Evidence level: B
150 米这个数字不是风水说的,是从氡研究里借来的。拿一张印着断层线的地图去对一张吉凶标签,首先要问的不是”对不对得上”,而是”这把尺子的刻度,比要量的东西还粗”。
正文
一、两个不同物种的变量
风水评级是离散的、序数的、由人给出的;断层距离是连续的、以米为单位的、由地图或实测给出的。把它们放进同一个检验,第一步就已经做了两次”切”:把距离切成”150 米内/外”,把评级切成”吉/凶”。
PMID 16217841 说得很直白:把连续预测变量二分,代价是明显的功效损失和残余混杂,而用数据自己找出来的”最优切点”会带来严重偏倚。PMID 16675816 的标题更直接——”二分化连续变量的代价”。两次切叠加,损失不是相加,是相乘。
二、150 米是怎么”长”出来的
回到 PMID 26630035 原文:640 处地下室氡实测,用 Haversine 公式算到最近断层的距离,然后用 Kruskal-Wallis 秩和检验逐段扫描,找出断层”开始有显著影响”的距离——150 米。结果是真的:150 米内氡超标率(>200 Bq/m³)22.8%,150-700 米降到 15.2%;只看富铀基岩上的点,是 30.7% 对 17.5%。方向是真的,梯度也是真的。
但”150″这个数字本身,是扫描出来的一个最优分箱边界,不是物理常数。换一批房子、换一种基岩筛选、换一个分箱粒度,它会移动。把它当成一个可以拿去验证的”阈值”,是把估计值当成了真值——这正是 PMID 16217841 警告的那类偏倚的产生机制。
三、断层线在图上本来就有几十米误差
这是本篇最要紧的一句。1:50000 地质图上的断层线,位置精度通常在图上 0.5-1 毫米量级,折算到地面就是 25-50 米;再算上断层带本身有宽度、倾角不明、隐伏段只能推断,实际不确定度更大。
于是”150 米阈值”这个数字,和”这条线到底画在哪儿”的误差,是同一个量级。阈值的效应宽度,可能还小于测量噪声的宽度。 这不是说效应不存在,是说用这个精度去分辨 150 米内外的差别,尺子先不够用。PMID 16286689 讲的正是同一件事:氡潜势在地质单元之间和之内都有变异,只按单元贴一个标签,会漏掉单元内部的大半信息。
四、真要验断点,工具不是”分两组比均值”
断点存在性有专属工具:断点回归(RDD)与分段回归。PMID 25579639 系统综述了 RDD 在医学与公共卫生中的用法,指出这类设计长期被低估,并列出五项必须报告的内容,包括有效性条件检验与稳健性检查;PMID 25061922、27547695、41190057 是同一件事的不同版本教程。
RDD 最怕的前提在这里反而是满足的:一个宅子无法把自己”安排”到 149 米或 151 米,运行变量不可操纵。但 RDD 对断点附近的样本密度极度敏感——如果 140 到 160 米之间只有三五个宅点,断点真假你都测不出来。所以正确的产出不是一个 p 值,而是断点位置的置信区间,以及”这个断点是否可识别”的判断。如果剖面似然曲线是平的,结论就是”没有证据支持存在阈值”——这本身是一个结果,不是失败。
五、kappa 会被”凶地太少见”毁掉
一致性系数也埋着坑。PMID 8501467 早已把 kappa 拆成偏倚、患病率、观察一致三部分,明确建议:只报 kappa 会误导。PMID 29238424 与 38283651 展示的是同一个悖论——当某一类极度常见(比如九成宅点都是”吉”),两位评级者几乎全说”吉”时,观察一致率可以很高,而 kappa 趋近于零。
所以”kappa=0.3,说明风水评级不可靠”是错误推论,它可能只说明”凶地太稀有”。PMID 33145661 提出的信息一致性指标,就是把患病率的影响从一致性核心里剥离出来的一种做法。要报,就同时报偏倚指数、患病率指数、观察一致率,三者一起看。
六、那到底该怎么做
同一批宅点:①断层距离用同一版地质图加野外校核,把地图精度作为显式误差项写进模型;②风水评级不二分,用 5 级序数,盲法进行(评级者不知物探与氡结果);③主分析用分段回归估计断点位置与置信区间,RDD 做稳健性检查;④一致性同时报加权 kappa、偏倚/患病率分解与信息一致性;⑤把”150 米”作为待估参数放进模型,而不是已知条件。
一句话:这条线真正能回答的不是”150 米能不能预测凶地”,而是”断层距离与吉凶评级之间,到底有没有一个数据分辨得出的转折点”。前者是先射箭再画靶,后者才有被证伪的可能。
证据与局限
已验证: 断层距离与室内氡水平的存在梯度关系,150 米附近为数据驱动的显著边界(PMID 26630035);连续预测变量二分会导致功效损失与残余混杂,数据驱动切点引入严重偏倚(PMID 16217841、16675816);kappa 受患病率与偏倚影响,存在高一致低 kappa 的悖论(PMID 8501467、29238424、38283651);RDD 用于因果推断的有效条件与报告规范(PMID 25579639、25061922)。
部分验证: 断层带、富铀基岩与室内氡升高的空间关联在多个地质背景中复现(PMID 40819484、42322771、41012418),但各研究的”影响半径”不一致,提示阈值本身随地质背景变化;氡潜势在地质单元内部亦存在大幅变异(PMID 16286689、33992858、26042833)。
待验证: 断层距离与风水”吉/凶”评级之间是否存在可识别的断点,及其位置置信区间;1:50000 地质图断层线位置误差量级是否已接近或超过阈值效应宽度,导致阈值不可识别;风水评级的一致性在控制患病率后是否仍低于可接受水平。
相关阅读
- 风水吉凶评级、大地电导率与室内氡暴露:三变量联合盲法队列能证明什么
- 风水穴与断层、花岗岩体、土壤氡的距离结构
- 形峦凶地评级与断层距离/岩性/土壤氡/电导率的定量映射模型
- 风水盲法评级与室内氡的配对测量设计
- 研究池子:环境-生命频率接口
证据等级: B
参考文献: PMID 26630035, PMID 16217841, PMID 16675816, PMID 21062456, PMID 8501467, PMID 29238424, PMID 38283651, PMID 33145661, PMID 25579639, PMID 25061922, PMID 27547695, PMID 41190057, PMID 16286689, PMID 33992858, PMID 26042833, PMID 40819484, PMID 42322771, PMID 41012418
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。
Can a 150-meter fault predict Feng Shui \”unlucky sites\”: How to measure discrete labels of good or bad fortune with a continuous geological scale
Evidence level: B
The figure “150 meters” is not derived from Feng Shui; it is borrowed from radon research. When overlaying a map marked with fault lines against a label of auspiciousness or inauspiciousness, the first question should not be “does it match,” but rather “is the scale of this ruler coarser than the object being measured.”
Main Text
I. Variables from Two Different Species
Feng Shui ratings are discrete, ordinal, and human-assigned; fault distances are continuous, measured in meters, and derived from maps or field measurements. Placing them into the same test immediately involves two “cuts”: slicing distance into “within/without 150 meters” and slicing ratings into “auspicious/in Auspicious.”
PMID 16217841 states plainly: dichotomizing a continuous predictor incurs obvious loss of power and residual confounding, while using data-driven “optimal cut-points” introduces severe bias. The title of PMID 16675816 is even more direct—”The Cost of Dichotomizing a Continuous Variable.” With two cuts stacked, the loss is not additive but multiplicative.
II. How the “150 Meters” Was “Grown”
Returning to the original text of PMID 26630035: 640 basement radon measurements were taken, distances to the nearest fault were calculated using the Haversine formula, and a Kruskal-Wallis rank-sum test scanned segment by segment to find the distance where faults “begin to have a significant effect”—150 meters. The result is real: the rate of radon exceedance (>200 Bq/m³) within 150 meters is 22.8%, dropping to 15.2% between 150–700 meters; looking only at points on uranium-rich bedrock, it is 30.7% versus 17.5%. The direction is real; the gradient is real.
However, the number “150” itself is an optimal binning boundary scanned out of the data, not a physical constant. Changing a batch of houses, changing a bedrock filter, or changing the binning granularity would shift it. Treating it as a verifiable “threshold” treats an estimate as truth—precisely the mechanism of bias warned against in PMID 16217841.
III. Fault Lines on Maps Already Carry Tens of Meters of Error
This is the most critical sentence in this piece. On 1:50,000 geological maps, fault line positions typically have a precision on the order of 0.5–1 millimeters on the map, translating to 25–50 meters on the ground; adding the width of the fault zone, unknown dip angles, and inferred hidden segments increases the actual uncertainty further.
Thus, the “150-meter threshold” number and the error in “where exactly this line is drawn” are of the same magnitude. The width of the threshold effect may be smaller than the width of the measurement noise. This does not mean the effect is absent; it means using this precision to distinguish differences within versus outside 150 meters renders the ruler insufficient. PMID 16286689 addresses exactly this: radon potential varies both between and within geological units; assigning a single label to a unit misses a large portion of the information within that unit.
IV. If Verifying the Discontinuity, the Tool Is Not “Comparing Means in Two Groups”
There are specialized tools for discontinuity existence: Regression Discontinuity Design (RDD) and Piecewise Regression. PMID 25579639 provides a systematic review of RDD usage in medicine and public health, noting that such designs have been long underestimated and listing five required reporting items, including validity condition tests and robustness checks; PMID 25061922, 27547695, and 41190057 are different versions of tutorials on the same topic.
The premise that RDD fears most is actually satisfied here: a house cannot “arrange” itself to be 149 meters or 151 meters away; the running variable is non-manipulable. However, RDD is extremely sensitive to sample density near the discontinuity—if there are only three or four houses between 140 and 160 meters, the truth of the discontinuity cannot be measured. Therefore, the correct output is not a p-value, but a confidence interval for the discontinuity position and a judgment on whether “this discontinuity is identifiable.” If the profile likelihood curve is flat, the conclusion is “no evidence supports the existence of a threshold”—which itself is a result, not a failure.
V. Kappa Can Be Destroyed by “Too Few Inauspicious Sites”
The consistency coefficient also hides pitfalls. PMID 8501467 has long decomposed kappa into bias, prevalence, and observed agreement, explicitly advising: reporting only kappa misleads. PMID 29238424 and 38283651 demonstrate the same paradox—when one category is extremely common (e.g., 90% of sites are “auspicious”), two raters may almost always say “auspicious,” resulting in high observed agreement while kappa approaches zero.
Therefore, the inference “kappa=0.3 indicates Feng Shui ratings are unreliable” is a false deduction; it may only indicate “inauspicious sites are too rare.” PMID 33145661 proposes an information consistency index, which is a method of stripping the prevalence influence from the core of consistency. If reporting, simultaneously report bias index, prevalence index, and observed agreement; view all three together.
VI. So What Should Actually Be Done
For the same batch of sites: ① Use the same geological map with field verification for fault distance, explicitly writing map precision as an error term in the model; ② Do not dichotomize Feng Shui ratings; use a 5-level ordinal scale, conducted blindly (raters unaware of geophysical and radon results); ③ Use piecewise regression for the main analysis to estimate discontinuity position and confidence intervals, using RDD for robustness checks; ④ Report weighted kappa, bias/prevalence decomposition, and information consistency simultaneously for consistency; ⑤ Treat “150 meters” as a parameter to be estimated in the model, not a known condition.
In one sentence: The line truly answers not “Can 150 meters predict inauspicious sites,” but “Is there a data-resolvable turning point between fault distance and Feng Shui ratings.” The former is shooting the arrow before drawing the target; only the latter has the possibility of being falsified.
Evidence and Limitations
Verified: Existence of a gradient relationship between fault distance and indoor radon levels, with a data-driven significant boundary near 150 meters (PMID 26630035); dichotomizing continuous predictors leads to loss of power and residual confounding, while data-driven cut-points introduce severe bias (PMID 16217841, 16675816); kappa is affected by prevalence and bias, exhibiting a paradox of high agreement with low kappa (PMID 8501467, 29238424, 38283651); RDD validity conditions and reporting standards for causal inference (PMID 25579639, 25061922).
Partially Verified: Spatial associations between fault zones, uranium-rich bedrock, and elevated indoor radon are replicated across multiple geological backgrounds (PMID 40819484, 42322771, 41012418), but inconsistent “influence radii” across studies suggest the threshold itself varies with geological context; radon potential also exhibits substantial variation within geological units (PMID 16286689, 33992858, 26042833).
To Be Verified: Whether an identifiable discontinuity exists between fault distance and Feng Shui “auspicious/in Auspicious” ratings, and its position confidence interval; whether the magnitude of error in fault line positions on 1:50,000 geological maps has approached or exceeded the width of the threshold effect, rendering the threshold unidentifiable; whether Feng Shui rating consistency remains below an acceptable level after controlling for prevalence.
Related Reading
- Feng Shui Auspicious/Inauspicious Ratings, Ground Electrical Conductivity, and Indoor Radon Exposure: What Can a Three-Variable Joint Blind Cohort Prove
- Feng Shui Xue Sites and Faults, Granite Bodies, Soil Radon Distance Structures
- Quantitative Mapping Models for Xingsha Inauspicious Site Ratings and Fault Distance/Lithology/Soil Radon/Electrical Conductivity
- Blind Feng Shui Rating and Indoor Radon Paired Measurement Design
- Research Pool: Environment-Life Frequency Interface
Evidence Level: B
References: PMID 26630035, PMID 16217841, PMID 16675816, PMID 21062456, PMID 8501467, PMID 29238424, PMID 38283651, PMID 33145661, PMID 25579639, PMID 25061922, PMID 27547695, PMID 41190057, PMID 16286689, PMID 33992858, PMID 26042833, PMID 40819484, PMID 42322771, PMID 41012418
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.
Leave a Reply