2. The German unemployment and social benefit system before and after 2005

In the comparative welfare state literature, Germany is considered a key representative of the conservative welfare state regime (Esping-Andersen, Reference Esping-Andersen1990). Organised based on the principle of social insurance, its welfare institutions centre around the worker and are underpinned by strong tripartite partnerships between employees, employers, and the state.

Until the early 2000s, Germany’s protection of the unemployed consisted of two pillars (Hassel and Schiller, Reference Hassel and Schiller2010). Both aimed at safeguarding the individual from experiencing a drop in living standards due to unemployment. The first pillar was ‘Arbeitslosengeld’, i.e. unemployment benefits from social insurance. These were generally paid for 12 months to individuals with social insurance contributions of at least 12 months before becoming unemployed. Depending on overall contributions, older individuals could even receive benefits for up to 36 months. Benefit levels were equivalent to 67 and 60 percent of previous earnings for individuals with and without children, respectively. The second pillar was ‘Arbeitslosenhilfe’, i.e. tax-funded unemployment assistance. Transfers from unemployment assistance were principally unlimited and paid to individuals whose claim to unemployment benefits expired. They were therefore also conditional on social insurance contributions. Replacement rates amounted to 57 and 53 percent for individuals with and without children, respectively. Payments only ceased if recipients repeatedly refused to accept work that was comparable to their previous employment or current benefit levels (Bäcker and Koch, Reference Bäcker and Koch2004).

Implemented on January 1, 2005, the Hartz IV reform combined the second pillar of unemployment protection with means-tested basic income support. Before the reform, ‘Sozialhilfe’, i.e. tax-funded social assistance, provided all individuals in need who were not covered by unemployment insurance with a minimum social security net. Its main purpose was to reduce poverty and guarantee a legally defined socio-cultural subsistence level within households. Thus, individuals received social assistance only upon passing a means test. Refusal to take up work was strictly sanctioned with benefit cuts. There was hence a clear difference in benefit generosity and conditionality for individuals whose entitlement to social benefits was based on previous employment and social contribution records vis-à-vis financial hardship. After the reform, this distinction eroded. Since 2005 entitlement to earnings-related benefits ends with the expiration of unemployment benefits, which older workers can now receive for a shorter maximum duration of 18 months (Bäcker, Reference Bäcker2017). Individuals who remain unemployed longer are only eligible for means-tested basic income support. Figure 1 illustrates how benefit schemes and levels change by unemployment duration before and after the reform.

Figure 1. Benefit schemes and benefit levels by unemployment duration, pre- and post-reform.

The main objective behind this dramatic policy shift was to reduce long-term unemployment and welfare dependance (Jacobi and Kluve, Reference Jacobi and Kluve2006). During the 2000s, the proportion of long-term unemployed among all unemployed had increased to almost 40 percent (Bundesagentur für Arbeit, 2019). For this purpose, the reform also redefined the category of people in need and divided them into two groups based on their ability to work (Goebel and Richter, Reference Goebel and Richter2007). Individuals who can work for at least 3 hours per day need to apply to the newly created unemployment assistance scheme called ‘Arbeitslosengeld II’ (ALGII).

Importantly, ALGII is accessible to individuals without employment as well as those on low incomes as an in-work benefit (Eichhorst and Marx, Reference Eichhorst and Marx2011). Individuals who cannot work for at least 3 hours per day, and who do not share a household with someone eligible for unemployment assistance, continue to receive social assistance. While benefit levels are similar, the main difference between the two schemes is that unemployment assistance is now highly conditional on active job search and participation in activation measures as well as related labour market programmes. Accordingly, harsh sanctions for non-compliance were introduced for recipients of unemployment assistance. By contrast, sanctions for recipients of social assistance were omitted after the reform as they are no longer subject to labour market integration. Appendix Table A1 gives an overview of the main characteristics of the unemployment and social benefit schemes that were in place before and after the reform.

3. Hartz IV and educational inequality

Over the past decades, levels of education have greatly increased in Germany due to vast educational expansion (Pollak and Müller, Reference Pollak, Müller, Breen and Müller2020). Nevertheless, family background is still a strong predictor of academic achievements (Shavit and Blossfeld, Reference Shavit and Blossfeld1993; Blossfeld, Reference Blossfeld2018) and social mobility low by international comparison (Bukodi et al., Reference Bukodi, Paskov and Nolan2020). The reform might have contributed to these persisting inequalities by reducing the educational outcomes of children of unemployment assistance recipients against concurrent efforts towards greater educational equalisation.

Two mechanisms are conceivable through which the reform reduced educational outcomes for children living in families that receive unemployment assistance. First, the reform changed the composition of unemployment assistance recipients towards the more disadvantaged by altering eligibility criteria. Second, the reform adversely affected parental resources and wellbeing by altering benefit levels and benefit conditionality.

As described in the previous section, the introduction of strict means-testing implies that a smaller group of the long-term unemployed is eligible for unemployment assistance since 2005. Only those receive the benefit, whose incomes fall below subsistence level. Families that can maintain a higher level of income, e.g. because one parent is still employed and earns sufficiently well, are excluded. Families receiving unemployment assistance after the reform could therefore form a more socio-economically disadvantaged group than families receiving unemployment assistance before the reform. Specifically, the proportion of low educated, working class, or single parents could be higher. Consequently, a decline in the transition rates into the academic school track among children of unemployment assistance recipients might be observed.

However, there is evidence that the reform also had a direct impact on family resources and wellbeing. Goebel and Richter (Reference Goebel and Richter2007) show that more than 50 percent of individuals who switched unemployment assistance schemes in 2005 experienced a loss in income following changes in benefit levels. In the first year of the reform, two in three households receiving unemployment assistance had incomes below 60 percent of median income, and the poverty gap was with 18.5 percent almost twice as high as the poverty gap for households receiving unemployment assistance one year before.

These reductions in the financial resources of unemployment assistance recipients could lead to lower education in the second generation by affecting their performance in school as well as choices regarding educational careers (Boudon, Reference Boudon1974). On the one hand, parents have fewer means to advance their children’s learning through buying additional material such as books and computers. They are also less likely to afford private tutoring lessons or summer schools (Duncan and Brooks-Gunn, Reference Duncan and Brooks-Gunn1997; Kaushal et al., Reference Kaushal, Magnuson, Waldfogel, Murnane and Duncan2011). On the other hand, their ability to finance longer periods of education and training is reduced, even if their children performed well in school. They are hence less likely to opt for the academic educational trajectory that takes significantly more time to complete than the vocational qualification route. The former requires at least three additional years of school even before starting university.

Moreover, the reform turned the receipt of unemployment assistance into a more taxing experience, aggravating the well-documented negative effects of unemployment on subjective wellbeing (Clark, Reference Clark2003; Oesch and Lipps, Reference Oesch and Lipps2013). Deter (Reference Deter2021) reports that the reform led to lower levels of life satisfaction among the long-term unemployed. As explained earlier, recipients of unemployment assistance are now strictly monitored by public employment agencies in their efforts to find work and are required to participate in active labour market programmes. Additionally, policymakers used language of blame and contempt, creating a dooming narrative around social benefit recipients, when introducing the Hartz reforms (Fohrbeck et al., Reference Fohrbeck, Hirseland and Ramos Lobato2014).Footnote ¹ Over and above the psychological costs of lower incomes (Kahneman and Deaton, Reference Kahneman and Deaton2010), parents receiving unemployment assistance after the reform are therefore exposed to greater stress and stigmatisation compared to parents receiving unemployment assistance before the reform.

This in turn can lead to poorer academic performance among their children. Parental emotional wellbeing creates a stable home environment for children’s flourishing development (Kalil, Reference Kalil2013) and ensures parents’ capacity to offer effective help with homework and other school-related activities (Murray et al., Reference Murray, Woolgar, Martins, Christaki, Hipwell and Cooper2006). Thus, given the detrimental effects of lower benefit levels and stricter benefit conditionality on family resources and wellbeing, children of unemployment assistance recipients might be less likely to transition to academic secondary school after the reform.

In some part, I hence expect negative selection to account for lower educational outcomes among children of unemployment assistance recipients. Nevertheless, this process is to be considered an integral part of the reform. As the pool of eligible families is restricted to those on low incomes, the reform’s potential negative consequences on family resources and wellbeing are exclusively channelled towards the most disadvantaged and economically vulnerable, worsening their conditions even further. If this is the case, the reform increased educational inequality by reinforcing the intergenerational transmission of disadvantage at the very bottom of society.

4. Analytical strategy

4.3. Modelling

To identify the effect of the reform on changes in the probability to enter the academic school track for children of unemployment assistance recipients, I adopt a quasi-experimental research design based on difference-in-differences estimation (Angrist and Pischke, Reference Angrist and Pischke2008). Although the reform was implemented simultaneously across Germany affecting the entire population at the same time, it seems plausible that families that did not receive any of the benefits discussed were least affected by the reform. Children living in such families are therefore taken as reference group against which pre- and post-reform differences in transitions to the academic track of children with parents receiving unemployment assistance are compared. For completeness and to account for potential spill-over effects, I also compare pre- and post-reform differences between children living in families, where parents do not receive benefits and where parents receive unemployment benefits or social assistance.

For unbiased identification, the difference-in-differences estimator rests on the so-called ‘common trends assumption’. This assumption states that in the absence of treatment, trends in the outcome of interest are similar for the two groups that are compared against one another. Accordingly, the probability to enter the academic track for children of benefit recipients and those who do not receive benefits should follow a similar trend before and after the reform. Besides visually inspecting these trends, I formally scrutinise the validity of this assumption using two placebo identification tests. The first assumes that the reform was already implemented in 2001. Prior to the 2005 Hartz reform, Germany’s unemployment and social benefit system was modified for the last time in 1997. Children who transitioned to secondary school between 1998-2001 (N=1,113) are hence identified as pre-treatment group, whereas children who transitioned between 2002-2005 (N=1,206) form the post-treatment group. As mentioned earlier, Hartz IV was part of an entire reform package. This placebo test thus also inspects whether changes in educational inequalities could be ascribed to any of the other three reforms that were introduced before. Analogously, the second placebo analysis assumes that the reform was not implemented until 2009. Children who transitioned to secondary school between 2006-2009 (N=1,004) are now assigned to the pre-treatment group, and children who transitioned between 2010-2013 (N= 2,057) are chosen as post-treatment group. Comparing these cohorts examines whether observed changes in educational inequality might be driven by some event other than the reform, which occurred after 2005 (e.g. financial crisis).Footnote ⁵ As shown in appendix Table A2, I do not find indication for a violation of the common trends assumption as the placebo reform interactions are all statistically insignificant and negligible in size. Similarly, appendix Figure A2 shows that raw transition rates into the academic track of children of unemployment assistance recipients and children of parents who did not receive benefits ran in parallel before and after the reform.

Using ordinary-least-squares regression, I estimate linear probability models of the following form:Footnote ⁶

$$\matrix{ {{\pi _{it}} = \;\;\alpha + {\beta _b}BE{N_{ibt - 1}} + \sigma REFOR{M_{t - 1}}\; + \;{\lambda _b}\left( {BE{N_{ibt - 1}} \times REFOR{M_{t - 1}}} \right)\;} \cr {\;\;\;\;\;\;\;\;\;\;\;\; + \;\varphi {X_{it - 1}} + \phi \left( {{X_{it - 1}} \times REFOR{M_{t - 1}}} \right)\; + {\tau _{t - 1}} + {\varepsilon _{it}}\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;} \cr } $$

where ${\pi _{it}}$ stands for the probability for child i to enter the academic track in year t. $BE{N_{ibt - 1}}$ denotes a set of dummies for parental benefit receipt $\;b$ in year $t - 1$ , i.e. one year before the child’s transition. $REFOR{M_{t - 1}}$ is a dummy that indicates whether parental benefit receipt is observed before or after the reform. It takes the value 0 if parental benefit receipt is observed in year $t - 1$ < 2005, and 1 otherwise. $BE{N_{ibt - 1}} \times REFOR{M_{t - 1}}$ identifies whether the association between children’s transition to the academic track in $t\;$ and parental benefit receipt $\;b$ in $t - 1\;$ changed after the reform relative to children whose parents did not receive any benefits. ${X_{it - 1}}$ and ${X_{it - 1}} \times REFOR{M_{t - 1}}$ denote a vector of covariates and their interaction with the reform dummy. The interaction is included to model potential changes in the association between these covariates and secondary school transitions across the two periods. ${\tau _{t - 1}}$ is a year fixed-effect that takes note of other trends occurring during the same period, such as the rise in comprehensive schools.Footnote ⁷ ${\varepsilon _{it}}\;$ stands for the error term.

The baseline model includes controls for the child’s sex, age at the transition from primary to secondary school as well as the state where the transition took placeFootnote ⁸ . I then add a term for net equivalised household income minus transfers from unemployment and social assistance in $t - 1$ . Controlling for pre-benefit household income allows to examine how far observed changes in the probability to transition to the academic track for children of unemployment assistance recipients are driven by the introduction of strict means-testing, and thus selection. To take account of further differences between families receiving unemployment assistance before and after the reform, controls for a range of socio-demographic characteristics are added to the model. These are the number of children younger than 15 in the household, whether living with a single parent, mean parental age (squared), parental migration background, parental education, and parental social class in $t - 1$ . Social class is measured using the European Socio-Economic Classification (Rose and Harrison, Reference Rose and Harrison2007). In cases where mother’s and father’s education or class position differ, the higher is taken according to the dominance approach (Erikson, Reference Erikson1984).Footnote ⁹

To precisely estimate the short-term association between parental benefit receipt and transitions to the academic track, I further control for benefit duration by adding a term that measures the number of consecutive years parents received the benefit during their child’s primary education, i.e. whether the benefit had been received for one, two, three or four years. Where parents did not receive benefits one year before the transition, the number of consecutive years spent without benefits is counted.Footnote ¹⁰

Next, I include terms for total net equivalised household income as well as parental life satisfaction in $t - 1.\;$ I do so to test for the mediating effects of changes in family resources and wellbeing due to reduced benefit levels and increased benefit conditionality. Parental life satisfaction is measured on a scale from 0 to 10, where higher values indicate higher life satisfaction. In cases where mother’s and father’s life satisfaction differ, the lower value is taken. Unfortunately, the SOEP does not provide more detailed information on parents’ stress level and parent-child interactions for the observation period in question. Table 1 displays the distributions of all variables included in the analysis, separately for the pre- and post-reform groups.

Table 1. Descriptives

Source: SOEP 1984-2017, v.34.

5. Empirical findings

Table 2 presents OLS estimates for children’s probability to transition to the academic school track between 2001 and 2010 in Germany. Column (1) shows the main effects of parental benefit receipt, accounting for the reform dummy and set of baseline controls. Living in families that received unemployment or social benefits is associated with substantively lower chances to transition to the academic track. On average, children whose parents received unemployment benefits are 16 percentage points less likely to transition to the academic track than children whose parents did not receive benefits. For children of unemployment and social assistance recipients the gap amounts to 26 and 31 percentage points, respectively.

Table 2. Transition to academic secondary school track (OLS)

Note: * p<0.10, ** p<0.05, *** p<0.01. Pre-treatment cohorts: 2001-2005, post-treatment cohorts: 2006-2010. Robust standard errors in parentheses. Total HH income in thousands (2005€). All models control for child’s sex, age at transition from primary to secondary school, state- and year-fixed effects. Socio-demographic controls include number of children age 0-14 in the household, living with single parent, parental age(-squared), parental migration background, parental education, and parental social class one year before transition. Full results shown in appendix Table A4.

Source: SOEP 1984-2017, v.34.

These large differences, which are statistically significant at 1 percent, are not surprising. Whether parents receive benefits or not depends on their position in the labour market and financial situation. Both are known to be highly predictive of children’s educational attainment (Becker and Nietfied, Reference Becker and Nietfied1999; Lohmann and Groh-Samberg, Reference Lohmann and Groh-Samberg2017). Moreover, the observed gradient in the probability to transition to the academic track corresponds to differences in the characteristics and target groups of the three benefit schemes. Given that recipients of unemployment benefits enjoy more generous replacement rates and are exposed to shorter periods of unemployment than recipients of unemployment assistance, educational disadvantages are plausibly smaller for children of the former. Conversely, families that claim social assistance generally struggle the most with financial hardship and poverty. It is thus reasonable that the greatest disadvantage is observed for children of social assistance recipients.

In column (2), the interaction with the reform dummy is included. The main effects for parental benefit receipt now denote average differences in the probability to transition to the academic track during the pre-reform period, and the interactions with the reform dummy identify changes in these differences after the reform. Their sums hence indicate post-reform inequalities in the probability to transition to the academic track. In line with expectations, I observe a sizable drop in the probability to enter the academic track for children of unemployment assistance recipients that is statistically significant at 10 percent. For children who made the transition before, 2006, parental unemployment assistance receipt is associated on average with a 22 percentage-point lower chance to enter the academic track relative to having parents that did not receive benefits. After the reform, this gap increased to 35 percentage points. Differences in the probability to attend the academic track between children of unemployment assistance recipients and children living in families, where no benefits were claimed, thus increased by 13 percentage points (see appendix Table A3 for predicted probabilities). As discussed earlier, this could be the result of changes in the composition of unemployment assistance recipients, decreases in family resources and wellbeing due to lower benefit levels and stricter benefit conditionality, or all three mechanisms combined.

By contrast, the interaction terms for the other two benefit types are much smaller. The coefficient for parental unemployment benefit receipt is negative and the coefficient for parental social assistance receipt is positive. This suggests that inequalities for children of unemployment benefit recipients also increased, while inequalities for children of social assistance recipients somewhat declined. However, both coefficients are not statistically significant.

In columns (3) and (4), terms for net equivalised household income before transfers from unemployment and social assistance as well as further socio-demographic characteristics are added. I thereby control for potential differences between unemployment assistance recipients before and after the reform that arise from the introduction of means-testing. The interaction between the reform dummy and parental unemployment assistance decreases by almost one fourth. Although somewhat imprecisely estimated (p=0.16), the magnitude of the coefficient indicates that inequalities in the probability to transition to the academic track still increased by 10 percentage points. The post-reform gap in academic track attendance for children of unemployment assistance recipients now amounts to 15 percentage points, whereas the pre-reform gap shrinks to less than 6 percentage points and is no longer statistically significant.

Accounting for benefit duration in column (5) further reduces pre-reform differences between children of unemployment assistance recipients and parents that did not receive benefits. The size of the interaction between the reform dummy and parental unemployment assistance slightly increases and is again statistically significant. This is because the association between benefit duration and the probability to transition to the academic track becomes negative after the reform (see appendix Table A4), while the association between parental unemployment assistance receipt and benefit duration remains negative across the two periodsFootnote ¹¹ (parents tend to receive unemployment assistance for a shorter period than not receiving any benefits). Estimated post-reform differences between these two groups therefore remain at similar levels as in column (4).

For better illustration, Figure 2 plots the predicted probabilities to transition to the academic track by parental benefit receipt before and after the reform based on column (5). Controlling for observed levels of pre-benefit income, socio-demographic characteristics and benefit duration, children of parents receiving unemployment assistance one year before the transition to secondary school had a probability of 33 percent to enter the academic track before the reform, corresponding to a less than 3 percentage-point gap compared to children of parents not receiving any kind of benefits. After the reform, this probability declined to 23 percent and translated into a 15 percentage-point gap, which is statistically significant at 5 percent.

Figure 2. Predicted probabilities to transition to academic secondary school track by parental benefit receipt, pre- and post-reform.

Note: Based on Table 2, column (5). Predicted probabilities calculated at mean values of covariates. 95 per cent confidence intervals shown.

Hartz IV thus shifted the composition of unemployment assistance recipients towards the more disadvantaged, which in turn accounts for more than 20 percent of the estimated reform effect on inequalities in secondary school track attainment. This change in the composition of unemployment assistance recipients is an important observation. It means that it is the most disadvantaged families who are affected by the reform’s potential adverse consequences on family resources and wellbeing. If so, the reform perpetuated the intergenerational transmission of inequality at the very bottom of society.

Indeed, adding a term for total net equivalised household income in column (6) drastically reduces the interaction coefficient to less than minus 2 percentage points. I therefore no longer observe meaningful pre- and post-reform differences in the gap to enter the academic track between children of unemployment assistance recipients and those living in no-benefit families. Controlling for parental life satisfaction in column (7) does not change the overall picture further. Although the conditional association between parental life satisfaction and the probability to enter the academic track is positive and statistically significant at 5 percent, a lack of change in the interaction between parental unemployment assistance receipt and the reform dummy between columns (6) and (7) makes it unlikely that parental life satisfaction plays a substantive mediating role. Rather a worsening in family’s financial resources following reductions in benefit levels appears as the main driver given the substantial reduction in the interaction coefficient between columns (5) and (6). Although the association between household income and academic track attendance is small after controlling for pre-benefit income, socio-demographic characteristics, and benefit duration, the coefficient for household income and its interaction with the reform dummy are not informative about the underlying driver of the reform effect. Instead, it is the decline in the interaction between the reform dummy and parental unemployment assistance that provides evidence for the importance of reductions in families’ financial resources for observed increases in educational inequality.