5D Library › Power-Law Weighting and Bottom-Up Controllability in Multi-Tier Conditional Excitation Systems · 中文

Chapter 2 Variance Shares: The Author’s Weights and Three Ways to Split

The author’s weights

The author assigns the five tiers a set of top-down, decreasing “variance contributions”: C1C_1 50 %, C2C_2 25 %, C3C_3 12.5 %, C4C_4 7.5 %, C5C_5 5 %, and calls them a “binary power law”. The intuition is clear: the higher the tier, the more it weighs, and by multiples. This chapter first checks the numbers themselves and then asks a more basic question: in what sense can a tier be said to “contribute” a share of the variance?

Note

These weights are the author’s assumption, not an empirical estimate. No published study has measured all five tiers together and found these numbers; the evidence collected in Chapters 3 and 4 concerns only some measurable parts of some tiers. The book treats the weights as a model parameter open to discussion, not as a fact.

Calculation

The program lists the author’s weights, the strict binary sequence 2−k2^{-k} (k=1,…,5k = 1, \ldots, 5), a version renormalized to 100 %, and a version in which the last tier absorbs the remainder; it then compares them with a genuine power law k−αk^{-\alpha}. In an additive model on the log scale with independent tiers, tier kk’s share is σk2/∑jσj2\sigma_k^2/\sum_j \sigma_j^2, so the shares also fix the ratio between the standard deviations of neighbouring tiers.

# The author's "binary power-law" variance shares for the five layers (Chapter 2).
# A strict halving sequence 2^-k is compared with the author's 50/25/12.5/7.5/5, and with
# a true power law w_k proportional to k^-a. In an additive model on the log scale with
# independent layers, share_k = sigma_k^2 / sum sigma^2, so the shares fix the SD ratios.
import math

author = [50.0, 25.0, 12.5, 7.5, 5.0]
dyadic = [100 / 2 ** k for k in range(1, 6)]
s = sum(dyadic)
print("layer     author   strict 2^-k   2^-k renormalised   last layer takes the rest")
rest = dyadic[:4] + [100 - sum(dyadic[:4])]
for k in range(5):
    print(f"  C{k + 1}  {author[k]:9.2f} {dyadic[k]:13.3f} {100 * dyadic[k] / s:19.2f}"
          f" {rest[k]:27.3f}")
print(f"  sum {sum(author):8.2f} {s:13.3f} {100.0:19.2f} {sum(rest):27.3f}")
print("ratios of neighbouring shares:")
print("  author: " + "  ".join(f"{a / b:.2f}" for a, b in zip(author, author[1:])))
print("  2^-k:   " + "  ".join(f"{a / b:.2f}" for a, b in zip(dyadic, dyadic[1:])))
print("SD ratio between neighbouring layers implied (independent layers, log scale):")
print("  author: "
      + "  ".join(f"{math.sqrt(a / b):.3f}" for a, b in zip(author, author[1:])))
print(f"  2^-k:   {math.sqrt(2):.3f} at every step")
print("geometric (2^-k) versus power law (k^-a), both normalised to 100:")
for a in (1.0, 2.0):
    w = [k ** -a for k in range(1, 6)]
    print(f"  k^-{a:.0f}: " + "  ".join(f"{100 * x / sum(w):5.1f}" for x in w))
w = [2.0 ** -k for k in range(1, 6)]
print("  2^-k: " + "  ".join(f"{100 * x / sum(w):5.1f}" for x in w))
print("log-log slope of 2^-k between k = 1 and k = 5: "
      f"{(math.log(w[4]) - math.log(w[0])) / (math.log(5) - math.log(1)):.2f};"
      f" between k = 4 and 5: {(math.log(w[4]) - math.log(w[3])) / math.log(5 / 4):.2f}")
# layer     author   strict 2^-k   2^-k renormalised   last layer takes the rest
#   C1      50.00        50.000               51.61                      50.000
#   C2      25.00        25.000               25.81                      25.000
#   C3      12.50        12.500               12.90                      12.500
#   C4       7.50         6.250                6.45                       6.250
#   C5       5.00         3.125                3.23                       6.250
#   sum   100.00        96.875              100.00                     100.000
# ratios of neighbouring shares:
#   author: 2.00  2.00  1.67  1.50
#   2^-k:   2.00  2.00  2.00  2.00
# SD ratio between neighbouring layers implied (independent layers, log scale):
#   author: 1.414  1.414  1.291  1.225
#   2^-k:   1.414 at every step
# geometric (2^-k) versus power law (k^-a), both normalised to 100:
#   k^-1:  43.8   21.9   14.6   10.9    8.8
#   k^-2:  68.3   17.1    7.6    4.3    2.7
#   2^-k:  51.6   25.8   12.9    6.5    3.2
# log-log slope of 2^-k between k = 1 and k = 5: -1.72; between k = 4 and 5: -3.11

The check yields three points. First, the strict binary sequence is 50, 25, 12.5, 6.25 and 3.125, totalling 96.875 % and leaving 3.125 % unaccounted for; the author’s 7.5 and 5 are not binary: the first two neighbouring ratios are 2, the last two are 1.67 and 1.5. The author’s numbers are an approximation that makes the total 100 %. A strict halving sequence admits two self-consistent versions: renormalize it (51.6, 25.8, 12.9, 6.5, 3.2), or let the last tier take the remainder (50, 25, 12.5, 6.25, 6.25). Second, 2−k2^{-k} is a geometric sequence (decaying exponentially in the tier number kk), not a power law: a power law k−αk^{-\alpha} is a straight line on log–log axes, whereas 2−k2^{-k} gets steeper and steeper (average slope −1.72 from tier 1 to tier 5, −3.11 between tiers 4 and 5). A power law with α=1\alpha = 1 gives 43.8, 21.9, 14.6, 10.9 and 8.8, very different from the binary sequence in the lower tiers. Third, in an independent, log-additive model, “variance halving from tier to tier” means “standard deviation divided by 2≈1.41\sqrt{2} \approx 1.41 from tier to tier”: a concrete claim about the spread of each tier that could in principle be tested.

The book keeps the term “power-law weighting” that the author uses in the title, but in the text it calls the numbers “geometric weights” or “the author’s weights”; the text discusses the author’s original 50/25/12.5/7.5/5 and lists the other three sets alongside for comparison.

What variance shares depend on

Even when the variance of each tier is known, “how much each tier contributes” has a unique answer only if the tiers are uncorrelated. In reality they are likely to be correlated: people who grow up in a certain environment have priors that correlate with it; people with good reputations find it easier to enter good networks; knowledge investment affects reputation. Once there is correlation, the shared variance must be assigned to someone, and the assignment depends on the rule used.

Evidence

Methods for splitting relative importance. Lindeman, Merenda and Gold proposed computing, for each predictor in a regression, the increase in R2R^2 it brings in every possible order of entry, and averaging (the LMG method) (Lindeman et al. 1980). This equals the Shapley value of cooperative game theory: treat R2R^2 as the total payoff of a “coalition” of predictors and give each its average marginal contribution over all orderings (Shapley 1953). Grömping compared several variance-based estimators of relative importance and pointed out that they answer different questions; no single split is right for all purposes (Grömping 2007). For non-linear models, Sobol’ decomposed output variance into contributions of inputs and their interactions (Sobol’ indices) (Sobol’ 2001); Owen showed that for independent inputs the Shapley effect equals sharing each interaction’s variance equally among the inputs involved (Owen 2014).

Calculation

In the log-additive model ln⁡S=x1+⋯+x5\ln S = x_1 + \cdots + x_5, let the variances of the tiers equal the author’s weights exactly (if independent, the shares would be 50/25/12.5/7.5/5), then add some illustrative correlations: r(C1,C3)=0.4r(C_1, C_3) = 0.4 (prior correlated with the environment one grows up in), r(C3,C4)=0.5r(C_3, C_4) = 0.5, r(C4,C5)=0.4r(C_4, C_5) = 0.4, r(C1,C5)=0.3r(C_1, C_5) = 0.3. Compare four splits: each tier’s own variance as a share of the total; the R2R^2 gained when entering top-down; when entering bottom-up; and the Shapley value (LMG, averaged over all 120 orders).

# How much of the variance of ln S = x1 + ... + x5 does each layer "explain"? (Ch. 2)
# Layer SDs are set so that, IF the layers were independent, the shares would be the
# author's 50/25/12.5/7.5/5; then illustrative correlations are added. Four splits of R^2:
# own variance var(x_k)/var(ln S); R^2 gained entering top-down (C1 first); bottom-up
# (C5 first); Shapley / LMG = average gain over all 120 orders (Lindeman, Merenda & Gold).
import math, itertools

def solve(M, v):
    n = len(v); A = [row[:] + [v[i]] for i, row in enumerate(M)]
    for c in range(n):
        p = max(range(c, n), key=lambda r: abs(A[r][c])); A[c], A[p] = A[p], A[c]
        for r in range(n):
            if r != c: A[r] = [a - A[r][c] / A[c][c] * b for a, b in zip(A[r], A[c])]
    return [A[i][n] / A[i][i] for i in range(n)]

def shares(cov):
    n = len(cov); c = [sum(row) for row in cov]; vy = sum(c); memo = {}
    def R(S):                                 # R^2 of ln S regressed on the layers in S
        key = frozenset(S)
        if key not in memo:
            S = sorted(S)
            b = solve([[cov[i][j] for j in S] for i in S], [c[i] for i in S]) if S else []
            memo[key] = sum(bi * c[i] for bi, i in zip(b, S)) / vy
        return memo[key]
    def seq(order):
        out, have = [0.0] * n, []
        for k in order: out[k] = R(have + [k]) - R(have); have.append(k)
        return out
    orders = list(itertools.permutations(range(n)))
    shap = [sum(g) / len(orders) for g in zip(*(seq(o) for o in orders))]
    return ([cov[k][k] / vy for k in range(n)], seq(range(n)), seq(range(n - 1, -1, -1)),
            shap)

def covariance(sd, corr):
    R = [[1.0 if i == j else corr.get((min(i, j), max(i, j)), 0.0) for j in range(5)]
         for i in range(5)]
    return [[R[i][j] * sd[i] * sd[j] for j in range(5)] for i in range(5)]

SD = [math.sqrt(w) for w in (0.50, 0.25, 0.125, 0.075, 0.05)]
CORR = {(0, 2): 0.4, (2, 3): 0.5, (3, 4): 0.4, (0, 4): 0.3}

if __name__ == "__main__":
    for title, corr in (("independent layers", {}), ("correlated layers: " + ", ".join(
            f"r(C{i + 1},C{j + 1}) = {r}" for (i, j), r in CORR.items()), CORR)):
        own, td, bu, sh = shares(covariance(SD, corr))
        print(title)
        print("  layer  own var  top-down  bottom-up  Shapley/LMG")
        for k in range(5):
            print(f"  C{k + 1}   {own[k]:7.1%} {td[k]:9.1%} {bu[k]:10.1%} {sh[k]:12.1%}")
        print(f"  sum   {sum(own):7.1%} {sum(td):9.1%} {sum(bu):10.1%} {sum(sh):12.1%}")
# independent layers
#   layer  own var  top-down  bottom-up  Shapley/LMG
#   C1     50.0%     50.0%      50.0%        50.0%
#   C2     25.0%     25.0%      25.0%        25.0%
#   C3     12.5%     12.5%      12.5%        12.5%
#   C4      7.5%      7.5%       7.5%         7.5%
#   C5      5.0%      5.0%       5.0%         5.0%
#   sum    100.0%    100.0%     100.0%       100.0%
# correlated layers: r(C1,C3) = 0.4, r(C3,C4) = 0.5, r(C4,C5) = 0.4, r(C1,C5) = 0.3
#   layer  own var  top-down  bottom-up  Shapley/LMG
#   C1     34.7%     58.2%      20.0%        39.2%
#   C2     17.4%     17.4%      17.4%        17.4%
#   C3      8.7%     13.7%      33.4%        22.3%
#   C4      5.2%      8.8%       8.6%        10.3%
#   C5      3.5%      2.0%      20.6%        10.9%
#   sum     69.4%    100.0%     100.0%       100.0%
Variance shares of the five tiers under five rules. Grey: the author’s weights (with independent tiers every rule gives these numbers). The other four come from the same correlated model: top-down order, bottom-up order, Shapley value (LMG), and the first-order Sobol’ index of the multiplicative model on the raw scale (log variance 1, independent tiers).

With independent tiers, the four methods give identical numbers. Once correlation is added they part ways: C1C_1’s share is 58.2 % top-down, 20.0 % bottom-up and 39.2 % by the Shapley value; C5C_5’s is only 2.0 % top-down, 20.6 % bottom-up and 10.9 % by Shapley. The “own variance” shares add up to only 69.4 %, because correlation makes the total variance larger than the sum of the tiers’ variances.

This shows something important: a top-down split systematically assigns the shared variance to the upper tiers. If one first decides that “endowment matters most” and then enters endowment first in a regression, the large share it receives is to a great extent an artefact of the order. The Shapley value treats all orders alike and is a fairer convention, but it is still a convention, not a law of nature.

Scale: logarithmic or raw

The decomposition above is on the log scale. On the raw scale the multiplicative model is not additive: S=f1f2⋯f5S = f_1 f_2 \cdots f_5 contains interactions between tiers, and the same model gives different shares on a different scale.

Calculation

Let the ln⁡fk\ln f_k be independent with variances wkVw_k V, where wkw_k are the author’s weights and VV is the total variance of ln⁡S\ln S. On the log scale the shares are exactly wkw_k. On the raw scale, the Sobol’–Hoeffding decomposition gives the first-order index (each tier acting alone), the total-effect index (including every interaction it takes part in) and the Shapley effect (each interaction split equally among its participants), for spreads V=0.25V = 0.25, 1 and 2 (the table lists C1C_1 and C5C_5).

# The same multiplicative model, decomposed on two scales (Chapter 2).
# S = f1 f2 f3 f4 f5 with independent ln f_k ~ N(0, w_k V): on the log scale the shares
# are exactly the weights w_k (author's 50/25/12.5/7.5/5). On the raw scale S itself has
# interactions. For independent inputs the Sobol'-Hoeffding (ANOVA) term of a set u of
# layers has variance V_u = prod_{j not in u} mu_j^2 * prod_{j in u} v_j, where mu_j and
# v_j are the mean and variance of f_j. First-order index: V_{k} / Var S; total index:
# sum of V_u over u containing k; Shapley effect (Owen 2014): each V_u split equally
# among the |u| layers it involves.
import math, random, itertools

W = [0.50, 0.25, 0.125, 0.075, 0.05]

def indices(V):
    s2 = [w * V for w in W]
    mu = [math.exp(s / 2) for s in s2]
    var = [math.exp(2 * s) - math.exp(s) for s in s2]
    terms = {}
    for r in range(1, 6):
        for u in itertools.combinations(range(5), r):
            terms[u] = math.prod(var[j] if j in u else mu[j] ** 2 for j in range(5))
    tot = sum(terms.values())
    first = [terms[(k,)] / tot for k in range(5)]
    total = [sum(v for u, v in terms.items() if k in u) / tot for k in range(5)]
    shap = [sum(v / len(u) for u, v in terms.items() if k in u) / tot for k in range(5)]
    return first, total, shap, tot, 1 - sum(first)

if __name__ == "__main__":
    print("V = Var(ln S); raw scale: first-order, total and Shapley indices of C1 and C5")
    print("   V     Var(S)  interactions | C1: first  total  Shapley | C5: first  total"
          "  Shapley")
    for V in (0.25, 1.0, 2.0):
        first, total, shap, tot, inter = indices(V)
        print(f"  {V:4.2f}  {tot:8.3f} {inter:11.1%}   |"
              + "  |".join(f"    {first[k]:6.1%} {total[k]:6.1%} {shap[k]:7.1%}"
                           for k in (0, 4)))
    print("log-scale shares (all V): " + " ".join(f"{w:.1%}" for w in W))
    rng = random.Random(41); V = 1.0
    xs = [math.prod(math.exp(rng.gauss(0, math.sqrt(w * V))) for w in W)
          for _ in range(200000)]
    m = sum(xs) / len(xs); mc = sum((x - m) ** 2 for x in xs) / len(xs)
    print(f"Monte Carlo check (V = 1, 200,000 draws): Var(S) = {mc:.3f}"
          f" (exact {indices(1.0)[3]:.3f})")
# V = Var(ln S); raw scale: first-order, total and Shapley indices of C1 and C5
#    V     Var(S)  interactions | C1: first  total  Shapley | C5: first  total  Shapley
#   0.25     0.365        8.1%   |     46.9%  53.1%   50.0%  |      4.4%   5.6%    5.0%
#   1.00     4.671       30.5%   |     37.8%  62.2%   49.3%  |      3.0%   7.7%    5.1%
#   2.00    47.209       54.3%   |     26.9%  73.1%   47.5%  |      1.6%  11.0%    5.4%
# log-scale shares (all V): 50.0% 25.0% 12.5% 7.5% 5.0%
# Monte Carlo check (V = 1, 200,000 draws): Var(S) = 4.753 (exact 4.671)

With little spread (V=0.25V = 0.25), interactions account for only 8.1 % of the raw-scale variance and all three indices of C1C_1 and C5C_5 are close to the author’s weights. With more spread, interactions grow fast: 30.5 % at V=1V = 1 and 54.3 % at V=2V = 2. Then C1C_1’s first-order index is only 26.9 % while its total-effect index is 73.1 %; the difference is C1C_1’s interaction with the other tiers, and there is no unique answer to whom an interaction “belongs”. The Shapley effect splits interactions equally and stays close to the log-scale weights (47.5 %). The Monte Carlo variance of SS, 4.753, agrees with the exact 4.671 (variance estimates of log-normal variables converge slowly).

For the author’s framework this has a direct implication: in a multiplicative model, “the upper tiers contribute a large share of variance” and “the lower tiers’ effects depend on the upper tiers” are two sides of the same coin. The stronger the interactions, the more “how much this tier contributes” depends on how one assigns the interactions, and the less it resembles a quantity that can be measured on its own.

Hypothesis · not established

The geometric weights hypothesis (the author’s hypothesis, not established). For log career outcomes, the Shapley shares of the five tiers decline roughly geometrically (neighbouring ratios close to 2). Testable prediction: in a large sample with reasonable measures of all five tiers, with the measures and the Shapley split registered in advance, the confidence intervals of the neighbouring ratios include 2 and the ordering matches the author’s. If the ordering of the shares differs clearly from the author’s (for example, environment larger than the prior) or the ratios are far from 2, the hypothesis is rejected. One caution must be stated in advance: shares depend on measurement reliability, and a tier measured more crudely receives a smaller share; the protocol must deal with that bias.

Chapter summary

The author’s 50/25/12.5/7.5/5 is not a strict binary sequence (which sums to 96.875 %), and 2−k2^{-k} is a geometric sequence rather than a power law; the numbers are the author’s assumption, not an empirical estimate. With independent tiers every splitting method gives the same shares; once the tiers correlate, the shares depend on the rule, a top-down order systematically favours the upper tiers, and the Shapley value (LMG) is the convention of averaging over all orders. On the raw scale the multiplicative model has interactions; the greater the spread, the more they matter and the more the shares depend on scale and convention. The next chapter asks what published research can say about endowment and timing.

References

Grömping, Ulrike. 2007. “Estimators of Relative Importance in Linear Regression Based on Variance Decomposition.” The American Statistician 61 (2): 139–47. https://doi.org/10.1198/000313007X188252.
Lindeman, Richard H., Peter F. Merenda, and Ruth Z. Gold. 1980. Introduction to Bivariate and Multivariate Analysis. Scott, Foresman.
Owen, Art B. 2014. “Sobol’ Indices and Shapley Value.” SIAM/ASA Journal on Uncertainty Quantification 2 (1): 245–51. https://doi.org/10.1137/130936233.
Shapley, Lloyd S. 1953. “A Value for nn-Person Games.” In Contributions to the Theory of Games, Volume II, edited by Harold W. Kuhn and Albert W. Tucker. Annals of Mathematics Studies 28. Princeton University Press. https://doi.org/10.1515/9781400881970-018.
Sobol’, Ilya M. 2001. “Global Sensitivity Indices for Nonlinear Mathematical Models and Their Monte Carlo Estimates.” Mathematics and Computers in Simulation 55 (1–3): 271–80. https://doi.org/10.1016/S0378-4754(00)00270-6.

Fan, Y. (2026). Power-Law Weighting and Bottom-Up Controllability in Multi-Tier Conditional Excitation Systems: A Mathematical Synthesis of Endogenous Priors, Timing Windows, Network Topology and Reputational Games [多层条件激发系统中的幂律权重与逆向控制可及性] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23810093 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).