![Sample our Education journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5932/TOC-Subject-Sample-2021-Education.jpg"})
Abstract
Few studies have investigated the causal spillover effects of compulsory education on children's siblings. Using a regression discontinuity method, I find that Hong Kong's 1971 free compulsory primary education policy reduced the dropout probability for the eldest siblings of full policy beneficiaries, especially for children in low-income families. Having younger brothers who were full policy beneficiaries did not affect the elder child's educational attainment. Having younger sisters who were full policy beneficiaries increased the eldest child's educational attainment, especially for the eldest sister. The results shed light on how sibling sex composition might affect intra-household resource allocation of human capital investment among children.
Acknowledgements
I thank the editor and two anonymous referees for very helpful comments on an earlier draft of the paper. I also thank Dora Choi, Yue-ping Chung, Weili Ding, Victor Lavy, Steve Lehrer, Patrick McEwan, Sandra McNally, Suet-ling Pong, David Post, Olmo Silva, Ngai-ying Wong, and participants at CASER seminar at the Hong Kong University of Science and Technology, CEP Education Group meeting at the London School of Economics, Department of Economics seminar at Lingnan University and 2012 SWUFE International Workshop on Applied Microeconomics for their comments and suggestions. I am grateful to Pekky Chiu, Horace Lee, and Ivan Wong for their help on using the Hong Kong Census data. Irina Chong and Elaine Lam provided excellent research assistance.
Disclosure statement
No potential conflict of interest was reported by the author.
Supplemental data
Supplemental data for this article can be accessed at doi:10.1080/09645292.2015.1017447
Notes
1. For example, scholars have evaluated the policy impact on child labor reduction in developing countries (Dessy Citation2000), dropout prevention, and school enrollment (Oreopoulos Citation2007; Goldin and Katz Citation2011), long-term impact on the labor market (Harmon and Walker Citation1995; Meghir and Palme Citation2001; Oreopoulos Citation2006; Pischke and Wachter Citation2008; Devereux and Hart Citation2010; Grenet Citation2013), and other social outcomes, such as mortality rates (Meghir, Palme and Simeonova Citation2012; Clark and Royer Citation2013), teenage marriages and births (Black, Devereux and Salvanes Citation2004; Kirdar, Tayfur and Koç Citation2011), fertility (Leon Citation2004), and crime (Meghir, Palme, and Schnabel Citation2012).
2. Please see detailed discussions in Parish and Willis (Citation1993) and Kaestner (Citation1997).
3. Garg and Morduch (Citation1996, Citation1998) refer to this change in parents’ distribution of resources as an inequality-aversion story; specifically, inequality-averse parents will invest more in children with higher returns (such as boys in pro-male societies), ‘but they will also put additional resources into children with lower returns in order to maintain a degree of fairness' (Garg and Morduch Citation1996, 4).
4. Son preference is very common in Asian countries, including India and China. Besides educational resources, sons have an advantage in receiving family resources, and parents want to provide more health inputs to sons in India (Jayachandran and Kuziemko Citation2011).
5. Government schools can be further divided into English-language schools and Chinese-language schools.
6. Post (Citation1993) asserts that the official declaration of free compulsory primary education seemed to be guided by paternalistic concerns, with no mechanisms for mass participation in government.
7. In 1961, there were 423,000 spots in government, aided, and private primary schools, which was equivalent to 86.5% of 6- to 11-year olds. By 1971, there were 862,000 spots, representing 136.6% of 6- to 11-year olds. There was a significant decrease in the number of children not attending school from 1961 to 1971, which may be attributed to the lowering of primary school fees in Hong Kong and the provision of sufficient places for all children of primary-school age. The 1961 and 1971 Hong Kong Census data show that there has been a great reduction in the number of children not at school from 1961 to 1971. For example, in the spring of 1961, 34.4% of 6-year olds were not in school, but by 1971 the percentage had decreased to 8.8. More than 10% of primary-school-age children (i.e. 6–12 years old) were not in school in 1961, but in 1971, the figure declined to around 5–6%. In total, about 90% of the primary-school-age children were enrolled in school (either primary or lower secondary school).
8. Other difficulties include children being too old to be enrolled in primary school, not being ready for school, and children not being able to attend school due to physical illness (Overseas Chinese Daily News of Hong Kong 1970).
9. Discussions can also be found in the Hong Kong Yearbook from the 1970s and 1980s.
10. The 1971 Hong Kong Census data show that, among children aged 10–16 years who were not in school, 76% of the boys and 75% of the girls were working.
11. See Online Appendix 2 for more details on the description of fees.
12. In 1971, there were 759 government and aided schools, including five English-language government schools and 691 private schools (Hong Kong Census and Statistics Department Citation1982).
13. Enrollment in government and aided primary schools increased by 10.6% during the same period. This implies that families showed increased demand for public schools after the policy instead of opting for more private schooling.
14. It was not until a government white paper in 1981 that the government allowed children who reached age 6 in December of the academic year to be eligible to enter Grade 1 in primary school in September of the same academic year (Hong Kong Government Citation1981).
15. The earliest Hong Kong Census data available are of 1981. However, because I look at the long-term educational attainment of the eldest siblings within the family, the 1986 data provide more suitable observations than the 1981 data.
16. I checked the 1981 data for the number of siblings of the full policy beneficiaries of the 1971 reform, and the mean of the sibship size is about 0.5 bigger than the mean of the sibship size in the 1986 data. It is possible that the eldest child in some families might have moved out of the household in 1986. However, the data-set does not include information that would allow me to identify whether the eldest child living in the household is the first child in the family (e.g. age of first birth for women). Nonetheless, I can gauge the potential sample bias by looking at the percentage of children of a particular age who did not live with their parents in 1986. For example, for the eldest siblings of the eldest cohort of policy full beneficiaries (i.e. the 27-year olds), 20% were living in a separate household other than with their parents. For the younger cohorts, the percentage of living in a separate household is smaller. Despite the data limitation, my results still show some interesting spillover effects on elder siblings, even if not all of them are actually the eldest siblings.
17. Please refer to Ou (Citation2013) for the direct impact of the 1971 compulsory-education law on children's educational attainment. Because below I present local linear estimates using the optimal bandwidths obtained by the Imbens and Kalyanaraman (IK) approach and because the IK approach is based on a linear probability model, I estimate all outcome variables through linear models. Nonetheless, results are similar using probit or logit regressions.
18. Another straightforward estimation would be to use the 1971 policy as an instrument to estimate younger siblings’ schooling and then look at the eldest siblings’ educational outcomes. However, the data-set provides information only on younger siblings’ educational attainment up to the survey time, at which point many were still in school; therefore, I could not identify the total years of schooling of the younger siblings for the sample of the eldest siblings that I estimate in the sample.
19. Manipulation is not of concern in this study because the policy was announced in July 1971 and made effective in September 1971 for those who were aged six years entering Grade 1 in primary school. That is, these students were born six years before 1971, so their mothers were not timing their childbirth patterns for policy reasons, and of course the children cannot control their age eligibility. A histogram of month-of-birth does not show substantial differences in the density for individuals born before or after the birthday eligibility cutoff either (see online Appendix Figure A1).
20. This variable is included in the test because the compulsory-education policy might have induced families to have more children after 1971, when schooling became free. However, please note that the demographic controls for all regressions include dummies for female, urban residence, low-income household, having a low-educated mother, having a low-educated father, living in a single-parent family, and living with grandparents.
21. I show in and that including younger siblings further away from the policy cutoff (i.e. a larger window at the right side of the policy cut-off) decreases the likelihood of being able to observe the eldest siblings’ educational attainment. Furthermore, it is not desirable to include the eldest siblings who were still living in the same household in the window further away from the cutoff. The unobserved characteristics of these eldest siblings could be a potential threat to the causal RD estimates (for example, if there is a relationship between education and age of marriage and if more educated siblings are likely to postpone marriage, this selection may lead to overestimates, since my sample will contain more educated siblings than the actual population). Thus, it is not preferred in my study to include data too far from the cutoff even though the advantage of doing so is to allow a flexible function form in month-of-birth.
22. To better determine whether educational resource re-allocation among siblings is plausible, we can look at the relative costs of primary and secondary education prior to the reform. Information and discussion are provided in online Appendix 2. However, the analysis is limited because the Census data do not survey the school enrollment information and fees charged by schools, which varied substantially even in the same school type.
23. This could be because the quadratic form fits better than the linear form for the pattern in the eldest brothers’ local average probability of leaving school by age 12 or 15 in our data. Furthermore, because the actual data show substantial curvature, larger bandwidths will not improve efficiency but generate bias (Lee and Lemieux Citation2010). So it is not surprising to see some differences among the local linear estimates under different bandwidths.
24. Powell and Steelman (Citation1989, 145) also claim that ‘if a daughter enrolls despite this general tendency to favor sons, then parents may support her ambition [of attending college].'
25. Results in might capture the different policy impacts on households with different numbers of children. The eldest brother or sister on either side of the policy cutoff should have the same number of younger siblings who are affected by the reform, with the only difference being the sibling whose date of birth is used as the forcing variable in the RD estimates. Put another way, if families re-allocate the freed-up resources from the younger siblings to their children who are not policy beneficiaries, the eldest sibling on the right-hand side of the RD cutoff will have more resources because that sibling has one more younger sibling who is a policy beneficiary; and the eldest sibling on the left-hand side of the RD cutoff will have fewer resources because that sibling needs to share those freed-up resources with the sibling who just missed eligibility to receive the policy benefit. Assuming that the freed-up resources will be split equally among the eldest child and the sibling born just before the policy, then the greater the total resources, the more the eldest child will receive. That is to say, in families with more younger siblings who are policy beneficiaries, the eldest sibling will receive more freed-up resources and is likely to obtain more education. I also estimate Equation (1) based on different ranges for the number of younger siblings for low-income households. Results are given in Table A3. This table reveals the same trend: the spillover effect is greater for families having more policy beneficiaries.
26. The magnitude is considerably large given that the policy's actual impact may be substantially smaller than the intention-to-treat if a non-negligible share of children belonging to the pre-reform cohorts were able to receive all six years of primary education because they started school late or repeated a grade and/or if a non-negligible fraction of children among the post-reform cohorts did not in fact receive all six years of free primary education. Unfortunately, the Census data do not contain information on school-entry age (Ou Citation2013, 32).
27. Except that there were some effects on the eldest sisters’ probability of obtaining education beyond high school.
28. The school day was split into morning and afternoon sessions, and class size was increased to accommodate the increasing number of students (Jul 1971, Education in Hong Kong. Hong Kong Government Information Service Publication).