Li, Liu and Zhang (2012) estimated the return to education in urban China using the 2002 Chinese Twins Survey: ordinary least squares gives 8.4% per year of schooling, within-monozygotic-twin fixed effects give 2.7%, and 3.8% after instrumental correction of measurement error, with academic high school returning only about 4.0-5.4% while vocational high school, vocational college and college return roughly 20-40%. The authors themselves flag identification problems behind these numbers: within-pair fixed effects may deliver only a tightened bound because residual within-pair differences in ability and personality remain correlated with within-pair schooling differences; nonclassical measurement error in education-level indicators has no general implemented correction, so level-specific returns are reported only as bounds; monozygotic zygosity is identified by self-reported appearance and may be misclassified; and the proposed explanation that exam-oriented academic high school mainly selects future college students rather than building workplace skills lacks a structural causal test. This problem asks for point identification (or sharp bounds) of the causal returns to years and levels of schooling in China within or improving upon the twins design, robust to residual ability bias, nonclassical and family-correlated misreporting of education categories, and zygosity misclassification, together with causal evidence on the screening-versus-human-capital mechanism behind the level-specific pattern.

Background

The return to education is the causal percentage increase in earnings produced by an additional year of schooling or by completing a given educational level. The benchmark tool is the Mincer wage equation, a regression of log earnings on years of schooling and experience; ordinary least squares (OLS) estimates of its schooling coefficient are biased upward if unobserved ability or advantaged family background raises both schooling and earnings. Monozygotic (MZ) twins are genetically identical and typically share their childhood family environment, while dizygotic (DZ) twins share family environment but only about half of segregating genes. Differencing the wage equations of the two twins in a pair (a within-pair fixed-effects, FE, estimator) cancels the shared family and genetic components, which is why twins data have been used since the 1990s to purge ability bias from estimated returns. Krueger and Ashenfelter (1992) additionally collected multiple measurements of each twin's schooling - including each twin's report of the co-twin's schooling - and used the disagreement between these reports to correct classical measurement error (random reporting error, which biases the FE coefficient toward zero), an adjustment that later twins studies implemented with cross-report instrumental variables. A later critique symposium (Neumark 1999; Rouse 1999; Behrman and Rosenzweig 1999; Bound and Solon 1999) argued that within-pair schooling differences are themselves partly driven by unobserved ability, that measurement error in schooling reports is not classical (reports can be mean-reverting or differentially related to true schooling), and that twins samples are small and fragile, so twins-based returns should be interpreted cautiously.

In China the question is sharper because the education system is highly selective and exam-oriented: students are tracked after junior high school by entrance-examination scores into academic high school (preparing for the college entrance examination) or vocational high school (teaching occupation-specific skills), and tertiary capacity was rationed. Cross-sectional estimates for urban China rose from roughly 2-4% per year in the 1980s-1990s to about 8-10% by the 2000s, but their causal status was unclear. Zhang, Liu and Yung (2007) had earlier used twins to study Cultural-Revolution-era schooling disruption. Li, Liu and Zhang (2012) collected the Chinese Twins Survey (June-July 2002) in five cities (Chengdu, Chongqing, Harbin, Hefei, Wuhan), obtained 488 complete MZ pairs with earnings and education data, and estimated: OLS 8.4% per year; within-MZ FE 2.7%; and, after correcting measurement error with two cross-report instrumental-variable FE (IVFE) estimators in the Krueger-Ashenfelter tradition, 3.2-3.8%. A generalized-least-squares (seemingly unrelated regression) specification estimates the family/selection component at about 3.3% per year, as large as the within-pair causal return. Because the within-DZ FE estimate (4.4%) exceeds the within-MZ estimate, unobserved ability appears positively correlated with schooling, so even the corrected MZ estimate is best read as an upper bound. Level-specific estimates relative to junior high school are reported as bounds between a restricted-sample FE estimate and an IVFE estimate: academic high school 4.0-5.4% (often statistically indistinguishable from zero), vocational high school 19.6-21.9%, vocational college 21.5-23.0%, and college 35.7-40.0%. The authors interpret this pattern as evidence that exam-oriented academic high school mainly serves to select future college students (a screening or signaling role) rather than to add workplace human capital, in contrast to Bonjour et al. (2003) and other Western twins studies surveyed in their appendix.

Problem Statement

The common objective is a defensible causal answer to what education returns in urban China - overall, at each education level, and through which mechanism - by overcoming the identification obstacles that Li, Liu and Zhang (2012) acknowledge in their twins-based analysis. Concretely, the goal is point identification, or sharp bounds under stated assumptions (the tightest bounds those assumptions imply), of the causal returns to years of schooling and to education levels (junior high, academic high school, vocational high school, vocational college, college) within or improving upon the within-monozygotic-twin fixed-effects design, robust to the obstacles the paper acknowledges: residual within-pair ability, personality or early-experience differences correlated with within-pair schooling differences - under which the published within-MZ FE estimate of 2.7% and the measurement-error-corrected IVFE estimate of 3.8% are best read as upper bounds on the true return rather than as bracketing it; nonclassical measurement error in categorical educational-attainment indicators, covering the case where self-reports and co-twin cross-reports share family-correlated reporting errors so that existing cross-report instruments fail; and appearance-based self-reported zygosity classification, under which misclassified dizygotic pairs attenuate the estimated ability bias. The goal also includes causal evidence on whether the low return to exam-oriented academic high school (about 4.0-5.4% versus roughly 20-40% for vocational and tertiary levels) reflects a screening or selection mechanism - mainly selecting future college students rather than building workplace skills - as opposed to human-capital formation. The objective requires that the identification obstacles the paper acknowledges be addressed rather than assumed away, and that any institutional interpretation of the mechanism be causally identified rather than inferred from correlation patterns.

References

  1. Li, Hongbin; Liu, Pak Wai; Zhang, Junsen (2012). Estimating returns to education using twins in urban China. Journal of Development Economics 97(2), 494-504. DOI 10.1016/j.jdeveco.2011.05.009. https://doi.org/10.1016/j.jdeveco.2011.05.009
  2. Zhang, Junsen; Liu, Pak-Wai; Yung, Linda (2007). The Cultural Revolution and returns to schooling in China: Estimates based on twins. Journal of Development Economics 84(2), 631-639. DOI 10.1016/j.jdeveco.2006.12.006. https://doi.org/10.1016/j.jdeveco.2006.12.006
  3. Krueger, Alan; Ashenfelter, Orley (1992). Estimates of the Economic Return to Schooling from a New Sample of Twins. NBER Working Paper 4143. DOI 10.3386/w4143. https://doi.org/10.3386/w4143
  4. Rouse, Cecilia Elena (1999). Further estimates of the economic return to schooling from a new sample of twins. Economics of Education Review 18(2), 149-157. DOI 10.1016/s0272-7757(98)00038-7. https://doi.org/10.1016/s0272-7757(98)00038-7
  5. Behrman, Jere R.; Rosenzweig, Mark R. (1999). "Ability" biases in schooling returns and twins: a test and new estimates. Economics of Education Review 18(2), 159-167. DOI 10.1016/s0272-7757(98)00033-8. https://doi.org/10.1016/s0272-7757(98)00033-8
  6. Bound, John; Solon, Gary (1999). Double trouble: on the value of twins-based estimation of the return to schooling. Economics of Education Review 18(2), 169-182. DOI 10.1016/s0272-7757(98)00048-x. https://doi.org/10.1016/s0272-7757(98)00048-x
  7. Neumark, David (1999). Biases in twin estimates of the return to schooling. Economics of Education Review 18(2), 143-148. DOI 10.1016/s0272-7757(97)00022-8. https://doi.org/10.1016/s0272-7757(97)00022-8
  8. Bonjour, Dorothe; Cherkas, Lynn F.; Haskel, Jonathan E.; Hawkes, Denise D.; Spector, Tim D. (2003). Returns to Education: Evidence from U.K. Twins. American Economic Review 93(5), 1799-1812. DOI 10.1257/000282803322655554. https://doi.org/10.1257/000282803322655554
  9. Black, Dan A.; Berger, Mark C.; Scott, Frank A. (2000). Bounding Parameter Estimates with Nonclassical Measurement Error. Journal of the American Statistical Association 95(451), 739-748. DOI 10.1080/01621459.2000.10474262. https://doi.org/10.1080/01621459.2000.10474262
  10. Light, Audrey; Flores-Lagunes, Alfonso (2006). Measurement Error in Schooling: Evidence from Samples of Siblings and Identical Twins. Contributions in Economic Analysis and Policy 5(1). DOI 10.2202/1538-0645.1522. https://doi.org/10.2202/1538-0645.1522
  11. Mahajan, Aprajit (2006). Identification and Estimation of Regression Models with Misclassification. Econometrica 74(3), 631-665. DOI 10.1111/j.1468-0262.2006.00677.x. https://doi.org/10.1111/j.1468-0262.2006.00677.x
  12. Hu, Yingyao (2008). Identification and estimation of nonlinear models with misclassification error using instrumental variables: A general solution. Journal of Econometrics 144(1), 27-61. DOI 10.1016/j.jeconom.2007.12.001. https://doi.org/10.1016/j.jeconom.2007.12.001
  13. Molinari, Francesca (2008). Partial identification of probability distributions with misclassified data. Journal of Econometrics 144(1), 81-117. DOI 10.1016/j.jeconom.2007.12.003. https://doi.org/10.1016/j.jeconom.2007.12.003
  14. Battistin, Erich; Sianesi, Barbara (2011). Misclassified Treatment Status and Treatment Effects: An Application to Returns to Education in the United Kingdom. Review of Economics and Statistics 93(2), 495-509. DOI 10.1162/rest_a_00175. https://doi.org/10.1162/rest_a_00175
  15. van Hasselt, Martijn; Bollinger, Christopher R. (2012). Binary misclassification and identification in regression models. Economics Letters 115(1), 81-84. DOI 10.1016/j.econlet.2011.11.031. https://doi.org/10.1016/j.econlet.2011.11.031
  16. Amin, Vikesh (2011). Returns to Education: Evidence from UK Twins: Comment. American Economic Review 101(4), 1629-1635. DOI 10.1257/aer.101.4.1629. https://doi.org/10.1257/aer.101.4.1629
  17. Sandewall, Orjan; Cesarini, David; Johannesson, Magnus (2014). The co-twin methodology and returns to schooling - testing a critical assumption. Labour Economics 26, 1-10. DOI 10.1016/j.labeco.2013.10.002. https://doi.org/10.1016/j.labeco.2013.10.002
  18. Xiu, Lin; Gunderson, Morley (2013). Credential Effects and the Returns to Education in China. LABOUR: Review of Labour Economics and Industrial Relations 27(2), 225-248. DOI 10.1111/labr.12009. https://doi.org/10.1111/labr.12009
  19. Battistin, Erich; De Nadai, Michele; Sianesi, Barbara (2014). Misreported schooling, multiple measures and returns to educational qualifications. Journal of Econometrics 181(2), 136-150. DOI 10.1016/j.jeconom.2014.03.002. https://doi.org/10.1016/j.jeconom.2014.03.002
  20. Nakamuro, Makiko; Inui, Tomohiko; Yamagata, Shinji (2017). Returns to Education Using a Sample of Twins: Evidence from Japan. Asian Economic Journal 31(1), 61-81. DOI 10.1111/asej.12113. https://doi.org/10.1111/asej.12113
  21. Andersson, Matthew A. (2018). A Discordance Weighting Approach Estimating Occupational and Income Returns to Education. Twin Research and Human Genetics 21(3), 191-202. DOI 10.1017/thg.2018.20. https://doi.org/10.1017/thg.2018.20
  22. Ye, Maoliang; Zhang, Junsen; Li, Hongbin (2023). Twins, income, and happiness: Evidence from China. Proceedings of the National Academy of Sciences 120(25), e2221884120. DOI 10.1073/pnas.2221884120. https://doi.org/10.1073/pnas.2221884120
  23. Dai, Li; Martins, Pedro S. (2024). Does vocational education pay off in China? Evidence from city-level education supply shocks. Economic Modelling 140, 106863. DOI 10.1016/j.econmod.2024.106863. https://doi.org/10.1016/j.econmod.2024.106863
  24. Chen, Jie; Pastore, Francesco (2024). Dynamics of returns to vocational education in China: 2010-2017. Humanities and Social Sciences Communications 11, article 118. DOI 10.1057/s41599-024-02616-2. https://doi.org/10.1057/s41599-024-02616-2
  25. Liu, Hongyan; Xu, Yali; Zhang, Ruyue; Nie, Jingchun; Rule, Andrew; Rozelle, Scott; Yuan, Lingran (2024). Comparing the payoff to vocational and academic secondary education in China over time. International Journal of Educational Development 107, 103038. DOI 10.1016/j.ijedudev.2024.103038. https://doi.org/10.1016/j.ijedudev.2024.103038
  26. Peeters, Hilde; Van Gestel, Sofie; Vlietinck, Robert; Derom, Catherine; Derom, Robert (1998). Validation of a Telephone Zygosity Questionnaire in Twins of Known Zygosity. Behavior Genetics 28(3), 159-163. DOI 10.1023/a:1021416112215. https://doi.org/10.1023/a:1021416112215
  27. Li, Hongbin; Liu, Pak Wai; Ma, Hong; Zhang, Junsen (2005). Does Education Pay in Urban China? Estimating Returns to Education Using Twins. CUHK Department of Economics Discussion Paper 00013 (earlier working-paper version of the same project). https://core.ac.uk/download/6271019.pdf