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

Chapter 1 Products and Cascades: The Conditional Excitation Equation

Five conditions, two ways to combine them

Imagine an outcome SS (the earnings of a career, the output of a research programme, the survival of a small firm) that depends on five conditions: the endogenous prior C1C_1, timing and cycles C2C_2, environmental topology C3C_3, reputational games C4C_4 and knowledge investment C5C_5. The two simplest ways to write it are

Sadd=15∑i=15fi(Ci),Smult=∏i=15fi(Ci).S_{\text{add}} = \frac{1}{5}\sum_{i=1}^{5} f_i(C_i), \qquad S_{\text{mult}} = \prod_{i=1}^{5} f_i(C_i).

The additive model says the tiers can compensate for one another: if one is weak, a little more of another makes up for it. The multiplicative model says each tier is indispensable: if any one tier is close to zero, so is the product, however strong the others. The author calls the second structure “conditional excitation”: only when the condition of one tier is met does the next tier get a chance to act.

Note

This is the author’s conceptual model. The division into five tiers, the meaning of each and the multiplicative combination are the author’s modelling choices, not empirical findings. The “five-tier tensor product” of the framework is, mathematically, the ordinary product of five numbers: a tensor product is an operation between vector spaces and is not needed to multiply five scalars. Only if each tier were itself a vector (say, “environment” with city, industry and family-network components) would a more general form be needed; this book does not attempt that generalization.

Three properties of a product

The multiplicative model has three properties that can be derived exactly.

First, the weakest link. If fk→0f_k \to 0 then Smult→0S_{\text{mult}} \to 0, whatever the other tiers. The partial derivative is ∂S/∂fi=S/fi\partial S/\partial f_i = S/f_i: the marginal value of a tier is inversely proportional to its own size, so the weakest tier has the largest marginal value.

Second, log-additivity. ln⁡Smult=∑iln⁡fi\ln S_{\text{mult}} = \sum_i \ln f_i. On the log scale the multiplicative model is an additive one, so the “contribution of each tier” can be discussed on the log scale by decomposing variance (Chapter 2). The elasticity ∂ln⁡S/∂ln⁡fi=1\partial \ln S / \partial \ln f_i = 1 is the same for every tier: the same percentage improvement has the same effect wherever it occurs, while the same absolute improvement has the largest effect in the weakest tier.

Third, a right-skewed distribution. If the ln⁡fi\ln f_i are independent with finite variance, ln⁡S\ln S is a sum of random variables and tends to a normal distribution by the central limit theorem, so SS is approximately log-normal: most people are in the middle and a few are far above average.

Evidence

Shockley’s multiplicative model. In 1957 the physicist William Shockley analysed publication counts of scientists in several research institutions and found the distribution of publication rates to be highly right-skewed and roughly log-normal. His explanation was that publishing a paper requires clearing a series of hurdles: thinking of a good problem, being able to work on it, recognizing a worthwhile result, deciding when to stop, writing adequately, profiting from criticism, being determined to submit, and persisting in making changes. If the probability of clearing hurdle ii is pip_i, output is proportional to p1p2⋯p8p_1 p_2 \cdots p_8; someone 50 % above average on each of the eight factors would produce about 25 times the average (Shockley 1957). The idea that products produce log-normal distributions goes back to Gibrat’s work on firm sizes (Gibrat 1931); for the wide occurrence of log-normal distributions across the natural and social sciences see the review by Limpert and colleagues (Limpert et al. 2001).

The economics of multiplicative production. Kremer’s “O-ring” theory writes production as the product of the qualities of many tasks: if any one step fails, the value of the whole product is badly reduced, so highly skilled workers tend to be matched with one another (Kremer 1993).

Is individual performance normal? Pooling 198 samples covering 633,263 researchers, entertainers, politicians and athletes, O’Boyle and Aguinis argued that individual performance follows a Paretian (power-law) rather than a normal distribution (O’Boyle and Aguinis 2012). Beck, Beatty and Sackett replied that much of the departure from normality is an artefact of how performance is measured: when the measures meet conditions such as covering the full range of performance, a normal distribution fits better (Beck et al. 2014). So “performance is highly skewed” is itself a disputed empirical question.

Shockley’s model and the author’s five-tier model have the same structure: several conditions multiplied together. But Shockley’s eight hurdles were an explanatory assumption; he did not measure the eight factors separately and then show that they multiply. A log-normal distribution is compatible with “several factors multiplied”, but it cannot prove in reverse that the factors are these eight, or that they must multiply.

Calculation

The program first checks Shockley’s arithmetic and then simulates 100,000 invented people: each has five factors with ln⁡fi\ln f_i independently normal with mean 0 and standard deviation 0.35 (illustrative values), combined additively and multiplicatively. Finally it looks at a person with four strong tiers and one weak one.

# Additive versus multiplicative combination of five layer scores (the author's model,
# Chapter 1). Each person has five positive factors f1..f5 (one per layer); success is
#   additive:        S_add  = (f1 + f2 + f3 + f4 + f5) / 5
#   multiplicative:  S_mult = f1 * f2 * f3 * f4 * f5,   so  ln S_mult = sum of ln f_i
# The factors are simulated as ln f_i ~ Normal(0, 0.35^2), independent. Illustrative only.
import math, random
from statistics import mean, median

def top_share(xs, q):
    xs = sorted(xs, reverse=True)
    k = max(1, int(len(xs) * q))
    return sum(xs[:k]) / sum(xs)

def skew(xs):
    m = mean(xs); s = math.sqrt(sum((x - m) ** 2 for x in xs) / len(xs))
    return sum(((x - m) / s) ** 3 for x in xs) / len(xs)

if __name__ == "__main__":
    print("Shockley's arithmetic: 8 factors, each 50 % above average")
    print(f"  multiplicative: {1.5 ** 8:.1f} x average (+{100 * (1.5 ** 8 - 1):,.0f} %);"
          f" additive: 1.5 x average (+50 %)")
    rng = random.Random(41)
    people = [[math.exp(rng.gauss(0, 0.35)) for _ in range(5)] for _ in range(100000)]
    add = [sum(f) / 5 for f in people]
    mul = [math.prod(f) for f in people]
    print("100,000 simulated people, ln f_i ~ N(0, 0.35^2)")
    print("  model          mean/median  skewness  top 1 % share  top 20 % share")
    for name, xs in (("additive", add), ("multiplicative", mul)):
        print(f"  {name:14s} {mean(xs) / median(xs):11.2f} {skew(xs):9.2f}"
              f" {top_share(xs, 0.01):13.1%} {top_share(xs, 0.20):15.1%}")
    lm = [math.log(x) for x in mul]
    print(f"  ln S_mult: mean {mean(lm):+.3f}, skewness {skew(lm):+.3f}"
          f" (a sum of 5 normals: lognormal S)")
    print("weakest link: layers (0.9, 0.9, 0.9, 0.9, 0.1)")
    base = [0.9, 0.9, 0.9, 0.9, 0.1]
    for i, lab in ((4, "weakest layer 0.1 -> 0.2"), (0, "a strong layer 0.9 -> 1.0")):
        up = base[:]; up[i] += 0.1
        print(f"  {lab}: multiplicative x{math.prod(up) / math.prod(base):.2f},"
              f" additive x{sum(up) / sum(base):.3f}")
    print(f"  product {math.prod(base):.3f} vs average {mean(base):.2f};"
          f" any layer -> 0 sends the product to 0")
# Shockley's arithmetic: 8 factors, each 50 % above average
#   multiplicative: 25.6 x average (+2,463 %); additive: 1.5 x average (+50 %)
# 100,000 simulated people, ln f_i ~ N(0, 0.35^2)
#   model          mean/median  skewness  top 1 % share  top 20 % share
#   additive              1.01      0.51          1.5%           24.8%
#   multiplicative        1.36      3.36          6.1%           47.8%
#   ln S_mult: mean -0.000, skewness +0.005 (a sum of 5 normals: lognormal S)
# weakest link: layers (0.9, 0.9, 0.9, 0.9, 0.1)
#   weakest layer 0.1 -> 0.2: multiplicative x2.00, additive x1.027
#   a strong layer 0.9 -> 1.0: multiplicative x1.11, additive x1.027
#   product 0.066 vs average 0.74; any layer -> 0 sends the product to 0
The same five factors combined in two ways. Left: the multiplicative combination produces a long right tail (share of people above S on log–log axes), the additive combination almost none. Right: for a person with four tiers at 0.9 and one at 0.1, the gain from adding 0.1 to each tier; in the multiplicative model the gain in the weakest tier is 100 %, in the additive model it is the same for all five.

The program shows several things. 1.58≈25.61.5^8 \approx 25.6, about 2,463 % above average, so Shockley’s arithmetic is right. With the same factors, the additive combination has a mean-to-median ratio of about 1.01, skewness 0.51, and the top 1 % hold 1.5 % of the total; the multiplicative combination has a mean 1.36 times the median, skewness 3.36, the top 1 % hold 6.1 % and the top 20 % hold 47.8 %. The logarithm of the product has skewness close to zero, exactly as log-normality requires. For the person with four tiers at 0.9 and one at 0.1, raising the weakest tier from 0.1 to 0.2 doubles the multiplicative outcome, while raising a strong tier from 0.9 to 1.0 adds only 11 %; in the additive model both add only 2.7 %.

The conditional-probability chain

The other form in the author’s framework is a chain of conditional probabilities:

P(S)=P(C1)P(C2∣C1)P(C3∣C1,C2)P(C4∣C1,C2,C3)P(C5∣C1,…,C4).P(S) = P(C_1)\, P(C_2 \mid C_1)\, P(C_3 \mid C_1, C_2)\, P(C_4 \mid C_1, C_2, C_3)\, P(C_5 \mid C_1, \ldots, C_4).

Here each tier is a met / not-met event and SS means all five are met. The formula is the multiplication rule of probability: it holds for any joint distribution and in any order; with C5C_5 placed first, it can be written as a chain just as well. The chain therefore implies no hierarchy with “endowment above, effort below”; the hierarchy is the author’s assumption about causal order and needs separate evidence.

Calculation

The program generates the five yes/no tiers of 200,000 invented people from correlated latent scores, computes the conditional probabilities along the chain top-down and bottom-up, and then uses an illustrative set of conditional probabilities to compare raising each tier by 0.10 in absolute terms with raising it by 10 % of its value.

# The conditional-probability chain P(S) = P(C1) P(C2|C1) ... P(C5|C1..C4) (Chapter 1).
# Part 1: the chain rule is an identity. Five yes/no layers are generated from correlated
# latent scores; the chain is computed top-down (C1 first) and bottom-up (C5 first).
# Both products equal P(all five), but the factors differ: order is a modelling choice.
# Part 2: marginal gains. dP/dp_i = P / p_i, so a fixed absolute gain is worth most at the
# smallest factor, while a fixed proportional gain is worth the same at every factor.
import math, random

def latent_people(n, rho, thresholds, seed):
    # layer k is "met" when z_k > threshold_k; z_k = sqrt(rho) common + sqrt(1-rho) own
    rng = random.Random(seed); out = []
    for _ in range(n):
        c = rng.gauss(0, 1)
        a, b = math.sqrt(rho), math.sqrt(1 - rho)
        z = [a * c + b * rng.gauss(0, 1) for _ in thresholds]
        out.append(tuple(int(zk > t) for zk, t in zip(z, thresholds)))
    return out

def chain(people, order):
    factors, keep = [], people
    for k in order:
        met = [p for p in keep if p[k]]
        factors.append(len(met) / len(keep)); keep = met
    return factors

if __name__ == "__main__":
    ppl = latent_people(200000, 0.3, [0.0, 0.2, 0.4, 0.6, -0.4], seed=41)
    p_all = sum(all(p) for p in ppl) / len(ppl)
    print(f"200,000 simulated people; P(all five layers met) = {p_all:.4f}")
    for name, order in (("top-down  C1..C5", range(5)),
                        ("bottom-up C5..C1", range(4, -1, -1))):
        f = chain(ppl, list(order))
        print(f"  {name}: factors " + " ".join(f"{x:.3f}" for x in f)
              + f"  product {math.prod(f):.4f}")
    p = [0.6, 0.5, 0.4, 0.3, 0.7]
    P = math.prod(p)
    print(f"illustrative chain factors {p}, P(S) = {P:.4f}")
    print("  layer  p_i   dP/dp_i   P after +0.10   P after +10 % of p_i")
    for i, pi in enumerate(p):
        a = p[:]; a[i] += 0.10
        r = p[:]; r[i] *= 1.10
        print(f"  C{i + 1}   {pi:.2f}  {P / pi:8.4f}  {math.prod(a):13.4f}"
              f"  {math.prod(r):18.4f}")
    print("  elasticity d ln P / d ln p_i = 1 for every layer")
# 200,000 simulated people; P(all five layers met) = 0.0596
#   top-down  C1..C5: factors 0.501 0.517 0.514 0.499 0.897  product 0.0596
#   bottom-up C5..C1: factors 0.656 0.330 0.522 0.673 0.785  product 0.0596
# illustrative chain factors [0.6, 0.5, 0.4, 0.3, 0.7], P(S) = 0.0252
#   layer  p_i   dP/dp_i   P after +0.10   P after +10 % of p_i
#   C1   0.60    0.0420         0.0294              0.0277
#   C2   0.50    0.0504         0.0302              0.0277
#   C3   0.40    0.0630         0.0315              0.0277
#   C4   0.30    0.0840         0.0336              0.0277
#   C5   0.70    0.0360         0.0288              0.0277
#   elasticity d ln P / d ln p_i = 1 for every layer

Both chains multiply to the share of people meeting all five tiers, 0.0596, but their factors differ completely: top-down, the conditional probability of C5C_5 is 0.897; bottom-up, C5C_5 comes first and is only 0.656. “Which tier has the lowest conditional probability” is therefore an order-dependent question and cannot by itself tell us which tier “matters most”. In the illustrative chain (0.6,0.5,0.4,0.3,0.7)(0.6, 0.5, 0.4, 0.3, 0.7), P(S)=0.0252P(S) = 0.0252; raising the weakest tier C4C_4 by 0.10 gives 0.0336, raising the strongest, C5C_5, by 0.10 only 0.0288; a 10 % relative increase gives 0.0277 whichever tier it is applied to. These are the two properties derived above.

Hypothesis · not established

The conditional excitation hypothesis (the author’s hypothesis, not established). For the career outcomes the author cares about, the five tiers combine more nearly multiplicatively than additively: without any one tier, the others lose much of their effect. Testable prediction: in a longitudinal sample that measures several tiers, with the log outcome as dependent variable, interactions between tiers (product terms on the raw scale) are clearly non-zero, and an additive model on the log scale fits better than an additive model on the raw scale in pre-registered cross-validation. If the raw-scale additive model fits as well, or the interactions are close to zero, the hypothesis is weakened. The difficulty is that none of the five tiers has an accepted measure, which is itself the first hurdle of any test.

Chapter summary

Multiplying five conditions gives three exact properties: any tier close to zero drags down the whole outcome; the model is additive in logarithms, so variance can be decomposed on the log scale; and a product of independent factors has a right-skewed, log-normal distribution. Shockley explained the skewed distribution of research output in 1957 by multiplying several hurdles, but his factors were an explanatory assumption, and the shape of the distribution of individual performance is itself disputed. The conditional-probability chain is an identity that holds in any order and implies no hierarchy. The next chapter turns to the weights the author assigns to the five tiers, and to why “how much does each tier explain” has no single answer.

References

Beck, James W., Adam S. Beatty, and Paul R. Sackett. 2014. “On the Distribution of Job Performance: The Role of Measurement Characteristics in Observed Departures from Normality.” Personnel Psychology 67 (3): 531–66. https://doi.org/10.1111/peps.12060.
Gibrat, Robert. 1931. Les inégalités économiques. Recueil Sirey.
Kremer, Michael. 1993. “The O-Ring Theory of Economic Development.” Quarterly Journal of Economics 108 (3): 551–75. https://doi.org/10.2307/2118400.
Limpert, Eckhard, Werner A. Stahel, and Markus Abbt. 2001. “Log-Normal Distributions Across the Sciences: Keys and Clues.” BioScience 51 (5): 341–52. https://doi.org/10.1641/0006-3568(2001)051[0341:LNDATS]2.0.CO;2.
O’Boyle, Jr., Ernest, and Herman Aguinis. 2012. “The Best and the Rest: Revisiting the Norm of Normality of Individual Performance.” Personnel Psychology 65 (1): 79–119. https://doi.org/10.1111/j.1744-6570.2011.01239.x.
Shockley, William. 1957. “On the Statistics of Individual Variations of Productivity in Research Laboratories.” Proceedings of the IRE 45 (3): 279–90. https://doi.org/10.1109/JRPROC.1957.278364.

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).