What a family actually transmits — and why the Chinese lineage record is the under-used arbitrator

Macheng Shen × agent workflow · 2026-08-02 · cognitive state: speculative · every number below was checked against its source; corrections invited

中文摘要

两组都很硬的证据表面上互相矛盾:(A) 双生子与收养研究里,「共享家庭环境」对成年人格变异的解释力接近于零;(B) 姓氏/世系方法测出的潜在社会地位每代衰减系数约 0.79,而单代可测收入或教育只有 0.3–0.4,且这个速率在英格兰 1600–2022 年间基本没变(样本 422,374 人)。家庭几乎不让自己的孩子彼此分化,家族线却能守住位置几百年。

本文的结论:调和这两者的最强解释里,占大头的那部分完全不需要基因 —— 单代地位测量是潜在地位的噪声代理,噪声会把单代相关衰减到零附近,却不会跨代累积;再加上高度的择偶同质性(配偶潜在值相关 0.57)会对称地放大文化传递与遗传传递,因此它不能用来仲裁二者。真正有争议的那一步 —— 把「高而稳定的持续性」读成「可遗传的社会能力」—— 在 2025 年被 Proc. R. Soc. B 的一篇论文正面挑战,而原作者当年 10 月发表了工作论文回应。这场交锋是活的,不是已经判完的。

本文另一半论点:能仲裁这件事的最好数据在中国,而且已经机读化 —— CGED-Q 收录 4,629,016 条清代官员记录、CMGPD-辽宁 约 150 万条观测、CBDB 约 65.8 万人物。清代辽宁的证据显示地位沿宗族而非核心家庭传递(村内宗族间差异至少是村际差异的两倍),而 1949 年后的土改与文革构成一次近乎理想的自然实验:物质资本被摧毁而家庭结构保留,半个世纪后旧精英的孙辈收入仍高约 16%、受教育年限多约 11%。但必须同时说:中国家谱是被建构的社会文书,不是血统档案 —— 明清有职业「谱匠」卖预印族谱模板,伪托欧阳修、朱熹的序文在毫无关系的姓氏间逐字重复。全文引用 30+ 条文献,均已逐条核实。

1 · Side A: the family barely differentiates its own children

The canonical summary is Turkheimer's three laws: all behavioural traits are heritable; the effect of being raised in the same family is smaller than the effect of genes; and a substantial portion of variance is explained by neither [1]. For adult personality specifically, the shared-environment component is near zero.

Two things almost every popular retelling gets wrong.

• This is a claim about sources of variation, not about levels. Your family fully determines your language, your nationality, your religion, your starting wealth and your passport. It just does not make siblings differ from each other less than random strangers do on personality measures. "Parents don't matter" does not follow and is not what the studies say.
• The result is conditional on the range of environments sampled. In impoverished American families, roughly 60% of the variance in young children's IQ is shared environment and heritability approaches zero — exactly reversing in affluent families [2]. The headline finding is, to a substantial degree, a middle-class-sample finding. (The specific SES interaction replicates unevenly outside the US, which is itself worth knowing.)

And the number is design-dependent in a way that single citations conceal: a meta-analysis gives personality heritability just under .47 from twin designs and just above .22 from family and adoption designs [3]. Quoting "40–50%" without saying which design produced it hides a factor-of-two methodological gap.

2 · Side B: lineages hold position for four centuries

Clark's method estimates the persistence of a latent status variable using rare surnames and, in the 2023 study, individually linked English genealogies. The headline results: correlations detectable out to fourth cousins; a per-generation decline factor of 0.79; a correlation of 0.57 between spouses in the underlying value; and a rate essentially unchanged from 1600 to 2022 despite the Industrial Revolution, mass schooling, and the welfare state [5][6].

That last part is the genuinely startling claim, and it is the one that gets the least scrutiny in popular retellings: not that status persists, but that nothing anyone did in four centuries changed the rate.

3 · How the two sides reconcile — mostly without genetics

3.1 Measurement error, which is the boring answer and does most of the work

Any single-generation measure of status — this year's income, highest degree — is a noisy proxy for latent status. Noise attenuates a single-generation correlation toward zero. But noise does not accumulate along a lineage: averaging over a surname group or over many descendants cancels it. So the same underlying process can look like 0.35 measured one generation at a time and 0.75 measured across many. This reconciles a large part of the gap with zero genetic content whatsoever, and it is the part of Clark's argument that critics generally grant.

3.2 Assortative mating amplifies — symmetrically

A spousal correlation of 0.57 in latent status [6] slows regression to the mean substantially. But note carefully: assortative mating amplifies cultural and environmental transmission exactly as much as it amplifies genetic transmission. It cannot be used to arbitrate between them, and an argument that invokes it in support of a genetic reading has quietly done nothing.

3.3 Genetic nurture: the molecular version of "家风"

In 21,637 Icelandic parent–offspring sets, parental alleles that were not transmitted to the child still predict the child's educational attainment, at 29.9% of the effect size of the transmitted alleles [4]. The mechanism is that parents' genotypes shape the environment they build, which then acts on the child.

This is a beautiful result and it cuts in a direction people rarely notice: it is simultaneously the strongest molecular evidence for heritable family influence and a demonstration that "genetic" and "environmental" transmission cannot be cleanly separated inside a family at all. A non-transmitted allele acting through the household is, at the same time, entirely genetic in origin and entirely environmental in mechanism.

4 · The contested step, and the exchange that is still live

Everything above is broadly agreed. The contested inference is Clark's step from "persistence is high and stable" to "therefore status is transmitted as an additive-genetic trait of social competence".

Against. Benning, Carlson, Smith, Shaw & Harpak argue that the additive model conflates genetic with non-genetic transmission (wealth being the obvious case), that statistical artefacts inflate the family correlations, that persistence in fact varies and declines over the period, and conclude that the analysis "provides no information" about the genetic versus non-genetic split. This is now peer-reviewed in Proceedings of the Royal Society B (2025), not a preprint [8].

For. Clark replied in October 2025 in a working paper titled, directly, "Hereditarian Fallacies?", arguing there is no simple hereditarian fallacy in the 2023 analysis [9]. We link it because publishing the critique without the reply would be the mirror image of the error the critique is about.

An independent empirical challenge. Clark's framework requires that single-measure estimates are biased downward. Vosters tested exactly this using multiple status measures and found no evidence of substantial bias [10]. That is a direct hit on the attenuation story rather than a philosophical objection, and it is under-cited relative to its importance.

Our reading, stated as a number rather than a posture: P(the persistence is real and large) ≈ 0.70. P(Clark's specific additive-genetic interpretation of it) ≈ 0.20. Those are different claims and conflating them is how this literature goes wrong in both directions — hereditarians treat the first as proof of the second, and their opponents treat doubts about the second as grounds to deny the first.

5 · Why the Chinese record is the under-used arbitrator

5.1 Status travelled along the lineage, not the parent–child link

In Qing-dynasty Liaoning, father–son status transmission was weaker than in contemporaneous Western societies — and yet kin-group persistence was strong. Variation between descent groups within a village was at least twice the variation between villages; nineteenth-century descent-group rank correlations were r = 0.38 (within district, n = 562) and r = 0.47 (within village, n = 149) [15]. A study of 20,000 reconstructed patrilineages found that lineages founded by high-status males kept growing faster for a hundred and fifty years [16].

This is a structural blind spot in the behaviour-genetics designs, which are built around the nuclear family. If the transmitting unit is the descent group — with its shared land, its charitable estate, its school, its network, and its marriage market — then a design that measures only sibling similarity is measuring the wrong object, and finding c² ≈ 0 tells you less than it appears to.

5.2 A natural experiment that destroyed capital and left families standing

The 1949 land reform and the Cultural Revolution destroyed material capital on a scale with few parallels, while leaving families biologically and socially intact. Half a century later, grandchildren of the pre-revolution elite earn about 16% more and have about 11% more years of schooling [14].

Two honest caveats, both load-bearing. First, this is still an NBER working paper six years on, not a published article; treat the numbers as provisional and do not invent a journal for it. Second, the authors' proposed mechanism — transmitted values, aspirations, and peer effects — is their interpretation of correlational evidence. The finding that the effect is detectable in adolescents who have not yet finished schooling does argue against a simple returns-to-education story. It does not rule out surviving networks or unmeasured assets that were never actually confiscated.

5.3 The data are real, large, and machine-readable

• CGED-Q (China Government Employee Database – Qing): 4,629,016 records of officials, 1760–1912 [23][24].
• CMGPD-Liaoning: ~1.5 million triennial observations of ~260,000 individuals, 1749–1909, deposited at ICPSR [21]; a Shuangcheng companion covers 1866–1913 [22].
• CBDB (China Biographical Database): ~658,000 individuals with ~483,000 kinship links, freely downloadable, with a known elite bias inherited from its sources [25].
• 《中国家谱总目》: a catalogue of 52,401 genealogies covering 608 surnames [26]. Shanghai Library alone holds roughly 40,000 original genealogies. FamilySearch has digitised over 13 million images of Chinese genealogies — mostly browse-only, with limited name indexing [27].

[our speculation] The frontier, as of mid-2026, is below the catalogue layer: entity extraction and record linkage across the genealogy corpus itself. We state that as an assessment rather than a finding, and we note evidence it may already be going stale — CGED-Q's examination records have added roughly 900,000 linked kin of 30,000 degree-holders, which is precisely this kind of extraction.

6 · The counterweight: a genealogy is a social document, not a pedigree

Any argument built on Chinese genealogies has to survive this section, so it comes before the conclusions rather than after.

Lineages were strategically composed. Szonyi's work on the Fuzhou region and Faure's on the Pearl River Delta both show late-imperial lineages constructing themselves — back-filling respectable ancestors to claim orthodox status and, concretely, land rights [28][29]. This is mainstream historiography, not revisionism. [widely repeated tradition] The associated colour — professional 谱匠 selling pre-printed lineage templates with the surname filled in afterwards, and forged prefaces attributed to Ouyang Xiu or Zhu Xi recurring across unrelated surnames — is consistently attested in Chinese writing on the subject; we could not locate a refereed source for the specific anecdote and mark it accordingly. The underlying claim it illustrates is carried by [28] and [29].

Y-DNA cuts both ways, and the famous case cuts against its own subject. Six of eight unrelated modern lineages claiming descent from Cao Cao share the rare haplogroup O2*-M268 [31], and ancient DNA from the tooth of his granduncle Cao Ding matches it [32]. That is a genuine positive result. But the same work's primary finding was negative: Cao Cao's own claimed descent from Marquis Cao Can is refuted by the data. The Aisin Gioro study makes the general point cleanly — the ritual patriline and the biological patriline are different objects [33]. And for the most-claimed lineage of all, testing of Kong claimants returns many divergent haplogroups; no "Confucius haplogroup" has been established, and anything circulating online that names one is unsourced.

7 · Cold water on the most-cited Chinese success case

The Zeng Guofan family is invariably offered as proof that explicit written family precepts work. The individual achievements are real and independently verifiable: Zeng Jize renegotiated the Treaty of Saint Petersburg in February 1881, recovering a large part of the Tekes valley; Zeng Jihong computed π to 100 digits; Zeng Zhaolun, chemist, was elected to the Chinese Academy of Sciences in 1955; Zeng Zhaoyu directed the Nanjing Museum; Zeng Xianzhi served as a vice-chair of the All-China Women's Federation.

The famous statistic is another matter. "Eight generations without a wastrel, 240-odd people of note" traces to a family-association and family-virtue publicity lineage — a 2016 newspaper series originating from the Zeng Guofan research society, subsequently recycled through encyclopedias and discipline-themed outlets. It pools the direct line with the collateral lines, counts only successes, and has never been independently audited. Note also the genealogical correction that this framing obscures: Zeng Zhaolun and Zeng Zhaoyu descend from Zeng Guofan's brother Zeng Guohuang, and Zeng Xianzhi from his brother Zeng Guoquan — not from Zeng Guofan himself.

The written precepts genuinely exist as primary sources, which makes this a rare case where the transmission method is documented rather than inferred — the 八本堂 and 日课十二条, and the eight-character maxim attributed to his grandfather, 书蔬鱼猪早扫考宝. But whether they worked has never been causally identified, and "shared environment ≈ 0 for adult personality" is a real headwind for the claim that a written rule set is what did the work, as against wealth, network, marriage market, and selection on who gets counted.

8 · Three questions this literature leaves open

Stated so that each could be answered and each could come out against us.

1. Is the transmitting unit the nuclear family or the descent group? The Liaoning result [15] says descent group in that setting. Testable elsewhere: compare within-lineage and within-household status correlations in a corpus where both are linkable. Falsified if descent-group correlations collapse to household correlations once household is controlled.
2. Does the 1949–1976 destruction of capital identify a non-material channel? [14] argues yes. Falsified if "class status" labels turn out to have tracked human or social capital that was never actually confiscated — urban household registration, surviving credentials, retained professional networks.
3. Does persistence measured on Chinese lineages match the English rate, and is it stable across the same kind of institutional upheaval? Surname-based estimates for China exist and are broadly consistent with high persistence [17], with lower coefficients reported for the Communist era; the Tongcheng study finds mobility slow but visibly rising in the seventeenth century [18]. Falsified if a Chinese estimate with individually linked records lands near the single-generation 0.3–0.4 rather than the lineage 0.7–0.8.

9 · Calibrated confidence

References

Each reference below was checked to exist and checked against what it actually reports. Where our working draft had an error — a wrong DOI, a misattributed blog post, a mischaracterised finding — the corrected version is what appears here. If any citation is still wrong, please write.

1. Turkheimer, E. (2000). Three laws of behavior genetics and what they mean. Current Directions in Psychological Science 9(5):160–164. doi:10.1111/1467-8721.00084
2. Turkheimer, E., Haley, A., Waldron, M., D'Onofrio, B. & Gottesman, I. I. (2003). Socioeconomic status modifies heritability of IQ in young children. Psychological Science 14(6):623–628.
3. Vukasović, T. & Bratko, D. (2015). Heritability of personality: a meta-analysis of behavior genetic studies. Psychological Bulletin 141(4):769–785.
4. Kong, A., Thorleifsson, G., Frigge, M. L. et al. (2018). The nature of nurture: effects of parental genotypes. Science 359(6374):424–428. doi:10.1126/science.aan6877
5. Clark, G. (2014). The Son Also Rises: Surnames and the History of Social Mobility. Princeton University Press.
6. Clark, G. (2023). The inheritance of social status: England, 1600 to 2022. PNAS 120(27):e2300926120. doi:10.1073/pnas.2300926120
7. Lee, J. J. (2023). The heritability and persistence of social class in England. PNAS 120(29):e2309250120 — a supportive commentary on [6], not Clark's own paper.
8. Benning, J. W., Carlson, J., Smith, O. S., Shaw, R. G. & Harpak, A. (2025). Confounding fuels misinterpretation in human genetics. Proceedings of the Royal Society B 292(2058):20251615. doi:10.1098/rspb.2025.1615
9. Clark, G. (2025). Hereditarian fallacies? Inheritance of social status in England, 1600–2022. EHES Working Paper No. 285, October 2025.
10. Vosters, K. (2018). Is the simple law of mobility really a law? Testing Clark's hypothesis. The Economic Journal 128(612):F404–F421. doi:10.1111/ecoj.12516
11. Corak, M. (2014). Review of The Son Also Rises. Science 344(6186):812.
12. Henderson, D. R. (2014). The great-grandson also rises. Econlib / EconLog, 7 May 2014.
13. Kling, A. (2014). Heritability of social status. Econlib feature article.
14. Alesina, A. F., Seror, M., Yang, D. Y., You, Y. & Zeng, W. (2020). Persistence despite revolutions. NBER Working Paper 27053. Unpublished working paper as of August 2026.
15. Campbell, C. D. & Lee, J. Z. (2011). Kinship and the long-term persistence of inequality in Liaoning, China, 1749–2005. Chinese Sociological Review 44(1):71–103.
16. Song, X., Campbell, C. D. & Lee, J. Z. (2015). Ancestry matters: patrilineage growth and extinction. American Sociological Review 80(6):1150–1174.
17. Hao, Y. & Clark, G. (2021). Social mobility in China, 1645–2012: a surname study. China Economic Quarterly International 1(3):208–221. doi:10.1016/j.ceqi.2021.08.002
18. Shiue, C. H. (2025). Social mobility in the long run: an analysis of Tongcheng, China, 1300 to 1900. Journal of Economic History 85(2):370–410.
19. Wang, Y. (2022). Blood is thicker than water: elite kinship networks and state building in imperial China. American Political Science Review 116(3):896–910. doi:10.1017/S0003055421001490
20. Wen, F., Wang, E. H. & Hout, M. (2024). Social mobility in the Tang dynasty as the imperial examination rose and aristocratic family pedigree declined, 618–907 CE. PNAS 121(4):e2305564121
21. CMGPD-Liaoning, 1749–1909. ICPSR/DSDR study 27063.
22. CMGPD-Shuangcheng, 1866–1913. ICPSR/DSDR study 35292.
23. CGED-Q Jinshenlu public release, Lee–Campbell Group. HKUST DataSpace doi:10.14711/dataset/E9GKRS; Harvard Dataverse mirror doi:10.7910/DVN/GMQWVZ
24. Chen, B., Campbell, C., Ren, Y., Hao, X. & Lee, J. Z. (2020). Big data for the study of Qing officialdom: the China Government Employee Database-Qing (CGED-Q). Journal of Chinese History.
25. China Biographical Database (CBDB), Harvard University / Academia Sinica / Peking University.
26. 《中国家谱总目》(上海图书馆编), 上海古籍出版社, 2009 — 52,401 genealogies, 608 surnames.
27. FamilySearch, China, Collection of Genealogies, 1239–2014 — over 13 million images, largely browse-only.
28. Szonyi, M. (2002). Practicing Kinship: Lineage and Descent in Late Imperial China. Stanford University Press.
29. Faure, D. (2007). Emperor and Ancestor: State and Lineage in South China. Stanford University Press.
30. 王明珂,《英雄祖先与弟兄民族:根基历史的文本与情境》.
31. Wang, C.-C., Yan, S., Hou, Z. et al. (2012). Present Y chromosomes reveal the ancestry of Emperor Cao Cao of 1800 years ago. Journal of Human Genetics 57:216–218.
32. Wang, C.-C. et al. (2013). Ancient DNA of Emperor Cao Cao's granduncle matches those of his present descendants. Journal of Human Genetics 58:238–239.
33. Wang, C.-C., Yan, S. et al. (2015). Y chromosome of Aisin Gioro, the imperial house of the Qing dynasty. Journal of Human Genetics 50(3).
34. Chen, N. (2018). Dreams of imperial descent drive Chinese to DNA testing. Sixth Tone, 25 April 2018 — on consumer DNA testing marketed on famous-ancestor claims.
35. 23魔方 (consumer genetics company), analysis of Y-haplogroup consistency within claimed genealogies — an unrefereed commercial blog post with no sampling frame; see the boxed caveat in §6.