Poverty alleviation through the lens of SI policy functions

In recent decades, changes in demographics, new social risks, and changing societies have required many welfare states to implement elements of social investment alongside the traditional redistributive social protection systems; work-family reconciliation, skill formation, and activation (Van Kersbergen and Hemerijck, Reference Van Kersbergen and Hemerijck2012; Hemerijck, Reference Hemerijck2017; Hemerijck and Ronchi, Reference Hemerijck, Ronchi, Béland, Leibfried, Morgan, Obinger and Pierson2021). Hence, recent studies have used a distinction between the ‘old’ compensatory social protection and ‘new’ social investment policies to bring forth positive macro- and micro-level poverty returns to these policies (Maldonado and Niewenhuis, Reference Maldonado and Nieuwenhuis2015; Plavgo and Hemerijck, Reference Plavgo and Hemerijck2021), some being more cautious on the ability of SI supporting the most disadvantaged groups in the society (Cantillon, Reference Cantillon2011; Vandenbroucke and Vleminckx, Reference Vandenbroucke and Vleminckx2011; Van Vliet and Wang, Reference van Vliet and Wang2015). Considering the complex phenomenon of poverty, single policies or policy types alone are not able to alleviate poverty, but cash transfers, services, and skill enhancement may all be required simultaneously. Hence, this article extends from the dichotomous approach in SI literature and uses SI as a lens to analyse the policy complementarities across various policies on poverty risk.

This article employs the policy function framework of buffers, flows, and stocks from the social investment framework to analyse policy impacts with a policy complementary perspective (Hemerijck, Reference Hemerijck2013, Reference Hemerijck2017). This approach combines policies together based on their main function, allowing to analyse policy impacts through their conjoint effects. Policies that provide safety nets and income compensation for labour market failures or personal crises, aiming to maintain adequate household financial resources for a sufficient livelihood, represent the buffer function. Policies that smooth labour market transitions in employment and promote more gender equal labour market attainment through work-family reconciliation represent the flow function. Policies that foster human capital development from compulsory education to higher education and lifelong learning (LLL) represent the stock function. Each function is expected to have varying impacts on individual poverty risk based on demographic factors, such as family status, and life course stage of the individual (Hemerijck et al., Reference Hemerijck, Ronchi and Plavgo2023). Although all policies are equally available to both men and women, opportunities in the labour market, gender norms, care responsibilities, and other societal issues cause gender disadvantages and differences in experiencing poverty. Hence, the article focuses on the differences across age groups, family types, and gender in individual poverty risk.

The buffer function provides a safety net to compensate for income loss during labour market failures, such as unemployment, and social problems, such as falling ill. Such policies include, for example, social assistance, unemployment benefits, and housing allowance, all of which have a direct and imminent impact on individual poverty risk by sustaining resource-sufficient livelihoods. These have traditionally been the main welfare state tools to tackle poverty, and previous research has found that generous social security transfers can moderate the negative consequences of job loss and reduce individual poverty risk (Brady et al., Reference Brady, Fullerton and Cross2010; O’Campo et al., Reference O’Campo, Molnar, Ng, Renahy, Mitchell, Shankardass, John, Bambra and Muntaner2015). While many policies with a buffer function often have universal and targeted elements, monetary social benefits are found to have a particularly positive impact on single mothers’ poverty risk (Brady and Burroway, Reference Brady and Burroway2012; Zagel et al., Reference Zagel, Hübgen and Nieuwenhuis2021).

The flow function includes policies that support individuals during labour market and family transitions. For clarity, this article focuses on the policies that promote work-family balance. Early childhood education and care (ECEC) and parental leave with sufficient income replacement are key policies within the flow function. They emphasise mothers as both caregivers and earners and are associated with higher maternal employment and more working hours (Boeckmann et al., Reference Boeckmann, Misra and Budig2015), a reduction in precarious employment (Zoch and Hondralis, Reference Zoch and Hondralis2017) and diminished poverty risk (Misra et al., Reference Misra, Budig and Moller2007; Zagel and Van Lancker, Reference Zagel and Van Lancker2022). Family cash transfers, such as child allowances, compensate for the costs of child-rearing and alleviate the pressures on parents’ labour market returns, and have been found to reduce poverty, particularly among single mothers (Lohmann, Reference Lohmann2009; Maldonado and Niewenhuis, Reference Maldonado and Nieuwenhuis2015; Leventi et al., Reference Leventi, Sutherland and Tasseva2019) and low-income households (Van Lancker and Van Mechelen, Reference Van Lancker and Van Mechelen2015). Although the child benefit, as a cash transfer, could also be analysed within the buffer function, here its main function is considered to be promoting work-life balance over a long-time span enabling parents to maintain existing labour market situation without a need for improved returns, instead of buffering only the increased costs of child-rearing for a short period.

Policies with a stock function provide the means to create, maintain, update, and upgrade one’s skills and human capital. ALMP training programmes and adult education are central policies. ALMP training programmes are found to be promoting higher employment and educational participation among young adults (Caliendo et al., Reference Caliendo, Künn and Schmidl2011; Cefalo and Scandurra, Reference Cefalo and Scandurra2023), and increased employment for unemployed single mothers (Zabel, Reference Zabel2013). Adult education can increase one’s earnings (Stenberg, Reference Stenberg2010) and promote employment prospects of low-skilled youth (Knipparth and De Rick, Reference Knipprath and De Rick2014). These are vital in poverty alleviation as social protection or employment promotion alone might be insufficient to tackle the multidimensional issue of poverty and the needs of individuals in in-work poverty (Lohmann, Reference Lohmann2009; Cantillon et al., Reference Cantillon, Van Mechelen, Pintelon, Van den Heede, Cantillon and Vandenbroucke2014). Yet, research has provided scarce evidence of the impacts of the stock function on poverty.

The above section has presented SI policy functions separately, expressing how they may influence different groups and life dimensions, yet they can also impact individuals simultaneously, creating policy complementarities. Some scholars have found multiplicative mechanisms of policies; for example, Plavgo (Reference Plavgo2023) found that investments in education increased the effectiveness of ALMP on employment, and Hemerijck et al. (Reference Hemerijck, Burgoon, di Pietro and Vydra2016) demonstrated an employment-boosting influence of ECEC when the social security transfers were strong. On the other hand, policies can also substitute for an effect when another policy is weak. For example, Niewenhuis (Reference Nieuwenhuis2022) found that while ALMP investment was low, ECEC was able to promote female employment even more strongly than when ALMP investment was high. Although all these studies focus on employment outcomes, it can be assumed that the policy complementarities would also have an influence on poverty, considering its alleviation requires welfare states to adopt various mechanisms for different social risks and demographic groups. Therefore, this study uses a novel two-level approach to examine policy complementarities. First, various policy indicators are used to create packages that reflect the SI policy functions, enabling an analysis of multiple policies simultaneously. Second, the interaction analyses between the policy functions examine further complementary relations.

Based on the theoretical and empirical evidence presented above, this study poses the following research questions (RQs) on the linkages between social investment policy functions and individual poverty risk:

Policy changes in twenty-first-century Germany

The poverty rates in Germany increased rapidly at the beginning of the twenty-first century (Jenkins and Schluter, Reference Jenkins and Schluter2003; OECD, 2008), compounded by high unemployment and low fertility rates. To tackle these demographic and social issues, Germany took major steps in reforming the welfare state. Previously characterised as the main model for a male-breadwinner system, the German welfare state shifted its focus towards providing stronger support for two-earner families and female employment, boosting the working-age population’s economic activity through active labour market policy reforms (Trzcinski and Camp, Reference Trzcinski, Camp and Robila2014). This article focuses on nine different policies, which are organised into the SI policy functions: Buffers consist of unemployment benefits, housing allowance, and minimum wage; flow function includes parental leave, child allowance, and childcare; and stocks consist of adult education, institutional ALMP training, and non-continuation of compulsory education.Footnote ¹

The unemployment benefits represent the social assistance part of the buffer function in Germany, particularly after the Hartz reforms, which took place around the mid-2000s. These reforms implemented a substantial change in unemployment benefits, merging them with social assistance with means-tested criteria and establishing stricter long-term benefits and sanctions (Trzcinski and Camp, Reference Trzcinski, Camp and Robila2014; Seeleib-Kaiser, Reference Seeleib-Kaiser2016). Although some positive returns in employment were observed, the reform also led to deregulating temporary and atypical forms of work, promoting any work over sanctions and increasing in-work poverty (Seeleib-Kaiser, Reference Seeleib-Kaiser2016; Gerlitz, Reference Gerlitz2018). Housing allowance was integrated into the unemployment benefit within the Hartz reforms, making it a benefit for low-income households above the social minimum (Haffner et al., Reference Haffner, Hoekstra, Oxley and van der Heijden2009). The allowance amount was increased in 2009 and again in 2016 to compensate housing costs for people with low incomes (Haffner et al., Reference Haffner, Hoekstra, Oxley and van der Heijden2009; Federal Ministry for Economic Affairs and Energy, 2016). A statutory hourly minimum wage was introduced in Germany in 2015 to alleviate poverty. The reform positively impacted low-wage workers, but broader poverty reduction has not been observed (Caliendo et al., Reference Caliendo, Schröder and Wittbrodt2019).

Regarding the flow function, the policies promoting better work-life balance in Germany have seen significant reforms. With more generous and earnings-related pro-dual-earner parental leave reform in 2007 and a significant but gradual increase in ECEC availability for under three-year-olds since 2005, the aim has been to promote maternal employment, reduce inequalities and poverty, and generate ‘sustainable’ human capital for the future (Ostner, Reference Ostner2010; Trzcinski and Camp, Reference Trzcinski, Camp and Robila2014). The parental leave reform has increased mothers’ labour market attachment (Ziefle and Gangl, Reference Ziefle and Gangl2014). On the other hand, child allowances have been stagnant and have suffered from inflation significantly despite an increase in 2010 and some small occasional increases later (Gauthier, Reference Gauthier2011; Raschke, Reference Raschke2012; Sozialpolitik-aktuels.de, 2023). However, the benefit system allows parents to choose between the direct cash benefit of child allowances or a tax deduction, with those who are well-off tending to opt for the latter (Blome et al., Reference Blome, Keck and Alber2009; Van Lancker and Ghysels, Reference Van Lancker and Ghysels2012).

Within the stock function, the German education and training systems have seen reforms in both general adult education and youth-targeted programmes. There have been attempts to reform vocational training towards lifelong learning, yet the system has remained very complex (European Commission, 2022). Overall, the adult education participation rates have been increasing steadily since the 2000s, and as such, it remains a central part of the German stock function (Eurostat, 2022). Although the overall government spending on the training as a part of ALMP has decreased sharply throughout the 2000s, specific programmes and employer subsidies for disadvantaged youth, as well as training programmes for low-skilled and unemployed individuals, were established since the early 2000s to improve employment of marginalised groups and to reduce high poverty rates (European Commission, 2022; OECD, 2023). The training programmes were found to increase educational participation, and the number of young adults without post-compulsory education has decreased steadily since the mid-2000s (Eurostat, 2023). However, the impact of the training programmes was less influential, or more prolonged, for employment returns and hence their influence on reducing the high poverty rates of youth could be in question (Caliendo et al., Reference Caliendo, Künn and Schmidl2011).

Data and methods

Policy function measures

This article aims to study the policy complementarities through social investment policy functions. These are operationalised using nine policy indicators (see Appendix A2 for detailed information and sources), which, based on their main policy function, are categorised into one of three policy packages: buffer, flow, and stock functions. Although each policy indicator could have multiple functions, they have been categorised based on their most substantial intended function. For example, child allowance could be seen as buffering inadequate labour market incomes for families, but because the policy aim is to compensate for the increased costs of child-rearing, often despite their income levels, the flow function is stronger. The policy packages aim to represent various policy types or relate to multiple life stages, yielding measures with extensive coverage of German welfare state policies between 2001Footnote ³ and 2018.

Each policy package aims to capture some of the main policy changes of each policy function in Germany (see the previous section for literature). The buffer package includes policy indicators that alleviate the negative economic impacts of labour market failures and personal events: unemployment benefits (replacement rate after twenty-four months of unemployment), housing allowance (monthly amount), and minimum wage (percentage of median full-time wages). The flow package contains policies that promote work-family balance: parental leave (replacement rate), ECEC (participation rate), and child allowances (monthly amount for a first child). The stock package covers policies that both build and maintain human capital: the rate of early leavers from education (the percentage of young adults with maximum lower secondary education and not currently in education or training); expenditure on institutional training (targeted programmes for disadvantaged groups); and adult education (participation rate).Footnote ⁴

Given that the exact time of policy changes cannot be pinpointed to ensure the actual policy provision at the time of the interview, the macro variables are lagged by one year (t−1), except for institutional training, which measures annual expenditure, and can be assumed to have a more direct impact on individuals’ poverty risk. Further, the policy indicators are harmonised to be comparable and scaled so that negative values indicate low or weak public provision and positive values indicate high/strong provision. Hence, the repetition and early education leavers rates were modified such that the higher the value, the better.

The policy function packages were created using principal component analysis (PCA), which uses a correlation matrix of the annual information on multiple policy variables, creating only one measure that captures the original over-time variance as much as possible while minimising information loss. The PCA component explains 67.5 per cent of the variance in buffer variables, 73.5 per cent of flow variables and 87.4 per cent of stock variables (see Appendix A3 for the correlation matrices, the eigenvalues and coefficients of the PCA components). The macro variables are standardised to have equal weights in the PCA. Figure 1 presents the annual information of each standardised policy measure and the policy packages across the observation period. Notably, although the policy measures that have been reduced over time (unemployment benefit and institutional training) are not well reflected in the PCA components, higher values of the components can still be interpreted as stronger functions.

Figure 1.

Changes in buffer, stock, and flow measures in Germany 2002–18 (variables are harmonised, standardised, and all, except for institutional training, are also lagged for analytical purposes).

Methods

The analyses are based on non-linear logistic regression modelling with clustered standard errors (SEs) by individuals to investigate the probability of individual poverty risk. This methodology allows to examine a phenomenon (poverty) that occurs in only a small portion of the population, i.e., the dependent variable does not vary over time for most individuals. The method acknowledges the longitudinal nature of the data by correcting SEs and enables to analyse both constant (i.e., gender, education, region, and immigrant status) and time-varying (labour market status, partnership status) variables.

The micro-macro modelling examines how the change in the independent variable (aggregate policy function measures) impacts the probability of the outcome (individual poverty risk) among specific demographic groups, accounting for individual-level covariates. In other words, the analytical framing examines cross-level interactions between policy functions and demographic factors such as gender, age, and family type. To examine individual effects within population subgroups, the analyses apply average marginal effects (AMEs).

Although the fixed-effects model would be ideal to counter the unobserved heterogeneity bias, it would drop out individuals who have no over-time change in their poverty risk and hence would not allow the examination of the phenomenon with a partial sample. Further, unlike hierarchical models, this analytical approach enables using cross-sectional individual weights that account for the initial sampling probability and all response probabilities for each survey year, which are crucial when analysing a marginal phenomenon and a population that often has a low participation rate in surveys.

The results are presented separately for each research question. First, the results of the interaction effects between the policy functions and age group (Figure 2) or family status (Figure 4) are presented as AME graphs from a logistic regression model with a complete set of controls for men and women separately (see Appendix A4 for tabled AME results of the figures and for AME graphs of individual policies). All models control for the other SI policy functions than the policy in focus to derive more apparent impacts for each policy. Using AMEs enables a clearer interpretation of the logistic regression coefficients and mitigates issues related to comparing different logistic regression models (Breen et al., Reference Breen, Karlson and Holm2018). Second, the policy complementarity effects, namely interaction effects between all the policy functions, are presented as predicted probabilities of flow and stock functions at low and high levels of buffer functions among young adults (Figure 3) or among single mothers (Figure 5). Low buffers reflect the values below, and high buffers above the median value, reflecting 2002–11 and 2012–18, respectively. In these graphs, confidence intervals (CIs) are omitted to emphasise the general associations and differences instead of statistically significant differences across (small) groups to clarify interpretation. Appendix A5 includes the tabled results of all regression models focusing on policy functions.

Figure 2.

Impacts of SI policy functions on poverty risk by age group and gender (AMEs of separate logistic regression models with clustered SEs).

Figure 3.

Policy complementarities on the poverty risk among young (17–29 years) men and women separately for low (2002–11) and high buffers (2012–18) (predicted probabilities of logistic regression models with clustered SEs, CIs omitted).

Figure 4.

Impacts of SI policy functions on poverty risk by family type and gender (AMEs of separate logistic regression models with clustered SEs).

Figure 5.

Policy complementarities on the poverty risk of single mothers, separately for low (2002–11) and high buffers (2012–18) (predicted probabilities of logistic regression models, CIs omitted).

Approximately 21 per cent of the analytical sample is from East Germany; hence, as a robustness check, the analyses were run for West Germany only. The results were very similar to those presented below for the total sample. Since policies are measured after unification, and most policies are implemented at the national level, including the sample from East Germany increased the robustness and generalisability of the results on policy impacts.