What does The Commuting Paradox and Relocation conclude, in numbers, about “The Case Against: Estimators, Sample Sizes and an Effect That Comes and Goes”?

Answer: Chapter 2 sets out the opposing evidence and shows that detecting an effect of 0.20 points per hour needs 13,440 person-waves of within variation.

Scope of the numbers: Its teams, sites, rents, pay and probabilities are illustrative; all calculations are generated by its programs; it is not individualised business, legal or financial advice. Marked boxes separate published evidence, runnable Python calculations (19 programs), classical sources and the author's hypotheses.

中文版

Data table

The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 13440

z(0.975) + z(0.80)2.8016
n SE/hour MDE/hour
2,0000.1851 0.5184
5,0000.1170 0.3279
20,0000.0585 0.1639
103,2700.0258 0.0721
detect -0.20/hour: n13,440 (s_x 8.7 min)
detect -0.10/hour: n53,757 (s_x 8.7 min)
if s_x were 4.0 min: n254,304 for -0.10/hour
a null in a small panel is weak evidence of no effect
Show the full program
# Chapter 2. p22: Can a panel see the commuting effect at all? In a
# fixed-effects regression only within-person changes in the commute
# identify the slope. With within-person SD of commuting s_x (8.7 min in
# Stutzer and Frey 2008), residual within-person SD of satisfaction
# s_e (illustrative 1.2 points) and n person-waves of within variation,
# SE(beta per hour) = s_e / (s_x / 60 * sqrt(n)). We print the minimum
# detectable effect (two-sided 5 %, 80 % power) and the person-waves
# needed to detect -0.20 and -0.10 per hour. 103,270 is the size of
# Stutzer and Frey's main sample; the smaller sizes are illustrative.
# (103,270 person-waves from 25,171 people in their table 2; Clark et
# al. 2020 follow over 26,000 workers in England for six waves.)
import math
from statistics import NormalDist

SX, SE_RES = 8.7, 1.2
Z = NormalDist().inv_cdf(0.975) + NormalDist().inv_cdf(0.80)


def se_beta(n, sx=SX):
    return SE_RES / (sx / 60.0 * math.sqrt(n))


def n_needed(beta, sx=SX):
    return math.ceil((Z * SE_RES / (beta * sx / 60.0)) ** 2)


print("z(0.975) + z(0.80) = {:.4f}".format(Z))
print("{:>9} {:>10} {:>12}".format("n", "SE/hour", "MDE/hour"))
for n in (2000, 5000, 20000, 103270):
    s = se_beta(n)
    print("{:>9,} {:>10.4f} {:>12.4f}".format(n, s, Z * s))
for beta in (0.20, 0.10):
    print("detect -{:.2f}/hour: n = {:,} (s_x 8.7 min)".format(
        beta, n_needed(beta)))
print("if s_x were 4.0 min: n = {:,} for -0.10/hour".format(
    n_needed(0.10, 4.0)))
print("a null in a small panel is weak evidence of no effect")

Single source and citation

Fan, Y. (2026). The Commuting Paradox, Compensating-Income Equations, Remote-Work Location Choice and the Foolish Old Man's Relocation Decision (Version 1.0.0), Chapter 2: The Case Against: Estimators, Sample Sizes and an Effect That Comes and Goes. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-204 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026204,
  author    = {Fan, Yafu},
  title     = {The Commuting Paradox, Compensating-Income Equations, Remote-Work Location Choice and the Foolish Old Man's Relocation Decision},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book