Journal of Family Psychology 2007, Vol. 21, No. 3, 510 –519
Copyright 2007 by the American Psychological Association 0893-3200/07/$12.00 DOI: 10.1037/0893-3200.21.3.510
Families, Economies, Cultures, and Science Achievement in 41 Countries: Country-, School-, and Student-Level Analyses Ming Ming Chiu Chinese University of Hong Kong This study examines the links between students’ families and science achievement across many countries. Science tests and questionnaire responses of 107,834 fifteen-year-olds in 41 countries were analyzed with multilevel analyses. Students had higher science scores if they were native born, lived with two parents, lived without grandparents, lived with fewer siblings (especially older ones), had more educational resources, had more family involvement, lived in wealthier countries, or lived in countries with more equal distributions of household income. In wealthier countries, family involvement, blended families, and number of siblings showed stronger links to science scores. Science achievement was more strongly linked to family socioeconomic status (SES) and educational resources in more egalitarian cultures and to single parents, family SES, resident grandparents, and birth order in more individualistic cultures. Hence, family constructs were linked to academic achievement in all 41 countries, and the links were stronger in more economically and culturally developed countries. Keywords: Comparative education, Cross cultural studies, hierarchical linear modeling, family structure, family involvement Supplemental materials: http://dx.doi.org/10.1037/0893-3200.21.3.510.supp
Past studies have linked family constructs and academic achievement. Children tend to have higher achievement when they have more parents, higher family socioeconomic status (SES), more educational resources, or more family involvement (Entwisle & Alexander, 1995). Whether these links differ across countries (richer vs. poorer, degree of household income inequality, various cultural values) have remained open questions. Hence, this study examines these issues in the context of science achievement. After summarizing the links between family constructs and academic achievement, several hypotheses for country differences are discussed.
see Figure 1, middle column; Amato, 2001; Entwisle & Alexander, 1995; Horvat, Weininger, & Lareau, 2003). Families with more parents typically have higher SES, more educational resources at home (e.g., books), and more parent time to spend with their children (greater parent communication and involvement). In contrast, separated parents have fewer resources and face more challenges in caring for their children, so the children might receive less attention (e.g. from stepparents in blended families). Meanwhile, immigrant parents, especially those that speak a foreign language, likely have less social and cultural capital to share with their children (e.g., cultural possessions and communication), which limits children’s learning opportunities and often their academic achievement (Coleman, 1994; Portes & MacLeod, 1996). On the other hand, additional family members who primarily compete for family resources (such as grandparents and siblings) reduce the available resources for a child, often yielding fewer learning opportunities and lower academic achievement (resource dilution hypothesis; see Figure 1, middle column; Downey, 2001). Some children benefit from affluent grandparents’ resources (Bengston, 2001) and show higher achievement (DeLeire & Kalil, 2002). However, children who live with poor or ill grandparents compete with them for limited family resources (PatilloMcCoy, Kalil, & Payne, 2003). Furthermore, children with more resident siblings often have fewer resources at home and achieve less than those with fewer resident siblings (Downey, 2001). Older siblings tend to receive more family
Family Constructs Family members can give children extra resources or compete for them. On the one hand, additional family members who provide extra resources, especially parents, provide more learning opportunities on which children can capitalize to achieve more (resource provider hypothesis;
This research was partially funded by a Spencer Foundation grant. I appreciate Yik Ting Choi’s research assistance and Bonnie Wing Yin Chow and Sung Wook Joh’s suggestions on an earlier version of this article. Correspondence concerning this article should be addressed to Ming Ming Chiu, Department of Educational Psychology, 314 Ho Tim Building, The Chinese University of Hong Kong, Shatin, N.T., Hong Kong. E-mail:
[email protected]
510
FAMILY, ECONOMICS, AND CULTURE Country
Student
Resources GDP per capita (+)
Family
Science
Inequality: GINI (–)
Resource Provider
Achievement
Living with no parents (–) Cultural Values
Single parent (–)
Egalitarian (+)
Blended family (–)
Individualism (–)
Family SES (+) 1st generation immigrant (–) 2nd generation immigrant (–) Speak foreign language at home (–) Number of books at home (+) Cultural possessions (+) Cultural communication (+)
Resource Dilution Grandparent (–) Number of siblings (–) Older siblings (–)
Figure 1. Model of country and family effects on students’ science achievement. GDP ⫽ gross domestic product; GINI ⫽ equality of household incomes.
resources than do younger siblings because they compete with younger siblings for family resources only after the latter’s births (e.g., parental time, energy, and engagement; Powell & Steelman, 1993). Hence, parents are often resource providers, whereas resident grandparents and siblings often compete for family resources, thereby diluting their impact.
Country Resources and Cultural Values Country properties such as average income, income inequality, and cultural values might also be linked to student achievement indirectly via family constructs. Students in countries with higher real gross domestic product (GDP) per capita generally show higher academic achievement (see Figure 1, left column; Baker, Goesling, & Letendre, 2002; Heyneman & Loxley, 1983). Wealthier countries can raise student achievement directly through education spending (e.g., books, teacher training, better curricula) or indirectly through higher nutritional standards or better health care (UNICEF, 2001). Apart from the amount of resources, overall student achievement might be linked to the distribution of those resources, as a result of diminishing marginal returns or homophily. For many resources, additional identical items are worth less than the initial items (diminishing marginal returns). Hence, an extra $100 in books likely improves a poor student’s reading score more than a rich student’s reading score. In countries with greater equality of household incomes (lower GINI), poorer students
511
have more resources and benefit more from them, resulting in higher education outcomes overall (Chiu & Khoo, 2005). As people prefer to interact with others of similar socioeconomic status (homophily; McPherson, Smith-Lovin, & Cook 2001), greater income equality within a country might also encourage greater cooperation among students, resulting in higher overall academic performance. In wealthier countries, the links between family variables and student achievement might be weaker, similar, or stronger, thereby yielding three hypotheses: public resources substitution, social reproduction, and complementary intangibles. Wealthier countries might provide more public resources (e.g., school teachers, library books) that can substitute for family resources (Blossfeld & Shavit, 1993; Schiller, Khmelkov & Wang, 2002), thereby weakening the link between family resources and student achievement (public resources substitution hypothesis). However, highSES families might use their superior resources to create other advantages for their children (e.g., hiring admissions coaches for their children’s college applications), resulting in similar links between family variables and achievement across poor and wealthy countries (social reproduction hypothesis; Baker et al, 2002; Blossfeld & Shavit, 1993). Lastly, the widespread availability of physical resources (such as public libraries and museums) might increase the value of less tangible resources (such as parent time and attention). For example, a child benefits from reading an extra book, but that benefit can be substantially magnified by discussing the book with a parent. Thus, parental involvement and other intangible family resources might be more strongly linked to academic outcomes in richer countries (complementary intangibles hypothesis). Countries differ not only by income and income inequality but also by how they address basic societal issues, which result in different cultural values (Schwartz & Ros, 1995). Consider two basic societal issues: (a) inducing responsible individual behavior and (b) prioritizing the interests of individuals versus groups. To encourage responsible behavior, a society might assign hierarchical roles and teach its citizens to obey authority (hierarchical). Alternatively, citizens might learn to view, value, and act toward one another as equals on the basis of their common humanity (egalitarian). Likewise, a society may favor group interests (collective) or individual interests (individualist). These cultural values might be linked to students’ behaviors and subsequent learning. Nations with more egalitarian cultures might promote close relationships among students and reduce status differences, both of which are linked to student achievement (see Figure 1, left column; e.g., Ames & Ames, 1985; E. G. Cohen, 1994). As egalitarian societies have less rigid societal status, individuals have greater economic and social mobility through their individual skills and talents (Hofstede, 2003). Thus, students might view academic achievement as more linked to future success, which might yield higher achievement scores. Furthermore, students might be more motivated to make better use of their resources in egalitarian societies, yielding stronger links between student achievement and family constructs (egalitarian motivation hypothesis).
512
CHIU
Meanwhile, students in collective societies might be more inclined to cooperate, support, and learn from one another, thereby enhancing their academic achievement (see Figure 1, left column). In more collective societies, people tend to rely more on their extended family members who often live nearby, and thus students are more likely to benefit from them (Georgas et al., 2001; Hofstede, 2003). As extended family resources dilute the effects of immediate family resources in collective societies, the link between immediate family resources and academic achievement might be weaker (collective dilution hypothesis). In short, this study examined the links between family constructs and science achievement and how these links might differ across economic and cultural contexts (see Figure 1). Four sets of hypotheses (resource providers, resource dilution, country resources, and cultural values) were tested (see Table 1). This study addressed these issues through analyses of science test scores and questionnaire responses of 107,834 fifteen-year-old students in 41 countries.
Method Data The Organization for Economic Cooperation and Development (OECD) consists of countries sharing the principles of a market economy, pluralist democracy, and respect for human rights. OECD initiated a study, the Program for International Student Assessment (PISA), to help participating countries evaluate and improve their school systems. In this study, OECD assessed 107,834 fifteen-year-olds’ science skills and asked them to fill out questionnaires. Written informed consent was obtained from each student or student’s parent. International experts from 28 participating OECD countries defined science literacy, built assessment frameworks, created test items, forward-translated and backward-
translated these items, and pilot-tested these items to ensure their validity and reliability (for details, including reliability and validity tests, see OECD, 2002). Fifteen non-OECD countries joined the study within two years. Canada and Japan had country-wide missing data, so they were not included in this analysis. The following countries participated: Albania, Argentina, Australia, Austria, Belgium, Bulgaria, Brazil, Chile, Czech Republic, Germany, Denmark, Finland, France, Greece, Holland, Hong Kong, Hungary, Iceland, Indonesia, Ireland, Israel, Italy, Korea, Latvia, Liechtenstein, Luxembourg, FYR Macedonia, Mexico, Norway, New Zealand, Peru, Poland, Portugal, Romania, Russian Federation, Spain, Sweden, Switzerland, Thailand, United Kingdom, and the United States. These experts defined science literacy as the ability to understand, use, and reflect on science concepts to achieve one’s goals, develop one’s knowledge and potential, and participate effectively in society. Example assessment items are available at the PISA Website (http://pisa-sq.acer .edu.au/). Each participating student completed a 2-hr assessment booklet and a 30- to 40-minute questionnaire (for questionnaire items, see OECD, 2002). Students responded to questions on family structure, SES, immigrant status, possessions at home, and communication at home. Three additional data sets were also used: economic data from OECD (2000), and cultural values data from Hofstede (2003) and Inglehart, Basanez, Diez-Medrano, Halman, and Luijkx (2004). See Table 2 for variable descriptions and summary statistics.
Methodological Design Investigating these research questions across many countries and schools requires representative sampling of 15year-olds, precise tests and questionnaire items for data collection, and suitable statistical models of the relationships within the data. To represent a broad spectrum of
Table 1 Seven Sets of Hypotheses Regarding the Links Between Family Constructs and Science Achievement and How They Might Differ Across Economic and Cultural Contexts Hypothesis Resource provider Resource dilution
Richer countries More equal countries Cultural values Egalitarian Collective dilution Country resources Public resources substitution Social reproduction Complementary intangibles Cultural values Egalitarian motivation Collective dilution
Description
Effect on student achievement
More parents More competitors Resident grandparents Resident siblings Older resident siblings More public resources Homophily or diminishing marginal returns
Higher Lower Lower Lower Higher Higher
More equal status Prioritize group interests
Higher Higher
Richer Rich or poor Richer
Weaker family-achievement links Similar family-achievement links Stronger family-achievement links
More equal status Prioritize group interests
Stronger family-achievement links Weaker family-achievement links
Variable
M
51.75
Individualism
SD
0.98
⫺0.01 1.02
1.58
4.17
0.07
1.45 0.98
1.94 1.80
1.04
23.35
0.01 21.61
8.42
0.61
108
Description
“How many books are there in your home?” Choices ⫽ None (2%); 1–10 (13%), 11–50 (23%); 51–100 (21%); 101–250 (18%); 251–500 (13%); over 500 (10%). Analysis with dummy variables showed nearly linear results, so a single ordered variable was used to ease readability. Index of availability of 3 items: “classical literature,” “books of poetry,” and “works of art.” Choices were: yes or no. Reliability ⫽ .59 (OECD, 2002b); minimum ⫽ ⫺1.65; maximum ⫽ 1.16. Index of frequency of 3 items: “discuss political or social issues,” “discuss books, films or TV programs,” and “listen to music.” Choices ⫽ never or hardly ever, a few times a year, about once a month, several times a month, or several times a week. Reliability ⫽ .55; minimum ⫽ ⫺2.20; maximum ⫽ 2.72.
1 ⫽ first-generation immigrant, 4%; the student, mother, and father were all born outside the country 1 ⫽ second generation immigrant, 4%; same as first generation, except the student was born in the country 1 ⫽ foreign language spoken at home, 11%. 89% of students speak the official language at home. 1 ⫽ living with at least one grandparent, 19%. 81% of students do not live with a grandparent. Min ⫽ 0, Max ⫽ 12. Values ⫽ 1: youngest child, 35%; 2: middle child, 25%; 3: oldest child, 32%; 4: only child, 8%. 1 ⫽ girl, 51%; boys ⫽ 49%
1 ⫽ single parent (14%) 1 ⫽ one parent and one step-parent (6%) 1 ⫽ student does not live any parents (4%) Standardized congeneric factor score of 3 items: Mother’s years of schooling, father’s years of schooling, and highest parent job status (Ganzeboom, De Graaf, & Treiman 1992). Reliability ⫽ 0.74, minimum ⫽ ⫺3.59; maximum ⫽ 2.37.
Minimum ⫽ 7.63, maximum ⫽ 9.90 (Heston, Summers, & Aten, 2002). We also tested linear GDP per capita, but it did not fit the data as well. GINI is a measure of inequality. Its scores range from 0 (perfect equality; everyone has exactly the same income) to 100 (perfect inequality, where one person has all the income, and everyone else’s income is zero). Minimum ⫽ 24.4; maximum ⫽ 59.1 (World Bank, 2004). % of a country’s GDP spent on public schools. Minimum ⫽ 0.01. Max ⫽ 0.08. (OECD, 2000). Minimum ⫽ 11; maximum ⫽ 93. Inverted power-distance scale from Hofstede (2003). Cultural values index created from factors analyses of responses to 4 questionnaire items in each country. Minimum ⫽ 14; maximum ⫽ 91 (Hofstede, 2003). See egalitarian scale regarding index construction; created from responses to 4 questionnaire items in each country.
1 ⫽ Some remedial courses in school (18%). These variables are rough proxies of past achievement. 1 ⫽ Many remedial courses in school (5%).
Science test scores created from 68 items. Reliability ⫽ .90, minimum ⫽ 28, maximum ⫽ 845. The student science scores estimated by the Rasch models were calibrated to a mean of 500 and a standard deviation of 100 (based on data from only the OECD countries; OECD, 2002). (Many non-OECD countries scored below the OECD mean.)
Note. All data are from the Program for International Student Assessment (PISA), unless otherwise specified. The Organization for Economic Cooperation and Development (OECD, 2002) created Warm (1989) indices and tested them for reliability. PISA indices were initially standardized (M ⫽ 0, SD ⫽ 1) for OECD countries. Non-OECD countries were added later, so negative means indicate lower values for non-OECD countries. GDP ⫽ gross domestic product; GINI ⫽ greater inequality of household incomes.
Cultural communication
Cultural possessions
(Baseline ⫽ native-born student and nativeborn parents, 92%) First-generation immigrant Second-generation immigrant Foreign language Grandparent Number of siblings Birth order Girl Parent investment and involvement variables at the student level Number of books at home
⫺0.05
0.05 51.08
% GDP school spending Egalitarian
Family variables at the student level (Baseline ⫽ two-birth parents, 76%) Single parent Blended family No parents SES
35.17
9.10
472
GDP GINI (inequality)
Remedial course variables at the student level (Baseline ⫽ no remedial courses, 77%) Some remediation Much remediation Country-level variables Log GDP per capita
Science score
Table 2 Summary Statistics of Variables
FAMILY, ECONOMICS, AND CULTURE 513
514
CHIU
schools, OECD (2002) used stratified sampling via neighborhood SES and student intake to select about 150 schools. They then sampled about 35 students from each of these schools. Each country sampled at least 4,500 students. OECD (2002) then weighted the participant test scores and variables accordingly to represent the schools and the 15year-old student populations of each country. For sampling details, see OECD (2002). These students were then given precise tests and questionnaires. Traditional tests that cover a lot of science content are often very long, resulting in student fatigue and learning effects. To reduce these effects and to maximize evaluative precision, OECD (2002) used a balanced incomplete block (BIB) test. In a BIB test, each student answers only a subset of questions from the overall test (Lord, 1980). Because each pair of subtests shared overlapping questions, OECD (2002) analyzed the test scores by fitting a graded response Rasch model to the BIB data. The Rasch model estimated the difficulty of each item and the achievement score of each student on the basis of the subtest responses (adjusting for the difficulty of each test item and after calibrating all test items, Lord, 1980). In addition to multiple-choice questions, the test included open-ended questions, so partial credit was captured by the graded response model (Samejima, 1969). Like the test, the questionnaire should also maximize precision. Using one question with a limited number of possible responses (e.g., yes/no, Likert scale) to measure an underlying theoretical construct often results in large measurement error. To minimize this error, OECD (2002) included multiple measures for each construct and computed a single value from these measures with a Rasch model (Warm, 1989). This method is more precise than the traditional method of summing the response values of multiple measures (Rowe & Rowe, 1997). To model the relationships in this nested data (students within schools within countries), multilevel analyses were used (Goldstein, 1995; also called hierarchical linear modeling [HLM], Bryk & Raudenbush, 1992). In comparison with ordinary least squares regressions, multilevel models estimate more precisely the standard errors of regression coefficients in nested data. Because students did not respond to all questionnaire items, there were missing data (5%) that could reduce estimation efficiency, complicate data analyses, and bias results (Rubin, 1996). Markov Chain Monte Carlo multiple imputation addressed these problems more effectively than other approaches (such as deletion, mean substitution, and simple imputation; Rubin, 1996).
nificant variance at all three levels, a three-level model was needed. Students’ science scores were modeled with sequential sets of variables (also known as hierarchical sets: J. Cohen, West, Aiken, & Cohen, 2003) to estimate the variance explained by each set (see mathematics equations in the Appendix). First, proxies were used to capture past student achievement. Country variables could affect family variables, which in turn might affect achievement. Hence, the variables were entered in this order: past achievement variables, country variables, and family variables. The past achievement variables were the following: “enrolled in some remedial courses in school” and “enrolled in many remedial courses in school.” Self-reported attendance in remedial courses served as a coarse proxy for past achievement to enable an analysis of learning (rather than only achievement). Next, country-level variables were added: log GDP per capita, GDP GINI, percentage of a country’s GDP spent on public schools, individualism, and egalitarianism. (Note that log GDP per capita fits the data better than does linear GDP per capita, as in other cross-country studies, such as Baker et al., 2002). Next, family structure variables and gender were entered: living with no parents, living with a single parent, blended families, SES, first-generation immigrant, second-generation immigrant, foreign language spoken at home, resident grandparent, number of siblings, birth order, and gender. Lastly, family process variables were entered: number of books at home, cultural possessions, and cultural communication. The significance of each set of predictors was tested with a nested hypothesis test (chi-square log likelihood; J. Cohen et al., 2003). Next, interaction effects among pairs of significant variables were modeled. If the predictors’ effects differed across countries, these differences were modeled with the above country variables. An alpha level of .05 was used for all statistical tests. Doing many tests on one set of data increases the likelihood of a spurious correlation. This likelihood is reduced by adjusting the alpha level based on the number of predictors through Hochberg’s (1988) variation on Holm’s (1979) method.
Results This sample included a variety of countries. They ranged from poor, unequal, hierarchical, collective nations (e.g., Indonesia) to wealthy, equal, egalitarian, individualistic ones (e.g., Switzerland). See Table 2 for summary statistics (see supplemental data online, Table A1 for correlation– variance– covariance matrices).
Analysis The students’ science scores were modeled with multilevel analyses via MLn software (Rasbash & Woodhouse, 1995). Multi-level models separate unexplained error into student (Level 1), school (Level 2), and country (Level 3) components, thereby removing the correlation among error terms resulting from students nested within schools within countries. As the variance components model showed sig-
Explanatory Model Country and family variables accounted for some of the differences in students’ science scores (see Table 3). Students’ science scores varied substantially at the student level (52% of the variance), at the school level (26%), and at the country level (22%). None of the effects discussed below differed significantly across schools within a country. All
FAMILY, ECONOMICS, AND CULTURE
515
Table 3 Summaries of Three Regression Models Predicting Students’ Science Scores, With Unstandardized Coefficients (and Standard Errors) Regressions predicting science Model 1 Predictor
b
Some remedial courses in school Much remedial courses in school Log GDP per capita GDP GINI Individualism Egalitarian Living with no parents Single parent Single parent ⫻ Individualism Blended families Blended families ⫻ Log GDP per capita SES SES ⫻ Individualism SES ⫻ Egalitarian First generation immigrant Second generation immigrant Foreign language spoken at home Grandparent Grandparent ⫻ Individualism Number of siblings Number of siblings ⫻ Log GDP per capita Birth order Birth order ⫻ Individualism Girl Number of books at home Number of books at home ⫻ Egalitarian Cultural possessions Cultural communication Cultural communication ⫻ Log GDP per capita Explained variance Country level School level Student level Total explained variance
⫺31.01 ⫺33.82*** 52.97** ⫺1.74* 0.11 ⫺0.18 ***
.60 .06 .02 .16
Model 2 (SE)
b
(0.68) (1.20) (16.52) (0.67) (0.42) (0.35)
Model 3 (SE)
b
***
⫺28.54 ⫺30.33*** 50.55* ⫺1.37* 0.01 ⫺0.11 ⫺17.42*** ⫺5.31*** ⫺0.13*** ⫺5.19*** ⫺4.83* 15.93*** 0.12*** 0.16*** ⫺21.25*** ⫺7.67*** ⫺17.05*** ⫺13.93*** ⫺0.32*** ⫺4.17*** ⫺1.70*** 4.21*** 0.04** ⫺5.46***
.60 .29 .06 .24
(SE)
(0.67) (1.18) (16.36) (0.66) (0.42) (0.35) (1.29) (0.71) (0.03) (1.05) (1.80) (0.30) ⫺0.02 ⫺0.02 (1.45) (1.45) (0.98) (0.69) (0.03) (0.19) (0.27)
***
⫺28.50 ⫺31.29*** 48.25* ⫺1.18 0.03 ⫺0.14 ⫺15.06*** ⫺2.78*** ⫺0.10** ⫺2.85* ⫺3.48* 9.25*** 0.08*** 0.09*** ⫺16.83*** ⫺6.12*** ⫺15.64*** ⫺13.26*** ⫺0.30*** ⫺4.65*** ⫺1.99***
(0.65) (1.15) (16.30) (0.66) (0.42) (0.35) (1.27) (0.69) (0.03) (1.03) (1.76) (0.31) (0.02) (0.02) (1.42) (1.42) (0.96) (0.68) (0.03) (0.19) (0.27)
(0.25) (0.01) (0.51)
4.89*** 0.03* ⫺7.86*** 9.62*** 0.09*** 3.78*** 6.22*** 4.00***
(0.25) (0.01) (0.50) (0.19) (0.01) (0.29) (0.25) (0.38) .60 .40 .10 .29
Note. Model 1 includes remedial courses as a proxy for past achievement and country-level variables. Model 2 also includes family structure variables, family socioeconomic status, significant interaction effects with country variables, and gender. Model 3 also includes family process variables (investment in books and cultural possessions, and family conversations about culture). GDP ⫽ gross domestic product; GINI ⫽ inequality of household incomes; SES ⫽ socioeconomic status. * p ⬍ .05. ** p ⬍ .01. *** p ⬍ .001.
results discussed below describe first entry into the regression, controlling for all previously included variables. Ancillary regressions and statistical tests are available upon request. Past achievement. Students who attended some or many remedial courses in school scored 31 or 34 points lower in science, respectively, than did other students (Table 3, Model 1). Countries’ economies and cultural values. Controlling for past achievement, science scores were linked to economic conditions but not cultural values. Students in richer countries scored higher than those in other countries. If a country’s GDP per capita exceeded that of another country by 10%, the former country’s students averaged 5 points higher in science than those in the latter country did, 5 ⫽
ln(1 ⫹ 10%) ⫻ 52.97; log variable computation, ln(1 ⫹ 10%) ⫻ b; Table 3, Model 1). Log GDP per capita accounted for 60% of the differences in science scores between countries. Students in countries with more income equality scored higher than those in countries with less equal incomes. If a country’s GINI exceeded that of another country by 10% (more unequal), the more unequal country’s students averaged 4 points lower in science compared with the latter country’s students (⫺4 ⫽ ⫺1.74 ⫻ 8.42 ⫻ [10% / 34%]; Table 3, Model 1), consistent with diminishing marginal returns or homophily. Meanwhile, cultural values were not significantly linked to science scores. Past achievement and country variables accounted for 16% of the variance in students’ science scores.
516
CHIU
Family constructs and gender. Controlling for each country’s economic conditions and cultural values, the results supported both the resource provider and the resource dilution hypotheses. Children had higher science scores if they lived with more parents, lived in higher SES families, or were native born. Children with two parents averaged 17 points higher than those living without parents, 5 points higher than those with single parents, and 5 points higher than those in blended families (Table 3, Model 2). Students averaged 5 points higher with an extra 10% increase in their family SES. First- and second-generation immigrant students averaged 21 and 8 points lower in science, respectively, compared with other students (Table 3, Model 2). Students speaking a different language at home than at school averaged 17 points lower in science than did other students (Table 3, Model 2). In contrast, children had lower science scores if they lived with their grandparents, had more siblings, or had older siblings. All of these results support the resource dilution hypothesis. Children living without grandparents averaged 14 points higher in science than those living with grandparents (Table 3, Model 2). Children with more siblings scored lower, averaging 4 points lower in science per extra sibling (Table 3, Model 2). Students who were born earlier tended to score higher (Table 3, Model 2). These results support the view that additional competition for resources is linked to lower science scores. Also, boys outscored girls by 5 points in science (Table 3, Model 2). Family structure and family SES account for an additional 8% of the variance in science scores. Controlling for family structure and family SES, the family process results also support the resource provider hypothesis. Students had higher science scores if their homes had more books, more cultural possessions, or more cultural communication (Table 3, Model 3). Students with an extra 10% of cultural possessions in their homes scored 1 point higher on average. Furthermore, students with an extra 10% of cultural communication with their parents averaged 2 points higher. All of these family results support the view that extra family resources are linked to higher science scores. Family processes account for an additional 5% of the variance in science scores. Country properties moderate family–achievement links. The results supported the social reproduction and complementary intangibles hypotheses. Most of the links between students’ science achievement and family variables showed no significant differences, supporting the social reproduction hypothesis (Table 3, Models 2 and 3). In richer countries, science scores had stronger negative links to blended families and to number of siblings (Table 3, Model 2). Furthermore, the positive link between cultural communication and science scores was also stronger in richer countries (Table 3, Model 3). These last three results support the view that intangible resources are more strongly linked to student achievement in richer countries. No family variable had a significantly weaker effect in richer countries. Hence, the results do not support the public resources substitution hypothesis (Table 3, Models 2 and 3). In addition to economic properties, a country’s cultural
values also moderated the links between family constructs and science scores. Specifically, the results supported both the egalitarian motivation and collective dilution hypotheses. In countries with stronger egalitarian values, science scores were more positively linked to family SES and number of books at home (Table 3, Models 2 and 3). These results support the view that students value academic achievement more in countries with more egalitarian values. The results of the interaction terms between individualism and family variables support the collective dilution hypothesis. In countries with stronger individualistic cultural values, students’ science scores had stronger negative links to living with a single parent or living with more siblings (Table 3, Model 2). Likewise, in more individualistic countries, science scores had stronger positive links to family SES and to resident grandparents (Table 3, Model 2). These results are consistent with the view that extended family members are more likely to share resources, thereby reducing the effects of the student’s immediate family.
Discussion Researchers have shown that many family constructs are linked to academic achievement (Amato, 2001; Entwisle & Alexander, 1995; Horvat, Weininger, & Lareau, 2003). This study extended past research by showing that country properties were linked to academic achievement and moderated the links between family constructs and academic achievement. See hypotheses in Figure 1 and Table 1.
Family Constructs This study showed that family constructs were linked to science achievement in many countries. Specifically, the results generally supported the resource provider and resource dilution hypotheses. Overall, children had higher science scores if they lived in families with two parents, had higher family SES, were native born, had more books, had more cultural possessions, or had more cultural communication at home. These results showed that children in families with more resources had more learning opportunities on which they often capitalized to attain higher academic achievement, supporting the resource provider hypothesis. Meanwhile, children had lower science scores if they lived with grandparents, had more siblings, or had older siblings. These results suggest that additional family members who competed for limited family resources reduced a child’s learning opportunities and academic achievement, supporting the resource dilution hypothesis. The empirical support for these two hypotheses shows that students in privileged families have substantial advantages over other students.
Country and School Differences Although the results generally supported the resource provider and resource dilution hypotheses, the results were not universal. Instead, the results differed across countries both economically and culturally. Students scored higher in countries with greater incomes or more equal incomes,
FAMILY, ECONOMICS, AND CULTURE
showing that total resources and equal distribution of resources were both important. Also, the size of the links between family constructs and science scores differed across countries with respect to GDP per capita and cultural values. Specifically, the results supported the social reproduction and complementary intangibles hypotheses over the public resources substitution hypothesis. The links between most family variables and science score did not differ significantly across richer and poorer countries, suggesting that high SES families used their superior resources for their children’s academic advantage despite the extra public resources in richer countries (social reproduction hypothesis). Regardless of a country’s economic condition, students in privileged families had higher science scores. Meanwhile, the links between science scores and blended families, number of siblings, and cultural communication were stronger in richer countries, supporting the view that widespread availability of public physical resources (e.g., books) increased the value of less tangible family resources for learning (complementary intangibles hypothesis). Cultural values did not directly affect students’ science scores, but they moderated the links between family constructs and science achievement. In more egalitarian cultures, science scores were more positively linked to family SES and number of books at home. This is consistent with the hypothesis that in egalitarian cultures with less rigid societal statuses, students viewed academic achievement as more strongly linked to future success, and were more motivated to better utilize family resources (SES, books; egalitarian motivation hypothesis). Also, single parents, family SES, resident grandparents, and birth order were all less negatively linked to science scores in more collectivist cultures. As people in collective societies tended to rely more on their extended family members who often live nearby, students were more likely to benefit from them, diluting the link between immediate family constructs and academic achievement (collective dilution). Collective dilution might not apply as much to students living without parents, students living in blended families, or nonresident grandparents. Many of the students living without parents lived alone or with friends, so they might not have access to extended family members. Meanwhile, extended family members might believe that blended families, with two parents, require less assistance than those with a single parent. Selection bias might also contribute to different links between resident grandparents and science achievement across cultures. If resident grandparents in more individualistic cultures are often poorer or sicker than nonresident grandparents, they require more family resources, thereby diluting family resources and lowering student achievement. If so, the results might differ for highly involved, nonresident grandparents. Within a country, the links between family constructs and student achievement did not differ much. This result is consistent with the literature showing that the shared culture of a country is more important than that of local institutions such as schools, churches, and corporations (Hofstede, Neuijen, Ohayv, & Sanders, 1990; Inglehart & Baker,
517
2000). Specifically, the result suggests that interactions between schools and families are too weak to influence family interactions that are linked to student achievement.
Trends Across Countries These cross-sectional differences across countries suggest several possible trends as countries develop over time. Higher GDP per capita was correlated with several family constructs (see Supplemental Data online, Table A1, fourth column). In richer countries, fewer children live with no parents. As more elderly people in richer countries can afford their own housing, students are less likely to live with their grandparents (United Nations, 2003). Also, parents in wealthier countries often have fewer children, so students often have fewer siblings (Schultz, 1997). Thus, students in wealthier countries likely face less resource dilution. As economies grow, families often become both wealthier and more educated. As a result, children in wealthier countries often receive more educational resources, have more learning opportunities, and achieve more. Still, some trends in wealthier countries are disturbing. For example, wealthier countries have higher divorce rates, yielding more single parents and blended families (United Nations, 2003). In wealthier countries, children in these families have lower academic achievement than those in traditional two-parent families. These children score lower in part because they have fewer intangible family resources (such as parent time and cultural communication), which are more important to science achievement in richer countries. As countries develop economically, their cultural values also change, typically becoming more egalitarian and individualistic (barring wars or natural catastrophes; Inglehart & Baker, 2000). The results suggest that as countries become more egalitarian, family constructs also become more important to science achievement, specifically, family SES and number of books at home. Likewise, as countries become more individualistic, immediate family constructs also become more important, specifically, single parents, family SES, resident grandparents, and birth order. In short, the results suggest that family constructs become more important to academic achievement as countries develop both economically and culturally, thereby raising the importance of supporting students with fewer family resources in these countries.
Limitations and Future Research This study had some limitations. First, the students sampled were not fully representative of all 15-year-olds. For example, students with very low achievement levels or very poor children might not attend school (e.g., UNICEF, 2001). Second, this correlational study does not warrant causal interpretations. Third, the reliabilities of cultural possessions and cultural communication are low, so results involving these two measures should be interpreted cautiously. Fourth, the cross-sectional data on 15-year-old students in this study does not address developmental effects. Fifth, the data did not include measures of family process
518
CHIU
quality (e.g., authoritative vs. authoritarian parenting styles) and thus cannot address how the quality of family processes was linked to science achievement. Sixth, the absence of other possibly relevant variables might result in omitted variable bias. In addition to the above issues, future research can also address the links between academic achievement and schoolmates’ family structures and processes as well as school or neighborhood contexts.
References Amato, P. R. (2001). Children of divorce in the 1990s: An update of the Amato and Keith (1991) meta-analysis. Journal of Family Psychology, 15, 355–370. Ames, C., & Ames, R. E. (1985). Research on motivation in education. San Diego, CA: Academic Press. Baker, D. P., Goesling, B., & Letendre, G. K. (2002). Socioeconomic status, school quality, and national economic development. Comparative Education Review, 46, 291–312. Bengston, V. (2001). Beyond the nuclear family: The increasing importance of multi-generational bonds. Journal of Marriage and the Family, 63, 1–19. Blossfeld, H. P., & Shavit, Y. (1993). Persisting barriers. In Y. Shavit & H. P. Blossfeld (Eds.), Persistent inequality (pp. 1–23). Boulder, CO: Westview Press. Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical linear models: Applications and data analysis. Newbury Park, CA: Sage Publications. Chiu, M. M., & Khoo, L. (2005). Effects of resources, inequality, and privilege bias on achievement. American Educational Research Journal, 42, 575– 603. Cohen, E. G. (1994). Restructuring the classroom. Review of Educational Research, 64, 1–35. Cohen, J., West, S. G., Aiken, L., & Cohen, P. (2003). Applied multiple regression/ correlation analysis for the behavioral sciences (2nd ed.). Mahwah, NJ: Erlbaum. Coleman, J. S. (1994). Family, school, and social capital. In T. Husen & T. N. Postlethwaite (Eds.), International encyclopedia of education (2nd ed., pp. 2272–2274). Oxford: Pergamon Press. DeLeire, T., & Kalil, A. (2002). Single-parent multigenerational family structure and adolescent adjustment. Demography, 39, 393– 413. Downey, D. B. (2001). Number of siblings and intellectual development. American Psychologist, 56, 497–504. Entwisle, D. R., & Alexander, K. L. (1995). A parent’s economic shadow. Journal of Marriage and the Family, 57, 399 – 409. Georgas, J., Mylonas, K., Bafiti, T., Poortinga, Y. H., Christakopoulou, S., Kagitcibasi, C., et al. (2001). Functional relationships in the nuclear and extended family: A 16-culture study. International Journal of Psychology, 36, 289 –300. Goldstein, H. (1995). Multilevel statistical models. Sydney, New South Wales, Australia: Arnold. Heyneman, S. P., & Loxley, W. A. (1983). The effect of primaryschool quality on academic performance across twenty-nine high- and low-income countries. American Journal of Sociology, 88, 1162–1194. Hochberg, Y. (1988). A sharper Bonferroni procedure for multiple tests of significance. Biometrika, 75, 800 – 802. Hofstede, G. (2003). Culture’s consequences. Thousand Oaks, CA: Sage. Hofstede, G., Neuijen, B., Ohayv, D. D., & Sanders, G. (1990).
Measuring organizational cultures. Administrative Science Quarterly, 35, 286 –316. Holm, S. (1979). A simple sequentially rejective multiple test procedure. Scandanavian Journal of Statistics, 6, 65–70. Horvat, E. M., Weininger, E. B., & Lareau, A. (2003). From social ties to social capital. American Educational Research Journal, 40, 319 –351. Inglehart, R., & Baker, W. E. (2000). Modernization, cultural change, and the persistence of traditional values. American Sociological Review, 65, 19 –51. Inglehart, R., Basanez, M., Diez-Medrano, J., Halman, L., & Luijkx, R. (2004). Human beliefs and values. Mexico City, NM: Siglo XXI. Lord, F. M. (1980). Applications of item response theory to practical testing problems. Hillsdale, NJ: Erlbaum. McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27, 415– 444. Organization for Economic Cooperation and Development [OECD]. (2000). Education at a glance: OECD Indicators. Paris: OECD. Organization for Economic Cooperation and Development [OECD]. (2002). Manual for the PISA 2000 database. Paris: OECD. Patillo-McCoy, M., Kalil, A., & Payne, M. (2003). Intergenerational assets and the Black/White test score gap. In D. Conley & K. Albright (Eds.), After the bell. New York: Routledge. Portes, A., & MacLeod, D. (1996). Educational progress of children of immigrants. Sociology of Education, 69, 255–275. Powell, B., & Steelman, L. C. (1993). The educational benefits of being spaced out: Sibship density and educational progress. American Sociological Review, 58, 367–381. Rasbash, J., & Woodhouse, G. (1995). MLn command reference. London: University of London, Institute of Education. Rowe, K. J., & Rowe, K. S. (1997). Norms for parental ratings on Conners’ abbreviated parent–teacher questionnaire: Implications for the design of behavioral rating inventories and analyses of data derived from them. Journal of Abnormal Child Psychology, 25, 425– 451. Rubin, D. B. (1996). Multiple imputation after 18 years. Journal of the American Statistical Association, 91, 473– 489. Samejima, F. (1969). Estimation of latent ability using a response patter of graded scores. Psychometrika Monograph Supplement, 34(4, Pt. 2). Schiller, K. S., Khmelkov, V. T., & Wang, X. Q. (2002). Economic development and the effects of family characteristics on mathematics achievement. Journal of Marriage and the Family, 64, 730 –742. Schultz, T. P. (1997). Demand for children in low-income countries. In M. R. Rosenzweig & O. Stark (Eds.), Handbook of population and family economics (pp. 349 – 430). Amsterdam: Elsevier Science. Schwartz, S. H., & Ros, M. (1995). Values in the West. World Psychology, 1, 91–122. UNICEF. (2001). The state of the world’s children 2001. Geneva, Switzerland: United Nations Publications. United Nations. (2003). Demographic yearbook 2003. New York: United Nations. Warm, T. A. (1989). Weighted likelihood estimation of ability in item response theory. Psychometrika, 54, 427– 450.
FAMILY, ECONOMICS, AND CULTURE
519
Appendix Regression Specification Equations A variance components model tests whether the variances are significant at each level. Yijk ⫽ 000 ⫹ eijk ⫹ f0jk ⫹ g00k .
(1)
Yijk is the students’ science scores for student i in school j in country k. 000 is the grand mean intercept. Error terms (residuals) at the student, school, and country levels are eijk, f0jk, and g00k, respectively. Proxies were used for past student achievement, namely a vector of some remedial courses in school and many remedial courses in school (V). Yijk ⫽ 000 ⫹ eijk ⫹ f0jk ⫹ g00k ⫹ vjk Vijk .
(2)
A vector of w variables was entered at the country level: log GDP per capita, GDP GINI, percentage of a country’s GDP spent on public schools, individualism, and egalitarian (W). Yijk ⫽ 000 ⫹ eijk ⫹ f0jk ⫹ g00k ⫹ vjk Vijk ⫹ 00w W00k .
(3)
Then, interaction effects among pairs of significant variables in W were tested. Next, x family structure variables were added at the student level: no parents, living with a single parent, blended families, SES, first-generation immigrant, second-generation immigrant, foreign language spo-
ken at home, resident grandparent, number of siblings, birth order, and gender (X). Yijk ⫽ 000 ⫹ eijk ⫹ f0jk ⫹ g00k ⫹ vjk Vijk ⫹ 00w W00k ⫹ xjk Xijk .
(4)
As with W, a nested hypothesis test was conducted for X, and interaction effects of significant variables were modeled. Next, the x student-level regression coefficients (xjk ⫽ x00 ⫹ gx0k) were tested for significant country differences (gx0k ⫽ 0?). If so, they were tested for dependence on W.  xjk ⫽  x00 ⫹ gx0k ⫹ x0w W00k .
(5)
Lastly, the above procedures for X were repeated on the family process variables at the student level: number of books at home, cultural possessions, and cultural communication (Z). Yijk ⫽ 000 ⫹ eijk ⫹ f0jk ⫹ g00k ⫹ vjk Vijk ⫹ 00w W00k ⫹ xjk Xijk ⫹ zjk Zijk .
(6)
Received May 31, 2006 Revision received February 15, 2007 Accepted February 15, 2007
䡲
