Social Science Research 51 (2015) 163–173

Contents lists available at ScienceDirect

Social Science Research

journal homepage: www.elsevier .com/locate /ssresearch

Do grandparents matter? A multigenerational perspectiveon educational attainment in Taiwan

http://dx.doi.org/10.1016/j.ssresearch.2014.09.0130049-089X/� 2014 Elsevier Inc. All rights reserved.

⇑ Corresponding author. Fax: +1 (215) 573 2081.E-mail addresses: yilinc@sas.upenn.edu (Y.-L. Chiang), hypark@sas.upenn.edu (H. Park).

1 Fax: +1 (215) 573 2081.

Yi-Lin Chiang ⇑, Hyunjoon Park 1

Department of Sociology, University of Pennsylvania, 3718 Locust Walk, Philadelphia, PA 19104-6299, USA

a r t i c l e i n f o a b s t r a c t

Article history:Received 6 August 2013Revised 19 September 2014Accepted 29 September 2014Available online 20 October 2014

Keywords:GrandparentsEducational attainmentEducational expansionTaiwan

In response to the growing interest in multigenerational effects, we investigate whethergrandparents’ education affects grandchildren’s transitions to academic high school anduniversity in Taiwan. Drawing on social capital literature, we consider potential heteroge-neity of the grandparent effect by parents’ characteristics and propose that grandparents’education yields differential effects depending on parents’ education. Our results show ten-uous effects of grandmother’s and grandfather’s years of schooling, net of parents’ educa-tion. However, the positive interaction effects between grandparents’ and parents’ years ofschooling indicate that grandparents’ additional years of schooling are more beneficial tostudents with more educated parents than for students with less educated parents. Thediverging gap in the likelihood of attending academic high school or university betweenstudents with parents in higher and lower ends of the educational hierarchy, along withincreased levels of grandparents’ education, supports our hypothesis that grandparents’education augments educational inequality by parents’ education.

� 2014 Elsevier Inc. All rights reserved.

1. Introduction

Sociologists have long been interested in the relationship between social origin, represented by parents’ status, and socialdestination, represented by individuals’ own standing in educational, occupational, or economic hierarchies. Building on thestatus attainment model developed by Blau and Duncan (1967), numerous studies on educational attainment in a variety ofsocieties have shown that parents’ education significantly affects child’s educational outcomes (Shavit and Blossfeld, 1993;Buchmann and Hannum, 2001). Recently, a growing number of studies point to the limitations of the two-generation frame-work that constrains investigation to the parent–child relationship, and propose an alternative framework that considers theeffects of grandparents and even further ancestors beyond the effects of parents in educational and social stratification pro-cesses (Mare, 2011, 2014; Pfeffer, 2014). Compared to the current two-generation paradigm, a multigenerational view ofinequality allows a more holistic understanding of educational attainment. While parents are adults who directly influencechildren’s educational success, extended kin, especially grandparents, may independently affect children’s education as well.Grandparents may provide help with childcare, supervision, and other emotional, social, and economic resources, all ofwhich can be beneficial for grandchildren’s educational outcomes.

However, despite the expected positive effects of grandparents on grandchildren’s educational and occupational out-comes, empirical evidence of a direct effect of grandparents, net of parents, is mixed (Chan and Boliver, 2013; Erola and

164 Y.-L. Chiang, H. Park / Social Science Research 51 (2015) 163–173

Moisio, 2007; Warren and Hauser, 1997; Zeng and Xie, 2014). The ambiguity of the results suggests that it is important toextend research on this issue to other settings, especially beyond Western societies. Anticipating considerable variation inthe effect of grandparents across contexts, for instance, Mare (2011) and Pfeffer (2014) called for more research on multi-generational effects across a range of societies that vary in institutional arrangements, which would help identify societalcontexts in which the effects of grandparents should be strong, weak, or negligible.

In this paper, we examine multigenerational educational attainment in a non-Western setting—Taiwan. The Taiwanesecase provides an opportunity to assess multigenerational effects in an interesting context of family arrangements and edu-cational expansion (described in detail later). In particular, our study is motivated by the hypothesis that the null effect ofgrandparents on grandchildren’s educational outcomes, found in some previous studies, might result from offsetting effectsfor different subgroups of the grandchildren generation. Specifically, we explore heterogeneity in the grandparent effect byinvestigating whether the effect of grandparents’ education on grandchildren’s educational attainment differs by parents’levels of education. Our focus on the interaction between parents’ and grandparents’ levels of education in affecting grand-children’s schooling is inspired by research that suggests possible variation of the grandparent effect under various familycontexts (Jæger, 2012; Solon, 2013; Zeng and Xie, 2014). In other words, ours is an attempt to move beyond previous studiesthat have assumed a uniform effect of grandparents without considering parental characteristics.

Drawing on literature of social capital, we propose that grandparents’ education ‘‘augments’’ parents’ education in affect-ing grandchildren’s educational outcomes. Our hypothesis points to the possibility that grandparents’ additional years ofschooling may be particularly beneficial to grandchildren whose parents have relatively high levels of education. In contrast,when parents have relatively low levels of education, grandparents’ education may not have a substantial effect or may evenresult in an adverse effect on grandchildren’s education. In other words, we anticipate that the interaction effect betweengrandparents’ and parents’ education on grandchildren’s education should be positive, implying that grandparents’ educa-tion augments educational gaps between children of parents with more education and children of parents with less educa-tion. These differential effects of grandparents’ education on grandchildren by parents’ education may offset each other sothat the overall effect of grandparents as a whole can appear to be tenuous. As explained in subsequent sections, the rapidexpansion of education during the past few decades in Taiwan makes the Taiwanese case particularly useful to test our aug-mentation hypothesis.

In the sections below, we first review existing literature on multigenerational educational attainment and highlight thelack of attention to the potentially heterogeneous effects of grandparents by parents’ socioeconomic status. We then intro-duce literature on social capital, from which we derive our hypothesis of the augmentation effect. Next, we provide a briefintroduction to the Taiwanese context, particularly focusing on the degree of educational expansion and its implications forthe multigenerational effect of education. In the data and methods section, we describe our data and modelling strategies totest our hypothesis. Using a Taiwanese longitudinal data set of middle school students who were followed up to five yearsafter middle school graduation, we assess the effects of grandparents’ education on grandchildren’s high school and univer-sity attendance. We compare models with and without interaction effects between grandparents’ and parents’ education totest whether the grandparent effect is conditioned by parents’ education. Finally, we summarize our findings and point tosome implications and limitations of our study.

2. Literature review

While grandparents’ socioeconomic status may affect grandchildren’s socioeconomic attainment in various ways, animportant question is whether grandparents independently affect grandchildren after controlling for parents’ socioeconomicstatus. Studies in the U.S. and Finland showed that grandparents’ socioeconomic positions were not significantly associatedwith grandchildren’s socioeconomic positions, after parents’ socioeconomic positions were taken into account (Warren andHauser, 1997; Erola and Moisio, 2007). These findings suggest that the effects of grandparents are likely present only throughtheir impacts on parents, which in turn affect children’s education. In contrast, studying class mobility across three gener-ations in Britain, Chan and Boliver (2013) found that social class position of grandparents had a significant effect on the classposition of grandchildren even after taking into account parental characteristics. Research from Sweden also showed asignificant grandparent (and great-grandparent) effect on grandchildren’s education and occupation, net of parents’ effect(Lindahl et al., 2012; Hällsten, 2014).

However, regardless of whether they found independent effects of grandparents, both sides of researchers often assumedhomogeneity in the effect of grandparents’ education on grandchildren’s education and failed to consider the possibility thatgrandparent effects may depend on parental characteristics or other factors. Investigating the effect of grandparents ongrandchildren’s schooling in rural China, Zeng and Xie (2014) showed that the effects of grandparents were contingent uponmultigenerational coresidence. Specifically, the authors found that only coresident grandparents significantly increasedgrandchildren’s chances of staying in school, while non-coresident and deceased grandparents did not. This finding suggeststhat grandparents may not uniformly affect grandchildren’s education, but may differently influence grandchildren depend-ing on family contexts.

We argue that parents’ characteristics, particularly parents’ levels of education, condition the effect of grandparents’ edu-cation on grandchildren’s educational attainment. Parents directly influence their children’s education and connect grand-parents with grandchildren. Therefore, depending on how parents utilize support and resources from grandparents, such

Y.-L. Chiang, H. Park / Social Science Research 51 (2015) 163–173 165

support and resources from grandparents may or may not be useful in fostering children’s education. Our augmentationhypothesis is derived from literature on social capital, which suggests that support and resources embedded in social net-work and relationships do not automatically translate into educational benefits for children, but depend on both the natureof family relationship and parents’ capacity to activate such social capital (Bourdieu 1986; McNeal 1999; Portes 1998; Wong,2002). The activation of social capital requires the possession of economic or cultural capital (Bourdieu, 1986). More edu-cated parents with professional skills and knowledge may better use resources from grandparents to maximize their impactson children’s education than less educated parents, whose limited knowledge and information on the educational systemmay hinder them from effectively mobilizing grandparental resources to produce substantial effects on children’seducational success.

Studies of parental involvement in children’s education, often considered as a kind of social capital, illustrate that moreeducated parents not only have a higher level of parental involvement but also make their involvement more effective inboosting their children’s educational outcomes than less educated parents (Lareau and Horvat, 1999; Lareau, 1987;McNeal, 1999; Perna and Titus, 2005; Ren and Hu, 2013). On the other hand, parents who obtained lower levels ofeducation, particularly those who experienced downward educational mobility compared to their own parents (i.e., grand-parents of children), may exhibit feelings of embarrassment or guilt associated with their downward mobility that can leadto strained relationships with their parents (i.e., the grandparents), thus preventing effective use of grandparents’ resourcesfor their children’s education (Newman, 1988). Further, tensions in family relations are found to have adverse effect onchildren’s academic outcome (Amato, 2001; Amato and Booth, 1997). Considering grandparents as resources embeddedin the family’s network, we expect that, compared to parents with lower levels of education, parents with higher levelsof education should be able to reap more benefits for their children’s education from grandparents’ additional years ofschooling.

While we propose the augmentation hypothesis, we acknowledge that the opposite pattern of the interaction betweengrandparents’ and parents’ education may be possible as well. In other words, grandparents’ education may compensatefor parents’ low levels of education to enhance grandchildren’s education. For instance, using the data from Wisconsin Lon-gitudinal Study, Jæger (2012) showed that the effect of grandparents’ education on grandchildren’s education was strongerfor grandchildren whose parents had relatively lower levels of education (and lower family income). However, as the authoracknowledged, the negative interaction between grandparents’ and parents’ education is difficult to be distinguished fromthe pattern of regression to the mean. Moreover, it is uncertain how the negative interaction found in Wisconsin can begeneralized to other contexts.

We expect that the greater benefits of grandparents’ education for children with more educated parents than childrenwith less educated parents can be particularly evident in a context where the educational system has rapidly expanded, suchas Taiwan. Studies showed that educational expansion led to a shift of educational competition to higher levels (Boudon,1974; Raftery and Hout, 1993). and educational expansion coexisted with sustained class advantage (Breen andGoldthorpe, 1997). One possibility is that more educated parents may recognize the necessity for children to achieve evenhigher levels of education in order to maintain high status, and thus heavily draw on grandparents to assist the process. Inthe context of rapid educational expansion where grandparents have seen significant returns of their investment in theirown children who successfully attained high levels of education, grandparents may willingly support grandchildren’seducation. In particular, when both grandparents and parents have relatively high levels of education and are consistentin their approaches to educational success, support and resources from grandparents can be particularly effective in fosteringgrandchildren’s education.

In contrast, when the parent generation experienced a rapid expansion of education and thus the majority of the parentgeneration improved their education levels over the generation of grandparents, grandparents whose children failed in edu-cational competition may have particularly unfavorable attitudes towards investing in grandchildren’s education. The feelingof relative failure in educational competition could be even more substantial for grandparents who had higher levels of edu-cation relative to other peers in their generation, but whose children made considerable downward mobility by ending upwith relatively lower levels of education compared with peers in the respective generation. In this situation, grandparentsmay be particularly pessimistic about values of education for their grandchildren and instead encourage grandchildren toenter labor market sooner than later. In other words, when parents have lower levels of education, having grandparentswho were more successful in education relative to their peers may adversely affect grandchildren’ educational continuation,if having any effect.

3. The Taiwanese context

This paper scrutinizes the prominence of grandparents’ education on grandchildren’s education in Taiwan. Taiwan pro-vides an interesting context to examine whether and how grandparents’ education affects grandchildren’s educationalattainment under the rapid expansion of education. Although educational expansion occurred in many parts of the worldduring the second half of the 20th century (Schofer and Meyer, 2005), the degree of expansion was particularly considerablein Taiwan. Taiwanese mandatory education increased from six to nine years in 1968. Since then, secondary and tertiary edu-cational systems have expanded substantially, such that the proportion of population (ages 15 and above) with a collegedegree more than quintupled over three decades: from 7% in 1976 to 33% in 2006 (Ministry of Education, 2013).

166 Y.-L. Chiang, H. Park / Social Science Research 51 (2015) 163–173

With regard to educational transitions, government statistics (Ministry of Education, 2013) show that high school edu-cation, grades 10–12, has become near-universal in Taiwan—the proportion of middle school graduates enrolling in highschools increased from 51% to 96% between 1950 and 2006. This transition rate is impressive, given that high school isnot compulsory education and enrollment is determined by student performance in the standardized entrance examination,administered to students upon middle school graduation in 9th grade. Although the overall high school attendance hasbecome near-universal, it is important to note that middle school graduates are separated into two different types of highschool: academic and vocational high schools. Students’ admissions to academic and vocational high schools are determinedby students’ test scores in the entrance examination. Typically, academic high schools are perceived as more prestigious andrequire higher admission cutoff scores than vocational high schools. Students in the academic track are also more likely tocome from higher socioeconomic backgrounds than their peers in the vocational track (Lin, 1999).

After high school, students can move to tertiary education that consists of four-year universities and two-year junior col-leges. The transition rates from high school to tertiary institutions have increased substantially to the extent that 91% ofacademic high school graduates and 70% of vocational high school graduate made transition to tertiary education in 2006(Ministry of Education, 2013). In other words, in Taiwan, even tertiary education has become mass education in whichthe majority of high school graduates enroll in tertiary education after high school. This trend indicates that the trackdifference between four-year universities and two-year junior colleges has become more meaningful in educational strati-fication than the mere attendance in tertiary education.

Another context of Taiwan to be highlighted in light of multigenerational effects is the comparably high level of coresi-dence with grandparents. Households in which grandchildren coresided with grandparents accounted for as high as one-third of all households in Taiwan (Taiwan Social Change Survey, 2005). The comparable figure in the U.S. was approximately16% in 2008 (Pew Research Center, 2010). The high level of coresidence with grandparents suggests that the effect of grand-parental characteristics on grandchildren’s educational outcomes can be particularly evident in Taiwan. As Zeng and Xie(2014) suggested, living in the same household implies frequent contact and allows grandparents to establish close relation-ship, which may facilitate direct influences of grandparents on grandchildren’s educational outcomes. Yet, it is important tonote that Zeng and Xi focused on the possibility of school dropout in rural China. Grandparents’ coresidence may increasegrandchildren’s chances of staying in school by reducing grandchildren’s time for household chores and other possible familylabor demands. However, grandparents with fairly low levels of education such as those in rural China or even in Taiwan(reflecting the recent expansion of education) may not know much about school curriculum and schooling processes, andtherefore may not be able to directly benefit grandchildren’s academic achievement or transitions to the next levels of edu-cation. In other words, it remains an empirical question whether grandparents coresiding with grandchildren can positivelyaffect grandchildren’s educational outcomes other than staying longer in school.

4. Data and methods

4.1. Data and variables

Our data come from the Taiwan Youth Project (TYP), a longitudinal survey of students and parents in northern Taiwan.The data sampled 5711 students in 7th (1st year in middle school) and 9th grades (3rd and last year in middle school)through multi-stage stratified cluster sampling. The students were surveyed annually since the launch of the project in2000. Parents of the students answered parental questionnaires to provide information on their own parents (i.e., grandpar-ents) and other family environments. In other words, with the TYP dataset, we can obtain information on educational attain-ment over three generations. Moreover, the longitudinal TYP survey contains information on students’ timing of transitionbetween educational institution, which allows us to retain the students who do not go through standard transitions to highschool and college in the sample.2

We combine student and parent data from Wave 1 to Wave 8 (2000–2007) for the 7th grade cohort and Wave 1 to Wave 6(2000–2005) for the 9th grade cohort. By doing so, the combined data contains student information up to two years afterhigh school graduation for both cohorts. As many longitudinal surveys, TYP suffers from attrition. For the purpose of thisstudy, we dropped the students whose high school enrollment status was not identified due to either attrition or non-response (N = 522). We further excluded students for whom we could not identify whether a mother or a father answeredthe parent questionnaire where we can obtain information on grandparents (N = 218). With reasons explained later, we carryout separate analyses for students whose mothers answered the parent questionnaire from students whose fathersanswered the parent questionnaire. The final sample used to analyze transition from middle school to high school consistsof 4971 students. We distinguish students who moved from middle school to academic high school (N = 2271) from theircounterparts who transited to vocational high school (N = 2438). Among our 4971 respondents, only 262 did not advanceto high school. Considering its small number, we combine those who did not attend high school with those who transitionedto vocational high school.3

2 Examples of non-standard transitions include taking fewer or additional years to prepare for the high school and college examination, boys enteringmilitary service before college, or students switching between vocational and academic track.

3 Note that we will call them vocational high school students for convenience even though 262 cases did not attend high school.

Y.-L. Chiang, H. Park / Social Science Research 51 (2015) 163–173 167

The second outcome examined in the current study is whether students attended university by up to two years after highschool graduation. Given the substantial proportion of Taiwanese high school graduates making transition to tertiary edu-cation as described earlier, it seems more meaningful to examine university attendance vs. no university attendance ratherthan to examine tertiary education vs. no tertiary education. In our analysis, the category of no university attendanceincludes those who attended junior college and those who did not attend tertiary education at all. After excluding respon-dents who did not attend any high school and those whose college enrollment status was not identified, the sample wasreduced to 4047. The final sample for the analysis of transition from high school to university consists of 3907 studentsfor whom we could identify whether a mother or a father filled out the parent questionnaire. In the final sample, 1632attended university by two years after high school and the remaining 2275 enrolled in junior college (N = 1568) or didnot attend any tertiary education (N = 707).

Our key independent variables are grandparents’ highest levels of education. These variables come from parents’ answersto the question ‘‘What are your father and mother’s levels of education?’’ in 2000 (Wave 1). The parent interviewees in Wave1 consist of two-thirds of the mothers and one-third of the fathers. Therefore, depending on who answered the parent ques-tionnaire, we have information on either maternal (when a mother answered the parent questionnaire) or paternal (when afather answered the parent questionnaire) grandparents, but not both sides of grandparents. Considering the possibility thatstudents whose mothers answered the parent questionnaire may differ in their characteristics from students whose fathersdid, we conduct separate analysis for the two types of students. In order to investigate how grandmothers and grandfathersmay have different effects, moreover, we include both grandmother’s and grandfather’s years of schooling in the same model(see Pfeffer, 2014). Given that grandparents’ levels of education were reported retrospectively by parents, missing informa-tion on grandparents’ education could be non-random. Fortunately, the percentage of students who have missing informa-tion on either grandmother’s or grandfather’s education is relatively small (6% in each group of students in maternal andpaternal lineages). We retain those students missing on grandparents’ education using multiple imputation.

In estimating the effects of grandparents’ years of schooling on grandchildren’s educational outcomes, we control for sev-eral individual and family background characteristics. Parents’ years of schooling are based on parents’ reports of their ownand spouse’s levels of education. Our measure of log family income utilizes answers from multiple waves of survey. We firstuse parents’ reported family income (per thousand TWD) when the students were in 9th grade. The missing values arereplaced with answers in the other waves of the parental survey, starting from the nearest. We then use students’ reporton family income in 9th grade if the parents did not respond to family income questions in any survey. Three-generationcoresidence is a dummy variable of whether students co-resided with any grandparent at 9th grade (0 = no, 1 = yes).Stratification research in Taiwan has demonstrated the relevance of ethnicity in educational attainment, showing that Main-landers are advantaged over other ethnic groups (Chen, 2005; Jao and McLeever, 2006). Thus, we control for ethnicity usingfather’s ethnicity as a categorical variable (1 = Minnan, 2 = Mainlander, 3 = Hakka or Aboriginal). Other control variablesinclude student’s gender (0 = male, 1 = female), residential area (1 = urban, 2 = suburban, 3 = rural), and sampled cohort(0 = 7th grade cohort, 1 = 9th grade cohort). For all control variables, except for residential area and cohort variables with

Table 1Descriptive statistics for student samples.

Students: maternal grandparents (n = 3343)a Students: paternal grandparents (n = 1628)b

percentage or mean SD percentage or mean SD

Academic high school attendance 48.8 – 39.4 –University attendancec 44.7 – 35.3 –Grandfather’s years of schooling 6.7 4.3 6.8 4.4Grandmother’s years of schooling 4.5 4.2 4.8 4.3Mother’s years of schooling 10.6 3.0 9.5 3.0Father’s years of schooling 11.0 3.3 10.7 3.1Log family income 3.9 0.8 3.9 0.8Female 53.0 – 43.2 –Coresidence with grandparents at 9th grade 25.0 – 29.1Father’s ethnicity

Minnan 77.5 – 80.5 –Mainlander 14.0 – 11.4 –Hakka or aboriginal 8.6 – 8.1 –

LocationUrban 38.7 – 33.7 –Suburban 39.8 – 41.2 –Rural 21.5 – 25.2 –

Cohort: 9th grade (vs. 7th grade) 52.1 53.4 –

a This group of students refer to those whose mothers answered the parent questionnaire so that only maternal grandparents’ years of schooling areknown.

b This group of students refer to those whose fathers answered the parent questionnaire so that only paternal grandparents’ years of schooling are known.c The same size for university attendance is 2692 for students of maternal lineage and 1215 for students of paternal lineage, respectively.

168 Y.-L. Chiang, H. Park / Social Science Research 51 (2015) 163–173

no missing cases, we use imputed values from multiple imputation with five imputed datasets.4 Table 1 presents the descrip-tive statistics for all the variables used in our analyses.

4.2. Methods

Using five imputed datasets, we run logistic regression to predict academic high school attendance and university atten-dance for students of maternal lineage and students of paternal lineage separately. Three separate models are estimated foreach outcome. Model 1 includes grandmother’s and grandfather’s years of schooling as well as all other independent vari-ables, except for mother’s and father’s years of schooling. This model shows whether grandmother’s and grandfather’s yearsof schooling are related to grandchildren’s education before controlling for parents’ education. Model 2 additionally controlsfor mother’s and father’s years of schooling to test whether grandmother’s and grandfather’s years of schooling remainassociated with grandchildren’s education net of parents’ years of schooling.

In Model 3, we add two interaction terms to examine whether the effects of grandparents’ education vary by parents’education. Specifically, the two interaction terms in the analysis of students of maternal lineage are: (1) maternal grand-mother’s years of schooling�mother’s years of schooling; and (2) maternal grandfather’s years of schooling �mother’syears of schooling. The corresponding two interaction terms in the analysis of students of paternal lineage are: (1) paternalgrandmother’s years of schooling � father’s years of schooling; and (2) paternal grandfathers’ years of schooling � father’syears of schooling. Positive coefficients of the interaction terms indicate that the beneficial effects of grandparents’ educationare stronger for parents with more years of schooling, while the negative coefficients indicate an opposite relation. For easierinterpretation of the results in the interaction terms, we provide a graphic presentation to demonstrate changes in log oddsfor grandchildren to attend academic high school or university across different years of grandparents’ and parents’schooling.5

5. Results

5.1. Academic high school attendance

We first discuss the results for students of the maternal lineage. Model 1 in Table 2 shows that, for students who have infor-mation on maternal grandparents, maternal grandfather’s education is positively associated with the odds for grandchildrento attend academic high school when controlling for all independent variables except for mother’s and father’s education. Incontrast to maternal grandfather’s education, maternal grandmother’s education is not associated with increased odds ofattending academic high school. Moreover, once mother’s and father’s years of schooling are included in Model 2, the positiveeffect of grandfather’s education disappears. However, as pointed out earlier, this null effect of grandparents’ education inModel 2 may disguise varying effects of grandparents’ education across different levels of parents’ education.

To investigate the hypothesis of differential effects of grandparents’ education by parents’ education, we include interac-tion terms between maternal grandparents’ education and mother’s education in Model 3. The results show that the inter-action term between maternal grandfather’s education and mother’s education is significantly positive (0.008), whichindicates that children whose mothers have higher levels of education reap larger benefits from additional years of maternalgrandfather’s schooling than children whose mothers have lower levels of education. In other words, the pattern of positiveinteraction is consistent with our augmentation hypothesis. The interaction term between maternal grandmother’s educa-tion and mother’s education is also positive, although it is not significant.

To facilitate interpretation of interaction coefficients in Model 3, we illustrate how the log odds for grandchildren toattend academic high school vary across different years of maternal grandfather’s schooling (0–18) for three different yearsof mother’s schooling (9, 12, and 16 years of schooling) in Fig. 1.6 For students whose mothers received university education(i.e. 16 years of schooling), maternal grandfather’s additional years of schooling increase the log odds of grandchildren attendingacademic high school. For students whose mothers received a high school diploma (i.e. 12 years of schooling), maternal grand-father’s years of schooling hardly make any change in the log odds of attending academic high school. Similar to students whosemothers have 12 years of schooling, maternal grandfathers’ additional years of schooling do not increase log odd for studentswith middle school-educated mothers (i.e. 9 years of schooling).

Although the log odds seem to decrease along with additional years of grandfather’s schooling for the group of studentswith middle school-educated mothers, we caution the interpretation of this finding. In our data, 90% of the maternal grand-fathers of students with middle-school educated mothers have 9 or less years of schooling. Therefore, while the figure shows

4 We do not use imputed values for the dependent variables, which are academic high school attendance and university attendance. Those who were missingon dependent variables were also included for multiple imputation for independent variables but were excluded for the analysis of predicting the educationaloutcomes of students. This strategy was recommended by von Hippel (2007).

5 As an anonymous review pointed out, comparisons of logit coefficients across different groups can be affected by differences in residual variation acrossgroups (Allison, 1999; Mood, 2010). To test whether our results were robust, we estimated linear probability models as well as logit models. The interactionterms in linear probability models showed similar patterns with those in logit models.

6 For Fig. 1, we estimated Model 3 again but without the interaction term between grandmother’s years of schooling and mother’s years of schooling, whichwas insignificant in Model 3 in Table 2. We fixed grandmother’s years of schooling as 6, father’s years of schooling as 12, log family income as the mean value,coresidence with grandparents as not coresiding, father’s ethnicity as Minnan, gender as male, location as urban, and cohort as 7th grade cohort.

Table 2Logit models of grandchildren’s academic high school attendance by grandparents’ years of schooling.

Independent variables Maternal grandparents Paternal grandparents

M1 M2 M3 M1 M2 M3

Grandfather’s years of schooling 0.028* �0.014 �0.103* 0.005 �0.035 �0.174*

(0.012) (0.012) (0.046) (0.017) (0.018) (0.068)Grandmother’s years of schooling �0.019 �0.022 �0.077 0.000 �0.001 �0.006

(0.011) (0.012) (0.052) (0.017) (0.018) (0.071)Mother’s years of schooling 0.115*** 0.044 0.063** 0.058*

(0.017) (0.025) (0.024) (0.024)Father’s years of schooling 0.126*** 0.124*** 0.173*** 0.100**

(0.016) (0.016) (0.025) (0.036)Grandfather’s years of schooling �mother’s years of schooling 0.008*

(0.004)Grandmother’s years of schooling �mother’s years of schooling 0.005

(0.005)Grandfather’s years of schooling � father’s years of schooling 0.012*

(0.006)Grandmother’s years of schooling � father’s years of schooling 0.001

(0.006)Log family income 0.507*** 0.212*** 0.208*** 0.472*** 0.161 0.158

(0.057) (0.057) (0.057) (0.082) (0.082) (0.083)Female 0.070 0.092 0.084 0.116 0.076 0.071

(0.072) (0.074) (0.075) (0.106) (0.106) (0.110)Coresidence with grandparents at 9th grade �0.052 �0.085 �0.074 �0.317** �0.383** �0.386**

(0.084) (0.087) (0.087) (0.121) (0.125) (0.125)Father’s ethnicity (ref: Minnan)Mainlander 0.308** 0.040 0.004 0.127 �0.118 �0.168

(0.110) (0.117) (0.119) (0.168) (0.176) (0.179)Hakka or aboriginal �0.236 �0.324* �0.331* 0.105 �0.098 �0.093

(0.131) (0.135) (0.136) (0.192) (0.198) (0.200)Location (ref: urban)Suburban �0.393*** �0.105 �0.092 �0.251* �0.021* �0.016

(0.082) (0.087) (0.087) (0.121) (0.128) (0.128)Rural �0.561*** �0.305** �0.293** �0.693*** �0.533** �0.505**

(0.102) (0.107) (0.107) (0.148) (0.154) (0.155)9th Grade cohort �0.134 �0.094 �0.101 �0.208* �0.175 �0.170

(0.072) (0.074) (0.075) (0.105) (0.109) (0.109)Constant �1.829*** �3.138*** �2.382*** �1.902*** �2.967*** �2.131***

(0.247) (0.261) (0.319) (0.359) (0.368) (0.469)Observations 3343 3343 3343 1628 1628 1628

Note: All the missing values on independent variables were imputed by multiple imputation (5 different datasets). Standard errors are in parentheses.* p < 0.05.

** p < 0.01.*** p < 0.001.

-1.2

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

1.2

0 2 4 6 8 10 12 14 16 18

log

odds

Grandfather's years of schooling

Mother's years of schooling = 9Mother's years of schooling = 12Mother's years of schooling = 16

Fig. 1. Log odds of attending academic high school by materna grandfather’s education and mother’’s education.

Y.-L. Chiang, H. Park / Social Science Research 51 (2015) 163–173 169

170 Y.-L. Chiang, H. Park / Social Science Research 51 (2015) 163–173

the expected log odds up to 18 years of maternal grandfather’s schooling, the line for students with middle school-educatedmothers would mostly end before 10 years of grandfather’s schooling. In other words, the seemingly declining pattern maybe artificial.

On the other hand, in our earlier discussion on the augmentation hypothesis, we pointed out the possibly adverse effect ofhaving educationally successful grandparents on grandchildren’s educational continuation when parents experienced down-ward mobility in education despite the rapid expansion of educational system. Our data do not allow us to systematicallytest this potentially negative effect of grandparents’ education, and therefore we are cautious in making a strong claim thatthe log odds for grandchildren to attend academic high school indeed decrease with additional years of grandfather’s school-ing. With regard to mothers who were university graduates, however, more than 30% of these mothers have fathers with sixor less years of schooling, which reflects the rapid expansion of education in Taiwan. Therefore, the solid line in Fig. 1 is a fairpresentation for students whose mothers have university education (16 years of schooling).

An important consequence of the differential effects of maternal grandfather’s education by mother’s education in Fig. 1 isthe increased gap in the likelihood of attending academic high school between students with more educated mothers andstudents with less educated mothers, as grandfather’s education increases. We find that the gap between students with moreeducated mothers and their counterparts with less educated mothers is much larger among students whose grandfathershave relatively more years of schooling than students whose grandfathers have relatively fewer years of schooling. Thisdiverging gap in relation to increased years of grandfather’s schooling is consistent with the pattern expected by ouraugmentation hypothesis.

Turning to the results for the paternal lineage, we do not find any positive relationship between grandparents’ and grand-children’s education in either Model 1 or 2. However, similar to the results for the maternal lineage, the interaction termbetween paternal grandfather’s education and father’s education is significantly positive in Model 3. In other words, thereis evidence that the effect of paternal grandfather’s education depends on father’s education. Students of fathers with moreyears of schooling particularly benefit from grandfathers’ additional years of schooling. We do not repeatedly presentanother figure to show changes in log odds of academic high school attendance for the paternal lineage. However, similarto Fig. 1, we find that the gap in the likelihood of attending academic high school increases between students with fathershaving higher and lower levels of education as paternal grandfathers’ years of schooling increase.

Other control variables associated with academic high school attendance in Table 2 are log family income, ethnicity, andresidential area. However, we also find that the ways in which those control variables are related to academic high schoolattendance somewhat differ by lineages. For instance, father’s ethnicity (Hakka or aboriginal) is significantly related to aca-demic high school attendance for students whose mothers answered the questionnaire (and thus only maternal grandpar-ents’ years of schooling are known) but not for students whose fathers did (and thus only paternal grandparents’ years ofschooling are known). On the other hand, coresidence with grandparent(s) is not significantly associated with academic highschool attendance for students whose mothers answered the questionnaire, but is negatively associated for students whosefathers were the responding parent. These differences between the two types of students confirm our decision to conductseparate analysis by lineages.

It is worth discussing the effect of coresidence with grandparent(s) in more detail. As described above, Zeng and Xie(2014) found that grandparents’ educational levels were significant for grandchildren’s staying in school only whengrandparents coresided with grandchildren in rural China. To address the proposed pattern by Zeng and Xie, we includedinteraction terms between coresidence and grandparents’ education in supplementary analysis but found them to be insig-nificant. However, our coresidence variable is not ideal in that it does not necessarily indicate that students lived with thegrandparent(s) whose years of schooling were available in the dataset. For instance, students could live with paternal grand-parent(s) but only have information on maternal grandparents’ years of schooling because mothers answered thequestionnaire.

To check for the robustness of our result, we conducted an additional analysis in which we selected only students forwhom coresident grandparents were those whose years of schooling were known. For these students, we consistently foundno significant interaction between coresidence and grandparents’ education. Given that grandchildren in Taiwan are morelikely to live with paternal grandparents than maternal grandparents, most students of paternal lineages have informationon coresiding grandparents (who are mostly paternal grandparents). As seen in Table 2, in this subgroup of students, we finda negative relationship between coresidence and academic attendance. In a supplementary analysis that included interactionterms (not reported), we found no significant interaction between coresidence and grandparent’s education. In short, ourresult is not consistent with Zeng and Xie’s finding of positive interaction between coresidence and grandparent’s education.

5.2. University attendance

We turn to the results for university attendance in Table 3. Model 1 shows that, for students whose maternal grandpar-ents’ years of schooling are available, grandfather’s education is significantly associated with increased odds of attendinguniversity. Similar to the result of Model 1 in Table 2 for high school attendance, grandmother’s education does not increasethe odds of attending university. Both grandfather’s and grandmother’s years of schooling are not related to university atten-dance once mother’s and father’s years of schooling are included in Model 2. Furthermore, Model 3 shows that the null effectof grandmother’s education in Model 2 is the result of differential effects of grandmother’s education by mother’s education.The positive interaction coefficient (0.01) is significant at the .1 level (p = 0.057), indicating that students of mothers who

Table 3Logit models of grandchildren’s university attendance by grandparents’ years of schooling.

Independent variables Maternal grandparents Paternal grandparents

M1 M2 M3 M1 M2 M3

Grandfather’s years of schooling 0.028* �0.016 �0.042 0.002 �0.048* �0.210*

(0.013) (0.014) (0.054) (0.020) (0.022) (0.085)Grandmother’s years of schooling �0.016 �0.019 �0.135* 0.002 0.006 0.068

(0.013) (0.013) (0.062) (0.019) (0.021) (0.087)Mother’s years of schooling 0.133*** 0.079** 0.061* 0.056

(0.020) (0.029) (0.029) (0.029)Father’s years of schooling 0.120*** 0.118* 0.218*** 0.162***

(0.018) (0.019) (0.031) (0.043)Grandfather’s years of schooling �mother’s years of schooling 0.003

(0.005)Grandmother’s years of schooling �mother’s years of schooling 0.010

(0.005)Grandfather’s years of schooling � father’s years of schooling 0.014*

(0.007)Grandmother’s years of schooling � father’s years of schooling �0.005

(0.007)Log family income 0.515*** 0.186** 0.181** 0.692*** 0.259* 0.253*

(0.066) (0.066) (0.067) (0.112) (0.114) (0.115)Female 0.296*** 0.350*** 0.344*** 0.107 0.066 0.064

(0.081) (0.084) (0.085) (0.126) (0.131) (0.132)Coresidence with grandparents at 9th grade �0.085 �0.114 �0.103 �0.092 �0.146 �0.147

(0.093) (0.096) (0.097) (0.140) (0.146) (0.146)Father’s ethnicity (ref: Minnan)Mainlander 0.332** 0.076 0.053 0.107 �0.188 �0.222

(0.122) (0.129) (0.130) (0.197) (0.209) (0.211)Hakka or aboriginal �0.172 �0.246 �0.253 0.078 �0.181 �0.185

(0.154) (0.158) (0.159) (0.234) (0.247) (0.248)Location (ref: urban)Suburban �0.616** �0.320** �0.311** �0.531*** �0.263 �0.254

(0.093) (0.098) (0.099) (0.145) (0.154) (0.155)Rural �0.479*** �0.217 �0.205^ �0.692*** �0.511** �0.477**

(0.112) (0.118) (0.118) (0.170) (0.180) (0.180)9th Grade cohort �0.056 �0.023 �0.037 �0.035 �0.001 0.000

(0.081) (0.084) (0.084) (0.125) (0.131) (0.131)Constant �2.159*** �3.509*** �2.902*** �2.991*** �4.063*** �3.387***

(0.289) (0.303) (0.370) (0.481) (0.488) (0.608)Observations 2692 2692 2692 1215 1215 1215

Note: All the missing values on independent variables were imputed by multiple imputation (5 different datasets). Standard errors are in parentheses.* p < 0.05.

** p < 0.01.*** p < 0.001.

Y.-L. Chiang, H. Park / Social Science Research 51 (2015) 163–173 171

have more years of schooling benefit more from maternal grandmothers’ additional years of schooling than do students ofmothers who have less years of schooling. Compared to the results for academic high school attendance in Table 2, in Table 3the interaction between maternal grandmother’s education and mother’s education is significant, but not the interactionbetween maternal grandfather’s education and mother’s education. Nevertheless, the significantly positive coefficients areconsistent with our augmentation hypothesis.

The result for students who have information on their paternal grandparents in Table 3 is also similar to the result foracademic high school attendance in Table 2. Both paternal grandfather’s and grandmother’s years of schooling are not sig-nificantly related to university attendance even prior to controlling for mother’s and father’s years of schooling; parents’ lev-els of education are significantly associated with increased odds for their children to attend university. However, Model 3lends support to the augmentation hypothesis by showing a significantly positive interaction between paternal grandfather’seducation and father’s education in affecting students’ university attendance. For students who have information on paternalgrandparents’ education, we find a significantly positive interaction between grandfather’s education and father’s educationfor both academic high school attendance and university attendance. We do not find any significant interaction betweenfather’s education and paternal grandmother’s education.

6. Conclusion

In response to the growing interest in multigenerational effects in educational stratification processes, we have examinedhow grandmother’s and grandfather’s years of schooling are related to grandchildren’s educational outcomes (academic highschool and university attendance), net of parents’ education in Taiwan. In particular, we have attempted to expand existing

172 Y.-L. Chiang, H. Park / Social Science Research 51 (2015) 163–173

literature on multigenerational effects, which tends to ignore potential heterogeneity of the grandparent effect, by investi-gating how the effects of grandparents’ education differ by parents’ education. Drawing on literature on social capital, wehave proposed an augmentation hypothesis, which suggests that grandparents’ education should augment parents’ educa-tion in affecting grandchildren’s educational outcomes. In other words, we expect that students with more educated parentsare more likely to reap benefits from grandparents’ education than students with less educated parents.

We find evidence supporting our augmentation hypothesis for both educational outcomes examined in this study. Addi-tional years of grandparents’ schooling increase the odds of attending academic high school and university only for studentswhose parents have high levels of education. In contrast, additional years of grandparents’ education do not increase the logodds of attending academic high school or university for students whose parents have middle or low levels of education. Dueto the differential effect of grandparent’s education by parents’ education, the gap in the likelihood of attending academichigh school (and university) between students with parents in higher and lower ends of educational hierarchy diverges alongwith the increased years of grandparents’ education. In other words, grandparents’ education augments inequality in edu-cational outcomes posed by parents’ education.

Our finding contradicts Jæger’s (2012) alternative hypothesis on the function of grandparents’ education, which wasfound to compensate for the lack of parents’ education. According to the compensation hypothesis, grandparents’ educationmay ameliorate educational inequality associated with parents’ education. However, we do not find evidence supporting thecompensation hypothesis in Taiwan. Rather, in our study, grandparents’ education is found to increase educational inequal-ity by parents’ education because students with more educated parents benefit more from grandparents’ education than stu-dents with less educated parents. Our review of literature on social capital suggests that grandparents’ education as a socialcapital embedded in family networks should be more likely to augment, rather than to compensate, parents’ education.

Given the important implications multigenerational studies hold for educational inequality, more research that examineswhether grandparents compensate or augment parents in different contexts is needed. We acknowledge that our conclusionfrom Taiwan cannot be easily generalized to other Western contexts that vastly differ in education and family systems. How-ever, we also note that several countries, especially neighboring countries like Japan and Korea, share with Taiwan some keyaspects of social institutions, particularly the degrees of educational expansion, features of the educational system (e.g.,Ishida 2007; Park 2007), and family structures, all of which are relevant for the investigation of multigenerational effects.Therefore, our understanding of the specific context in which grandparents’ education augments educational inequalityby parental education would be significantly expanded if future comparative research in Asian countries provides empiricalevidence that supports the augmentation hypothesis.

In addition to heterogeneity of the grandparent effect by parents’ education, we have separately examined grandmother’sand grandfather’s years of schooling for each group of students, for whom either maternal grandparents’ years of schoolingor paternal grandparents’ years of schooling are known. Compared to many previous studies, our study addressed potentialheterogeneity of the grandparent effect in various aspects, including the difference between grandmothers and grandfathersas well as the difference between lineages. Without controlling for parents’ levels of education, grandfather’s education butnot grandmother’s education is significantly associated with both academic high school attendance and university atten-dance for students whose maternal parents’ years of schooling are known. Moreover, grandfather’s education interacts witheither mother’s or father’s education to the extent to which educational inequality by parents’ education further diverges. Anexceptional case is university attendance for students with information on maternal grandparents, for whom it is not grand-father’s education but grandmother’s education that interacts with mother’s education. Our analysis does not provide anyapparent conclusion on differences between grandmothers and grandfathers effects or by lineage. However, as Pfeffer(2014: 2) suggested, more research on multigenerational effects should explore ‘‘heterogeneity across groups and popula-tions,’’ as the mechanisms and processes through which multigenerational effects are generated may differ.

A limitation of our data is that we only have information on either maternal grandparents or paternal grandparentsdepending on whether the mother or the father responded to the parent questionnaire. Due to this lack of information onthe other side of family, we could not directly compare the effects of maternal grandparents and paternal grandparentsfor the same group of students. Students whose mothers answered the parent questionnaire and their counterparts whosefathers answered likely differ in their individual and family characteristics (a glimpse of which shown is in Table 1). Since wedo not have additional information that gauges the correlation between maternal and paternal grandparents’ characteristics,we are not comfortable using multiple imputation to impute the years of schooling of maternal or paternal grandparents thatare completely missing. How maternal and paternal grandparents may have different impacts on grandchildren’s educationawaits data with information on both sides of grandparents.

Acknowledgments

Hyunjoon Park acknowledges support from the Academy of Korean Studies Grant funded by the Korean Government(MEST) (AKS-2010-DZZ-2101).

References

Allison, P.D., 1999. Comparing logit and probit coefficients across groups. Soc. Method Res. 28, 186–208.Amato, P.R., 2001. Children of divorce in the 1990s: an update of the Amato and Keith (1991) meta-analysis. J. Fam. Psychol. 15 (3), 355–370.

Y.-L. Chiang, H. Park / Social Science Research 51 (2015) 163–173 173

Amato, P.R., Booth, A., 1997. A Generation at Risk: Growing Up in an Era of Family Upheaval. Harvard University Press, Cambridge, MA.Blau, P., Duncan, O.D., 1967. The American Occupational Structure. John Wiley & Sons Inc., New York.Boudon, R., 1974. Education, Opportunity and Social Inequality: Changing Prospects in Western Society. Wiley, New York.Breen, R., Goldthorpe, J.H., 1997. Explaining educational differentials towards a formal rational action theory. Ration. Soc. 9, 275–305.Bourdieu, P., 1986. The forms of capital. In: Richardson, J. (Ed.), Handbook of Theory and Research for the Sociology of Education. Greenwood Press,

Westport, pp. 241–258.Buchmann, C., Hannum, E., 2001. Education and stratification in developing countries: a review of theories and research. Ann. Rev. Soc. 27, 77–102.Mood, C., 2010. Logistic regression: why we cannot do what we think we can do, and what we can do about it. Eur. Sociol. Rev. 26, 67–82.Chan, T.W., Boliver, V., 2013. The grandparents effect in social mobility: evidence from British birth cohort studies. Am. Sociol. Rev. 78, 662–678.Chen, W., 2005. Ethnicity, gender and class: ethnic difference in Taiwan’s educational attainment revisited. Taiwanese Sociol. 10, 1–40.Erola, J., Moisio, P., 2007. Social mobility over three generations in Finland, 1950–2000’’. Eur. Sociol. Rev. 23, 169–183.Hällsten, M., 2014. Inequality across three and four generations in egalitarian Sweden: 1st and 2nd cousin correlations in socio-economic outcomes. Res.

Soc. Stratif. Mobil. 35, 19–33.Von Hippel, P.T., 2007. Regression with missing Ys: an improved strategy for analyzing multiply imputed data. Sociol. Methodol. 37, 83–117.Ishida, H., 2007. Japan: educational expansion and inequality in access to higher education. In: Shavit, Y., Arum, R., Gamoran, A. (Eds.), Stratification in

Higher Education: A Comparative Study. Stanford University Press, Stanford, CA.Jæger, M.M., 2012. The extended family and children’s educational success. Am. Sociol. Rev. 77, 903–922.Jao, J., McLeever, M., 2006. Ethnic inequalities and educational attainment in Taiwan. Sociol. Educ. 79, 131–152.Lareau, A., 1987. Social class differences in family-school relationships: the importance of cultural capital. Sociol. Educ. 60, 73–85.Lareau, A., Horvat, E.M., 1999. Moments of social inclusion and exclusion: race, class, and cultural capital in family-school relationships. Sociol. Educ. 71, 37–

53.Lin, Da.-Sen., 1999. The effects of family background on tracking of secondary education in Taiwan: a study of the distinction between ‘‘academic/

vocational’’ and ‘‘public/private’’ tracking. Soochow J. Sociol. 8, 35–77.Lindahl, Mikael, Palme, Mårten, Sandgren Massih, Sofia, Sjögren, Anna, 2012. The intergenerational persistence of human capital: an empirical analysis of

four generations. IZA Discussion Paper No. 6463.Mare, R.D., 2011. A multigenerational view of inequality. Demography 48, 1–23.Mare, R.D., 2014. Multigenerational aspects of social stratification: issues for further research. Res. Soc. Stratif. Mobil. 35, 121–128.McNeal Jr., R.B., 1999. Parental involvement as social capital: differential effectiveness on science achievement, truancy, and dropping out. Soc. Forces 78,

17–144.Ministry of Education, Republic of China (Taiwan), 2013. 2013 Education Statistical Indicators. <http://www.edu.tw/pages/

detail.aspx?Node=1052&Page=16582&wid=31d75a44-efff-4c44-a075-15a9eb7aecdf&Index=1>. (Accessed 22.06.13).Newman, K.S., 1988. Falling from Grace. The Experience of Downward Mobility in the American Middle Class. Vintage Books, New York.Park, H., 2007. South Korea: educational expansion and inequality of opportunity for higher education. In: Shavit, Y., Arum, R., Gamoran, A. (Eds.),

Stratification in Higher Education: A Comparative Study. Stanford University Press, Stanford, CA.Perna, L.W., Titus, M.A., 2005. The relationship between parental involvement as social capital and college enrollment: an examination of racial/ethnic

group differences. J. High. Educ. 76, 485–518.Pew Research Center, 2010. The return of the multi-generational family household. Pew Research Center, Washington, DC.Pfeffer, F.T., 2014. Multigenerational approaches to social mobility: a multifaceted research agenda. Res. Soc. Stratif. Mobil. 35, 1–12.Portes, A., 1998. Social capital: its origins and applications in modern sociology. Ann. Rev. Sociol. 24, 1–24.Raftery, A.E., Hout, M., 1993. Maximally maintained inequality: expansion, reform, and opportunity in Irish education, 1921–75. Sociol. Educ. 66, 41–62.Ren, L., Hu, G., 2013. A comparative study of family social capital and literacy practices in Singapore. J. Early Childhood Lit. 13, 98–130.Schofer, E., Meyer, J.W., 2005. The worldwide expansion of higher education in the twentieth century. Am. Sociol. Rev. 70, 898–920.Shavit, Y., Blossfeld, H. (Eds.), 1993. Persistent Inequality: Changing Educational Attainment in Thirteen Countries. Social Inequality Series. Robert Westview

Press, Boulder.Solon, G., 2013. Theoretical models of inequality: transmission across multiple generations. Res. Soc. Stratif. Mobil. 35, 13–18.Warren, J.R., Hauser, R.M., 1997. Social stratification across three generations: new evidence from the Wisconsin longitudinal study. Am. Sociol. Rev. 62,

561–572.Wong, R.S.K., 2002. Cultural and social capital in educational research: commentary on the Buchmann paper. Res. Sociol. Educ. 13, 163–171.Zeng, Z., Xie, Y., 2014. The effects of grandparents on children’s schooling: evidence from Rural China. Demography 51, 599–617.