Family resources and children’s skills: development of a skills attainment model

ABSTRACT

The association between parental SES and children’s educational outcomes is one of the key topics in stratification research. Although most differences between social groups are explained by initial differences in performance, the influence of family resources associated with children’s basic skills is still poorly understood. We therefore developed a skills attainment model, focusing on the relative contribution of three family resources: parents’ own basic skills at age 12, other parental skills developed in education and financial resources in the household. In addition, we examine potential heterogeneity across social groups. We develop a unique dataset, the Intergenerational Transmission of Skills dataset, covering 25,000 Dutch parents and 41,000 children. It includes high-quality test scores in math and language, similarly measured among parents and children, and detailed register information on educational attainment and income. Using structural equation modeling, we find that about one-fifth of children’s basic skills is explained by the three parental resources. Of this explained variance, 69 percent is related to parent’s basic skills, 21 percent to other parental skills developed in education and 10 percent to household income. We find no substantial differences in the transmission across sex, between low- and high-income families and between low- and high-educated parents.

KEYWORDS:

• Intergenerational transmission of skills
• basic skills
• family resources
• education
• financial resources

Introduction

The relation between parental socioeconomic status (SES) and children’s educational outcomes is one of the key topics in stratification research (e.g. Buis 2013; Bukodi et al. 2014; Bukodi and Goldthorpe 2013; Kloosterman et al. 2009). A strong association between parents’ socioeconomic status and their offspring’s educational attainment reflects a high level of inequality of educational opportunities (IEO), whereas a weak association is indicative of a high degree of social mobility and openness of societies. Previous research has shown that most of the differences in educational attainment between social groups, about two-third, is explained by initial differences in school performance (Büchner et al. 2013; Bukodi et al. 2021; Erikson et al. 2005; Jackson et al. 2007; Karlson and Holm 2011).¹ While this is a robust result, found in various western societies and time periods (e.g. Büchner et al. 2013; Bukodi et al. 2021; Erikson et al. 2005; Jackson et al. 2007; Karlson and Holm 2011), the underlying mechanisms explaining these differences are still poorly understood.

Twin studies show that a considerable part is related to genetic differences (Bartels et al. 2002; Silventoinen et al. 2020), but the causal chain is much more complicated than just genetic inheritance of intelligence (Harden 2021). The genetic potential needs to be developed by meaningful interactions in the environment (Harden 2021) in order to result in skills² that determine success in education. In this process, parents play a crucial role with different resources that contribute to children’s skill development (Björklund and Salvanes 2011; Farkas 2003). The findings by Bukodi et al. (2021), for example, suggest that initial differences in school performance across social groups are primarily related to parental education and status, with a smaller role for parental income and class. In their study, however, it is not clear where the effect of parents’ education comes from. Is it related to the transmission of their basic skills or to things like familiarity with the educational system? This holds for many other stratification studies as well, as measures of basic skills of the parents are usually lacking (Marks 2021). The studies that do include cognitive measures of parents usually rely on general cognitive measures (Brown et al. 2011), rather than basic skills, thus priming the results as genetic inheritance. With this paper, we aim to add a new perspective to the literature by focusing on the intergenerational transmission of basic skills. By basic skills we mean acquired math and language skills that form the basis to be successful in education.

Interestingly, the focus on basic skills came fully in the center of research due to OECD-led projects like PISA and PIAAC (Schleicher 2007). These international projects defined literacy and numeracy as ‘key information-processing skills’ directly related to a good understanding of daily problems that people encounter and which directly affect their success in education or in society (OECD 2016). The research in this area shows that literacy and numeracy are the most important predictors of educational attainment (e.g. Fischbach et al. 2013; Knighton and Bussière 2006), labor market success (e.g. Hanushek et al. 2015; Levels et al. 2014; McIntosh and Vignoles 2001), and even non-economic outcomes such as health and social engagement (e.g. Borgonovi et al. 2016; Kakarmath et al. 2018; Martin 2018; Vera-Toscano et al. 2017).

Our study can be seen as bridging the two lines of research, namely social stratification research and skills research. Our main research questions are: What is the relative contribution of different parental resources in the development of basic skills of young children? And to what extent are there differences in the intergenerational transmission of basic skills across social groups? We will use the term basic skills instead of key skills, to refer to performance in math and language at the end of primary education. We develop a skills attainment model, focusing on three possible resources that parents have at their disposal that may affect their offspring’s basic skills at age 12. These parental resources are the parental basic skills at age 12, other parental skills developed in education and household financial resources. First, basic skills, such as math and language skills, are crucial to be successful in education. These skills form the basis for further skill development in education (e.g. Fischbach et al. 2013; Knighton and Bussière 2006). Second, there are also other skills acquired in education by parents, next to basic skills, that can affect children’s basic skill development. Educational sociologists have argued that familiarity with the culture at school, knowledge of the education system, feeling at home at school, and the acquisition of certain positive norms and values toward education, make it easier to successfully navigate through education (Barone 2006; Forster and Van de Werfhorst 2020; Kloosterman et al. 2009; Sullivan 2001). Third, financial resources available to the family enable parents to ensure a stimulating learning environment in terms of equipment and extra tuition (Bray 2020; Schneider et al. 2018; Zwier et al. 2021).

To answer our research questions, we developed a unique dataset, the Intergenerational Transmission of Skills (ITS) dataset, covering more than 25,000 Dutch parents and their children, which allow us to carefully distinguish the three different types of resources. One of the salient features of this dataset is that basic skills were measured with similar tests for both parents and children around the same age (age 12), just before entering secondary education. This allows us to distinguish the effect of parent’s basic skills from the effect of other parental skills developed in further education.

Our contribution is threefold. First, we have a strong measure of basic skills. We use high-stakes test in math and language that are used in education for track placement in secondary education. This test is highly predictive of educational success (Feron et al. 2016; Lek 2020). Second, we have similar tests for parents and children measured at about the same age. Third, by combining high-quality survey data on the parents with reliable register data for parents and their children, we avoid the panel mortality that often plagues this kind of longitudinal research.

Insights into the relative importance of these three forms of family resources are crucial to the development of policy interventions that could reduce educational inequalities. Each of the previously mentioned resources is likely connected to a different set of interventions. If the intergenerational transmission of basic skills is the dominant mechanism, an effective intervention should aim at stimulating the acquisition of such skills in early childhood (Cunha and Heckman 2007; Cunha et al. 2010) or in primary education, specifically targeting disadvantaged youth. An example of such an intervention is high-dosage tutoring stimulating math and language skills (De Ree et al. 2021; Fryer and Howard-Noveck 2020). However, if other parental skills developed in education, such as familiarity with the culture at school and knowledge about the educational system, are more influential for the children’s basic skills, effective interventions should likely aim at increasing parental engagement and involvement (Kim and Hill 2015; Sénéchal and Young 2008). Finally, when financial resources play a dominant role, governments could make learning equipment, extra tuition, or grants available for disadvantaged children (Bloome et al. 2018).

Developing the skills attainment model

Stratification research has a long tradition of looking at socioeconomic differences in educational attainment. As highlighted by Boudon (1974), these socioeconomic differences in educational attainment result from two sources: differences in basic skills (primary effects of social stratification) and differences in educational choices given these basic skills (secondary effects of social stratification). This is illustrated in the simple educational attainment model in Figure 1.³

Figure 1. Boudon’s educational attainment model.

Primary effects, i.e. the indirect paths in Figure 1 from family socioeconomic status (SES) to educational attainment child, refer to socioeconomic differences in educational attainment due to differences in basic skills. Children from higher strata perform on average better at school than children from lower strata (Breen and Goldthorpe 1997). These differences in performance originate from differences in genetic endowments as well as differences in skill development in early childhood (Goldthorpe 1996). The secondary effects of social stratification, i.e. the direct path from family SES to educational attainment, refer to differential educational choices across social strata after controlling for basic skills. In this article, we focus on the primary effects of social stratification.

To do so, we first look at what determines the socioeconomic position of the family. Parental SES is in itself the result of a classic status attainment model of parents, in which basic skills affect educational outcomes and in turn affect later earnings or socioeconomic status (Figure 2). Previous research has shown the relevance of basic skills in predicting educational attainment (e.g. Fischbach et al. 2013; Knighton and Bussière 2006) and wages (e.g. Hanushek et al. 2015; Levels et al. 2014; McIntosh and Vignoles 2001; Strenze 2007).

Figure 2. Classic status attainment model for parents.

Integrating the educational attainment model with this status attainment model of parents leads to the skills attainment model as depicted in Figure 3. The different paths leading to children’s basic skills in this model will now be discussed in more detail. The first path is the direct effect of parental skills on children’s skills. While studies on the intergenerational transmission of basic skills are scarce, more work has been done in the area of intergenerational transmission of intelligence. Previous studies show intergenerational correlations between .20 and .50 (Anger 2012; Anger and Heineck 2010; Björklund et al. 2010; Black et al. 2009; Grönqvist et al. 2017; Silventoinen et al. 2020). Moreover, findings on sibling correlations (Björklund et al. 2010; Björklund and Salvanes 2011) indicate that about half of the correlation between brothers can be attributed to shared family background. In adoption studies, where no genetic transmission can take place, typically lower correlations are found, namely .12 for fathers and .20 for mothers (Grönqvist et al. 2017). However, this approach faces (selectivity) problems when it comes to the type of parents, the characteristics of the adopted child and potential behavioral and attachment problems associated with adoption.

Figure 3. Skills attainment model.

Research looking at the intergenerational transmission of basic skills instead of intelligence is scarcer. These studies show that parental skills in math and language strongly influence those same skills in their children (Brown et al. 2011; Crawford et al. 2011; De Coulon et al. 2011; Sullivan et al. 2021). For example, Brown et al. (2011) exploit an additional wave in the British National Child Development Study (NCDS) that includes a random sample of the NCDS original respondents’ offspring. The parents were tested at age seven in reading and arithmetic. All the children of a random group of one-third of the NCDS respondents were tested in 1991, when the parents were aged 33, and the children were aged between five and 18. The researchers study the intergenerational transmission of test scores with an instrumental variable approach, exploiting regional differences in the age at which parents first entered formal education. They find evidence of strong intergenerational transmission of reading skills that is not explained by genetic differences. However, they do not find such an effect for math skills in the instrumental variable analysis, suggesting that the intergenerational transmission of math skills is more genetic. A related study was carried out by De Coulon et al. (2011), using multiple waves of the British Cohort Study (BCS) that started in the 1970s. Since skill levels are not static over the life course, parental skills were measured at two points in time, at ages five and 34. The basic skills of the offspring were measured between ages three and five with an ability test including a verbal and numeric part. De Coulon et al. (2011) show that the correlation between the parents’ childhood basic skills and their children’s basic skills measured at age five is around .20. After the parental basic skills at age 34 are included, this correlation decreases to around .10.

In studies looking at the effect of parental background on children’s basic skills, taking into account parental skills, findings demonstrate that parental skills are a strong predictor of children’s skills, and also strongly mediate the relation between parental education and children’s skills (Crawford et al. 2011; Sullivan et al. 2021). More precisely, Crawford et al. (2011) use data from the BCS to study the extent to which the influence of families’ SES on children's skills is explained by skill transmission next to other parental characteristics. Besides a substantive correlation between parents’ and children’s basic skills, the authors demonstrate that about one-sixth of the influence of family socioeconomic position on the children’s skills is driven by parental skills. Sullivan et al. (2021) utilize the Millennium Cohort Study (MCS), including vocabulary test scores of mothers, their partners and children. They find that children's language skills at age 14 are particularly strongly linked to their mothers’ language skills (.19), and a bit less strongly linked to her partners’ (.14) skills. These effects become slightly smaller after adding children’s language skills at age 5, but remain significant and substantial. On top of that, their mediation analysis shows that most of the effect of parental education on children’s language skills is indirect (75%) through the language skills of the mother (35%) and her partner (20%). They conclude that it is essential to include parental skills in research on educational inequalities.

Although the above studies shed a first, meaningful light on the intergenerational transmission of basic skills, they also have some limitations. The most important limitations are that the tests used for the children seem to capture general cognitive ability rather than school-related knowledge on math and reading, and that the tests were not the same for the parents and the children. A major contribution of our study is that we use very similar tests to measure basic skills at age 12 for both parents and their children. As mentioned before, the test is used to determine track placement in secondary education, and is thus administered during one of the most important educational transitions.

The second path is from family income to children’s basic skills. Parents differ in the investments they make in the educational careers of their children (Doren and Grodsky 2016; Farkas 2003; Schneider et al. 2018). The availability of financial resources in the family provides material stimuli for children’s learning processes in the form of learning equipment (e.g. a laptop, place to study) or pay for extra tuition. While financial resources did not used to be stressed as the most important mechanism (De Graaf 1986; Huang 2013), recent developments in education, such as the rise in shadow education (Bray, 2011) and increase in tuition fees (OECD 2020), have brought renewed attention to this area (Bray, 2020). Research shows that participation in so-called shadow education, that is, extra out-of-school educational activities to improve one’s school performance, has increased strongly worldwide (Bray 2011; Mori and Baker 2010). Children from lower social strata participate less in shadow education than children from higher strata, and this relation between social origin and participation in shadow education is stronger in countries with high-stakes tests, such as the Netherlands (Zwier et al. 2021). This path from family income to the offspring’s skills thus represents the effect of the family ensuring a stimulating learning environment in terms of equipment and extra tuition that may contribute to an optimal skills development of young children.

The third path is from the educational attainment of the parents to their offspring’s skills. As parents’ basic skills and income are taken into account, this path represents the effect of further skill development of parents in education on children’s basic skills. This does not only include a further development of key skills, such as literacy and numeracy, but also soft skills such as familiarity with the culture at school, knowledge of the educational system, feeling at home at school, and the acquisition of certain positive norms and values toward education. It is important that parents understand how the education system is organized, what the important decisions are, and what the short- and long-term consequences of certain choices are. Using Dutch panel data, Forster and Van de Werfhorst (2020) show that parents’ knowledge is a significant predictor of the educational success of their children net of parents’ education and other sociodemographic characteristics. Another important type of soft skills are the attitudes that foster a successful educational career, such as perseverance, grit, achievement motivation, and curiosity (for an overview, see Borghans et al. 2008). Duckworth and Seligman (2005), for example, report on the importance of self-discipline in predicting academic performance. These attitudes play an important role, since they can amplify the role of skills (making the impact of skills more relevant for educational success), as was already argued in the early 1960s by the Canadian psychologist Vroom (1964). A third relevant type of soft skill is familiarity with the school’s culture (i.e. alignment between the culture at home and that at school). For example, Calarco (2014) shows that middle- and working-class parents express contrasting beliefs regarding appropriate classroom behavior, beliefs that shape parents’ coaching of their children, which affects the children’s behavior and problem-solving strategies in class.

Identifying the relative contribution of each of these three paths has been difficult, mainly because data on the intergenerational transmission of basic skills was lacking (Marks 2021). Researchers have been able to identify the role of financial resources, generally pointing to a significant albeit modest contribution (De Graaf 1986; Huang 2013), compared to the effect of parental education. However, parental education in such models usually captures both parental basic skills as well as parental further skill development in education. The key contribution of this study is that we include measures of school-related knowledge in the key domains of math and language for both generations, as well as indicators for educational attainment and household income, allowing us to differentiate between the three different paths. Note that in this study we lack direct measures of soft skills and further skills gained in education by the parent. Instead, we look at the association between parental education and the children’s basic skills, controlled for parental basic skills and income. Our interpretation of this association is that it captures the further development of these basic skills in education, as well as soft skills that are needed to be successful in education. By doing so, we aim to give a better understanding of the relative contribution of the different types of parental resources to the development of children’s basic skills. Moreover, we will examine whether the intergenerational transmission of basic skills differs across social groups (i.e. fathers, mothers, sons, and daughters; low- versus high-income families; and low- versus high-educated parents).

Dutch context

It is important to understand some key features of the Dutch educational system relevant to our data. The Dutch education system is a so-called early stratifying system (Bol and van de Werfhorst 2013), where students are allocated to different tracks in secondary education after the final year of primary education (grade 6, at age 12). The allocation is based on two factors: the performance of students on a national test, often the so-called Central Institute for Test Development (CITO) test,⁴ and the advice of the primary school teacher. The Cito test is a high-stakes test measuring school performance in math and the Dutch language. Although the tests are updated annually and occasionally cover other domains as well, the core principle behind the test and its aims have not changed over the years, namely, to provide an objective indication of students’ performance in the key domains of math and language. These domains are considered crucial for the successful completion of secondary education. It is important to note that the use of a national test was not mandatory before the 2014/2015 school year, but since the introduction of the Cito test in 1970, a large majority (around 85%) of the schools in primary education have used it. The second factor is the primary school teacher’s advice. This advice is partly based on the results of the previously mentioned national test and partly based on the teacher’s assessment that the student will succeed in secondary education at the advised level.

Data and variables

To answer our research question, we developed the so-called ITS dataset which is a combination of Dutch panel survey data gathered in the 1970s and 1980s complemented with register data available from Statistics Netherlands. The panel data consist of three cohorts that started their secondary education in 1977, 1983,⁵ and 1989, respectively. Each of these longitudinal surveys consist of nationally representative panels of 37,280 (1977), 16,813 (1983), and 19,524 (1989) students entering Dutch secondary education (grade 7, age 13). The surveys were carried out using a two-stage sampling design, with, first, a random sample of schools and, second, a random sample of classes within these schools (for more information, see Jacobs et al. 2021).

Each cohort started at the beginning of the school year with an assessment of math and language skills, using a short version of the above-mentioned Cito test. Questionnaires were also sent to the parents of the cohort members to gather basic background information. All the students were subsequently followed during their school career, assessing their position in education (i.e. tracks and grades) annually until they left education.

For most of the students in the original cohorts, basic identifying information such as their name and address at the time of the survey allowed us to link these cohort data to the register data from Statistics Netherlands. For the latest cohort, we could rely on a unique personal identifier that allowed for a successful linking process in 98% of the cases. For the other cohorts, the percentage of original students who could be linked to the register is lower, namely 81% (1977 cohort) and 88% (1983 cohort).

In the first two decades of the new millennium, the original students from the three cohorts were in their 30s or 40s, an age at which their own children could enter secondary education. Statistics Netherlands has register data for all schools that participated in the Cito test from the school year 2005/2006 onward.⁶ As indicated earlier, this Cito test is taken in the final year (grade 6) of primary education. We linked the original cohort data to the register data of their children, including test scores, as well as other information on the children’s educational careers. The latter information is available from the so-called Netherlands Cohort Study on Education, in which multiple sources of register data on education are linked (for an overview, see Haelermans et al. 2020).

The ITS dataset currently covers 25,287 unique parents with a total of 41,326 unique children. Table 1 provides an overview (see Jacobs et al. (2021) and the supplementary material for a detailed description of the linkage process and the representativeness).⁷

Table 1. Overview combining all data.

	Education cohort 1977	Education cohort 1983	Education cohort 1989
A. Original file	37,280	16,813	19,524
B. Respondents with ID that can be linked	30,171 (81% of A)	14,764 (88% of A)	19,111 (98% of A)
C. Respondents with test data	25,462 (84% of B)	14,613 (99% of B)	18,015 (92% of B)
D. Respondents with children	19,633 parents (77% of C) 42,616 children	11,315 parents (77% of C) 24,330 children	13,657 parents (76% of C) 29,089 children
E. Children with test data	13,248 parents (67% of D) 22,213 children (52% of D)	7,649 parents (68% of D) 12,775 children (53% of D)	4,586 parents (34% of D) 6,358 children (22% of D)

Note: Own calculation based on data available at Statistics Netherlands.

Table 2. Descriptive statistics.

	Education cohort 1977			Education cohort 1982			Education cohort 1989			All education cohorts
	N	Mean/%	SD	N	Mean/%	SD	N	Mean/%	SD	N	Mean/%	SD
Children
Language skills child (std)	22,213	0.149	0.925	12,755	0.019	0.963	6,358	−0.133	1.003	41,326	0.066	0.955
Math skills child (std)	22,213	0.119	0.946	12,755	0.010	0.972	6,358	−0.125	0.995	41,326	0.048	0.966
Female child	22,213	49.9		12,755	50.5		6,358	51.5		41,326	50.4
Parent
Language skills parent (std)	13,175	0.146	0.948	7,558	0.007	0.986	4,548	−0.097	0.973	25,281	0.061	0.969
Math skills parent (std)	13,175	0.158	0.970	7,558	−0.010	0.985	4,548	−0.164	0.963	25,281	0.050	0.981
Highest education in years	12,938	11.682	2.824	7,301	11.457	2.982	4,366	10.875	2.920	24,605	11.472	2.904
Tertiary education other (legal) parent (dummy)	7,334	47.9		4,596	0.495	0.500	2,852	41.2	0.492	14,782	47.1
Intelligence score parent (std)	7,545	0.137	0.958	7,532	0.013	0.989	4,548	−0.131	0.607	19,625	0.027	0.907
Education grandmother in years (std)	12,043	0.093	1.029	6,744	−0.025	0.979	4,196	−0.151	0.938	22,983	0.014	1.003
Education grandfather in years (std)	12,438	0.105	1.008	6,493	−0.045	0.970	3,976	−0.153	0.967	22,907	0.018	0.995
Female parent	13,175	49.5		7,558	55.9		4,548	61.6		25,281	53.6
Low education parent (dummy)	12,938	18.7		7,301	23.2		4,366	27.4		24,605	21.6
Household income (in percentiles)	22,028	58.023	24.394	12,673	54.267	24.869	6,290	46.452	23.804	40,991	55.086	24.783
Below median household income (dummy)	22,028	37.4		12,673	43.7		6,290	56.7		40,991	42.3

Source: ITS dataset.

As explained earlier, sample sizes vary across cohorts, mainly due to differences in the original sample sizes and in the possibility of linking the original respondents to the register data. In addition, there is missing information on the test data of some parents, specifically those in the 1977 cohort (N = 4,709). The reason for this is that in the 1977 education cohort, it was impossible for Statistics Netherlands to link part of the Cito test data because the necessary identifiers were missing or unreadable on the test sheets (CBS 1982). Moreover, in the 1977 and 1989, some classes in the selected schools were not able to take the test, so that whole class dropped out. These parents were also left out of the subsequent analyses. On average 77% of all parents for whom we have test data, have children of their own, and this does not vary across the three cohorts. However, for many of these children we do not have test data because their tests fall outside the 2006–2019 observation window. This holds in particular for the children of the 1989 education cohort who are often still too young to have taken the Cito test (22% of the cases compared to 52% (cohort 1977) and 53% (cohort 1983)). The representativeness of the ITS dataset was further examined in the technical report of the dataset (Jacobs et al. 2021). This showed that the pooled dataset is fairly representative.

Basic skills

The short version of the Cito test consisted of 25 math items and 45 language items for the 1977 cohort and 20 items for each domain in the 1983 and 1989 cohorts. The test results are standardized for each domain–cohort combination separately, using the complete original data set.⁸ The Cito test data of the children are available from school year 2005/2006 to 2018/2019. Depending on the year, the test consists of 60 to 85 math items and 100 to 135 language items. The test results are standardized for each domain–test year combination, using the complete original data sets. As of school year 2014/2015, other test suppliers entered the market. Because schools that switched to a different test supplier could be selective in terms of population characteristics (Jacobs et al. 2023), the standardization is conducted using the parameters of the schools that participated in the Cito test every year. We use the math and language scores separately to construct measures for basic skills.

Education variables

Parents’ education was annually monitored by Statistics Netherlands using a detailed coding scheme. In addition, education registers have been kept in higher education from 1983 (academic universities) and 1986 (universities of applied science) onwards, and since school year 2004/2005 also in upper secondary vocational education. For the education of the parent whose test score is available, the cohort information is supplemented by the education registry information if it recorded a higher level of education. The categories are converted into years of schooling using the so-called educational ladder of Bosker and Velden (1989).

For the other legal parent, a dummy-variable measuring the completion of tertiary education is included. We use this dummy, because for most parents the educational attainment is not known if it was lower than tertiary education, due to the later introduction of these education registers.

Financial resources

We use household income for the operationalization of financial resources. Household income is measured as the average standardized disposable private household income in percentiles in the period one to three years before the child took the Cito test.⁹ The disposable income of the household consists of the gross income excluding transfer payments, such as alimony, income insurance contributions, health insurance premiums, and taxes on income and assets. The measure is adjusted for the sizes and composition of the households.

Variables for verification checks

Non-verbal intelligence – In the original cohorts, the parents also took a non-verbal intelligence test, namely, the Test di Intelligenza Breve for the 1977 cohort and the Prüfsystem fur Schul- und Bildungsberatung-3 test for the other two cohorts. The test scores are standardized within each cohort, using the full cohort.

Education grandparents – The grandparents provided their highest education at the beginning of the cohort studies in the questionnaire. As the answer categories differed across the education cohorts, the original categories were harmonized in the following categories: primary education, lower secondary education, upper secondary education and tertiary education. These categories were converted into years of education.

Variables for heterogeneity analyses

Sex – The cohort parent’s and child’s sex distinguishes between male (0) and female (1).

Low-educated parent – This dummy-variable makes a distinction between low-educated parents (8 years or less) and non-low-educated parents (more than 8 years of education).

Below median household income – This dummy-variable makes a distinction between below-median-income households (in the 50th percentile group or below) and household with income above the median. Again, we look at the average household income one to three years before the child’s administration of the Cito test.

Method & models

To estimate the relative contribution of the different parental resources on children’s basic skills (the first research question), structural equation modeling (SEM) is performed. The advantage of using SEM is that it can simultaneously include path models and measurement models. In path models, complex models can be tested, while in measurement models, unobserved constructs (that is, latent variables), can be estimated with factor analysis. The inclusion of a measurement model thus enables us to correct for measurement errors in the skill tests (Ramlall 2016). These measurement errors would otherwise cause a downward bias in the observed relations (see also Büchner et al. 2013). Note that we do not claim to find causal relationships with this estimation.

In the main model, measurement models for both parents and children are included to measure the latent construct basic skills with the observed math and language skills. In the structural model, the association between these two latent variables is estimated, as well as the direct and indirect associations with parents’ education and household income. The cohort parent's basic skills are associated with their obtained level of education as well as (part of) the household income. In addition, the cohort parent's obtained level of education, interpreted as measuring the parent’s skills gained in further education, is directly associated with children’s basic skills and the household income. The other parent's tertiary education is included in the structural model since this is also associated with household income and children’s basic skills. Because partner choice is not random as a result of educational homogamy (Domański and Przybysz 2007), we include a covariation between the education of the cohort parent and the other legal parent.

Four additional checks are performed to verify the results from the main model. To compare the effects of educational attainment of both parents, we do an additional check where we leave out the parent’s basic skill measure (Check 1). Another verification of the relative contributions of the parental resources is done by leaving out the completion of tertiary education by the other legal parent (Check 2). Although in practice children benefit from parental resources from both parents, this summarizes, as it were, the relative contribution of resources of the total explained variance for one of the two parents.

The estimation of the relative contribution of the parental resources of the explained variance is dependent on the reliability of the different measures. While most of the variables are based on high-quality register data, there might be concerns about the quality of the basic skills measure for the parents as the test was not high-stake, as was the case for children. The issue is that if parent’s basic skills are not measured well, the effect of their basic skills will be picked up by the other paths, mainly the education effect. We address this issue in a verification check by including the non-verbal intelligence score in the cohort parent’s measurement model (Check 3), since this may improve the measurement of the latent variable basic skills for the parents. Finally, we estimate whether the intergenerational transmission of basic skills is affected by grandparental education (Check 4). Next to these verification checks, several robustness checks are done (see footnotes 10 to 14). The results are included in the supplementary materials (Tables S13– S19).

To answer the second research question, we examine whether the intergenerational transmission of basic skills differs across social groups. Potential heterogeneous effects in the intergenerational transmission of skills are estimated between different combinations of fathers, mothers, sons, and daughters; low- versus high-income families; and low- versus high-educated parents.

All structural equation models are estimated in Mplus 8.0 (Muthén and Muthén 2017) using maximum likelihood estimation with robust standard errors (MLR). We cluster children within their cohort parent. The model fit is determined by looking at the root mean square error of approximation (RMSEA), which needs to be below .05 to indicate a good fit (Steiger 1990); the standardized root mean square residual (SRMR), which should be below .08; and the Tucker–Lewis index (TLI) and the comparative fit index (CFI), which should be above .95 (Bentler 1990).

Results

Descriptive results

Table 2 displays the descriptive statistics of the main variables in our analyses, separately for the three cohorts and for the total sample. Table 2 reflects the earlier differences in the numbers of observations for the different cohorts (see Table 1). There is missing information for some of the key variables, such as the highest level of education of the parents, household income and non-verbal intelligence scores of the parents. However, the percentage of missing information is relatively low, and all missing information is imputed using full information maximum likelihood (FIML) estimation in the structural equation models (Muthén and Muthén 2017).¹⁰

Although the math and language skills for both parents and children are standardized within the cohort or test year using the full original cohorts, we observe in Table 2 that the means of these skills are slightly above average for the 1977 cohort and below average for the 1989 cohort. This result reflects the fact that, for the 1977 cohort, we mainly observe older parents, who are more likely to be high-educated, whereas, in the 1989 cohort, we mainly observe younger parents, who are more likely to be low-educated. However, the distributions of the math and language skills for the 1983 cohort and for the combined cohorts are close to normal. This result should provide confidence in our main analysis in which we work with the combined cohorts.¹¹ It is also notable that the percentage of females in the 1989 cohort is higher than in the other two cohorts. This is perhaps not surprising because women marry or cohabitate and get children at an earlier age than men, and we mainly have younger parents in this cohort.

Main results

In Figure 4, we present the graphical results of the structural equation model with the parental resources that are directly or indirectly associated with the child’s basic skills. All reported associations are standardized, meaning that an increase of one standard deviation on the independent variable is associated with a change in the dependent variable of the particular coefficient in standard deviations. The model fit indicators demonstrate an overall good model fit (RMSE = .072; SRMR = .071; CFI = .965; TLI = .908). Additionally, the measurement model of basic skills is satisfactory for both parent and child, with roughly similar and high factor loadings for math and language. This means that the latent construct is picking up both types of skills. Furthermore, the assortative mating, measured by the covariance between the parents’ education attainment is quite strong (.322). The parental resources together explain about one-fifth (R²= 0.207) of the variation in basic skills of the children. It is important to keep that in mind when we discuss the relative contribution of these resources to the explained variation.

Figure 4. Parent’s basic skills, parental education, and household income on basic skills child.

Looking at different paths that are associated with child’s basic skills, it is clear that parent’s basic skills at age 12, expressed as proficiency in math and language, are the strongest predictor of child’s basic skills also at age 12. One standard deviation increase in parent’s basic skills is associated with more than one-third (.343) of a standard deviation increase in the basic skills of the child. This association between child’s and parent’s basic skills is underestimated as this is only based on the estimate of one parent. If we assume that the other parent’s basic skills have a similar association, the total effect of parent’s basic skills would be larger.

Evidence for an additional effect of the other parent’s basic skills is found in the fact that the direct association between this parent’s educational attainment and the child’s basic skills (.177) is three times higher compared to the association of the parent for whom we have a basic skill measure (.061). For this other parent whose basic skills were not measured, this association (.177) also picks up the association with this other parent’s basic skills.

Check 1: leave out basic skills of the cohort parent

To compare the associations of educational attainment for both parents, we ran a similar model as in Figure 4, but now leaving out the basic skills of the cohort parent. The results in Figure 5 show that the standardized coefficient for the parent for whom we have this skills measure is .238, while it is only .171 for the parent for whom the skills measure is not available. The difference in coefficient size can be attributed to the fact that for the other parent, we have a weaker measure of educational attainment, namely a dummy distinguishing tertiary versus non-tertiary education instead of a continuous variable. In addition, this measure has a lot of missing values because in most cases the educational registers were only introduced after the parent had finished their educational career. Another reason for the difference may therefore be the imputation scores. If both variables are turned into a dummy, the sizes of the coefficients are more or less similar (see Table S4 in the supplementary material), namely .187 for the parent with the basic skill measure and .183 for the other parent.

Figure 5. Parental education and household income on basic skills child.

Table 3 provides an overview of the mechanisms affecting child’s basic skills and is based on the direct estimates of the parental resources from Figure 1. As indicated above, the three parental resources together explain one-fifth of the total variation in the basic skills of children. If we look at the relative contributions to the total explained variance of each of the parental resources, we see that 53 percent is related to parent’s basic skills, 9 percent to parent’s education, 28 percent to other parent’s education and 10 percent to financial resources. As indicated before, this is an underestimation of the role of basic skills as parental resource¹² and an overestimation of the role of educational attainment, because a measure of the basic skills of the other parent is missing.¹³

Table 3. Parental resources affecting child’s basic skills.

	Direct effects on basic skills child	% (relative contribution to explained variance)
Mechanism
Parent’s basic skills	.343	53%
Parent’s education	.061	9%
Other parent’s education	.177	28%
Household income	.062	10%
Total effect	.643	100%

Source: ITS dataset.

Note: Standardized model coefficients are displayed.

Check 2: leave out tertiary education of other legal parent

To provide more insight in the relative contribution of parental resources on child's basic skills, particularly for the transmission of the parent’s basic skills, we ran similar models as in Figure 4, but leaving out the educational information of the other parent. This will lead to an underestimation of the total influence of parental resources, as in practice children will get resources from both parents. However, it may give us a better indication of the relative contribution for at least basic skills and educational attainment. Since in Dutch society relatively many women work part-time and thus contribute less to the household income (OECD 2019), the analyses were done separately for mothers and fathers. These results together with the pooled averages are presented in Table 4 (the full model is presented in Table S5 of the supplementary material).

Table 4. Contribution of parental resources when only including parents for whom we have a basic skills measure.

	Direct effects on basic skills child	%	Direct effects on basic skills child	%	Direct effects on basic skills child	%
	Fathers		Mothers		Pooled average
Mechanism
Parent’s basic skills	0.332	62%	0.335	61%	0.334	61%
Parent’s education	0.094	18%	0.111	20%	0.103	19%
Household income	0.107	20%	0.107	19%	0.107	20%
Total effect	0.533	100%	0.553	100%	0.543	100%

Source: ITS dataset.

Note: Standardized model coefficients are displayed. The percentages refer to the relative contribution to the explained variance.

The results demonstrate that the relative contributions of parental resources on child’s basic skills change somewhat after leaving out the educational level of the other parent. For both fathers and mothers, the relative contribution of their basic skills to the explained variation child’s basic skills, is about 60 percent. Compared to Table 3, this is an increase from 53 percent to 61 percent. About one-fifth of the total explained variation of the child’s basic skills is related to parent’s education. This is obviously lower compared to the relative contribution of parents’ education, including the other parent’s education, from the previous model which was 37 percent. However, the relative contributions of parent’s basic skills and education as presented in Table 4 are probably closer to the ‘true’ proportions than the proportions as estimated in Table 3. The relative contribution of household income according to Table 4 is about 20 percent, which is twice as high as the 10 percent in the previous model. Because the other parent is no longer included, this is probably an overestimation of the financial resources mechanism. After all, the household income¹⁴ consists in most cases of the income of both parents. Given that Dutch mothers, in particular from this generation, more often work part-time and contribute therefore less to the household income, this is especially the case for mothers.

All in all, this additional exercise, together with the main model, leads us to assume that the relative contribution of household income from Table 3, which is not an underestimate as in Table 4, is plausible, and that the ratio of the contributions of parental education and basic skills¹⁵ from Table 4 comes closest to reality. Based on these assumptions, we conclude that the relative contributions of the three parental resources in explaining child’s basic skills are 69 percent for parents’ basic skills, 21 percent for other parental skills acquired in education, and 10 percent for household income.

Check 3: inclusion of non-verbal intelligence

As the basic skills of the cohort parents were measured with a low-stakes test, an alternative model was specified including the non-verbal intelligence score in the parent’s measurement model. Including this measure may improve the measurement of the latent variable basic skills for the parents. For the children, this information is not available, but also less needed as for them the test was high-stakes.

The results in Figure 6 demonstrate that non-verbal intelligence contributes, in terms of the factor loading (.405), to the latent construct measuring the parent’s basic skills, but not much. Looking at the intergenerational transmission of basic skills, we see that this is hardly affected by the inclusion of parent’s intelligence score. Overall, the model clearly demonstrates that our skills measures in math and language capture most of the construct. In terms of model fit, the inclusion of intelligence seems to improve the fit, but this improvement is negligible.

Figure 6. Parent’s basic skills and non-verbal intelligence, parental education, and household income on basic skills child. Source: ITS dataset. Note: Standardized model coefficients are displayed. The standard errors are provided in Table S6 of the supplementary material. Model fit: RMSEA = .054; CFI = .965; TLI = .931; SRMR = .066. *p<.05; **p< .01; ***p< .001 (two-tailed tests).

Check 4: grandparental education

In a final check, we ran a model that included grandparents’ education, to check whether the associations between the parents’ resources and the child’s basic skills were affected by the resources in the extended family. This model is presented in Figure 7.

Figure 7. Parent’s skills, education and household income on skills child including highest attainment of grandparents.

The results show that grandparents’ education is directly associated with their grandchildren’s basic skills, but the coefficient of this direct association is small (.030). We are cautious with overinterpreting this small association, as our study faces the previously described problems with multigenerational research (Breen 2018; Engzell et al. 2020), such as collider bias due to omitted variables, for which it is impossible to control in the current design. More importantly, however, our key parameters of interest are not substantially affected. For the association between parent’s basic skills and child’s basic skills, we observe a small decrease, from .343 to .334; for parent’s educational attainment and family’s financial resources, we see only minor differences (for educational attainment from .061 to .056, and for household income from .062 to .060).

Heterogeneity analyses

We examined whether the intergenerational transmission of basic skills differs across social groups. The potential heterogeneous effects were explored between different combinations of fathers, mothers, sons, and daughters; low- versus high-income families; and low- versus high-educated parents. A summary of the results is presented in Table 5 (full results are shown in Tables S7 to S10 of the supplementary material).

Table 5. Summary table heterogeneous effects intergenerational transmission of basic skills.

Heterogeneous effects
	N	Estimate	Wald-test
Gender
Mother & daughter	11,076 daughters 8,764 mothers	.324***	n.s.
Mother & son	10,785 sons 8,499 mothers	.322***
Father & daughter	9,732 daughters 7,635 fathers	.324***
Father & son	9,733 sons 7,530 fathers	.337***
Household income
Below-median income	17,344 children 11,875 parents	.337***	n.s.
Median income and above	23,647 children 15,328 parents	.323***
Education
Low education	8,018 children 5,307 parents	.321***	n.s.
High education	32,252 children 19,298 parents	.371***

Source: ITS dataset.

Note: Standardized coefficients are displayed.

All standardized coefficients, standard errors and model fit are provided in Tables S8–S11 of the supplementary material.

*p < .05; **p < .01; ***p < .001 (two-tailed tests).

Our parameter of interest, namely the association between parent’s basic skills and their children’s basic skills does not change across the social groups. This holds for all groups, with a small exception: in the case of high-educated parents, we find a slightly higher estimate of parent’s basic skills on child’s basic skills (.371). However, the Wald test indicates that this difference is not significant. Overall, we conclude that the model presented in Figure 1 holds for the different subgroups as well.

Conclusions and discussion

For decades, stratification research has focused on the role of parental resources on various educational outcomes of their offspring. In this article, we focused on the relative contributions of parental resources on the child’s basic skills, to better understand the underlying mechanisms. In particular, we examined the intergenerational transmission of basic skills in the key domains math and language, as these skills are the most important predictors of success in education (e.g. Fischbach et al. 2013; Knighton and Bussière 2006), success on the labor market (e.g. Hanushek et al. 2015; McIntosh and Vignoles 2001) and success in other life outcomes (e.g. Borgonovi et al. 2016; Vera-Toscano et al. 2017).

This study aimed to bring together stratification and skills research by developing a skills attainment model distinguishing three types of parental resources that may affect children’s basic skills. First, parent’s basic skills affect children’s basic directly. While part of this transmission is due to genetic inheritance, it is also widely accepted that genes are not deterministic (Harden 2021). This means that genes still need to be stimulated in an encouraging environment in order to fully reach their potential. As such, parents’ basic skills are crucial in creating such an environment and help their children to develop their full potential (Marks 2021). A second major resource are the other skills that parents develop in education, such as knowledge of the system, knowing how to navigate through education and appreciating the value of education (e.g. Forster and Van de Werfhorst 2020). The third parental resource that may affect child’s basic skills are financial resources that may be used to provide learning equipment and tuition (e.g. Zwier et al. 2021).

To date, scholars have not been able to accurately distinguish the effects of these three types of resources. In particular, the role of parent’s basic skills in the intergenerational transmission of education has been underexposed, due to the absence of adequate data (Marks 2021). In this study, we contribute to the literature by including a direct measure of the basic skills of the parents and their offspring, and estimate the relative importance of the three parental resources for the child’s basic skills. We developed a unique and unparalleled dataset that includes parents’ and children’s basic skills measured with a similar test at age 12. The tests are taken in the domains of math and language. This information is linked to detailed information on educational pathways and to the household income.

Using SEM, we find that about one-fifth of the variation in basic skills of children is explained by the parental resources. Of this explained variance, about 69 percent is attributable to parent’s basic skills, about 21 percent to other parental skills developed in education and about 10 percent to household income. These relative contributions are based on the main model and four verification models, in which the relative contribution of each of the parental resources was further substantiated. Although the study by De Coulon et al. (2011) hinted at possible heterogeneity in the intergenerational transmission of basic skills, we found little evidence of this being the case. There are no substantial differences in the intergenerational transmission of basic skills from fathers to sons, from fathers to daughters, from mothers to sons, or from mothers to daughters; nor did we find substantial differences in the intergenerational transmission of basic skills between low – and high-income families or between low – and high-educated parents.

All in all, this study shows that in order to understand the role of parental resources in education inequality, it is crucial to include parental skills in math and language. This is consistent with the limited number of other studies that have looked at basic skills transmission (Crawford et al. 2011; Sullivan et al. 2021). However, the other parental resources in this study, that is, other skills acquired in education and household income, should not be overlooked, as these still contribute about one-third to the explained variance of children's basic skills at age 12.

The findings can, to a certain extent, be placed in a larger debate about IEO. In Michael Young’s (1958) prophetic essay ‘The Rise of the Meritocracy’, the adverse effect of meritocracy was already described. Young’s critical view on meritocracy was echoed in a recent publication by Sandel (2020). In Sandel’s view, meritocracy generates ‘winners’ and ‘losers’ based on diplomas, which are easier to obtain for the fortunate who have high-educated parents themselves. Social mobility is restricted not by inheritance of social positions in an aristocracy, but by credentialism in a parentocracy (Brown 1990). Markovits (2019) goes even further by arguing that meritocracy has become what it was conceived to resist: a mechanism for the concentration and dynastic transmission of wealth and privilege across generations. In this view, strong intergenerational transmission of skills is not the result of deviations or retreats from meritocracy, but, rather, stems directly from meritocracy’s successes. High-skilled parents know how important skills are and use this knowledge to help their offspring succeed in education.

What are the possible policy implications of our findings? Previous research has already shown that key skills are predictive of a wide range of individual economic and non-economic outcomes (e.g. Hanushek et al. 2015; Levels et al. 2014; McIntosh and Vignoles 2001). We now have evidence of the strong intergenerational transmission of such key skills. Even if a substantial part of this intergenerational transmission is related to genetic differences, this does not imply interventions would be pointless, as Harden (2021) argued in the book The Genetic Lottery. It makes it even more crucial that interventions aiming to increase social mobility and equal educational opportunities should focus on stimulating these basic skills among disadvantaged youth in early childhood or during primary education. This recommendation fits well with evidence from the economic literature (Cunha et al. 2010; Heckman 2006) stating that investment in skill development in early childhood is more beneficial than interventions later in life. The benefits of such early investment in skill formation are particularly effective if they focus specifically on disadvantaged children. A meta-study seems to suggest that indeed disadvantaged children profit most from Early Childhood Education and Care (ECEC) (Van Huizen and Plantenga 2018). Since we now have found strong evidence of the intergenerational transmission of basic skills, the returns to early childhood interventions are, in fact, underestimated. These interventions can also affect the skills of their offspring, leading to a multiplier effect in reducing inequality for future generations. This possible multiplier effect means that the social returns of investments in skills at an early age could be even higher than previously thought.

The strong intergenerational transmission of basic skills does not imply that schools or the environment outside the nuclear family do not matter. Key skills such as math and literacy are malleable (Aucejo and James 2021) and schools play an important role in their development. Moreover, there is substantial evidence that targeted interventions, such as high-dosage tutoring or summer schools, are effective interventions to increase math and language skills for disadvantaged youth (De Ree et al. 2021; Fryer and Howard-Noveck 2020).

The importance of basic skills is even greater, since parents’ basic skills also provide the basis for the acquisition of other skills developed in education and financial resources. However, the underlying mechanism of how they affect children’s basic skills and educational outcomes is different. In the case of parents’ further education, the mechanism is related to things such as positive norms and values towards education, familiarity with the culture at school, and knowledge about the educational system (Forster and Van de Werfhorst 2020), while, in the case of financial resources, it is related to providing a stimulating learning environment in terms of equipment and extra tuition (Zwier et al. 2021, Farkas 2003). Neither of these mechanisms are the strongest drivers of the child’s basic skill formation, but both are substantial, and, together, their contribution is about one-third of the total effect. Moreover, the importance of these two mechanisms will probably increase due to school segregation (Boterman et al. 2019; Reardon and Owens 2014; Vogels et al. 2021), differences in school quality and the emergence of elite schools (Merry and Boterman 2020), the increasing role of shadow education (Bray 2011), and increases in tuition fees (OECD, 2020). The growing importance of these two other mechanisms implies that policy measures should also be aimed at decreasing the negative impact of low parental education and the lack of financial resources on children’s performance in school. In this case, effective interventions are likely aimed at increasing parental engagement and involvement (Kim and Hill 2015; Sénéchal and Youn, 2008) or in offering grants for disadvantaged students, lowering tuition fees, or providing free learning equipment (Bloome et al. 2018).

Limitations and future research

Despite these new insights in the role of family resources for the children’s basic skills, this study also faces some limitations. First of all, test results on basic skills were only available for one of the parents. Future research should aim to include direct measures of basic skills for both parents, as both parents contribute to the child’s development of these skills. Second, we had no direct measure for other skills developed in education, such as further obtained cognitive skills and soft skills. Instead we had to argue that the remaining effect of education, after controlling for basic skills and income, is a proxy of this further skills development. Future research strive to include direct measures of such skills. Lastly, we did not differentiate between math versus language skills, while the intergenerational transmission might differ across these subjects. The study by Hanushek et al. (2023) using the same dataset, showed no differences between the two domains in the strength of the intergenerational transmission. Nevertheless it is likely that the underlying mechanism of the transmission might be different as families may be more effective in stimulating the development of language skills than math skills.

For future research, it would also be interesting to look at how family dynamics, potentially in interplay with the role of schools, affect the formation of skills. The intergenerational transmission of basic skills is strong, but this transmission might be hampered in case of parental separation or death. Another mechanism to be explored is that high-skilled parents might choose high-quality schools influencing the child’s basic skills further. Research in the future, in particular using the ITS dataset, can also focus on the educational pathway of the offspring and see how the intergenerational transmission of skills affects the secondary effects of social stratification. All in all, the findings in this study on the role of different parental resources on the child's basic skills are considered to be a valuable starting point for follow-up research.

Acknowledgements

This study is part of the Intergenerational Transmission of Skills (ITS) program carried out at the Research Centre for Education and the Labour Market (see project link). We are grateful to Per Bles, Arie Glebbeek, Rick Hanushek, Carla Haelermans, Tim Huijts, Suzanne de Leeuw, Guido Schwerdt, Stan Vermeulen, Herman van de Werfhorst and Simon Wiederhold for their valuable feedback on earlier drafts of this manuscript. We would also like to express our gratitude to the anonymous reviewers and the journal editors for their helpful comments. We gratefully acknowledge a grant received from the Dutch Ministry of Education, Culture and Science and the Netherlands Initiative for Education Research (NRO: grant 405-17-900) to develop the ITS database.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The used data set is available in the highly secure environment of Statistics Netherlands. To gain access, a number of steps must be completed, which can be found on the following website: https://www.cbs.nl/en-gb/ourservices/customised-services-microdata/microdata-conducting-your-own-research. The replication package including the code to replicate the findings presented in this article can be found on Zenodo (DOI: https://doi.org/10.5281/zenodo.8338137).

Additional information

Funding

This work was supported by the Dutch Ministry of Education, Culture and Science and the Netherlands Initiative for Education Research under Grant 405-17-900.

Notes on contributors

Babs Jacobs

Babs Jacobs is a PhD candidate at the Research Centre for Education and the Labour Market (ROA) at the Maastricht University. Her research focuses on the development and intergenerational transmission of basic skills in math and language.

Rolf van der Velden

Rolf van der Velden is emeritus professor at the Research Centre for Education and the Labour Market (ROA) of Maastricht University. His current research focuses on the intergenerational transmission of skills, skills development in education, transition from school to work, knowledge economy and the demand for 21st century skills, skills mismatches and the acquisition and decline of skills over the life course.

Notes

1 Note that there are differences across countries with regard to the relative contributions of primary versus secondary effects in explaining differences between social groups. These differences also change over time across countries. In general, however, the role of secondary effects of social stratification seems to decline in most western societies. Primary effects explain between 42% to 83% of the total effect of social origin depending on the country and the educational transition in question.

2 We use the term skills to refer to knowledge, skills and attitudes that determine success in education.

3 Note that Figures 1–3 are theoretical models and not SEM models like Figures 4–7.

4 From school year 2014/2015 onwards, the name of the test is changed into the ‘Central End Test’ (in Dutch ‘Centrale Eindtoets’), which is abbreviated with CET.

5 The 1983 cohort actually started in 1982, in the final year of primary education. From this cohort we selected those respondents who were in secondary education in the following school year, so that all three cohorts have the same characteristics in terms of sample design. We therefore refer to this cohort as the 1983 cohort.

6 Up until school year 2013/'14, we have Cito test data from pupils in schools that gave permission to Cito to provide the data to Statistics Netherlands. From school year 2014/’15 onwards, all pupils who made the Cito test in the particular school year are included in the data.

7 Note that the total number of parents at the bottom of the table is slightly higher than the number of observations in the descriptive Table 2 and subsequent analyses. This is because some members in the original cohorts married or cohabitated with other cohort members and had children. We randomly select one parent-child combination in this instance.

8 A possible problem is that the cohorts are grade cohorts rather than age cohorts. This means that, while most respondents were 13 at the time of the test, some were older due to grade repetition in primary education and some were younger due to acceleration. We conducted additional analyses using test scores corrected for these age differences. This proved not to affect the estimates in the main model (see Table S12 in the supplementary material).

9 We look at the period from one to three years prior for household income because the household income of the self-employed fluctuates considerably on a yearly basis.

10 As a robustness check, we also ran the main model without imputing the missing data using FIML. Our main conclusions remain the same (see Table S13 in the supplementary material).

11 In a robustness check, we checked whether the results are substantially different if we analyze each cohort separately. That is not the case (see Tables S14 to S16 in the supplementary material).

12 Just as a thought experiment, imagine we would have a skill measure for the other parent which would have the same impact, a continuous measure for the other parent’s educational attainment again having the same impact while we assume no covariation between the skills of both parents. In that case, 78 percent of the explained variation in children’s basic skills would be related to the basic skills of the parents, 14 percent to other skills attained by the parents in education, such a knowledge of the system and knowing how to maneuver, and 7 percent due to differences in financial resources available in the family. However, this would constitute an overestimation of the role of basic skills, as the skills of both parents are correlated due to assortative mating.

13 For some couples we do have information on the skills of both parents as they were both member of the three original cohorts. As the descriptives in Table S18 of the supplementary material show, this subsample is somewhat biased with higher test scores for the mothers. We ran a model for this subsample including skills measures of both parents, and this shows an effect of .275 for father’s basic skills and .395 for mother’s basic skills (see Table S19 in the supplementary material) and no significant effects for the other two mechanisms. Although this corroborates the main finding of a strong intergenerational transmission of skills, we should be careful drawing conclusions on the relative distributions as the subsample is small and not representative (e.g., high-school lovers).

14 Because we also have information on the personal income, we ran the same model changing household income in personal income (see Table S17 in the supplementary material). We observe that for mothers the direct association between personal income and the child’s basic skills is not present, and that the child’s basic skills are only associated with the mother’s basic skills (71%) and her educational attainment (29%). For fathers, 7 percent of the total direct associations is attributable to personal income, 24 percent to educational attainment and 69 percent to the transmission of basic skills. However, the measurement of personal income does not take into account household size, includes income before certain tax payments have been made and refers to the personal income of the parent for whom we have the skill measure regardless of whether that parent lives in the same household of the child or not. Thus, the measure of personal income does not fully reflect the financial resources available in the child's household and therefore is a less optimal measure.

15 The ratio of the coefficients of educational attainment (19%) and basic skills (61%) from Table 4 is about one to three. Assuming that the relative contribution of income is 10%, this means that the ratio should be calculated on the remaining 90%.