五运六气的周期相位会漂移吗:如何用统计框架判定\u201c漂移多大才算真\u201d

Written by

in

Language

Evidence level: B

运气周期相位漂移检测的统计框架:分段相位角差异的置换检验与效应量阈值设定,回答”漂移多大才算显著”。

正文

五运六气理论认为,气候存在约五年、六年、十年等准周期振荡,这些振荡的”节奏”(相位)会随岁月推移而改变。现代气候研究也观察到类似的相位漂移:有研究用全球出生与死亡季节数据发现,随气候变暖,人类的季节生育与死亡节奏正发生系统性偏移(PMID 41648148)。问题在于——当我们在气象站数据里”看到”某个运气周期提前或推迟了几个月,这个漂移是真实的物理信号,还是纯属随机波动的巧合?

这正是统计检验要回答的问题。判定”漂移多大才算真”,需要一套由四道工序构成的框架。

第一道工序:把”相位”变成可比较的圆形变量。 周期的相位角是一个 0 到 360 度的圆形量,0 度和 360 度是同一个点。对圆形数据的统计分析不能套用普通的线性均值——比如 350 度和 10 度的平均值,线性算法会算出 180 度,而正确的圆形均值是 0 度。方法学文献明确指出,实施圆形数据统计需要专门的程序与检验(PMID 30100666)。因此,先要把变暖前/后两个时段的相位角分别提取出来,用圆形统计(圆形均值、平均合向量长度 R̄、Watson-Williams 检验)计算两段之间的相位差及其置信区间。这是整个框架的地基:相位差必须是圆形变量之差,而不是线性相减。

第二道工序:用置换检验/替代数据建立”无漂移”的零假设。 关键陷阱在于,气象时间序列高度自相关,普通的重排(把数据顺序彻底打乱)会破坏这种自相关结构,产生虚假的显著性。正确的做法是生成保留自相关结构的替代数据(surrogate data),再检验真实数据是否显著偏离替代数据的分布(PMID 29595296)。针对节律数据,有一类专门的多变量置换检验方法,直接对傅里叶变换后的数据做置换,能够在保持自相关的同时检验周期成分的变化(PMID 19212840)。这与脑电等生物节律研究中验证相位/谱成分显著性的思路一脉相承(PMID 35222076)。如果真实观测到的相位差落在替代数据分布的中心,说明这个”漂移”只是红噪声背景下的随机起伏;只有落在分布尾部,才谈得上统计显著。

第三道工序:用蒙特卡洛模拟设定效应量阈值——”漂移多大才算显著”。 零假设检验只能回答”有没有”,不能回答”多大才有意义”。这需要合成已知相位漂移量的时间序列(模拟数据),逐一施加不同幅度的相位偏移,跑置换检验,统计检验的检出率(功效),画出一条”真实漂移量 → 被检出的概率”的功效曲线。曲线会给出两个关键数字:最小可检出漂移(在这个样本量下能可靠分辨的最小相位偏移)和效应量阈值(比如”漂移超过 X 个月才算显著”)。没有这道工序,研究者很容易在短序列、高噪声的数据上把随机波动误读为运气周期在漂移。

第四道工序:多重比较校正。 气象站分析往往同时扫描多个周期带(5 年、6 年、10 年)、多个季节、多个站点,假设检验次数动辄成百上千。此时若不校正,假阳性会急剧膨胀。现代方法学共识是用错误发现率(FDR)控制替代 Bonferroni 式校正,前者在健康研究等多个领域被推荐为首选(PMID 24831050)。多周期带 × 多站点的相位差检验,必须做 FDR 校正,并在报告中明确报告校正后的 q 值。

把这四道工序连起来,就得到一个可复现的流水线:提取相位角(圆形)→ 替代数据置换检验(自相关保持)→ 蒙特卡洛功效标定(效应量阈值)→ FDR 多重比较控制。这个框架的最大价值在于它把”运气周期在漂移”从一句印象式的观察,变成一个有明确判定标准的统计断言。它也天然带有证伪机制:若某站点在替代数据检验下相位差不显著,或效应量小于标定阈值,我们就不能宣称该站的运气周期发生了漂移——哪怕小波谱上”看起来”有个峰在移动。

对五运六气研究者而言,这套框架的落地意味着一个纪律:先定阈值,再看数据。预先用合成数据标定”多小的漂移会被当成噪声”,就避免了事后挑站、挑周期带、挑季节的”钓鱼式”分析。这也呼应了传统术数”先立规矩,再断吉凶”的治学精神——判定标准先行,结论才有可辩护的基础。

证据与局限

已验证: 圆形数据的统计实施需要专门方法(PMID 30100666);替代数据置换检验能在保持自相关的条件下检验时间序列的统计显著性(PMID 29595296);针对节律数据存在专门的多变量置换检验(PMID 19212840);FDR 控制是多重比较校正的推荐首选(PMID 24831050);全球出生/死亡季节周期相位随气候偏移是真实可测现象(PMID 41648148);小波域替代数据可用于谱成分显著性检验(PMID 35222076)。

部分验证: 生物医学时间序列的同步性测量已有成熟统计体系(PMID 33754758);改进的替代数据方法已成功用于检测步态混沌等非线性动力学(PMID 16325825)——证明替代数据框架在真实生理数据上可落地,但直接用于气候”运气周期”相位漂移的功效标定尚未有公开发表的系统研究。

待验证: 中国多气象站长序列中 5 年/6 年运气周期相位漂移的站间一致性;分段相位差在 FDR 校正后的稳健性;最小可检出漂移量随序列长度的变化曲线。

相关阅读

证据等级: B
参考文献: PMID 19212840, PMID 24831050, PMID 29595296, PMID 16325825, PMID 33754758, PMID 30100666, PMID 41648148, PMID 35222076


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。

Will the Cycle Phase of Five Movements and Six Qi Drift: How to Determine with a Statistical Framework Whether the Degree of Drift Is Real

Evidence level: B

Statistical Framework for Detecting Phase Drift in Cyclical Phases: Permutation testing of segmented phase angle differences and setting effect size thresholds to address the question, “How much drift constitutes significance?”

Main Text

Yijing Studies theory posits that climate exhibits quasi-periodic oscillations with cycles of approximately five, six, or ten years. The “rhythm” (phase) of these oscillations shifts over time. Modern climate research also observes similar phase drifts: studies utilizing global birth and death season data have found that with climate warming, human seasonal rhythms of birth and death are undergoing systematic shifts (PMID 41648148). The core question remains: when we observe a specific Yijing cycle advancing or delaying by several months in meteorological station data, is this drift a genuine physical signal, or merely a coincidence of random fluctuation?

This is precisely the question statistical testing must answer. Determining “how much drift constitutes reality” requires a framework composed of four procedural steps.

Step 1: Convert “Phase” into a comparable circular variable. The phase angle of a cycle is a circular quantity ranging from 0 to 360 degrees, where 0 and 360 represent the same point. Statistical analysis of circular data cannot apply standard linear means; for instance, averaging 350 degrees and 10 degrees via linear algorithms yields 180 degrees, whereas the correct circular mean is 0 degrees. Methodological literature explicitly states that implementing circular data statistics requires specialized software and tests (PMID 30100666). Therefore, phase angles for pre-warming and post-warming periods must be extracted separately, and circular statistics (circular mean, mean resultant length $\bar{R}$, Watson-Williams test) must be used to calculate the phase difference and its confidence interval between the two periods. This forms the foundation of the entire framework: the phase difference must be the difference between circular variables, not linear subtraction.

Step 2: Establish a “no-drift” null hypothesis using permutation testing/surrogate data. A critical trap lies in the high autocorrelation of meteorological time series; standard shuffling (completely randomizing data order) destroys this autocorrelation structure, generating false significance. The correct approach is to generate surrogate data that preserves the autocorrelation structure, then test whether the real data significantly deviates from the distribution of the surrogate data (PMID 29595296). For rhythmic data, a specific class of multivariate permutation testing methods directly permutes data after Fourier transformation. This allows testing for changes in cyclic components while maintaining autocorrelation (PMID 19212840). This aligns with the logic used in biological rhythm research, such as EEG studies, to verify the significance of phase or spectral components (PMID 35222076). If the observed phase difference falls within the center of the surrogate data distribution, the “drift” is merely random fluctuation against a red noise background; only if it falls in the tail can statistical significance be claimed.

Step 3: Use Monte Carlo simulation to set an effect size threshold—”How much drift constitutes significance.” Null hypothesis testing can only answer “whether” there is a difference, not “how much constitutes significance.” This requires synthesizing time series with known phase drift magnitudes (simulated data), applying various amplitudes of phase shifts sequentially, running permutation tests, and calculating the detection rate (power) of the statistical test. A power curve plotting “true drift magnitude → probability of detection” should be generated. The curve provides two key numbers: the minimum detectable drift (the smallest phase shift reliably resolvable given the sample size) and the effect size threshold (e.g., “drift exceeding X months constitutes significance”). Without this step, researchers are prone to misinterpreting random fluctuations as Yijing cycle drifts in short sequences with high noise.

Step 4: Multiple comparison correction. Meteorological station analyses often simultaneously scan multiple cycle bands (5 years, 6 years, 10 years), multiple seasons, and multiple stations, resulting in hundreds or thousands of hypothesis tests. Without correction, false positives expand drastically. Modern methodological consensus recommends controlling the False Discovery Rate (FDR) rather than Bonferroni-style correction; the former is recommended as the primary choice in fields like health research (PMID 24831050). Phase difference tests across multiple cycle bands and stations must undergo FDR correction, and the corrected q-values must be explicitly reported in the findings.

Linking these four steps creates a reproducible pipeline: Extract Phase Angles (Circular) → Surrogate Data Permutation Testing (Preserve Autocorrelation) → Monte Carlo Power Calibration (Effect Size Threshold) → FDR Multiple Comparison Control. The greatest value of this framework lies in transforming the impressionistic observation that “Yijing cycles are drifting” into a statistically assertive claim with clear criteria. It also inherently carries a falsification mechanism: if the phase difference is not significant under surrogate testing at a specific station, or if the effect size is below the calibrated threshold, we cannot claim that the Yijing cycle at that station has drifted—even if a peak appears to be moving in the wavelet spectrum.

For researchers in Five Transmissions and Six Gases, implementing this framework implies a discipline: Set the threshold first, then examine the data. Pre-calibrating “how small a drift is treated as noise” using synthetic data avoids post-hoc “fishing expeditions” selecting specific stations, cycle bands, or seasons. This also resonates with the scholarly spirit of traditional numerology—”establish rules first, then judge auspiciousness”—where criteria precede conclusions, providing a defensible basis for results.

Evidence and Limitations

Verified: Implementation of circular data statistics requires specialized methods (PMID 30100666); surrogate data permutation testing can assess statistical significance in time series while preserving autocorrelation (PMID 29595296); multivariate permutation tests exist specifically for rhythmic data (PMID 19212840); FDR control is the recommended primary choice for multiple comparison correction (PMID 24831050); global birth/death seasonal cycle phases shifting with climate are a measurable phenomenon (PMID 41648148); surrogate data in the wavelet domain can be used for spectral component significance testing (PMID 35222076).

Partially Verified: Maturity exists for statistical systems measuring synchronicity in biomedical time series (PMID 33754758); improved surrogate data methods have successfully been applied to detect chaos in gait and other nonlinear dynamics (PMID 16325825)—proving the surrogate data framework is feasible for real physiological data, but systematic studies calibrating power specifically for climate “Yijing cycle” phase drifts have not yet been published.

To Be Verified: Inter-station consistency of 5-year/6-year Yijing cycle phase drifts in long sequences from multiple Chinese meteorological stations; robustness of segmented phase differences after FDR correction; the curve showing how the minimum detectable drift varies with sequence length.

Related Reading

Evidence Level: B
References: PMID 19212840, PMID 24831050, PMID 29595296, PMID 16325825, PMID 33754758, PMID 30100666, PMID 41648148, PMID 35222076


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.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *