Tag: AUC

  • 命理判定的门槛:个体命运断言需要多大的效应量与判别力

    Evidence level: A

    如果”用出生时刻判定个体命运”是一个可以检验的科学命题,那么它必须先跨过一道门槛:其携带的判别信息至少要达到 AUC 0.75 甚至更高的量级,对应的效应量(OR)至少要 3 倍以上——而现代流行病学能测量的出生时刻效应,连这个门槛的十分之一都够不着。

    正文

    上一篇文章《出生时刻的信息量上限》从”实测”端算出出生季节效应携带的信息量(约 0.0002 比特),与命理”个体命运判定”所需信息量(10+ 比特)之间存在 3-4 个数量级的鸿沟。本文换一个角度,从”理论”端问:如果命理的个体断言是真的,它需要多大的效应量与判别力?

    一、判别力的理论上限:遗传力与患病率决定天花板

    基因组学个体化预测有一套成熟理论:任何预测因子的判别力(AUC)都存在绝对上限,由结局的遗传力(heritability)与患病率共同决定。Wray 等人 2010 年给出了最大 AUC 与遗传力、患病率的显式方程,并计算了 17 种复杂疾病的”理论天花板”:要达到 AUC 0.75,各疾病需要解释的遗传方差从 10% 到 74% 不等——高遗传力疾病相对”容易”,大多数复杂性状几乎不可能靠单一类别的预测因子达到高判别力。Dreyfuss 等人 2012 年进一步证明,即使不做任何分布假设,遗传预测准确度也存在数学上的硬上限。这条理论给命理命题的第一个约束:个体命运判定必须建立在”出生时刻信息”与命运结局足够强的关联之上,而这个关联强度有理论天花板。 出生时刻(八字)作为预测因子,其判别力上限同样受限于它解释了命运结局多少方差——如果它像实测那样只解释百分之零点几,那么无论命理逻辑多精致,其判别力都不可能超过 0.51。

    二、”够用”的门槛:AUC 0.7 不够,至少 0.75-0.8

    临床预测领域对”多少 AUC 才算够用”有大量讨论。流行的经验阈值是 0.7(可接受)、0.8(优秀)、0.9(杰出)——但 White 等人 2023 年在 BMC Medicine 上分析了 PubMed 中 306,888 个报告过的 AUC 值,发现分布在这三个阈值处出现明显”堆积”,而阈值之下”短缺”,提示大量研究者为报告”漂亮”的 AUC 而反复重分析数据(AUC hacking)。这说明 0.7/0.8/0.9 不仅是经验分界,还成了被系统博弈的对象——真正经得起检验的模型,门槛应该更高。

    Vickers 的决策曲线分析(DCA)方法论给出了更严格的回答:AUC 本身不回答”模型是否值得用”,必须结合净获益(net benefit)与阈值概率(threshold probability)评估——”根据预测采取行动”的好处是否超过”误判”的代价。Vickers 与 Holland 2021 年的综述明确指出:“AUC 要多少才够用”没有统一答案——取决于使用场景与代价结构。

    那为什么本文仍坚持”至少 AUC 0.75″?因为命理不是医疗筛查,它的断言(职业、婚期、健康事件)没有明确的”行动与代价”框架,却承诺了极高粒度的个体判定。对照临床最强的预测工具多基因风险评分(PRS)——Dudbridge 2013 年的理论分析表明,即使标记面板解释了全部遗传力,要达到临床有用的 AUC(约 0.75),多数复杂疾病也需要数十万到数百万的样本规模来估计效应。命理的输入只是出生时刻,若它的判别力要达到 PRS 的天花板,意味着出生时刻必须携带相当于”全部遗传信息”的预测力——这在效应量上是不可能的。

    三、把门槛换算成效应量:AUC 0.75 需要 OR 4 以上

    AUC 与效应量(OR)的换算有标准公式。对一个二分类暴露(如”这个八字是否吉命”),AUC 0.60 约对应 OR 2.0;AUC 0.70 对应 OR 约 4.0;AUC 0.80 对应 OR 约 9-10。也就是说,命理断言要达到 AUC 0.75 的门槛,出生时刻暴露的效应量至少需要 OR 4-5——”吉命”与”凶命”之间的风险差 4 倍以上。

    对照现实:出生季节与精神分裂症的关联 OR 1.05,出生月份与数十种疾病的关联 OR 1.02-1.10,即使按效应最强的纬度带计算,出生时刻相关暴露的 OR 也极少超过 1.3。实测效应与所需效应量之间隔着约 20-30 倍。更严峻的是:即使 OR 达 3.0 的强生物标志物,加入模型后的 AUC 增量也可能微乎其微,因为判别力还取决于暴露在人群中的分布与结局患病率。

    四、另一个被低估的门槛:多重断言的一致性

    命理通常一次性做出数十个独立断言(事业、婚姻、子女、健康、财运、流年……)。一个很少被讨论的统计学含义:每个断言都是一个独立的判别问题,都要跨过各自的 AUC 门槛。 如果单个断言的判别力只有 AUC 0.55(已远超实测的 0.51),”同时准确命中 10 个维度”的概率接近零。要让”整套判词”在统计上可辩护,每个维度都要相当高,这进一步抬高了总门槛。还有一个实践陷阱:用”事后觉得准”评价命理是无效的。临床预测领域尚且存在 AUC 系统性高估,命理断言更缺乏预注册、盲法与多重检验校正。任何关于命理判别力的主张,都必须先通过盲法+预注册的检验,才能进入与 AUC 门槛对话的环节。

    五、结论:门槛立起来了,现实够不着

    把这条理论链收拢:命理”个体命运判定”若要成为可辩护的科学命题,出生时刻暴露的判别力至少要达到 AUC 0.75(对应 OR 4-5),且每个断言维度都要独立达标;而现代流行病学实测的出生时刻效应(OR 1.02-1.10,AUC 0.51)连门槛的十分之一都不到。这条鸿沟不是”统计方法不够先进”造成的,而是理论天花板与实测效应量共同决定的。科学能做的,是把门槛本身立清楚——不是断言命理”不可能”,而是给出一个可以被检验、被反驳的量化标准。

    证据与局限

    已验证: 遗传预测判别力存在由遗传力与患病率决定的绝对理论上限(PMID 22827772, 20195508);PRS 达到 AUC 0.75 需要解释大部分遗传方差且需海量样本(PMID 23555274);决策曲线分析框架下”AUC 多高才够用”取决于净获益与代价结构而非 AUC 本身(PMID 17099194, 33676020);报告中的 AUC 存在 0.7/0.8/0.9 阈值堆积的系统性高估证据(PMID 37667344);稀有事件下 AUC 的可靠性由事件数而非事件率决定(PMID 41224138)。

    部分验证: AUC-OR 换算公式有成熟方法学(Chinn 2000 等),但实际判别力还受暴露分布与患病率影响,需个案计算。

    待验证: 出生时刻(时辰粒度)暴露的效应量是否在任何疾病上达到 OR 3 以上;命理多个断言维度之间的相关性是否允许”部分维度达标”;命盘评分操作化后在临床模型基线之上能否带来 NRI/净获益增量(对应 backlog #419)。

    相关阅读

    证据等级: A
    参考文献: PMID 22827772, PMID 20195508, PMID 23555274, PMID 17099194, PMID 33676020, PMID 37667344, PMID 41224138


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    Thresholds for Fate Determination: Required Effect Sizes and Discriminatory Power for Individual Destiny Claims

    Evidence level: A

    If the proposition “judging individual fate based on birth time” is a testable scientific claim, it must first cross a threshold: the discriminative information carried must reach an AUC of at least 0.75, or higher. The corresponding effect size (Odds Ratio) must be at least 3-fold—yet modern epidemiological measurements of birth-time effects fall short of even one-tenth of this threshold.

    Main Text

    The previous article, “Information Content Upper Limits of Birth Time,” calculated the information content carried by birth season effects from an “empirical” perspective (approx. 0.0002 bits). This stands in a gap of 3–4 orders of magnitude compared to the information required for “individual fate determination” in Yijing Studies (10+ bits). This article approaches the problem from a “theoretical” perspective: If individual fortune assertions in Yijing Studies are true, what effect size and discriminative power are required?

    I. Theoretical Upper Limit of Discriminative Power: Heritability and Prevalence Set the Ceiling

    Genomic individualized prediction has a mature theoretical framework: any predictor’s discriminative power (AUC) has an absolute upper limit determined jointly by the heritability of the outcome and its prevalence. Wray et al. (2010) provided explicit equations for maximum AUC based on heritability and prevalence, calculating the “theoretical ceilings” for 17 complex diseases. To achieve an AUC of 0.75, the genetic variance explained by each disease ranges from 10% to 74%; high-heritability diseases are relatively “easier,” while most complex traits almost certainly cannot achieve high discriminative power via a single categorical predictor. Dreyfuss et al. (2012) further demonstrated that even without distributional assumptions, genetic prediction accuracy exists a mathematical hard ceiling. This theory imposes the first constraint on the Yijing Studies proposition: Individual fortune determination must be built upon a sufficiently strong association between “birth time information” and the outcome, and this association strength has a theoretical ceiling. As a predictor, birth time (Bazi) is subject to the same limit; its discriminative power depends on how much variance it explains regarding the outcome. If it explains only a fraction of a percent, as empirical data suggests, its discriminative power cannot exceed 0.51 regardless of how refined the Yijing logic is.

    II. The “Sufficient” Threshold: AUC 0.7 Is Insufficient; At Least 0.75–0.8 Is Required

    There is extensive discussion in clinical prediction regarding “how much AUC is sufficient.” Popular empirical thresholds are 0.7 (acceptable), 0.8 (excellent), and 0.9 (outstanding). However, White et al. (2023), analyzing 306,888 reported AUC values from PubMed in BMC Medicine, found significant “clumping” at these thresholds while there is a “shortage” below them, suggesting researchers repeatedly reanalyze data to report “pretty” AUCs (AUC hacking). This indicates that 0.7/0.8/0.9 are not merely empirical boundaries but targets of systematic gaming—models that truly withstand scrutiny should have higher thresholds.

    Vickers’ Decision Curve Analysis (DCA) methodology provides a stricter answer: AUC alone does not answer “whether the model is worth using”; it must be evaluated alongside net benefit and threshold probability—determining whether the benefit of acting on predictions exceeds the cost of misclassification. A review by Vickers and Holland (2021) explicitly states: “There is no single answer to ‘how much AUC is sufficient’—it depends on the context and cost structure.”

    Then why does this article persist in insisting on “at least AUC 0.75”? Because Yijing Studies is not medical screening; its assertions (career, marriage date, health events) lack a clear “action and cost” framework yet promise high-granularity individual determination. Comparing to the strongest clinical predictor, polygenic risk scores (PRS)—Dudbridge’s (2013) theoretical analysis shows that even if a marker panel explains all heritability, achieving a clinically useful AUC (approx. 0.75) for most complex diseases requires hundreds of thousands to millions of sample sizes to estimate effects. Yijing Studies input is merely birth time; if its discriminative power were to reach the PRS ceiling, birth time would need to carry predictive power equivalent to “all genetic information”—which is impossible in terms of effect size.

    III. Converting Thresholds to Effect Size: AUC 0.75 Requires OR > 4

    Conversion between AUC and effect size (OR) follows standard formulas. For a binary exposure (e.g., “Is this Bazi auspicious?”), an AUC of 0.60 corresponds roughly to OR 2.0; AUC 0.70 corresponds to OR approx. 4.0; AUC 0.80 corresponds to OR approx. 9–10. Thus, to meet the AUC 0.75 threshold, the effect size of birth time exposure must be at least OR 4–5—a risk difference of more than four-fold between “auspicious” and “inauspicious” fates.

    Contrast with reality: The association between birth season and schizophrenia has an OR of 1.05; associations between birth month and dozens of diseases range from OR 1.02 to 1.10. Even calculating based on the strongest latitudinal bands, birth-time related exposures rarely exceed an OR of 1.3. There is a gap of approx. 20–30 times between empirical effects and required effect sizes. More severely, even with strong biomarkers reaching OR 3.0, the incremental AUC added to a model may be negligible because discriminative power also depends on the distribution of exposure in the population and outcome prevalence.

    IV. Another Underestimated Threshold: Consistency Across Multiple Assertions

    Yijing Studies typically makes dozens of independent assertions at once (career, marriage, children, health, wealth, yearly trends…). A statistical implication rarely discussed: Each assertion is an independent discriminative problem requiring its own AUC threshold. If the discriminative power of a single assertion is only AUC 0.55 (already far exceeding the empirical 0.51), the probability of “simultaneously hitting 10 dimensions accurately” is near zero. To make the “entire set of judgments” statistically defensible, each dimension must be quite high, further raising the total threshold. Another practical trap: evaluating Yijing Studies via “retrospective feeling of accuracy” is invalid. Clinical prediction fields already suffer from systematic overestimation of AUC; Yijing assertions lack preregistration, blinding, and multiple testing corrections. Any claim regarding Yijing Studies discriminative power must first pass blind + preregistered testing before entering the dialogue with AUC thresholds.

    V. Conclusion: The Threshold Is Established; Reality Falls Short

    Summarizing this theoretical chain: For “individual fate determination” in Yijing Studies to become a defensible scientific proposition, birth time exposure discriminative power must reach at least AUC 0.75 (corresponding to OR 4–5), with each assertion dimension independently meeting the standard. Modern epidemiological empirical measurements of birth time effects (OR 1.02–1.10, AUC 0.51) fall short of even one-tenth of this threshold. This gap is not caused by “insufficient statistical methods,” but jointly determined by theoretical ceilings and empirical effect sizes. Science can do is clarify the threshold itself—not assert that Yijing Studies is “impossible,” but provide a quantifiable standard that can be tested and refuted.

    Evidence and Limitations

    Verified: Genetic prediction discriminative power has an absolute theoretical upper limit determined by heritability and prevalence (PMID 22827772, PMID 20195508); PRS requires explaining most genetic variance and massive sample sizes to reach AUC 0.75 (PMID 23555274); under Decision Curve Analysis, “how high AUC is sufficient” depends on net benefit and cost structure rather than AUC itself (PMID 17099194, PMID 33676020); systematic overestimation evidence exists regarding AUC clumping at 0.7/0.8/0.9 thresholds in reports (PMID 37667344); reliability of AUC under rare events is determined by event counts rather than event rates (PMID 41224138).

    Partially Verified: AUC-OR conversion formulas have mature methodological support (Chinn et al. 2000), but actual discriminative power is also influenced by exposure distribution and prevalence, requiring case-specific calculation.

    To Be Verified: Whether birth time (hourly granularity) exposure effect sizes reach OR > 3 in any disease; whether correlations between multiple Yijing assertion dimensions allow “partial dimension compliance”; whether operationalized chart scores can bring NRI/net benefit increments above clinical model baselines (corresponding to backlog #419).

    Related Reading

    Evidence Level: A
    References: PMID 22827772, PMID 20195508, PMID 23555274, PMID 17099194, PMID 33676020, PMID 37667344, PMID 41224138


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

  • 出生季节效应的判别力:从统计显著到个体命运的鸿沟

    Evidence level: A

    出生季节效应在群体层面真实存在、稳定可复制,但效应量小到几乎无法用于判断任何一个人的命运——”统计显著”与”个体可判别”之间隔着一条巨大的鸿沟。

    正文

    命理体系(八字、紫微斗数)有一个最核心的假设:一个人出生的时刻,本身就携带了足以判定其一生吉凶祸福的信息。现代科学恰好有一个与之可对撞的研究领域——出生季节效应(season of birth effect):出生月份确实与精神分裂症、多发性硬化、代谢特征等数十种健康结局存在统计关联。那么关键问题来了:出生时刻如果真的携带信息,这个信息量有多大?够不够支撑”个体命运判定”?

    一、效应量级:真实但极小

    先看最可靠的证据。2023年一项覆盖43个国家与地区、纳入440,039名精神分裂症患者的系统综述与meta分析显示,冬季出生者的患病风险显著升高,但合并效应量仅为OR 1.05(95%CI 1.03–1.07);夏季出生者风险降低(OR 0.96)。这是目前全球最大规模的出生季节-精神疾病证据。更早的北半球meta(126,196名患者、8,660万对照出生人口)得到类似结果:冬春出生超额OR 1.07,但人群归因风险只有3.3%——即使把”冬春出生”视为暴露,它也只能解释精神分裂症约3%的病例,远低于家族史、城市出生地等因素。

    全表型扫描给出同样的图景。哥伦比亚大学团队对174.9万人的电子病历做系统扫描(SeaWAS),在1,688种疾病中找到55种与出生月份显著相关,但多数效应的相对风险仅在1.02–1.10之间徘徊。英国生物库(UK Biobank)50.2万人的分析同样显示,出生季节对基础代谢率、反应速度、睡眠等指标存在显著但微小的效应——2024年神经影像子研究里,夏季出生与复发性抑郁的关联系数β仅为0.026,作者直言”效应量过小,需在更大范围验证”。

    二、判别力:接近掷硬币

    效应量小意味着什么?把OR换算成判别力指标即可量化。对OR 1.05的暴露,其对应的AUC(ROC曲线下面积)大约只有0.51——满分1.0、随机0.5,出生季节对”这个人将来会不会得精神分裂症”的判别能力,只比掷硬币好了约1个百分点。即使按效应最强的纬度带(OR约1.28)计算,AUC也不超过0.56。

    判别力方法学对此早有成熟工具。净重分类改善指数(NRI)与综合判别改善指数(IDI)常被用来评估”新标记物加入模型后是否提升个体风险分层”,但Pencina等人的理论分析指出,这类指标对弱效应极其敏感,容易给出”显著改善”的假象;Pepe等人的模拟更警告,NRI即使在新标记毫无预测信息时也可能为正值。Kerr等人建议回归更朴素的标尺:净获益(net benefit)与真实/假阳性率。用这些更严格的尺度衡量,出生季节几乎不可能通过任何有临床意义的判别力门槛——因为门槛的另一端是”给个体贴风险标签”这一有代价的动作。

    三、脑影像的间接印证

    一个令人印象深刻的对照实验:Pantazatos用550例脑结构MRI尝试”仅凭大脑形态判断一个人的出生季节”。结果显示脑形态确实携带季节印记(女性整体分类准确率35%,显著高于25%的机会水平),但即便在最好区分的”冬季vs夏季”二元对比中,AUC也只有0.71——这是”群体印记真实存在、个体判别依然困难”的最直观例证。另一个角度:UK Biobank 13.6万人的基因分析显示,精神分裂症多基因风险评分与出生季节完全无关,说明出生季节效应是真实的环境印记、而非基因-环境混杂的假象——但它依然只是一个小效应。作为对照,同一样本中家族史/多基因评分这些”个体级”风险标记的效应量级远高于出生季节,这正是”群体印记”与”个体可判别”差异的最直接注脚。

    四、对命理体系的启示

    这些证据指向一个可以被命理界认真回应的量化问题:出生时刻的”信息量”确实存在(群体层面可复制),但量级决定了它无法支撑”个体命运判定”。把出生季节的判别力上限(AUC约0.51)与八字/紫微”精准判命”的宣称并置,鸿沟是显性的。科学能做的,一是把这个上限算清楚,二是给出一个可对话的坐标系:什么样的效应量级才能支撑个体级别的断言——那至少需要AUC达到0.7以上,而目前出生季节效应连0.55都够不着。

    证据与局限

    已验证: 出生季节-精神分裂症关联在44万人meta中稳定(OR 1.05);SeaWAS在174.9万人中发现55种疾病与出生月份相关;效应量换算的AUC增量不足0.02;NRI/IDI在弱效应下的夸大倾向已被方法学文献反复警示。

    部分验证: 脑结构MRI可超随机水平分类出生季节(女性35% vs 机会25%),但个体判别准确率仍低——”群体印记vs个体判别”的分离有实证;UK Biobank 13.6万人证实出生季节效应非基因-环境混杂。

    待验证: 出生季节在多大比例的疾病上能带来AUC/净获益的可检验增量;与家族史、遗传PRS等个体风险标记的判别力头对头对比;命理判定粒度与出生时刻信息量之间的信息论上界(互信息/可判别性)量化。

    相关阅读

    证据等级: A
    参考文献: PMID 36682315, PMID 14609251, PMID 26041386, PMID 27426413, PMID 26504496, PMID 38776272, PMID 29501066


    免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

    Discriminative Power of Season of Birth: The Chasm from Statistical Significance to Individual Fate

    Evidence level: A

    Birth season effects are real, stable, and replicable at the population level, yet their effect sizes are so small that they are practically useless for predicting the fate of any single individual—a vast chasm separates “statistical significance” from “individual discriminability.”

    Main Text

    Yijing Studies systems (such as Bazi and Zi Wei Dou Shu) rest on a core assumption: the exact moment of birth inherently carries sufficient information to determine an individual’s lifetime fortune and misfortune. Modern science possesses a corresponding research field to confront this hypothesis: the birth season effect, which shows statistical associations between birth month and dozens of health outcomes, including schizophrenia and multiple sclerosis. The critical question remains: if birth moments truly carry information, how much information do they hold? Is it sufficient to support “individual fate prediction”?

    I. Magnitude of Effect: Real but Negligible

    First, consider the most reliable evidence. A 2023 systematic review and meta-analysis covering 43 countries/regions and including 440,039 patients with schizophrenia found that winter-born individuals had a significantly elevated risk, yet the pooled effect size was only an Odds Ratio (OR) of 1.05 (95%CI 1.03–1.07); summer-born individuals showed reduced risk (OR 0.96). This represents the largest global dataset regarding birth season and mental illness. An earlier Northern Hemisphere meta-analysis (126,196 patients, 86.6 million birth controls) yielded similar results: winter/spring births showed an excess OR of 1.07, but the population attributable risk was only 3.3%—even treating “winter/spring birth” as an exposure, it accounts for only about 3% of schizophrenia cases, far below factors such as family history or urban birth location.

    Whole-genome phenotype scanning presents a similar picture. A team from Columbia University performed a systematic scan of electronic medical records for 1.749 million people (SeaWAS), finding 55 diseases significantly associated with birth month among 1,688 conditions, but most effect sizes hovered only between relative risks of 1.02 and 1.10. Similarly, an analysis of 502,000 individuals in the UK Biobank showed significant but minute effects of birth season on baseline metabolic rate, reaction time, and sleep. In a 2024 neuroimaging sub-study, the correlation coefficient (β) between summer birth and recurrent depression was merely 0.026, with authors explicitly stating, “The effect size is too small and requires validation in a larger scope.”

    II. Discriminative Power: Approaching Coin Tossing

    What does a small effect size imply? Converting ORs into discriminative metrics quantifies this. For an exposure with an OR of 1.05, the corresponding Area Under the Curve (AUC) is approximately 0.51—where a perfect score is 1.0 and random chance is 0.5. Thus, the discriminative ability of birth season for predicting “whether this individual will develop schizophrenia” is only about 1 percentage point better than flipping a coin. Even calculating based on the strongest latitude band (OR ~1.28), the AUC does not exceed 0.56.

    Methodological tools for discriminative power are well-established. The Net Reclassification Improvement (NRI) and Integrated Discrimination Improvement (IDI) are commonly used to assess whether adding new markers improves individual risk stratification; however, theoretical analyses by Pencina et al. indicate that these metrics are highly sensitive to weak effects and may falsely suggest “significant improvement.” Simulations by Pepe et al. further warn that NRI can yield positive values even when the new marker possesses no predictive information. Kerr et al. recommend more straightforward scales: net benefit and true/false positive rates. Measured against these stricter criteria, birth season almost certainly fails any clinically meaningful discriminative threshold—because the other side of that threshold involves the costly action of “labeling an individual with a risk.”

    III. Indirect Corroboration from Neuroimaging

    A striking counter-experiment involved Pantazatos attempting to “judge an individual’s birth season solely based on brain morphology” using 550 structural MRI scans. Results showed that brain morphology indeed carries seasonal imprints (overall classification accuracy for females was 35%, significantly higher than the 25% chance level), yet even in the best binary contrast (“winter vs. summer”), the AUC was only 0.71—this serves as the most direct evidence that “population imprints are real, but individual discrimination remains difficult.” From another angle, a genetic analysis of 136,000 individuals in the UK Biobank showed that polygenic risk scores for schizophrenia were completely unrelated to birth season, confirming that the birth season effect is a genuine environmental imprint rather than a gene-environment confounding artifact—yet it remains merely a small effect. As a control, within the same sample, the effect sizes of family history and polygenic scores (individual-level risk markers) were far higher than those of birth season; this is the most direct annotation of the difference between “population imprint” and “individual discriminability.”

    IV. Implications for Yijing Systems

    These evidence points to a quantifiable question that the Yijing community should seriously address: while an “information content” at birth exists (replicable at the population level), its magnitude determines that it cannot support “individual fate prediction.” Placing the upper limit of discriminative power for birth season (AUC ~0.51) alongside claims of “precise fate judgment” in Bazi/Zi Wei reveals an explicit chasm. Science can do two things: first, calculate this upper limit clearly; second, provide a dialogable coordinate system: what effect magnitude is required to support individual-level assertions—at least an AUC of 0.7 is needed, whereas current birth season effects fall short of even 0.55.

    Evidence and Limitations

    Verified: Association between birth season and schizophrenia is stable across a meta-analysis of 440,000 individuals (OR 1.05); SeaWAS identified 55 diseases associated with birth month in 1.749 million people; AUC increments from effect size conversions are insufficient (<0.02); the tendency of NRI/IDI to exaggerate weak effects has been repeatedly warned against in methodological literature.

    Partially Verified: Structural MRI can classify birth season above random levels (35% for females vs. 25% chance), yet individual discrimination accuracy remains low—”population imprint vs. individual discrimination” separation is empirically supported; UK Biobank data from 136,000 individuals confirms the birth season effect is not a gene-environment confounding artifact.

    To Be Verified: The proportion of diseases where birth season yields a testable increment in AUC/net benefit; head-to-head comparison of discriminative power against family history and genetic PRS; the information-theoretic upper bound (mutual information/discriminability) between Yijing judgment granularity and birth moment information content.

    Related Reading

    Evidence Level: A
    References: PMID 36682315, PMID 14609251, PMID 26041386, PMID 27426413, PMID 26504496, PMID 38776272, PMID 29501066


    Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.