Category: 科学易学

  • 身份层和状态层,谁的聚类结构更像「同一家人」:四路签名跨人一致性的分层裁决

    Evidence level: B

    把脉搏、心电、脑电里的参数分成「身份层」(基因底色深、跨年不变)和「状态层」(受当下状态调制、容易漂移),然后用同一个聚类结构检验去问:身份层的参数是不是真的”组内像一家人、组间分得开”,而状态层相反? ——这是把”望闻问切合参”从整体印象变成可分层的科学动作。

    正文

    为什么要分两层? 上一篇文章(#290,四路签名分层检验)提出了一个假设:人的生理信号里,低阶脉搏谐波、个体α频(IAF)、QRS 形态这类基因底色深厚的参数属于”身份层”,跨年稳定、能认出一个人;HRV、高阶谐波这类受年龄、情绪、疾病调制的参数属于”状态层”,随状态漂移、只适合读当下。这个假设听起来合理,但当时只验证了”单参数自己稳不稳”(重测信度),没有验证”参数们的相对结构”——也就是:同一层里的参数,是不是真的聚成一团、层与层之间是不是真的分得开。本文要裁决的就是这个缺口:身份层与状态层的”聚类结构”(同层内跨年相关 vs 层间差异)是否如假设所预言。

    “组内高、组间低”这个判据为什么关键? 一个参数如果真是”身份层”,它必须同时满足两个条件:同一个人今年测和明年测高度相关(组内一致),不同的人之间差异明显(组间可分)。这是聚类结构检验(轮廓系数、F 统计量、组内 ICC)的标准问题。反过来,一个”状态层”参数因为随当下状态漂移,同一个人跨年两次测量的相关会偏低,不同人的分布会互相重叠——组内不高、组间不分,聚类结构散架。所以”身份层聚类紧、状态层聚类散”一旦被数据证实,就等于给”分层假说”装上了骨架;如果相反(比如某些身份层参数其实组间也不分),假说就要打补丁。

    现有证据已经为”分层能成立”铺好了半条路。 身份层侧证据最扎实的是 IAF:237 对同卵 + 282 对异卵双胞胎研究显示 IAF 遗传度高达 0.81、alpha 功率 0.82,个体差异几乎全由加性遗传解释,非遗传方差全是测量误差(PMID 16338015)——遗传度这么高,意味着”组间差异”物理上就大,这是聚类可分的前提。更大规模的行为学证据也一致:20-31 岁与 65-80 岁两组健康人经过约 100 小时认知训练,IAF 纹丝不动,重测稳定性高到”测量误差解释不了”(PMID 23551082)。2024 年一项 MEG 静息谱研究把这事量化成数字:中位频率、IAF、谱边缘频率、谱熵四个参数的组内 ICC 高达 0.83-0.96(PMID 38249648)——“同一个人测 N 次像同一个人”这一半,身份层已经基本坐实。

    但状态层侧的证据恰恰是反过来的,这正是分层最有力的佐证。 HRV 的重测信度早已被系统性地证明”不稳定”:63 名健康成人中,ICC 随记录时长浮动,6 分钟频域指标刚过 0.7、2 分钟频域指标掉到 0.5 以下(PMID 15286929);体位影响更大,卧位 ICC 0.65-0.89、立位直接跌到 0.35-0.89(PMID 31411277)——同一个人的 HRV,换个姿势、换个时长就变样,这就是典型的”组内不一致”,与身份层的组内高相关形成鲜明对照。 更直接的反差来自 PPG 生物识别的长期追踪:同一天内识别错误率(EER)可低至 1%,但换一天再测,EER 最高飙到 23.2%(PMID 29751681)——“跨天就认不出”,正是状态层参数污染身份层识别的实锤。

    于是本文的裁决设计浮出水面: ① 用千人级跨年度队列,四路签名(脉搏谐波、ECG、IAF、可穿戴)每个参数算个体内跨年 ICC 与个体间差异(F 统计量);② 把参数按假设分进身份层/状态层两组,分别做聚类,算轮廓系数——看身份层的轮廓系数是否显著高于状态层,是否呈现”组内高、组间低”,而状态层相反甚至不显著;③ 检验层间聚类分离度(簇中心距 vs 簇内散布)是否支持”两层真的分得开”。裁决规则:若身份层轮廓系数显著高于状态层且两者方向符合假设 → 分层假说获结构性支持;若两者不分上下 → “分层”可能只是给参数贴标签,假说降级。

    一个必须提前防的坑: 聚类结构检验天然容易被”总方差大小”骗。状态层参数(如 HRV)绝对方差大,聚类时可能显得”散”,但那是量纲问题不是结构问题;所以必须先标准化、再用与量纲无关的指标(轮廓系数、F 比),否则会把”方差大”误读成”聚类散”。这也正是为什么这个检验不能省。

    证据与局限

    已验证: IAF 遗传度 0.81 且跨年龄稳定(PMID 16338015, 23551082);MEG 静息谱参数组内 ICC 0.83-0.96(PMID 38249648);HRV 重测信度低且受时长/体位调制(PMID 15286929, 31411277);PPG 识别跨天 EER 升至 23.2%(PMID 29751681)。

    部分验证: “身份层参数跨年聚类紧、状态层参数聚类散”这一结构命题本身尚无直接证据——现有文献只分别证明了”身份层自己稳”与”状态层自己不稳”,没有人把两层放进同一个聚类检验里比轮廓系数。

    待验证: ① 千人队列下身份层/状态层的轮廓系数差异是否显著且方向符合假设;② 层间聚类分离度在跨季节、跨时段的重复采样中是否稳定;③ 若聚类结构随年龄/疾病漂移,分层边界是否需要重标。

    相关阅读

    L1 网络状态:该主题(个体频率签名”身份层 vs 状态层”聚类结构)尚未进入 L1 对应网络——CORRESPONDENCE_NETWORK.md 边表仅登记 E-001~E-050(五行↔五脏/五色/五音/五方/五季/五志、八卦↔五行、天干地支↔五行、纳音↔频率),无任何以频率签名、身份层/状态层、ICC 或聚类结构为源或目标的边;EDGE_TABLE_science-yixue.md 仅有概念计数(频率 407 篇/A级70、HRV 214 篇/A级19、脉象 73 篇/A级5),无 L1/L2/L3 层级与 active/open/conflict 状态判定。已登记至 science-yixue-l1-gaps.md。

    证据等级: B
    参考文献: PMID 16338015, PMID 23551082, PMID 38249648, PMID 15286929, PMID 31411277, PMID 29751681


    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。

    Identity Layer and Status Layer, Whose Clustering Structure Resembles \”Same Family\”: Four-Way Signature Cross-Person Consistency Hierarchical Arbitration

    Evidence level: B

    Partitioning parameters from pulse, ECG, and EEG into an “identity layer” (deep genetic baseline, stable across years) and a “state layer” (modulated by current condition, prone to drift), then testing them against the same clustering structure, asks: Do identity-layer parameters truly cluster tightly within groups and separate between groups, while the state layer does the opposite? — This transforms the holistic “inspection, listening, questioning, and palpation” (望闻问切) from a general impression into a stratified scientific procedure.

    Body

    Why split into two layers? The previous article (#290, Stratified Validation of Four Modalities) proposed a hypothesis: physiological signals contain low-order pulse harmonics, individual alpha frequency (IAF), and QRS morphology—parameters with deep genetic underpinnings—belonging to the “identity layer”, stable across years and identifiable for a specific individual; whereas HRV and high-order harmonics—modulated by age, emotion, and disease—belong to the “state layer”, drifting with condition and suitable only for reading the present. This hypothesis sounds reasonable, but at that time it only validated “stability of single parameters” (test-retest reliability), not the “relative structure among parameters”—that is, whether parameters within the same layer truly cluster together and whether layers separate from one another. This article aims to arbitrate this gap: Do the “clustering structures” of the identity layer and state layer (intra-layer cross-year correlation vs. inter-layer difference) match the hypothesis prediction.

    Why is the criterion “high within-group, low between-group” critical? If a parameter truly belongs to the “identity layer”, it must satisfy two conditions simultaneously: high correlation between measurements of the same person across years (within-group consistency) and clear differentiation among different individuals (between-group separability). This is the standard question in clustering structure validation (silhouette coefficients, F-statistics, intra-group ICC). Conversely, because “state layer” parameters drift with current conditions, the correlation between cross-year measurements for the same person is low, and distributions of different individuals overlap—low within-group cohesion and poor between-group separation lead to a collapsed clustering structure. Therefore, once data confirms “tight clustering in the identity layer and scattered clustering in the state layer”, it provides the skeleton for the “stratification hypothesis”; if the opposite occurs (e.g., certain identity-layer parameters show no between-group separation), the hypothesis requires refinement.

    Existing evidence has laid half the groundwork for “stratification being feasible.” The strongest identity-layer evidence comes from IAF: studies of 237 monozygotic and 282 dizygotic twin pairs show IAF heritability reaching 0.81 and alpha power 0.82, with individual differences almost entirely explained by additive genetic factors, leaving non-genetic variance as measurement error (PMID 16338015)—such high heritability implies physically large “between-group differences”, a prerequisite for cluster separability. Larger-scale behavioral evidence also aligns: healthy adults aged 20–31 and 65–80 underwent ~100 hours of cognitive training, yet IAF remained unchanged, exhibiting high test-retest stability “beyond what measurement error could explain” (PMID 23551082). A 2024 MEG resting-spectrum study quantified this numerically: intra-group ICC for median frequency, IAF, spectral edge frequency, and spectral entropy reached 0.83–0.96 (PMID 38249648)—“the same person measured N times looks like the same person”—half of the identity layer is now firmly established.

    But the evidence for the state layer is precisely the reverse, which serves as the strongest corroboration of stratification. HRV test-retest reliability has long been systematically proven “unstable”: among 63 healthy adults, ICC fluctuates with recording duration; 6-minute time-domain metrics barely exceed 0.7, while 2-minute time-domain metrics drop below 0.5 (PMID 15286929); posture effects are even more pronounced, with supine ICC ranging 0.65–0.89 and standing ICC plummeting to 0.35–0.89 (PMID 31411277)—HRV of the same person changes with posture and duration, representing classic “within-group inconsistency,” sharply contrasting with the high within-group correlation of the identity layer. More direct contrast comes from long-term PPG biometric tracking: same-day error rate (EER) can be as low as 1%, but retesting the next day raises EER up to 23.2% (PMID 29751681)—“unrecognizable across days” is the smoking gun of state-layer parameters contaminating identity-layer recognition.

    Thus, the arbitration design emerges: ① Using a thousand-person cross-year cohort, compute intra-individual cross-year ICC and inter-individual differences (F-statistics) for each parameter across four modalities (pulse harmonics, ECG, IAF, wearable); ② Partition parameters into identity-layer and state-layer groups per hypothesis, perform clustering separately, calculate silhouette coefficients—check whether identity-layer silhouette coefficients are significantly higher than state-layer ones, exhibiting “high within-group, low between-group,” while the state layer shows the opposite or no significance; ③ Test inter-layer cluster separation (distance between cluster centers vs. within-cluster spread) to confirm “two layers truly separate.” Arbitration rule: if identity-layer silhouette coefficients are significantly higher than state-layer ones and both directions align with the hypothesis → stratification hypothesis gains structural support; if neither layer dominates → “stratification” may merely be labeling parameters, downgrading the hypothesis.

    A pitfall to preempt: clustering structure validation is inherently susceptible to deception by “total variance magnitude.” State-layer parameters (e.g., HRV) have large absolute variance and may appear “scattered” during clustering, but this is a scale issue, not a structural one; therefore, standardization must precede the use of scale-independent metrics (silhouette coefficient, F-ratio) to avoid misreading “large variance” as “scattered clustering.” This is precisely why this validation cannot be skipped.

    Evidence and Limitations

    Verified: IAF heritability 0.81 and cross-age stability (PMID 16338015, 23551082); MEG resting-spectrum parameters intra-group ICC 0.83–0.96 (PMID 38249648); HRV low test-retest reliability modulated by duration/posture (PMID 15286929, 31411277); PPG recognition cross-day EER rising to 23.2% (PMID 29751681).

    Partially Verified: The structural proposition itself—”identity-layer parameters cluster tightly across years, state-layer parameters scatter”—lacks direct evidence; existing literature only separately proves “identity layer is stable” and “state layer is unstable,” with no study placing both layers into the same clustering test to compare silhouette coefficients.

    To Be Verified: ① Whether identity-layer/state-layer silhouette coefficient differences are significant and directionally aligned with the hypothesis in the thousand-person cohort; ② Whether inter-layer cluster separation remains stable across seasonal/time-based repeated sampling; ③ Whether stratification boundaries need re-scaling if clustering structure drifts with age/disease.

    Related Reading

    L1 Network Status: This topic (clustering structure of “identity layer vs state layer” in individual frequency signatures) has not yet entered the corresponding L1 network—CORRESPONDENCE_NETWORK.md edge table only registers E-001~E-050 (Five Phases ↔ Five Organs/Five Colors/Five Tones/Five Directions/Five Seasons/Five Emotions, Eight Trigrams ↔ Five Phases, Heavenly Stems/Earthly Branches ↔ Five Phases, Na Yin ↔ Frequency), with no edges sourced or targeted by frequency signatures, identity layer/state layer, ICC, or clustering structure; EDGE_TABLE_science-yixue.md contains only concept counts (Frequency 407 articles/A-level 70, HRV 214 articles/A-level 19, Pulse 73 articles/A-level 5), lacking L1/L2/L3 level and active/open/conflict status determination. Registered in science-yixue-l1-gaps.md.

    Evidence Level: B
    References: PMID 16338015, PMID 23551082, PMID 38249648, PMID 15286929, PMID 31411277, PMID 29751681


    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.

  • 「等频率」到底等住了多少:伪呼吸与丹田呼吸的机械匹配残差,会不会把 ΔX5 整段传染

    Evidence level: C

    等频率对照只锁住了一个变量(呼吸频率),而潮气量、呼气末容积、腹式占比、吸气波形这四个维度一个都没锁——四维残差乘以各自的传导系数,就是「气感特异增量」里可能混进来的机械噪音。这个量级从未被标定过。

    正文

    研究「丹田呼吸有没有特异效应」的标准做法,是让受试者跟节拍器做一种胸式伪呼吸,把呼吸频率对齐到丹田呼吸的频率,然后用两组条件下电抗 X5(5 Hz 振荡法测出的呼吸系统电抗)的差值 ΔX5 当作「气感特异增量」。这个设计的全部底气,都押在「等频率」三个字上——但频率只是呼吸的诸多维度之一。一次呼吸至少有四个独立维度同时在场:潮气量(每次吸进多少)、呼气末容积(吸完剩下多少)、腹式占比(胸腹如何分工)、吸气波形(吸气的上升时间与形态)。「等频率」只锁住了周期,其余四维完全自由。若丹田呼吸与胸式伪呼吸在这四维上本来就差得很大,那么 ΔX5 里混进去的就不是「气感」,而是纯粹的机械状态差。

    这个担忧不是纸面上的。呼吸力学里有明确传导链:胸壁的弹性与阻力随呼吸频率和潮气量变化,肺本身只有在平均肺容积基本不变时才近似恒定(PMID 1731572);气道阻力与电抗都随肺容积改变,低于功能残气量(FRC)时频率依赖性会自发出现(PMID 7204163);高频段阻抗随肺容积下降(PMID 17943446)。物理模型实测显示,呼吸频率从 10 提到 30 次/分、潮气量从 0.5 提到 1.5 L 时,X5 单调抬高(PMID 25034109);COPD 患者把呼吸频率从 20 提到 40 时,呼气相电抗显著下降(PMID 25251948)。这些数字说明一件事:X5 是呼吸力学状态的灵敏探针,而呼吸力学状态恰恰是「等频率」对照没有锁住的那部分。

    2024 年的一项研究把这件事量化得更直接:12 名健康受试者在六个操作肺容积水平(从残气量 RV 到肺总量 TLC)上分别测 FOT 阻抗,结果 Xrs 对肺容积高度响应——从潮气量位过度膨胀到 TLC 时,5 Hz 电抗平均下降 0.56 cmH₂O·s/L(95%CI 0.25-0.86),从潮气量位过度缩小到 RV 时,5 Hz 电抗下降更多,达 2.53 cmH₂O·s/L(95%CI 1.04-4.02)(PMID 38036081)。注意这是同一个人的肺容积操作就能引起的电抗变化——如果丹田呼吸把呼气末容积抬高了几个百分点,而伪呼吸没有,这个差就会原样流进 ΔX5。

    更有意思的是呼吸模式本身对阻抗就有主效应。2024 年一项 69 名耐力运动员的研究发现,膈式(腹式主导)呼吸模式的人,其 5 Hz 吸气相电抗 X5%pred 显著低于胸式呼吸模式的人,且膈式模式的肺功能测试指标整体更好(PMID 38212427)。这直接说明:「腹式 vs 胸式」不是一个可以忽略的细节,它本身就是会改变 X5 的机械变量。丹田呼吸本质上是深度腹式呼吸,胸式伪呼吸本质上是胸式呼吸——两者在「腹式占比」这个维度上的差异,很可能远大于「等频率」所锁住的差异。

    那要怎么办?答案是两段式标定。第一步测匹配残差:让同一批有经验的练习者分别做丹田呼吸和跟节拍器等频率的胸式伪呼吸,用呼吸带加电阻抗断层成像(EIT)同步记录每一次呼吸的潮气量、呼气末容积、腹式占比、吸气上升时间,得到四维残差的分布和相关矩阵。第二步估传染系数:在同一批人身上,做「频率锁定、只系统改变单一维度」的操纵——潮气量 ±20%、呼气末容积 ±5% TLC、腹式占比 ±15 个百分点——回归出每个维度对 X5/AX 的偏效应,然后用第一步的残差乘第二步的系数,外推「预期机械泄漏幅度」。

    判据分四档。若外推泄漏幅度小于预期组间差的三分之一,说明等频率伪呼吸作为对照可接受,原设计成立——因为受试者跟节拍器呼吸时潮气量往往会自然收敛(这正是节拍器的一个隐藏好处);若泄漏幅度与组间差同量级,则 ΔX5 可能完全由机械残差解释,「气感特异增量」的名头必须摘掉,改走四维逐点匹配或容积夹持/体描箱定容积条件;若某一维度单独贡献超过总泄漏的一半,协议只需锁住那一个维度即可,不必四维全配;若四维残差之间高度相关(共线),则「逐维匹配」在操作上不可实现,课题只能改判为「自由呼吸下无法对齐机械条件」。

    这个课题没有现成的标准答案,但它有一个非常清晰的量纲逻辑:X5 的单位是压力/流量,机械泄漏的单位也是压力/流量,两者可以直接比大小。只要把残差和系数都测出来,ΔX5 里有多少是机械的、有多少可能是「剩下的东西」,就是一道算术题。至于「剩下的东西」是不是气感,那是下一步的问题——但在此之前,我们连「等频率」到底等住了多少都不知道。

    L1 网络状态:该主题尚未进入 L1 对应网络。CORRESPONDENCE_NETWORK.md 节点表中「呼吸率 / 呼吸模式」仅为生理侧观测节点,边表中无任何以「伪呼吸 vs 丹田呼吸的机械匹配残差 / 四维呼吸状态(潮气量、呼气末容积、腹式占比、吸气波形)对齐 / 电抗 X5 机械泄漏传染」为源或目标的边。该边属于「丹田 ↔ 呼吸频率」方法学前置(第 0 层),与 E-001~E-005(五行↔五脏骨架边)及呼吸相关概念边均无对应,证据等级受限于 C。

    证据与局限

    已验证: 呼吸系统阻抗(阻力与电抗)强烈依赖肺容积与呼吸频率——胸壁弹性阻力随频率与潮气量变化(PMID 1731572);电抗在低于 FRC 时出现频率依赖(PMID 7204163);高频阻抗随肺容积下降(PMID 17943446);物理模型显示频率 10→30、潮气量 0.5→1.5 L 时 X5 单调抬高(PMID 25034109);COPD 患者频率 20→40 时呼气相电抗显著下降(PMID 25251948);2024 年研究量化了同一人六个操作肺容积下 Xrs 的变化量级(PMID 38036081);膈式 vs 胸式呼吸模式对 X5%pred 有显著主效应(PMID 38212427);节拍器引导的自愿呼吸会系统性改变胸腹协调动力学(PMID 35918415)。「等频率」只锁频率、不锁其余四维,这一点在呼吸力学上是确定的。

    部分验证: 丹田呼吸与胸式伪呼吸在四维(潮气量/呼气末容积/腹式占比/吸气波形)上的实际匹配残差量级未被直接测量过——现有证据只证明「这些维度会改变 X5」的传导系数存在,不证明「丹田 vs 伪呼吸在这几个维度上具体差多少」。ΔX5 的机械泄漏幅度(残差 × 系数)属于推断而非实测。

    待验证: ① 四维匹配残差在 ≥24 名练习者身上的实测分布与相关矩阵;② 各维度(VT ±20%、EELV ±5% TLC、腹式占比 ±15pp)对 X5/AX 的偏效应回归系数;③ 外推泄漏幅度与预期组间差的比例(<1/3 可接受 / 同量级须改设计 / 单维过半只锁该维 / 四维共线则改判);④ 「等频率」对照在自由呼吸下能否真正对齐机械条件(若四维残差高度相关则不可行)。

    相关阅读

    证据等级: C
    参考文献: PMID 1731572, PMID 7204163, PMID 17943446, PMID 25034109, PMID 25251948, PMID 6653086, PMID 38036081, PMID 38212427, PMID 35918415


    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。

    How much \”equal frequency\” actually equalizes: whether the mechanical matching residual between pseudo-breathing and dantian breathing will infect the entire ΔX5 segment

    Evidence level: C

    Equal-frequency control locks only one variable (breathing rate), while tidal volume, end-expiratory volume, abdominal contribution ratio, and inspiratory waveform—four dimensions—remain completely unconstrained. Multiplying these four-dimensional residuals by their respective transmission coefficients yields the mechanical noise potentially mixed into the “Qi-sensation specific increment.” This magnitude has never been calibrated.

    Main Text

    The standard approach to investigating “whether abdominal breathing has a specific effect” involves having subjects perform chest-based pseudo-breathing synchronized with a metronome, aligning the breathing rate to match abdominal breathing, and then using the difference in electrical reactance X5 (measured via 5 Hz oscillation) between the two conditions, denoted as ΔX5, as the “Qi-sensation specific increment.” The entire rationale for this design rests solely on the phrase “equal frequency“—yet frequency is only one of many dimensions of breathing.

    A single breath involves at least four independent dimensions simultaneously present: tidal volume (how much is inhaled each time), end-expiratory volume (what remains after inhalation), abdominal contribution ratio (how the chest and abdomen divide labor), and inspiratory waveform (the rise time and shape of inhalation). “Equal frequency” locks only the cycle; the other four dimensions remain entirely free. If abdominal breathing and chest-based pseudo-breathing differ significantly across these four dimensions, then what mixes into ΔX5 is not “Qi-sensation” but pure mechanical state difference.

    This concern is not merely theoretical. Respiratory mechanics has clear transmission chains: chest wall elasticity and resistance vary with breathing rate and tidal volume; lung impedance itself remains approximately constant only when average lung volume is essentially unchanged (PMID 1731572); airway resistance and reactance both change with lung volume, with frequency dependence spontaneously appearing below functional residual capacity (FRC) (PMID 7204163); and high-frequency impedance decreases as lung volume drops (PMID 17943446). Physical model measurements show that when breathing rate increases from 10 to 30 breaths per minute and tidal volume from 0.5 to 1.5 L, X5 rises monotonically (PMID 25034109); COPD patients exhibit a significant decrease in expiratory-phase reactance when breathing rate increases from 20 to 40 (PMID 25251948). These figures demonstrate one fact: X5 is a sensitive probe of respiratory mechanical state, and this mechanical state is precisely the portion not locked by “equal-frequency” control.

    A 2024 study quantified this matter more directly: 12 healthy subjects had FOT impedance measured at six operational lung volume levels (from residual volume RV to total lung capacity TLC), revealing that Xrs responds highly to lung volume—from excessive inflation at tidal volume levels to TLC, the 5 Hz reactance averaged a decrease of 0.56 cmH₂O·s/L (95%CI 0.25-0.86); from excessive reduction at tidal volume levels to RV, the 5 Hz reactance decreased more, reaching 2.53 cmH₂O·s/L (95%CI 1.04-4.02) (PMID 38036081). Note that this is electrical reactance change caused by manipulating lung volume in the same individual—if abdominal breathing elevates end-expiratory volume by several percentage points while pseudo-breathing does not, this difference flows directly into ΔX5.

    More interestingly, breathing pattern itself has a main effect on impedance. A 2024 study of 69 endurance athletes found that individuals using diaphragmatic (abdominal-dominant) breathing patterns had significantly lower 5 Hz inspiratory-phase reactance X5%pred compared to chest-dominant breathing patterns, and their pulmonary function test metrics were overall better (PMID 38212427). This directly demonstrates: “Abdominal vs. Chest” is not a negligible detail; it itself is a mechanical variable that alters X5. Abdominal breathing is essentially deep diaphragmatic breathing, while chest-based pseudo-breathing is essentially chest breathing—the difference between them in the “abdominal contribution ratio” dimension may far exceed the difference locked by “equal frequency.”

    So what is the solution? The answer is two-stage calibration. First, measure matching residuals: have the same group of experienced practitioners perform both abdominal breathing and metronome-synchronized chest-based pseudo-breathing, using respiratory belts combined with electrical impedance tomography (EIT) to simultaneously record tidal volume, end-expiratory volume, abdominal contribution ratio, and inspiratory rise time for each breath, thereby obtaining the distribution and correlation matrix of the four-dimensional residuals. Second, estimate transmission coefficients: on the same group, perform manipulation where “frequency is locked but only one dimension is systematically altered”—tidal volume ±20%, end-expiratory volume ±5% TLC, abdominal contribution ratio ±15 percentage points—regressing out the partial effects of each dimension on X5/AX, then multiplying the residuals from Step One by these coefficients to extrapolate the “expected mechanical leakage amplitude.”

    Judgment criteria are divided into four tiers. If the extrapolated leakage amplitude is less than one-third of the expected inter-group difference, equal-frequency pseudo-breathing as control is acceptable and the original design holds—because subjects breathing with a metronome naturally converge tidal volume (this is actually a hidden benefit of the metronome); if the leakage amplitude is comparable in magnitude to the inter-group difference, ΔX5 may be entirely explained by mechanical residuals, and the title “Qi-sensation specific increment” must be removed, shifting instead to four-dimensional point-by-point matching or volume clamping/body plethysmography for fixed volume conditions; if one dimension alone contributes more than half of the total leakage, the protocol need only lock that single dimension without requiring all four to match; if the four-dimensional residuals are highly correlated (collinear), “dimension-by-dimension matching” becomes operationally impossible, and the study must be reclassified as “unable to align mechanical conditions under free breathing.”

    This topic has no ready-made standard answer, but it possesses a very clear dimensional logic: X5 units are pressure/flow, mechanical leakage units are also pressure/flow, so they can be directly compared. Once both residuals and coefficients are measured, how much of ΔX5 is mechanical and how much might be “the remaining stuff” becomes a simple arithmetic problem. Whether “the remaining stuff” constitutes Qi-sensation is the next question—but before that, we do not even know how much “equal frequency” actually locks.

    L1 Network Status: This topic has not yet entered the L1 corresponding network. In the CORRESPONDENCE_NETWORK.md node table, “breathing rate / breathing pattern” are merely physiological-side observation nodes, with no edges in the edge table originating from or targeting “mechanical matching residuals of pseudo-breathing vs. abdominal breathing / four-dimensional respiratory state (tidal volume, end-expiratory volume, abdominal contribution ratio, inspiratory waveform) alignment / electrical reactance X5 mechanical leakage transmission.” This edge belongs to the “Abdominal ↔ Breathing Rate” methodological prerequisite (Layer 0), and has no correspondence with E-001~E-005 (Five Phases ↔ Five Organs skeleton edges) or breathing-related concept edges; evidence level is constrained to C.

    Evidence and Limitations

    Verified: Respiratory system impedance (resistance and reactance) strongly depends on lung volume and breathing rate—chest wall elastic resistance varies with frequency and tidal volume (PMID 1731572); reactance exhibits frequency dependence below FRC (PMID 7204163); high-frequency impedance decreases with lung volume reduction (PMID 17943446); physical models show X5 rises monotonically when frequency increases from 10→30 and tidal volume from 0.5→1.5 L (PMID 25034109); COPD patients show significant decrease in expiratory-phase reactance when frequency increases from 20→40 (PMID 25251948); a 2024 study quantified the magnitude of Xrs change across six operational lung volumes in the same individual (PMID 38036081); diaphragmatic vs. chest breathing patterns have significant main effects on X5%pred (PMID 38212427); metronome-guided voluntary breathing systematically alters chest-abdominal coordination dynamics (PMID 35918415). “Equal frequency” locks only rate, not the other four dimensions—a fact confirmed by respiratory mechanics.

    Partially Verified: The actual magnitude of matching residuals between abdominal breathing and chest-based pseudo-breathing across the four dimensions (tidal volume / end-expiratory volume / abdominal contribution ratio / inspiratory waveform) has not been directly measured—existing evidence only confirms that “these dimensions affect X5” transmission coefficients exist, but does not prove “how much abdominal breathing differs from pseudo-breathing on these specific dimensions.” The mechanical leakage amplitude of ΔX5 (residuals × coefficients) is inferred, not directly measured.

    To Be Verified: ① Measured distribution and correlation matrix of four-dimensional matching residuals across ≥24 practitioners; ② Partial effect regression coefficients for each dimension (VT ±20%, EELV ±5% TLC, abdominal contribution ratio ±15pp) on X5/AX; ③ Ratio of extrapolated leakage amplitude to expected inter-group difference (<1/3 acceptable / same magnitude requires design revision / single dimension >50% lock only that dimension / four dimensions collinear then reclassify); ④ Whether “equal frequency” control can truly align mechanical conditions under free breathing (impossible if four-dimensional residuals are highly correlated).

    Further Reading

    Evidence Level: C
    References: PMID 1731572, PMID 7204163, PMID 17943446, PMID 25034109, PMID 25251948, PMID 6653086, PMID 38036081, PMID 38212427, PMID 35918415


    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.

  • 哪三个指标归哪个构造:六指标 3+3 切分的因子归属稳定性,与换一组指标就换一个答案的风险

    Evidence level: C

    六个呼吸指标要拆成「腹式化」与「协调化」两组,但「哪三个指标归哪一组」这个归属从未被检验过——探索性因子分析在小样本下会把归属跳来跳去,换一组指标就可能换一个答案。

    正文

    上一篇文章指出,「腹式化」和「协调化」这两个词,可能只是给两组参数起的名字,而不是发现了两个机制。当时给出的出路是:把指标从四个扩到六个(3+3),让模型自由度从 1 升到 8,好让「两个构造」的假设可以被检验。但这条路默认了一个从未被验证的前提——哪三个指标归哪个构造。

    这个归属现在只有两条软线索。一是「同源」:%AB(腹式化指标之一)与 CoV(协调化指标之一)都可由躯干运动解释;二是「同语义」:φ(相位离散度)与 X5 都带着「协调」的字眼。两条线索都不构成证据。「同源」说的是测量通道一样,不等于两个指标测的是同一个构造;「同语义」更是循环论证——先按名字分组,再用分组结果证明名字有意义。

    问题的核心是:探索性因子分析(EFA)在指标少、载荷相近时,因子归属本身就不稳定。六个变量、两个因子,旋转后的归属会随样本和初始值跳。这不是理论上的担心,而是有实测支撑的事实。一项针对 EFA 小样本行为的模拟研究发现,心理学期刊里 40%–60% 的应用研究样本量在 200 人以内,而小样本下 EFA 的因子载荷估计会出现严重偏倚(PMID 27929657)。另一项比较研究进一步显示,在样本量小、因子数少时,不同估计方法对因子结构的恢复能力差异巨大,普通最小二乘在「因子数少」的条件下优于正则化方法,但没有任何方法能保证小样本下的归属稳定(PMID 23541772)。

    更深的麻烦在于「未建模复杂度」。模拟研究表明,当数据里存在未建模的交叉载荷或相关残差时,模型之间的可分辨性会随未建模复杂度的上升而急剧下降(PMID 29230085);两个看似可分辨的层级模型,其可比性只在载荷成比例时成立,一旦比例被违反,即使两个模型拟合都好,它们的因子分数相关性也不高(PMID 35677113)。把这些结论翻译到丹田参数上:如果 %AB 和 CoV 共享躯干运动这个来源、φ 和 X5 共享「协调」这个语义,那么「按测量通道分」和「按语义分」两套真值结构在 EFA 眼里可能长得一模一样——你按哪个规则切,EFA 就给你哪个答案,而数据本身无法裁决。

    「每因子至少三个指标」也不构成安慰。三个指标只是模型可估的下限,不是归属稳健的保证——纵向因子分析研究明确指出,每因子三指标只是识别的最低要求,样本一波动、载荷一接近,归属就会翻转(PMID 41511670)。

    所以需要做的,是一次双轨判定。仿真轨道:设定两套真值结构(「按测量通道分」= 三个 RIP 指标一因子、三个 EIT/FOT 指标一因子;「按语义分」= 腹式化三项、协调化三项),在载荷 {0.5, 0.7}、相关 {0.3, 0.6}、样本量 {120, 300, 800} 的组合下各生成至少 500 次,跑两因子 promax 旋转的 EFA,看归属正确率。实测轨道:用真实六参数数据做 bootstrap 至少 1000 次,每次重跑 EFA,记录每个指标「落在同一因子」的次数比。

    判定标准很直接:若实测归属一致率低于 0.75,或仿真在样本量 120 时归属正确率低于 0.6,那么「六个指标分两组」这个前提不成立,加指标这条路作废,必须改走多质多法——每个构造各由两种方法测,而不是在同一台仪器上加同方法的指标。若归属一致率高于 0.9 且与「按测量通道分」一致,则问题从「构造成员」降级为「方法因子」,「腹式化」「协调化」这两个词应直接改写为「RIP 通道量」「非 RIP 通道量」。若归属稳定但与「按语义分」一致,则两个构造的说法可以保留,但必须声明它与测量通道正交。

    这个判定的意义在于:它把「命名」和「发现」分开。如果六指标方案下归属一致率低,「腹式化」「协调化」就只是命名,丹田呼吸研究里所有基于这两个词的结论都要重新标注;如果归属稳定且与测量通道一致,那么这两个词掩盖的其实是「一台仪器的两种读法」,后续研究应该按通道重新组织变量;只有当归属稳定且与语义一致时,「两个构造」才第一次有了数据依据。

    L1 网络状态:该主题尚未进入 L1 对应网络——CORRESPONDENCE_NETWORK 边表止于 E-050(「六十甲子纳音 ↔ 60 个频率」,L1、open,现代证据栏原文「尚无任何可测来源」);节点表生理侧仅登记「呼吸率 / 呼吸模式|可穿戴」,无「腹式化 / 协调化 / 因子归属 / 构造效度」任何字段;第 45 条缺口(母文 #1189)已登记「构造层」整体缺席,本条为其深化:即便补上构造层,仍缺「构造成员的归属依据与稳定性检验」这一整条边。

    证据与局限

    已验证:EFA 在样本量 200 以内时载荷估计严重偏倚(PMID 27929657);小样本下不同估计方法的因子结构恢复能力差异大(PMID 23541772);未建模复杂度上升使模型可分辨性下降(PMID 29230085);低阶因子与 bifactor 特异因子的可比性只在载荷成比例时成立(PMID 35677113);每因子三指标只是可估下限而非稳健保证(PMID 41511670)。

    部分验证:丹田六参数的「按测量通道分」与「按语义分」两套归属规则在 EFA 下可能不可分辨——这个推论建立在文献关于未建模复杂度的模拟之上,尚未在真实丹田数据上直接检验。

    待验证:① 仿真轨道中「按测量通道分」与「按语义分」两套真值结构在载荷 {0.5, 0.7}、相关 {0.3, 0.6}、N {120, 300, 800} 下的归属正确率差异;② 实测轨道中六参数 bootstrap 的归属一致率是否达到 0.75 门槛;③ 若归属稳定且与测量通道一致,「腹式化 / 协调化」改写为「RIP 通道量 / 非 RIP 通道量」后,既往基于这两个词的结论需要重标哪些。

    相关阅读

    证据等级: C
    参考文献: PMID 27929657, PMID 23541772, PMID 29230085, PMID 35677113, PMID 41511670


    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。

    Which three indicators belong to which construct: six indicators split 3+3 for factor assignment stability, and the risk of changing the answer when swapping to another set of i…

    Evidence level: C

    Six respiratory indicators must be split into two groups: “abdominalization” and “coordination.” However, the assignment of “which three indicators belong to which group” has never been tested. Exploratory Factor Analysis (EFA) on small samples causes factor assignments to jump around; changing the set of indicators may yield a different answer.

    Main Text

    The previous article pointed out that the terms “abdominalization” and “coordination” might simply be names given to two sets of parameters, rather than discoveries of two mechanisms. The proposed solution at the time was: expand the indicators from four to six (3+3), increasing the model degrees of freedom from 1 to 8, so that the hypothesis of “two constructs” could be tested. However, this path assumes a premise that has never been verified: which three indicators belong to which construct.

    Currently, this assignment relies on only two soft clues. First is “same origin”: %AB (one of the abdominalization indicators) and CoV (one of the coordination indicators) can both be explained by trunk motion. Second is “same semantics”: φ (phase dispersion) and X5 both carry the word “coordination.” Neither clue constitutes evidence. “Same origin” means the measurement channels are identical, but it does not mean the two indicators measure the same construct; “same semantics” is circular reasoning—grouping by name first, then using the grouping result to prove the names are meaningful.

    The core issue is that EFA, when faced with few indicators and similar loadings, produces inherently unstable factor assignments. With six variables and two factors, the rotated assignment jumps depending on the sample and initial values. This is not just theoretical concern; it is supported by empirical findings. A simulation study on EFA behavior with small samples found that 40%–60% of applied studies in psychology journals have sample sizes under 200, and under such conditions, EFA factor loading estimates suffer severe bias (PMID 27929657). Another comparative study further showed that with small samples and few factors, different estimation methods vary greatly in their ability to recover the factor structure; ordinary least squares performs better than regularized methods when the number of factors is small, but no method may support stable assignment under small samples (PMID 23541772).

    A deeper complication lies in “unmodeled complexity.” Simulation research indicates that when data contains unmodeled cross-loadings or correlated residuals, model distinguishability drops sharply as unmodeled complexity increases (PMID 29230085). Two seemingly distinguishable hierarchical models are comparable only when loadings are proportional; once proportionality is violated, even if both models fit well, their factor score correlations remain low (PMID 35677113). Translating these conclusions to dantian parameters: if %AB and CoV share the trunk motion source, and φ and X5 share the “coordination” semantic, then in EFA’s view, the two true structures—”grouped by measurement channel” and “grouped by semantics”—may look identical. Whichever rule you use to split, EFA gives you whichever answer; the data itself cannot adjudicate.

    The rule “at least three indicators per factor” also fails to provide comfort. Three indicators represent only the lower bound for a factor to be estimable, not a may support of robust assignment. Longitudinal factor analysis research explicitly states that three indicators per factor is merely the minimum requirement for identification; any fluctuation in sample or closeness in loadings can flip the assignment (PMID 41511670).

    Therefore, what is needed is a dual-track determination.
    – Simulation track: Set up two true structures (“grouped by measurement channel” = three RIP indicators per factor, three EIT/FOT indicators per factor; “grouped by semantics” = three abdominalization indicators, three coordination indicators). Generate at least 500 runs under combinations of loadings {0.5, 0.7}, correlations {0.3, 0.6}, and sample sizes {120, 300, 800}, performing two-factor promax rotation EFA to check assignment accuracy.
    – Empirical track: Apply bootstrap resampling at least 1,000 times on real six-parameter data, re-running EFA each time, and record the proportion of times each indicator falls into the same factor.

    The criteria are straightforward: if the empirical assignment consistency rate falls below 0.75, or the simulation accuracy at sample size 120 drops below 0.6, then the premise of “six indicators split into two groups” fails; adding indicators is futile, and we must shift to multi-method multi-construct: each construct measured by two different methods, rather than adding indicators of the same method to one instrument. If assignment consistency exceeds 0.9 and aligns with “grouped by measurement channel,” the problem degrades from “construct membership” to “method factor”; the terms “abdominalization” and “coordination” should be directly rewritten as “RIP channel quantity” and “non-RIP channel quantity.” If assignment is stable but aligns with “grouped by semantics,” the claim of “two constructs” can be retained, but it must be declared orthogonal to measurement channels.

    The significance of this determination lies in separating “naming” from “discovery.” If assignment consistency under the six-indicator scheme is low, then “abdominalization” and “coordination” are merely labels; all conclusions in dantian breathing research based on these terms must be relabeled. If assignment is stable and matches measurement channels, then these terms mask “two readings of one instrument”; future studies should reorganize variables by channel. Only when assignment is stable and matches semantics do we have the first data-backed evidence for “two constructs.”

    L1 Network Status: This topic has not yet entered the L1 Correspondence Network—the CORRESPONDENCE_NETWORK edge table stops at E-050 (“60 Jiazi Na Yin ↔ 60 frequencies,” L1, open, original text of modern evidence column states “no measurable source yet“); the physiological side node table registers only “respiration rate / respiration pattern | wearable,” with no fields for “abdominalization / coordination / factor assignment / construct validity”; Gap #45 (main text #1189) has been logged as “construct layer entirely missing,” and this entry deepens it: even if the construct layer is filled, the entire edge of “assignment basis and stability test for construct members” remains absent.

    Evidence and Limitations

    Verified: EFA loading estimates suffer severe bias when sample size is under 200 (PMID 27929657); different estimation methods vary greatly in recovering factor structure under small samples (PMID 23541772); rising unmodeled complexity reduces model distinguishability (PMID 29230085); comparability between low-order factors and bifactor specific factors holds only when loadings are proportional (PMID 35677113); three indicators per factor is merely the minimum for estimation, not a robust may support (PMID 41511670).

    Partially Verified: The two assignment rules—”grouped by measurement channel” and “grouped by semantics”—for the dantian six parameters may be indistinguishable under EFA. This inference rests on simulations of unmodeled complexity from the literature and has not yet been directly tested on real dantian data.

    To Be Verified: ① the difference in assignment accuracy between the two true structures (“grouped by measurement channel” vs. “grouped by semantics”) under loadings {0.5, 0.7}, correlations {0.3, 0.6}, and N {120, 300, 800} in the simulation track; ② whether the empirical bootstrap assignment consistency rate reaches the 0.75 threshold; ③ if assignment is stable and matches measurement channels, which prior conclusions based on “abdominalization” and “coordination” must be relabeled after rewriting them as “RIP channel quantity” and “non-RIP channel quantity.”

    Related Reading

    Evidence Level: C
    References: PMID 27929657, PMID 23541772, PMID 29230085, PMID 35677113, PMID 41511670


    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.

  • 两个读数之间的误差相关到底有多大:扰动臂实验把责任指向定标还是算法

    Evidence level: C

    同一根呼吸曲线、同一个容积轴,两次读数之间的误差相关 ρ 从未在本领域被测过——它的大小、符号和方向,决定了整套「余量」终点的最小可检测变化是放宽还是收严。

    正文

    在呼吸功能检测里,「余量」这个终点是呼气末容积(EELV)与肺活量变化(DR2)两个读数的差。上一篇文章(《余量的最小可检测变化,一直按两分量独立在算》)已经证明:一旦两个读数的测量误差相关(ρ≠0),独立假设就会把最小可检测变化高估 1.58 倍到 4.47 倍(ρ=0.6 到 0.95);反过来,若 ρ 为负,名义 5% 的假阳性率会涨到 8.6%–11%。换句话说,整道判据门槛都挂在 ρ 这一个数上。

    可问题是:这个 ρ 在本领域从来没有人测过。「同一条 SVC-FOT 曲线、同一个容积轴」这条设计理由,只能说明 ρ 应当为正,给不出量级;而 ρ 的符号和大小恰好决定结论方向——ρ>0 意味着独立假设高估门槛(自罚),ρ<0 意味着低估门槛(自宽)。更麻烦的是,ρ 越大越估不准:ρ=0.9 时估偏 0.05 就能让 MDC95 差 22%,ρ=0.95 时差 41%;样本量 n=30 时 ρ=0.8 的 95% 置信区间半宽约 0.14,映射到 MDC95 上上下相差 2.29 倍。所以「估出 ρ」这件事本身需要一份精度预算,不能只报一个点估计。

    本主题给出的方案是:用扰动臂实验把「ρ 归因」拆开。第一步,取 ≥30 名健康成人,隔天各测 5 次 SVC-FOT,同一算法逐次同时检出 DR2 与呼气末容积(均折算为 %TLC),把每次的两个读数配成对,用两次测量之差的 2×2 协方差阵直接估计 s₁、s₂ 与 ρ,并给出 ρ 的 Fisher z 变换 95% 置信区间。配对测量误差的完整表达式必须含 2ρs₁s₂ 这一项——这是经典测量学早已给出的框架(Bruynesteyn 在类风湿关节进展判定中明确提出:分数统计上的相依不等于误差上的相关,两者要分开处理,PMID 15286006);分量层面的 s₁、s₂ 与 ICC 则由标准重测信度框架给出(Weir 的 ICC+SEM 框架,PMID 15705040);每个个体多次观测的一致性评估方法(Bland–Altman 扩展)则提供了「把重复测量之差当新变量」的统计出处(PMID 17613642)。

    第二步是关键:设两个扰动臂。(a) 换容积轴定标方法——例如以已知容积注射器标定 vs 以潮气量积分标定;(b) 换 DR2 检出算法——换切线判据或换平滑窗。每个臂内重估 ρ,比较臂间 ρ 的差。这个设计的逻辑是:如果 ρ 主要由「共用采集通道」这一物理事实决定,那么换定标、换算法都不该改变它太多,责任在测量原理;如果某一臂让 ρ 明显下降,那么 ρ 的实际来源是被该环节注入的,责任就在那一环。

    判据分四档。第一档:主臂 ρ>0.6 且两个扰动臂的 ρ 都同向为正,(a) 臂使 ρ 下降的幅度大于 (b) 臂的 2 倍 → 责任在定标,须把定标方法登记为误差结构的一部分并优先改造。第二档:(b) 臂使 ρ 下降更多 → 责任在检出算法,优先换算法。第三档:两臂 ρ 变动都 <0.1 且 ρ 的 95% 置信区间宽度 >0.3 → ρ 在当前样本量下不可估,须先扩样(或改用不依赖 ρ 的组水平终点),在此之前任何含 ρ 的 MDC95 都不得写入结论。第四档:主臂 ρ<0.2 或为负 → 与「同曲线同轴」的设计理由相冲,须逐条排查两个读数是否实际来自不同处理链(如不同平滑窗、不同采样率),排查不出则视为未测。

    这个设计的价值在于:它把「误差相关」从一个隐假设变成可测量的量,再把「ρ 偏大」的责任定位到具体环节。肺功能检测文献已经给出 ρ 可能量级的旁证:FOT 的会话内重复性在分量层面有明确数据(PMID 33914403),日间变异 5%–15%(PMID 23598954);但分量层面的重复性好,恰恰不能推出「两分量误差互不相关」——两个高度可重复的读数,可以因为共用同一采集通道而高度相关。这也是为什么必须有扰动臂,而不是只看 ICC。

    需要特别说明:本项目上一篇文章已经指出,含 ρ 的 MDC95 公式一旦启用,必须同时检查差值信度 r_DD 是否为正——ρ 升高会让差值信度塌得比门槛降得快(分量信度 0.80 时,ρ=0.6 已让 r_DD=−0.25)。因此本文测出的 ρ 不是终点,而是下一篇文章(差值信度会不会先破产)的输入。

    L1 网络状态:本主题对应边 E-004「金 → 肺」,《素问·阴阳应象大论》,现代证据「待补」,层级 L1(同构),状态 open——L1 网络中该边尚未进入实证阶段,现代证据列标注为「待补」。

    证据与局限

    已验证: 配对测量误差须用含 2ρs₁s₂ 的完整表达式(PMID 15286006);分量信度可用 ICC+SEM 标准框架估计(PMID 15705040);多次观测下的一致性评估方法成熟(PMID 17613642);FOT 会话内重复性有数据(PMID 33914403);日间变异 5%–15%(PMID 23598954)。
    部分验证: 含 ρ 的 MDC95 修正公式、ρ 的 Fisher z 置信区间方法学均有文献支持;但 ρ 本身在本领域(呼吸余量终点)从未实测,其量级、符号均未知。
    待验证: 主臂 ρ 的实测值与符号;两个扰动臂(换定标 / 换算法)各自对 ρ 的影响幅度;「ρ 主要来自共用通道」与「ρ 由某环节注入」两种假说的区分。下一步检验差值信度 r_DD 在实测 ρ 下是否仍为正(对应关联课题 #1283)。

    相关阅读

    证据等级: C
    参考文献: PMID 15286006, PMID 15705040, PMID 17613642, PMID 33914403, PMID 23598954


    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。

    How Large Is the Error Correlation Between Two Readings: Does the Perturbation Arm Experiment Point to Calibration or Algorithm?

    Evidence level: C

    The error correlation $\rho$ between two readings on the same respiratory curve and volume axis has never been measured in this field—its magnitude, sign, and direction determine whether the minimum detectable change (MDC) endpoint of the entire “margin” is relaxed or tightened.

    Body

    In respiratory function testing, the endpoint “margin” is defined as the difference between end-expiratory lung volume (EELV) and the change in vital capacity (DR2). A previous article (“The Minimum Detectable Change of the Margin Has Always Been Calculated Assuming Two Components Are Independent”) demonstrated that once measurement errors between two readings are correlated ($\rho \neq 0$), the assumption of independence inflates the MDC by 1.58 to 4.47 times (for $\rho = 0.6$ to $0.95$). Conversely, if $\rho$ is negative, the nominal false-positive rate of 5% rises to 8.6%–11%. In short, the entire decision threshold hinges on this single value, $\rho$.

    The problem is: this $\rho$ has never been measured in this field. The design rationale—”the same SVC-FOT curve, the same volume axis”—only implies that $\rho$ should be positive; it provides no magnitude. Yet the sign and size of $\rho$ precisely determine the conclusion’s direction: $\rho > 0$ means the independence assumption overestimates the threshold (self-penalty), while $\rho < 0$ means it underestimates it (self-loosening). More critically, larger $\rho$ leads to greater estimation error: at $\rho = 0.9$, an estimation bias of just 0.05 causes the MDC${95}$ to differ by 22%; at $\rho = 0.95$, the difference reaches 41%. With a sample size $n=30$, the 95% confidence interval half-width for $\rho = 0.8$ is approximately 0.14, which maps to a 2.29-fold difference in MDC$$. Therefore, “estimating $\rho$” itself requires an error budget; reporting a single point estimate is insufficient.

    This topic proposes a solution: use perturbation arms to decompose the “attribution of $\rho$.”

    Step 1: Recruit $\ge 30$ healthy adults, each tested 5 times on alternate days using SVC-FOT. Using the same algorithm, simultaneously detect DR2 and EELV (both converted to %TLC) on each occasion. Pair the two readings from each test, and directly estimate $s_1$, $s_2$, and $\rho$ using the $2 \times 2$ covariance matrix of the difference between two measurements. Provide the Fisher z-transformed 95% confidence interval for $\rho$. The complete expression for paired measurement error must include the term $2\rho s_1 s_2$—a framework long established in classical metrology (Bruynesteyn explicitly stated in rheumatoid joint progression assessment that statistical dependence of scores does not equal error correlation; the two must be handled separately, PMID 15286006). Component-level $s_1$, $s_2$, and ICC are derived from the standard retest reliability framework (Weir’s ICC+SEM framework, PMID 15705040). The method for assessing consistency across multiple observations per individual (Bland–Altman extension) provides the statistical basis for “treating the difference of repeated measurements as a new variable” (PMID 17613642).

    Step 2 is critical: Set up two perturbation arms. (a) Change the volume axis calibration method—for example, calibrate using a known-volume syringe versus calibrating via tidal volume integration. (b) Change the DR2 detection algorithm—for example, switch from tangent-based criteria to smoothing windows. Re-estimate $\rho$ within each arm and compare the difference in $\rho$ between arms. The logic of this design is: if $\rho$ is primarily determined by the physical fact of “shared acquisition channel,” then changing calibration or algorithm should not alter it significantly; responsibility lies with the measurement principle. If one arm causes $\rho$ to drop noticeably, then the actual source of $\rho$ is injected at that stage; responsibility lies there.

    The decision criteria are divided into four tiers:

    1. Tier 1: Main arm $\rho > 0.6$ and both perturbation arms yield positive $\rho$, but arm (a) reduces $\rho$ by more than twice the amount arm (b) does → Responsibility lies with calibration. The calibration method must be registered as part of the error structure and prioritized for modification.
    2. Tier 2: Arm (b) reduces $\rho$ more than arm (a) → Responsibility lies with the detection algorithm. Prioritize changing the algorithm.
    3. Tier 3: Both arms show $\rho$ variation $< 0.1$ and the 95% confidence interval width for $\rho$ exceeds 0.3 → $\rho$ cannot be estimated under the current sample size; sampling must be expanded first (or switch to group-level endpoints not dependent on $\rho$). Until then, any MDC$_{95}$ involving $\rho$ must not be included in conclusions.
    4. Tier 4: Main arm $\rho < 0.2$ or negative → This contradicts the design rationale of “same curve, same axis.” Each reading must be checked individually to determine if they actually originate from different processing chains (e.g., different smoothing windows, different sampling rates). If such discrepancies cannot be identified, $\rho$ is considered unmeasured.

    The value of this design lies in transforming “error correlation” from an implicit assumption into a measurable quantity, and precisely locating responsibility for “overestimated $\rho$” to specific stages. Respiratory function literature already provides collateral evidence of possible magnitudes for $\rho$: FOT intra-session repeatability has clear component-level data (PMID 33914403), and daily variation ranges from 5%–15% (PMID 23598954). However, good repeatability at the component level does not imply “errors between two components are uncorrelated.” Two highly repeatable readings can be strongly correlated simply because they share the same acquisition channel. This is why perturbation arms are necessary, rather than relying solely on ICC.

    Special note: A previous article in this project indicated that once the MDC${95}$ formula involving $\rho$ is activated, the difference reliability $r = -0.25$). Thus, the $\rho$ measured here is not the endpoint but input for the next article (whether difference reliability fails first).}$ must simultaneously be checked for positivity—rising $\rho$ causes $r_{DD}$ to collapse faster than the threshold drops (when component reliability is 0.80, $\rho = 0.6$ already yields $r_{DD

    L1 Network Status: This topic corresponds to edge E-004 “Metal → Lung” in Su Wen · Yin-Yang Ying Xiang Da Lun. Modern evidence is “pending,” level L1 (isomorphic), status open—the edge has not yet entered the empirical phase in the L1 network; modern evidence is marked as “pending.”

    Evidence and Limitations

    Verified: Paired measurement error requires the complete expression including $2\rho s_1 s_2$ (PMID 15286006); component reliability can be estimated using the ICC+SEM standard framework (PMID 15705040); consistency assessment under multiple observations is mature (PMID 17613642); FOT intra-session repeatability has data (PMID 33914403); daily variation is 5%–15% (PMID 23598954).

    Partially Verified: The MDC$_{95}$ correction formula involving $\rho$ and the Fisher z confidence interval methodology for $\rho$ are supported by literature; however, $\rho$ itself has never been directly measured in this field (respiratory margin endpoint), so its magnitude and sign remain unknown.

    Pending Verification: The measured value and sign of $\rho$ in the main arm; the magnitude of $\rho$ reduction caused by each perturbation arm (changing calibration vs. changing algorithm); distinguishing between the hypothesis that “$\rho$ mainly stems from the shared channel” and that “$\rho$ is injected at a specific stage.” Next, verify whether difference reliability $r_{DD}$ remains positive under the measured $\rho$ (corresponding to related topic #1283).

    Related Reading

    Evidence Level: C
    References: PMID 15286006, PMID 15705040, PMID 17613642, PMID 33914403, PMID 23598954


    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.

  • 为了藏住装置而加的那点不适,值多少度:装置反应性与主动规整量的相对大小标定

    Evidence level: C

    一句话核心命题:”知情减遮蔽”这个差值里混着两个方向相反的分量——知情组多主动规整、遮蔽组多装置不适——而这个被低估的比例,至今没有任何数字。

    正文

    研究丹田呼吸的人最常问的一个问题是:资深修炼者静息态的呼吸模式,和普通人到底差多少?要回答这个问题,就必须先把”测量本身会改变被测对象”这件事算清楚。呼吸测量领域早就知道:当一个人知道自己正在被测量时,他的呼吸会不自觉地变规整。Han 等在 74 名健康人身上用三条件设计证明了这一点——不知情(伪装成校准期)、知情(明确告知正在记录呼吸)、以及口含件加流速计,三条件下呼吸模式各不相同(PMID 9032510)。Hill 等的实验更进一步:82 名成人随机分到”知晓被数呼吸”与”不知晓”两组,知晓组的呼吸率平均低了 2.13 次/分(PMID 28426897)。这两项研究合起来说明一件事:对呼吸的”知道”,本身就会压平呼吸。

    但这里藏着一个被几乎所有研究忽略的坑。”知情减遮蔽”这个差值,被当作”主动规整量”——即练功者在知道自己被观察时,主动把呼吸调整向”标准样子”的量。然而这个差值里其实混着两个方向相反的分量:知情组多出来的主动规整,和遮蔽组为了藏住装置而承受的”装置不适”。为了不让受试者察觉在测呼吸,研究者常常被迫使用不那么舒服的装置——紧的呼吸带、口含件、鼻夹。而这些装置本身就是强干预。

    口含件的效应有多强?Weissman 等用无创 canopy 系统作对照,发现 17mm 口径的口含件加鼻夹使每分钟通气量、潮气量和平均吸气流量都显著增加,且把口径缩小到 9mm 就能消除大部分改变(PMID 6469818)。Perez 和 Tobin 进一步拆解了机制:改变呼吸路径(被迫经口呼吸)是口含件效应最主要的来源,而非单纯的口腔刺激(PMID 4066581)。Askanazi 团队的另一项研究显示,口含件与面罩在安静呼吸和化学刺激下会产生不同的呼吸模式(PMID 6816769)。即使在日常活动中,口含件也会让每分钟通气量增加 16%、潮气量增加 34%、呼吸周期增加 20%(PMID 1400052)。换句话说:你为了”测得更干净”而藏起来的装置,可能本身就是最大的干扰源。

    这正是 #1187 第二节指出的两条件设计死穴:三组做”知情 vs 遮蔽”两条件比较时,差值被装置反应性系统性压低。知情组在主动规整,遮蔽组在被装置不适干扰,两者相减,真实规整量被吃掉一块,而且这块被吃掉的比例没有任何数字。

    要把它测出来并不难,只需要把两条件换成三条件:① 裸露装置+不知情(装置反应性最纯)② 隐藏装置+不知情(接近理想遮蔽)③ 隐藏装置+知情(规整量+装置残余)。三条件两两相减,就能把装置反应性与主动规整量分离,还能算出两条件设计低估了真实规整量百分之多少。判据是:若装置反应性(①−②)达到差值(③−②)的 50% 以上,两条件设计整体作废,必须改三条件;若在 20-50%,差值须乘校正系数并报系数的不确定区间;若低于 20%,两条件设计可用,但必须在方法里写明装置型号与贴身程度。

    还有一层性别差异不能忽略。Han 等的研究特别指出,口含件引发的呼吸不适甚至焦虑在女性中更明显(PMID 9032510)。若性别×条件交互显著,校正系数必须分性别给。这也意味着,”装置反应性有多大”本身就是一个因人群而异的量,不能拿一个总体平均值糊弄过去。

    L1 网络状态:CORRESPONDENCE_NETWORK 生理/体验侧节点表仅列有「呼吸率 / 呼吸模式|可穿戴」观测项与「气感报告|结构化自评」,无边承载「装置反应性 × 主动规整量」的相对大小标定;该主题尚未进入 L1 对应网络(见 science-yixue-l1-gaps.md「测量反应性」簇第 5 条)。

    证据与局限

    已验证: 知晓呼吸被记录会改变呼吸模式——吸气/呼气时间延长、叹息与呼气末停顿减少(PMID 9032510);知晓被数呼吸使呼吸率平均低 2.13 次/分(PMID 28426897);口含件+鼻夹显著增加通气量与潮气量、缩小口径可消除大部分效应(PMID 6469818);口含件效应主因是呼吸路径改变(PMID 4066581);口含件使通气量+16%/潮气量+34%(PMID 1400052)。

    部分验证: 口含件与面罩在不同化学刺激下产生不同呼吸模式(PMID 6816769);装置不适在女性中更明显(PMID 9032510)。

    待验证: “装置反应性 vs 主动规整量”的相对大小——两条件设计低估真实规整量的百分比,至今无任何实测数字;三条件设计(裸装置/隐藏装置/隐藏+知情)是否能成功分离两个分量;校正系数是否随性别、BMI、修炼年限变化。

    相关阅读

    证据等级: C(传统文化共识/初步探索——该主题属测量学标定议题,L1 网络无边承载,证据链为呼吸测量领域既有实证的延伸应用)
    参考文献: PMID 9032510, PMID 28426897, PMID 6469818, PMID 6816769, PMID 4066581, PMID 1400052


    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。

    The discomfort added to hide the device, worth how many degrees: calibration of the relative magnitude of device reactivity versus active rectification

    Evidence level: C

    Core proposition: Within the “Awareness-minus-Blinding” differential lie two opposing components—greater active regulation in the aware group and greater device discomfort in the blinded group. The proportion of this differential attributable to device reactivity remains entirely unquantified.

    Body

    The most frequently asked question in Dantian breathing research is: How much does the resting breathing pattern of experienced practitioners differ from that of ordinary people? Answering this requires first quantifying the fact that measurement itself alters the measured object. The breathing measurement field has long established that when a person knows they are being measured, their breathing unconsciously becomes more regular. Han et al. demonstrated this in 74 healthy subjects using a three-condition design—unaware (disguised as calibration), aware (explicitly informed of recording), and mouthpiece plus flowmeter—showing distinct breathing patterns under each condition (PMID 9032510). Hill et al. extended this: 82 adults were randomly assigned to “knowing they were being counted” versus “not knowing,” with the aware group showing an average breathing rate 2.13 breaths per minute lower (PMID 2842510). Together, these studies establish: knowing about breathing flattens the breath.

    Yet here lies a pitfall ignored by almost all studies. The “Awareness-minus-Blinding” differential is often treated as “active regulation amount”—the quantity by which practitioners consciously adjust breathing toward a “standard pattern” when they know they are being observed. In reality, this differential mixes two opposing components: the extra active regulation in the aware group and the “device discomfort” endured by the blinded group to conceal the apparatus. To prevent subjects from noticing they are being measured, researchers are often forced to use uncomfortable devices—tight breathing belts, mouthpieces, nasal clips. These devices themselves constitute strong interventions.

    How powerful is the mouthpiece effect? Weissman et al., using a non-invasive canopy system as control, found that a 17mm mouthpiece plus nasal clip significantly increased minute ventilation, tidal volume, and average inspiratory flow, while reducing the aperture to 9mm eliminated most of these changes (PMID 6469818). Perez and Tobin further dissected the mechanism: altering the breathing pathway (forced oral breathing) is the primary source of mouthpiece effects, not mere oral stimulation (PMID 4066581). Another study by Askanazi’s team showed that mouthpieces and masks produce different breathing patterns under quiet breathing and chemical stimulation (PMID 6816769). Even in daily activities, mouthpieces increase minute ventilation by 16%, tidal volume by 34%, and respiratory cycle by 20% (PMID 1400052). In other words: the device you hide to “measure cleaner” may itself be the largest source of interference.

    This is precisely the fatal flaw in the two-condition design noted in #1187 Section 2: when comparing “aware vs. blinded,” the differential is systematically suppressed by device reactivity. The aware group actively regulates, while the blinded group suffers device discomfort; subtracting them consumes part of the true regulation amount, and the proportion consumed remains entirely unquantified.

    Measuring this is not difficult; simply convert the two-condition design into three: ① exposed device + unaware (purest device reactivity), ② hidden device + unaware (near-ideal blinding), ③ hidden device + aware (regulation plus residual device effect). Subtracting these pairwise separates device reactivity from active regulation and calculates how much the two-condition design underestimates true regulation. The criterion is: if device reactivity (①−②) exceeds 50% of the differential (③−②), the two-condition design is invalid and must be replaced with three conditions; if between 20–50%, the differential must be multiplied by a correction coefficient with its uncertainty interval reported; if below 20%, the two-condition design is usable but must explicitly state device model and fit tightness in methods.

    One more layer of gender difference cannot be ignored. Han et al. specifically noted that mouthpiece-induced breathing discomfort and anxiety are more pronounced in women (PMID 9032510). If gender×condition interaction is significant, the correction coefficient must be gender-specific. This means “how much device reactivity exists” itself is a population-dependent quantity that cannot be approximated by a single overall average.

    L1 Network Status: The CORRESPONDENCE_NETWORK physiological/experience-side node table currently lists only “Breathing rate / Breathing pattern | Wearable” and “Qi sensation report | Structured self-assessment,” lacking any carrier for “Device Reactivity × Active Regulation” relative magnitude calibration. This topic has not yet entered the L1 corresponding network (see science-yixue-l1-gaps.md “Measurement Reactivity” cluster, item 5).

    Evidence and Limitations

    Verified: Awareness of breathing being recorded alters breathing patterns—prolonged inhalation/exhalation, reduced sighs and end-expiratory pauses (PMID 9032510); awareness of being counted reduces breathing rate by 2.13 breaths/min (PMID 2842510); mouthpiece + nasal clip significantly increases ventilation and tidal volume, with smaller aperture eliminating most effects (PMID 6469818); mouthpiece effect primarily due to breathing pathway alteration (PMID 4066581); mouthpiece increases ventilation +16%, tidal volume +34% (PMID 1400052).

    Partially Verified: Mouthpiece and mask produce different breathing patterns under varying chemical stimulation (PMID 6816769); device discomfort more pronounced in women (PMID 9032510).

    Pending Verification: Relative magnitude of “device reactivity vs. active regulation”—the percentage by which two-condition design underestimates true regulation remains entirely unmeasured; whether three-condition design (exposed/hidden/hybrid) successfully separates the two components; whether correction coefficients vary by gender, BMI, or years of practice.

    Further Reading

    Evidence Level: C (traditional consensus / preliminary exploration—this topic belongs to measurement calibration issues, currently lacking L1 network carriers; evidence chain extends existing breathing measurement research)
    References: PMID 9032510, PMID 2842510, PMID 6469818, PMID 6816769, PMID 4066581, PMID 1400052


    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.

  • 加设备是摊薄还是暴露:设备偏倚的结构决定功效的走向

    Evidence level: C

    多台设备一起测,到底是”样本变多了”还是”噪声变大了”?答案不在设备数量本身,而在这些设备的偏倚是朝同一个方向、还是各偏各的。

    正文

    问题的起点:研究”气沉丹田”这类现象时,想用多台仪器同时采集——三台、五台、八台一起上,以为这样等于把样本量乘了几倍。这个直觉来自一个默认假设:设备只是”多个版本的同一把尺子”,多一把尺子就多一批独立读数。但现实里设备不是同一把尺子。每一台设备都带着自己的一套系统性偏移——有的刻度整体偏高,有的偏低;有的对呼吸敏感,有的对温度敏感。这个偏移叫设备偏倚。它一进来,上面的直觉就崩了:多设备到底帮你还是害你,不取决于设备数量,而取决于这些偏倚怎么排列。

    同向偏倚与异向偏倚,是两种完全不同的世界。假设三台设备都系统性偏高 0.5 个标准差(全同向):它们的读数互相相关,每台都往同一个方向撒谎。此时把三台的数据合起来,等于把同一个错误重复了三遍——均值不会因此更接近真值,只会更稳地停留在错误的位置。更糟的是,多中心研究里这种偏倚会直接改变结论:磁共振成像(MRI)研究中,来自不同扫描仪的数据异质性就足以改变疾病关联分析的结论,而常见的 ComBat 类校正算法只能部分补偿这种异质性(PMID 41380987)。反过来说,如果三台设备各偏各的(随机异向 ±0.5σ):一台偏高、一台偏低、一台居中,它们的错误互相抵消,合起来确实更接近真值——这时”加设备 = 加样本”才成立。同一种数量操作,两种截然不同的结局,分野全在偏倚的方向结构。

    更反直觉的是,存在”设备越多功效越低”的实证区间。功效(power)是”真有效应能被测出来”的概率。在簇随机试验里有一个已知规律:簇大小不等时,要达到同样的功效,需要的簇数要多出四成以上(PMID 32970522)——也就是说,结构不均匀会让”增加单元数”这件事贬值。设备偏倚结构不均匀同理:全同向偏倚且没有校准样本时,加设备只是在积累同一方向的错误,功效不仅不升,反而可能随设备数上升而下降。这不是推理,是模拟里可以看到的区间:设备数从 3 台加到 5 台,功效掉 5 个百分点以上。“设备数不是样本量”——这句看起来违背直觉的结论,在特定偏倚结构下是字面为真的。

    那什么时候”加设备”才真正划算?统计学给出了精确的条件。多中心试验的样本分配存在一个最优配比:中心数与每中心人数必须联合优化,而这个最优配比本身依赖未知的方差参数,需要用最大化最坏情形(maximin)的策略来兜底(PMID 42479089)。两层嵌套设计(人嵌套在设备里)在固定总样本量下同样存在最优分配比——均衡设计(每台设备配同样多人)并非在所有假设下都最优(PMID 24030312)。翻译成大白话:加多少台设备、每台配多少人,是一道有解的优化题,不是”越多越好”。最优解由三类未知参数共同决定:设备间偏倚的方差(偏倚有多大)、偏倚的方向结构(是否同向)、以及个体测量噪声(人本身的波动)。这三个参数不估出来,任何”加设备”的决策都是掷骰子。

    回到气功研究:丹田参数(呼吸比、相位同步率等)的跨设备测量,目前既没有估过设备间偏倚的方向结构,也没有配校准样本。本文主张一个可执行的全因子模拟:设备偏倚结构设三档(全同向 +0.5σ / 随机异向 ±0.5σ / 半同向半异向),设备数取 1/2/3/5/8,受试者取 20/40/80,是否配备校准样本两档,每格跑 2000 次模拟算功效与校正后的残余偏倚。判据四条:① 若存在”设备数从 3 增至 5 功效下降 >5 个百分点”的格子,则”多设备 = 更多样本”被证伪,跨设备设计必须设设备数上限并逐台报告偏倚;② 若仅在同向偏倚且无校准样本时功效下降、有校准后消失,则问题不在设备数而在校准缺失,方案须强制每台设备配校准样本;③ 若功效随设备数单调不降,则设备异向偏倚可被随机效应模型完全吸收,原”设备数不是样本量”的断言过强、须撤回;④ 若功效下降幅度随受试者数增大而缩小,则设备数是”小样本时代的伪问题”——只在大样本下才可忽略,文章须写明适用样本区间。

    一句话总结:多设备是”摊薄”还是”暴露”,取决于设备偏倚是同向还是异向。同向偏倚下加设备是在放大同一个错误,功效可以真的下降;异向偏倚下加设备才是摊薄噪声。判定偏倚结构、配校准样本、按最优配比分配——这三步做完,”加设备”才是研究者的朋友。

    证据与局限

    已验证: 多扫描仪异质性可直接改变 MRI 关联分析结论,ComBat 类校正仅部分补偿(PMID 41380987);簇大小不等时须增加四成以上簇数才补回效率损失(PMID 32970522);多中心试验存在最优中心-人数配比且依赖未知方差参数(PMID 42479089);两层嵌套设计在固定总量下存在最优分配比、均衡设计并非处处最优(PMID 24030312)。
    部分验证: 以上证据来自医学影像与临床试验设计领域,属同构类比——气功/丹田参数测量尚无任何设备偏倚结构实测数据。
    待验证: ① 丹田参数(呼吸比、相位同步率)设备间偏倚的方向结构实测——全因子模拟的核心输入,从未估过;② “设备数 3→5 功效下降 >5 个百分点”是否存在于丹田参数的真实参数空间;③ 校准样本(每台设备配一组标准受试者)能否完全消除同向偏倚对功效的侵蚀;④ 最优配比公式(PMID 42479089 / 24030312)在气功测量场景的参数化移植。

    L1 网络状态:该主题对应”设备偏倚结构 → 跨设备功效走向”边,在 CORRESPONDENCE_NETWORK / EDGE_TABLE_science-yixue 中均无此边——网络边表只有”源—关系—目标”单一对应,无”设备间偏倚结构(同向/异向)”节点,亦无”多设备设计的最优配比”约束位。该边尚未进入 L1 对应网络,本文证据等级受限为 C。

    相关阅读

    证据等级: C
    参考文献: PMID 41380987, PMID 32970522, PMID 42479089, PMID 24030312


    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。

    Adding Equipment: Bias or Exposure? Structural Bias Determines the Direction of Effectiveness

    Evidence level: C

    When multiple devices are used simultaneously, does the sample size effectively increase or does the noise amplify? The answer lies not in the quantity of devices, but in the structure of their bias: whether they share a common directional error or exhibit independent deviations.

    Body

    The starting point of the problem: When investigating phenomena such as “Qi sinking into the Dantian” (a traditional concept referring to the concentration of vital energy in the lower abdomen), researchers often intend to collect data using multiple instruments simultaneously—three, five, or eight devices—to assume this multiplies the sample size. This intuition stems from a default assumption: devices are merely “multiple versions of the same ruler”, implying that adding another device adds another set of independent readings. However, in reality, devices are not the same ruler. Each device carries its own systematic offset—some scales may be generally higher, others lower; some are sensitive to breathing, others to temperature. This offset is called device bias. Once this bias enters the analysis, the initial intuition collapses: whether adding devices helps or harms depends not on their number, but on how these biases are structured.

    Co-directional bias and counter-directional bias represent two entirely different worlds. Suppose three devices are all systematically high by 0.5 standard deviations (fully co-directional): their readings are correlated, each lying in the same direction. Combining their data merely repeats the same error three times—the mean does not approach the true value; it simply stabilizes at the wrong position. More critically, in multi-center studies, such bias can directly alter conclusions: in MRI research, heterogeneity from different scanners is sufficient to change disease association results, while common ComBat-style correction algorithms can only partially compensate for this heterogeneity (PMID 41380987). Conversely, if three devices each have random independent biases (stochastic counter-directional ±0.5σ)—one high, one low, one centered—their errors cancel out, and the combined result indeed approaches the true value. The same numerical operation yields two radically different outcomes, distinguished entirely by the structure of the bias.

    More counterintuitively, there exists an empirical region where “more devices means lower effectiveness.” Effectiveness (power) is the probability of detecting a true effect. In cluster-randomized trials, a known rule applies: when cluster sizes are unequal, more than 40% additional clusters are required to recover the efficiency loss (PMID 32970522)—meaning structural heterogeneity devalues the act of “adding units.” Device bias structure behaves similarly: with fully co-directional bias and no calibration samples, adding devices merely accumulates errors in the same direction; effectiveness not only fails to rise but may actually decline as the number of devices increases. This is not mere reasoning; it is an observable interval in simulations: increasing the number of devices from 3 to 5 can cause effectiveness to drop by more than 5 percentage points. “The number of devices is not the sample size”—this conclusion, which seems to defy intuition under normal assumptions, becomes literally true under specific bias structures.

    So when is “adding devices” truly worthwhile? Statistics provides precise conditions. Multi-center trials have an optimal allocation ratio: the number of centers and the number of subjects per center must be jointly optimized, and this optimal ratio itself depends on unknown variance parameters, requiring a maximin strategy to safeguard against the worst-case scenario (PMID 42479089). Similarly, in two-level nested designs (subjects nested within devices), an optimal allocation ratio exists under fixed total sample size—balanced design (equal subjects per device) is not always optimal across all assumptions (PMID 24030312). In plain terms: how many devices to add and how many subjects per device is a solvable optimization problem, not “the more the better.” The optimal solution is jointly determined by three unknown parameters: the variance of inter-device bias (how large the bias is), the direction structure of the bias (whether it is co-directional or counter-directional), and individual measurement noise (natural variation among subjects). Without estimating these three parameters, any decision to “add devices” is essentially a gamble.

    Returning to Qi research: Cross-device measurement of Dantian parameters (breathing ratio, phase synchronization rate, etc.) currently lacks estimation of the direction structure of inter-device bias and allocation of calibration samples. This paper proposes an executable full-factorial simulation: device bias structure set at three levels (fully co-directional +0.5σ / stochastic counter-directional ±0.5σ / half co-directional half counter-directional), number of devices at 1/2/3/5/8, number of subjects at 20/40/80, and presence of calibration samples at two levels. Each cell runs 2,000 simulations to compute effectiveness and residual bias after correction. Four criteria determine the outcome: ① If any cell shows effectiveness dropping by >5 percentage points when devices increase from 3 to 5, then “more devices = larger sample” is falsified; cross-device design must impose an upper limit on device count and report bias per device; ② If effectiveness drops only under co-directional bias without calibration samples but recovers after calibration, then the problem lies not in device count but in missing calibration; the protocol must mandate calibration samples for each device; ③ If effectiveness monotonically does not decline with increasing devices, then inter-device counter-directional bias is fully absorbed by random effects models; the original claim “device count is not sample size” is too strong and must be withdrawn; ④ If the magnitude of effectiveness decline shrinks as subject count increases, then device count is a “pseudo-problem of the small-sample era”—only negligible under large samples; the article must explicitly state the applicable sample range.

    One-sentence summary: Whether adding devices dilutes or exposes depends on whether device bias is co-directional or counter-directional. Under co-directional bias, adding devices amplifies the same error, potentially reducing effectiveness; under counter-directional bias, it dilutes noise. Determining the bias structure, allocating calibration samples, and optimizing the ratio—these three steps make “adding devices” a friend of the researcher.

    Evidence and Limitations

    Verified: Heterogeneity across scanners can directly alter MRI association conclusions, with ComBat-style corrections only partially compensating (PMID 41380987); unequal cluster sizes require >40% more clusters to recover efficiency loss (PMID 32970522); multi-center trials possess an optimal center-subject ratio dependent on unknown variance parameters (PMID 42479089); two-level nested designs under fixed total sample size have an optimal allocation ratio, and balanced design is not universally optimal (PMID 24030312).
    Partially verified: The above evidence comes from medical imaging and clinical trial design domains, representing isomorphic analogy—no empirical data on device bias structure exists for Qi/Dantian parameter measurement.
    To be verified: ① Empirical measurement of the direction structure of inter-device bias for Dantian parameters (breathing ratio, phase synchronization rate)—the core input of the full-factorial simulation, never estimated; ② Whether the “effectiveness drop >5 percentage points when devices increase from 3 to 5” exists within the true parameter space of Dantian parameters; ③ Whether calibration samples (one set of standard subjects per device) can fully eliminate the erosive effect of co-directional bias on effectiveness; ④ Parameterization transfer of the optimal ratio formula (PMID 42479089 / 24030312) to the Qi measurement context.

    L1 Network Status: This topic corresponds to the “device bias structure → cross-device effectiveness trajectory” edge, which is absent in both CORRESPONDENCE_NETWORK and EDGE_TABLE_science-yixue—the network edge table only contains single “source–relation–target” mappings, lacking nodes for “inter-device bias structure (co-directional/counter-directional)” and constraint positions for “optimal allocation ratios in multi-device design.” This edge has not yet entered the L1 correspondence network; thus, the evidence level of this article is limited to C.

    Related Reading

    Evidence level: C
    References: PMID 41380987, PMID 32970522, PMID 42479089, PMID 24030312


    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.

  • 差值终点够不够格:ΔX5/ΔAX 自身的重测一致性门槛,卡在”气感特异增量”这个主判据的命门上

    Evidence level: C

    要证明”气感带来的呼吸力学增量是真的”,先要回答一个更基础的问题:用来证明它的那个差值,本身量得稳不稳?如果 ΔX5 的重测信度连 0.5 都不到,”增量不显著”就永远分不清是”真的没有”还是”测不出来”。

    正文

    “丹田呼吸让电抗 X5 往负向走、让 AX 变大”——这是气功呼吸力学研究里最常被当成结论的一句话。但要把它从”现象描述”升级成”可检验的主判据”,中间还隔着一道几乎没人走的路:用来承载这个结论的差值终点 ΔX5、ΔAX,它们自己的重测一致性到底是多少?

    这个问题不是理论洁癖。强迫振荡技术(FOT)测的是呼吸系统的阻抗与电抗,而电抗这一侧公认比电阻一侧更难测稳。欧洲呼吸学会的技术标准反复强调,同一会话内电阻 R5 的变异系数要控制在 10% 以内才算合格,却对电抗侧几乎没有同等强度的质控(PMID 33033141)。健康成人的会话间变异分析也显示,X5 与 AX 的会话间波动明显宽于 R5——即便在同一批受试者、同一批设备、相隔 7±3 天的条件下(PMID 35907930)。也就是说,我们打算拿来当”主判据”的那两个量,恰恰是整套 FOT 参数里最不稳的两个。

    更麻烦的是数学。经典测量理论给差值信度写过一个精确的公式:

    $$r_{DD} = \frac{(r_{XX} + r_{YY})/2 – r_{XY}}{1 – r_{XY}}$$

    当两个分量(真呼吸条件与伪呼吸条件)的观测相关 r_XY 越高,差值信度就越往零塌(PMID 36842498)。而”丹田呼吸 vs 等频率伪呼吸”恰恰是最容易高相关的一对——同一个人的肺、同一个呼吸频率、同一套胸廓,两者的读数本来就该高度相关。如果分量的重测信度只有 0.7 上下(FOT 电抗侧的现实水平),而两条件相关 r_XY 又高(0.8-0.9 不稀奇),那么代入公式,ΔX5 的信度很可能已经落到 0.5 以下。认知任务领域的实证也给出了同样的警告:在所有候选指标里,差值指标的重测信度排在最末,而同任务里不用差值的单维指标信度良好(PMID 39890691)——虽然那是在认知任务上测的,但它提醒我们,差值信度崩塌是一个普遍的结构性风险,不挑领域。

    这就是为什么”气感特异增量不显著”这句判读现在处于一种悬空状态:它可能是真的(气感确实没有带来增量),也可能只是测不出来(差值本身太不稳,任何真实增量都被噪声淹没)。没有 ΔX5/ΔAX 的重测一致性标定,这两种解释在统计上无法区分,整条研究线的结论就都盖着”未标定”的章。

    要补上这道门槛,需要的是一次纯信度标定:不分组、不干预,只回答”这个差值本身稳不稳”。≥40 名受试者,隔 2-7 天重复 4 次,每次在同一会话内完成”真/伪两条件”的模拟测量,逐人计算 X5_真、X5_伪 与 ΔX5 的 ICC(2,1)、SEM、MDC95,同时报告两条件的相关 r 与方差比 σ_真/σ_伪——因为公式告诉我们,这两个量决定了差值信度的走向。

    判据可以做成四档:若 ΔX5 的 ICC ≥0.75 且 MDC95 小于预期效应量,差值终点可用、原设计保留;若 ICC 落在 0.5-0.75 之间,则不能直接作差,须改为”以分量为协变量的回归”,并把结论降为组水平;若 ICC <0.5,差值终点不可用,课题须改判为”该构念在现有测量精度下不可检验”、转入方法学储备;若两条件相关 r >0.9 且方差比接近 1,则差值信度崩塌可由公式直接预言,无须实测即可判定原设计无效——这一步其实是在省样本,把注定要失败的设计提前判死。

    这里还要点破一个被普遍忽略的事实:“多次取均值”救不了差值信度。 会话内重复取均值压的是当天随机误差,而差值信度公式里的 r_XY 是两个条件下的相关,不会因为多测几次就变小——如果两条件共享同一个驱动源(同一套胸廓、同一个呼吸节律),那个高相关是结构性的,不是采样误差。

    L1 网络状态:CORRESPONDENCE_NETWORK 边表止于 E-050(六十甲子纳音↔频率,层级 L1、状态 open,现代证据栏原文”尚无任何可测来源”),无以”差值信度 / 重测一致性 / ICC / FOT 电抗”为源或目标的边;节点表生理侧仅有”呼吸率 / 呼吸模式”等信号节点,无”测量学维度(信度/ICC/MDC95)”这一层。该主题尚未进入 L1 对应网络,已追加至 science-yixue-l1-gaps.md 第 62 条。

    证据与局限

    已验证: 差值信度公式 r_DD = [(r_XX+r_YY)/2 − r_XY] / (1 − r_XY) 有明确的测量理论出处(PMID 36842498);FOT 电抗侧重复性弱于电阻侧、会话间变异更宽(PMID 33033141、PMID 35907930);认知任务领域差值指标重测信度排末位(PMID 39890691)。

    部分验证: 上述规律在呼吸力学参数上是否同样成立尚无直接实测——这正是本文设计的标定要回答的问题;FOT 参数日间波动 5%-15% 的常识量级来自综述性文献(PMID 35140105),具体到丹田呼吸场景未见专项数据。

    待验证: ① ΔX5/ΔAX 在实际丹田呼吸 vs 伪呼吸条件下的 ICC、SEM、MDC95 数值;② 两条件相关 r_XY 与方差比是否如公式预言的把差值信度压到 0.5 以下;③ 若差值不可用,”分量作协变量的回归”方案能否保住”气感特异增量”这个构念;④ 正交化/因子法(如可靠分量法)能否改善差值信度(PMID 15327736 提示正交解优于斜交解)。

    相关阅读

    证据等级: C
    参考文献: PMID 39890691, PMID 36842498, PMID 15327736, PMID 35140105, PMID 33033141, PMID 35907930, PMID 35944717


    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。

    Is the Difference Endpoint Qualified: The Retest Reliability Threshold of ΔX5/ΔAX Hinges on the \”Qi-Sensation Specific Increment\” as the Primary Criterion

    Evidence level: C

    To prove that “the respiratory mechanics increment caused by qi-sensation is real,” one must first answer a more fundamental question: Is the difference endpoint itself measured stably? If the retest reliability of ΔX5 falls below 0.5, it is impossible to distinguish whether “the increment is not significant” means “there truly is none” or “it could not be measured.”

    Body

    The statement “Dantian breathing drives the reactance X5 negative and increases AX” is the most frequently cited conclusion in qigong respiratory mechanics research. However, elevating this from a “phenomenological description” to a “testable primary criterion” requires crossing a gap almost no one has traversed: what is the retest reliability of the difference endpoints ΔX5 and ΔAX themselves?

    This is not a matter of theoretical fastidiousness. Forced Oscillation Technology (FOT) measures respiratory impedance and reactance, with the reactance side universally acknowledged as harder to stabilize than the resistance side. European Respiratory Society technical standards repeatedly emphasize that the coefficient of variation for resistance R5 must be kept under 10% to qualify, yet they impose almost no equivalent quality control on the reactance side (PMID 33033141). Session-to-session variability analysis in healthy adults also shows that X5 and AX fluctuate significantly more than R5—even under identical subjects, equipment, and intervals of 7±3 days (PMID 35907930). Thus, the two quantities intended as “primary criteria” are precisely the most unstable parameters in the entire FOT set.

    Compounding this is the mathematics. Classical measurement theory provides an exact formula for difference reliability:

    $$r_{DD} = \frac{(r_{XX} + r_{YY})/2 – r_{XY}}{1 – r_{XY}}$$

    As the observed correlation $r_{XY}$ between the two conditions (true breathing vs. pseudo-breathing) increases, the difference reliability collapses toward zero (PMID 36842498). “Dantian breathing vs. equal-frequency pseudo-breathing” represents the pair most prone to high correlation—same lungs, same breathing rate, same thoracic cage mean the readings should inherently be highly correlated. If component retest reliability is only around 0.7 (the realistic level for FOT reactance), and the condition correlation $r_{XY}$ is high (0.8–0.9 is not unusual), substituting these into the formula likely drives ΔX5 reliability below 0.5. Empirical evidence from cognitive tasks also warns of this: among candidate indicators, difference metrics rank last in retest reliability, while single-dimension indicators without differences maintain good reliability (PMID 39890691)—even though that study focused on cognitive tasks, it highlights that difference reliability collapse is a universal structural risk, not domain-specific.

    This explains why the judgment “qi-sensation specific increment is not significant” currently hangs in limbo: it may be true (qi-sensation genuinely yields no increment) or merely unmeasurable (the difference endpoint itself is too unstable, drowning any real increment in noise). Without calibrating the retest consistency of ΔX5/ΔAX, these two explanations cannot be statistically distinguished, leaving the entire research line stamped “un-calibrated.”

    To bridge this gap requires a pure reliability calibration: no grouping, no intervention, simply answering “is this difference endpoint stable?” With ≥40 subjects, repeat measurements 4 times over 2–7 days, performing simulated “true/pseudo” dual-condition measurements within each session. For each subject, calculate ICC(2,1), SEM, and MDC95 for X5_true, X5_pseudo, and ΔX5, while reporting the correlation $r$ between conditions and variance ratio $\sigma_{true}/\sigma_{pseudo}$—because the formula dictates these two quantities determine the trajectory of difference reliability.

    The criterion can be structured into four tiers:
    – If ΔX5 ICC ≥0.75 and MDC95 is below the expected effect size, the difference endpoint is valid and the original design retained.
    – If ICC falls between 0.5–0.75, direct differencing cannot be used; instead, switch to “regression with components as covariates,” downgrading conclusions to group level.
    – If ICC <0.5, the difference endpoint is invalid; the study must be reclassified as “this construct cannot be tested under current measurement precision” and archived for methodological reserve.
    – If the condition correlation $r >0.9$ and variance ratio approaches 1, difference reliability collapse can be predicted directly by the formula without empirical testing—this step actually saves samples by pre-declaring doomed designs as failed.

    Here, we must also expose a widely overlooked fact: “Averaging multiple trials does not rescue difference reliability.” Averaging within sessions suppresses random daily error, but the $r_{XY}$ in the difference reliability formula reflects correlation between conditions, which does not shrink with more trials. If both conditions share the same driver (same thoracic cage, same breathing rhythm), that high correlation is structural, not sampling error.

    L1 Network Status: The CORRESPONDENCE_NETWORK edge list terminates at E-050 (Sixty Jiazi Na Yin ↔ Frequency, Level L1, Status open, modern evidence section originally stated “no measurable source yet”). There are no edges with “difference reliability / retest consistency / ICC / FOT reactance” as source or target; the physiological node list includes only signals like “breathing rate / breathing pattern,” lacking the “measurement dimension (reliability/ICC/MDC95)” layer. This topic has not yet entered the L1 corresponding network and has been added to item 62 of science-yixue-l1-gaps.md.

    Evidence and Limitations

    Verified: The difference reliability formula $r_{DD} = [(r_{XX}+r_{YY})/2 − r_{XY}] / (1 − r_{XY})$ has clear measurement theory origins (PMID 36842498); FOT reactance exhibits weaker repeatability than resistance with wider session-to-session variability (PMID 33033141, PMID 35907930); difference metrics rank last in retest reliability in cognitive tasks (PMID 39890691).

    Partially Verified: Whether these patterns hold for respiratory mechanics parameters lacks direct empirical testing—this is precisely the calibration question designed here; FOT parameter daily fluctuations of 5–15% are common knowledge from review literature (PMID 35140105), but specific data for Dantian breathing scenarios is absent.

    To Be Verified: ① ICC, SEM, MDC95 values for ΔX5/ΔAX under actual Dantian breathing vs. pseudo-breathing conditions; ② whether condition correlation $r_{XY}$ and variance ratio suppress difference reliability below 0.5 as the formula predicts; ③ if the difference endpoint is unusable, whether “regression with components as covariates” can preserve the construct of “qi-sensation specific increment”; ④ whether orthogonalization/factor methods (e.g., reliable component method) can improve difference reliability (PMID 15327736 suggests orthogonal solutions outperform oblique ones).

    Related Reading

    Evidence Level: C
    References: PMID 39890691, PMID 36842498, PMID 15327736, PMID 35140105, PMID 33033141, PMID 35907930, PMID 35944717


    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.

  • 83% 那条线有多少是规定、多少是惯例:判据门槛的出处核查与替代方案

    Evidence level: C

    同一批人的三模态方向投票,判「通过 / 失败」用的 10/12、ICC 0.3、相关 0.5、功效 30% 四个门槛,到底哪来的?追到底会发现:其中三个是「惯例」不是「规定」,而且零假设本身设得太保守——换成等效性检验后,「三角验证」的结论形式可能根本不可达。

    正文

    气功外气研究里,「三角验证」是个常用话术:同一构造,用 %AB、φ、X5、CoV 四个参数、12 个格子同时测,格子同向率超过某条线就算「验证通过」。但这条线是怎么定出来的,几乎没人问过。本文要做的,就是把这几个数字的出处一个个挖出来,看它们是「规定」(有方法学依据、有出处)还是「惯例」(大家这么用,没人为它负责)。

    先看第一个数字:ICC ≥ 0.3 算「有信度」。这个数在测量学界其实有一整条可查的来源链——最权威的出处是 COSMIN(健康测量工具遴选共识标准)。COSMIN 对信度的评级标准是 ICC ≥ 0.70 为「充分」、< 0.70 为「不充分」、没报告为「不确定」(PMID 17161752,Terwee 2007 年发表在 J Clin Epidemiol 的《Quality criteria were proposed for measurement properties of health status questionnaires》)。注意:0.3 根本不在 COSMIN 的档位里。学界流传的「0.3 及格」来自更早的心理测量文献里「临床 vs 研究用途」的粗略分层(研究用途可放宽),但那不是 COSMIN 的判定标准。所以 ICC 0.3 这条线,是「惯例」——它对应的是「研究用途」的宽口径,不是「这个测量能不能用」的判定。

    第二个数字:相关 ≥ 0.5。这个数同样能追到 COSMIN 的构造效度标准:与「测同一构造」的工具相关 ≥ 0.50 视为充分(同一份 COSMIN 文件里有明确条款,PMID 17161752)。也就是说「相关 0.5」是构造效度的收敛标准,被拿来当「两模态方向一致」的线——语义上是近亲,但不是一回事。收敛效度问「两个工具测的是不是同一个东西」,投票一致性问「两个格子方向朝不朝同一个方向」,后者没有 COSMIN 条款。

    第三个数字:功效 30%。这个最离谱。功效(power)在统计学的任何标准里都是越高越好,公认下限是 80%(这是统计教材级的规定,不是惯例——Cohen 的经典框架里 0.80 是「想要但常达不到」的目标)。30% 作为「判据门槛」出现在这里,是把「探测功效」误用成了「通过线」:30% 其实意味着「有真效应时也只有三成概率能检出」,这个数字当门槛,等于允许七成概率漏报。它既不是规定,也不是惯例,更像是口误。

    第四个数字:10/12(83% 同向率)。这是唯一一个真正由二项分布算出来的数:12 格、每格独立、零假设同向率 0.5,双侧 α=0.05 时,≥10 格同向才显著(P≈0.019)。它的问题是零假设本身:二项检验假设每一格都在「独立抛硬币」,可 12 个格子测的是同一个构造,同向率低于 0.9 都不该算「验证通过」——零假设设成 0.5,等于把「同向率 0.6-0.8 的弱证据」也放进「可能通过」的区间。这正是#1266 那篇「12 格折成 6.3 格甚至 4.8 格」指控的统计侧根源:格子不独立,二项框架的整体假设就已经打折。

    那替代方案是什么?最干净的一条是等效性检验(TOST,two one-sided tests)。TOST 的原始出处是 Schuirmann 1987 年在 J Pharmacokinet Biopharm 上的论文(PMID 3450848):把零假设从「无差异」翻过来,改成「差异太大不可接受」,上下两条单侧检验都拒绝,才能宣布「等效」。用在投票判据上就是:不问「同向率是否显著大于 0.5」,而问「同向率是否显著落在 [0.7, 0.9] 的等效区间内」——等效界取多宽,直接决定 12 格需要几格同向。如果等效界取 0.1(即同向率 0.8±0.1 算通过),所需格数会显著高于 10/12,甚至超过现有可行上限——那「三角验证通过」这个结论形式本身就不可达,只能退而逐模态独立报告。

    还有一层:投票计数的口径混乱。系统综述方法学里有实证记录:非元分析的合成方法(包括投票计数)被广泛使用却极少被说明(PMID 36758885,2023 年 J Clin Epidemiol 调查),投票计数在少量研究下异质性检验表现很差(PMID 10412966,Am J Epidemiol 1999)。这三条合起来的意思是:把「方向投票」当验证结论的载体,方法学本身就不稳——这跟符号检验在平票不确定时功效大跌(PMID 20974291)是同一个病根:投票法的功效对「平票格怎么算」极其敏感。

    落到研究设计上,可做一次纯方法学标定(无人体数据即可):把零假设 p0 从 0.5 扫到 0.9、格数从 6 扫到 65、门槛从「精确二项」换成 TOST,输出三张对照表——(a) 不同 p0 下的最小同向格数 (b) 达到 80% 功效所需格数 (c)「判失败 / 未检出 / 判通过」三区的位置;再用 ≥40 人实测同向率作锚点校准 p0。三条核心判据:若 p0 提到 0.7 后 9/12 就能显著,则现有「7/12 判失败」的结论在更合理零假设下翻成「无法判断」,此前所有基于 p0=0.5 的失败判定须整体重估;若 TOST 等效界 0.1 时所需格数 > 现有可行上限,「三角验证通过」结论形式不可达,须改逐模态独立报告;若三个惯例数能追到方法学出处,则 #1188 的自认应改成引用具体 GRADE/COSMIN 条款。

    证据与局限

    已验证: 二项检验的保守性有方法学文献支撑——投票计数在少量研究下的缺陷(PMID 10412966)与平票不确定性对符号检验功效的打击(PMID 20974291)都是公开文献结论;COSMIN 的质量标准(ICC ≥ 0.70、相关 ≥ 0.50)有明确出处(PMID 17161752)。

    部分验证: TOST 等效性检验作为「支持无效应」的标准做法有原始论文(PMID 3450848)和后续方法学综述(PMC5502906)支撑,但它在方向投票这类「格子非独立」场景的适用性,只有理论推演、没有实证标定——本文的核心判据(p0 提到 0.7 后门槛落到几格)需要蒙特卡洛模拟才能落地。

    待验证: ① 12 格同向率的真实分布——用 ≥40 人有经验者实测,校准 p0 该取 0.5 还是 0.7/0.9;② TOST 等效界取 0.1 时所需格数与可行上限的对照表——纯数值模拟即可产出;③ 三个惯例数(ICC 0.3 / 相关 0.5 / 功效 30%)在库内既往文章中的实际使用范围——需要一次库内全量统计,判断「惯例」是否已经渗进结论措辞。

    L1 网络状态:该主题尚未进入 L1 对应网络。CORRESPONDENCE_NETWORK 逐词检索「ICC / 门槛 / 功效 / TOST / 等效 / 二项 / 同向 / 投票 / 符号检验 / 异质性 / 判据」均无对应边;EDGE_TABLE_science-yixue 反抽表亦无此边——「方向投票的判据门槛出处」目前是网络外的方法学规则层议题,本文为该方法学缺口补上第一层标定。

    相关阅读

    证据等级: C
    参考文献: PMID 3450848, PMID 17161752, PMID 20974291, PMID 36758885, PMID 10412966


    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。

    83% How much of that line is regulation and how much is convention: tracing the origin of criterion thresholds and exploring alternatives

    Evidence level: C

    In studies of the same group using three modalities, four thresholds are applied to judge “pass/fail”: 10/12, ICC ≥ 0.3, correlation ≥ 0.5, and power ≥ 30%. Where do these come from? Tracing them reveals that three are conventions rather than regulations, and the null hypothesis itself is set too conservatively—switching to equivalence testing may make the “triangulation” conclusion form unattainable.

    Body

    In Qigong external energy research, “triangulation” is a common phrase: the same construct is measured simultaneously using four parameters (%AB, φ, X5, CoV) across 12 cells, and “verification passed” is declared if the rate of same-direction cells exceeds some threshold. Yet how this threshold was determined has rarely been questioned. This article traces the origins of these numbers to determine whether they are “regulations” (methodologically grounded with cited sources) or “conventions” (widely used without specific responsibility).

    First number: ICC ≥ 0.3 counts as “reliable.” In the measurement community, there is a complete traceable chain of sources for this. The most authoritative source is COSMIN (Consensus-based Standards for the selection of Measurement Instruments). COSMIN’s reliability rating standards define ICC ≥ 0.70 as “sufficient,” < 0.70 as “insufficient,” and no reporting as “uncertain” (PMID 17161752, Terwee 2007 published in J Clin Epidemiol titled “Quality criteria were proposed for measurement properties of health status questionnaires”). Note: 0.3 is not in COSMIN’s tiers at all. The widely circulated “0.3 passes” comes from an earlier psychometric literature’s rough stratification of “clinical vs research use” (research use allows relaxation), but this is not COSMIN’s decision criterion. Therefore, ICC ≥ 0.3 is a convention—it corresponds to the broad “research use” scope, not a judgment of whether “this measurement can be used.”

    Second number: correlation ≥ 0.5. This also traces back to COSMIN’s construct validity standards: tools correlated with those measuring the same construct ≥ 0.50 are considered sufficient (explicitly stated in the same COSMIN document, PMID 17161752). Thus “correlation 0.5” is the convergence criterion for construct validity, repurposed as the threshold for “two modalities agreeing in direction.” Semantically they are close relatives, but not identical. Convergent validity asks “do two tools measure the same thing,” whereas voting consistency asks “do two cells point in the same direction”; the latter has no COSMIN clause.

    Third number: power 30%. This is the most problematic. Power in any statistical standard should be as high as possible, with an accepted floor of 80% (a textbook-level regulation, not a convention—Cohen’s classic framework sets 0.80 as the “desired but often unattainable” target). Using 30% as a “decision threshold” misapplies “detection power” as the “pass line”: 30% actually means “only a 30% probability of detecting a true effect,” so setting this as the threshold allows a 70% chance of missing it. It is neither a regulation nor a convention; it reads more like a slip of the tongue.

    Fourth number: 10/12 (83% same-direction rate). This is the only number genuinely derived from binomial distribution: 12 cells, each independent, null hypothesis same-direction rate = 0.5, two-sided α = 0.05 yields significance at ≥10 same-direction cells (P ≈ 0.019). Its problem lies in the null hypothesis itself: binomial tests assume each cell is like “independently flipping a coin,” yet 12 cells measure the same construct, so a same-direction rate below 0.9 should not count as “verification passed.” Setting the null at 0.5 effectively includes weak evidence (same-direction rates 0.6–0.8) in the “possibly passed” interval. This is precisely the statistical root of accusation #1266 that “12 cells fold into 6.3 or even 4.8 cells”: the cells are not independent, so the overall binomial assumption is already discounted.

    What is the alternative? The cleanest option is equivalence testing (TOST, two one-sided tests). TOST’s original source is Schuirmann 1987 in J Pharmacokinet Biopharm (PMID 3450848): flipping the null hypothesis from “no difference” to “difference too large to be acceptable,” requiring rejection of both upper and lower one-sided tests to declare “equivalence.” Applied to voting criteria, this means: instead of asking “is the same-direction rate significantly greater than 0.5,” ask “does the same-direction rate fall significantly within an equivalence interval [0.7, 0.9]?” The width of the equivalence bounds directly determines how many of the 12 cells must align. If the equivalence bound is set at 0.1 (i.e., same-direction rate 0.8 ± 0.1 passes), the required number of cells would significantly exceed 10/12, possibly surpassing existing feasible limits—making the conclusion form “triangulation passed” itself unattainable, forcing a retreat to reporting each modality independently.

    Another layer: inconsistency in voting count methodology. Systematic review methodology literature contains empirical records: non-meta-analytic synthesis methods (including voting counts) are widely used yet rarely explained (PMID 36758885, 2023 J Clin Epidemiol survey), and voting counts perform poorly in heterogeneity testing under small study numbers (PMID 10412966, Am J Epidemiol 1999). Together these imply that using “direction voting” as the carrier of verification conclusions is methodologically unstable—this shares the same root cause as the dramatic power drop in sign tests under tie uncertainty (PMID 20974291): voting power is extremely sensitive to how “tie cells” are counted.

    In terms of research design, a pure methodological calibration (without human data) can be performed: sweep the null hypothesis p0 from 0.5 to 0.9, grid count from 6 to 65, replace the threshold from “exact binomial” to TOST, and output three comparison tables: (a) minimum same-direction grid count under different p0 values, (b) grid count needed for 80% power, (c) positions of the three zones “fail / not detected / pass.” Then anchor p0 using observed same-direction rates from ≥40 participants. Three core decision criteria emerge: if raising p0 to 0.7 makes 9/12 significant, the existing “7/12 fails” conclusion flips to “cannot decide” under the more reasonable null hypothesis, requiring overall re-evaluation of all prior failures based on p0 = 0.5; if the TOST equivalence bound of 0.1 requires more grids than currently feasible, the “triangulation passed” conclusion form becomes unattainable, necessitating independent per-modality reporting; if the three conventional numbers can be traced to methodological sources, then citation #1188’s self-reference should become specific GRADE/COSMIN clause citations.

    Evidence and Limitations

    Verified: The conservatism of binomial tests is supported by methodological literature—deficiencies of voting counts in small studies (PMID 10412966) and the impact of tie uncertainty on sign test power (PMID 20974291) are documented conclusions; COSMIN quality standards (ICC ≥ 0.70, correlation ≥ 0.50) have explicit sources (PMID 17161752).

    Partially verified: Equivalence testing (TOST) as the standard approach for “supporting no effect” has original papers (PMID 3450848) and subsequent methodological reviews (PMC5502906), but its applicability to direction-voting scenarios where “cells are not independent” relies only on theoretical derivation, without empirical calibration. The core decision criteria (how many grids after raising p0 to 0.7) require Monte Carlo simulation to implement.

    To be verified: ① the true distribution of same-direction rates across 12 cells—calibrate p0 by testing with ≥40 experienced participants to determine whether p0 should be 0.5 or 0.7/0.9; ② comparison table of required grids versus feasible limits when TOST equivalence bound is set at 0.1—can be produced via pure numerical simulation; ③ actual usage scope of the three conventional numbers (ICC 0.3, correlation 0.5, power 30%) in prior articles within the corpus—requires full corpus statistics to judge whether “conventions” have already seeped into conclusion wording.

    L1 Network Status: This topic has not yet entered the corresponding L1 network. CORRESPONDENCE_NETWORK word-by-word search for “ICC / threshold / power / TOST / equivalence / binomial / same-direction / voting / sign test / heterogeneity / decision criteria” yields no corresponding edges; EDGE_TABLE_science-yixue reverse-extraction table also lacks this edge—”decision threshold origins for direction voting” remains a methodological rule layer outside the network. This article fills the first layer of calibration for this methodological gap.

    Related Reading

    Evidence Level: C
    References: PMID 3450848, PMID 17161752, PMID 20974291, PMID 36758885, PMID 10412966


    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.

  • 那 12 个格子到底有几个是独立的:丹田三模态相关矩阵与中心内 ICC 的有效格数反解

    Evidence level: C

    三角验证的前提是”三个模态各自独立地摇头”;若它们其实在同步点头,12 个格子就只剩 6 个、甚至 4 个。

    正文

    上一篇给丹田呼吸的方向投票设计算了检验力:3 中心 × 4 模态 = 12 个格子,按二项检验去数”几个格子同向”。但那个计算藏着一个没有量过的假设——12 个格子彼此独立。格子一旦相关,”12″这个分母就要打折。折扣率是多少,取决于两件事:① 同一个人身上四个模态(%AB、φ、X5、CoV)之间到底有多像(模态间相关);② 同一个中心里的受试者到底有多像(中心内 ICC)。这两个数在丹田研究里从未被同一批人实测过,此前所有”12 格”的推算都只是代入假设值(ICC=0.3-0.5、相关=0.5)的结果。

    先看”相关会让样本打几折”这件事在方法学里的正式身份。聚类数据里有一个标准概念叫设计效应:只要同一簇里的个体互相像(ICC>0),真实的信息量就不是”人数”,而是”人数 ÷ 设计效应”,后者叫有效样本量(effective sample size)。经典推演给出的量级很具体:128 名受试者、簇内 ICC 只有 0.017(人类研究里常见的微小值),设计效应就已到 1.527,有效样本量直接掉到 84——功效从预期水平摔到 61%。ICC 越小,损失越温和;但”小”不等于”没有”。

    这个框架搬到丹田 12 格上,问题就变得尖锐:格子的”簇”是谁? 中心内的受试者互相像,格子的有效人数会缩水;模态之间的相关,则会让”格子数”本身缩水。后一种缩水更致命——它不是”每个格子里的样本变少了”,而是”4 个模态其实是 2 个模态在说话”。

    模态间相关高不高?已有的生理证据其实给出了方向。自主神经系统的多路信号(心血管、肺、外分泌、平滑肌)在静息态低频段呈现广泛的同步共波动,同一个时空模式贯穿脑-体轴,一次被唤醒的深快呼吸就能同时调动所有通路(PMID 39131291)。呼吸与脑全局信号的功能地形高度重叠,且这个重叠与行为指标相关——说明”呼吸一动、全身跟着动”不是测量噪声,而是生理常态(PMID 40419776)。心肺相互作用甚至直接改写心跳波形形态:左室射血时间与形态相似度指数随呼吸周期节律性起伏,与参考呼吸信号的回归 R² 高达 0.86 和 0.97(PMID 36502041)。把这三条叠起来看,%AB(呼吸分配)、φ(相位)、X5(电抗)、CoV(变异系数)共享同一个呼吸驱动源,模态间相关很可能不低——这正是 12 格折价的最危险情形。

    那么”相关已知时该怎么折算”有现成的统计工具吗?有。纵向聚类试验设计现在有专门的教程:三种相关结构(可交换、嵌套块、指数衰减)下,估算 ICC 与簇自相关系数有一套完整的模型框架和 R/SAS/Stata 实现(PMID 37196320)。更早的方法学文献给过”稀疏聚类”的边界:当簇内人数很少、ICC 又小,多层模型可能退化成单层模型而不损失太多(PMID 18621963),但稀疏聚类(每簇 2-6 人)恰恰是最需要小心设计导向法的场景(PMID 25110903)。而一旦测量在三个水平上嵌套(人/次/模态),效应量的拆分必须用三层模型,否则依赖效应的方差会被错误归到某一层(PMID 23055166)。这些工具存在的意义正是:折扣率不是靠”拍脑袋代入 0.5″来定的,是可实测、可反解的。

    回到 12 格。假设真的去测:≥40 名有经验者,同一次静息记录里同步导出四个量,每人隔日重复 3 次,用三项水平模型拆”人/次/模态”三层方差——得到模态间相关矩阵与中心内 ICC 之后,把折后有效格数代回上一篇的二项框架,重算门槛与功效。届时三种结局对应三种截然不同的处置:

    • 若实测 ICC ≥ 0.5 或模态间平均相关 ≥ 0.6:有效格数不足 5,”12 格设计”在任何真实同向率下功效都低于 20%。这等于宣布”7/12 判失败 / 12/12 判通过”这两句话在数学上不成立,必须从所有既往结论中撤回——三角验证的”三角”名存实亡。
    • 若 ICC 落在 0.2-0.5:按折后格数重写门槛表(很可能要求全票),并同步下调论文里”三角验证通过”的措辞强度。12 格可能只是”6 格说了 12 格的话”。
    • 若 ICC < 0.2 且模态间相关 < 0.3:12 格设计基本可用,但门槛必须从习惯的”过半”改成更严的 10/12;此前按过半判读、结论落在 7-9 格区间的,一律改判为”未检出”。

    无论落在哪一档,有一条底线是不变的:“三角验证通过”这句话的成立条件不是”三个模态都点头”,而是”三个模态可以各自独立地摇头”。 相关矩阵与 ICC 测的就是”独立摇头”的能力还剩多少。没有这一步实测,12 格只是一个好看的数字。

    证据与局限

    已验证: 聚类数据的有效样本量随设计效应收缩,ICC 仅 0.017 时 128 人即折为 84 人(经典设计效应推演);自主神经多通路信号存在广泛的同步共波动(PMID 39131291);呼吸与脑全局信号的功能地形显著重叠(PMID 40419776);心肺相互作用周期性改写心跳波形形态指标(PMID 36502041);纵向聚类试验的 ICC 与相关结构估算有成熟教程与实现(PMID 37196320)。

    部分验证: 稀疏聚类在簇人数极少时可能退化为单层模型(PMID 18621963),说明”折价”在 ICC 很小时不总是灾难;三层模型是拆分依赖效应量的标准框架(PMID 23055166)。

    待验证: ① 丹田四模态(%AB/φ/X5/CoV)在同一批人、同一次静息记录下的模态间相关矩阵实测值;② 中心内 ICC 实测值(多中心时再拆中心层);③ 折后有效格数代入二项框架后,10/12 与 7/12 两条线的真实功效;④ 模态间相关随练功年限、体位、采集时长是否漂移(相关矩阵本身是否稳定)。

    相关阅读

    证据等级: C
    参考文献: PMID 23055166, PMID 18621963, PMID 25110903, PMID 37196320, PMID 39131291, PMID 40419776, PMID 36502041

    L1 网络状态:「该主题尚未进入 L1 对应网络」——CORRESPONDENCE_NETWORK.md 节点表仅把「相位/节律」列为物理侧节点、「呼吸率/呼吸模式」列为生理侧观测项,均未与易学侧连边;边表 3.1–3.4 只登记五行↔五脏/五色/五音/五方/五季/五志、八卦↔五行、干支↔五行等传统对应边,无「丹田呼吸—模态间相关结构—有效格数」类边。EDGE_TABLE_science-yixue.md 仅有概念文章计数(丹田 51 篇 / 气功 48 篇 / 呼吸 189 篇),无模态间相关或 ICC 结构边。故本文封顶 C。


    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。

    How Many Independent Cells Are There in the Effective Grid of the Dantian Tri-modal Correlation Matrix and the Center Internal ICC?

    Evidence level: C

    Triangulation presupposes that “the three modalities each independently shake their heads”; if they are actually nodding in synchrony, the 12 cells collapse to just 6, or even 4.

    Main Text

    The previous post calculated statistical power for the Dantian breathing direction vote design: 3 centers × 4 modalities = 12 cells, counting “how many cells align” via binomial testing. However, that calculation hides an unmeasured assumption: the 12 cells are mutually independent. Once cells become correlated, the denominator “12” must be discounted. The discount rate depends on two factors: ① how similar the four modalities (%AB, φ, X5, CoV) are within the same individual (inter-modal correlation); and ② how similar subjects are within the same center (center-intra ICC). These two numbers have never been measured on the same cohort in Dantian research; all prior “12-cell” calculations merely substituted assumed values (ICC=0.3–0.5, correlation=0.5).

    Let us first address the formal identity of “how much does correlation discount the sample” within methodology. Clustered data have a standard concept called design effect: as long as individuals within the same cluster are similar (ICC > 0), true information is not “number of people,” but “number of people ÷ design effect”; the latter is termed effective sample size. Classic derivations yield specific magnitudes: with 128 subjects and an intra-cluster ICC of only 0.017 (a common small value in human studies), the design effect reaches 1.527, dropping the effective sample size directly to 84—power falling from expected levels to 61%. The smaller the ICC, the gentler the loss; yet “small” does not mean “none.”

    Transferring this framework to the Dantian 12 cells sharpens the problem: whose “cluster” are the cells? Subjects within a center are similar, shrinking the effective number of cells; inter-modal correlation shrinks the “number of cells” itself. The latter shrinkage is more fatal—it is not that “fewer samples exist within each cell,” but rather “the 4 modalities are actually just 2 modalities speaking.”

    How high is inter-modal correlation? Existing physiological evidence points the way. Multi-pathway signals of the autonomic nervous system (cardiovascular, pulmonary, exocrine, smooth muscle) exhibit widespread synchronous co-fluctuation in the low-frequency band during rest, with a single spatiotemporal pattern traversing the brain-body axis; one deep-fast breath awakening simultaneously activates all pathways (PMID 39131291). Functional topography of breathing and brain global signals overlaps heavily, and this overlap correlates with behavioral markers—indicating “breath moves, the whole body follows” is not measurement noise but physiological norm (PMID 40419776). Cardio-pulmonary interaction even directly rewrites heart waveform morphology: left ventricular ejection time and morphological similarity index fluctuate rhythmically with the breathing cycle, yielding regression R² of up to 0.86 and 0.97 against reference breathing signals (PMID 36502041). Stacking these three findings, %AB (breath allocation), φ (phase), X5 (reactance), and CoV (coefficient of variation) share the same breathing drive source, making inter-modal correlation likely non-negligible—this is the most dangerous scenario for discounting the 12 cells.

    When “correlation is known, how do we discount?” ready statistical tools exist. Longitudinal clustered trial designs now have dedicated tutorials: under three correlation structures (exchangeable, nested blocks, exponential decay), estimating ICC and cluster autocorrelation coefficients has a complete model framework and R/SAS/Stata implementations (PMID 37196320). Earlier methodology literature set boundaries for “sparse clustering”: when cluster size is small and ICC is low, multilevel models may degenerate into single-level models without much loss (PMID 18621963), yet sparse clustering (2–6 people per cluster) is precisely the scenario requiring careful design-oriented methods (PMID 25110903). Once measurement nests at three levels (person/time/modal), effect decomposition must use a three-level model; otherwise variance of the effect will be wrongly attributed to one layer (PMID 23055166). The existence of these tools means: the discount rate is not determined by “guessing 0.5,” but is measurable and reverse-solvable.

    Back to the 12 cells. Suppose we actually measure: ≥40 experienced subjects, synchronously exporting four quantities from one rest record, repeating every other day three times; use a three-level model to decompose “person/time/modal” variance—yielding the inter-modal correlation matrix and center-intra ICC. Plug the discounted effective cell count back into the previous binomial framework to recompute thresholds and power. Three outcomes correspond to three distinct treatments:

    • If measured ICC ≥ 0.5 or average inter-modal correlation ≥ 0.6: effective cells fall below 5; “12-cell design” yields power <20% under any realistic alignment rate. This declares the statements “7/12 fails / 12/12 passes” mathematically invalid; they must be withdrawn from all prior conclusions—the “triangulation” of triangulation is merely nominal.
    • If ICC falls between 0.2–0.5: rewrite the threshold table based on discounted cells (likely requiring unanimous votes), and simultaneously downgrade the wording intensity of “triangulation passed” in papers. The 12 cells may simply mean “6 cells spoke the words of 12.”
    • If ICC < 0.2 and inter-modal correlation < 0.3: the 12-cell design remains basically usable, but thresholds must shift from the customary “majority” to stricter 10/12; all prior readings based on majority, with conclusions falling in the 7–9 cell range, must be reclassified as “not detected.”

    Regardless of which tier applies, one bottom line remains unchanged: the condition for “triangulation passed” is not that “all three modalities nod,” but that “all three modalities can independently shake their heads.” The correlation matrix and ICC measure how much capacity for “independent shaking” remains. Without this measurement step, the 12 cells are merely an attractive number.

    Evidence and Limitations

    Verified: Effective sample size in clustered data contracts with design effect; ICC at 0.017 reduces 128 subjects to 84 (classic design effect derivation); widespread synchronous co-fluctuation exists across autonomic multi-pathway signals (PMID 39131291); functional topography of breathing and brain global signals shows significant overlap (PMID 40419776); cardio-pulmonary interaction periodically rewrites heart waveform morphology indicators (PMID 36502041); mature tutorials and implementations exist for ICC and correlation structure estimation in longitudinal clustered trials (PMID 37196320).

    Partially Verified: Sparse clustering may degenerate into single-level models when cluster size is very small (PMID 18621963), indicating “discounting” is not always catastrophic when ICC is low; three-level models provide the standard framework for decomposing effect variance (PMID 23055166).

    To Be Verified: ① measured inter-modal correlation matrix for the four Dantian modalities (%AB/φ/X5/CoV) on the same cohort during a single rest record; ② measured center-intra ICC (further decomposing the center layer for multi-center studies); ③ true power of the 10/12 and 7/12 lines after substituting discounted effective cell counts into the binomial framework; ④ whether inter-modal correlation drifts with years of practice, posture, and acquisition duration (stability of the correlation matrix itself).

    Further Reading

    Evidence Level: C
    References: PMID 23055166, PMID 18621963, PMID 25110903, PMID 37196320, PMID 39131291, PMID 40419776, PMID 36502041

    L1 Network Status: “This topic has not yet entered the L1 corresponding network”—the CORRESPONDENCE_NETWORK.md node table lists only “phase/rhythm” as physical-side nodes and “breathing rate/breathing pattern” as physiological-side observation items, with no edges connecting to Yijing Studies; Edge Table 3.1–3.4 registers only traditional correspondences such as Five Phases ↔ Five Organs/Five Colors/Five Tones/Five Directions/Five Seasons/Five Emotions, Eight Trigrams ↔ Five Phases, Heavenly Stems and Earthly Branches ↔ Five Phases, lacking edges like “Dantian breathing—inter-modal correlation structure—effective cell count.” EDGE_TABLE_science-yixue.md contains only concept article counts (Dantian 51 / Qigong 48 / Breathing 189), with no inter-modal correlation or ICC structure edges. Hence this article caps at C.


    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.

  • Δ 的方差里,有多少是「隔天再测」抖出来的:个体内与个体间方差成分拆分,与年限斜率还剩多少可测空间

    Evidence level: C

    「练得越久 Δ 越窄」这句话要成立,先得回答一个问题:同一个人隔天再测一次,Δ 能抖出多大动静?这个数字一天没测出来,年限斜率就一天说不清「可测空间」有多大。

    一、问题的形状:两个「已知数字」,没有一个来自 φ 或 %AB

    上一轮(#1181)论证年限斜率的功效陷阱时,引了两个关键数字:场次之间的变异平均相当于个体间变异的 25%–50%,与横断面比较中跨越 12–27 岁年龄范围的差异相当(PMID 15014087);以及个体斜率估计的方差由两部分构成——个体内变异(测量误差、随访长度与频率决定)与真实的个体间斜率差异(PMID 2362980)。

    这两个数字支撑了「场次噪声能伪造十几年功力」的核心担忧。但这里有一个必须戳破的缺口:这两个数字,没有一个是在胸腹相位角 φ 或腹式占比 %AB 上测出来的。PMID 15014087 测的是知觉—运动任务,PMID 2362980 是纵向研究功效的通用统计框架——它们给出的是「认知任务世界」的方差结构,不是「呼吸力学参数世界」的。

    而呼吸力学参数自己的方差结构,可能完全不同。同一批早产儿在静息睡眠里,φ 的个体内变异系数(intrasubject CV)可以横跨 11%–123%(PMID 24716708),而且作者明确归因于胸腹相位关系本身的不稳定,不是记录或分析方法造成的。如果 φ/%AB 的个体内成分占总量 60% 以上,那么年限斜率的可测空间几乎为零——这不是”样本量不够”能救的,是分子结构的问题。

    二、为什么必须做方差成分拆分,而不是直接算功效

    把总方差直接当功效公式的分母,等于假设「总方差 = 个体间方差」。但母文已经点破:横断面四层设计把个体内成分推到了最大——每人只测一次,个体内抖动和个体间差异在同一个数里无法分离,真实斜率差异被埋在下面。

    方差成分拆分(variance component decomposition)做的正是这件事:用线性混合模型把总方差拆成 个体间(between-person)、个体内/场次(within-person/session)、残余(residual) 三层。拆完之后的关键输出是两个数:

    1. 个体内成分占总方差的比例——它直接决定 ICC 的高低。个体内占比 >60%(ICC <0.4)时,按年限分层的任何斜率结论都落在噪声里。
    2. 设计效应(design effect = 1 + (m−1)×ICC)——它决定「重复测 m 次取均值」能把有效样本量放大多少。设计效应越大,同样的 n=120 能检出的斜率越小。

    这两个数互为表里:ICC 是方差的比值,design effect 是这个比值在设计里的放大倍数。不拆方差,两个都算不出来;而这两个数在 φ/%AB 上至今没有任何实测——这就是「先算功效」还是「先拆方差」的分水岭。

    三、φ 的个体内 CV 11%–123%:方差结构可能完全不同于认知任务

    PMID 24716708 的量级提示值得单独说。同一批受试者、同一场记录、同一个算法口径,个体内 CV 可以从 11% 到 123%——这不是”测量不准”,而是胸腹相位关系本身就是个不稳定的生理量(作者结论原话:variability arises from the instability of the rib cage and abdominal phase relationship, not from the recording and analytical methods)。

    把这个数字放进 #1181 的框架里,后果是具体的:如果 φ 的个体内成分在成人丹田练习者身上也占到 60% 以上,那么「场次噪声等量于十几年」在 φ 上字面成立——而 #1181 引用的 25%–50% 只是认知任务的数字,在 φ 上可能更高。反过来,若实测个体内占比 <30%(ICC >0.7),”噪声等量于十几年”在 φ 上不成立,#1181 的核心担忧可撤回。这正是方差成分拆分这个纯测量学实验的价值:它能把「担忧」从修辞变成可判定的数。

    四、设计:纯方差成分研究,零干预、零分组

    这个研究不需要干预、不需要分组、不涉及气功练习本身,只需要一份干净的重复测量数据:

    • ≥40 名健康成人,覆盖新手到 10 年以上(不按年限层等分,连续覆盖即可)
    • 同体位、同倾角(体位和倾角会改变 φ,必须锁死)
    • 隔日 ×3 + 隔周 ×2,共 5 次,每次 ≥10 分钟
    • 每次取稳定段算 %AB 与 φ,互相关法、固定单一算法口径(口径不同会让 φ 差出几十度)
    • 用线性混合模型拆三层方差,算个体内占比、ICC、design effect
    • 最后据拆分结果反推「四层年限设计 n=120 时的最小可检出斜率」

    五、核心判据:三档裁决

    1. 个体内成分 >60%(ICC <0.4):按年限分层的任何斜率结论都落在噪声内——现有横断面证据整体降级为「未可解释」,功效计算必须改用 ICC 校正后的有效样本量(n_eff = n / design effect)。
    2. 个体内占 30%–60%(ICC 0.4–0.7):斜率可测,但必须报 design effect 与有效 n,且重复测量次数须 ≥3 次取均值才允许进回归——单次测量不能进。
    3. 个体内 <30%(ICC >0.7):§二 所引的「噪声等量于十几年」在 φ 上不成立,#1181 的核心担忧可撤回,年限斜率按常规功效计算即可。

    六、与 #1255 的关系:一个数据的两次使用

    1254(方差成分拆分)与 #1255(retest gain 饱和曲线)共用同一批 40 人 × 5 次数据,零额外采集成本。方差成分拆分回答「Δ 里多少是隔天抖的」,retest gain 回答「抖的方向是不是单向抬升、几次饱和」。两者合起来才是完整的测量学体检:一个管方差结构(ICC 决定可测空间),一个管系统性偏移(retest 决定基线定义)。只拆方差不测 retest,会把”练习效应”当成”个体差异”;只测 retest 不拆方差,会把”噪声底”当成”可测空间”。

    证据与局限

    已验证: 场次间变异在知觉—运动任务上相当于个体间的 25%–50%、与跨 12–27 岁横断面差相当(PMID 15014087);斜率方差由个体内与个体间两成分构成(PMID 2362980);φ 在静息睡眠中个体内 CV 横跨 11%–123% 且源于相位关系不稳而非测量方法(PMID 24716708);纵向设计中 retest gain 与年龄-队列差异的混淆已被仿真证实(PMID 21639642)。

    部分验证: 「场次噪声等量于十几年」在认知任务域成立,但在 φ/%AB 上尚无任何直接测量——25%–50% 与 11%–123% 这两个数字来自不同人群与不同任务,无法直接外推到成人丹田练习者。

    待验证: φ/%AB 的个体内/个体间方差成分拆分(≥40 人 × 5 次、混合模型三层拆分);个体内占比三档判据(>60% / 30–60% / <30%)的落地数值;ICC 校正后四层年限设计的实际最小可检出斜率;design effect 在 φ 上的实测值。

    L1 网络状态: 「丹田」概念在 L1 对应网络中有 51 篇文章、最高证据级 A(代表 PMID 10593633,腹部呼吸体腔共振);但 「方差成分 / 个体内-个体间拆分 / ICC / design effect」这一测量学维度整条边缺失——CORRESPONDENCE_NETWORK.md 节点表三域只登记「观测什么量」,无任何位置承载「这个量能被估到多准、方差里多少是隔天再测抖出来的」;边表止于 E-050(六十甲子纳音↔频率,L1、open,现代证据栏「尚无任何可测来源」)。本条已登记 L1 gap 表(第 70 条,「丹田参数 ↔ 测量学维度」簇第 13 条)。

    相关阅读

    证据等级: C
    参考文献: PMID 15014087, PMID 2362980, PMID 24716708, PMID 21639642


    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。

    How much of the variance in Δ is due to ‘retest after one day’: decomposition of within-subject and between-subject variance components, and how much measurable space remains fo…

    Evidence level: C

    For the statement “the longer one practices, the narrower Δ becomes” to hold, we must first answer: if the same person is tested again the next day, how much movement can Δ produce? Until this number is measured, we cannot clarify how large the “measurable space” is for the tenure slope.

    I. The Shape of the Problem: Two “Known Numbers,” Neither from φ or %AB

    In the previous round (#1181), when arguing about the efficacy trap of the tenure slope, two key numbers were cited: between-session variability averages 25%–50% of between-person variability, and is comparable to differences across the 12–27 year age range in cross-sectional comparisons (PMID 15014087); and the variance of individual slope estimates consists of two components—within-person variability (determined by measurement error, follow-up length, and frequency) and true between-person slope differences (PMID 2362980).

    These two numbers support the core concern that “session noise can fake decades of practice.” However, there is a gap that must be broken here: neither of these numbers was measured on the thoraco-abdominal phase angle φ or the abdominal ratio %AB. PMID 15014087 measures perceptual-motor tasks, and PMID 2362980 is a general statistical framework for longitudinal study efficacy—they provide the variance structure of the “cognitive task world,” not the “respiratory mechanics parameter world.”

    The variance structure of respiratory mechanics parameters themselves may be entirely different. In the same cohort of preterm infants at rest during sleep, the within-subject coefficient of variation (CV) for φ can span 11%–123% (PMID 24716708), and the authors explicitly attribute this to instability in the thoraco-abdominal phase relationship itself, not to recording or analytical methods. If the within-person component of φ/%AB accounts for more than 60% of the total, then the measurable space for the tenure slope is nearly zero—this cannot be saved by “insufficient sample size”; it is a problem of molecular structure.

    II. Why Variance Component Decomposition Is Necessary, Rather Than Direct Power Calculation

    Treating total variance directly as the denominator of the power formula assumes “total variance = between-person variance.” But the original text already reveals: the cross-sectional four-level design pushes the within-person component to its maximum—each person is measured only once, so within-person jitter and between-person differences cannot be separated in the same number, burying true slope differences underneath.

    Variance component decomposition does exactly this: using linear mixed models, it splits total variance into between-person, within-person/session, and residual layers. The key outputs after decomposition are two numbers:

    1. The proportion of within-person component to total variance—this directly determines the ICC. When the within-person proportion exceeds 60% (ICC <0.4), any slope conclusion stratified by tenure falls into noise.
    2. The design effect (design effect = 1 + (m−1)×ICC)—this determines how much the effective sample size is amplified by “taking the mean of m repeated measurements.” The larger the design effect, the smaller the slope detectable with the same n=120.

    These two numbers are interdependent: ICC is the ratio of variances, and the design effect is the amplification factor of this ratio in the design. Without decomposing variance, neither can be calculated; and these two numbers have never been directly measured on φ/%AB—this is the watershed between “calculating power first” and “decomposing variance first.”

    III. φ’s Within-Subject CV of 11%–123%: Variance Structure May Differ Completely from Cognitive Tasks

    The magnitude reported in PMID 24716708 warrants separate discussion. For the same subjects, the same recording session, and the same algorithmic scope, the within-subject CV ranges from 11% to 123%—this is not “imprecise measurement,” but rather the thoraco-abdominal phase relationship itself is an unstable physiological quantity (the authors’ conclusion: “variability arises from the instability of the rib cage and abdominal phase relationship, not from the recording and analytical methods”).

    Placing this number into the framework of #1181 yields concrete consequences: if the within-person component of φ accounts for more than 60% in adult dantian practitioners, then “session noise equals decades” holds literally for φ—and the 25%–50% cited in #1181, being from cognitive tasks, may be even higher for φ. Conversely, if direct measurement shows the within-person proportion is less than 30% (ICC >0.7), then “noise equals decades” does not hold for φ, and the core concern of #1181 can be withdrawn. This is precisely the value of variance component decomposition as a pure measurement experiment: it transforms “concern” from rhetoric into determinable numbers.

    IV. Design: Pure Variance Component Study, Zero Intervention, Zero Grouping

    This study requires no intervention, no grouping, and does not involve qigong practice itself; it only needs clean repeated measurement data:

    • ≥40 healthy adults, covering novices to over 10 years (not evenly stratified by tenure, but continuously covered)
    • Same posture, same tilt angle (posture and tilt angle affect φ, so these must be locked)
    • Every other day ×3 + every week ×2, totaling 5 sessions, each ≥10 minutes
    • Calculate %AB and φ from stable segments each time, using mutual correlation method with a fixed single algorithmic scope (different scopes would cause φ to differ by tens of degrees)
    • Use linear mixed models to decompose the three layers of variance, calculating the within-person proportion, ICC, and design effect
    • Finally, reverse-calculate from the decomposition results to determine “the minimum detectable slope for a four-level tenure design with n=120”

    V. Core Criteria: Three-Tier Verdicts

    1. Within-person component >60% (ICC <0.4): Any slope conclusion stratified by tenure falls into noise—existing cross-sectional evidence overall downgrades to “uninterpretable,” and power calculation must use ICC-corrected effective sample size (n_eff = n / design effect).
    2. Within-person accounts for 30%–60% (ICC 0.4–0.7): Slope is measurable, but the design effect and effective n must be reported, and repeated measurements must average ≥3 times before entering regression—single measurements are not allowed.
    3. Within-person <30% (ICC >0.7): The “noise equals decades” claim cited in §II does not hold for φ, so the core concern of #1181 can be withdrawn, and the tenure slope can be calculated using conventional power methods.

    VI. Relationship with #1255: Two Uses of One Dataset

    1254 (variance component decomposition) and #1255 (retest gain saturation curve) share the same dataset of 40 people × 5 sessions, with zero additional acquisition cost. Variance component decomposition answers “how much of Δ is due to testing again the next day,” while retest gain answers “does the direction of ‘jitter’ unidirectionally rise, and does it saturate after several times.” Together they constitute a complete measurement examination: one manages variance structure (ICC determines measurable space), and the other manages systematic shift (retest determines baseline definition). Decomposing variance without measuring retest would mistake “practice effect” for “individual difference”; measuring retest without decomposing variance would mistake “noise floor” for “measurable space.”

    Evidence and Limitations

    Verified: Between-session variability in perceptual-motor tasks equals 25%–50% of between-person variability and matches cross-sectional differences across 12–27 years (PMID 15014087); slope variance consists of within-person and between-person components (PMID 2362980); φ’s within-subject CV spans 11%–123% at rest during sleep, attributed to phase relationship instability rather than measurement methods (PMID 24716708); retest gain confounding with age-cohort differences in longitudinal designs has been simulated and confirmed (PMID 21639642).

    Partially Verified: “Session noise equals decades” holds in the cognitive task domain, but no direct measurement exists for φ/%AB—the figures 25%–50% and 11%–123% come from different populations and tasks, so they cannot be directly extrapolated to adult dantian practitioners.

    Pending Verification: Decomposition of within-person/between-person variance components for φ/%AB (≥40 people × 5 sessions, three-layer mixed model); the three-tier thresholds for within-person proportion (>60% / 30–60% / <30%); the actual minimum detectable slope for the four-level tenure design after ICC correction; and the measured design effect for φ.

    L1 Network Status: The concept of “Dantian” corresponds to 51 articles in the L1 network, with the highest evidence level A (representing PMID 10593633, abdominal breathing cavity resonance); however, the entire measurement dimension of “variance component / within-person-between-person decomposition / ICC / design effect” is missing—the CORRESPONDENCE_NETWORK.md node table’s three domains only record “what quantity is observed,” with no position carrying “how accurately this quantity can be estimated, and how much of the variance is due to testing again the next day”; the edge table stops at E-050 (Sixty Jiazi Na Yin ↔ Frequency, L1, open, modern evidence column states “no measurable source yet”). This gap has been registered in the L1 gap table (item 70, “Dantian parameters ↔ measurement dimension” cluster item 13).

    Further Reading

    Evidence Level: C
    References: PMID 15014087, PMID 2362980, PMID 24716708, PMID 21639642


    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.