in terms of the public support that they may enjoy. Insights such as those of van Oorschot (2000; 2006) about welfare deservingness would be impossible to arrive at, using redistribution preferences.4 Spending preferences, then, should be a sufficiently valid as well as useful operationalisation of individual welfare state support. And although we will partly be using an index – in practice a form of single indicator – we will have the opportunity to parse out the effects on each of its component variables.

4 One can see why this approach has been so frequent within the literature, though. The ESS, for instance, has measured redistribution preferences continuously since 2002, and so, using redistribution preferences might be reasonable from a methodological standpoint, since it allows for pooled analyses.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 26

Independent variables: country level Here, we discuss the independent variables that will be key to testing our hypotheses. Since we rely upon observational data, it will be necessary to introduce a set of covariates alongside our main explanatory variable. This, then, is in order to deal with OVB, and thereby meeting the CIA. While our dependent variables were found on the individual level, our main independent variable is a country’s elderly dependency ratio (DR). Specifically, the DR is calculated as the ratio of persons in a country aged 65 and over, to those aged 15 to 64. This ratio can then be multiplied by a hundred to provide a percentage ratio that straightforwardly describes the balance between the working-age and the elderly, according to

Pop. ≥65 ( ) x 100 Pop. 15-64

The figures for this important variable are drawn from the OECD Population database, and there is of course no doubt that the DR has increased within the 13 countries included in our sample. This shows in figure 3, where it has been plotted between 1960 and 2016, with dashed vertical lines at the survey years to highlight the variation between these years (see figure A8 in Appendix A for country-separated graphs). Japan, for instance, a country that started off in 1960 with a DR lower than 10 percent, has ended up with one of more than 40 percent in 2016 – following an exponential growth over the past 60 years.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 27

50

40

30

Dependency (%) Ratio

20 10

1960 1980 2000 2020 Year

Australia Denmark Finland France Germany Japan Norway New Zealand Spain Sweden Switzerland United Kingdom United States

Figure 2: Dependency ratio in sample countries, 1960-2016

We can also see the exact measures of each country’s dependency ratio at the years of the surveys in table 2. In terms of the overall change in the DR between these years, we see an increase of 7,4 percentage units – from 22,2 to 29,6 percent. As we can see, though, this development in the sample average masks some differences between countries. Japan, then, has had a very rapid increase, but New Zealand, for instance, has seen a much more modest increase, with the lowest DR in 2016 at 22,9 percent.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 28

Table 2: Dependency ratio at survey years Country 1996 2006 2016 Australia 18,1 19,3 23,8 Denmark* 22,4 23,1 29,5 Finland* 21,6 24,4 32,8 France 23,6 25,4 30,5 Germany 24,0 29,6 32,2 Japan 21,8 31,8 45,7 Norway 24,5 22,3 25,2 New Zealand 17,6 18,4 22,9 Spain 22,6 24,1 28,5 Sweden 27,4 26,4 31,5 Switzerland 22,0 23,7 26,9 United Kingdom 25,5 23,2 28,0 United States 19,3 18,5 23,1 Overall 22,2 23,6 29,6 * Were not part of the 1996 round of the ISSP Role of Government.

Regarding the validity of the dependency ratio as an indicator of a pressure to reform existing systems, one issue is that some working-age people do not work (for various reasons), and that not all people stop working (and become “dependent”) when they turn 65. True enough, the number of working elderlies is even bound to increase, thanks to increasing longevity. In reality, then, the idea of an increasing reform pressure is inevitably not as simple as suggested by the dependency ratio. Still, it should be a valid indicator seen on aggregate, and using it allows us to properly relate our study to the previous literature. One alternative would have been a country’s proportion of retirees to its working population – but a country’s mandatory retirement age, for instance, is in itself very much a product of its welfare state. Therefore, it would be harder to argue for the exogeneity of this variable in relation to our dependent variables, particularly the one related to old-age benefits. The DR, then, despite these mentioned issues, should be a reasonable choice as our main independent variable. Going beyond the DR, it is necessary to introduce a set of country-level covariates that are thought to be able to influence both the dependency ratio and individuals’ spending preferences – and that are not fully absorbed by our country fixed effects. First, we include a variable for a country’s log real gross domestic product (GDP) per capita – an overall measure of economic affluence. This may influence the dependency ratio in that affluence tends to reduce fertility rates – in line with developments towards modernisation and individualisation – as well as cause increased longevity. Log GDP per capita is thus expected to be positively correlated with the DR. It is also expected to affect individual support for the welfare state, although the expected direction is somewhat uncertain. Individuals in poorly performing countries may seek increased protection, but they may also see the necessary tax-effort as

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 29 too burdensome given the circumstances. Individuals in countries that perform well, conversely, may feel less need for an extensive welfare state, but they may likewise have a higher tolerance for it. In the literature, Dallinger (2010) showed its correlation with redistribution preferences to be negative (cf., Blekesaune, 2007). But, as we know, redistribution is not necessarily the same thing as welfare, so nothing is certain. Data for real GDP per capita are provided by the Penn World Table. A country’s average life expectancy is expected to exert a direct, positive influence on the dependency ratio. The longer people live on average, the larger our share of individuals over 65 (i.e., our numerator) becomes. It may also cause a higher support for old-age benefits, in that the need for it would increase with longevity. Data for life expectancy are provided by the OECD’s Health Statistics. A country’s , however, is expected to have a direct, negative influence on the dependency ratio, seeing we can expect our share of individuals between 15-64 (i.e., our denominator) to be relatively larger. The fertility rate may have a positive effect on support for education, which would become more important following large youth cohorts. Data for total fertility are provided by World Bank’s World Development Indicators. A country’s infant mortality rate is expected to have a positive influence on the dependency ratio, in that the share of individuals between 15-64 can be expected to be smaller. The infant mortality rate is expected to increase support for health care – the reason being arguably quite apparent. Data for infant mortality are also provided by the OECD’s Health Statistics. The levels of immigration in a country are thought to have a direct, negative influence on the dependency ratio, since those who immigrate to a country tend to be young. The relationship between immigration and individual support for the welfare state remains contested, though. The current tally possibly indicates a negative relationship (Mau & Burkhardt 2009; Eger 2010; Dahlberg et al. 2012; Burgoon 2014; Alesina et al. 2019), although some have found a null-relationship or even a positive relationship (Brady & Finnigan 2014; Sumino 2014; Kwon & Curran 2016). Notably, these last studies use the same attitudes data that we do here, and so we might perhaps expect a positive relationship. We thus include data for the countries’ stock of international migrants, as provided by the World Bank’s World Development Indicators. The data are provided in five-year intervals, and the attitudes data thus have to be paired with immigration data for the year before each survey. We also incorporate a country’s unemployment rate. Unemployment rates can clearly vary over time within countries – and will thus not be absorbed by fixed effects – and one could imagine that high unemployment rates, could drive emigration among younger people, which would increase the DR. It is also expected to have influence on attitudes, especially in relation to unemployment benefits, and perhaps in relation to other variables as well. The data for unemployment rates are provided by the OECD’s Annual Labour Force Statistics.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 30

Independent variables: individual level Following contributions such as those of Emery (2012), and Hess et al. (2017) we, too, make sure to incorporate individual age as an individual-level covariate, but also, and more importantly, as a moderator. In the data, we have access to respondents’ ages in years within all rounds, as well as in all countries – except for Denmark in 2016. These Danish respondents’ ages were instead measured in ten-year categories. This would be a slight issue for variable consistency, were it not for the fact that we may be more interested in these kinds of categories than a continuous scale. In relation to theory, it makes more sense to focus on “the old” and the “the young”, and we can also simplify the interpretations of cross- level interactions significantly by using these truncated categories.5 However, we must still discuss how our categories should be specified – especially with regards to categorising respondents as being old. While there are nowadays good reasons to discuss this on a more conceptual level (Neugarten & Neugarten, 1996; Kohli, 2006), our discussion is more practical. For sure, the most logical and straightforward choice here is to use a cut-off at 65. This is the standard retirement age in many of our sample countries, and to leave work should mark a sharp and significant change in one’s life, which could have effects on one’s inner life in general (Osborne, 2012), as well as one’s political attitudes and behaviour (e.g., Melo & Stockemer, 2014). In our empirical analyses, this is the cut-off that we will rely on. Likewise, though, one could envisage old age becoming relevant earlier or later than at 65, and so, our choice in this regard is arbitrary to some extent. As for the remaining respondents, we categorise those between 18 and 35 as being young, and those between 36 and 65 as being middle-aged. In the end, we arrive at two age-category dummy variables: one for the middle-aged, and one for the old, with the young constituting the reference category for both dummies. Using the age categories in this way, then, will simplify our interpretations of cross-level interactions. As a standard control variable on the individual level, that can increase the precision of our models, we include a dummy variable for gender, with men as reference category, and with women as the dummy value. Past research (e.g., Gelissen, 2000; Blekesaune & Quadagno, 2003) have made it very clear that women are generally more supportive of welfare than men, and so we can expect this variable to have a positive effect on the outcome(s). We also include educational background, which acts as a proxy for a respondent’s socioeconomic background. The ISSP allows for the use of both degree-based and year- based variables for measuring educational attainment, but to simplify the use of this control variable, we opt for the year-based approach. We expect this variable to be negatively related to overall welfare state support, since education tends to introduce one into the more affluent

5 The reader will later get to assess alternative models where for instance a continuous age variable is used.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 31 middle classes, although it might still be positively correlated with support for education policy (Garritzmann et al., 2018). A variable that we might also have wanted to include is the respondents’ personal income – a factor that clearly affects individual support for the welfare state (Gelissen, 2000). However, this can be argued to constitute what authors like Angrist & Pischke (2008) refer to as a “bad control” (p. 47). This is because personal income levels tend to follow age quite closely, and it ought to be strongly correlated with educational background as well. We will thus not use this variable as an individual-level covariate. However, we will attempt to use this variable in an exploratory manner later on in our analyses. And so, regarding the construction of this variable, the respondents’ personal incomes have been combined into a decile variable. This allows us to compare respondents not on their income levels in absolute numbers, as measured in euros, pounds or dollars, but rather their income relative to other respondents’ incomes (Busemeyer et al., 2009).

4.5. Estimation Strategy

Testing our hypotheses, we will rely on a set of fixed-effects, ordinary least squares (OLS) multivariate regression models, where we assume individual support for the welfare state to be a linear function of the dependency ratio. OLS, as we know, is a classic tool for assessing statistical associations between variables, and while generally being susceptible to endogeneity, we come quite a long way towards robustness by including country and time fixed effects (Kmenta, 1990). We also supplement our models with further country-level, and individual-level covariates. We include cross-level interaction terms in order to test our third hypothesis, whereby we assume the main relationship to be moderated by individual age (as captured by our age- category dummies). Following Brambor et al. (2006, p. 66), when we include these, we also include their constitutive terms, which is something that, if neglected, can cause incorrect estimates. Following the advice of Yzerbyt et al. (2004; see Österman, 2018, for an empirical application), moreover, we also interact our country-level covariates with our individual-level moderator. The intuition behind this is that since we include country-level covariates to isolate the general relationship between the DR and individual support for the welfare state, we should also control for their potential influence on the interaction between the DR and individual age. It may be that one or more of the other country-level factors drive the interaction that we see between these two variables – thus making it spurious – or, instead, suppress one that we should be seeing. The empirical value of this strategy will become clear when the analyses are presented. Following the established literature, our regressions are computed using clustered standard errors on the country level. The intuition behind this is that survey respondents

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 32 from the same country are likely to be similar due to their common country. In statistical terms, this is saying that there may exist a degree of intragroup error correlation (Moulton 1986; Angrist & Pischke, 2008), and, if left unadjusted for, this may provide standard errors that are deceivingly small. Failing to cluster standard errors would thus cause misleadingly low p- values, which would naturally increase the risk of incorrectly rejecting the null hypothesis (i.e., making a type-I error). We can also note that the clustering takes care of the risk that residuals – the deviations from the regression line – are unequally distributed along the horizontal axis. This is to say that it protects us from the potential heteroskedasticity that would otherwise disturb our analyses, while also invalidating one of the key assumptions behind OLS regression (Lewis-Beck & Lewis-Beck, 2015, p. 24). To summarise our estimation strategy, we can formulate our model according to

Yijt = α + β1Xjt + β2Zit + β3(Xjt × Zit) + Лjt + Γijt + γj + δt + εijt

where the left-hand term Yijt represents the outcome (for each respondent i, in each country j, in each year t). The first term on the right-hand side, α, represents the intercept of the model. The second term, Xjt, represents the country-level variable of interest: the dependency ratio (for each country j, in each year t). Zijt represents respondents’ individual age (for each individual i, in each country j, for each year t). (Xjt × Zit) represents the cross-level interaction between the DR and individual age. Лjt, and Γijt represents vectors of the country-level, and individual-level covariates, respectively. γj represents the country dummies, and δt represents 6 the survey year dummies. εijt, lastly, denotes the error term.

6 Included in the model (but not shown) are then also the terms for the interactions between the country-level covariates, and individual age.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 33

4.6. Descriptive Statistics

Before moving on to the empirical analyses, we can examine some summary statistics for all of our variables. Table 3 provides measures of central tendency (mean) and dispersion (standard deviation), the number of observations, and minimum vs. maximum values, for both our individual-level variables and our country-level variables.

Table 3: Descriptive statistics for key variables Variable Obs. Mean Std. Dev. Min Max Individual level Spending index 52313 3.639 .592 1 5 Health care spending 55406 3.949 .836 1 5 Education spending 54987 3.879 .822 1 5 Unemployment spending 54064 3.045 .951 1 5 Old-age spending 54734 3.689 .824 1 5 Age (years) 55691 48.742 17.224 15 98 Age (groups) 56829 1.932 .673 1 3 Gender 56844 1.518 .5 1 2 Years of education 51982 12.72 3.897 0 65 Income (deciles) 47413 5.04 2.92 1 10

Country level* Dependency ratio 57070 .25 .053 17,6 45,7 Log GDP per capita 57083 10.514 .367 9.806 11.463 Immigration stock 57083 12.874 7.188 1.109 29.387 Unemployment rate 57083 7.377 4.518 3.117 22.149 Life expectancy 57083 80.107 2.048 75.7 84.1 Total fertility rate 57083 1.664 .258 1.15 2.1 Infant mortality rate 55823 4.205 1.354 1.9 7.3 * Country-level statistics are calculated on the individual-level observations, as to be appropriately weighted.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 34

5. Results and Analysis

This fifth section presents the main results of this thesis. We begin by assessing the relationship between demographic ageing and individual support for the welfare state, as measured by our general spending index. Then, we consider the possibility of policy divergences, i.e., that respondents evaluate specific welfare state areas in substantially different ways. Performing these analyses, we also assess the cross-level interaction between demographic ageing, and individual age. Afterwards, we present some further, and more investigative analyses.

5.1. Demographic Ageing and General Support for the Welfare State

Table 4 helps us test our first hypothesis, which stipulates that individuals’ general support for the welfare state increases, following demographic ageing. Assessing the results, however, there is little that suggests any form of significant relationship between the DR, and respondents’ general spending preferences. The bivariate relationship is null, and as we build our robust regression, beta coefficients remain small and insignificant. Assessing the coefficient for the DR in the fifth model, which includes fixed effects, and the full set of covariates (on both country and individual level), we do see that a one percentage-unit increase in the DR would increase spending preferences, but this coefficient is very small, and not at all significant (p=0,675). Hence, we do not have the sufficient empirical basis on which to confirm hypothesis 1, although one could argue that these null results can be suggestive of an overall preference for the status quo. In any case, the results do not present us with a clear “popular verdict” regarding the should-be of the welfare state, as captured by respondents’ general spending preferences. We will not comment on our covariates in any great detail, since they are merely included to isolate our main independent variable. We can name immigration, total fertility, and infant mortality as seemingly the most influential country-level confounders; immigration, interestingly, is positively correlated with general support for the welfare state. But since we cannot be sure to have isolated this covariate from its potential confounders, it is unwise to give it a substantial interpretation (cf. Hünermund & Louw, 2020). Echoing previous research, however, we can say that being a woman seems to have a positive effect, the extent of one’s education has a negative effect, and, importantly, age in itself is very much positively related to general spending preferences.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 35

Table 4: Effect of dependency ratio on general spending preferences (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio 0 -.009 -.003 .004 .004 .003 .007 (.005) (.007) (.009) (.008) (.008) (.009) (.009) Log GDP per capita .253 -.095 -.205 -.204 -.2 (.144) (.14) (.233) (.236) (.226) Immigration -.003 .05*** .048*** .048*** .047*** (.004) (.013) (.013) (.013) (.013) Unemployment rate .006 .001 .001 .001 -.001 (.009) (.005) (.007) (.007) (.007) Life expectancy .018 .014 .028 .028 .049 (.017) (.06) (.052) (.052) (.047) Total fertility rate -.324 .411** .362* .362* .223 (.239) (.17) (.186) (.187) (.198) Infant mortality .044 .133*** .166*** .166*** .225*** (.032) (.034) (.044) (.044) (.045) Individual level

Woman .105*** .105*** .105*** (.008) (.008) (.008) Middle age .026 .031 2.124** (.019) (.08) (.826) Old age .03 .005 3.657** (.026) (.105) (1.292) Years of education -.015*** -.015*** -.015*** (.003) (.003) (.003) DR × Middle age 0 -.006* (.003) (.003) DR × Old age .001 -.003 (.003) (.005) Constant 3.646*** 3.713*** -.06 .77 .73 .732 -1.061 (.135) (.128) (1.679) (4.291) (4.211) (4.259) (4.045) Observations 52300 52300 51121 51121 46664 46664 46664 R-squared 0 .08 .038 .09 .107 .107 . 108 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

And while mentioning age, we may assess its moderating influence on the main relationship. True, the idea of age-moderated effects has been deemed most relevant in relation to the specific policy areas, but here we see that age might have something of a moderating impact on the relationship between the DR and general spending preferences, too. The main effect in model 7 represents the effect among young respondents (since they constitute the reference group in relation to the two age-group dummies), and this coefficient is (slightly) larger than that in model 5 – suggesting that the effect on general spending

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 36 preferences might be higher among the young. Indeed, both interaction coefficients for middle-aged and old are negative, and the middle-age interaction is significant at p<0,1 (for old-age, p=0,552). But, in line with the main relationship’s coefficient in model 5, these interaction coefficients are very small, and, notably, it is not in the old-age category that we find the significant interaction. We rather expected the opposite, namely that older individuals would be keener on supporting the welfare state, following demographic ageing. Still, the old-age interaction coefficient does arguably not indicate that they become less keen on supporting it. Taken on the whole, then, it is most likely the case that the DR does not exacerbate age-group differences with regards to general support for the welfare state in a substantial way.

5.2. Demographic Ageing and Support for Specific Welfare Policies

In this section, we analyse the effects within the specific policy areas. The results for the general spending index were somewhat noneventful – but a reason for this may very well be our theorised discrepancies in how demographic ageing affects peoples’ spending preferences towards different policies. We may then have analysed an index where different effects cancelled each other out to some degree. Hence, we will now assess the effect of the dependency ratio within our theoretically most interesting policy areas: education, and old- age benefits. Health care and unemployment benefits are omitted as they are the least relevant to theory, and, inevitably, because of length-restrictions. The results for these variables can be assessed in tables B1 and B2 in Appendix B, though, and we will make some references to them as we go on.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 37

Table 5: Effect of dependency ratio on spending preferences for education (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio -.001 .017 .017 .036** .039** .039** .054*** (.009) (.015) (.011) (.015) (.016) (.016) (.015) Log GDP per capita .555*** -.15 .029 .033 .052 (.091) (.247) (.529) (.531) (.518) Immigration .012** .074*** .078*** .078*** .083*** (.005) (.022) (.023) (.023) (.021) Unemployment rate -.003 .002 0 0 .001 (.004) (.008) (.01) (.01) (.011) Life expectancy -.004 .032 .011 .011 .002 (.021) (.083) (.09) (.09) (.085) Total fertility rate -.462*** .365 .415 .412 .471 (.121) (.317) (.375) (.378) (.393) Infant mortality .098** .148** .123* .123* .173** (.034) (.054) (.063) (.062) (.072) Individual level

Woman .062*** .062*** .062*** (.017) (.017) (.018) Middle age -.045 -.037 .52 (.026) (.128) (1.249) Old age -.117*** -.189 .653 (.036) (.135) (1.422) Years of education .013*** .013*** .013*** (.004) (.004) (.004) DR × Middle age 0 -.02*** (.004) (.004) DR × Old age .003 -.022*** (.004) (.005) Constant 3.895*** 3.604*** -1.874 -.781 -1.316 -1.33 -1.567 (.222) (.286) (1.592) (6.164) (7.807) (7.894) (7.527) Observations 54974 54974 53740 53740 48927 48927 48927 R-squared 0 .06 .043 .067 .078 .078 .08 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Effects on Spending Preferences for Education Beginning with education (see table 5), there are in fact clear indications that the dependency ratio is associated with higher spending preferences. Although the bivariate relationship is negligible, as we introduce country and year fixed effects, and the other country-level covariates, there is a rather marked effect. Indeed, in model 4, we receive a coefficient of 0,036, significant at p<0,05. Adding individual-level variables in model 5, which can

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 38

potentially increase precision, the coefficient remains similar in size at 0,039, and remains significant at p<0,05. As we know, model 5 represents our fully specified model with regards to the main effect of demographic ageing, and the basic interpretation of this is that one percentage-unit increase in the DR is associated with an increased spending preference by 0,039, on average. We can show the meaning of this coefficient graphically by plotting the average predicted spending preferences among respondents, for different values of the DR (see figure 3). As indicated by the width of the confidence intervals, most values of the DR lie between 20 and 30 percent, with fewer values above this range. This means that the predictions for higher values of the DR are much more uncertain. The graph should still be useful as a general illustration of the relationship, though. We can also substantiate the relationship in words by noting that the difference in the 2016 dependency ratio between Finland and New Zeeland – two countries with a high and low DR, respectively – makes for a difference in spending preferences of roughly 0,4 on average.7 This corresponds to almost half a step on the response scale (noting that this is based on the partial relationship between the DR and the outcome, conditional on covariates being held constant). We are thus looking at quite a considerable effect, and this is especially the case when considering the rigidity in spending preferences over time that we observed earlier (figure 2).

5.5

5

4.5

4

Predictedmean spending preferences

3.5 3

15 20 25 30 35 40 45 Dependency ratio (%)

Figure 3: Predicted spending preferences for education based on different values of DR

7 (-1,316 + [0,039 × 32,8]) – (-1,316 + [0,039× 22,9]) = 0,3861. Comparing New Zealand with Japan, which may be something of an outlier here, the difference amounts to an impressive 0,89 units.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 39

Importantly with regards to hypothesis 3, we also see evidence of a significant cross- level interaction between the DR and individual age. In model 7, then, the main coefficient of 0,054 represents that of the young respondents. This is higher than the main effect in model 5, and the two interaction coefficients – which are both significant at p<0,01 – convey that the effect of the DR is on average 0,02 units lower among the middle-aged and the old, compared to the young. Comparing these to the equivalent interaction coefficients in model 6, we see how there are indeed other cross-level interactions that (in this case) suppress the interaction between the DR and individual age. Controlling for these therefore appears to have been well-founded. What we see here, then, is that the effect is not only lower among retirees, but among working-age people too. This suggests that individual age has a rather quick fade-off effect – as shown in figure 4, where we can observe the different beta coefficients for the DR across the three age categories.

.08

.06

.04

.02

Marginal effectsof Marginalratio dependency 0

18-35 35-64 ≥65 Age

Figure 4: Marginal effects of dependency ratio on spending preferences for education

Following the advice of the American Statistical Association (ASA) – to always try to supplement significant results with further evidence (Wasserstein & Lazar, 2016, p. 132) – we may note that another question in the survey, that asks whether it is the “government’s responsibility to give financial aid to university students from low-income families”, provides similar results (see figure A2 in Appendix A for questionnaire, and table B5 in Appendix B for results). This question has a more redistributive character, and the effect is weaker (also

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 40 because the variable is measured on only 4 steps, allowing less variation). But it is nevertheless significant, and age has the same moderating influence: the effect fades with age, to the point that the marginal effect within the old-age group is in fact clearly insignificant (p=0,527). The effect among those individuals may therefore be null, and this clearly corroborates our previous findings. In sum, it seems that individuals in older populations express a higher support for education policy (conditional on covariates held constant). As has been argued, education represents one of the most important ways to invest in the future workforce, and, similar to what Busemeyer & Garritzmann (2019) found with regards to economic globalisation, these results are arguably suggestive of that that demographic ageing causes citizens to express a stronger support for social investment. This provides initial support for our second hypothesis, although we need to show the opposite relationship for old-age benefits to fully be able to confirm it. Beyond this, we have also seen that the positive effect is indeed lower among older individuals. While it remains positive across the age categories, the results for the supplementary variable contests this interpretation somewhat. In fact, one may also assess the interaction between the DR and individual age, when measured on a continuous scale (see figure C1 in Appendix C). We see there that the effect turns insignificant for higher ages, even before the number of observations becomes very low (which could otherwise explain insignificant coefficients). In any case, these results provide initial support for hypothesis 3, although here we must also assess the results for old-age benefits before drawing our final conclusions.

Effects on Spending Preferences for Old-age Benefits As for old-age benefits (see table 6), there are no coefficients as strong as there were for education, although there are some indications that the DR is associated with lower spending preferences. In model 2, which includes fixed effects, the coefficient is negative and significant on p<0,05. However, it is not robust to the adding of the other covariates: it turns insignificant in model 3, and remains so in the remaining models. This includes the fully specified model 5, even though this coefficient is actually not that far from being significant (p=0,156). In model 7, which includes the cross-level interactions (while also controlling for interactions with country-level covariates), we see no substantial signs of interaction effects either (p=0,425 for the middle-aged, and p=0,740 for the old). The strict interpretation here is that demographic ageing does not affect individuals’ spending preferences towards old-age benefits in any direction, although we are somewhat close to being able to interpret the negative coefficient in model 5. Moreover, there is little here that speaks for there being a noteworthy cross-level interaction between the DR and individual age.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 41

Table 6: Effect of dependency ratio on spending preferences for old-age benefits (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio .001 -.019** -.011 -.011 -.013 -.017 -.008 (.006) (.007) (.009) (.01) (.009) (.012) (.016) Log GDP per capita .107 .055 -.092 -.084 -.05 (.153) (.203) (.34) (.352) (.355) Immigration -.008* .027 .022 .022 .029 (.004) (.022) (.017) (.017) (.02) Unemployment rate .005 -.004 -.006 -.006 -.009 (.01) (.004) (.006) (.006) (.007) Life expectancy .046** -.073 -.043 -.043 -.033 (.017) (.05) (.037) (.038) (.041) Total fertility rate -.142 .534** .47** .461** .2 (.248) (.183) (.163) (.172) (.17) Infant mortality .03 .13 .176 .175 .249** (.042) (.08) (.104) (.105) (.108) Individual level

Woman .112*** .112*** .114*** (.01) (.011) (.011) Middle age .099** .039 1.751 (.035) (.154) (1.362) Old age .166*** -.092 4.201** (.049) (.22) (1.803) Years of education -.036*** -.036*** -.036*** (.005) (.005) (.005) DR × Middle age .002 -.009 (.006) (.011) DR × Old age .01 -.005 (.008) (.016) Constant 3.661*** 3.892*** -.689 6.633 6.031* 6.062* 4.698 (.148) (.105) (1.75) (4.226) (3.16) (3.198) (3.057) Observations 54721 54721 53498 53498 48650 48650 48650 R-squared 0 .039 .012 .045 .087 .087 .088 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Remembering the results for education, it would have been interesting had we arrived at similarly clear-cut results here. This would have suggested a clear verdict in favour of education, but against old-age benefits. What we can do at this point too, though, is to try to triangulate these results with further tests. True enough, when assessing the equivalent “government’s responsibility”-variable regarding to support the elderly, coefficients are clearly negative across most models, and they are overwhelmingly significant (see table B6 in

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 42

Appendix B). The seventh model also shows some signs of negative interaction effects, although only significant with regards to the middle-aged. While not matching our research objectives perfectly, these results provide some valuable supplementary evidence. We can also mention the effects on spending preferences for health care (table B1 in Appendix B), a policy with something of an old-age bias. The coefficients for this variable are also negative, and the same goes for the supplementary variable about the government’s responsibility for providing health care (table B4 in Appendix B). Overall, then, there is some supplementary evidence that could perhaps encourage us to trust in the main results for old-age benefits. Alone, the results are certainly shaky in terms of significance, but when viewed in relation to these other results, one could argue that there emerges a fuller picture that is in fact interpretable. Indeed, demographic ageing may well induce individuals to reduce their support for old-age benefits – a part of the welfare state that they may perceive will be very costly moving forward, and/or that they expect will yield too little to be worth supporting. These indications thus corroborate the results of Emery (2012), showing further signs of withdrawal from elderly-oriented policies within ageing populations, even in countries outside Europe. In the context of this paper, specifically, they arguably provide us with enough evidence to consider our second hypothesis confirmed. The moderating effect of individual age is also not as strong here as it was for education. We expected demographic ageing to cause older people to become more supportive of old- age benefits than young people, and despite indications of a negative main effect, we could very well have seen a less negative, or even a positive marginal effect within the old-age group. The combined evidence shows little indication of this, though. By all accounts, demographic ageing does not cause older individuals to become more supportive of (or less opposed to) old-age benefits than young individuals. Consequently, we can only provide partial evidence in favour of hypothesis 3. Nevertheless, this section has shown that, in line with the idea of an intensified intergenerational cleavage, individual age can very well moderate the relationship between demographic ageing and support for welfare policy – although it clearly matters what policy one is analysing.

5.3. Exploratory Analyses and Robustness Tests

An important issue that has been raised in the literature (e.g., Garritzmann et al., 2018), is to what extent social investment is something that is in fact supported across socioeconomic groups. Indeed, it has been argued that social investment may actually fall short of its intended purpose, i.e., invest in those who would otherwise have difficulty taking place in the labour market, and instead favour those who are already quite well-off, thus potentially increasing various inequalities. This has pointedly been termed the “Matthew effects” of social investment (see Bonoli et al., 2017; Bonoli, 2020). In our case, it becomes pertinent to

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 43 assess whether the positive, and significant effects that we found for education, first and foremost, do in fact represent the entire socioeconomic spectrum, or if they are merely driven by a strong social investment-sentiment among, say, the middle classes. The same goes for the cross-level interaction between demographic structure and individual age – do we see the same dynamic playing out between age groups, regardless of socioeconomic group? Beyond this, it is of course also important to see whether this moderates the effects on general spending preferences, and old-age benefits. To assess this, we could add respondent income (which we would want to see as a proxy for “class”) as a second individual-level moderator in our models, and then analyse two-way and three-way interactions – but this would quickly become unnecessarily complicated. A more straightforward approach is to divide our sample, and analyse sub-samples based on income. In these analyses, the income decile-variable – that we chose not to include in the multivariate regressions, and that we can call Iijt – has been split across the middle, with one half constituting low-income and the other constituting high-income, according to

Low, I <5 I = { ijt High, I ≥5

Importantly, this does little with regards to spending preferences towards education: when comparing coefficients from model 5 in each sub-sample, low-income earners and high- income earners are affected equally. For general spending preferences – where we could yet discover theoretically interesting results within sub-samples – we do see opposite coefficient signs for low- and high-income earners. The coefficients are small, however, and far from being significant (p=0,885, and p=0,649, respectively), and so this is likely not of substantial importance. For old-age benefits, we can see that high-income earners are negatively affected (p=0,009), whereas the equivalent coefficient is not significant for low-income earners (p=0,165). This could indicate that the withdrawal from old-age benefits is indeed driven by high-income earners, but the still relatively low p-value for low-income earners indicates that the sentiment is probably shared across the income spectrum. For the cross-level interactions, income does not moderate the negative, and significant fade-off effect on support for education. For general spending preferences, we see the same kind of negative (but mostly insignificant) tendencies for both groups. For old-age benefits, we see that among low-income earners, both interactions are negative, while among high-income earners, the interaction coefficient for the oldest category changes sign and becomes positive. None of these coefficients are significant, however. In sum, we can say that the results for education are seemingly the most robust when it comes to the potential influence of income, and in extension, class. The positive effects are thus distinguishable across the socioeconomic

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 44 spectrum. The other results seem slightly more vulnerable, but sub-sample sizes may play some part, since we are in effect dividing our sample on income as well as age groups. All of these results can be reviewed in tables D1-D6 in Appendix D.8 In a similar test (results available on request), which might add some further depth to our results, the sample is divided according to the number of persons in a respondents’ household, where households with more than two persons are assumed to include children. The idea is that parenthood could drive the effect on spending preferences for education in a positive direction, whereas childlessness might instead dampen or neutralise the (potentially) negative effect on spending preferences for old-age benefits. No important differences can be reported for education, which yet again speaks to the policy’s wide popularity. Interestingly, we do get oppositely signed coefficients for old-age benefits, but these are both small, and far from being significant. Similar results are also obtained when alternative household assumptions are made. An important factor that could affect the robustness of our most substantial results is the fact that the development in the DR follows a clear, and almost linear trend over time in each one of our countries (see figure 2). It may therefore be the case that there are other background factors that affect all countries, that also increase over time, and that could affect the within-country variation of the DR. This is another way of saying that there may be something else in the background of our model that could cause us to detect an effect of the DR that is still spurious. To get at this linear time-trend, we can add an interaction between each country and each of its available years, according to

(γj + δt + [γj × δt])

where we include the interaction term (γj × δ)t, as well as its constitutive country (γj) and year

(δt) dummies. Controlling for this linear time-trend ought to be quite a hard test for our data, considering that we are only using variation between three years in total, or even just two in the cases of Denmark and Finland. We should therefore not be surprised if we end up with considerably larger standard errors (i.e., less precision) following this test. Performing this test, (see tables E1-E3 in Appendix E), the main coefficient for education (model 5) turns insignificant (p=0,286), but the cross-level interactions remain significant, which might suggest that the main relationship remains too. The relationship for old-age benefits is clearly diminished, but this is not unexpected due to the hard test that this linear-trend control imposes on the data, and overall, while we do end up with less precision, it should suffice to

8 We may also note that a second specification, with three income-groups (low, middle, high), provides similar results (available on request).

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 45 say that the adding of this linear trend-control does not change our results in a way that would call for their immediate reinterpretation. Another factor that is commonly assessed in TSCS analyses is the potential influence of lagging key variables. The idea is that, by moving the causal factor back from t to t-k, we can isolate it from OVB, as well as simultaneity (Reed, 2015). According to Bellemare et al. (2017), this can indeed be reasonable in order to isolate against simultaneity, but regarding OVB, it “merely moves the channel through which endogeneity affects the estimates” (p. 949). We should not have to worry about simultaneity, though, since it is unfeasible that individuals’ spending preferences can significantly affect the demographic structure at t, nor other country-level factors. Indeed, the causal effect of spending preferences is realised further away in time. In any case, it may be interesting to compare results using different time subscripts. Perhaps the past demographic structure is more influential than the current one? Regarding main effects, we can say for general spending preferences that the positive, but small (and insignificant) coefficient remains similar in size when regressed on a lagged DR (1-5 years). For education, the results remain very similar, although for old-age benefits, we can say that the negative tendencies become slightly less noteworthy. Importantly, however, when assessing the “government’s responsibility”-variable with regards to the elderly – which we have previously used as supplementary evidence – the negative coefficients remain significant regardless of the DR’s time-subscript. In sum, we can say that a lagged DR does not radically transform our results – an important finding in terms of robustness (results available in tables F1-F3 in Appendix F). Lastly, we note the warning of Angrist & Pischke (2008) about clustering standard errors using too few clusters. We have used 13 countries as clusters, and Angrist & Pischke, for instance, suggest no fewer than 42 (2008, p. 238). Crucially, using too few clusters may cause standard errors to remain underestimated. To examine this further, the main results – which have been obtained using the above technique – have been compared with the results when using the standard “robust”-option available in Stata (which can also correct for underestimated standard errors). Suffice it to say that none of our discovered relationships lose out in terms of statistical significance – if anything, the relationships that have been the least convincing (general spending preferences, most notably), become “more” significant when using this alternative correction (results available on request). We will not use this as proof of anything, though, since clustering is generally a better approach to account for underestimated standard errors, since it also manages heteroskedasticity.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 46

6. Conclusions

In this thesis, we have asked ourselves what role a worsening demographic structure may play for individual support for the welfare state. Accordingly, we have examined the relationship between countries’ dependency ratios at different points in time (1996, 2006, and 2016, specifically), and the spending preferences of respondents from 13 advanced democracies, towards the welfare state in general, as well as specific policy areas. We have also analysed whether the dependency ratio interacts with individual age, as to give rise to an age-dependent relationship between demographic structure and spending preferences. The empirical analyses have shown, first, that demographic ageing does not affect people’s general support for the welfare state in a substantial way. There is no general withdrawal from the welfare state, as would be predicted coming from a rational-choice standpoint – thereby stressing issues of higher taxation, and the hollowing-out of the welfare state. And while there is no significant increase either, the null-results suggest that citizens might prefer the status quo with regards to total spending – and this is arguably supportive of the nowadays classic New politics-thesis, as laid out by Pierson (e.g., 1994). On a more fundamental level, this may simply reflect a perceived want within populations to continue with a comprehensive welfare state in these times of demographic ageing. Moving down one step on the ladder of abstraction, though, we have seen strong evidence that it increases support for education policy. Clearly, it is these effects that drive general spending preferences in a more positive direction. Interestingly, however, individual age has also been shown to be a negative moderator of this relationship. We can therefore give some credit to those who have indeed assumed that the intergenerational cleavage will intensify, following demographic ageing (e.g., Busemeyer et al., 2009) – although our results are probably not so compelling as to suggest an emergence of a serious generational conflict. Indeed, the positive relationship is seemingly present across all age groups – thus illustrating the popularity of social investment-type policies within ageing populations. Its popularity cuts through socioeconomic boundaries too, meaning that such policies may hold quite broad public legitimacy. Conversely, despite established notions of elderlies’ welfare deservingness, we have seen indications that demographic ageing decreases individuals’ support for old-age benefits – a core representative of the traditional, compensatory welfare state, and a policy whose demand is expected to grow considerably in the coming years. Our results corroborate earlier evidence, and the so-called “support fatigue” (Emery, 2012, p. 20) within ageing populations may thus be quite real. Moreover, the negative age-interaction for education is not mirrored by a positive one for old-age benefits. The old are not less negatively affected than the young. This illustrates the limitations of a self-interest-based approach to individual welfare attitudes on the basis of age, at least regarding this one policy,

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 47 and it shows how the indications of a verdict against old-age benefits are present within all age-groups. Clearly, other values than self-interest are at work here – notions of a public interest (Lewin, 1991), for example, might perhaps be one of them. What are the main contributions of this thesis – besides having generally contributed to the empirical understanding of a research problem that, until recently, has been quite low? First, we have managed to assess the role of demographic ageing regarding general spending preferences, as well as spending preferences towards specific policies. As we have shown, there are good theoretical and empirical reasons to do both, and our approach has allowed us to approach the issue on the aggregate level, as well as within specific policy areas. Doing the former, we have been able to assess the general relationship between demographic ageing and welfare state support, without having to resort to less valid indicators (like redistribution preferences). Doing the latter, we have been able to corroborate previous findings with new sources of data, as in the case of old-age benefits (Emery, 2012), but also extend beyond them by presenting novel evidence regarding individuals’ support for education policy. We have also been able to show something resembling a deepening intergenerational cleavage in the area of education policy. Previous inquiries have provided little evidence of these sorts of cross-level interactions, but this paper has illustrated that their conclusions may have been premature. Second, we have gone beyond previous cross-country studies by assessing the relationship between the dependency ratio and welfare state attitudes as it has developed within countries over time. The temporal dimension of our design represents a clear step forward, since it has allowed us to apply more data to the research problem, but also to be more certain that we have been able to account for the most important sources of endogeneity. Indeed, by using a fixed-effects within-country estimator, we have been able to remedy a significant weakness of past research: the risk of observable and unobservable between-country bias. While being no experiment, this thesis’ TSCS design represents a notable improvement compared to the previous literature, and as was noted in the beginning, this is (to the author’s knowledge) the first time that a temporal-comparative analysis has been made on this important topic. Third, we have also helped to bridge the literature on the attitudinal effects of demographic ageing with that of economic globalisation (e.g., Busemeyer & Garritzmann, 2019), illustrating how both these structural developments may have similar effects on welfare state attitudes. To be sure, the fact that both globalisation and demographic ageing has now been found to be positively related to spending preferences for education – a key form of social investment – could be instructive for future inquiries into the effects of these sorts of macro-level developments on welfare state attitudes.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 48

Regarding future research, much could certainly be gained from inquiries into the role of welfare institutions. In this thesis, we have used a robust within-country estimator, and considering that institutions do not change considerably over shorter time periods, our model has absorbed much of the variation that would have been needed to properly test for their effects. It is nevertheless well-established that welfare states themselves tend to create their own sorts of welfare politics, and they might therefore have some moderating influence on the relationships that we have uncovered. So, a focus on welfare regimes more generally (e.g., Esping-Andersen, 1990), or different institutional, intergenerational welfare contracts (Birnbaum et al., 2017) – that are either balanced, or biased towards certain age groups – would be a way to add some valuable, institutional complexity to these relationships. The policy feedback literature (e.g., Jordan, 2013) should be a good departure-point for such a study. As is often the case, one could also benefit from conducting some type of experiment. Survey experiments, like that of Naumann (2017), could allow us to assess the causal effects of (information about) demographic ageing, and such an experiment could provide a good triangulation of evidence. The interaction with individual age would be particularly important to pursue, since it has not been featured in previous experiments. Sample sizes and geographical representation may be a disadvantage of such a study, however, since there quickly arises logistical and financial issues with regarding creating original survey experiments in sufficient scale. One may also remember that when presenting the first hypothesis, we referenced the literature on the attitudinal effects of economic crises. Thinking ahead, there are arguably some interesting questions about where one should place demographic ageing in relation to more proximate, short-term crises. Considering the world’s current predicament, for instance, it would be pertinent to assess how the demographic structure of a country moderates the popular response to various short-term crises. A worsening demographic structure may severely limit the de facto opportunities for financially expansive policy responses, and there would surely be many insights to be gained from studies that exploit exogenously given crises, and assess the moderating influence of demographic context on the (potential) changes in public welfare opinion that follow. This would be a very interesting and timely issue to investigate within a longer thesis project, for instance. True enough, demographic ageing is a structural process. But it is one that many still see as an impending crisis – a gathering, grey storm, to be sure. However, and perhaps contrary to intuition, our results may indicate that citizens in ageing populations might be inclined to withdraw their support from elderly-oriented welfare state policies, for which the public demand (as well as the costs that come with it) is bound to increase. Crucially, this is while they seem to be instead favouring policies that – if properly invested in – can come with

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 49 considerable future interest for the economy, and the welfare state. The implication of this is that there may be, at least on the aggregate level, some public legitimacy to the political reforms whose aim is to restructure old-age policies, and that the public veto – that some fear will block future attempts at restructuring these – may not be as strong as one might believe. It is no exaggeration that future generations undoubtedly face some considerable challenges, but this thesis does at least suggest that the generations of today see their future as one that is worth investing in. True, this may be one of the key popular responses to the ageing crisis.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 50

7. References

Alber, J. (1988). Is there a crisis of the welfare state? Crossnational evidence from Europe, North America, and Japan. European Sociological Review, 4(3), 181-205. Alesina, A., Murard, E., & Rapoport, H. (2019). Immigration and preferences for redistribution in Europe (No. w25562). National Bureau of Economic Research. Allan, J. P., & Scruggs, L. (2004). Political partisanship and welfare state reform in advanced industrial societies. American Journal of Political Science, 48(3), 496-512. Angrist, J. D., & Pischke, J. S. (2008). Mostly harmless econometrics: An empiricist's companion. Princeton university press. Armingeon, K., Wenger, W., Wiedemeier, F., Isler, C., Knöpfel, L., Weisstanne, D., & Engler, S. (2020). Comparative Political Data Set 1960-2018. Zurich: Institute of Political Science, University of Zurich. Bartels, M. (1996). ‘Uninformed Votes: Information Effects in Presidential Elections’. American Journal of Political Science 40 (1): 194–230. Bellemare, M. F., Masaki, T., & Pepinsky, T. B. (2017). Lagged explanatory variables and the estimation of causal effect. The Journal of Politics, 79(3), 949-963. Birnbaum, S., Ferrarini, T., Nelson, K., & Palme, J. (2017). The generational welfare contract. Edward Elgar Publishing. Black, D. (1948). On the rationale of group decision-making. Journal of political economy, 56(1), 23-34. Blekesaune, M. (2007). Economic conditions and public attitudes to welfare policies. European Sociological Review, 23(3), 393-403. Blekesaune, M., & Quadagno, J. (2003). Public Attitudes toward Welfare State Policies. A Comparative Analysis of 24 Nations. European sociological review, 19(5), 415-427. Bonoli, G. (2010). The political economy of active labor-market policy. Politics & Society, 38(4), 435-457. Bonoli, G. (2020). Social investment, active labour market policies and migration. In The European Social Model under Pressure (pp. 193-206). Springer VS, Wiesbaden. Bonoli, G., Cantillon, B., & Van Lancker, W. (2017). Social Investment and the Matthew effect. The uses of social investment, 66-76. Brady, D., & Finnigan, R. (2014). Does immigration undermine public support for social policy?. American Sociological Review, 79(1), 17-42. Brady, D., Huber, E., Stephens, J. D. (2020). Comparative Welfare States Data Set. Brambor, T., Clark, W. R., & Golder, M. (2006). Understanding interaction models: Improving empirical analyses. Political analysis, 63-82. Brooks, C., & Manza, J. (2006a). Social policy responsiveness in developed democracies. American Sociological Review, 71(3), 474-494. Brooks, C., & Manza, J. (2006b). Why do welfare states persist?. The Journal of Politics, 68(4), 816-827.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 51

Brooks, C., & Manza, J. (2008). Why welfare states persist: The importance of public opinion in democracies. University of Chicago Press. Burgoon, B. (2001). Globalization and welfare compensation: disentangling the ties that bind. International organization, 55(3), 509-551. Burgoon, B. (2014). Immigration, integration, and support for redistribution in Europe. World Pol., 66, 365. Busemeyer, M. R., & Garritzmann, J. L. (2019). Compensation or social investment? Revisiting the link between globalisation and popular demand for the welfare state. Journal of social policy, 48(3), 427-448. Busemeyer, M. R., Goerres, A., & Weschle, S. (2009). Attitudes towards redistributive spending in an era of demographic ageing: the rival pressures from age and income in 14 OECD countries. Journal of European Social Policy, 19(3), 195-212. Castles, F. G. (2004). The future of the welfare state: Crisis myths and crisis realities. OUP Oxford. Chung, H., Taylor‐Gooby, P., & Leruth, B. (2018). Political legitimacy and welfare state futures: Introduction. Social Policy & Administration, 52(4), 835-846. Clasen, J., & Siegel, N. A. (Eds.). (2007). Investigating welfare state change: the dependent variable problem in comparative analysis. Edward Elgar Publishing. Cortina, J. M. (1993). What is coefficient alpha? An examination of theory and applications. Journal of applied psychology, 78(1), 98. Dahl, R. A. (1989). Democracy and its Critics. Yale university press. Dahlberg, M., Edmark, K., & Lundqvist, H. (2012). Ethnic diversity and preferences for redistribution. Journal of Political Economy, 120(1), 41-76. Dallinger, U. (2010). Public support for redistribution: what explains cross-national differences?. Journal of European Social Policy, 20(4), 333-349. De la Porte, C., & Jacobsson, K. (2011). Social investment or recommodification? Assessing the employment policies of the EU member states. Towards a social investment welfare state? Ideas, policies and challenges, 117-234. De Mello, L., Schotte, S., Tiongson, E. R., & Winkler, H. (2016). Greying the budget: Ageing and preferences over public policies. The World Bank. Deutsch, K. W. (1981). The crisis of the state. Government and Opposition, 331-343. Disney, R. (2007). Population ageing and the size of the welfare state: Is there a puzzle to explain? European Journal of Political Economy, 23(2), 542-553. Downs, A. (1957). An economic theory of political action in a democracy. Journal of political economy, 65(2), 135-150. Eger, M. A. (2010). Even in Sweden: the effect of immigration on support for welfare state spending. European Sociological Review, 26(2), 203-217 Emery, T. (2012). Intergenerational conflict: evidence from Europe. Journal of Population Ageing, 5(1), 7-22. Esping-Andersen, G. (1990). The three worlds of welfare capitalism. Princeton University Press.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 52

Esping-Andersen, G. (Ed.). (1996). Welfare states in transition: National adaptations in global economies. Sage. Esping-Andersen, G., & Sarasa, S. (2002). The generational conflict reconsidered. Journal of European social policy, 12(1), 5-21. Esping-Andersen, G., Gallie, D., Hemerijck, A., & Myles, J. (2002). Why we need a new welfare state. OUP Oxford. Finseraas, H. (2009). Income inequality and demand for redistribution: A multilevel analysis of European public opinion. Scandinavian Political Studies, 32(1), 94-119. Fisman, R., Jakiela, P., & Kariv, S. (2015). How did distributional preferences change during the great recession?. Journal of Public Economics, 128, 84-95. Freeman, G. P. (1986). Migration and the political economy of the welfare state. The Annals of the American Academy of Political and Social Science, 485(1), 51-63. Galasso, V., & Profeta, P. (2007). How does ageing affect the welfare state?. European Journal of Political Economy, 23(2), 554-563. Garritzmann, J. L., Busemeyer, M. R., & Neimanns, E. (2018). Public demand for social investment: new supporting coalitions for welfare state reform in Western Europe?. Journal of European Public Policy, 25(6), 844-861. Gelissen, J. (2000). Popular support for institutionalised solidarity: a comparison between European welfare states. International Journal of Social Welfare, 9(4), 285-300. Gerber, A. S., & Green, D. P. (2008). Field experiments and natural experiments. In The Oxford handbook of political science. Giesselmann, M., & Schmidt-Catran, A. W. (2019). Getting the within estimator of cross- level interactions in multilevel models with pooled cross-sections: Why country dummies (sometimes) do not do the job. Sociological Methodology, 49(1), 190-219. Goerres, A., & Prinzen, K. (2012). Can we improve the measurement of attitudes towards the welfare state? A constructive critique of survey instruments with evidence from focus groups. Social Indicators Research, 109(3), 515-534. Goerres, A., Karlsen, R., & Kumlin, S. (2020). What makes people worry about the welfare state? A three-country experiment. British Journal of Political Science, 50(4), 1519-1537. Gonthier, F. (2017). Parallel publics? Support for income redistribution in times of economic crisis. European Journal of Political Research, 56(1), 92-114. Green-Pedersen, C. (2004). The dependent variable problem within the study of welfare state retrenchment: Defining the problem and looking for solutions. Journal of Comparative Policy Analysis: Research and Practice, 6(1), 3-14. Gusmano, M. K., & Okma, K. G. (2018). Population aging and the sustainability of the welfare state. Hastings Center Report, 48, S57-S61. Hemerijck, A. (2013). Changing welfare states. Oxford University Press. Hemerijck, A. (2018). Social investment as a policy paradigm. Journal of European public policy, 25(6), 810-827.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 53

Hess, M., Naumann, E., & Steinkopf, L. (2017). Population ageing, the intergenerational conflict, and active ageing policies – A multilevel study of 27 European countries. Journal of Population Ageing, 10(1), 11-23. Hofäcker, D. (2015). In line or at odds with active ageing policies? Exploring patterns of retirement preferences in Europe. Ageing and Society, 35(7), 1529. Hünermund, P., & Louw, B. (2020). On the Nuisance of Control Variables in Regression Analysis. arXiv preprint arXiv:2005.10314. Irmen, A., Cömertpay, R., & Litina, A. (2019). Individual Attitudes towards Immigration in Aging Populations. ISSP Research Group (1999). International Social Survey Programme: Role of Government III - ISSP 1996. GESIS Data Archive. Cologne. ZA2900 Data file Version 1.0.0 ISSP Research Group (2008). International Social Survey Programme: Role of Government IV - ISSP 2006. GESIS Data Archive. Cologne. ZA4700 Data file Version 1.0.0 ISSP Research Group (2018). International Social Survey Programme: Role of Government V - ISSP 2016. GESIS Data Archive. Cologne. ZA6900 Data file Version 2.0.0 ISSP Research Group (2020). http://w.issp.org/menu-top/home/ [Retrieved 2020-12-22] Jæger, M. M., & Kvist, J. (2003). Pressures on State Welfare in Post‐industrial Societies: Is More or Less Better?. Social Policy & Administration, 37(6), 555-572. Jensen, C. (2011). Less bad than its reputation: Social spending as a proxy for welfare effort in cross-national studies. Journal of Comparative Policy Analysis: Research and Practice, 13(3), 327-340. Jordan, J. (2013). Policy feedback and support for the welfare state. Journal of European Social Policy, 23(2), 134-148. Kmenta, J. (1990). Elements of Econometrics, 2nd Edn. London: Collier Macmillan. Kohli, M. (2006). Aging and justice. In Handbook of aging and the social sciences (pp. 456-478). Academic Press. Kohli, M. (2015). Generations in aging societies: Inequalities, cleavages, conflicts. In Challenges of Aging (pp. 265-288). Palgrave Macmillan, London. Korpi, W., & Palme, J. (1998). The paradox of redistribution and strategies of equality: Welfare state institutions, inequality, and poverty in the Western countries. American sociological review, 661-687. Korpi, W., & Palme, J. (2003). New politics and class politics in the context of austerity and globalization: Welfare state regress in 18 countries, 1975-95. American Political Science Review, 425-446. Kotlikoff, L. J., & Burns, S. (2005). The coming generational storm: What you need to know about America's economic future. MIT press. Kotlikoff, L. J., & Burns, S. (2012). The clash of generations: saving ourselves, our kids, and our economy. MIT press. Kramer, G. H. (1971). Short-term fluctuations in US voting behavior, 1896-1964. The American Political Science Review, 65(1), 131-143.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 54

Kwon, R., & Curran, M. (2016). Immigration and support for redistributive social policy: Does multiculturalism matter? International Journal of Comparative Sociology, 57(6), 375-400. Larsen, C. A. (2008). The institutional logic of welfare attitudes: How welfare regimes influence public support. Comparative political studies, 41(2), 145-168. Lewis-Beck, C., & Lewis-Beck, M. (2015). Applied regression: An introduction (Vol. 22). Sage publications. Lindbom, A. (2011). Systemskifte?: den nya svenska välfärdspolitiken. Studentlitteratur. Lindgren, K. O., & Vernby, K. (2016). The electoral impact of the financial crisis: Evidence using district-level data. Electoral Studies, 44, 214-224 Lindh, T. (2011). Social investment in the ageing populations of Europe. Towards a social investment welfare state, 261-84. Lindh, T., Malmberg, B., & Palme, J. 2005. Generations at War or Sustainable Social Policy in Ageing Societies? Journal of Political Philosophy 13(4): 470–489. Lindvall, J. (2014). The electoral consequences of two great crises. European Journal of Political Research, 53(4), 747-765. Lindvall, J. (2017). Economic downturns and political competition since the 1870s. The Journal of Politics, 79(4), 1302-1314. Lutz, P. (2020). Welfare states, and immigration policies. In The European Social Model under Pressure (pp. 331-347). Springer VS, Wiesbaden. March, J. G., & Olsen, J. P. (2004). The logic of appropriateness. In The Oxford handbook of political science. Marshall, T. H. (1950). Citizenship and social class (Vol. 11, pp. 28-29). Cambridge: CUP. Mau, S., & Burkhardt, C. (2009). Migration and welfare state solidarity in Western Europe. Journal of European Social Policy, 19(3), 213-229. Meier, V., & Werding, M. (2010). Ageing and the welfare state: securing sustainability. Oxford Review of Economic Policy, 26(4), 655-673. Melo, D. F., & Stockemer, D. (2014). Age and political participation in Germany, France and the UK: A comparative analysis. Comparative european politics, 12(1), 33-53. Meltzer, A. H., & Richard, S. F. (1981). A rational theory of the size of government. Journal of political Economy, 89(5), 914-927. Midgley, J. (1999). Growth, redistribution, and welfare: Toward social investment. Social Service Review, 73(1), 3-21. Morel, N., & Palier, B. (Eds.). (2011). Towards a social investment welfare state?: ideas, policies and challenges. Policy Press. Moulton, B. R. (1986). Random group effects and the precision of regression estimates. Journal of econometrics, 32(3), 385-397. Myles, J. (2002). A new social contract for the elderly. Why we need a new welfare state, 1, 130- 173. Myrdal, A., & Myrdal, G. (1934). Crisis in the population question. Stockholm: Bonniers.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 55

Naumann, E. (2017). Do increasing reform pressures change welfare state attitudes? An experimental study on population ageing, pension reform preferences, political knowledge and ideology. Ageing and Society, 37(2), 266. OECD. (2019). Pensions at a Glance 2019. OECD Pensions at a Glance. Olivera Angulo, J. (2014). Preferences for redistribution after the economic crisis. Economics and Business Letters, 3(3), 137-145. Osborne, J. W. (2012). Psychological effects of the transition to retirement. Canadian Journal of Counselling and Psychotherapy, 46(1), 45-58. Österman, M. (2018). Varieties of education and inequality: how the institutions of education and political economy condition inequality. Socio-Economic Review, 16(1), 113-135. Page, B. I., & Shapiro, R. Y. (1983). Effects of public opinion on policy. The American political science review, 175-190. Pampel, F. C. (1994). Population aging, class context, and age inequality in public spending. American Journal of Sociology, 100(1), 153-195. Pennings, P. (2016). Quantitative data analysis in political science. In Handbook of Research Methods and Applications in Political Science. Edward Elgar Publishing. Pierson, P. (1994). Dismantling the welfare state?: Reagan, Thatcher and the politics of retrenchment. Cambridge University Press. Pierson, P. (1996). The new politics of the welfare state. World politics, 48(2), 143-179. Pierson, P. (2001). The New Politics of the Welfare State. Oxford; New York: Oxford University Press. Prinzen, K. (2014). Intergenerational ambivalence: new perspectives on intergenerational relationships in the German welfare state. Ageing & Society, 34(3), 428-451. Prinzen, K. (2017). The Moral Economy of Intergenerational Redistribution in an Ageing Society: A Qualitative Analysis of Young Adults' Beliefs in the United States. Social Policy & Administration, 51(7), 1267-1286. Razin, A., Sadka, E., Swagel, P. (2002). The ageing population and the size of the welfare state. Journal of Political Economy 110, 900–918. Reed, W. R. (2015). On the practice of lagging variables to avoid simultaneity. Oxford Bulletin of Economics and Statistics, 77(6), 897-905. Rubin, D. B. (1978). Bayesian inference for causal effects: The role of randomization. The Annals of statistics, 34-58. Scott, A. (2018). “The myth of an “ageing society”. Project Syndicate. https://www.project- syndicate.org/onpoint/the-myth-of-the-aging-society-by-andrew-scott-2018- 05?barrier=accesspaylog [Retrieved 2020-12-19] Soroka, S. N., & Wlezien, C. (2010). Degrees of democracy: Politics, public opinion, and policy. Cambridge University Press. Stegmueller, D. (2013). How many countries for multilevel modeling? A comparison of frequentist and Bayesian approaches. American Journal of Political Science, 57(3), 748-761.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 56

Stimson, J. A., MacKuen, M. B., & Erikson, R. S. (1995). Dynamic representation. American political science review, 543-565. Street, D., & Cossman, J. S. (2006). Greatest generation or greedy geezers? Social spending preferences and the elderly. Social problems, 53(1), 75-96. Sumino, T. (2014). Does immigration erode the multicultural welfare state? A cross-national multilevel analysis in 19 OECD member states. Journal of Ethnic and Migration Studies, 40(3), 436-455. Svallfors, S. (2008). The generational contract in Sweden: Age-specific attitudes to age-related policies. Policy & Politics, 36(3), 381-396. Svallfors, S. (Ed.). (2012). Contested welfare states: Welfare attitudes in Europe and beyond. Stanford University Press. Swank, D. (2002). Global capital, political institutions, and policy change in developed welfare states. Cambridge University Press Taylor-Gooby, P. (2002). The silver age of the welfare state: perspectives on resilience. Journal of Social Policy, 31(4), 597-621. Taylor-Gooby, P. (2004). New risks, new welfare: the transformation of the European welfare state. Oxford University Press. Taylor-Gooby, P., Leruth, B., & Chung, H. (Eds.). (2017). After austerity: Welfare state transformation in Europe after the great recession. Oxford University Press. Teorell, J., & Svensson, T. (2007). Att fråga och att svara: samhällsvetenskaplig metod. Liber. Tepe, M., & Vanhuysse, P. (2009). Are aging OECD welfare states on the path to gerontocracy? Evidence from 18 democracies, 1980-2002. Journal of Public Policy, 1-28. Titmuss R. (1958). Essays on the Welfare State, London: Allena and Unwin. Van Oorschot, W. (2000). Who should get what, and why? On deservingness criteria and the conditionality of solidarity among the public. Policy & Politics, 28(1), 33-48. Van Oorschot, W. (2006). Making the difference in social Europe: deservingness perceptions among citizens of European welfare states. Journal of European social policy, 16(1), 23-42. Walter, S. (2010). Globalization and the welfare state: Testing the microfoundations of the compensation hypothesis. International Studies Quarterly, 54(2), 403-426. Wasserstein, R. L., & Lazar, N. A. (2016). The ASA statement on p-values: context, process, and purpose. Wilensky, H. (1975). The Welfare State and Equality. Structural and Ideological Roots of Public Expenditure, Berkeley/Los Angeles/London. Yzerbyt, V. Y., Muller, D., & Judd, C. M. (2004). Adjusting researchers’ approach to adjustment: On the use of covariates when testing interactions. Journal of Experimental Social Psychology, 40(3), 424-431.

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 57

Appendix A: Questionnaire Excerpts and Descriptive Graphs

Figure A1: Excerpt from the ISSP Role of Government 2016 master questionnaire (1) Source: ISSP Research Group (2018)

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 58

Figure A2: Excerpt from the ISSP Role of Government 2016 master questionnaire (2) Source: ISSP Research Group (2018)

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 59

AUL DEN

4

3.5 3

FIN FRA

4

3.5 3

FRG JPN

4

3.5 3

NOR NZL

4

3.5 3

SPA SWE

4

3.5

General spending preferences (index) preferences spending General 3

SWZ UKM

4

3.5 3

1995 2000 2005 2010 2015

USA

4

3.5 3

1995 2000 2005 2010 2015 Year

Figure A3: General spending preferences (index) in sample countries at survey years (1996-2016)

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 60

AUL DEN

4.5

4

3.5 3

FIN FRA

4.5

4

3.5 3

FRG JPN

4.5

4

3.5 3

NOR NZL

4.5

4

3.5 3

SPA SWE

4.5

4

3.5

Spending preferences for health care for health preferences Spending 3

SWZ UKM

4.5

4

3.5 3

1995 2000 2005 2010 2015

USA

4.5

4

3.5 3

1995 2000 2005 2010 2015 Year

Figure A4: Spending preferences for health care in sample countries at survey years (1996-2016)

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 61

AUL DEN

4.5

4 3.5

FIN FRA

4.5

4 3.5

FRG JPN

4.5

4 3.5

NOR NZL

4.5

4 3.5

SPA SWE

4.5

4

Spending preferences for education preferences Spending 3.5

SWZ UKM

4.5

4 3.5 1995 2000 2005 2010 2015

USA

4.5

4 3.5 1995 2000 2005 2010 2015 Year

Figure A5: Spending preferences for education in sample countries at survey years (1996-2016)

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 62

AUL DEN

4

3.5

3

2.5 2

FIN FRA

4

3.5

3

2.5 2

FRG JPN

4

3.5

3

2.5 2

NOR NZL

4

3.5

3

2.5 2

SPA SWE

4

3.5

3

2.5 2

Spending preferences for unemployment benefits for unemployment preferences Spending SWZ UKM

4

3.5

3

2.5 2

1995 2000 2005 2010 2015

USA

4

3.5

3

2.5 2

1995 2000 2005 2010 2015 Year

Figure A6: Spending preferences for unemployment benefits in sample countries at survey years (1996- 2016)

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 63

AUL DEN

4

3.8

3.6

3.4 3.2

FIN FRA

4

3.8

3.6

3.4 3.2

FRG JPN

4

3.8

3.6

3.4 3.2

NOR NZL

4

3.8

3.6

3.4 3.2

SPA SWE

4

3.8

3.6

3.4

3.2 Spending preferences for old-age benefits for old-age preferences Spending

SWZ UKM

4

3.8

3.6

3.4 3.2 1995 2000 2005 2010 2015

USA

4

3.8

3.6

3.4 3.2 1995 2000 2005 2010 2015 Year

Figure A7: Spending preferences for old-age benefits in sample countries at survey years (1996-2016)

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 64

AUL DEN

50

40

30

20 10

FIN FRA

50

40

30

20 10

FRG JPN

50

40

30

20 10

NOR NZL

50

40

30

20 10

SPA SWE

Dependency ratio (%)ratio Dependency

50

40

30

20 10

SWZ UKM

50

40

30

20 10

1960 1980 2000 2020

USA

50

40

30

20 10

1960 1980 2000 2020 Year

Figure A8: Dependency ratio in sample countries (separated), 1960-2016

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 65

Appendix B: Results for Supplementary Variables

Table B1: Effect of dependency ratio on spending preferences for health care (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio -.009 -.016** -.018 -.011 -.01 -.009 -.011 (.006) (.007) (.01) (.007) (.007) (.007) (.009) Log GDP per capita .302 .014 -.11 -.112 -.112 (.177) (.182) (.212) (.21) (.207) Immigration -.007 .021 .022 .022 .016 (.008) (.017) (.016) (.016) (.015) Unemployment rate .009 .003 .002 .002 -.003 (.014) (.006) (.007) (.007) (.007) Life expectancy .009 .047 .056 .056 .086 (.027) (.063) (.06) (.06) (.056) Total fertility rate .151 .732*** .702*** .705*** .545*** (.337) (.142) (.133) (.132) (.154) Infant mortality -.031 .121*** .156*** .156*** .203*** (.048) (.032) (.032) (.031) (.035) Individual level

Woman .133*** .133*** .134*** (.016) (.016) (.015) Middle age .002 .021 1.605 (.022) (.059) (.973) Old age -.021 .053 5.87** (.029) (.108) (2.067) Years of education -.017*** -.017*** -.017*** (.005) (.005) (.004) DR × Middle age -.001 -.001 (.002) (.004) DR × Old age -.003 .005 (.004) (.008) Constant 4.182*** 4.391*** .373 -1.993 -1.575 -1.587 -3.722 (.195) (.112) (2.045) (4.464) (4.287) (4.307) (4.186) Observations 55393 55393 54151 54151 49246 49246 49246 R-squared .003 .085 .04 .094 .111 .111 .113 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 66

Table B2: The effect of the dependency ratio on spending preferences for unemployment benefits (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio .008 -.015** 0 .003 -.001 0 -.007 (.01) (.006) (.011) (.008) (.007) (.007) (.007) Log GDP per capita .041 -.282 -.692*** -.697*** -.723*** (.244) (.232) (.172) (.174) (.16) Immigration -.008 .075*** .067*** .066*** .058** (.008) (.019) (.02) (.02) (.021) Unemployment rate .01 .003 .006 .006 .006 (.014) (.009) (.009) (.009) (.01) Life expectancy .02 .06 .096 .096 .142* (.033) (.097) (.079) (.079) (.079) Total fertility rate -.83** -.045 -.197 -.192 -.396 (.317) (.255) (.234) (.232) (.248) Infant mortality .078 .141** .225*** .226*** .282*** (.084) (.046) (.05) (.05) (.05) Individual level

Woman .108*** .108*** .109*** (.013) (.013) (.013) Middle age .055* .056 4.027*** (.026) (.088) (1.272) Old age .088* .223 3.083* (.043) (.134) (1.688) Years of education -.019*** -.02*** -.02*** (.006) (.006) (.006) DR × Middle age 0 .007 (.004) (.005) DR × Old age -.005 .013** (.004) (.006) Constant 2.852*** 2.94*** 2.104 -1.575 -.106 -.105 -3.042 (.298) (.078) (3.526) (7.022) (5.672) (5.638) (5.673) Observations 54051 54051 52836 52836 48144 48144 48144 R-squared .002 .12 .068 .126 .132 .132 .135 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 67

Table B3: Descriptive statistics for supplementary variables Variable Obs. Mean Std. Dev. Min Max Gov. responsibility for health care 55369 4.561 .628 1 4 Gov. responsibility to support 54374 3.238 .761 1 4 disadvantaged students Gov. responsibility for the old 55424 3.517 .625 1 4

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 68

Table B4: The effect of the dependency ratio on support for government’s responsibility for health care (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio -.01 -.016*** -.016* -.023** -.02** -.022** -.017** (.009) (.004) (.008) (.008) (.007) (.008) (.007) Log GDP per capita .174 -.236 -.106 -.11 -.068 (.112) (.172) (.184) (.184) (.185) Immigration -.002 -.021 -.015 -.015 -.011 (.005) (.019) (.017) (.017) (.017) Unemployment rate .02** -.007 -.009 -.009 -.013* (.008) (.006) (.007) (.007) (.006) Life expectancy -.043 .088 .084 .085 .082 (.028) (.061) (.064) (.064) (.062) Total fertility rate .207 -.108 -.059 -.059 -.197 (.165) (.167) (.157) (.157) (.149) Infant mortality -.105* .064 .047 .047 .077 (.057) (.044) (.052) (.052) (.054) Individual level

Woman .064*** .064*** .065*** (.007) (.007) (.007) Middle age -.004 -.088 -.024 (.023) (.09) (.908) Old age -.006 -.051 1.827 (.028) (.112) (1.214) Years of education -.006* -.006* -.006* (.003) (.003) (.003) DR × Middle age .003 -.005 (.003) (.003) DR × Old age .002 -.002 (.004) (.004) Constant 4.811*** 4.82*** 6.522*** .829 -.411 -.347 -.623 (.249) (.059) (2.11) (3.463) (4.363) (4.379) (4.117) Observations 55356 55356 54106 54106 49101 49101 49101 R-squared .007 .113 .057 .111 .111 .111 .113 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 69

Table B5: The effect of the dependency ratio on support for government’s responsibility to support disadvantaged students (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio -.009 .007* 0 .012** .011* .009 .021*** (.008) (.003) (.009) (.004) (.005) (.006) (.005) Log GDP per capita .269* -.645*** -.735*** -.736*** -.596** (.144) (.175) (.24) (.24) (.24) Immigration .006 .031** .031* .031* .038** (.008) (.014) (.014) (.014) (.016) Unemployment rate .034** 0 0 0 -.01 (.012) (.006) (.007) (.008) (.009) Life expectancy .006 .08 .091 .091 .083 (.034) (.086) (.082) (.083) (.086) Total fertility rate -.137 -.073 -.118 -.12 -.234 (.265) (.181) (.207) (.208) (.23) Infant mortality .055 .112*** .129*** .129*** .166*** (.034) (.029) (.035) (.035) (.045) Individual level

Woman .06*** .06*** .062*** (.013) (.013) (.013) Middle age -.042 -.099 1.615* (.026) (.112) (.744) Old age -.019 -.095 2.205 (.038) (.159) (1.701) Years of education -.003 -.003 -.002 (.003) (.003) (.003) DR × Middle age .002 -.012*** (.004) (.003) DR × Old age .003 -.016*** (.005) (.005) Constant 3.469*** 3.038*** -.391 2.143 2.193 2.254 1.134 (.186) (.05) (2.118) (6.163) (6.214) (6.293) (6.396) Observations 54361 54361 53137 53137 48242 48242 48242 R-squared .004 .099 .067 .103 .088 .088 .091 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 70

Table B6: The effect of the dependency ratio on support for government’s responsibility for the old (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio -.008 -.016*** -.016** -.02** -.021*** -.022*** -.017** (.009) (.004) (.007) (.007) (.006) (.006) (.007) Log GDP per capita .097 .099 .013 .012 .083 (.113) (.132) (.201) (.202) (.198) Immigration -.005 -.016 -.017 -.017 -.015 (.005) (.015) (.013) (.013) (.014) Unemployment rate .025*** 0 0 0 -.007 (.008) (.005) (.006) (.006) (.006) Life expectancy -.013 .011 .022 .022 .026 (.025) (.051) (.047) (.047) (.044) Total fertility rate .173 .034 -.001 -.001 -.153 (.182) (.137) (.139) (.138) (.15) Infant mortality -.063 .016 .043 .043 .077 (.047) (.035) (.051) (.051) (.055) Individual level

Woman .093*** .093*** .095*** (.013) (.013) (.013) Middle age .049** .033 .884 (.019) (.086) (.836) Old age .07** .05 3.305** (.032) (.121) (1.459) Years of education -.013*** -.013*** -.013*** (.003) (.003) (.003) DR × Middle age .001 -.005* (.003) (.003) DR × Old age .001 -.005 (.004) (.005) Constant 3.716*** 3.7*** 3.782* 2.074 2.041 2.057 1.049 (.245) (.067) (1.913) (3.443) (3.368) (3.402) (3.146) Observations 55411 55411 54164 54164 49154 49154 49154 R-squared .005 .096 .05 .094 .101 .101 .103 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 71

Appendix C: Results using Continuous Age Variable

Table C1: Effect of dependency ratio on general spending preferences (cont. age variable) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio 0 -.009 -.003 .004 .003 .003 .008 (.005) (.007) (.009) (.008) (.008) (.01) (.012) Log GDP per capita .253 -.095 -.172 -.171 -.171 (.144) (.14) (.251) (.251) (.23) Immigration -.003 .05*** .045*** .045*** .041*** (.004) (.013) (.013) (.013) (.013) Unemployment rate .006 .001 0 0 .001 (.009) (.005) (.007) (.007) (.006) Life expectancy .018 .014 .05 .05 .096* (.017) (.06) (.056) (.056) (.051) Total fertility rate -.324 .411** .344* .344* .168 (.239) (.17) (.173) (.173) (.207) Infant mortality .044 .133*** .173*** .173*** .286*** (.032) (.034) (.044) (.044) (.053) Individual level

Woman .104*** .104*** .105*** (.008) (.008) (.008) Years of education -.016*** -.016*** -.016*** (.003) (.003) (.003) DR × Age 0 0 (0) (0) Constant 3.646*** 3.713*** -.06 .77 -1.189 -1.2 -5.169 (.135) (.128) (1.679) (4.291) (4.82) (4.89) (4.657) Observations 52300 52300 51121 51121 45645 45645 45645 R-squared 0 .08 .038 .09 .109 .109 .111 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 72

Table C2: Effect of dependency ratio on spending preferences for education (cont. age variable)

(1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio -.001 .017 .017 .036** .037** .034* .058*** (.009) (.015) (.011) (.015) (.016) (.017) (.015) Log GDP per capita .555*** -.15 .099 .097 .115 (.091) (.247) (.551) (.553) (.52) Immigration .012** .074*** .07** .07** .079*** (.005) (.022) (.023) (.023) (.021) Unemployment rate -.003 .002 -.002 -.001 .001 (.004) (.008) (.009) (.009) (.009) Life expectancy -.004 .032 .054 .054 .033 (.021) (.083) (.092) (.092) (.088) Total fertility rate -.462*** .365 .376 .373 .524 (.121) (.317) (.345) (.349) (.386) Infant mortality .098** .148** .137** .137** .213** (.034) (.054) (.062) (.062) (.076) Individual level

Woman .064*** .064*** .063*** (.016) (.016) (.017) Years of education .013** .013** .013*** (.004) (.004) (.004) DR × Age 0 0*** (0) (0) Constant 3.895*** 3.604*** -1.874 -.781 -5.089 -4.974 -4.88 (.222) (.286) (1.592) (6.164) (8.639) (8.773) (8.584) Observations 54974 54974 53740 53740 47884 47884 47884 R-squared 0 .06 .043 .067 .08 .08 .082 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 73

Table C3: Effect of dependency ratio on spending preferences for old-age benefits (cont. age variable) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio .001 -.019** -.011 -.011 -.013 -.021 -.003 (.006) (.007) (.009) (.01) (.008) (.016) (.024) Log GDP per capita .107 .055 -.059 -.062 .088 (.153) (.203) (.354) (.358) (.344) Immigration -.008* .027 .02 .02 .029 (.004) (.022) (.016) (.016) (.021) Unemployment rate .005 -.004 -.007 -.007 -.013 (.01) (.004) (.006) (.006) (.008) Life expectancy .046** -.073 -.022 -.022 .011 (.017) (.05) (.038) (.039) (.048) Total fertility rate -.142 .534** .45** .443** -.019 (.248) (.183) (.153) (.161) (.195) Infant mortality .03 .13 .183 .182 .344** (.042) (.08) (.103) (.104) (.114) Individual level

Woman .113*** .113*** .116*** (.01) (.011) (.011) Years of education -.038*** -.037*** -.037*** (.005) (.005) (.005) DR × Age 0 0 (0) (0) Constant 3.661*** 3.892*** -.689 6.633 4.019 4.243 -.476 (.148) (.105) (1.75) (4.226) (3.465) (3.517) (3.731) Observations 54721 54721 53498 53498 47614 47614 47614 R-squared 0 .039 .012 .045 .091 .091 .094 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 74

.08

.06

.04

.02

Marginaleffectsof dependency ratio

0 -.02

15 24 33 42 51 61 70 79 88 97 Age

Figure C1: Marginal effects of dependency on support for spending on education (cont. age variable)

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 75

Appendix D: Results within Income Sub-samples

Table D1: Effect of dependency ratio on general spending preferences (low-income) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio 0 -.007 0 .004 .003 .002 .008 (.006) (.007) (.009) (.007) (.007) (.007) (.007) Log GDP per capita .222 -.176 -.276 -.278 -.265 (.13) (.153) (.226) (.229) (.227) Immigration .002 .049*** .048*** .048*** .049*** (.004) (.013) (.012) (.012) (.011) Unemployment rate .009 .003 .002 .002 -.002 (.008) (.005) (.006) (.006) (.005) Life expectancy -.005 .037 .053 .053 .063 (.018) (.055) (.047) (.047) (.043) Total fertility rate -.281 .427** .353** .353** .211 (.22) (.154) (.158) (.159) (.176) Infant mortality .012 .105*** .138*** .139*** .196*** (.032) (.032) (.042) (.042) (.038) Individual level

Woman .058*** .058*** .06*** (.012) (.012) (.012) Middle age .077*** .032 2.229* (.022) (.105) (1.116) Old age .05 .027 2.007 (.028) (.09) (1.521) Years of education -.009** -.009** -.009** (.003) (.003) (.003) DR × Middle age .002 -.007** (.004) (.003) DR × Old age .001 -.008 (.003) (.006) Constant 3.698*** 3.758*** 2.09 -.013 -.233 -.193 -1.246 (.146) (.123) (1.738) (4.116) (3.777) (3.832) (3.892) Observations 24027 24027 23313 23313 21168 21168 21168 R-squared 0 .072 .032 .08 .091 .091 .094 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 76

Table D2: Effect of dependency ratio on general spending preferences (high-income) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio .001 -.013* -.006 -.001 -.001 0 .003 (.005) (.006) (.01) (.007) (.008) (.008) (.01) Log GDP per capita .23 -.151 -.169 -.162 -.107 (.142) (.152) (.234) (.238) (.214) Immigration -.005 .046*** .045*** .045*** .043*** (.005) (.012) (.013) (.013) (.012) Unemployment rate .006 0 -.003 -.003 -.004 (.01) (.006) (.007) (.007) (.009) Life expectancy .039* -.003 .014 .013 .037 (.02) (.051) (.048) (.048) (.038) Total fertility rate -.357 .458** .4** .397** .251 (.245) (.158) (.179) (.179) (.184) Infant mortality .058* .153*** .173*** .171*** .235*** (.03) (.031) (.034) (.034) (.035) Individual level

Woman .105*** .105*** .105*** (.01) (.01) (.011) Middle age .042* .091 3.02** (.02) (.075) (1.056) Old age .054** -.004 3.313** (.024) (.073) (1.326) Years of education -.013*** -.013*** -.013*** (.003) (.003) (.003) DR × Middle age -.002 -.007 (.002) (.006) DR × Old age .002 -.003 (.003) (.008) Constant 3.52*** 3.675*** -1.466 2.6 1.456 1.456 -1.084 (.124) (.112) (1.756) (3.83) (3.808) (3.855) (3.315) Observations 19877 19877 19412 19412 18331 18331 18331 R-squared 0 .089 .042 .101 .11 .11 .112 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 77

Table D3: Effect of dependency ratio on spending preferences for education (low-income) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio -.002 .02 .018 .038** .039** .04** .057*** (.009) (.016) (.012) (.015) (.016) (.016) (.015) Log GDP per capita .542*** -.135 .026 .029 .043 (.091) (.217) (.481) (.482) (.48) Immigration .013** .07*** .073*** .073*** .08*** (.006) (.022) (.021) (.021) (.02) Unemployment rate -.001 .002 -.001 -.001 -.002 (.003) (.008) (.01) (.01) (.011) Life expectancy -.005 .071 .05 .05 .014 (.026) (.078) (.08) (.08) (.079) Total fertility rate -.399*** .437 .432 .434 .467 (.113) (.304) (.349) (.351) (.374) Infant mortality .098** .13** .108* .108* .142** (.038) (.05) (.054) (.055) (.064) Individual level

Woman .064*** .064*** .064*** (.021) (.021) (.02) Middle age -.055** -.013 -.579 (.024) (.133) (1.132) Old age -.124*** -.081 -.951 (.032) (.146) (1.27) Years of education .013*** .013*** .014*** (.003) (.003) (.003) DR × Middle age -.002 -.023*** (.004) (.004) DR × Old age -.002 -.031*** (.005) (.005) Constant 3.922*** 3.576*** -1.737 -3.941 -4.096 -4.161 -2.167 (.234) (.301) (1.794) (5.73) (7.294) (7.373) (7.394) Observations 25353 25353 24603 24603 22285 22285 22285 R-squared 0 .063 .042 .07 .083 .083 .086 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 78

Table D4: Effect of dependency ratio on spending preferences for education (high-income) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio .002 .016 .017 .034** .038** .038** .052** (.008) (.014) (.012) (.016) (.017) (.017) (.017) Log GDP per capita .548*** -.265 .003 .016 .061 (.102) (.343) (.591) (.592) (.562) Immigration .012* .071*** .075*** .075*** .078*** (.006) (.021) (.023) (.023) (.02) Unemployment rate -.003 -.005 -.008 -.008 -.008 (.006) (.009) (.01) (.01) (.013) Life expectancy .001 .038 .017 .015 .018 (.022) (.081) (.094) (.094) (.086) Total fertility rate -.552*** .242 .313 .306 .344 (.156) (.332) (.397) (.401) (.417) Infant mortality .097** .181** .134 .131 .195** (.037) (.06) (.081) (.08) (.087) Individual level

Woman .06** .06** .061** (.024) (.024) (.024) Middle age -.002 .032 1.871 (.027) (.099) (1.983) Old age -.049 -.219** 2.18 (.039) (.095) (2.113) Years of education .012** .012** .013** (.005) (.005) (.005) DR × Middle age -.001 -.019** (.003) (.007) DR × Old age .006* -.019* (.003) (.009) Constant 3.843*** 3.599*** -2.025 .048 -1.323 -1.264 -2.685 (.212) (.278) (1.753) (6.009) (7.456) (7.541) (7.138) Observations 20555 20555 20071 20071 18942 18942 18942 R-squared 0 .066 .044 .072 .078 .078 .08 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covaraite interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 79

Table D5: Effect of dependency ratio on spending preferences for old-age benefits (low-income) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio .002 -.02*** -.01 -.012 -.011 -.017 -.004 (.007) (.006) (.01) (.009) (.008) (.011) (.015) Log GDP per capita .049 -.116 -.159 -.17 -.134 (.14) (.194) (.326) (.335) (.349) Immigration -.004 .033 .033 .032 .038* (.004) (.023) (.019) (.019) (.019) Unemployment rate .009 -.002 -.005 -.005 -.009 (.009) (.005) (.006) (.006) (.007) Life expectancy .026 -.059 -.026 -.027 -.026 (.017) (.059) (.048) (.05) (.051) Total fertility rate -.078 .614*** .568*** .551*** .254 (.23) (.179) (.158) (.17) (.173) Infant mortality -.015 .119 .147 .15 .24** (.039) (.073) (.111) (.111) (.109) Individual level

Woman .059*** .057*** .061*** (.016) (.017) (.017) Middle age .18*** .085 2.662 (.041) (.167) (1.892) Old age .228*** -.032 3.021 (.06) (.194) (2.365) Years of education -.028*** -.028*** -.026*** (.006) (.006) (.006) DR × Middle age .004 -.012 (.007) (.011) DR × Old age .01 -.013 (.008) (.017) Constant 3.719*** 3.986*** 1.624 7.224 5.067 5.397 4.681 (.169) (.1) (1.798) (5.148) (4.035) (4.116) (3.959) Observations 25274 25274 24525 24525 22153 22153 22153 R-squared 0 .033 .009 .04 .074 .074 .078 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 80

Table D6: Effect of dependency ratio on spending preferences old-age benefits (high-income) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio .001 -.026*** -.014 -.021** -.023*** -.027** -.02 (.005) (.006) (.01) (.007) (.007) (.011) (.014) Log GDP per capita .109 .017 -.11 -.091 .022 (.151) (.165) (.251) (.266) (.279) Immigration -.01** .016 .013 .013 .018 (.005) (.016) (.015) (.015) (.018) Unemployment rate .003 -.005 -.009* -.009 -.011 (.01) (.004) (.005) (.005) (.007) Life expectancy .065*** -.08** -.042 -.046 -.034 (.018) (.033) (.029) (.03) (.032) Total fertility rate -.247 .569** .467** .452** .204 (.252) (.193) (.16) (.17) (.168) Infant mortality .053 .152** .199** .193** .253*** (.042) (.057) (.069) (.071) (.08) Individual level

Woman .109*** .109*** .109*** (.015) (.015) (.015) Middle age .069** .049 2.128 (.031) (.135) (1.318) Old age .109* -.283 2.316 (.054) (.185) (1.539) Years of education -.035*** -.035*** -.035*** (.004) (.004) (.004) DR × Middle age .001 -.006 (.005) (.009) DR × Old age .014* .003 (.007) (.011) Constant 3.533*** 3.874*** -2.134 7.725** 6.367** 6.571** 4.358* (.135) (.104) (1.768) (2.831) (2.348) (2.363) (2.004) Observations 20406 20406 19932 19932 18803 18803 18803 R-squared 0 .045 .016 .053 .083 .084 .087 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 81

Appendix E: Results with Control for Linear Time-trend

Table E1: Effect of dependency ratio on general spending preferences (linear time-trend control) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio 0 -.014 -.003 .02 .047 .047 .049 (.005) (.018) (.009) (.04) (.034) (.035) (.035) Log GDP per capita .253 .448 .936* .94* .955* (.144) (.603) (.499) (.499) (.52) Immigration -.003 .05 .083*** .083*** .081** (.004) (.036) (.027) (.027) (.028) Unemployment rate .006 .002 -.047** -.048** -.048** (.009) (.006) (.016) (.016) (.018) Life expectancy .018 .026 -.009 -.01 .011 (.017) (.138) (.097) (.097) (.099) Total fertility rate -.324 .672 .595 .59 .447 (.239) (.613) (.444) (.444) (.467) Infant mortality .044 .224*** .181*** .179*** .244*** (.032) (.063) (.051) (.05) (.062) Individual level

Woman .104*** .104*** .105*** (.008) (.008) (.008) Middle age .026 .027 2.098** (.019) (.086) (.81) Old age .032 -.017 3.587** (.027) (.108) (1.312) Years of education -.015*** -.015*** -.015*** (.003) (.003) (.003) DR × Middle age 0 -.005 (.003) (.003) DR × Old age .002 -.002 (.004) (.005) Constant 3.646*** -7.64 -.06 -10.041 43.395* 43.84* 38.366 (.135) (9.552) (1.679) (31.375) (24.028) (23.814) (25.415) Observations 52300 52300 51121 51121 46664 46664 46664 R-squared 0 .093 .038 .097 .115 .115 .117 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 82

Table E2: Effect of dependency ratio on spending preferences for education (linear time-trend control) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio -.001 .002 .017 .039 .046 .044 .059 (.009) (.021) (.011) (.035) (.041) (.041) (.041) Log GDP per capita .555*** .171 .284 .294 .332 (.091) (.516) (.591) (.591) (.599) Immigration .012** .061* .066* .066* .07** (.005) (.03) (.03) (.031) (.031) Unemployment rate -.003 -.003 -.006 -.007 -.002 (.004) (.005) (.019) (.019) (.019) Life expectancy -.004 .281** .267** .265** .252* (.021) (.113) (.118) (.118) (.119) Total fertility rate -.462*** .147 .188 .175 .267 (.121) (.529) (.555) (.556) (.581) Infant mortality .098** .237*** .239*** .235*** .29*** (.034) (.06) (.062) (.061) (.07) Individual level

Woman .06*** .06*** .06*** (.018) (.018) (.018) Middle age -.043 -.052 .331 (.026) (.133) (1.145) Old age -.114*** -.23 .319 (.035) (.148) (1.22) Years of education .012*** .012*** .013*** (.004) (.004) (.004) DR × Middle age 0 -.019*** (.004) (.004) DR × Old age .004 -.021*** (.005) (.005) Constant 3.895*** -1.617 -1.874 90.752** 95.865*** 97*** 91.901*** * (.222) (11.206) (1.592) (21.382) (25.751) (25.479) (26.518) Observations 54974 54974 53740 53740 48927 48927 48927 R-squared 0 .074 .043 .078 .089 .089 .091 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 83

Table E3: Effect of dependency ratio on spending preferences for old-age benefits (linear time-trend control) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio .001 -.043** -.011 -.018 .001 -.004 0 (.006) (.015) (.009) (.017) (.02) (.02) (.021) Log GDP per capita .107 .697** .998*** 1.02*** 1.023*** (.153) (.241) (.317) (.317) (.293) Immigration -.008* .025* .041* .042* .045* (.004) (.013) (.021) (.021) (.021) Unemployment rate .005 -.004 -.037** -.038** -.036** (.01) (.004) (.016) (.016) (.015) Life expectancy .046** -.157** -.175** -.179** -.162** (.017) (.057) (.06) (.06) (.062) Total fertility rate -.142 .501 .458* .426* .145 (.248) (.32) (.234) (.234) (.255) Infant mortality .03 .146* .112** .102** .182*** (.042) (.078) (.042) (.042) (.039) Individual level

Woman .112*** .112*** .114*** (.01) (.01) (.011) Middle age .094** .048 1.617 (.034) (.161) (1.346) Old age .173*** -.123 3.984* (.05) (.208) (1.852) Years of education -.037*** -.037*** -.036*** (.005) (.005) (.005) DR × Middle age .002 -.008 (.007) (.011) DR × Old age .012 -.003 (.008) (.016) Constant 3.661*** - -.689 - -58.525* -55.6* -65.274** 32.325*** 88.639** * (.148) (7.849) (1.75) (19.262) (27.746) (28.092) (26.158) Observations 54721 54721 53498 53498 48650 48650 48650 R-squared 0 .05 .012 .052 .094 .095 .097 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 84

Appendix F: Results with Lagged Dependency Ratio (5 years)

Table F1: Effect of dependency ratio on general spending preferences (5-year lag) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio (t-5) .004 .002 .009 .009 .01 .008 .013 (.008) (.014) (.013) (.01) (.01) (.01) (.01) Log GDP per capita .287* -.118 -.226 -.226 -.229 (.146) (.135) (.201) (.204) (.194) Immigration 0 .056*** .055*** .055*** .053*** (.004) (.014) (.013) (.013) (.013) Unemployment rate .006 .003 .003 .003 .001 (.009) (.006) (.007) (.007) (.008) Life expectancy .006 .002 .015 .015 .038 (.018) (.062) (.055) (.055) (.05) Total fertility rate -.33 .36** .311* .309* .16 (.236) (.155) (.153) (.154) (.163) Infant mortality .057* .133*** .166*** .166*** .226*** (.031) (.03) (.035) (.035) (.039) Individual level

Woman .105*** .105*** .105*** (.008) (.008) (.008) Middle age .026 0 2.26** (.019) (.088) (.798) Old age .03 -.032 3.673** (.027) (.113) (1.264) Years of education -.015*** -.015*** -.015*** (.003) (.003) (.003) DR × Middle age .001 -.005* (.003) (.003) DR × Old age .003 -.002 (.004) (.005) Constant 3.546*** 3.528*** .207 1.836 1.861 1.899 -.02 (.169) (.233) (1.651) (4.46) (4.482) (4.549) (4.229) Observations 52313 52313 51133 51133 46675 46675 46675 R-squared .001 .079 .04 .09 .107 .107 .109 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 85

Table F2: Effect of dependency ratio on spending preferences for education (5-year lag) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio (t-5) .006 .028 .031* .044** .046** .042** .053*** (.014) (.018) (.016) (.016) (.016) (.015) (.012) Log GDP per capita .562*** -.165 -.064 -.063 -.071 (.084) (.231) (.425) (.427) (.422) Immigration .013* .075*** .076*** .076*** .078*** (.006) (.021) (.021) (.021) (.02) Unemployment rate -.003 .01 .01 .01 .011 (.004) (.01) (.01) (.01) (.011) Life expectancy -.005 -.002 -.015 -.016 -.012 (.021) (.092) (.093) (.092) (.088) Total fertility rate -.492*** .123 .133 .127 .164 (.106) (.212) (.227) (.232) (.253) Infant mortality .116*** .116* .103 .103 .15 (.03) (.06) (.07) (.069) (.084) Individual level

Woman .062*** .062*** .062*** (.017) (.017) (.018) Middle age -.042 -.11 1.09 (.026) (.167) (1.396) Old age -.116*** -.299 1.319 (.035) (.19) (1.536) Years of education .013*** .013*** .013*** (.004) (.004) (.004) DR × Middle age .003 -.011* (.006) (.005) DR × Old age .008 -.011 (.007) (.008) Constant 3.741*** 3.443*** -2.204 2.433 2.243 2.339 1.604 (.318) (.331) (1.321) (6.403) (6.982) (7.116) (6.612) Observations 54987 54987 53752 53752 48938 48938 48938 R-squared .001 .063 .049 .07 .08 .08 .081 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 86

Table F3: Effect of dependency ratio on spending preferences for old-age benefits (5-year lag) (1) (2) (3) (4) (5) (6) (7)

Country level

Dependency ratio (t-5) .007 -.006 .003 -.004 -.004 -.006 .007 (.008) (.015) (.014) (.014) (.013) (.017) (.02) Log GDP per capita .155 .006 -.093 -.086 -.05 (.158) (.22) (.374) (.383) (.385) Immigration -.004 .039 .037* .038* .044* (.003) (.025) (.021) (.021) (.022) Unemployment rate .005 -.005 -.007 -.007 -.011 (.009) (.005) (.008) (.008) (.009) Life expectancy .028 -.081 -.058 -.059 -.047 (.017) (.054) (.041) (.042) (.042) Total fertility rate -.142 .555** .513** .506** .226 (.255) (.212) (.192) (.2) (.195) Infant mortality .043 .149* .186* .184* .268** (.047) (.072) (.099) (.1) (.101) Individual level

Woman .113*** .113*** .115*** (.01) (.01) (.011) Middle age .095** .09 1.867 (.034) (.181) (1.304) Old age .173*** -.079 4.172* (.05) (.272) (1.925) Years of education -.036*** -.036*** -.036*** (.005) (.005) (.005) DR × Middle age 0 -.015 (.008) (.012) DR × Old age .011 -.011 (.012) (.018) Constant 3.538*** 3.651*** -.164 7.273 6.605 6.666 5.07 (.187) (.256) (1.878) (4.745) (4.203) (4.292) (3.954) Observations 54734 54734 53510 53510 48661 48661 48661 R-squared .001 .036 .011 .044 .086 .087 .09 Year FE No Yes No Yes Yes Yes Yes Country FE No Yes No Yes Yes Yes Yes Covariate interactions No No No No No No Yes Clustered standard errors in parentheses. *** p<.01, ** p<.05, * p<.1

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 87

Appendix G: Stata Syntax

***Main multivariate regressions*** reg [variable] dr100 if (year==1996|year==2006|year==2016), cluster(country) reg [variable] dr100 i.country i.year if (year==1996|year==2006|year==2016), /// , cluster(country) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// if (year==1996|year==2006|year==2016), cluster(country) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// i.country i.year if (year==1996|year==2006|year==2016), cluster(country) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// gender i.agegroups educyrs i.country i.year /// if (year==1996|year==2006|year==2016), cluster(country) reg [variable] c.dr100##(i.agegroups) log_rgdpecap immigration unemr lifexp /// tfr infmort gender educyrs i.country i.year if /// (year==1996|year==2006|year==2016), cluster(country) reg [variable] c.dr100##(i.agegroups) i.agegroups##(c.log_rgdpecap /// c.immigration c.unemr c.lifexp c.tfr c.infmort) gender educyrs i.country /// i.year if (year==1996|year==2006|year==2016), cluster(country)

***Sub-sample analyses*** reg [variable] dr100 if (year==1996|year==2016|year==2016) & income==[0/1], cluster(country) reg [variable] dr100 i.country i.year if (year==1996|year==2016|year==2016) /// & income==[0/1], cluster(country) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// if (year==1996|year==2016|year==2016) & income==[0/1], cluster(country) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// i.country i.year if (year==1996|year==2016|year==2016) & income==[0/1], /// cluster(country) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// gender i.agegroups educyrs i.country i.year if /// (year==1996|year==2016|year==2016) & income==[0/1], cluster(country) reg [variable] c.dr100##(i.agegroups) log_rgdpecap immigration unemr lifexp /// tfr infmort gender educyrs i.country i.year if /// (year==1996|year==2006|year==2016), cluster(country) reg [variable] c.dr100##(i.agegroups) i.agegroups##(c.log_rgdpecap /// c.immigration c.unemr c.lifexp c.tfr c.infmort) gender educyrs i.country /// i.year if (year==1996|year==2016|year==2016) & income==[0/1], cluster(country)

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 88

***Linear time-trend control*** reg [variable] dr100 if (year==1996|year==2016|year==2016) & income==[0/1], cluster(country) reg [variable] dr100 i.country##c.year if (year==1996|year==2016|year==2016) /// & income==[0/1], cluster(country) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// if (year==1996|year==2016|year==2016) & income==[0/1], cluster(country) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// i.country##c.year if (year==1996|year==2016|year==2016) & income==[0/1], /// cluster(country) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// gender i.agegroups educyrs i.country##c.year if /// (year==1996|year==2016|year==2016) & income==[0/1], cluster(country) reg [variable] c.dr100##(i.agegroups) log_rgdpecap immigration unemr lifexp /// tfr infmort gender educyrs i.country##c.year if /// (year==1996|year==2006|year==2016), cluster(country) reg [variable] c.dr100##(i.agegroups) i.agegroups##(c.log_rgdpecap /// c.immigration c.unemr c.lifexp c.tfr c.infmort) gender educyrs i.country /// i.year if (year==1996|year==2016|year==2016) & income==[0/1], cluster(country)

***Lagged DR*** reg [variable] dr100_m5 if (year==1996|year==2016|year==2016) & income==[0/1], cluster(country) reg [variable] dr100_m5 i.country##c.year if (year==1996|year==2016|year==2016) /// & income==[0/1], cluster(country) reg [variable] dr100_m5 log_rgdpecap immigration unemr lifexp tfr infmort /// if (year==1996|year==2016|year==2016) & income==[0/1], cluster(country) reg [variable] dr100_m5 log_rgdpecap immigration unemr lifexp tfr infmort /// i.country##c.year if (year==1996|year==2016|year==2016) & income==[0/1], /// cluster(country) reg [variable] dr100_m5 log_rgdpecap immigration unemr lifexp tfr infmort /// gender i.agegroups educyrs i.country##c.year if /// (year==1996|year==2016|year==2016) & income==[0/1], cluster(country) reg [variable] c.dr100_m5##(i.agegroups) log_rgdpecap immigration unemr /// lifexp tfr infmort gender educyrs i.country##c.year if /// (year==1996|year==2006|year==2016), cluster(country) reg [variable] c.dr100_m5##(i.agegroups) i.agegroups##(c.log_rgdpecap /// c.immigration c.unemr c.lifexp c.tfr c.infmort) gender educyrs i.country /// i.year if (year==1996|year==2016|year==2016) & income==[0/1], cluster(country)

***Alternative option for robust standard errors***

Oskar Pettersson ∙ The Popular Response to the Ageing Crisis 89 reg [variable] dr100 if (year==1996|year==2016|year==2016) & income==[0/1], vce(robust) reg [variable] dr100 i.country##c.year if (year==1996|year==2016|year==2016) /// & income==[0/1], vce(robust) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// if (year==1996|year==2016|year==2016) & income==[0/1], vce(robust) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// i.country##c.year if (year==1996|year==2016|year==2016) & income==[0/1], /// vce(robust) reg [variable] dr100 log_rgdpecap immigration unemr lifexp tfr infmort /// gender i.agegroups educyrs i.country##c.year if /// (year==1996|year==2016|year==2016) & income==[0/1], vce(robust) reg [variable] c.dr100##(i.agegroups) log_rgdpecap immigration unemr lifexp /// tfr infmort gender educyrs i.country##c.year if /// (year==1996|year==2006|year==2016), vce(robust) reg [variable] c.dr100##(i.agegroups) i.agegroups##(c.log_rgdpecap /// c.immigration c.unemr c.lifexp c.tfr c.infmort) gender educyrs i.country /// i.year if (year==1996|year==2016|year==2016) & income==[0/1], vce(robust)