Risk factors of social assistance transitions: a case-control study for Germany

Abstract

Purpose

Poverty transitions can be explained by two opposing theories: the traditional sociological approach that focusses on social stratification and individualisation theory, which emphasises on life course risks for all strata. Both perspectives have been investigated extensively for income poverty while neglecting other important poverty indicators, such as deprivation or the receipt of social assistance. The purpose of this paper is to focus on the latter to investigate the impact of social stratification (e.g. social class), life course risks (e.g. health problems), and their interactions on the probability of social assistance entry for Germany.

Design/methodology/approach

The analysis utilises survey data containing a sample of first-time social assistance entrants and a sample of the residential population. Applying case-control methodology, logistic regression is conducted to model the impact of social stratification determinants, life course risks, and their interactions on the probability of social assistance entry.

Findings

Social stratification determinants, particularly social class, have a significant effect. However, their effect is weaker than the effect of life course risks. Contrary to the prediction of individualisation theory, the poverty-triggering impact of life course risks varies substantially by social stratum. The combination of both theories yields high predictive power.

Originality/value

This paper is the first to comprehensively test social stratification and individualisation theory with respect to social assistance receipt as a poverty indicator. It is the first paper that investigates the entire population at risk of social assistance entry in Germany.

Citation

Fuchs, B. (2017), "Risk factors of social assistance transitions: a case-control study for Germany", International Journal of Sociology and Social Policy, Vol. 37 No. 13/14, pp. 714-728. https://doi.org/10.1108/IJSSP-11-2016-0123

1. Introduction

In current sociological poverty research, two competing perspectives are prevalent among the various theoretical approaches. The traditional sociological perspective asserts that poverty is highly structured by social stratification determinants such as social class or ethnic origin. Conversely, individualisation theory states that the importance of such determinants has decreased within the twentieth century such that their association with poverty should be marginal in contemporary Western societies. Instead, individualisation theorists consider poverty a phenomenon that is induced by life course risks such as health problems or unemployment, which affect all social strata rather than only disadvantaged social groups. However, many empirical studies on poverty transitions indicate that these seemingly opposing theories are actually complementary. These studies suggest that a simultaneous investigation of both perspectives is a promising way to obtain a sociological understanding of poverty dynamics (e.g. Barbieri and Bozzon, 2016; Hill et al., 1998; Layte and Whelan, 2002; Walker, 1994).

However, the question of whether one or both of these theories is an appropriate representation of the social reality of poverty has not been sufficiently addressed. Most research has focussed on income poverty, but other important poverty indicators, such as social assistance receipt, deprivation, or subjective measures, have rarely been used in a combined investigation of these theories. One rare exception is the study of Kauppinen et al. (2014) on social assistance dynamics in Scandinavia, which did not investigate the impact of social class as the determinant for which sociological theory may bring its genuine contribution to the scientific analysis of poverty. Some authors have recently argued in favour of deprivation as the most appropriate measurement of poverty (e.g. Terraneo, 2016), whereas others prefer different indicators (Whelan and Maître, 2008). Ideally, a comprehensive sociological theory of poverty would yield high predictive value for various meaningful poverty indicators. When this notion is applied to social stratification and individualisation theories on poverty, it implies a research gap because the current state of research is limited to income poverty.

The current study is motivated by this research gap and applies both theories to another important poverty outcome, namely, the receipt of means-tested social assistance. Empirical studies that have utilised this long-established concept of poverty (e.g. Simmel, 1965) have recently been criticised for equating social assistance exits with poverty exits by showing that many individuals who stop receiving social assistance remain in precarious life circumstances (Groh-Samberg, 2004; for a comprehensive criticism of various relative poverty indicators, see Sen, 1983). However, social assistance entries can be regarded as a meaningful measurement of poverty transitions. Although there might be a significant share of poor people among non-recipients, recipients can be understood to be poor. Empirically, means-tested social assistance benefits in Western countries are significantly below the poverty line of 60 per cent of the median income in most household constellations (Avram, 2009; Eardley et al., 1996). This low amount, the means-tested character, the associated stigmatisation and the dependency dimension (Spicker, 1998) make the receipt of these benefits a meaningful indicator of relative poverty in societies that offer a minimum income support scheme.

This paper examines a country that offers social assistance benefits that are particularly well suited as a poverty indicator: Germany, which introduced a new social assistance scheme in the social and labour market reforms of 2005 (“Hartz legislation”). In contrast to what the administrative term “unemployment benefit II” (Arbeitslosengeld II) suggests, the reformed social assistance benefits are not linked to a previous or current employment status. A household is eligible if its net income[1] falls below the legally defined subsistence level and savings are mostly used. Both conditions are verified through a strict means test. These benefits refer to all households in need that have at least one person who is able to work. Therefore, “unemployment benefit II” can be considered a social assistance scheme for employable individuals and their household members. The amounts, which exactly meet the legally defined subsistence level in Germany, are substantially below the poverty line of 60 per cent of the median income in most household constellations (Lietzmann et al., 2011). The introduction of these benefits led to huge public protests, demonstrations and political agitation that argued that the new social assistance benefits imply “poverty by law” (Armut per Gesetz) (for an overview of the reform, see Promberger, 2015). Although the low amount, the means-tested character, and the association with deprivation (Christoph and Lietzmann, 2013) and reduced social integration (Beste et al., 2014) make the receipt of the reformed social assistance benefits a good indicator of poverty, the state of research concerning transitions into this form of poverty is limited.

There is currently only one paper on social assistance entries for Germany after the reforms of 2005 (Aldashev and Fitzenberger, 2009), and it has no reference to sociological theory. Furthermore, that study considers only employees. No analysis with a focus on first-time transitions exists. This issue is particularly critical because the first receipt is a crucial incident that can initiate a process of long-term welfare dependence or recidivism. Both phenomena are of high relevance for the occurrence of “poverty careers”, which are highly prevalent within Germany’s new social assistance scheme (Graf and Rudolph, 2009). Motivated by this research gap, the purpose of this paper is to investigate the predictive power of social stratification, individualisation theory, and their imbrication for the first social assistance entry.

2. The current study

This paper contributes to the literature by investigating the entire German population at risk of social assistance entry. This is the first paper to apply and comprehensively test sociological theory on this topic. Its research questions are motivated by the ongoing discussion of the poverty-triggering impact of social stratification determinants and life course risks. The latter are considered major biographical events or statuses that are believed to trigger undesired events within individual biographies, such as poverty transitions[2]. Social stratification determinants are considered social structural variables that are linked to positions within the social hierarchy and that are believed to shape individual biographies. Social class is considered to be the central social stratification determinant and is conceptualised in the sense of Erikson et al. (1979) as a combination of employment and occupational status which enduringly constitutes one’s relational position within the given labour market and production system in industrial societies. This paper investigates whether the social stratification determinants of education, social class, ethnic origin and gender are significant predictors of first-time social assistance transitions, their relative importance compared to the life course risks of unemployment, health problems and single parenthood, and whether the impact of the investigated life course risks varies for different social strata. The paper is structured as follows. First, it outlines the two theoretical perspectives and summarises the state of research. Then, the data set and the case-control methodology are described. The following section presents the variables that are used and interprets the multivariate results. In the final section, the results and limitations of the study are summarised, and directions for further research are given.

3. Social stratification, life course risks, and their relation to poverty

3.1 Theoretical considerations

In sociological research, certain determinants of social stratification, such as class affiliation or gender, have been considered the main factors within processes that explain social inequality. These determinants may be understood as resources for or barriers to a certain position within the social hierarchy, which is linked to the valued goods of a society, such as power and income (Breen, 1997) or, vice versa, to having considerably less of these goods and being poor. This approach assumes that the social structures that we inhabit have a substantial influence on our life chances (Alcock, 2004). Applied to poverty dynamics, this structural perspective suggests a significant and strong influence of social stratification determinants on poverty entry and exit. This perspective is reflected within various theoretical approaches that address poverty and social inequality, such as Marxism, with its focus on social class, or feminism, with its focus on gender. If this long tradition of sociological thought is still a valuable framework for contemporary poverty research, it requires a substantial impact of social stratification determinants on various indicators of poverty in modern societies, such as transitions into the last social safety net these societies provide, namely, means-tested social assistance. Less politicised theoretical approaches, such as (neo-)Weberian theory, also suggest a persistent importance of social stratification (e.g. see Sica, 2001).

In recent decades, this “traditional” perspective has been challenged by approaches that focus on life course risks. Ulrich Beck postulated a trend hypothesis that has been controversially discussed within sociological poverty research: the individualisation thesis. He stated that life course risks have become more important within the biographies of individuals in modern societies over time, whereas social stratification and, especially, social class have become less determinative. At the end of this development, these societies are supposed to have changed into what Beck calls a “risk society” (Beck, 1986), in which each individual can be affected by these risks at any time. Individualisation theory suggests that poverty no longer has a common societal cause in modern societies, such as discrimination by social class (Andreß and Schulte, 1998; Beck, 1986). These authors do not understand poverty as a result of social processes that have roots in social stratification; instead, they postulate a “democratisation of poverty”. Consequently, poverty should transcend traditional social boundaries and occur in biographies of individuals from various social strata. Applying this perspective to poverty entries and inspired by the methodology of dynamic poverty research (Bane and Ellwood, 1986), Leisering and Leibfried (1999) derived the biographisation hypothesis. This hypothesis states that poverty is far more likely to be induced by poverty-triggering circumstances that may occur within the life course than by social stratification determinants. Of these, social class is the central determinant that has produced the most intense discourse. A rigid interpretation of individualisation theory implies that class has lost its relevance as a social category. Some have even diagnosed a “death of class” (Pakulski and Waters, 1996) or a “zombie category” (Beck and Beck-Gernsheim, 2002).

3.2 State of research

The state of research shows some support for the individualisation perspective. Layte and Whelan (2002) theoretically applied the individualisation thesis to poverty and tested it empirically. The authors investigated various European countries by comparing the risk of relative income poverty of the manual working class compared to the non-manual classes. They found that this risk narrowed between 1989 and 1995 in countries such as the Netherlands, Germany and Luxembourg. Strong support for individualisation theory also comes from the studies of Leisering and Leibfried (1999), who used the framework of dynamic poverty research on claimants of former social assistance benefits in the German city of Bremen. Analysing two first-time social assistance inflow cohorts of this city, they applied the dynamic poverty research methodology, which was previously established for studies of relative income poverty in the USA. Their results identified a “risk society” in Germany in the 1980s and 1990s, with individualised and social boundary-crossing poverty (Leisering and Leibfried, 1999) in which social assistance served as an important final social safety net. This and other studies showed that most first-time social assistance episodes were accompanied by the occurrence of life course risks, such as health problems, changes in the family structure, and, especially, unemployment (Buhr and Voges, 1991; Leibfried et al., 1995). Life course risks have consistently been found to be important factors for social assistance entry or receipt in various countries. Empirical studies suggest a causal effect of family status and health on the social assistance entry probability for Sweden (Andrén and Gustafsson, 2004), Norway (Lorentzen et al., 2012), Britain (Cappellari and Jenkins, 2009) and Germany before the social assistance reform of 2005 (Castronova et al., 2001; Riphahn et al., 2013). In countries where the social assistance scheme is not directly linked to previous or current employment status, unemployment also appears to have a causal impact on the social assistance transition probability (Andrén and Gustafsson, 2004; Castronova et al., 2001; Lorentzen et al., 2012).

In contrast, some results suggest the persistent importance of social stratification determinants or contradict individualisation theory. Layte and Whelan (2002) found that although social class differences in income poverty risks narrowed over time in central European countries, they remained constant in countries such as the UK and Spain. For some countries, these differences even diverged. The authors concluded that “German social theorists have been partially right, though only in the context of their own country”. The study by Groh-Samberg (2004) raised doubts about the validity of studies on social assistance claimants in Bremen. Constructing a multi-dimensional measure of poverty that included the dimensions of deprivation, employment status, social assistance receipt and time, Groh-Samberg analysed representative German survey data for the 1996-2000 period. He found that poverty was highly structured by social class, and he harshly criticised individualisation theory. Conversely, Vandecasteele (2011) emphasised the complementarity rather than the contradictions between the structural and the individualisation perspectives with regard to transitions into relative income poverty in her study on thirteen European societies. She showed that although life course risks are strong predictors, the influence of social stratification is not negligible, especially in the case of social class. These results also appear to hold true for social assistance receipt as a poverty indicator in various countries (Andrén and Gustafsson, 2004; Cappellari and Jenkins, 2009; Castronova et al., 2001; Lorentzen et al., 2012). These studies have also found statistically significant interactions of social stratification determinants with life course risks, which had mostly weak effect sizes. Unfortunately, none of these studies have investigated the impact of social class.

At this point, it seems that no matter how individualised the modern European “risk societies” are, it is likely that social stratification determinants remain important. In sum, there is evidence for both the social stratification and the individualisation perspectives. However, a rigid interpretation of the latter in the sense of a “death of class” is not supported by the evidence.

3.3 Hypotheses

Considering the state of research, the hypotheses are derived as follows. Based on individualisation theory, it is assumed that Germany is a typical example of a “risk society” as described by Beck (1986) after the reforms of 2005 and that the life course risks of unemployment, (single) parenthood and health problems are deciding factors for transitions into poverty[3]. Transitions into the first social assistance receipt episode are investigated as a particularly good subject to test the individualisation perspective. The first transition should be strongly affected by life course risks, according to this theory. If social boundary-crossing poverty, as a typical phenomenon of a “risk society”, does exist, this must be visible at the first receipt episode, if not at all receipt episodes. However, due to the fact that there is considerable theoretical and empirical evidence for the persistent importance of social stratification, the current study combines the individualisation and the traditional perspectives, which leads to the following hypotheses:

H1.

Social stratification determinants should be statistically significant predictors.

but, on the other hand:

H2.

Social stratification determinants should be weaker predictors than life course risks.

In line with individualisation theory, it can be expected that different social strata are equally affected by life course risks so that:

H3.

There are no substantial interactions between social stratification determinants and life course risks.

4. Data and Methods

4.1 Data

The data used to test the hypotheses originate from the fourth wave of the Panel Study “Labour Market and Social Security” (PASS), which consists of several different samples (Trappmann et al., 2010). The first wave began in 2007 with one sample of households in which there was at least one person receiving social assistance in July 2006 and a stratified sample of the German residential population. In the second wave, an additional sample was drawn from the administrative social assistance data of the Federal Employment Agency, which contains households in which there was at least one person receiving social assistance in July 2007 but not in July 2006. Because this procedure was repeated in subsequent waves, three inflow samples of social assistance recipients are included in the Scientific Use File of the fourth wave. The population sample and the inflow sample of wave four will be the basis for the analyses. The latter contains only persons who received social assistance in July 2009 but not in July 2006, 2007 and 2008.

4.2 The case-control approach

The panel design of the survey might suggest fixed-effects panel regression models or other longitudinal methods that aim to reveal causal relationships. By using the population sample of PASS, these procedures might yield generalisable results about which causal risk factors for first-time social assistance transitions exist for the German working age population. Unfortunately, there are too few prospectively observed first-time social assistance transitions for such methods within this sample. The same holds true for other major panel surveys, such as the German Socio-Economic Panel. Administrative data of the Federal Employment Agency as used by Aldashev and Fitzenberger (2009), which would provide sufficient transitions, do not include the entire at-risk population and do not contain information about the household structure. Applying individualisation theory would thus not be possible because the impact of (single) parenthood cannot be assessed. To address these problems, the analysis follows the methodology of case-control studies (Schlesselman, 1982), which are used in epidemiology to identify risk factors for rare diseases. The problem with cohort studies, which is the methodology of choice for research about causal relationships between risk factors and diseases when experimental research is impossible in this research field, is that they often do not contain sufficient rare disease events. Within case-control studies, a sufficient number of persons who contracted the disease of interest (“cases”) are selected and interviewed to make quantitative analyses possible. Usually, these are retrospective interviews about exposure to perceived risk factors before the disease occurred. Additionally, a group of persons without the disease are interviewed (“controls”). The resulting estimation samples of case-control studies are characterised by an oversampling on the dependent variable; that is, the share of contracting persons is disproportionately high.

The empirical analysis in case-control studies is usually conducted using logit models with disease status as the dependent variable, which indicates a binary outcome (0 – affiliation with the control group; 1 – affiliation with the case group). Suspected risk and protection factors and control variables serve as predictors. The results of case-control studies cannot be causally interpreted (Breslow, 2005), but they are used to reveal statistical associations that, if strong enough, should be further investigated in longitudinal studies. The oversampling on the dependent variable in the logit model affects the estimated parameters in that the estimate for the constant term is incorrect, but all other slopes are unaffected (Donkers et al., 2003). Thus, all estimated coefficients (except for the constant term) can be interpreted in the same way as if there was a simple random sampling. A prominent example of this methodology is the causal relationship of smoking and lung cancer, which was first detected by case-control studies and then verified by longitudinal studies.

4.3 A case-control study on social assistance transitions

The case-control design can be applied to the PASS data to address the problem that the transition into first-time social assistance receipt is a rare event with correspondingly small transition numbers. The entry sample of the fourth wave of PASS serves as the case group. This group includes persons living in households that received social assistance in July 2009 but not in July 2006, July 2007, or July 2008. Using the social assistance spell data of PASS, all persons who had receipt episodes between or before these reference dates are excluded from the entry sample to retain only first-time inflows. In a second step, all persons receiving social assistance before and including July 2009 are excluded from the population sample. This sample of non-recipients serves as the control group. Furthermore, only persons who were of working age at the interview date and whose household members were all successfully interviewed are included in the estimation sample. This approach leads to a suitable sample size, that is, a sample of 2,988 persons of approximately 39.7 million persons at risk of first-time social assistance entry (own calculations using PASS, see Table I).

4.4 Method

Logit models are estimated using the fourth wave of PASS with group attachment as the outcome (0 for the control group, 1 for the case group). These models distinguish persons who entered social assistance receipt from persons who did not enter social assistance receipt between August 2008 and July 2009 for the part of the German residential population that had not previously received social assistance. The aim is to reveal which factors trigger or prevent the first transition into social assistance receipt. The analysis is restricted to persons of working age in July 2008, which is the population at risk of starting social assistance receipt according to the legal regulations. The aim is not to quantify the exact causal effects of the investigated predictors, which is not possible because data for this purpose are not available for Germany. Instead, this study investigates the strength and significance of statistical associations and their predictive value.

5. Multivariate results and discussion

5.1 Variables

In the first model, social stratification determinants are investigated without accounting for life course risks. Social class is included using a slightly modified version of the Goldthorpe scale (Erikson et al., 1979), with the higher professional-managerial class as the reference category. Social class is measured at the household level using the dominance principle (Erikson and Goldthorpe, 1992), which assigns the class position of the household head to each household member. In the case of households with no employed household head in July 2007, the class position of the most recent employment spell of any household member within two years before July 2007 is assigned. Migration background (measured by personal or parental immigration), gender and education are also included in the model and relate to the household head. A reduced version of the CASMIN classification (Brauns and Steinmann, 1999) serves as the measure for education: 1a (inadequately completed general education) and 1b (general elementary education) are combined to the category of low educational level, which serves as the reference category; 1c (basic vocational education), 2a (intermediate vocational qualification), 2b (intermediate general qualification), 2c_gen (general maturity certificate), and 2c_voc (vocational maturity certificate) are defined as the medium educational level; and 3a (lower tertiary education) and 3b (higher tertiary education) are combined into the category of high educational level.

The second model investigates the effects of life course risks without accounting for social stratification determinants. A combined parenthood and partnership status variable that measures the family context is included in the model to account for the case of single parenthood, where being single without having children under the age of 15 in the household serves as the reference. Unemployment is differentiated by its length: no short-term or long-term unemployed individual living in the household is the reference. A variable that indicates whether there is at least one person with severe health impairment in the household is also included.

In the third model, social stratification determinants, life course risks, and a set of control variables are included simultaneously. Age is included as a categorical control variable using the 44-53 year old age group as the reference. The place of residence also serves as a control variable to differentiate between West Germany and East Germany (the former German Democratic Republic). Finally, interactions between social stratification determinants and life course risks are included in the third model.

5.2 Stratification vs life course risks – what predicts better?

Table I displays the results of the three multivariate models. In the first model, only social stratification determinants are considered. In the second model, only the impact of life course risks is investigated. In this way, a comparison of the explanatory power is possible. The third model includes both elements and a set of control variables. To allow comparisons of the magnitude of effects between the models, which is not possible with logit coefficients or odds ratios, average marginal effects (AME) are reported[4]. The AMEs indicate the average effect of an independent variable on the probability of beginning social assistance receipt over all observations.

Model 1 indicates that all investigated social stratification determinants have a significant impact when we do not control for life course risks. The significant AME of 0.08 indicates that on average, individuals living in female-headed households have a 8 per cent higher probability of beginning social assistance receipt compared with individuals living in a male-headed household. Higher education is a protective factor that reduces the probability of beginning social assistance receipt (p<0.01). Migration background is a significant risk factor that increases the probability by 14 per cent on average. The results for social class, for which the discourse between individualisation theorists and advocates of the social stratification perspective has been most intense, are unambiguous. The unskilled manual working class has, on average, a 24 per cent higher probability of experiencing a first-time social assistance transition than the higher professional-managerial class does. Therefore, the results are more in accordance with Groh-Samberg’s (2004) diagnosis of a strongly poverty-threatened German working class rather than individualisation theory. However, the self-employed class is also strongly affected. Its probability is, on average, 21 per cent higher compared to the higher professional-managerial class. Overall, social stratification predicts social assistance transitions quite well, which is indicated by a good model fit (McFadden’s R²=0.126, BIC=2,671.4). That is, rigid interpretations of individualisation theory in the sense of social stratification as a negligible factor are incorrect. Despite these clear results, it might be that social stratification determinants are correlated with life course risks so that the revealed effects could be negligible and originate only from life course risks. This possibility will be investigated in the third model.

Model 2 indicates that life course risks also have a substantial impact, at least when social stratification determinants are not considered. Short-term and long-term unemployment are significant risk factors (p<0.001). Regarding health, a severe health impairment of one or more household members is also a substantial risk factor (AME=0.20). Concerning the impact of children, living in a single-parent household increases the probability by 18 per cent on average compared to a household with a single person and no children under the age of 15. The AMEs of couple households with and without children are basically the same (−0.22 vs −0.26). It seems that parenthood is a risk factor only for singles.

A comparison of Models 1 and 2 via BIC provides evidence for the relative importance of social stratification determinants and life course risks. Individualisation theory implies that life course risks are the main factor and are more important than social stratification determinants. This notion should be reflected by a better model fit of Model 2 compared to Model 1, which is indeed the case (2,524.2 vs 2,671.4). The BIC discrepancy is significantly above the value of 10, which according to Kass and Raftery (1995), provides “very strong evidence” for a better model fit. Thus, H2 cannot be rejected. Nevertheless, social stratification determinants are strong predictors, which contradicts rigid interpretations of individualisation theory (“death of class”).

At this point, one could argue that the empirical results might be sensitive to what is conceptually understood as a social stratification determinant or life course risk. This possible sensitivity can be illustrated for the case of age. Some theorists consider age an important social stratification determinant (Turner, 1998), whereas, others doubt that age can be understood as a social stratification determinant (Irwin, 1996). In the current study, age is not considered a social stratification determinant but is included in the third model as a control variable. However, further analyses revealed that if age is included as part of the social stratification determinants model, the BIC value decreases to 2,586.9. This BIC value is still far above the BIC value of the life course risks model (2,524.2), which means that the life course risks model has better predictive power and that the contentual result of life course risks as better predictors of social assistance transitions persists.

Model 3 includes all variables simultaneously. The results contradict the notion of the irrelevance of social stratification determinants because all investigated determinants except gender continue to have a significant impact. Concerning the impact of life course risks, all of these continue to have a significant impact with the exception of the health impairment variable, which becomes insignificant. However, the AME of 0.13 still points in the expected direction, suggesting that health impairments of household members raise the probability of a social assistance transition. The insignificant estimate might be attributed to the low number of persons in the sample who have a household member with a severe health impairment (the sample contained 22 respondents who had a household member with a severe health impairment).

5.3 Does the impact of life course risks vary by social strata?

Table II shows the results of the introduction of the respective interaction term(s) into Model 3. The first column shows the social stratification determinant (italicised) and the life course variable with which the interaction terms are built. The third column indicates whether any of the resulting interaction terms is statistically significant at the five per cent level. For example, as seen in the third column, at least one of the two resulting interaction terms between migration background and unemployment status are statistically significant. The logit coefficients of the interaction terms “migration background×short-term unemployment” and “migration background×long-term unemployment” are −1.28 (p<0.01) and −1.70 (p>0.05) (results not shown due to the high total number of interaction terms that would have to be listed)[5].

The fourth column shows the results of the likelihood ratio tests (LR tests) with which the baseline model and the baseline model with the respective interaction term(s) are tested against each other. Under the null hypothesis, the explanatory power of both models does not differ. For instance, concerning the interaction between the ethnic origin of the household head and the unemployment status, the introduction of the interaction terms significantly improves the explanatory power of the model (χ²=9.69; p<0.01).

However, the statistical significance of the interaction terms and the LR test together are not sufficient to answer the question of whether there are substantial interactions between social stratification determinants and life course risks. The higher explanatory power of the models with interaction terms could originate from the fact that they are more strongly parameterised than the baseline model. Therefore, it is useful to compare the BIC of the fully specified models to the BIC of the baseline model. One can only rule out such a parameterisation misapprehension if the BIC does not increase substantially due to the inclusion of additional variables. Column 5 shows the BIC of the interacted models, and column 6 shows the resulting difference to the BIC of the baseline model. For the model that includes the significant and model-improving interaction terms of education and family type, the BIC of 2,270.3 is considerably worse than in the baseline model (ΔBIC=20.5). It is obvious that this interaction enhances explanatory power only through stronger parameterisation. Nevertheless, the significant and model-improving interaction of ethnic origin and health status only moderately increases the BIC (ΔBIC=3.5). Moreover, the introduction of the interaction term of gender and family type even decreases the BIC (ΔBIC=−17.3). Therefore, it can be concluded that contrary to what individualisation theory suggests, the impact of life course risks does vary substantially for different social strata.

6. Summary and conclusion

This paper investigated the predictive power of social stratification and individualisation theories for social assistance transitions as a poverty indicator. The study utilised German survey data with a large share of social assistance recipients to investigate first-time social assistance entries. Although the longitudinal design of these data suggested the use of fixed-effects models or other methods that aim to reveal causal relationships, the empirical investigation had to be performed indirectly because there are too few prospectively observed transitions into first-time social assistance receipt in German panel surveys. The study applied case-control methodology to circumvent this problem, and the empirical analysis indicated that life course risks such as unemployment, health problems, and single parenthood have a strong impact on the transition probability. Contrary to the prediction of individualisation theory, these risks do not have the same poverty-triggering effects for all investigated social strata. This result is in line with the evidence for income poverty, which also found significant interactions of life course risks with social stratification determinants (e.g. Whelan and Maître, 2008), and underlines the importance of investigating multiple forms of poverty to retain a complete picture. Future research should apply the social stratification and individualisation perspective to poverty indicators that have not yet been investigated from these perspectives (e.g. subjective poverty measures).

The current study also shows that the traditional sociological perspective remains relevant because social stratification determinants (ethnic origin, education level, and especially social class) continue to have a strong influence on an important poverty indicator. However, life course risks have higher determinative power. The social processes that lead into indigence are obviously linked both to the occurrence of life course risks and to social stratification. This study and other studies show that focussing on the complementarities instead of the contradictions between the two perspectives is a suitable framework for poverty research. However, rigid interpretations of individualisation theory have been found to be wrong. This situation calls for more moderate theoretical interpretations of individualisation (e.g. Mayer, 2004). In summary, scholars should integrate both perspectives and eliminate the remaining contradictions within a comprehensive sociological theory of poverty.

This study has certain limitations. The associations revealed cannot be causally interpreted due to the case-control design. Nevertheless, the case-control design is fruitful and should be transferred to social policy research, where it is uncommon and may be unknown to many scholars. It allows a quantitative investigation of processes that are, in some countries, infrequently observed in longitudinal data but highly relevant to socio-political issues, such as the transition into social assistance receipt. In the current study, case-control methodology enabled the detection of strong associations that suggest causal relationships and that should be examined in further research. However, the design made it difficult to investigate the impact of life course risks as biographical events because no suitable longitudinal data are available for the German social assistance scheme. In the current study, life course risks concerning health, family situation and employment were conceptualised as critical statuses in which individuals may find themselves within their biographies. Investigating life course risks as events in the form of job losses, childbirths, partnership dissolutions and health deteriorations revealed interesting results for some measures of poverty (e.g. Vandecasteele, 2011); this approach should also be applied to transitions into social assistance as a further poverty measure. This longitudinal perspective can reveal which social processes actually cause people to apply for the final social safety net and improve our understanding of poverty dynamics.

Table I

Logistic regression models, average marginal effects (AME), dependent variable: first-time entry (1) / no first-time entry into social assistance receipt (0)

	Model 1	Model 2	Model 3
Social stratification determinants
Education level hh head
Low	RC		RC
Average	−0.09*		−0.11**
High	−0.14**		−0.15
Social class of the hh
Higher professional-managerial	RC		RC
Lower professional-managerial	0.00		0.01
Self-employed	0.21***		0.20***
Higher routine non-manual	0.04		0.02
Skilled manual	0.07*		0.05
Lower routine non-manual	0.20***		0.17***
Unskilled manual	0.24***		0.16***
Ethnic origin hh head
No migration background	RC		RC
Migration background	0.14***		0.14***
Gender hh head
Male	RC		RC
Female	0.08*		0.01
Life course risks
Health
No hh member with severe health impairment		RC	RC
⩾1 hh member(s) with severe health impairment		0.20*	0.13
Unemployment
No unemployed hh member		RC	RC
Short-term unemployed, but no  Long-term		0.33***	0.27***
Unemployed hh member(s)
⩾1 long-term unemployed hh member(s)		0.41***	0.33***
Family structure
Single without children <15 years		RC	RC
Single-parent hh		0.18**	0.14**
Couple without children		−0.26***	−0.20***
Couple with children		−0.22***	−0.18***
Control variables
Age
15-23			0.04*
24-33			0.13***
34-43			0.01
44-53			RC
54-62			0.02
Place of residence
West Germany			RC
East Germany			0.06**
Model summary
Sample size	2,988	2,988	2,988
Number of retrospective events	585	585	585
Pseudo R2 (McFadden)	0.126	0.165	0.300
BIC	2,671.4	2,524.2	2,249.8

Notes: Estimation sample: 15-62 year old persons who did not receive social assistance in July 2006, July 2007, or July 2008. Pupils, students or persons in professional training are excluded. *p<0.05, **p<0.01, ***p<0.001

Source: Scientific Use File PASS, wave 4 (own estimations)

Table II

Results of the models with interaction terms; baseline model is Model 3 (see Table I)

	Sign. interaction?	LR test χ2	BIC	ΔBIC
Educationa
Impaired health	No	0.50	2,263.8	14.0
Family type	Yes	27.41***	2,270.3	20.5
Unemployment	No	0.31	2,254.9	5.1
Social classa
Impaired health	No	0.40	2,340.4	90.6
Family type	Yes	44.44***	2,349.3	99.5
Unemployment	Yes	17.09	2,315.7	65.9
Migration backgrounda
Impaired health	Yes	4.46*	2,253.3	3.5
Family type	Yes	9.18**	2,264.5	14.7
Unemployment	Yes	9.69**	2,256.0	6.2
Femalea
Impaired health	No	0.37	2,257.3	7.5
Family type	Yes	41.21***	2,232.5	−17.3
Unemployment	No	3.87	2,261.8	12.0

Notes: Estimation sample: 15-62 year old persons who did not receive social assistance in July 2006, July 2007, or July 2008. Pupils, students or persons in professional training are excluded. *p<0.05; **p<0.01; ***p<0.001

Source: Scientific Use File PASS, wave 4 (own estimations)

Notes

1. In administrative terms, the reference is “benefit units”. This unit includes the employable individual and, if present in the household, the partner and unmarried children under the age of 25. For simplicity, this paper uses the term “households”.

2. Poverty can also be regarded as a life course risk, which should be associated with the occurrence of other life course risks like unemployment. See Heyne (2012) for this relationship.

3. The current study draws on the picture of a risk society, which does not require assumptions about the occurrence of social change that leads to this type of society. Thus, it does not implicitly take the individualisation thesis for granted, a deficit of various studies that adopt the individualisation perspective (Burzan, 2011).

4. Any other coefficients of interest (e.g., standardised regression coefficients) can be obtained from the author.

5. The estimates of the numerous interaction terms can be obtained from the author.