Maternal Education and Early Childhood Education across Affluent English-Speaking Countries

Abstract

Women who attain more education tend to have children with more educational opportunities, a transmission of educational advantages across generations that is embedded in the larger structures of families’ societies. Investigating such country-level variation with a life course model, this study estimated associations of mothers’ educational attainment with their young children’s enrollment in early childhood education and engagement in cognitively stimulating activities in a pooled sample of 36,400 children (n = 17,900 girls, 18,500 boys) drawn from nationally representative datasets from Australia, Ireland, United Kingdom, and United States. Results showed that having a mother with a college degree generally differentiated young children on these two outcomes more in the United States, potentially reflecting processes related to strong relative advantage (i.e., maternal education matters more in populations with lower rates of women’s educational attainment) and weak contingent protection (i.e., it matters more in societies with less policy investment in families).

Connections between mothers’ lives and children’s lives are a core component of the intergenerational transmission of inequality (Elder, 1998). Consider the example of education. When mothers attain advanced educational credentials, their children tend to have more formal and informal opportunities for learning and cognitive stimulation that support their eventual educational attainment. An intergenerational transmission such as this one occurs through many channels, including the socioeconomic resources that educational attainment often brings as well as what is going on inside the child’s proximate developmental ecology (Gennetian, Magnuson, & Morris, 2008; Davis-Kean, 2005). Such ecologically embedded transactions between mothers and children need to be situated within the countries in which families live, as different countries provide their own economic conditions, normative systems, and policy settings that can facilitate or block how mothers’ advantages become advantages for children (Bradbury et al., 2015).

This study investigates a life course model comparing linkages between individual mothers’ educational attainment (receiving a college degree) and their children’s early educational opportunities (enrollment in early childhood education, engagement in cognitively stimulating activities such as shared reading) across four affluent English-speaking countries at different levels of the modern “liberal welfare state” (Australia, Ireland, United Kingdom, United States). We do so through statistical analyses of representative datasets from each country following birth cohorts of children in the 2000s. Such a comparative approach can provide insights into how aspects of national context influence the relative standing of mothers within their societies and the scaffolding available to reduce inequalities among their children.

Mothers, Children, and Linked Lives

The intergenerational transmission of educational advantages can occur through both mothers and fathers but in different ways. For fathers, the transmission is often studied in terms of the statuses that their educational attainment bestows on children (i.e., middle class or higher position), but, for mothers, it is usually studied in terms of how their educational attainment shapes their interactions with children (i.e., parenting). Of course, both status-related and interactional processes are relevant to understanding fathers’ and mothers’ roles, and the two roles are connected through marriage, previous or current relationships, and shared or co-parenting (Jeong, McCoy, & Fink, 2017). The focus here is on mothers, given recent theoretical attention to how children’s future prospects increasingly have become tied to their mothers’ social and economic trajectories (see McLanahan, 2004) and the growing number of policies and program aiming to invest in mothers’ human capital as a means of promoting their children’s healthy development and socioeconomic attainment (Hernandez & Napierala, 2014).

Life course theory of human development, which encourages the treatment of human lives as dynamic and contextualized, is valuable for understanding the intergenerational transmission of educational advantages between mothers and children (Elder, 1998). One of its basic principles, linked lives, is that the unfolding trajectories of different family members need to be studied as intertwined across developmental time. Thus, the advantaged circumstances of mothers—such as having attained more education—can shape the future educational prospects of children. Another principle, sociohistorical context, is that linked lives are rooted in interpersonal interactions that are themselves embedded in larger social and historical forces (Crosnoe & Benner, 2015). Thus, the connections of mothers’ past and current circumstances with children’s current and future prospects emerge from macro-level structures (e.g., societal institutions, cultural traditions, policy regimes, stratification systems) (Silbereisen, 2005).

Delving deeper into the linked lives component of life course theory, multiple quasi-experimental analyses of national data in several countries suggest that observed associations between mothers’ educational attainment and indicators of children’s educational progress are at least partly causal (Carneiro, Meghir, & Parey, 2013; Oreopoulos, Page, & Stevens, 2006). Such associations are often attributed to three key mechanisms concerning how mothers’ educational experiences—particularly in higher education—translate into resources that allow them to invest in their children, elicit investments from others, and locate and secure opportunities for their children (LeVine et al., 2011).

The three mechanisms underlying the intergenerational transmission of educational advantages are in line with the rise in intensive mothering among college-educated mothers in many developed countries (Dotti Sani & Treas, 2016). First, such mothers are able to marshal practical resources that lift constraints on how they manage their children’s lives and open up new avenues for promoting their children’s interests. For example, more education can increase access to financial resources, with an earnings premium for college degrees evident in most developed countries (Chevalier et al., 2013; OECD, 2011). Education can also permit more control over one’s time, which helps mothers be more child-focused at home and secure supplemental supports and enriching activities for children (Davis-Kean, 2005; Lareau, 2004). With such resources, more educated mothers are better able to follow through on their own motivations for educationally supportive parenting with fewer obstacles or distractions. Second, more education can cultivate social psychological resources that shape how mothers proactively and reactively approach the educational experiences of their children to identify, maintain, and multiply advantages for them. Examples include the ways that educational attainment can increase the information mothers have about the educational system, expose them to models of parenting that “work the system”, and develop critical psychosocial skills (e.g., efficacy) (Kalil, Ryan, & Corey, 2012; Magnuson, 2007). With such resources, more educated mothers develop ideas of what their children need that are aligned with the structure and culture of schools and then actualize these ideas more consistently. Third, more education can raise mothers’ power in institutional systems by encouraging them to make demands and increasing the odds these demands are rewarded. Examples include the ways that the prestige and influence of a college degree can empower mothers to advocate for their children at school and have their concerns addressed (Augustine, 2014; Lareau, 2004). With such resources, more educated mothers can justify their working models of education-focused parenting, which reinforces their efforts.

College-educated mothers, therefore, are more likely to use practical, social psychological, and status-related resources to locate, create, and secure educational opportunities for young children. Opportunities of this kind are often outside the home, such as enrolling children in early childhood education programs. Such enrollment is associated with long-term educational attainment, although this association varies by program quality, social location of the family, and the dimension of attainment being considered (Duncan & Magnuson, 2013; Boocock, 1995). There are also many motivations behind enrollment, including a strict need to secure child care, a clear desire for early enrichment, and a mixture of both. Prior research in the United States has shown that more educated mothers consistently opt for early childhood education programs over other kinds of child care when child care is a necessity and access such programs even when they do not need child care because they view early enrichment as key to their children’s futures (Augustine et al., 2009; Waldfogel, 2006). Educational opportunities can also be within the home, such as cognitively stimulating children through parent-child interactions. Mothers can plan and engage in such stimulation in many ways, but much research has focused on shared reading as a concrete and teachable dimension. For example, evidence from the Programme for International Student Assessment has suggested that shared reading and other aspects of home literacy have the largest observed effects on academic achievement of any parenting behavior (OECD, 2012). Shared reading is associated with early literacy, which has cascading associations with academic indicators that then predict college enrollment (Lesnick et al., 2010; Cho, 2007). Thus, although early educational opportunities in and out of the home do not guarantee future educational attainment, they do tend to probabilistically increase the odds.

The Linked Lives of Mothers and Children in Country-Level Context

If the intergenerational transmission of educational advantages between mothers and children represents a linked lives phenomenon, then this phenomenon needs to be understood within sociohistorical context. Notably, the association between maternal education and children’s educational opportunities is robust in a variety of affluent Western countries (Chevalier et al., 2013). The cross-country similarity of this association is significant given differences in funding and quality of educational systems, even if the association is not deterministic in any country and does vary in magnitude across countries. Importantly, a similar pattern has been documented in developing countries and has long undergirded international aid and development (Grépin & Bharadwaj, 2015; Desai & Alva, 1998). Our focus here is not establishing this association across countries but rather examining how its magnitude might vary across countries. In other words, we approach country as a moderator.

Country-level moderation of the linkage between mothers’ educational attainment and children’s educational opportunities could be conceptualized in many ways, including levels of economic inequality (see Reeves, 2018) and gender egalitarianism (Geist, 2005). This study focuses on three dimensions of the country-level context that are theoretically driven and interdependent (see Gonzalez, Strohschein, & Crosnoe, 2018; Bradbury et al., 2015; Heuveline & Timberlake, 2004). First, countries can be characterized by their overall economic conditions (e.g., level of economic development, volatility, and inequality). Such economic conditions may magnify the significance of differences between mothers that have a differential impact on their children. The degree to which mothers are able to capitalize on their educational attainment to promote their children’s interests depends on the average economic opportunity where they live. Second, countries can be characterized by broad systems of norms about individual behavior, achievements, and statuses. Women likely judge themselves and are judged by others relative to such systems, which can make more apparent and consequential differences among mothers and subsequently their children. For example, the degree to which a mothers’ own educational attainment sets her apart from others and translates into a competitive edge for her children depends on how she compares to the norm of women’s educational attainment in her society. Third, countries can be characterized by the availability of social policies that buffer mothers and their children from the deleterious effects of disadvantaged circumstances (e.g., income assistance, child care subsidies). For example, policy supports can chip away at the differences in practical, social psychological, and status-related resources emerging from differential rates of educational attainment, so that divergence in maternal education matters less to children.

Directly testing these country-level mechanisms is challenging absent a large sample of countries with harmonized measurement. Still, we can conceptualize how they might converge to moderate links between mothers’ educational attainment and children’s educational opportunities and then examine that potential moderation, with the match between the data and hypotheses often quite telling (Yu, 2015). For that purpose, this study selected four countries that have basic commonalities but, within those commonalities, differ in important ways. Australia, Ireland, the United Kingdom, and the United States are predominantly white and English speaking, although the United States is the most racially/ethnically diverse. The four countries are all considered affluent relative to the rest of the world, although they vary in levels of affluence and economic inequality, and they each have strong educational systems through higher education, with different levels of public support but similar earnings premiums for college degrees. Finally, all are classified as “liberal” welfare regimes, which means that, compared to Nordic countries for example, they tend to have more market-based approaches to social policy and public attitudes more tolerant of inequality (Bradbury et al., 2015; Kamerman & Kahn, 1997).

To understand country-level variation in links between mothers’ education and children’s educational opportunities, the conceptual model of this study poses non-competing hypotheses that build on the three focal country-level parameters of economic conditions, normative systems, and social policies. The first—the relative advantage hypothesis—emphasizes that, within affluent countries, population norms will moderate these links (Gonzalez et al., 2018). The idea is that, if some status or circumstance facilitates access to academic opportunities, that advantage will increase as that status or circumstances becomes less common. If fewer mothers have that status or circumstances, those who do will have a greater competitive edge. Given rates of women’s educational attainment across the countries in question (see OECD, 2018a), this hypothesis suggests that the link between mothers’ college education and children’s educational opportunities would be strongestn the country with the lowest relative rate of female attainment of higher education (United States) and weakest in the country with the highest relative rate (Australia). The countries with intermediate rates (United Kingdom, Ireland) would fall in between. Because this hypothesis is about norms and how they shape social expectations and comparisons, it is likely to equally apply to mothers’ behaviors in and out of the home.

The second hypothesis—contingent protection—emphasizes that the generosity of state supports will moderate the links between mothers’ college education and children’s educational opportunities (Fomby, Cavanagh, & Goode, 2011). The idea is that, if some social status or family circumstance blocks access to academic opportunities and investments, that disadvantage will weaken as state support for families and children increases. In other words, the state can buffer against the intergenerational transmission of educational inequality (Bradbury et al., 2015), directly in terms of creating formal educational opportunities (e.g., publicly financing preschool programs) and indirectly by easing the barriers and challenges that can interfere with mothers’ engagement in more informal activities (e.g., work-family policies that reduce time constraints on mothers’ ability to spend quality time with children). Comparing countries in this way is difficult because of the many polices to consider, but one way is the per capita proportion of gross domestic product that is dedicated to family- and child focused policies (e.g., income assistance, state-funded health care) (OECD, 2018b). Thus, this hypothesis suggests that maternal education will do more to differentiate children’s educational opportunities in the country with the lowest proportional investment in such policies (United States) than in the three countries with more moderate spending (Australia, Ireland, United Kingdom).

After estimating the association of mothers’ college education with two kinds of educational opportunities within each country, this study compares the strength of these associations across countries. These comparisons will assess the value of each hypothesis.

Data and Methods

Data

This study drew on four country-level datasets. Each is nationally representative, following a focal child and her or his families from the first year of life through infancy and early childhood. They all commenced in the 2000s, although with varying starting points. The storage and secondary analyses of each was approved by the Institutional Review Board of the University of Texas at Austin.

The Longitudinal Study of Australian Children (LSAC) has followed a representative sample of Australian children since 2004. Data are collected every two years for two cohorts, with 5,000 in each cohort (Soloff, Lawrence, & Johnstone, 2005). This study used data from the first three waves of the birth cohort, which began when children were 0–1 in 2003–2004 and had an 86% response rate by wave 3. Growing Up in Ireland: National Longitudinal Study of Children (GUI) has a representative infant cohort that began in 2008 when the focal child was 9 months old (n = 11,100), with follow-up waves at child age 3, 5, 7/8, and 9 years (Williams et al., 2010). The Millennium Cohort Study (MCS) has followed 18,818 young people and their families in the United Kingdom (Plewis, 2007). The focal children were born in 2000–2001 and have been followed at two-year intervals. The wave 2 sample included families and children from the wave 1 sample and 1,389 families from the wave 1 sample that were not issued a survey. The eligible sample for wave 3 consisted of families who participated in wave 1 plus the new wave 2 families, for a total of approximately 15,000. Finally, the Early Childhood Longitudinal Study-Birth Cohort (ECLS-B) is a representative study of approximately 10,000 children born in the United States and followed up at ages 9 months, 2 years, and 4 years and kindergarten entry with 9% attrition by the last wave (Andreassen, Fletcher, & Park, 2007).

The analytical samples for each dataset included families who participated across all waves, which raised the issues of attrition bias. To gauge the nature of such bias, logistic regressions predicted whether a family exited the sample before the final wave by baseline maternal education, age, employment, and partnership status (Table 1). These models were estimated first with the sample that would later be used for analyses of enrollment in early childhood education and then with the sample that would later be used for analyses of engagement in home reading (see below). The most consistent findings across the four datasets were the underrepresentation of single women and a trend toward older mothers. The table also includes the percentage in pooled data, which illustrated the over-contribution of cases from the United Kingdom dataset and under-contribution of cases from the Australian dataset. To address such biases, longitudinal sampling weights provided by each dataset adjusted the respective samples for differential attrition across waves and other sources of sampling bias.

Table 1.

Results of logistic regressions predicting attrition from the sample by maternal characteristics for each country-level dataset

	Odds Ratio (SE)
	Australia		Ireland		U.K.		U.S.
	1 (n = 4,003)	2 (n = 4,143)	1 (n = 8,655)	2 (n = 6,565)	1 (n = 12,074)	2 (n = 12,766)	1 (n = 8,850)	2 (n = 8,850)
Educational Attainment (college)	1.32* (0.14)	1.11 (0.12)	0.10* (0.11)	1.46*** (0.10)	1.24 (0.40)	1.08 (0.12)	1.25** (0.09)	1.24** (0.09)
Age at Birth of Child	1.04** (0.01)	1.04** (0.01)	1.28* (0.13)	1.00 (0.01)	1.11*** (0.03)	1.07*** (0.01)	1.01** (0.00)	1.01 (0.00)**
Employment (ref: not employed)
Part-time	1.85*** (0.22)	1.66*** (0.21)	1.00 (0.00)	1.14 (0.09)	1.28 (0.49)	2.53*** (0.33)	1.23** (0.09)	1.23** (0.09)
Full-time	1.49** (0.20)	1.08 (0.14)	6.96 (7.54)	1.15 (0.09)	0.86 (0.44)	2.05*** (0.39)	1.06 (0.06)	1.06 (0.06)
Partnership Status (ref: married)
Cohabitating	0.77 (0.10)	0.69** (0.10)	1.00 (0.00)	0.45*** (0.04)	1.00 (0.00)	1.00 (0.00)	0.76*** (0.06)	0.75*** (0.06)
Single	0.44*** (0.07)	0.39*** (0.06)	1.00 (0.00)	0.05*** (0.01)	1.09 (0.34)	2.12*** (0.29)	0.71*** (0.05)	0.72*** (0.05)
Constant	1.61 (0.52)	2.03* (0.67)	0.58 (1.70)	4.66*** (1.13)	7.94* (6.73)	1.56 (0.49)	3.52*** (0.50)	3.52*** (0.50)
Percentage in pooled data	0.12	0.13	0.26	0.20	0.36	0.40	0.26	0.27

Note:

^*p < .05;

^**p < .01;

^***p < .001. Imputed data could not be used for this attrition analysis, so the sample sizes differ from what is reported for subsequent analyses. For each country-level dataset, the outcome was a binary variable differentiating exits from the sample by the final wave vs. remaining in the sample, with Model 1 based on the weighted sample used for the enrollment in early childhood education analyses and Model 2 based on the weighted sample used for the engagement in home reading analyses. The NCES guidelines for use of the U.S. data require rounding, so the n for the two ECLS-B analytical samples appears to be equivalent even though they differed slightly.

SOURCE (U.S. Data): U.S. Department of Education, National Center for Education Statistics, Early Childhood Longitudinal Study, Birth Cohort (ECLS-B), 2001–02, 2003–04, 2005–06, 2006–07.

Beyond attrition, item-level missingness can also bias results. Consequently, multiple imputation chained equations (MICE) and sample weighting were conducted separately for each of the four datasets so that information from one dataset would not influence the procedures for another (Johnson & Young, 2011). The subsequent pooling of the four imputed and weighted datasets into a single dataset was done with Stata’s mi append command. We also employed the multiple imputation, then deletion method (MID), which included the dependent variables in the MICE but excluded them from the final analytical models. Exclusion of imputed values for dependent variables from modeling removed unnecessary noise from model, but their inclusion during the multiple imputation stage maintained unbiased estimates (von Hippel, 2007). MID led to separate analytical samples for the set of models predicting enrollment in early childhood education and the set of models predicting engagement in home reading activities.

The pooled, imputed, and weighted analytical samples included 36,400 children (17,900 girls, 18,500 boys) for the first set of models predicting enrollment in early childhood education and 35,800 children (17,600 girls, 18,200 boys) for the second set of models predicting engagement in home reading activities (note that the n for any sample including U.S. cases was rounded to the nearest 50 to follow NCES guidelines).The corresponding numbers for the Australian LSAC were 4,225 children (2,062 girls, 2,163 boys) and 4,278 children (2,090 girls, 2,188 boys), respectively. For the Irish GUI, they were 8,705 children (4,297 girls, 4,408 boys) and 6,725 children (3,298 girls, 3,427 boys), respectively. For the British MCS, they were 14,613 children (7,155 girls, 7,458 boys) and 14,237 children (6,964 girls, 7,273 boys), respectively. For ECLS-B, they were 8,950 children (4,400 girls, 4,550 boys, rounded to the nearest 50, per NCES guidelines) and 10,550 children (5,150 girls, 5,400 boys, rounded to the nearest 50, per NCES guidelines), respectively.

Measures

Table 2 presents the measures by country-level dataset (note: for ease of presentation, descriptive statistics for all variables were drawn from the analytical sample imputed and weighted for subsequent analyses of enrollment in early childhood education, with the exception being the statistics for variable for engagement in home-reading activities). We also calculated the percentage of cases with imputed values for all variables. It was at 20% or below for all variables in the United Kingdom and United States datasets and between 20–22% in the Irish and Australian datasets.

Table 2.

Descriptive statistics for study variables for each country-level dataset

	M (SD) or %
	Overall (n = 36,400)	Australia (n = 4,225)	Ireland (n = 8,705)	U.K. (n = 14,613)	U.S (n = 8,950)
Outcomes
Early education enrollment	85.61	99.72a	71.11b	98.75a	70.74b
Home reading*	1.48 (0.60)	1.48b (0.60)	1.28c (0.62)	1.46b (0.60)	1.69a (0.52)
Maternal Characteristics
Educational attain. (college)	36.15	50.10a	37.66b	37.46b	27.79c
Changes in educational attain.	11.82	08.97a	09.78a	17.66a	05.41a
Employment status
Not employed	42.31	34.16a	46.32a	43.82a	39.62a
Part-time	32.79	41.81a	25.28b	42.37a	19.23c
Full-time	24.90	24.03bc	28.40b	13.81c	40.95a
Partnership status
Married	68.55	76.91a	74.13ab	62.36b	69.81ab
Cohabiting	16.51	11.64ab	13.87ab	22.54a	11.05b
Single	14.94	11.45b	12.00b	15.07a	19.14a
Age at childbirth	30.05 (6.36)	31.34c (5.27)	32.02a (5.15)	29.93d (5.89)	27.70b (6.36)
Child Characteristics
Race/ethnicity (white)	78.62	97.94a	94.30b	84.87b	43.56c
Gender (boys)	50.93	51.20a	50.64a	51.04a	50.94a
Age (in months)	58.83 (5.51)	57.67b (2.87)		62.63a (2.97)	52.95c (4.19)

Note:

^*The means and standard deviations for home reading came from the analytical sample imputed and weighted for models predicting this variable, but the descriptive statistics for all other variables came from the analytical sample imputed and weighted for models predicting children’s enrollment in early childhood education. For Ireland, child age was measured in years and is, therefore, not included in the table. Means or percentages with the same subscript did not significantly differ across countries (p <. 05), as determined by t-test for means and Chi-square tests for percentages, and means or percentages with a different subscript significantly differed from each other. The sample size for any sample including U.S. cases was rounded to the nearest 50 to follow NCES guidelines.

SOURCE (U.S. Data): U.S. Department of Education, National Center for Education Statistics, Early Childhood Longitudinal Study, Birth Cohort (ECLS-B), 2001–02, 2003–04, 2005–06, 2006–07.

Enrollment in early childhood education.

In each country, mothers reported at the final preschool wave whether children were enrolled in any type of early education program (i.e., preschool), which we used to create a binary measure (1 = enrolled). Each country has its own unique set of policies regarding early childhood care and education. Since 2008, the Australian government has offered preschool access to all children the year before they enter elementary school (education.gov.au/universal-access-early-childhood-education). Since 2010, Irish children age 3 and older have access to free preschool, and the age limit was recently lowered to age 2 (www.education.ie/en/The-Education-System/Early-Childhood). Since 2010, 3 and 4 year olds in England have had access to free childcare, with free access to part-time childcare beginning at age 3 in Wales and Scotland (assets.publishing.service.uk.gov). The United States does not have a universal plan at the federal level, although many states do, and there are subsidies available for financing early childhood education at the federal and state level for certain populations. Thus, three of the four countries have a universal approach to early childhood education. The policies in question followed the start of the dataset in each country, however, and none of the countries had yet achieved universal enrollment.

Cognitive stimulation at home.

Maternal reports of home reading activities (i.e., how often they read at home to or with their children) measured cognitive stimulation at home. To harmonize the construct across datasets, we collapsed categories in the Ireland and United Kingdom datasets, which had slightly more categories than in the Australia and United States datasets (see Table 3). The Irish measure originally had five categories, and the Australian measure originally had four categories that were collapsed into three to fit with United States and United Kingdom measures. The harmonized measure resulted into three ordered categories: never/rarely (comprised of categories from never to once/twice a month), occasionally (categories from 1–2 days a week, several times a week, or 3–6 times a week(, and everyday (categories from either everyday or 6–7 days a week.

Table 3.

Categorization of responses to home reading activities questions for each country-level dataset

	Categories for Final Harmonized Measures
	Never/Rarely	Occasionally	Everyday
Australia	Never	1–2 Days a Week, 3–5 Days a Week	6–7 Days
Ireland	Never, Hardly Ever	Occasionally, Once/Twice a Week	Everyday
U.K.	Not at All, Less Often, Once/Twice a Month	Several Times a Week	Everyday
U.S.	Not at All, Once/Twice a Month	3–6 Times a Week	Everyday

SOURCE (U.S. Data): U.S. Department of Education, National Center for Education Statistics, Early Childhood Longitudinal Study, Birth Cohort (ECLS-B), 2001–02, 2003–04, 2005–06, 2006–07.

Maternal education.

A binary measure differentiated women with a college degree or not, with that distinction operationalized in a country-specific way. This variable differentiated between graduating from college or not to reflect the emphasis on this distinction as a fundamental division in the credentialing process and to be in line with the interdisciplinary conceptualization of the college premium. A continuous measure of educational attainment was also created using years of education. As a sensitivity check, we re-estimated all models replacing the binary measure with this continuous operationalization. We retained the cleaner and more easily interpretable binary measure for the final models because the continuous operationalization obscured differences between Ireland and the United Kingdom but otherwise left the main pattern of results the same. The original measure was assessed at baseline. To account for changes in maternal education over time, we created another binary variable identifying mothers who reported completion of a college degree after baseline.

Other maternal characteristics.

For maternal employment, each dataset included maternal reports of their employment status and hours worked in the past week. To harmonize maternal employment across datasets, we created dummy variables for not employed for pay, employed part-time for pay (paid employment for 32 hours per week or less), and employed full-time for pay (paid employment exceeding 32 hours per week). We measured maternal employment at the same wave as the outcome variables in each dataset. For maternal partnership status, we drew on maternal responses to slightly different questions about household composition and marital status to create three dummy variables (single, cohabiting, married) at the same wave as the outcome variables in each dataset. For maternal age at childbirth, each dataset included mothers’ reports of their age when the focal child was born.

Child characteristics.

In each dataset, mothers reported their children’s race/ethnicity (1 = White, 0 = non-White, for consistency given the varying levels of diversity across the four populations), gender (1 = boys, 0 = girls), and child age in months (note that this variable was not available in the Irish GUI).

Plan of Analyses

With the first pooled, imputed, and weighted analytical sample, logistic regression models predicted enrollment in early childhood education by maternal education, child characteristics, and country in to order to test the focal linked lives phenomenon, followed by the inclusion of the other maternal characteristics in order to assess the robustness of the linked lives phenomenon to a number of potential observable confounds. The final set of models included interactions of maternal education with the country variables in order to test the sociohistorical contextualization of the linked lives phenomenon. We then conducted all possible pair-wise comparisons of the maternal education variables and the maternal education x country interactions by rotating the comparison category. Comparing the magnitude of the maternal education coefficients across countries to the rank ordering of the countries on gross domestic product (relative advantage) and policy investments in families and children (contingent protection) then assessed the strength of the two hypotheses. Ordinary least square linear regression models predicting home reading activities with the second pooled, imputed, and weighted analytical sample followed the same sequence.

Logistic models return an odds ratio; a significant odds ratio above 1 indicate that the likelihood of the 1 category of the dependent variable is more likely to occur as the values or categories of the independent variable increase (with values less than one indicating a decreasing likelihood). The closer an odds ratio is to approaching 1, the less likely any “effect” (in either direction) has occurred. The absolute value of the difference between the odds ratio and 1, multiplied by 100, represents the percent change of the “effect”. For example, an odds ratio of 1.8 for variable X represents an 80% increase ([1–1.8] × 100) in the odds of variable Y with every one-unit change in variable X. As another example, an odds ratio of .75 for variable X represents a 25% decrease ([1-.75] × 100) in the odds of variable Y with every one-unit change in variable X. Linear models return more straightforwardly interpretable unstandardized coefficients, where positive values indicate increasing values on the dependent variable as the independent variable increases and negative values indicate decreasing values on the dependent variable as the independent variable increases.

These primary modeling steps were followed by the calculation of two post-hoc statistics to gauge the sensitivity of results to unobserved confounds (the Impact Threshold for Confounding Variables) and to the varying composition and historical timing of each data collection (the Replacement Statistic). Each is described below.

Results

Refer back to Table 2 for the descriptive statistics by country. The subscripts in the table designate significant and non-significant differences in variables across the four countries. Following country-level patterns for women’s educational attainment overall, mothers were most likely to have graduated from college in Australia (nearly one-half) and least likely to have done so in the United States (just over one-quarter), with mothers in the United Kingdom and Ireland in between (around one-third). Reflecting the push for universal enrollment in Australia and the United Kingdom, the overwhelming majority of young children in each of those two datasets was enrolled in early childhood education programs, but the enrollment rates of children in Ireland and the United States lagged behind. Engagement in home reading activities was more consistent across countries, with the highest levels in the United States, followed by Australia and the United Kingdom, and then Ireland.

Linking Mothers’ Educational Attainment to Children’s Educational Opportunities

Beginning with the focal linked lives phenomenon (i.e., associations between mothers’ educational attainment and children’s educational opportunities), Table 4 presents the results of the logistic regressions predicting enrollment in early childhood education, with odds ratios above 1 indicating direct associations and odds ratios below 1 indicating inverse associations. Table 5 presents the corresponding results of the linear regressions predicting engagement in home reading activities, with coefficients with positive signs indicating direct associations and coefficients with negative signs indicating inverse associations. Recall that, for both tables, we conducted additional analyses rotating the reference categories of the country dummy variables to test all possible pair-wise comparisons among countries. We highlight all significant differences among countries from this full set of comparisons in the text below, even though only the comparisons and significant differences between the United States and the other countries are presented in the tables (see the supplemental online appendix for full results).

Table 4.

Results of logistic regressions predicting children’s enrollment in early childhood education

	Odds Ratio (SE)
	[95% Confidence Interval]
	Model 1	Model 2	Model 3
Maternal Characteristics
Educational attain. (ref: college)	1.37*** (0.04)	1.38*** (0.04)	3.12*** (0.07)
	[0.24, 0.39]	[0.25, 0.40]	[1.01, 1.27]
Changes in educational attain	1.26*** (0.06)	1.25*** (0.06)	1.24** (0.06)
	[0.11, 0.35]	[0.10, 0.35]	[0.09, 0.34]
Partnership status (ref: married)
Cohabiting		1.25*** (0.05)	1.23*** (0.05)
		[0.12, 0.32]	[0.10, 0.31]
Single		1.54*** (0.05)	1.54*** (0.05)
		[0.33, 0.53]	[0.33, 0.53]
Employment status (ref: not employed)
Part-time		1.22*** (0.04)	1.28*** (0.05)
		[0.12, 0.29]	[0.17, 0.34]
Full-time		1.22*** (0.04)	1.32*** (0.05)
		[0.13, 0.28]	[0.20, 0.36]
Age at childbirth	1.12*** (0.02)	1.17*** (0.02)	1.11*** (.02)
	[−0.00, −0.00]	[0.11, 0.35]	[0.06, 0.15]
Child Characteristics
Race/ethnicity (white)	1.00 (0.04)	0.97 (0.04)	0.90* (0.04)
	[−0.15, 0.02]	[−0.12, 0.05]	[−0.19, −0.27]
Gender (male)	0.84*** (0.03)	0.84*** (0.03)	0.84*** (0.03)
	[−0.24, −0.11]	[−0.24, −0.12]	[−0.24, −0.11]
Country (ref: U.S.)
Australia	140.48*** (0.29)	136.65*** (0.29)	132.54*** (0.34)
	[4.37, 5.52]	[4.35, 5.49]	[4.23, 5.54]
Ireland	1.38 (0.04)	0.99 (0.04)	1.59*** (0.05)
	[−0.07, 0.09]	[−0.09, 0.08]	[0.37, 0.56]
U.K.	9.60*** (0.08)	27.94*** (0.09)	39.66*** (0.16)
	[3.20, 3.53]	[3.16, 3.50]	[3.36, 4.00]
Maternal Education x Country (ref: college x U.S.)
College x Australia			1.23 (0.84)
			[−1.44, 1.85]
College x Ireland			0.25*** (0.08)
			[−1.53, −1.21]
College x U.K.			0.39*** (0.19)
			[−1.33, −0.58]
Constant	0.53* (0.32)	0.22*** (0.34)	0.39** (0.34)
	[−1.26, −0.00]	[−2.16, −0.84]	[−0.1.61, −0.27]

Note:

^*p < .05;

^**p < .01;

^***p < .001. n = 36,400 (the sample size for any sample including U.S. cases was rounded to the nearest 50 to follow NCES guidelines); Significant odds ratios above 1 indicated that the outcome became more likely to occur as the independent variable increased values or categories, and those below 1 indicated that the outcome became less likely to occur as the independent variable increased values or categories.

SOURCE (U.S. Data): U.S. Department of Education, National Center for Education Statistics, Early Childhood Longitudinal Study, Birth Cohort (ECLS-B), 2001–02, 2003–04, 2005–06, 2006–07.

Table 5.

Results of linear regressions predicting children’s engagement in home reading activities

	Unstandardized Coefficients (SE)
	[95% Confidence Interval]
	Model 1	Model 2	Model 3
Maternal Characteristics
Educational attain. (college)	0.17*** (0.01)	0.16*** (0.01)	0.20*** (0.01)
	[0.15, 0.18]	[0.14, 0.17]	[0.18, 0.23]
Changes in educational attain	0.03** (0.01)	0.03*** (0.01)	0.03*** (0.01)
	[0.01, 0.04]	[0.01, 0.05]	[0.01, 0.05]
Partnership status (married)
Cohabiting		−0.08*** (0.01)	−0.08*** (0.01)
		[−0.10, −0.07]	[−0.10, −0.06]
Single		−0.09*** (0.01)	−0.09*** (0.01)
		[−0.11, −0.07]	[−0.11, −0.07]
Employment status (not employed)
Part-time		0.02** (0.01)	0.03* (0.01)
		[0.01, 0.04]	[0.01, 0.04]
Full-time		−0.01 (0.01)	−0.01 (0.01)
		[−0.02, 0.01]	[−0.02, 0.01]
Age at childbirth	0.02*** (0.01)	0.00 (0.00)	0.00 (0.00)
	[0.01, 0.02]	[−0.00, 0.01]	[−0.01, 0.01]
Child Characteristics
Race/ethnicity (white)	0.18*** (0.04)	0.17*** (0.01)	0.18*** (0.01)
	[0.16, 0.20]	[0.16, 0.19]	[0.16, 0.19]
Gender (male)	−0.03*** (0.01)	−0.03*** (0.01)	−0.03*** (0.01)
	[−0.04, −0.02]	[−0.04, −0.02]	[−0.04, −0.02]
Country (U.S.)
Australia	−0.35*** (0.01)	−0.35*** (0. 01)	−0.37*** (0.02)
	[−0.37, −0.33]	[−0.37, −0.33]	[−0.40, −0.34]
Ireland	−0.55*** (0.01)	−0.55*** (0. 01)	−0.54*** (0.01)
	[0.57, −0.53]	[−0.57, −0.53]	[−0.57, −0.52]
U.K.	−0.40*** (0.01)	−0.39*** (0. 01)	−0.32*** (0.01)
	[−0.42, −0.38]	[−0.41, −0.37]	[−0.35, −0.30]
Maternal Education x Country (ref: college x U.S.)
College x Australia			0.02 b (0.02)
			[−0.03, 0.07]
College x Ireland			−0.04*** (0.08)
			[−0.08, −0.00]
College x U.K.			−0.11*** (0.02)
			[−0.11, −0.07]
Constant	1.26*** (0.06)	1.49*** (0.06)	1.51*** (0.06)
	[1.13, 1.38]	[1.36, 1.62]	[1.38, 1.63]

Note:

^*p < .05;

^**p < .01;

^***p < .001. n = 35,800 (the sample size for any sample including U.S. cases was rounded to the nearest 50 to follow NCES guidelines). Significant coefficients with a positive sign indicated that the outcome became more likely to occur as the independent variable increased values or categories, and those with a negative sign indicated that the outcome became less likely to occur as the independent variable increased values or categories.

SOURCE (U.S. Data): U.S. Department of Education, National Center for Education Statistics, Early Childhood Longitudinal Study, Birth Cohort (ECLS-B), 2001–02, 2003–04, 2005–06, 2006–07.

First, across all four countries, college-educated mothers were just under 40% ([1.37–1] × 100) more likely to enroll their children in early childhood education programs than other mothers, net of the control variables (Model 1 in Table 4). Adding the other maternal characteristics as covariates did not attenuate the initial odds ratio for maternal education when comparing Model 1 to Model 2 in Table 4 (note: early childhood education enrollment was also significantly associated with mothers being employed and unmarried). Models 1 and 2 also revealed a rank ordering of country “effects” that changed slightly with the control of other maternal characteristics. Reflecting country-level variation in early childhood education policies, children in Australia were the most likely to be enrolled in an early childhood education program, followed by children in United Kingdom, with the difference between these two countries statistically significant (as tested by rotating the reference category to capture all possible pair-wise comparisons). Children in Ireland and the United States were the least likely to be enrolled in an early childhood education program, and the difference between these two countries was not statistically significant. The addition of maternal characteristics did not change the rank ordering.

Second, across all four countries, college-educated mothers were significantly more likely to read with or to their children at home. The effect size for this coefficient (β = .17, p < .001) equaled 28% of a standard deviation of this outcome variable (Model 1 in Table 5). Again, comparing Models 1 and 2 in Table 5 revealed that adding the other maternal characteristics as covariates did not attenuate the initial coefficient for maternal education (note: home reading was also more frequent in families with married mothers). Reflecting country-level variation in engaging in home reading activities, children in the United States were more likely to be read to at home, followed by children in Australia and the United Kingdom, with children in Ireland the least likely to be read to at home. Rotating the reference category revealed that all differences among the four countries in the coefficient were statistically significant. The rank ordering did not change when maternal characteristics were added to the model.

Variation in Intergenerational Linked Lives Across Countries

Turning to the country-level contextualization of the focal linked lives phenomenon, Model 3 in Tables 4 and 5 presents the results of the final modeling step, which tested maternal education x country interactions to examine the moderation of the associations between mothers’ educational attainment and children’s educational opportunities by country context.

The association between mothers’ college education and children’s enrollment in early childhood education differed between the United States on one hand and the United Kingdom and Ireland on the other according to the two significant maternal education x country interactions in Model 3 in Table 4. Specifically, it was weaker in these two countries than in the United States. Figure 1 depicts the predicted probabilities of enrollment across these three countries based on the main effects and interaction terms. The association between maternal college education and children’s enrollment in early childhood education was non-significant in the United Kingdom and Ireland (and even slightly negative in the latter). It was significant and positive only in the United States, with college-educated mothers at least three times as likely to enroll their children in an early childhood education program. There was also a small but non-significant positive association in Australia (not shown in Figure 1), which could not be statistically differentiated from the the other countries, perhaps because of the large standard error likely related to near universal enrollment.

Figure 1.

Predicted probabilities of children’s enrollment in early childhood education, by maternal education and country

Note: n = 36,400 (the sample size for any sample including U.S. cases was rounded to the nearest 50 to follow NCES guidelines); The difference between college-educated mothers and other mothers was significant (p < .001) in the United States. The interaction between maternal education and United States significantly differed (p < .05) from the interaction between maternal education and Ireland and from the interaction between maternal education and United Kingdom. Australia was omitted because its interaction did not significantly differ from any other country. SOURCE (U.S. Data): U.S. Department of Education, National Center for Education Statistics, Early Childhood Longitudinal Study, Birth Cohort (ECLS-B), 2001–02, 2003–04, 2005–06, 2006–07.

Similarly, the association between maternal education and engagement in home reading activities differed by country (see the two significant maternal education x country interactions in Model 3 in Table 5). Basically, the advantage of maternal education was significantly stronger in the United States and Australia (which did not significantly differ from each other) than in Ireland, which had a significantly stronger association than the United Kingdom.

See the supplemental online appendix for results of significance testing for contrasts among all possible maternal education by country combinations for each domain of educational opportunity. Among other patterns, they show the higher level of home reading activities in the United States relative to other countries regardless of educational level.

Sensitivity Analyses

Investigating links between parents’ circumstances and children’s developmental ecologies across countries posed a number of analytical problems that undermine causal inference. Of course, analyses with observational data cannot prove causality, but they can be strengthened in a number of ways to raise confidence in the inference of causality. In this spirit, we calculated two sets of post-hoc robustness statistics.

First, maternal education is a highly endogenous variable, meaning that its apparent associations with a child outcome are likely confounded with other factors, some of which (e.g., genetics, local conditions) are difficult to observe and control (Magnuson & Duncan, 2002). Thus, other factors might explain why maternal education and children’s educational opportunities covary, so a lack of exploration of such factors reduces causal inference in maternal education results. The Impact Threshold for Confounding Variables test gauges confidence in causal inference by estimating how much some unknown confound would have to be correlated with predictor and outcome variables to reduce the observed association to non-significance (see Frank et al., 2008 for a description of this statistical procedure).

Using the new Stata procedure, we calculated the Impact Threshold for Confounding Variables for the maternal education coefficient in the pooled model for each outcome as well as in each country-level dataset when analyses indicated maternal education significantly predicted the outcome in that country. The only country with a significant association between maternal education and enrollment in early childhood education was the United States. The Impact Threshold for Confounding Variables for this coefficient was .36, which means that some unobserved confound would have to be correlated with maternal education at .60 or higher and with enrollment at .60 or higher (i.e., .60 × .60 = .36) for its inclusion to wash out the observed association between the two. Achieving that magnitude of correlation with two separate variables is difficult, although not impossible. Three countries yielded significant associations between maternal education and home reading activities, with the highest Impact Threshold for Confounding Variables values for the United Kingdom (.20, indicating the two correlations both be at magnitude of .45 or higher) and lowest for Ireland (.06, .24 × .24).

Second, pooling multiple datasets into one dataset is difficult because of differences in measurement and sampling. By simplifying variable responses, we roughly harmonized measurement across the datasets, and we could at least control for demographic differences across country. One difference that could not be controlled, however, was the different starting years across studies. Although they all began in the 2000s, small differences in starting years—especially the relatively late start of the Irish GUI—meant that the cohorts encountered the global economic recession at different ages, which could have differentially affected their mothers and children. To gauge how much of a problem this difference posed, we calculated a second post-hoc statistic that estimated how many cases in the sample would have to be replaced by cases with other characteristics to invalidate inferences based on model coefficients. In other words, how many of the ECLS-B children would have to be replaced by children born later in the decade—such as as 2008 in the the Irish GUI sample—to substantially alter the initially observed results (see Frank et al., 2013 for a description of this statistical procedure)?

Again, we calculated the Replacement Statistic for the Model 2 maternal education coefficient for each outcome when it was significant for a specific country. For enrollment in early childhood education in the United States, the Replacement Statistic was .86, indicating that fully 86% of the cases in the United States sample would have to be replaced with cases with no correlation between maternal education and this outcome (such as cases later in the decade) to eliminate the observed association between the two. Again, Ireland had the lowest Replacement Statistic value. It was .65 for Ireland, while the others exceeded .85.

The observed “effects” of maternal education on early educational opportunities, therefore, appeared to be most robust to threats to causal inference in the United States and least robust in Ireland.

Discussion

In the modern global economy, the lifelong returns to higher education have risen to historic levels, and these returns cross a number of domains beyond earnings (Hout, 2012). One domain is educational opportunity for children. In short, past research has shown that educational inequality in the mother generation can be passed down to the child generation in part because educational attainment enables mothers to marshal more resources to give their children educational advantages over other children. This study identified possible mechanisms of such an intergenerational transmission of inequality in the significant associations of having a college-educated mother with two early educational advantages that forecast future educational attainment. Past research also suggests that the linkage between maternal education and children’s educational opportunities—a clear illustration of the linked lives principle of life course theory—extends across a wide array of countries (see Grépin & Bharadwaj, 2015; Chevalier et al., 2013). Less clear is the extent to which different country contexts weaken or strengthen this general association, the possibility of which is suggested by the sociohistorical contextualization principle of life course theory. This study revealed such significant country-level variation in the intergenerational transmission of inequality, revealing that the associations of having a college-educated mother with two early educational advantages tended to be consistently stronger in the United States than in three other affluent English speaking countries.

To summarize the results in greater detail, this study analyzed pooled nationally representative data on children and families from four countries to document that the associations of maternal education with children’s enrollment in early childhood education and their engagement in a key home-based activity of cognitive stimulation (reading at home) varied across countries. For both types of educational opportunities, the associations were generally strong and robust in the United States, a country with the lowest overall rate of college education among women, the least generous policy context, and the absence of universal early childhood education policies. College-educated mothers were far more likely to enroll their children in early childhood education and to engage in reading activities at home with their children than other mothers in the United States. In Australia, college-educated mothers were no more likely than other mothers to enroll their children in early childhood education, their positive and significant association between maternal education and reading at home was similar to the United States. In Ireland, college educated mothers were not more likely than other mothers to enroll their children in early childhood education but were moderately more likely to read with them—these Irish patterns, however, were less robust to threats to causal inference than in the other countries. In the United Kingdom, the focal associations between maternal education and children’s educational opportunities were more inconsistent and/or weaker than in other countries. Thus, the United States was more of an outlier with the other countries in terms of the intergenerational linkage of education, but the differences among these countries varied depending on the type of educational opportunity being considered. How does this pattern of results line up with the two hypotheses that structured our conceptual model?

An overview of the results suggests partial but not complete support for both the relative advantage and contingent hypotheses. In short, support was fairly strong for the predictions about more pronounced inequality in the United States, but differentiating the relative standing of the other three countries was more challenging. Recall that the relative advantage hypothesis—which emphasized the importance of how much mothers resembled or differed from the population norm for women in their societies—predicted that the association between maternal education and children’s educational opportunities would be strongest in the United States (where women have lower rates of educational attainment) and weakest in Australia (where they have higher rates). This hypothesis was supported by evidence that the focal associations were indeed robust and relatively strong in the United States, but that support was only partial because the associations were not weakest in Australia or consistently intermediate in Ireland or the United Kingdom. Recall also that the contingent protection hypothesis—which emphasized the importance of social policies as buffers against family-based disadvantages—predicted that the association between maternal education and children’s educational opportunities would be stronger in the United States (which proportionally spends the least amount on policy supports for families and children) than in the other three more generous countries. Again, this association was robust and relatively strong in the United States, but there was some qualifier in support related to where Australia stood relative to the United States.

To be more specific, Australia should have had weaker associations between maternal education and children’s educational opportunities than the United States because of its higher rate of maternal educational (relative advantage) and greater proportional policy expenditures (contingent protection). Yet, it was hard to differentiate the two countries on the links between maternal education and children’s educational opportunities, even though the two countries clearly differed in both maternal education and children’s educational opportunities. At the same time, the United Kingdom should have had weaker associations between maternal education and children’s educational opportunities than Ireland because the former had a higher rate of maternal educational (relative advantage) and greater proportional policy expenditures (contingent protection) than the latter. Again, though, the two differed on overall levels of the key variables (i.e., the main effects of country on outcome) more consistently than on the links between them (i.e., the country x maternal education interactions predicting the outcome).

The partiality of the support for the two hypotheses could have resulted from methodology and operationalization. We measured maternal education at the country level by the proportion of college-educated mothers in the population in order to straightforwardly compare any individual woman to the norms in her country, but the magnitude of the educational gap (i.e., variance) and where mothers stood in this gap is also a country-level dimension of inequality for assessing relative advantage. Similarly, assessing contingent protection through per capita spending on family and child policies captures explicit governmental priorities about families and children but ignores the more indirect ways that other less targeted government programs (e.g., tax breaks) can act as buffers against family disadvantages. The absence of fathers as a focal point of analyses—necessitated by the inconsistent data collection on fathers across the national datasets—could also obscure differences among the four countries that speak to the value of the relative advantage and contingent protection hypotheses.

Beyond methods and measurement, the pattern of findings could reflect that the processes at the heart of the relative advantage and contingent protection hypotheses intersect in ways that are difficult to disentangle. Perhaps the more consistent findings for the United States arose because the processes underlying the two hypotheses work in the same direction and magnify each other while the more inconsistent findings for the other countries emerged because the two sets of processes worked at cross-purposes, elevating a country’s standing in one way and diluting it in another. That possibility would be better evaluated by modeling that explicitly contrasted multiple mechanisms apiece for each hypothesis, pooled data from a much larger sample of countries, and conceptualized and operationalized how maternal education works through and against parental preferences for early childhood education and home enrichment (i.e., does maternal education matter more because it influences preferences or because it protects those preferences from being disrupted by external challenges and setbacks?).

Overall, the clearest takeaway from the results concerns the United States. Specifically, a combination of strong relative advantage and weak contingent protection in the United States means that the stratification of children’s educational opportunities by mothers’ own educational attainment is comparatively powerful in this country. Such a pattern of a stronger stratifying role of maternal education in children’s lives reflects that the United States has the highest level of economic inequality of the four countries. Yet, the strength of stratifying role of maternal education within the United States is difficult to assess relative to the payoff of concrete policy levers. For example, a federal data base of programmatic educational interventions in the U.S. reveals that few experimental evaluations report an effect size greater than the correlations reported here (https://ies.ed.gov/ncee/wwc/). In that light, the findings of this study echo conclusions by developmental scientists that direct interventions in children’s lives are often less effective than boosting the socioeconomic circumstances of their families (Kalil & Ryan, 2020). As a socioeconomic benchmark, the effect size reported here for maternal education represents about half the effect size of a $1000 child care subsidy reported in one study in the United States, although that effect size was only for low-income families (Ros Pilarz, 2018).

Whether considering parental education, income, family structure, or other socioeconomic factors, better contextualizing the intergenerational transmission of inequality from mother to child within and across countries is a valuable part of developmental science. This science is driven by within-country investigations, but the only way to truly know the meaning of what is uncovered within a country is to compare that country to another. Most clearly, the complementary nature of within- and between-country investigation was illustrated here by the evidence that there is an intergenerational transmission of educational advantages in the United States (within-country investigation) that makes it stand out from even otherwise similar countries (between-country investigation).