What a social investment ‘litmus test’ must address: A response to Plavgo and Hemerijck

Abstract

Throughout the past decade, the notion that social policy should be applied as a tool to enhance the productive capacity of families, rather than to smooth consumption, has gained considerable traction under the moniker of social investment (SI). Against lingering critiques of the usefulness of SI research, Plavgo and Hemerijck primarily present bivariate, cross-sectional correlations of national-level outcomes to provide what they call a ‘social investment litmus test’. They conclude that SI policies are effective at achieving ‘positive returns in employment and poverty mitigation’. They also present evidence to suggest that ‘passive’ income supports are not effective at reducing poverty and promoting employment. Given the importance of their claims and the pre-eminence of SI in recent social policy literature, we feel compelled to respond and clarify that the authors’ findings do not support their conclusions. In doing so, we (1) review the central claims from the authors’ study, (2) provide evidence as to why the authors’ findings do not support their claims of a ‘social investment litmus test’, and (3) propose a path forward for a more critical assessment of the performance of policies labelled as SI.

Keywords

Employment poverty social investment social policy welfare states

Introduction

The authors’ claims are profoundly important: after 10 years of critiques and questions regarding the conceptual usefulness and empirical validity of SI research, the authors now offer a ‘litmus test’ – defined as a decisively indicative test – to dispel the primary criticisms of SI research.² Of these criticisms, the authors mainly take aim at scholars identifying a Matthew effect in SI policies. The authors put forward the central argument of the article that SI policies cannot be simply understood as redistributive cash-transfers, and that ‘tracking household disposable income before and after SI “transfers” cannot effectively capture the mid- to long-term effects of SI policy synergies’ (p. 2). And so, they set out to redress the ‘academic impasse by gauging the socioeconomic effect of SI policies in terms of employment inclusion and poverty mitigation in the medium term at a societal level’ (p. 2). Given the importance of their claims and the pre-eminence of SI in recent social policy literature, we feel compelled to respond and clarify that the authors’ findings do not support their conclusions.

In doing so, we (1) review the central claims from the authors’ study, (2) provide evidence as to why the authors’ findings do not support their claims of a ‘social investment litmus test’ and (3) propose a path forward for a more critical assessment of the performance of policies labelled as SI.

Points of agreement

We first note our points of agreement with PH and highlight the key findings from their study. Like PH, we agree that a critical function of social policy research is to investigate the effects of public policies on a range of socio-economic outcomes, such as employment and poverty. Similarly, we agree with PH, and a vast body of empirical literature, that reducing children’s exposure to poverty and hardship is not only critical from moral and normative perspectives, but also from a long-term economic perspective: children who grow up in poverty are more likely to face health and learning challenges, and are more likely to struggle in the labour market and live in poverty in adulthood (see Van Lancker and Vinck, 2020 for an overview of studies). Thus, we applaud PH for attempting to provide empirical clarity on whether SI policies do, indeed, generate favourable poverty and employment outcomes, and whether such results hold when evaluating policies that fall outside the SI umbrella.

In their attempt, PH seek specifically to answer whether SI policies ‘explain cross-country variation in work- and welfare-related life outcomes for families with children’ (p. 2). To do so, they operationalize Hemerijck’s (2015, 2017) conceptual framework of stocks, flows and buffers as the three-legged stool underpinning the idea of SI. Stock policies refer to those that promote the development of ‘human capital stock’, typically education or skill-development programmes. PH operationalize ‘stock’ as participation in and spending on early childhood education and care services. Flow policies ‘facilitat[e] life course and labour-market transitions’. PH operationalize ‘flows’ as paid-leave availability and flexible working arrangements. Buffer policies ‘make sure that individuals and families do not fall between the cracks of the economy when social and/or personal misfortune strikes’. PH operationalize ‘buffers’ as public spending on family benefits, social assistance and the replacement rate of paid leave benefits.

The authors primarily present a set of bivariate scatterplots demonstrating the correlation of these country-level indicators with country-level parental employment rates and child poverty rates. In their online appendix, they also present results while controlling for GDP per capita and spending on cash benefits as a percentage of GDP. The results show, for the most part, that countries with lower child poverty rates and higher parental employment rates tend to have higher childcare participation rates, more spending on childcare, and more spending on family benefits. As a result of these positive, cross-sectional associations, the authors conclude that SI is, indeed, effective at achieving long-term gains in employment and reductions in poverty. PH also present evidence that suggests that ‘passive’ income supports, which they operationalize primarily as spending on pensions and unemployment benefits, are not effective at reducing poverty and promoting employment.

Conceptual and empirical shortcomings of social investment and its litmus test

We argue that the set of evidence that PH provide does not support their conclusions and does not offer a ‘litmus test’ on the performance of policies labelled as SI. We discuss the conceptual and empirical shortcomings of their analysis, focusing on the following three questions: (1) Do the authors’ findings support their conclusions? (2) What is social investment and what is not? (3) Are stocks, buffers and flows conceptually and empirically distinguishable? We then conclude with steps toward what we believe offer a more critical assessment of the long-run effects of social policies in general and SI in particular.

Do the authors’ findings support their conclusions?

First, PH claim that ‘stock, flows and buffer policies concur positive returns in employment and poverty mitigation across OECD countries’ (p. 13). This conclusion is primarily derived from a series of bivariate scatter plots of country-level correlations, plus multivariate analyses adding two controls. However, correlational observations are clearly not suitable for evaluating the performance of a given policy intervention, even if these correlations are adjusted for GDP per capita or social spending. Moreover, aggregated, country-level poverty and employment outcomes in a single year are likewise insufficient for evaluating the ‘returns’ of a policy intervention on the individuals exposed to the policy. In turn, this study’s use of cross-national correlations cannot claim to provide a ‘litmus test’ on the comparative performance of SI policies.

Our conclusion is not novel. We agree with Hemerijck (2017) who notes that ‘Aggregate outcome indicators are not particularly well suited for the testing of actual policy performance’ (p. 18). This is in line with a similar observation from Nolan (2017) who writes that ‘studies based on such aggregate indicators/evidence face major difficulties in establishing robust statistical correlations, and even more in ascribing them to underlying causal mechanisms’ (p. 46).

If country-level indicators are to be used, examining how within-country variation in policy performance relates to within-country variation in poverty or employment is clearly a more empirically-useful approach. It should be noted that PH fail to mention articles that have adopted such an approach and that model the longer-term impact of SI spending on poverty outcomes using more advanced estimation techniques (Bakker and Van Vliet, 2019; Noël, 2020; Taylor-Gooby et al., 2015; Van Vliet and Wang, 2015). These studies generally find a modest yet positive relationship between changes in SI policies over time and employment outcomes, and a more complicated relationship with poverty outcomes. Van Vliet and Wang (2015), for instance, point to the different trajectories of the Nordic welfare states and conclude that ‘there might be a positive relationship between expenditure shifts towards new welfare state programmes and stagnating or even increasing poverty trends’ (p. 634).

In contrast, PH offer cross-sectional, cross-national correlations to provide their litmus test. That PH lag many of their country-level spending values does not remove the threats of omitted variable bias, reverse causality and ecological fallacies.³ Put simply, the authors’ methodological framework is not one that is set up to provide empirical clarity on the ‘returns’ to policies labelled as SI.

Second, PH claim that ‘the identified effects of SI policies’ (higher employment and lower poverty) ‘do not exist when passive social protection policies are considered’ (p. 9). Specifically, they write: ‘To test the plausibility of our identification assumptions, we verified if the effects of SI policies that we identified do not exist when we substitute them with alternate policies. SI policies were replaced with passive social protection programmes and the overall social cash benefits. Findings show that public spending on pensions and unemployment have a weak and statistically non-significant association with parental employment and child poverty rates’ (pp. 11–12).

Put differently, PH compare policies labelled as SI with ‘passive social protection programmes’ in their effects on parental employment and child poverty. This comparison builds on the supposed contrast between investment-oriented policies and ‘passive’ social protection programmes that are meant to smooth consumption rather than reap economic returns (we address the flaws in this distinction below).

The authors’ attempt to demonstrate that ‘the effects of SI policies that we identified do not exist’ when examining ‘passive’ social protection programmes, however, is misguided and misleading. PH compare the association of spending on pensions and unemployment benefits (passive benefits) with parental employment and child poverty rates (the desired SI outcomes). They also examine the association of total social benefits in cash as a proxy to the overall welfare state generosity. However, there is no compelling reason to believe that more spending on pension benefits, which are targeted at non-working retirees, should generally have an effect on employment rates or child poverty (the lone exception being pension spending in countries in which multigenerational households substitute for developed income protection systems, e.g. Diris et al., 2017). Finding a weak relationship between pension spending and employment/poverty outcomes simply signifies that pensions are targeted at individuals who are generally out of work and not living in a household with children.

Additionally, spending on unemployment benefits is endogenous to the level of unemployment in the country (and unemployment tends to be negatively associated with the country’s employment rate); when more people are unemployed, there is greater demand for unemployment benefits and total spending on unemployment benefits tends to increase. Of course, an increase in unemployment also tends to be correlated with an increase in poverty rates. It is no surprise then that PH find a weak association of spending on unemployment benefits with employment and poverty rates. A negative or weakly positive relationship between spending on unemployment benefits and the employment rate is not an indictment of ‘passive’ spending; it is the correlation we should generally expect.⁴

Third, PH conclude that ‘a swath of prior research has identified such Matthew effects when looking at the short-term SI policy return at the individual level’ (p. 13) and claim that ‘when extending the scope of this research to medium-term returns, positive employment- and wellbeing-related outcomes at the societal level are observed’ (p. 13). We are afraid that such a conclusion is unwarranted and arises from a misunderstanding of what the Matthew effect in social policy entails. The Matthew effect in social policy refers to the observation that middle- and higher-income groups in society tend to benefit more from public spending than do lower income groups. This can be problematic because such unequal distribution of resources risks igniting a feedback loop of cumulative (dis)advantage, in particular when the objective of policies is to benefit the most disadvantaged (Pavolini and Van Lancker, 2018).

Consider the example of ECEC services. Such services are promoted under the SI umbrella to help achieve social inclusion through the labour market by allowing parents of young children to engage in paid employment (increasing the flow). Being enrolled in high-quality ECEC services is also beneficial for children in terms of cognitive and non-cognitive development (increasing the stock), which will in turn increase later labour market opportunities. Both dimensions should be particularly helpful for children from disadvantaged backgrounds, who are most often living in jobless households, where they stand to gain the most from ECEC services in terms of school readiness. This obviously means that disadvantaged children should be enrolled in ECEC services in the first place. Unfortunately, the empirical evidence is clear: across many high-income countries, poor children are much less likely to participate in childcare compared with middle-class and higher-income children, and there has been little progress over time in reducing these inequalities in childcare use despite increasing overall enrolment rates (Blossfeld et al., 2017; OECD, 2016; Van Lancker, 2018). Studies exploiting ECEC expansions as natural experiments show that newly created places indeed tend to first benefit households higher up the income distribution (Bettendorf et al., 2015; Havnes and Mogstad, 2011; Lefebvre et al., 2009). When inequalities in ECEC are substantial, the aggregate returns of public investment in these services will be lower.

The fact that child poverty rates are lower and employment rates are higher in countries with higher spending levels on stock and flow policies do not counter these observations. In fact, previous research has shown that rises in individual employment did not necessarily translate into a decrease of household joblessness (Corluy and Vandenbroucke, 2014; de Beer, 2007). In other words, the returns to increasing employment were heterogeneous, benefiting households in which adult members were already connected to the labour market. As such, it is perfectly possible for mean employment rates to rise and for the most disadvantaged to not benefit from this rise. Matthew effects are about heterogeneous returns to policy investment. To appropriately study the existence of Matthew effects, one needs to exploit individual-level data (ideally life-course data, as we discuss below). To measure Matthew effects at the country level, one would at least need to go beyond the mean and study returns across the income distribution. Though PH claim that their findings ‘question. . .the Matthew effect critique that social investment reform disproportionately benefits better-off households at the expense of poor families’ (p. 13), their use of country-level data does not, in fact, provide any compelling evidence on the Matthew effect.

What is social investment and what is not?

A central tenet of the present article and the broader SI literature is that some social policies can be considered as ‘investment’ whereas other policies cannot. However, we argue that the distinction between investment-oriented and consumption-oriented policies remains flawed. We are not the first to point this out (Nolan, 2013, 2017). Nolan (2017) writes, for example, ‘with a definition of “investment” broad enough to include anything that might facilitate higher labour force participation or contribute (directly or indirectly) to the health and productive capacity of the workforce, what is it legitimate to exclude?’ (pp. 45–46).

Indeed, the authors seem to acknowledge the conceptual fuzziness in their own operationalization of SI policies. To give one clear example: PH include spending on child benefits as an SI policy. This spending measure constructed by the OECD includes spending on cash benefits, such as child allowances, spending on tax breaks and spending on services such as ECEC. For the majority of OECD countries, child allowances do the heavy lifting here.⁵ Child allowances (partially) account for the high costs of raising children; moreover, they are known to reduce child poverty which, in turn, contributes to more favourable long-run outcomes for children (Bradshaw, 2002; Cooper and Stewart, 2020; Immervoll et al., 2001; Van Lancker and Van Mechelen, 2015). Put differently, child allowances enable greater consumption, which tends to lead to the healthier development of children. For a jobless mother receiving unemployment insurance, the underlying mechanisms are nearly identical: the mother receives unemployment benefits to provide for her family while searching for a new opportunity. The benefits enable greater consumption which, in turn, contribute to more favourable child outcomes (at least compared to the counterfactual). Under the authors’ rubric, however, the first of these policies is considered investment while the second is not. In that regard, it is odd that social assistance benefits are considered to be social investment buffers by PH. Social assistance benefits are usually residual, means-tested benefits of last resort, not buffers automatically kicking in ‘when social and/or personal misfortune strikes’ (p. 4); that is exactly the role of social insurance policies, including unemployment benefits, which are derided as ‘passive’ expenditures.⁶

Such conceptual obfuscation is also evident from earlier attempts to classify SI policies. Häusermann and Palier (2017), for instance, in Hemerijck’s 2017 edited volume on SI, regard child allowance expansions as ‘purely consumption-oriented income transfers that pursue socially conservative, much more than “buffer”-functions’ (p. 346). De Deken (2017), in the same volume, classifies family benefits as an ‘ambiguous category’ (p. 191). Moreover, earlier attempts to flesh out social investment spending are not mentioned by PH, for instance Nikolai (2012) who compares compensatory versus investment policies, Vandenbroucke and Vleminckx (2011) who compare old forms of traditional social spending to new forms of spending, or De Deken (2013) who tries to outline the ‘skeleton’ of SI by means of OECD spending categories. Family allowances are classified differently in these different approaches; at the least, theory should be presented as to why child allowances, or family benefits in general, are regarded purely as an investment policy.

Adding child allowances under the SI umbrella also disregards the long history of child allowances as a key part of the traditional social protection system, long predating the social investment ‘turn’ in European and other welfare states. As Gauthier (1998) argues, child allowances have historically come to fruition in the 1930s and served the purpose of compensating for the cost of childrearing and to enhance the wellbeing of families. Over time, however, child benefit policies have been implemented in pursuit of different, often contradictory, objectives such as increasing fertility, reducing poverty, or discouraging maternal employment. In the past decades, government spending on child allowances has generally been declining across many EU countries while public spending on ECEC services rose, often substantially (Adema et al., 2020; Van Lancker and Ghysels, 2014). This demands that the following question is asked: If child allowances always have been investment policies, what is new about social investment? And if spending on child allowances is on the decline while spending on childcare is increasing, is this a turn toward or away from social investment?

Again, we are not the first to question the conceptual ambiguities of social investment. In fact, a fruitful academic debate has emerged on the question of to what extent a social investment turn can be distinguished empirically, and if so, whether this has crowded out spending on traditional social protection programmes. This debate is mentioned in passing by PH but not further discussed: ‘whether flow and stock policies crowd-in or crowd-out social protection buffers and for whom, we maintain, is a matter of methodologically perceptive empirical research’ (p. 5, our emphasis). It is unclear what is meant by the adjective ‘perceptive’, but the literature already converged that there is not much evidence for a crowding out of spending on income protection by spending on childcare or activation policies (Cantillon, 2011; De Deken, 2013; Kuitto, 2016; Noël, 2020; Vandenbroucke and Vleminckx, 2011), and that the determinants of these policies are by and large the same as determinants for traditional social protection programmes (Bonoli and Reber, 2010; Ferragina and Seeleib-Kaiser, 2015).

Few scholars concerned with socio-economic inequalities would argue against more spending on child allowances, expanding access to affordable childcare, and a strong set of minimum income protections for those who fall through the cracks of the labour market and welfare state. But classifying these policies as investment, while classifying other income supports as non-investment (such as the ‘alternate policies’ to SI that PH label as ‘passive social protection programmes’, pp. 11–12), is conceptually flawed; in turn, the framing of SI as a unique and useful policy paradigm is likewise tenuous.

Are stocks, buffers and flows conceptually and empirically distinguishable?

Underpinning recent conceptualizations of SI is the three-legged stool of ‘stocks, buffers, and flows’. Hemerijck (2017) notes that these three sub-classifications of SI policies interact in mutually-beneficial ways. Likewise, PH point to institutional complementarities among ECEC policies, paid leaves and income supports. PH claim that their study is ‘the first social science effort to examine stock, flows and buffer policy efforts on employment and redistribution’ (p. 13). They follow by noting that their ‘analyses of stock, flows and buffer policies concur positive returns in employment and poverty mitigation’ (p. 13).

That said, we question whether the distinction of these policies as stocks, buffers and flows is useful, either conceptually or empirically. To their credit, PH largely agree with the concerns of conceptual overlap. They write that ‘there is considerable overlap between the policy functions of stocks, flows and buffers’ (p. 4). Consider, as one example, that childcare participation rates are measured as a stock policy while childcare can easily be regarded a flow policy as well, given its importance in facilitating household employment (see also De Deken, 2017).

Given the recognized conceptual overlap, what is then the rationale for including one set of policies in the category of stock and the other in the flow category?

The confusion is apparent in the way the three dimensions are operationalized. We give two examples. First, the ‘flow’ measure ‘paid leave duration’ includes the duration of paid maternity, parental and home care leave available to mothers and fathers, measured in weeks. This one indicator includes three different sets of policies with different objectives and, as shown in previous work, with different outcomes (Ferragina, 2020; Thévenon and Solaz, 2013). For instance, maternity benefits are long-standing policies with the purpose of protecting maternal health, while home care leaves (or cash-for-care schemes) actually disincentivize parents, in reality mothers, to return to the labour market. It should be discussed why including these different policies into one indicator is useful. While it might be difficult to find proper measures to operationalize the concept of stocks, flows and buffers, at least these choices should be made more explicit.

Second, the ‘buffer’ measure ‘public spending on family benefits’ not only includes spending on child allowances but also spending on childcare services, which is defined as a stock measure. Spending on childcare services is also included as a separate ‘stock’ measure. Similarly, the buffer measure ‘replacement rate of paid leave benefits’ is for obvious reasons strongly related to the flow measure of ‘paid leave duration’.

It is then no surprise that the conceptual overlap leads to empirical overlap. The national-level buffer policy of the paid leave replacement rate and flow policy of duration of paid leave are correlated with a strength of r = 0.82. Childcare participation (stock) and social protection expenditures (buffer) are r = 0.67, and the latter is also tightly connected with the flow policy of flexible employment policies (r = 0.68) and the stock policy of childcare expenditures (r = 0.66). Put simply, stocks, buffers and flows are, conceptually and empirically, difficult to distinguish (Table 1).

Table 1.

Correlation table of stock, flow and buffer indicators included in PH.

		Stock			Flow		Buffer
		(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Stock	(1) ECEC enrolment, 0–2	1
	(2) ECEC enrolment, 3–5	0.53	1
	(3) Public expenditure on ECEC services	0.64	0.47	1
Flow	(4) Flexible working time	0.72	0.33	0.68	1
Flow	(5) Duration of paid leave	−0.30	0.07	−0.02	−0.07	1
Buffer	(6) Expenditure on housing and social exclusion	0.67	0.41	0.66	0.68	−0.22	1
	(7) Expenditure on family benefits and services	0.36	0.48	0.58	0.59	0.17	0.70	1
	(8) Paid leave replacement rates	−0.38	0.08	−0.12	−0.16	0.82	−0.45	0.03	1

Source: Indicators included in PH.

While it is true that there is overlap between stocks, flows and buffers, we do not understand how such operationalization helps further our understanding of the complementarities and outcomes of SI policies. For example, if spending on family benefits is associated with lower poverty rates (assuming for the moment an association is sufficient to examine this question, see below), is it due to the share of spending allocated for childcare services, due to the direct income effect of child cash benefits, or is there a role for tax credits as well? And if the underlying idea of complementarities is that buffer policies ‘back up the other functions in an interconnected fashion’ (p. 4), then the overlap in the operationalization of the indicators is ill-advised. If parents are not entitled to or do not make use of paid leave (flow), they will not benefit from paid leave replacement rates (buffer) either. Simply aggregating all sorts of different policy measures into one indicator is not useful to reliably gauge the effect and complementarities of social investment policies.

Beyond the conceptual and empirical overlap, we have concern with the theory underlying their policy classifications. Consider one of their stock indicators, the ‘percent of children enrolled in ECEC services or primary school, age 3 to 5’. Across most countries in PH’s study, children aged 3–5 are in preschool and in the education system. In countries such as France, Belgium or Spain, hardly prime examples of ‘social investment states’, enrolment of preschool reaches almost 100%, which has a longstanding history completely unrelated from the concept of social investment. In their primary analyses, PH do not present the correlation between the share of 3–5-year-olds enrolled in ECEC with employment or poverty, though they note in Footnote 3 that this indicator ‘has a weaker strength of association’. Indeed, our own analyses based on the PH data show that the indicator is not significantly related to employment among partnered mothers (Adj. R² = 0.03), and both ECEC spending (Adj. R² = −0.03) and the share of 3–5-year-olds in ECEC (Adj. R² = −0.03) indicators are not associated with the single parent employment outcomes.

Moreover, interpretations of cross-national correlations of stock, buffer and flow policies with employment and poverty outcomes should be made cautiously. In PH’s analyses, they find that not one of the stock, flow or buffer measures is significantly or substantially associated with employment or poverty outcomes of single parents. Indeed, PH acknowledge this and note that ‘lone-parent families need additional support beyond SI policies’ (p. 13). Yet following the logic of their analyses, our conclusion would be that social investment does not yield returns for single parents. Likewise, their regression models (Table C1 in their appendix) using the same data show that cash spending is significantly and substantially associated with lower poverty rates among single parents. So, traditional social protection spending appears most effective in shielding single parents from poverty, according to their results.

Taken at face value, the findings that stocks, flows and buffers yield positive returns for couple families but not for single parents would mean that the returns are selective, and do not help the most disadvantaged. This is exactly what the critics of social investment returns mean when they examine Matthew effects. In a way, PH have proven the critics’ point. Naturally, such a conclusion would be unwarranted. The conclusion is that the fuzzy concept of stocks, flows and buffers, the overlapping and undertheorized operationalization of the variables, and the methods used do not allow for such far-reaching conclusions to be drawn.

Toward a more critical assessment of the effects of social policies

Despite our conceptual and empirical concerns with PH’s study, we believe that the overarching questions this and other SI-focused studies ask are useful and important. How do we ensure the sustainability of the welfare state? How can social policies adapt to better serve the changing demographic and employment situations of families with children? How can we understand cross-national differences in employment and inequality? We do not believe recent SI research has been remarkably useful in answering these questions. Instead, we propose three pathways for a more fruitful social policy research agenda moving forward.

First, we propose that the SI and active labour market policy literatures, in particular, would benefit from greater use of individual-level panel data to investigate their theorized mechanisms. Indeed, much of the SI literature, in particular, is focused on ‘life-course multipliers’ and individual ‘returns’ to policy interventions. Yet in their SI litmus test PH opt for a cross-national, cross-sectional view of social outcomes as opposed to within-household or within-family, and the SI literature more broadly rarely measures such effects using life-course data. Our argument is not that comparative cross-sectional research designs are inherently flawed; with well-grounded theory, careful operationalization of key indicators and cautious interpretation of results, cross-sectional correlations can meaningfully inform the social policy literature. However, such analyses are not designed to offer a ‘litmus test’ on the efficacy of a given policy paradigm on national socio-economic outcomes.

In the pursuit of demonstrating ‘returns’ to policy investments, a life-course perspective would make for a more fruitful analytical approach (e.g. Kvist, 2015). To their credit, PH acknowledge this fact at the end of their study, concluding that future analyses should investigate ‘micro-level wellbeing in a longitudinal perspective’ (p. 13). Moreover, Hemerijck et al. (2016) have conducted micro-level analyses on the basis of the panel-component of EU-SILC to examine the relationship between SI and poverty and employment in a report commissioned by the European Commission. The results show, inter alia, that a year-on-year change in ECEC spending is associated with a higher poverty probability, in particular among women (Hemerijck et al., 2016: 61), while it is associated with a higher probability to be employed for women (Hemerijck et al., 2016: 53). Applying a longitudinal perspective thus leads to different interpretations of the returns to SI. Still, PH offer their SI ‘litmus test’ in the absence of longitudinal or life-course analysis.

Second, we propose such analyses of social policy interventions would continue to benefit from investigations of heterogeneous effects across different demographic and socio-economic groups, as well as the contextual effects that moderate the effectiveness of a given intervention. As noted already in reference to Matthew effects, examining the country-level mean of a given socio-economic outcome is not an appropriate test of whether ‘social investment reform disproportionately benefits better-off households at the expense of poor families’ (p. 13). Instead, investigating the distributional effects inherent within policy implementations (e.g. are poorer families less likely to participate in childcare?), rather than assuming them away, would contribute more to the body of social policy knowledge.

The same applies to the investigation of contextual effects. As Deaton (2020) writes, ‘Nothing works except in context, and finding out what works where and under what circumstances is a real scientific endeavour.’ Rather than analysing the correlation of national childcare spending and employment rates of single mothers, for example, we propose that it would be more fruitful to investigate the conditions under which an increase in childcare spending yields positive returns for single mothers, and under which conditions it does not. Likewise, analyses of institutional complementarities, on which the logic of SI is based, require an understanding of the policy and economic context in which a given policy is implemented.

We emphasize that our proposed pathways toward a more useful assessment of the returns to social policies are not new or uniquely ours. Indeed, PH themselves acknowledge the need to ‘uncover the concrete (joint) effects of SI policies’ (p. 13) and ‘address the channels through which stock, flow, and buffer policies affect. . . wellbeing’ (p. 13). Nonetheless, it is unfortunate that PH advance their SI ‘litmus test’ without putting these insights into practice.

Our third and broader concern, however, emphasizes that resolving empirical and methodological challenges can only go so far if built on a weak conceptual foundation. We thus propose that the SI literature should move away from a focus on the conceptual repackaging of different groups of policies and move toward a stronger focus on the theory, mechanisms and values underlying different policy instruments.

As noted before, applying a common theoretical framework to the relationship between a diverse set of policy indicators and employment/poverty outcomes is bound to lead to conceptual fuzziness. Moreover, such an approach inevitably skips over a potentially fruitful investigation of the mechanisms linking the given policies to the social and economic outcomes of interest. Consider that the SI framework includes everything from child allowances to paid family leave to enrolment in preschool. Despite the clear heterogeneity in the policies labelled as SI, PH nonetheless attempt to interpret the performance of the SI framework as if it were a homogenous bundle of interventions. This is clearly not the case. Instead, there is greater value in acknowledging that child allowances, paid family leave and preschool enrolment are likely to have different, sometimes conflicting types and magnitudes of effects on our social indicators of interest (poverty, employment and so on).⁷

That is not to say that policies should always be analysed in isolation. PH are right to note the existence of institutional complementarities. And we agree that contextual effects that may moderate the strength of a given policy on a social outcome should be investigated. Do child allowances have smaller labour supply effects when affordable childcare options are more accessible? Is childcare participation more widespread, and more egalitarian, when a country invests more into public-sector job creation? These are empirical questions that could help reveal the particularities of institutional complementarities, yet still get lost when packaging many heterogeneous policies under a broad policy label, as PH do in their SI ‘litmus test’.

The values underlying the SI framework are also worth greater inspection and discussion. If policy investments are meant to produce ‘positive returns in terms of economic growth, employment opportunities, and (child) poverty mitigation’ (Hemerijck, 2017: 24), then one can naturally wonder what this calculus means for domestic care work compared to formal employment, or the social value of protecting the old and/or infirm relative to the economic returns of reducing child poverty. On the point of care work, several scholars have already pointed out the ‘gender blindness’ of the SI framework (Cantillon and Van Lancker, 2013; Daly, 2011; Jenson, 2009; Saraceno, 2015). As Saraceno (2015) has argued forcefully, ‘the citizen envisaged by the social investment approach is first and foremost a paid worker, either in actuality or (when a child) in the making’ (p. 257). Given the implicit values underpinning SI, it tends to overlook gender inequalities and power relations within and across households.

Similarly, investments into the older-age population and individuals with disabilities are less likely to yield higher employment rates or lower child poverty rates (the key indicators in PH’s SI ‘litmus test’), but should that imply that we give lesser priority to such investments? How do we account for the social value of protecting the elderly, even if our investments reap little financial reward? What weight do we place on values such as justice, equality and the minimization of hardship, compared to the pursuit of returns to our policy investments? These are questions that cannot be resolved with data, yet our uses and interpretation of data often assume that these questions are resolved or not worth confronting.

Put simply, regardless of data or methods employed, we propose that the discipline should continue to give particular attention to ensuring (1) that policy indicators used in social policy analyses are theoretically grounded and properly operationalized and (2) that scholars, especially those with potential to influence public policy decisions, avoid over-interpreting empirical findings that are based on cross-sectional, cross-national correlations.

In sum, our views align closely with those of PH’s study: we should continue to investigate the effects of social policies on socio-economic outcomes. We should continue to investigate how different types of policies affect employment, poverty and other social and economic outcomes of interest. However, we disagree as to whether the present study offers a decisively indicative analysis as to whether policies branded as social investment offer a clear path to higher employment and less poverty.

Footnotes

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Adema

Clarke

Thevenon

(2020) ‘Family Policies and Family Outcomes in OECD Countries’, in Nieuwenhuis

Van Lancker

(eds) The Palgrave Handbook of Family Policy, pp.193–217. Cham: Palgrave MacMillan.

Bakker

V.B.

Van Vliet

(2019) ‘Social Investment, Employment Outcomes and Policy and Institutional Complementarities: A Comparative Analysis Across 26 OECD Countries’, Department of Economics Research Memorandum, Leiden University, Institute of Public Administration, Leiden.

Bettendorf

L.J.

Jongen

E.L.

Muller

(2015) ‘Childcare Subsidies and Labour Supply: Evidence from a Large Dutch Reform’, Labour Economics 36: 112–23.

Blossfeld

H.-P.

Kulic

Skopek

Triventi

(eds) (2017) Childcare, Early Education and Social Inequality: An International Perspective. Cheltenham: Edward Elgar Publishing.

Bonoli

Reber

(2010) ‘The Political Economy of Childcare in OECD Countries: Explaining Cross-national Variation in Spending and Coverage Rates’, European Journal of Political Research 49(1): 97–118.

Bradshaw

(2002) ‘Child Poverty and Child Outcomes’, Children and Society 16(2): 131–40.

Cantillon

(2011) ‘The Paradox of the Social Investment State: Growth, Employment and Poverty in the Lisbon Era’, Journal of European Social Policy 21(5): 432–49.

Cantillon

Van Lancker

(2013) ‘Three Shortcomings of the Social Investment Perspective’, Social Policy and Society 12(4): 553–64.

Cooper

Stewart

(2020) ‘Does Household Income Affect Children’s Outcomes? A Systematic Review of the Evidence’, Child Indicators Research. Epub ahead of print 4 November 2020. DOI: 10.1007/s12187-020-09782-0.

10.

Corluy

Vandenbroucke

(2014) ‘Individual Employment, Household Employment, and Risk of Poverty in the European Union: A Decomposition Analysis’, in Cantillon

Vandenbroucke

(eds) Reconciling Work and Poverty Reduction: How Successful are European Welfare States?, pp.94–130. Oxford: Oxford University Press. DOI: 10.1093/acprof:oso/9780199926589.003.0003

11.

Daly

(2011) ‘What Adult Worker Model? A Critical Look at Recent Social Policy Reform in Europe from a Gender and Family Perspective’, Social Politics: International Studies in Gender, State and Society 18(1): 1–23.

12.

Deaton

(2020) ‘Randomization in the Tropics Revisited: A Theme and Eleven Variations’, National Bureau of Economic Research Working Paper Series No. 27600. Cambridge, MA: National Bureau of Economic Research. DOI: 10.3386/w27600.

13.

de Beer

. (2007) ‘Why Work is Not a Panacea: A Decomposition Analysis of EU-15 Countries’, Journal of European Social Policy 17(4): 375–88.

14.

De Deken

. (2013) ‘Identifying the Skeleton of the Social Investment State: Defining and Measuring Patterns of Social Policy Change on the Basis of Expenditure Data’, in Cantillon

Vandenbroucke

(eds) Reconciling Work and Poverty Reduction: How Successful are European Welfare States?, pp.260–85. Oxford: Oxford University Press.

15.

De Deken

. (2017) ‘Conceptualizing and Measuring Social Investment’, in Hemerijck

(ed.) The Uses of Social Investment, pp.184–93. Oxford: Oxford University Press.

16.

Diris

Vandenbroucke

Verbist

(2017) ‘The Impact of Pensions, Transfers and Taxes on Child Poverty in Europe: The Role of Size, Pro-poorness and Child Orientation’, Socio-Economic Review 15(4): 745–75.

17.

Ferragina

(2020) ‘Family Policy and Women’s Employment Outcomes in 45 High-income Countries: A Systematic Qualitative Review of 238 Comparative and National Studies’, Social Policy and Administration 54: 1016–66.

18.

Ferragina

Seeleib-Kaiser

(2015) ‘Determinants of a Silent (R)evolution: Understanding the Expansion of Family Policy in Rich OECD Countries’, Social Politics: International Studies in Gender, State and Society 22(1): 1–37.

19.

Gauthier

A.H.

(1998) The State and the Family: A Comparative Analysis of Family Policies in Industrialized Countries. Oxford: Oxford University Press.

20.

Häusermann

Palier

(2017) ‘The Politics of Social Investment: Policy Legacies and Class Coalitions’, in Hemerijck

(ed.) The Uses of Social Investment, pp.339–48. Oxford: Oxford University Press.

21.

Havnes

Mogstad

(2011) ‘Money for Nothing? Universal Child Care and Maternal Employment’, Journal of Public Economics 95(11): 1455–65.

22.

Hemerijck

(2015) ‘The Quiet Paradigm Revolution of Social Investment’, Social Politics: International Studies in Gender, State and Society 22(2): 242–56.

23.

Hemerijck

(2017) The Uses of Social Investment. Oxford: Oxford University Press.

24.

Hemerijck

Burgoon

Di Pietro

Vydra

(2016) Assessing Social Investment Synergies (ASIS). Luxembourg: Directorate-General for Employment, Social Affairs and Inclusion, Publications Office of the European Union, Available at https://op.europa.eu/s/ozXB (accessed 1 October 2020).

25.

Immervoll

Sutherland

de Vos

(2001) ‘Reducing Child Poverty in the European Union: The Role of Child Benefits’, in Vleminckx

Smeeding

(eds) Child Well-being, Child Poverty and Child Policy in Modern Nations, pp.407–32. Bristol: Policy Press.

26.

Jenson

(2009) ‘Lost in Translation: The Social Investment Perspective and Gender Equality’, Social Politics 16(4): 446–83.

27.

Kuitto

(2016) ‘From Social Security to Social Investment? Compensating and Social Investment Welfare Policies in a Life-course Perspective’, Journal of European Social Policy 26(5): 442–59.

28.

Kvist

(2015) ‘A Framework for Social Investment Strategies: Integrating Generational, Life Course and Gender Perspectives in the EU Social Investment Strategy’, Comparative European Politics 13: 131–49.

29.

Lefebvre

Merrigan

Verstraete

(2009) ‘Dynamic Labour Supply Effects of Childcare Subsidies: Evidence from a Canadian Natural Experiment on Low-fee Universal Child Care’, Labour Economics 16(5): 490–502.

30.

Nikolai

(2012) ‘Towards Social Investment? Patterns of Public Policy in the OECD World’, in Morel

Palier

(eds) Towards a Social Investment Welfare State?: Ideas, Policies and Challenges, pp.91–116. Bristol: The Policy Press.

31.

Noël

(2020) ‘Is Social Investment Inimical to the Poor?’, Socio-Economic Review 18(3): 857–80.

32.

Nolan

(2013) ‘What Use is “Social Investment”?’, Journal of European Social Policy 23(5): 459–68.

33.

Nolan

(2017) ‘The Thin Line between Evidence-Based Research and Political Advocacy’, in Hemerijck

(ed.) The Uses of Social Investment, pp.43–50. Oxford: Oxford University Press.

34.

OECD (2016) Who Uses Childcare? Background Brief on Inequalities in the Use of Formal Early Childhood Education and Care (ECEC) Among Very Young Children. Paris: OECD, Available at https://www.oecd.org/els/family/Who_uses_childcare-Backgrounder_inequalities_formal_ECEC.pdf (accessed 1 October 2020).

35.

Pavolini

Van Lancker

(2018) ‘The Matthew Effect in Childcare Use: A Matter of Policies or Preferences?’, Journal of European Public Policy 25(6): 878–93.

36.

Plavgo

Hemerijck

(2020) “The Social Investment Litmus Test: Family Formation, Employment and Poverty”, Journal of European Social Policy. Epub ahead of print 12 September 2020. DOI: 10.1177/0958928720950627.

37.

Saraceno

(2015) ‘A Critical Look to the Social Investment Approach from a Gender Perspective’, Social Politics: International Studies in Gender, State and Society 22(2): 257–69.

38.

Taylor-Gooby

Gumy

J.M.

Otto

(2015) ‘Can “New Welfare” Address Poverty Through More and Better Jobs?’, Journal of Social Policy 44(1): 83–104.

39.

Thévenon

Solaz

(2013) Labour Market Effects of Parental Leave Policies in OECD Countries. OECD Social, Employment and Migration Working Papers No. 141. Paris: OECD.

40.

Vandenbroucke

Vleminckx

(2011) ‘Disappointing Poverty Trends: Is the Social Investment State to Blame?’, Journal of European Social Policy 21(5): 450–71.

41.

Van Lancker

. (2018) ‘Reducing Inequality in Childcare Service Use Across European Countries: What (if any) is the Role of Social Spending?’, Social Policy and Administration 52(1): 271–92.

42.

Van Lancker

Ghysels

. (2014) ‘Who Benefits from Investment Policies? The Case of Family Activation in European Countries’, in Cantillon

Vandenbroucke

(eds) Reconciling Work and Poverty Reduction: How Successful are European Welfare States?, pp.212–37. Oxford: Oxford University Press. DOI: 10.1093/acprof:oso/9780199926589.003.0007.

43.

Van Lancker

Van Mechelen

. (2015) ‘Universalism under Siege? Exploring the Association between Targeting, Child Benefits and Child Poverty across 26 Countries’, Social Science Research 5: 60–75.

44.

Van Lancker

Vinck

. (2020) ‘The Consequences of Growing Up Poor’, in Greve

(ed.) Routledge International Handbook of Poverty, pp.96–106. London; New York, NY: Routledge.

45.

Van Vliet

Wang

. (2015) ‘Social Investment and Poverty Reduction: A Comparative Analysis across Fifteen European Countries’, Journal of Social Policy 44(3): 611–38.