5D Library › Power-Law Weighting and Bottom-Up Controllability in Multi-Tier Conditional Excitation Systems · 中文
Chapter 4 Environment, Reputation and Luck
Environment: where one grows up
The author calls the third tier “environmental topology”: the structure of the city, neighbourhood, industry and personal network one is embedded in. The mathematics of network position (weak ties, structural holes, complex contagion) was treated in Volume 25, Social Network Topology, Structural Holes and Complex Contagion; here we look only at the strongest kind of evidence on the environment tier, randomized and quasi-experimental moves.
Evidence
The Moving to Opportunity experiment. Between 1994 and 1998 the US Department of Housing and Urban Development randomly assigned 4,604 families living in high-poverty public housing in Baltimore, Boston, Chicago, Los Angeles and New York to three groups; the experimental group received housing vouchers usable only in low-poverty neighbourhoods. Following the children’s tax records into adulthood, Chetty, Hendren and Katz found that children who moved before age 13 had clearly higher earnings in their mid-twenties: the treatment-on-the-treated effect was $3,477 a year, 31 % above the control mean of $11,270, and they were more likely to attend college. Children who moved at 13–18 did not gain and may have been slightly harmed, perhaps by the disruption of moving (Chetty et al. 2016).
Exposure time. Using more than seven million families who moved between areas in the United States, Chetty and Hendren found that children’s adult outcomes converge towards those of children who grew up permanently in the destination at about 4 % per year of exposure (Chetty and Hendren 2018).
This is the closest the author’s “environment” tier comes to causal evidence: under random assignment, the environment changed outcomes. It also shows two limits. First, the effect depends on age: the same move helps young children and not adolescents. In the author’s language, the environment tier interacts with the timing tier rather than adding to it independently. Second, the effect is an average: a randomized experiment tells us what vouchers do for a group of children on average, not what a change of environment would do for a particular child.
Reputation: cumulative advantage
Evidence
The Matthew effect. In a 1968 article Merton described the “Matthew effect” in science: scientists who already have standing receive disproportionate recognition for joint work and simultaneous discoveries, and recognition itself becomes a cause of further resources and attention (Merton 1968).
An artificial music market. Salganik, Dodds and Watts had 14,341 participants listen to and download songs by 48 unknown bands on a website. Participants were randomly assigned to an “independent” world (no download counts shown) or to one of eight “social influence” worlds (download counts within their world shown). Inequality in downloads was clearly higher in the social-influence worlds, and which songs became hits was hard to predict across worlds; the best songs rarely did badly and the worst rarely did well, but almost any other outcome was possible (Salganik et al. 2006).
Randomly given early success. Van de Rijt and colleagues ran field experiments on four platforms, randomly giving early success: donations to some crowdfunding projects, awards to some Wikipedia editors, positive ratings to some product reviews, signatures to some petitions. Recipients went on to succeed at significantly higher rates, but larger initial successes did not bring larger subsequent ones (Rijt et al. 2014).
A funding threshold. Bol, de Vaan and van de Rijt compared applicants to a Dutch early-career grant scheme whose review scores fell just above and just below the funding line: their scores were almost identical, yet over the following eight years those just funded accumulated more than twice as much research funding and were more likely to become full professors (Bol et al. 2018).
The designs differ (historical observation, laboratory experiment, field experiment, threshold comparison), but the conclusions point the same way: early, partly random success gets amplified. Evolutionary game models of reputation and indirect reciprocity were treated in Volume 35, Evolutionary Game Theory, Direct and Indirect Reciprocity and the Emergence of Cooperation; here a simple program shows how cumulative advantage raises inequality and unpredictability at the same time.
Calculation
48 songs with quality spread evenly from 0.15 to 0.85 (illustrative); 8 independent worlds of 700 listeners, each listener sampling 3 songs. With social-influence strength , a song is picked for listening with probability and then downloaded with probability . is the independent condition. Measures: the Gini coefficient of downloads (inequality), the average difference in market share between worlds (unpredictability), and the rank correlation between quality and market share.
# Cumulative advantage in an artificial market (layer C4), a toy version of the design of
# Salganik, Dodds & Watts (2006): 48 songs, 8 independent worlds of 700 listeners, each
# sampling 3 songs. With social-influence strength s a song is picked with probability
# (1 - s) / 48 + s (downloads + 1) / (total + 48) and downloaded with probability q, its
# fixed "quality" (illustrative). s = 0 is the independent condition. Measures: Gini of
# downloads; unpredictability U = mean |share_a - share_b| over songs and pairs of
# worlds; Spearman rank correlation between quality and market share.
import random, itertools
from statistics import mean
M, WORLDS, LISTENERS, SAMPLES = 48, 8, 700, 3
def gini(xs):
xs = sorted(xs); n = len(xs)
return sum((2 * i - n + 1) * x for i, x in enumerate(xs)) / (n * sum(xs))
def spearman(a, b):
rk = lambda xs: [sorted(range(len(xs)), key=lambda i: xs[i]).index(i)
for i in range(len(xs))]
ra, rb = rk(a), rk(b); m = mean(ra)
num = sum((x - m) * (y - m) for x, y in zip(ra, rb))
return num / sum((x - m) ** 2 for x in ra)
def world(q, s, rng):
d = [0] * M
for _ in range(LISTENERS * SAMPLES):
tot = sum(d)
w = [(1 - s) / M + s * (x + 1) / (tot + M) for x in d]
i = rng.choices(range(M), weights=w)[0]
if rng.random() < q[i]: d[i] += 1
return d
def experiment(s, seed=41):
rng = random.Random(seed)
q = [0.15 + 0.7 * k / (M - 1) for k in range(M)]
worlds = [world(q, s, rng) for _ in range(WORLDS)]
sh = [[x / sum(d) for x in d] for d in worlds]
U = mean(abs(a[i] - b[i]) for a, b in itertools.combinations(sh, 2) for i in range(M))
G, rho = mean(gini(d) for d in worlds), mean(spearman(q, d) for d in worlds)
return q, worlds, G, U, rho
if __name__ == "__main__":
print("social influence s | Gini | unpredictability U | rank corr(quality, share)")
for s in (0.0, 0.5, 0.9):
q, worlds, g, U, rho = experiment(s)
print(f" {s:.1f} | {g:.2f} | {U:18.4f} | {rho:.2f}")
q, worlds, *_ = experiment(0.9)
rank = lambda d, i: 1 + sorted(d, reverse=True).index(d[i])
print("s = 0.9: rank of the best song in each world (1 = most downloads): "
+ " ".join(str(rank(d, M - 1)) for d in worlds))
print(" rank of the worst song in each world: "
+ " ".join(str(rank(d, 0)) for d in worlds))
# social influence s | Gini | unpredictability U | rank corr(quality, share)
# 0.0 | 0.26 | 0.0051 | 0.91
# 0.5 | 0.40 | 0.0088 | 0.89
# 0.9 | 0.60 | 0.0171 | 0.81
# s = 0.9: rank of the best song in each world (1 = most downloads): 7 8 16 6 5 1 3 1
# rank of the worst song in each world: 39 47 36 35 47 36 47 45As social influence rises from 0 to 0.9, the Gini coefficient rises from 0.26 to 0.60, unpredictability from 0.0051 to 0.0171, and the rank correlation between quality and share falls from 0.91 to 0.81. At the best song ranks between 1st and 16th in the eight worlds and the worst between 35th and 47th: as Salganik and colleagues observed, the best do not do too badly and the worst do not do too well, while everything in between depends heavily on early chance. The toy model only illustrates the mechanism; its parameters correspond to no real market.
Luck: a model
Evidence
The “talent versus luck” model. Pluchino, Biondo and Rapisarda built an agent-based model: 1,000 agents with talent normally distributed with mean 0.6 and standard deviation 0.1, all starting with capital 10; over 40 years (80 half-year steps) agents randomly meet lucky and unlucky events; at a lucky event capital doubles with probability equal to the agent’s talent, at an unlucky event it halves. The model produces highly unequal capital, and the most successful agents usually have only moderately above-average talent (Pluchino et al. 2018).
This is a model, not evidence. Its conclusions follow from assumed rules: talent acts only when good luck arrives, and bad luck strikes everyone alike. Frank discusses many examples of the role of luck in success and its implications for public policy, but that too is argument rather than measurement (Frank 2016).
Calculation
Below is a mean-field simplification of the model: instead of events moving on a grid, each agent meets a lucky event with probability 0.03 and an unlucky event with probability 0.03 at every step. The other rules are unchanged.
# A mean-field simplification of the "talent versus luck" MODEL of Pluchino, Biondo &
# Rapisarda (2018). 1,000 agents, talent T ~ Normal(0.6, 0.1) clipped to [0, 1], capital
# 10, 80 half-year steps. In each step an agent meets a lucky event with probability p and
# an unlucky event with probability p (the original moves events on a 201 x 201 grid; here
# encounters are simply random). Lucky event: capital doubles with probability T.
# Unlucky event: capital halves. A model, not evidence about real careers.
import math, random
from statistics import mean
def run(p=0.03, steps=80, n=1000, seed=41):
rng = random.Random(seed)
T = [min(1.0, max(0.0, rng.gauss(0.6, 0.1))) for _ in range(n)]
C = [10.0] * n
for _ in range(steps):
for i in range(n):
if rng.random() < p and rng.random() < T[i]: C[i] *= 2
if rng.random() < p: C[i] /= 2
return T, C
def corr(x, y):
mx, my = mean(x), mean(y)
sxy = sum((a - mx) * (b - my) for a, b in zip(x, y))
return sxy / math.sqrt(sum((a - mx) ** 2 for a in x) * sum((b - my) ** 2 for b in y))
if __name__ == "__main__":
T, C = run()
order = sorted(range(len(C)), key=lambda i: -C[i])
top20 = sum(C[i] for i in order[:200]) / sum(C)
best = order[0]; most_talented = max(range(len(T)), key=lambda i: T[i])
print(f"top 20 % of agents hold {top20:.0%} of the capital")
print(f"richest agent: talent {T[best]:.2f}, capital {C[best]:.0f}"
f" (mean talent {mean(T):.2f})")
print(f"most talented agent: talent {T[most_talented]:.2f},"
f" capital {C[most_talented]:.1f}")
r = corr(T, [math.log(c) for c in C])
print(f"correlation of talent with log capital: {r:.2f}")
print(f"mean talent of the 10 richest: {mean(T[i] for i in order[:10]):.2f}")
hi = [i for i in range(len(T)) if T[i] >= 0.75]
print(f"agents with talent >= 0.75: {len(hi)}; median capital "
f"{sorted(C[i] for i in hi)[len(hi) // 2]:.0f}; overall median "
f"{sorted(C)[len(C) // 2]:.0f}")
wins = 0
for seed in range(100):
T, C = run(seed=seed)
wins += T[max(range(len(C)), key=lambda i: C[i])] >= 0.75
print(f"100 runs: the richest agent has talent >= 0.75 in {wins} runs")
# top 20 % of agents hold 69% of the capital
# richest agent: talent 0.54, capital 640 (mean talent 0.60)
# most talented agent: talent 0.89, capital 2.5
# correlation of talent with log capital: 0.09
# mean talent of the 10 richest: 0.66
# agents with talent >= 0.75: 76; median capital 5; overall median 5
# 100 runs: the richest agent has talent >= 0.75 in 21 runsIn this run the richest 20 % of agents hold 69 % of the capital; the richest agent has talent of only 0.54, below the mean of 0.60; the most talented agent (0.89) is left with capital 2.5; and the correlation between talent and log capital is only 0.09. Over 100 runs, the richest agent has talent of 0.75 or more in only 21. The mechanism is clear: under multiplicative rules that make talent conditional on good luck, the variance of outcomes comes mainly from the number of events, not from talent.
But this shows a property of the rules, not of the world. If talent also raised the chance of meeting good luck, or softened the losses from bad luck (in the author’s model, precisely the role of knowledge investment and reputation ), the correlation between talent and outcome would rise. Using a model to prove that “luck matters more than talent” is, like using one to prove that “effort decides everything”, only a restatement of the assumptions.
Effort: the last tier
The author’s fifth tier, “knowledge investment”, is the only one entirely in one’s own hands. Deliberate practice was treated in Volume 23; here we give only one figure directly relevant to this book.
Evidence
How much practice explains. Ericsson, Krampe and Tesch-Römer proposed that expert performance comes mainly from long deliberate practice (Ericsson et al. 1993). In a meta-analysis of 88 studies, Macnamara, Hambrick and Oswald found that deliberate practice explained on average about 12 % of the differences in performance: about 26 % in games, 21 % in music, 18 % in sports, 4 % in education and less than 1 % in professions (Macnamara et al. 2014).
The result is often quoted as evidence that “effort does not matter”. That is a misreading. First, it measures the correlation between hours of practice and performance across people, not the causal effect of practice on any individual: a small share of the differences between people does not mean a small effect for a given person. Second, in professions “performance” is strongly shaped by environment, reputation and luck, and practice’s share is small partly because the other tiers vary so much. Third, as Chapter 2 showed, in a multiplicative model shares depend on the splitting rule and the scale.
Chapter summary
The Moving to Opportunity experiment shows that under random assignment the environment changed children’s adult earnings, but the effect depends on age at the move: environment and timing interact. Studies of the Matthew effect, the artificial music market, randomly given early success and funding thresholds together show that early, partly random success is amplified, raising inequality and lowering predictability. “Talent versus luck” is a model whose conclusions follow from its assumed rules. Deliberate practice explains between about 26 % and under 1 % of the differences in performance in different domains, but these are variance shares across people, not the effect of practice on an individual. The next chapter turns to the second direction of the author’s framework: how far one can push from the bottom.
References
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).