(Un)equal from the start? A quantitative analysis of preschool children’s participation in organised activities in Germany

ABSTRACT In this article, I investigate preschool children’s participation in organised activities. Current political and academic debates consider informal education as a prime vehicle for potentially diminishing social class inequalities in educational outcomes before school entry. However, studies point to unequal participation rates between social classes, which means the activities might actually aggravate existing disparities. Various explanations have been offered for this social class gap. Some scholars argue that material resources play a pivotal role, while others say that culture is the decisive factor. This study uses the kindergarten cohort of the German National Educational Panel Study (NEPS) to test how far these two dimensions contribute to social class differences in preschool children’s participation in organised activities. My analysis shows that both dimensions are important determinants of children’s participation in organised activities. However, occupational characteristics also have a considerable effect, which suggests shortcomings in the current scholarly discussion.


Introduction
Historians and social scientists have frequently pointed out that childhood is a social construct which is not fixed through time or space (Ariés 1962;Keller et al. 2006). Children are an 'ideal projection screen for a society's self-perception, its dreams, and its visions for the future' (Gebhardt 2009, 13, my translation), and as such they are subject to competing demands and standards (Ramaekers and Suissa 2012). However, scholars have identified common components of a hegemonic model of a 'good' childhood in various societal spheres such as politics (Betz, de Moll, and Bischoff 2013;Gillies 2012) and science (Putnam 2016, 109-117). The emerging ideal has been termed 'intensive mothering' (Hays 1996), 'concerted cultivation' (Lareau 2011(Lareau [2003) and 'parenting for cognitive development' (Schaub 2010). Following Beck-Gernsheim (1997), this model has the following main drivers: science, politics, (education) professionals, and childrearing manuals (109f.). However, it has also been picked up by market actors (e.g. toy stores and the preschool enrichment market; see Smyth 2016; Vincent and Ball 2007). Looking at current debates on parenting, one gets the impression that these kinds of parenting practices have reached unexpected heights .
Firstly, policymakers have identified parenting as 'a mechanism for tackling social ills' (Gillies 2012, 13). It is thought that interventions in parenting might mitigate educational disparities from the outset (Hartas 2015), and various countries have proposed government programmes designed with this in mind (Daly 2013;Gillies 2008). This development has led scholars to speak of a 'politicization of parenthood' Richter and Andresen 2012). Secondly, hopes surrounding the equalising effect of informal education are fuelled by scholars who support interventions during early childhood (Edwards, Gillies, and Horsley 2016;Gillies 2012;Lee 2014). For instance, Nobel laureate Heckman (2008) stresses that '[g]aps in the abilities (…) open up very early across socioeconomic groups' (298) and that investments made during early childhood thus provide 'high benefit-cost ratios and rates of return' (ibid., 290; see also Esping-Andersen 2008). Thirdly, popular education guidebooks and parenting magazines promote the idea of encouraging cognitive development from an early age (BMFSFJ 2013, 100-104;Quirke 2006;Wrigley 1989).
In all these spheres, a prominent role is assigned to organised activities (Vincent and Maxwell 2016). These activities could potentially diminish disparities (Vandenbroeck and Lazzari 2014). A case in point are music lessons that have been shown to exert positive effects on school grades and cognitive skills (Hille and Schupp 2015;Mühler and Spieß 2008). However, organised activities could also be 'building blocks in class care strategies' (Stefansen and Farstad 2010, 121) and hence constitute 'a mechanism through which social inequalities are maintained and reproduced' (Bennett, Lutz, and Jayaram 2012, 131).
Given these contradictory perspectives, the scarcity of empirical evidence on the actual use of these activities is quite surprising. Yet despite the lack of evidence, political programmes have already been initiated (Gillies 2008). What is more, there is a danger of scapegoating less affluent and single parents. In this vein,  asserts that 'problems that would once have been conceived of as structural in origin, such as poverty, inequality, poor educational progress, or the ill health associated with social deprivation, have now come to be attributed to parental behavior' (83f.). This transformation has been described as an 'individualization of social class' (Gillies 2005). Against this background, the article proposes to take a step back by asking which factors contribute to the differential uptake of organised activities.
Germany is a prime case for studying parenting in times of changing demands. It has been pointed out that educational trends are often imported into Germany with a time lag (Gebhardt 2009). Early childhood education and care (ECEC) is a prominent example. In Germany, ECEC only recently gained attention from policymakers and became an aim of social policy (Betz 2012;Ostner and Stolberg 2015). Conflictingly, however, government documents also stress that the cognitive stimulation of children is primarily the responsibility of parents (Betz, de Moll, and Bischoff 2013).
By using the German National Educational Panels Study (NEPS) (Blossfeld, von Maurice, and Schneider 2011), my study contributes to the debate by analysing the empirical distribution of organised activities through 'a social class lens' (Hartas 2015, 33). I will begin by reviewing the current state of the field (Section 2) in order to trace the main lines of argument. I will then present the hypotheses and concepts (Section 3), introduce the data (Section 4) and present my results (Section 5). I will conclude by summarising the findings and integrating them into the scholarly discussion (Section 6).

State of the field
Although the relationship between social class and parenting attracts considerable attention, scholars do not share a common understanding of the underlying mechanism. Following Sherman and Harris (2012), a major dividing line is 'whether structural conditions or cultural understandings are the more important influences' (60).
The most prominent adherent of the cultural approach is certainly Anette Lareau (2011Lareau ( [2003). Based on ethnographic fieldwork in the US, Lareau (2011Lareau ( [2003) found that middle-class and working-class parents raise their children according to two distinct logics of childrearing: 'concerted cultivation' and 'accomplishment of natural growth', respectively. While middle-class parents offer their children a myriad of organised leisure activities, elicit their opinions and put emphasis on a good rapport with teachers, their working-class counterparts use more directives, maintain a looser schedule for organised activities and are less involved in institutional settings (Lareau 2011(Lareau [2003). This conceptualisation shows that organised activities are a core dimension (Vincent and Maxwell 2016).
Quantitative studies conducted in the US confirmed and expanded on Lareau's findings (Bodovski 2010;Bodovski and Farkas 2008;Cheadle and Amato 2011). In addition, Lareau herself was part of a team that examined her hypothesis by using quantitative time-survey data Weininger, Lareau, and Conley 2015). Weininger, Lareau, and Conley (2015) found that household income is positively related to annual expenditure on organised activities, but that maternal education level plays an even bigger role. Regarding their second dependent variable, 'weekly time spent in organised activities', the researchers found that income fails to reach significance while education still exerts a positive effect. The authors interpret the positive effect of maternal education as a cultural effect (Weininger, Lareau, and Conley 2015, 498f.).
Another strand of research considers parenting as an investment of time and money (Kornrich and Furstenberg 2013, 2). From this perspective, constraints on resources are thought to impede parental investments. While time investments are mainly required for managing children's schedules, monetary investments are assumed to be far more important. Regarding the latter, household income emerged as a strong predictor in the US (Kornrich and Furstenberg 2013). Using a qualitative sample of eighth-graders and their parents in the US, Bennett, Lutz, and Jayaram (2012) found that financial constraints were more significant than culture. This conclusion is also reached by Chin and Phillips (2004), who report that the social class gap in summer break activities stems mainly from differences in financial situations.
However, a major problem with the aforementioned approaches is that they treat social class as a kind of nuisance that persists after its material or cultural aspects have been stripped away. Even if a measure of social class position is used, it is rarely the focus of any theorising. Nonetheless, resources and culture are far from being the whole story when it comes to social class. 1 And in fact, a third approach is interested in the effect of occupational conditions on parenting (i.e. occupational effects). Following Kohn (1963) and Kohn andSlomczynski (1993 [1990]), the position in the occupational structure is correlated with certain demands of the jobi.e. to obey authority or exercise self-direction. These demands are, moreover, reflected in differing value commitments which then steer childrearing practices (see also Chan and Koo 2011). For instance, Weininger and Lareau (2009) have shown that middle-class parents perceive organised activities as a way of encouraging independence in their children, and as a way of subtly controlling the course of their children's curiosity and (self-)direction.
With regard to Germany, research on children's organised activity participation is still scarce (De Moll and Betz 2014, 238). However, using the German Socio-Economic Panel (GSOEP), Hille, Arnold, and Schupp (2013) report increasing rates of adolescent involvement in education-related leisure activities over time. This indicates that organised activities are becoming more important in Germany. In addition, by analysing Families in Germany (FID) data, Schroeder, Spieß, and Storck (2015) found that household income and parental education are positively related to monetary expenditure on children's formal and informal education. Interestingly, expenditure is highest for younger children, who are the target group of this study (Schroeder, Spieß, and Storck 2015). Mühler and Spieß (2008) also analysed GSOEP data and found that maternal education and household income are positively related to preschool children's participation in organised activities. They found negative effects for children whose mothers are in fulltime employment or who were born in foreign countries. The number of siblings also has a negative effect. The number of books in the household, a common measure for cultural capital, does not have a statistically significant effect (Mühler and Spieß 2008). In a similar study, Schober and Spiess (2013) indicate that differences persist between the former West and East Germany. A more recent study by De Moll and Betz (2014) used data from the Growing up in Germany survey (AID:A) to examine ECEC arrangements of parents with preschool children. Their analysis showed that working-class parents and immigrants are less likely to enrol their children in organised activities (De Moll and Betz 2014).
While De Moll and Betz (2014) focus on general enrolment in activities over and above kindergarten attendance, and Mühler and Spieß (2008) focus on the general tendency to participate in at least one activity, the present study adds nuance by considering different organised activities and hence supplements both of these studies. More importantly, the field is divided into scholars who focus on the structural constraints that less affluent parents face, and scholars who focus on the role of culture. To the detriment of the field, however, scholars on both sides rarely apply methods that can separate the effect of social class (if they consider it at all) into its cultural and material parts. This study attempts to enrich the debate by investigating whether money or culture matters (more).

Conceptualisation and hypotheses
This paper intends to shed light on how social class influences the organised activities promoted by the current ideal of parenting as cognitive stimulation. I have identified three common perspectives on this social class effect. First, scholars emphasising the role of material constraints argue that parents enrolling their children in organised activities incur considerable costs. As Vincent and Maxwell (2016) put it laconically, 'these classes cost and some cost a lot' (274). Accordingly, I hypothesise that there will be lower participation rates among children with less affluent parents (H1).
While the rationale for considering material resources is quite straightforward, the influence of culture remains open to interpretation (Weininger, Lareau, and Conley 2015, 498f.). How do different logics of childrearing come about? In a dense passage, Lareau (2011Lareau ( [2003) argues that education professionals play a major role in formulating the expectations that are placed on parents. She stresses that these expectations form a dominant set of cultural repertoires about how children should be raised. This widespread agreement among professionals about the broad principles for child rearing permeates our society. A small number of experts thus potentially shape the behavior of a large number of parents (Lareau 2011(Lareau [2003. In line with this interpretation, Schaub (2010) underlines that these 'legitimate' expectations are embedded in the education system, which not only impacts children but also their parents. Following this argument, maternal educational level can be understood not only as a resource in itself but also as a measure of exposure to 'legitimate' expectations about how to raise a child. I therefore assume that parents with a higher level of education are likely to be more engaged in the cognitive development of their child through organised activities than parents with a lower level of education are (H2.1). Furthermore, I apply a measure of cultural capital to shed light on intra-class heterogeneity. Following Jaeger and Breen (2016), parents with more cultural capital must transmit their cultural resources to their offspring in order to reproduce it. Thus, I hypothesise that parents with more cultural capital 2 will enrol their children in more organised activities than parents with less cultural capital (H2.2).
Besides these broader debates, scholars have criticised the fact that the term 'social class' often boils down to an 'umbrella concept that aims to capture all the various aspects of social inequality that we know exist in contemporary societies' (oral conference contribution by Goldthorpe cit. in Lareau 2008, 11, emph. i.o.). In the present study, I address this issue by examining the effect of different occupational conditions on participation in organised activities as a core variable. In addition to the indirect effect of social class through culture and resources, I assume that an occupational effect of social class persists. Accordingly, I hypothesise that parents in professional jobs will enrol their children in more organised activities than parents in intermediate and manual jobs (H3).

Data and estimation strategy
My study uses the German National Educational Panel Study (NEPS), which applies a multicohort sequence design in order to shed light on educational processes throughout a person's life (Blossfeld, von Maurice, and Schneider 2011). 3 The NEPS provides highquality data that complies with the comprehensive German legislation on data protection (Meixner et al. 2011). In particular, Starting Cohort Kindergarten (SC2) provides information on parental activities for preschool children. So far, data on six waves are available. The second wave includes information on organised activity participation among preschool children, and is thus the focus of this study.
As no direct sampling frame exists for kindergartens in Germany, the NEPS applies an indirect sampling approach that uses a link between elementary schools and kindergartens in order to create a sample of kindergarten children (Steinhauer et al. 2015). In more detail, this approach involves the following steps (Steinhauer et al. 2015, 133). Firstly, an existing sampling frame provided by the Federal Statistical Office of Germany was used to draw a nationwide sample of schools that enrol first-grade students. These schools were then asked to specify which kindergarten institutions their pupils attended previously to enrolment at school. Finally, this information rendered it possible to select a sample of German kindergartens.
As a result, the sample is a multilevel structure, with preschool children being nested in kindergartens. The independence assumption of linear regression is thus violated, and the data structure requires multilevel modelling or standard error correction for clustered data. I will use the latter approach for my analysis, as I have no hypothesis for the impact of the higher level.
I imputed missing values using a multiple imputation approach. The algorithm is described in King et al. (2001). Even though the number of missing values is quite low (listwise deletion would remove 284 of the 1771 cases), this approach retains a full set of observations and avoids the bias that has been shown to arise with more traditional methods, such as casewise deletion (Johnson and Young 2011). Following results from a simulation study, I calculated 25 multiply imputed datasets using all variables of the analysis model (Johnson and Young 2011). I then ran all multivariate analyses on each of the imputed datasets. Finally, I combined the point estimates and standard errors according to Rubin's rules (1987).
In terms of the analytical strategy, I took a two-part approach. Firstly, I applied logistic regression with cluster-corrected standard errors in order to retain a full model with the effects of social class, culture and material resources while considering control variables. I give the coefficients as average marginal effects (AME) because their interpretation comes closest to standard OLS interpretation. In addition, the substantive interpretation of logits, odds ratios (OR) and even relative risks (RR) is not straightforward (Best and Wolf 2012).
Another pitfall that has hindered scholars in their attempts to break down the social class effect into its material and cultural dimensions is the problem of rescaling in nonlinear models (Karlson, Holm, and Breen 2012). Accordingly, coefficients of nested models are not comparable as they are in the case of OLS regression. However, the Karlson-Holm-Breen method (KHB) (Karlson, Holm, and Breen 2012), which Kohler, Karlson, and Holm (2011) implemented in the statistical software package Stata, separates the total effect of an independent variable into its direct and indirect effects on an outcome variable. The method thus makes it possible to report the magnitude of the effect that social class has on organised activity participation broken down according to culture and monetary constraints. This approach is summarised in Figure 1 below: Figure 1 shows that the total effect of social class on organised activity participation consists of a direct effect (interpreted as occupational effect) and two indirect effects that run through culture and material resources. Hence, the direct and indirect effects sum up to the total effect of social class.

Independent variables
As shown in the literature review, some scholars see parenting as investments of material resources (Kornrich and Furstenberg 2013). To identify the effects of constraints on these resources, I have included the net household income in euroslog-transformed to normalise its distribution. I did not apply any adjustment for household size because household composition is incorporated into the model as a control variable.
Other scholars emphasise the effect of culture, so I included a measure of maternal educational level according to the CASMIN classification. Specifically, I distinguished between three groups: (1) a group with only elementary education, (2) a group with secondary education and (3) a group with at least one tertiary qualification. Figure 2 provides information on the distribution of all independent variables using appropriate sampling weights.
In addition, I constructed a latent variable in order to measure familiarity with highbrow culture. It is based on three manifest variables which measure participation in cultural events such as visiting (1) museums, (2) operas or classical concerts and (3) theatre performances on an ordinal scale ranging from never to more than five times in the last year. In order to derive a single latent variable, I inspected the three indicators using a Kernel smoothing approach (Mazza, Punzo, and McGuire 2014). This showed that a better fit is achieved when aggregating the item responses into three categories that indicate whether the above activities never happened (=1), happened between one and three times (=2), or happened more than five times (=3) in the last twelve months.
Finally, I applied a generalised partial credit model (GPCM) for polytomous data (De Ayala 2009). A parametric bootstrap goodness-of-fit test indicated a good fit for each dataset. This produced a single latent score which measures affinity to highbrow culture. Unfortunately, this information was only available for the main respondent of the questionnaire. Furthermore, the information stems from wave 1, as participation in highbrow activities is not included as a panel question. Nonetheless, a comparison of wave 1 with wave 3, when the question was asked again, indicates that the responses are quite stable.
As a measure of social class, I applied the Erikson-Goldthorpe-Portocarero (EGP) class scheme in an aggregated four-class format. The EGP scheme has the advantage of closely resembling Kohn's understanding of occupational conditions, which include 'the substantive complexity of work typically performed, the closeness of supervision that the individual experiences, and the degree of routinization (…) of his or her job activities' (Weininger and Lareau 2009, 681). The EGP is based on the difficulty of monitoring and the specificity of assets of job tasks as its defining features (Goldthorpe 2007(Goldthorpe [2000), and hence constitutes a good proxy for Kohn's thesis. It distinguishes between a service class (EGP I/II), an intermediate class (IIIa/V), a petty bourgeoisie (IV) and a manual class (IIIb/VI/VII) (Breen 2005). Class is assigned on the basis of the person with the highest social class in the household.

Control variables
In addition to these core variables of interest, my analysis also controlled for the place of residence (the former East Germany (GDR, coded as 1) or West Germany), the child's gender and the mother's age. As there is considerable debate on the role of immigration status, I constructed a dichotomous variable that differentiates between families where at least one person was born outside Germany (coded as 1) and families where all members are autochthonous. The model also includes the number of siblings in order to identify potential resource dilution between siblings (Strohschein et al. 2008).

Dependent variables
The dependent variables are four items that indicate whether children regularly attend certain organised activities outside kindergarten. Figure 3 shows that organised activities are quite common during kindergarten. Taking all activities into account, participation ranges from 73% (manual class) to 96% (service class). However, there is considerable variation between activities. The high participation rates are mainly for sporting activities, while activities such as music lessons differ heavily according to social class. In the following analysis, I will begin by focusing on the general probability of attending at least one of the activities. I will then analyse the other activities in more detail.

General participation in organised activities
I will assess the probability of participating in organised activities outside kindergarten by applying a logistic regression with cluster-corrected standard error. I will begin by discussing the full model, which includes all independent variables and controls (see Figure 4). The results are reported as AME. All analyses were estimated using 25 imputed datasets. I present the full model as a dot plot which visualises the point estimates and 95%-confidence intervals. Coefficients that do not overlap with the vertical line have significant effects on the probability of organised activity participation. In addition, the more detailed regression tables (Table A1) are provided in the Appendix. Regarding the results provided in Figure 4, my analysis confirms that both monetary resources and culture are statistically significant in determining children's organised activity participation. However, a more substantial interpretation shows that, on average, a 10% increase in income results in a minor increase (0.76 percentage points) in the probability of participating in organised activities (H1). Regarding the impact of maternal educational level (H2.1), mothers with secondary education are, on average, 6% more likely to enrol their children in informal education than mothers with elementary education are. However, the difference between mothers with elementary education and mothers with tertiary education is not statistically significant. In contrast, cultural capital has a significant and substantial effect (H2.2). Holding other variables constant, an increase in cultural capital from one standard deviation below its mean value (−0.69) to one standard deviation above (0.79) results, on average, in an 11 percentage points increase in the probability of organised activity participation.
Looking at the occupational effects proposed in H3, differences between the service class and the manual class persist even after taking structural and cultural constraints into account. More precisely, children of parents from the service class are, on average, 10 percentage points more likely to participate in organised activities than children of parents from the manual class. The same holds for children of parents from the intermediate class, who are still 9 percentage points more likely to participate in organised activities than children from the manual class. Thus, H3 finds support even after monetary resources and culture have been included.
Regarding the control variables, there is evidence to support the resource-dilution hypothesis, i.e. that having siblings decreases the probability of participating in organised activities.
In order to trace the confounding of the social class effect by culture and material resources, I used the KHB method. Table 1 provides information on the total effect (reduced model) and direct effect (full model) of social class on organised activity participation. The difference between these two coefficients indicates the strength of the indirect effect that runs through either culture or material constraints. In addition, column four shows the degree of confounding expressed as a percentage.
The differences between the manual class reference category and the petty bourgeoisie, the intermediate class and the service class decline considerably once the measures for constraints on resources and culture are added. With the petty bourgeoisie, the difference ceases to be statistically significant. As for the intermediate class and the service class, 29.45% and 40.57% of the net social class gap is attributable to differences in resources and culture. However, a social class gap persists even after adding those variables. The KHB method also makes it possible to separate the confounding of social class into its individual components. The individual confounding percentages are provided in Figure 5 and show, again, that material resources and culture do matter, with cultural capital being a particularly strong mediating effect. In contrast, maternal educational level is of minor importance compared to cultural capital and material resources.
Overall, the analysis tells us that children born into families with a service class background, higher cultural capital and education, and greater monetary resources are more likely to participate in organised activities. In this case, we can rightfully speak of unequal opportunities from the start. Nonetheless, it has also been emphasised that '[w]e still do not know much about (…) which groups of parents enrol their children in what kinds of activities' (Vincent and Maxwell 2016, 277). I will address these questions in the following section.

Specific activities
Having provided an overview of the probability of participating in at least one organised activity of any kind, I will now analyse the particular activities set out in Figure 3. Again, the results are presented in a dot plot with point estimates and 95%-confidence bands (see Figure 6). Instead of describing all results, I will focus on the most striking results and the differences between the different types of organised activities. Regression using 25 imputed datasets; ** p < 0.01, * p < 0.05, + p < 0.10. a Confounding percentage was computed using logit coefficients. Table 2 thus provides an overview of the hypothesis and the results of the analysis. A plus sign indicates that the variable has a significant effect in the proposed direction (at least p < .05).
Overall, my hypotheses are supported. Again, it is clear that culture and material resources do matter. In addition, an occupational effect is apparent for three out of four activitieswith the only exception being language courses. However, there are also differences between the activities. The decision to choose language courses in particular seems to follow a different pattern than expected. Although monetary resources emerge as a decisive predictor, neither culture nor occupation plays a role. In contrast, material resources do not seem to be an obstacle for participation in other activities such as handicrafts and dance lessons. 4 Surprisingly, maternal education only has a sizeable effect on the probability of attending music lessons. In this case, however, the variable has a remarkable effect: mothers who hold a tertiary degree are 22.4% more likely to enrol their children in music lessons than mothers with elementary education are. Still, the strong effect of maternal education observed in comparable studies might partly be caused by their use of a catch-all indicator for organised activities (see Mühler and Spieß 2008;Schober and Spiess 2013).

Results and discussion
With this paper, I set out to analyse preschool children's organised activity participation through 'a social class lens' (Hartas 2015, 33). I have shown that the dividing line between scholars who align themselves with either a structural or cultural approach to social class and parenting is an artificial one. Both paths function mutually rather than exclusively. Accordingly, further theoretical work is needed to integrate both perspectives into the research on social class and parenting, and to avoid falling back into a simplified narrative of socio-economic status. More importantly, a social class gap in organised activity participation persists even after taking both approaches into account. I interpret this finding as a call to continue research in the vein of Melvin Kohn, who was interested in the role of occupational conditions beyond the workplace (Kohn 1963;Kohn andSlomczynski 1993 [1990]). I have applied this perspective in this paper. The empirical foundation of social class differences in organised activity participation is still insufficient. Notwithstanding this fact, governments across Europe have identified informal education as a potential vehicle to mitigate social inequalities. For instance, the German government supports less affluent children's leisure activities with monetary subsidies (Federal Ministry of Labour and Social Affairs 2015). While these children can potentially benefit from this programme, its design neglects the roots that lead to an unequal uptake of organised activities. This paper proposed to take a step back in order to analyse in how far material constraints, culture, and occupational conditions contribute to the social class gap in organised activity participation. My analysis indicates that families' social position matters most for attending music lessons. Playing a musical instrument has been shown to exert strong effects on school grades and cognitive development (Hille and Schupp 2015). Accordingly, future research should take into account not only material constraints but also cultural understandings in order to inform policymakers. A final problem with the current state of the field is that wider societal changes are not incorporated into empirical analyses. As Faircloth (2014) noted, we are currently witnessing the emergence of 'a particular parenting style (…) in Euro-American contexts that is widely considered "ideal"' (48). Two issues could be raised here. On the one hand, this transformation can be interpreted as a struggle for the power to impose the 'right' or 'legitimate' way of raising children. In this sense, the inherent danger of scapegoating parents in less favourable positions is 'symbolic violence' (Bourdieu 1985) which yields 'high benefit-cost ratios and rates of return' (Heckman 2008, 290) for those who can conform to the 'legitimate' expectations embedded in institutions such as the education system. On the other hand, the discussions on 'legitimate' parenting which have been characterised as 'parenting out of control' (Nelson 2010) have also alienated middle-class parents (Perrier 2013). This development could easily be aggravated by a 'rug rat race' (Ramey and Ramey 2010) in which parents have to invest more and more into the cognitive development of their children without improving their family's social class position. In each case, greater consideration of the macro-level context is advisable in order to formulate appropriate hypotheses at the individual level. Table A1. Logistic regression on organised activity participation using AME and cluster-corrected SE, N = 1771.