Skip to main content

The dragon cohort of Hong Kong: traditional beliefs, demographics, and education


In Hong Kong, parental preference for babies born in the Chinese Zodiac dragon year causes spikes in fertility. The larger number of “dragon babies” born in 1988 resulted in a schooling cohort which was 5% larger on average every school-year. Using an innovative identification strategy that avoids selection bias, I find that dragon cohort students increase their time spent studying math by an average of 0.26 hours per week (a 9% increase relative to the mean). These effort responses are strongest for girls and for students whose parents do not have post-secondary education. Being in the dragon cohort also results in higher math scores. These empirical findings are consistent with competitive behavior changes of dragon cohort members responding to the presence of additional students. I cannot, however, rule out other possible mechanisms, such as cooperation, peer quality, and educational investments, acting in conjunction to improve academic outcomes. This paper is the first to document the test score and effort impacts of such zodiac cohorts; its findings highlight the importance of cultural forces in determining population changes, and their potential to influence education and other societal outcomes.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7


  1. 1.

    A similar double cohort occurred in Hong Kong in 2012 due to its transition from a 3- to 4-year university program. The timing of this change does not affect this paper’s analysis, which uses data collected prior.

  2. 2.

    More generally, there is a well-established literature on the relationship between demographic changes and labor market outcomes. See Welch (1979); Stapleton and Young (1984); Berger (1985); Bloom et al. (1988); Macunovich (1999); and Korenman and Neumark (2000).

  3. 3.

    Basten and Verropoulou (2013) argue that this increase was the result of families avoiding the restrictive fertility policies in mainland China, as well as the relaxation of travel restrictions for mainland Chinese into Hong Kong around that time. The subsequent drastic decrease in births after 2012 can be attributed to a new ban on mainland-Chinese mothers giving birth in Hong Kong hospitals instituted that year.

  4. 4.

    This seasonal pattern is almost identical to the one observed by Goodkind (1993) for Taiwan’s dragon year of 1976 (see Fig. 3 of that paper).

  5. 5.

    They do not want to gamble with the astrological fate of their child.

  6. 6.

    Similar challenges are also faced by other countries (Goodkind 1991; Sim 2015).

  7. 7.

    I use the measure “Total outbound internationally mobile tertiary students studying abroad, all countries, both sexes” (UIS.OE.56.40510) in the World Bank’s Education Statistics series. This is defined as the number of “students who have crossed a national or territorial border for the purpose of education and are now enrolled outside their country of origin” for tertiary education.

  8. 8.

    Extensions based upon the work of Lazear and Rosen (1981) include Green and Stokey (1983); and Nalebuff and Stiglitz (1983).

  9. 9.

    For a survey, see Konrad (2009).

  10. 10.

    Examples include Boudreau et al. (2012), Casas-Arce and Asis Martinez-Jerez (2009), Eriksson (1999), Main et al. (1993), Garcia and Tor (2009), Harbring and Irlenbusch (2003), and Orrison et al. (1997).

  11. 11.

    In the corresponding grade 9 PISA 2006 non-dragon cohort comparison group, students born outside of these two months are similarly excluded to make the cells of the difference-in-differences analysis symmetric. I include students born in February because only birth month, and not exact birth date, is observed. I argue this is reasonable since a majority of days in February 1988 were non-dragon baby birth-dates, and the inherently different parents who are intentionally targeting their childbirths seem to “play it safe” by not aiming for a date close to the lunar new year cutoff, as Fig. 2 suggests. In Appendix A, I conduct robustness checks that exclude February-born students; the estimates from these checks are very similar.

  12. 12.

    Occupational status is based on the International Socioeconomic Index (ISEI). Education levels are categorized using definitions in the International Standard Classification of Education (ISCED). The number of books variable is a categorical variable with the following categories: 0–10 books, 11–25 books, 26–100 books, 101–200 books, 201–500 books, and more than 500 books.

  13. 13.

    The categories were no time, less than 2 h a week, 2 or more but less than 4 h a week, 4 or more but less than 6 h a week, and 6 or more hours a week.

  14. 14.

    This procedure is described in Appendix C. There, I also repeat the regression analyses in this section using an interval regression procedure which accounts for the category nature of the unmapped outcome variable; the results are robust to the procedure used.

  15. 15.

    This is because grade 10 in 2003 is the untreated non-dragon cohort comparison group. The fact that the normalization is carried out separately across PISA rounds does not matter because year fixed effects are included in the regressions. The means and standard deviations calculated in Table 2 are not exactly zero and one respectively in the grade 10 columns for two reasons. First, the normalization procedure makes use of weights provided by the OECD, while the summary statistics are unweighted. Second, the normalization is done over all observations with non-missing test scores, whereas the summary statistics are calculated over the sample of analysis, which exclude observations with missing covariates. For the normalization procedure, plausible value raw scores reported by PISA are averaged across each student observation.

  16. 16.

    The PISA recruits students born within a specific 1-year period at the time of the survey. In PISA 2003, this group encompassed students born between March 1987 and February 1988. In PISA 2006, because of survey timing, this group encompassed students born between February 1990 and January 1991. Moreover, in 2003, PISA surveyed the “average” student in June 2003, while in 2006, PISA surveyed the “average” student in May 2006. This difference in survey timing interacts with the fact that recruited students had birth-months that did not align across the two PISA years, resulting in the observed difference-in-differences in age. To address potential concerns about differences in age and birth-month profiles across the two PISA years, I conduct robustness analyses in Appendix A.

  17. 17.

    Some parameters in this and subsequent regression equations will be omitted because of perfect collinearity with a base category. However, I include them in the equations nonetheless for expositional simplicity.

  18. 18.

    Since the PISA does not follow the same schools across years, the dataset comprises two separate cross sections of students in different schools. The clustered standard errors account for correlations within each school within a particular PISA round/year. Clustering at the school-by-year level in this way is equivalent to clustering at the school level.

  19. 19.

    Previous studies have found differences between the sexes in responding to competitive pressure. See Gneezy et al. (2003); Niederle and Vesterlund (2010); Jurajda and Munich (2011); Ors et al. (2013); and Morin (2015b).

  20. 20.

    Studies examining peer effects include Hanushek et al. (2003); Zimmerman (2003); Angrist and Lang (2004); Lefgren (2004); and Duflo et al. (2011). See Epple and Romano (2011) for a survey.

  21. 21.

    Empirical studies on the subject include Krueger (1999); Angrist and Lavy (1999); Krueger and Whitmore (2001); and Chetty et al. (2011).

  22. 22.

    East Asia stands out in international comparisons in that despite larger class sizes relative to other parts of the world, countries in the region still top test score performance rankings such as PISA and TIMSS. This suggests that unlike other countries, class sizes may not matter as much in generating negative peer effects (Woessmann 2005).

  23. 23.

    For instance, Witkow and Fuligni (2010) find that students who had more friends at schools obtained higher grade point averages.

  24. 24.

    Bray and Lykins (2012) note that particularly in Asia, “families that do not wish their children to [do poorly] may invest in private supplementary tutoring to secure an edge in the competition.” (p. 23)

  25. 25.

    In particular, Do and Phung (2010) include direct measures of cohort size as covariates in regression specifications that also include auspicious birth year dummies (their Table 5 on p. 248). They find that a 1% increase in local (commune) level cohort sizes is associated with a 0.2 year increase in education attainment, although higher level aggregates of cohort size are uncorrelated.


  1. Almond D, Chee CP, Sviatschi MM, Zhong N (2015) Auspicious birth dates among Chinese in California. Econ Hum Biol 18:153–159

    Article  Google Scholar 

  2. Angrist JD, Lavy V (1999) Using maimonides’ rule to estimate the effect of class size on scholastic achievement. Q J Econ 114(2):533–575

    Article  Google Scholar 

  3. Angrist JD, Lang K (2004) Does school integration generate peer effects? Evidence from boston’s metco program. Am Econ Rev 94(5):1613–1634

    Article  Google Scholar 

  4. Athey S, Imbens GW (2006) Identification and inference in nonlinear difference-in-differences models. Econometrica 74(2):431–497

    Article  Google Scholar 

  5. Basten S, Verropoulou G (2013) Maternity migration and the increased sex ratio at birth in Hong Kong SAR. Popul Stud 67(3):323–334

    Article  Google Scholar 

  6. Berelson B (1979) Romania’s 1966 Anti-Abortion decree: the demographic experience of the first decade. Popul Stud 33(2):209–222

    Article  Google Scholar 

  7. Berger MC (1985) The effect of cohort size on earnings growth: a reexamination of the evidence. J Polit Econ 93(3):561–573

    Article  Google Scholar 

  8. Bloom David E, Freeman RB, Korenman SD (1988) The Labour-market consequences of generational crowding. European Journal of Population / Revue Européenne de Démographie 3(2):131–176

    Article  Google Scholar 

  9. Boudreau KJ, Helfat CE, Lakhani KR, Menietti M (2012) Field evidence on individual behavior and performance in rank-order tournaments,” Working Paper 13-016, Harvard Business School

  10. Bound J (2007) Cohort crowding: how resources affect collegiate attainment. J Public Econ 91(5-6):877–899

    Article  Google Scholar 

  11. Bound J, Turner S (2002) Going to war and going to college: did world war II and the G.I. Bill increase educational attainment for returning veterans? J Labor Econ 20(4):784–815

    Article  Google Scholar 

  12. Bray M, Lykins C (2012) Shadow education: private supplementary tutoring and its implications for policy makers in asia, CERC Monograph Series in Comparative and International Education and Developmentonal 9

  13. Casas-Arce P, Asis Martinez-Jerez F (2009) Relative performance compensation, contests, and dynamic incentives. Manag Sci 55(8):1306–1320

    Article  Google Scholar 

  14. Chetty R, Friedman JN, Hilger N, Saez E, Schanzenbach DW, Yagan D (2011) How does your kindergarten classroom affect your earnings? Evidence from project star. Q J Econ 126(4):1593–1660

    Article  Google Scholar 

  15. Do Q-T, Phung TD (2010) The importance of being wanted. American Economic Journal: Applied Economics 2(4):236–253

    Google Scholar 

  16. Duflo E, Dupas P, Kremer M (2011) Peer effects, teacher incentives, and the impact of tracking: evidence from a randomized evaluation in kenya. Am Econ Rev 101(5):1739–1774

    Article  Google Scholar 

  17. Epple D, Romano RE (2011) Peer effects in education: a survey of the theory and evidence. In: Handbook of Social Economics, Vol. 1B: Carnegie Mellon University, pp 1053–1163

  18. Eriksson T (1999) Executive compensation and tournament theory: empirical tests on danish data. J Labor Econ 17(2):262–280

    Article  Google Scholar 

  19. Garcia SM, Tor A (2009) The N-Effect. Psychol Sci 20(7):871–877

    Article  Google Scholar 

  20. Gneezy U, Niederle M, Rustichini A (2003) Performance in competitive environments: gender differences. Q J Econ 118(3):1049–1074

    Article  Google Scholar 

  21. Goodkind DM (1991) Creating new traditions in modern chinese populations: aiming for birth in the year of the dragon. Popul Dev Rev 17(4):663–686

    Article  Google Scholar 

  22. Goodkind DM (1993) New zodiacal influences on chinese family formation: Taiwan, 1976. Demography 30(2):127–142

    Article  Google Scholar 

  23. Goodkind DM (1995) The significance of demographic triviality: minority status and zodiacal fertility timing among chinese malaysians. Popul Stud 49(1):45–55

    Article  Google Scholar 

  24. Goodkind DM (1996) Chinese lunar birth timing in Singapore: new concerns for child quality amidst multicultural modernity. J Marriage Fam 58(3):784–795

    Article  Google Scholar 

  25. Green JR, Stokey NL (1983) A comparison of tournaments and contracts. J Polit Econ 91(3):349–364

    Article  Google Scholar 

  26. Hanushek EA, Kain JF, Markman JM, Rivkin SG (2003) Does peer ability affect student achievement? J Appl Econ 18(5):527–544

    Article  Google Scholar 

  27. Harbring C, Irlenbusch B (2003) An experimental study on tournament design. Labour Economics 10(4):443–464. Special Issue on Labour Market Experiments

    Article  Google Scholar 

  28. Johnson ND, Nye JVC (2011) Does fortune favor dragons? J Econ Behav Organ 78(1-2):85–97

    Article  Google Scholar 

  29. Jurajda S, Munich D (2011) Gender gap in performance under competitive pressure: admissions to czech universities. Am Econ Rev 101(3):514–18

    Article  Google Scholar 

  30. Konrad KA (2009) Strategy and dynamics in contests. Oxford University Press, Oxford

    Google Scholar 

  31. Korenman S, Neumark D (2000) Cohort crowding and youth labor markets: a cross-national analysis. In: Blanchflower DG, Freeman RB (eds) Youth employment and joblessness in advanced countries. University of Chicago Press, pp 57–106

  32. Krueger AB (1999) Experimental estimates of education production functions. Q J Econ 114(2):497–532

    Article  Google Scholar 

  33. Krueger AB, Whitmore DM (2001) The effect of attending a small class in the early grades on college-test taking and middle school test results: evidence from project star. Econ J 111(468):1–28

    Article  Google Scholar 

  34. Lazear EP (2001) Educational production. Q J Econ 116(3):777–803

    Article  Google Scholar 

  35. Lazear EP, Rosen S (1981) Rank-order tournaments as optimum labor contracts. J Polit Econ 89(5):841–864

    Article  Google Scholar 

  36. Lee J, Paik M (2006) Sex preferences and fertility in South Korea during the year of the horse. Demography 43(2):269–292

    Article  Google Scholar 

  37. Lefgren L (2004) Educational peer effects and the Chicago public schools. J Urban Econ 56(2):169–191

    Article  Google Scholar 

  38. Macunovich DJ (1999) The fortunes of one’s birth: relative cohort size and the youth labor market in the United States. J Popul Econ 12(2):215–272

    Article  Google Scholar 

  39. Main BGM, O’Reilly CA III, Wade J (1993) Top executive pay: tournament or teamwork? J Labor Econ 11(4):606–628

    Article  Google Scholar 

  40. Morin L-P (2015a) Cohort size and youth earnings: evidence from a quasi-experiment. Labour Econ 32:99–111

    Article  Google Scholar 

  41. Morin L-P (2015b) Do men and women respond differently to competition? Evidence from a major education reform. J Labor Econ 33(2):443–491

    Article  Google Scholar 

  42. Nalebuff BJ, Stiglitz JE (1983) Prizes and incentives: towards a general theory of compensation and competition. Bell J Econ 14(1):21–43

    Article  Google Scholar 

  43. Niederle M, Vesterlund L (2010) Explaining the gender gap in Math test scores: the role of competition. J Econ Perspect 24(2):129–144

    Article  Google Scholar 

  44. Nye JV, Xue MM (2014) Raising dragons, Working Paper 15-18, George Mason University

  45. Orrison A, Schotter A, Weigelt K (1997) On the design of optimal organizations using tournaments: an experimental examination, working paper 97-26, C.V. Starr Center for Applied Economics, New York University

  46. Ors E, Palomino F, Peyrache E (2013) Performance gender gap: does competition matter?. J Labor Econ 31(3):443–499

    Article  Google Scholar 

  47. Pong WY, Chow JCS (2002) On the pedagogy of examinations in Hong Kong. Teach Teach Educ 18(2):139–149. The Pedagogical Task of Teaching

    Article  Google Scholar 

  48. Rohlfs C, Reed A, Yamada H (2010) Causal effects of sex preference on sex-blind and sex-selective child avoidance and substitution across birth years: evidence from the Japanese Year of the Fire Horse. J Dev Econ 92(1):82–95

    Article  Google Scholar 

  49. Sim N (2015) Astronomics in action: the graduate earnings premium and the dragon effect in Singapore. Econ Inq 53(2):922–939

    Article  Google Scholar 

  50. South China Morning Post (1994) Primary schools hit by demand, South China Morning Post, p 5

  51. South China Morning Post (1999) System pressured to find 2,000 Form One places, South China Morning Post, p 7

  52. South China Morning Post (2000) Enter the Dragon Babies, South China Morning Post, p 1

  53. South China Morning Post (2012) The 12-year hitch, South China Morning Post, p 4

  54. Stapleton DC (1988) Educational attainment and cohort size. J Labor Econ 6 (3):330–361

    Article  Google Scholar 

  55. Stapleton DC, Young DJ (1984) The effects of demographic change on the distribution of wages, 1967–1990. J Human Res 19(2):175–201

    Article  Google Scholar 

  56. Vere JP (2008) Dragon children: identifying the causal effect of the first child on female labour supply with the chinese lunar calendar. Oxf Bull Econ Stat 70 (3):303–325

    Article  Google Scholar 

  57. Welch F (1979) Effects of cohort size on earnings: the baby boom babies’ financial bust. J Polit Econ 87(5):S65–S97

    Article  Google Scholar 

  58. Witkow MR, Fuligni AJ (2010) In-school versus out-of-school friendships and academic achievement among an ethnically diverse sample of adolescents. J Res Adolesc 20(3):631–650

    Article  Google Scholar 

  59. Woessmann L (2005) Educational production in east asia: the impact of family background and schooling policies on student performance. Ger Econ Rev 6(3):331–353

    Article  Google Scholar 

  60. Wong K-F, Yung L (2005) Do dragons have better fate?. Econ Inq 43 (3):689–697

    Article  Google Scholar 

  61. Yip PS, Lee J, Cheung YB (2002) The influence of the chinese zodiac on fertility in Hong Kong SAR. Soc Sci Med 55(10):1803–1812

    Article  Google Scholar 

  62. Zimmerman DJ (2003) Peer effects in academic outcomes: evidence from a natural experiment. Rev Econ Stat 85(1):9–23

    Article  Google Scholar 

Download references


I would like to thank Alex Mas, Cecilia Rouse, Harvey Rosen, Henry Farber, Nicholas Lawson, Quynh Nguyen, and the seminar and workshop participants at the Industrial Relations Section and the Public Finance Working Group at Princeton University for invaluable insights and comments. I am grateful to anonymous referees for their helpful feedback and guidance.

Author information



Corresponding author

Correspondence to Yan Lau.

Ethics declarations

Conflict of interest

The author declares that there is no conflict of interest.

Additional information

The views expressed are those of the author and do not necessarily reflect those of the Federal Trade Commission or any individual Commissioner.

Responsible editor: Junsen Zhang


Appendix A: Robustness checks excluding february-born students

This appendix details robustness checks using an analysis sample which excludes February-born students. Using this restricted sample addresses two concerns: (1) that the February-born group in the dragon cohort treatment cell contains both dragon babies and non-dragon babies (see footnote 11) and (2) that there is a misalignment in birth-months between PISA 2003 (March to February birth-months) and PISA 2006 (February to January birth-months) interacting with delayed survey timing (see footnote 16).

In particular, excluding February-born students fixes the misalignment in birth-months between the two PISA years; now, students in each PISA round are born from March to January. This means that despite the survey being conducted later by a month in 2003, there is no peculiar interaction with misaligned birth-months anymore, and the differences in age will difference out between the two PISA years.

This is confirmed by the summary statistics in Table 5, which presents the means and standard deviations of age (in months) of the four difference-in-differences cells, as well as tests for differences, over the sample excluding February-born students. Basically, this is identical to the statistics in the 7th row of Table 2, except for the new sample exclusion.

Table 5 Age profile of sample excluding february-born

The calculations show that after removing February-born students from the sample, there is exactly a 1-month difference between 9th graders in the two PISA years, accounting for the misaligned birth-month profile in 2003. There is also no statistically significant difference in the differences between grades and PISA years. This suggests that it is the presence of February-born students in the full sample, and how their birth-months interact with the delay in surveying, that is causing the original difference-in-differences of − 0.70 in Table 2.

Even though it seems like the February-born students are making the treatment and control groups not exactly comparable, this should not matter in the regressions controlling for an age quadratic function, which would account for any differences in age between the groups. If regression estimates using the full sample versus the restricted sample excluding February-born students are similar, then that would suggest that age differences have been adequately accounted for, and including February-born students into the analysis sample is not an issue.

Furthermore, if the estimates from these two regressions are similar, it also suggests that including some dragon babies born in late February is not an issue from a birth-month targeting perspective.

To test this, Tables 6 and 7 report results using this sample excluding February-born students. In each table, row (A) shows the original OLS estimates for ease of comparison, while row (B) shows corresponding estimates using the restricted sample. Comparing the two sets of estimates, we see that they are very similar, although the latter are less statistically significant because of the reduced sample size. This gives me confidence that (1) the presence of February-born students in the full sample did not materially affect the findings and that (2) the inclusion of quadratic covariates in age was adequate in accounting for any remaining differences with respect to age between the treatment and control groups. This is also consistent with the evidence in Fig. 2 which shows that parents selecting into the dragon year are not doing so in these initial months of the dragon year.

Table 6 Robustness checks for effort regressions
Table 7 Robustness checks for math score regressions

Appendix B: Robustness checks excluding control variables

This appendix details robustness checks showing that the original estimates are robust to the inclusion/exclusion of control variables (sex, birth country, age and its square, highest parental occupational status, each parent’s education level, number of books at home). These robustness specifications nonetheless include the full set of difference-in-differences terms and school fixed effects. The estimates are shown in row (C) of Tables 6 and 7 respectively.

Compared to the original OLS estimates in row (A), the estimates from these specifications without controls remain essentially unchanged. In spite of the exclusion of student control variables, the estimates that were statistically significant before remain as such. (In a host of similar robustness checks not reported here, I run regressions that exclude the control variables successively. With each combination of controls, similar results persist.) These findings show that the main results are robust to specification changes, and suggest that explanatory power of the dependent variables is coming primarily from the difference-in-differences dragon cohort interaction term, as opposed to other observable characteristics, such as those found to be dissimilar in Table 2. If the difference-in-differences estimates were biased by non-comparable comparison groups, then one would expect significant shifts in point estimates when the controls were included/excluded, which is not the case.

Overall, the results from these robustness checks suggest that the difference-in-differences analysis is not affected by the slight differences in observable characteristics, and that the other non-dragon cohort students in other grades and PISA years serve as valid counterfactuals.

Appendix C: Mapping procedure and interval regressions

The mapping procedure is as follows. First, integer hours in 2003 are mapped into the 2006 categories. Next, an integer number of hours is assigned to each category. For “No time,” 0 is assigned. For ranged categories with two endpoints, the midpoint of the endpoints is assigned. (For example, “2 or more but less than 4 h...” is assigned 3 h) For the “6 or more hours...” category, 8 h is assigned, because in the 2003 data, this is the median number of hours conditional on reporting 6 or more hours.

To assess whether the mapping procedure used to generate the continuous dependent variable hoursisy influences the results, I repeat the effort regressions using an interval regression procedure, where the dependent variable hoursisy is treated as the categories defined in the PISA survey. The interval regression technique is a generalization of the Tobit method, and accounts for censoring of the data within intervals defined by ranged categories of the dependent variable.

Row (D) of Table 6 presents these estimates for the full sample and restricted sub-samples. Compared to the original OLS estimates (row (A)), they are very similar, showing that the findings are robust to mapping hours categories into exact number of hours.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Lau, Y. The dragon cohort of Hong Kong: traditional beliefs, demographics, and education. J Popul Econ 32, 219–246 (2019).

Download citation


  • Education
  • Chinese zodiac
  • Dragon year
  • Schooling cohort
  • Fertility
  • Birth rate

JEL Classification

  • J13
  • I20
  • Z10