What does The Shrinking Marshmallow Effect conclude, in numbers, about “Why the Coefficient Shrinks: Omitted Variables, Measurement Error and Bad Controls”?
Answer: Shrinkage after controls can come from three directions: confounding correctly removed, confounding left behind by imperfect measurement, and true effect wrongly deleted by a bad control. The omitted-variable formula shows that an unmeasured factor correlating 0.3 with waiting and moderately affecting achievement is enough to erase the 0.081; lowering the reliability of family resources from 1 to 0.6 lifts the controlled coefficient from near the truth to more than double it; and controlling for a concurrent variable that is itself an outcome cuts the total effect down to the direct effect. All three disputes take place within a range that is already small. The next chapter turns to measurement itself: what can a test given once, lasting seven minutes and involving one sweet, actually measure?
Scope of the numbers: Its parameters are illustrative; all calculations are generated by its programs; the book is not a psychological assessment or medical or parenting advice, and its suggestions for shops and communities are non-personalised business advice. Marked boxes separate published evidence, the author's hypotheses, classical sources and runnable Python calculations (18 programs).
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. printed values that match the answer: 0.081
| gap explained by early controls: 0.155 | ||
| rho | gamma for gap | gamma to erase 0.081 |
| 0.20 | 0.775 | 0.405 |
| 0.30 | 0.517 | 0.270 |
| 0.40 | 0.387 | 0.202 |
| 0.50 | 0.310 | 0.162 |
| 0.60 | 0.258 | 0.135 |
| bivariate R2 = 0.236^2 = 0.0557 | ||
| R_ctrl | Rmax | Oster beta* (delta = 1) |
| 0.30 | 0.39 | 0.024 |
| 0.40 | 0.52 | 0.027 |
| 0.50 | 0.65 | 0.029 |
Show the full program
# Chapter 4. p41: how strong must the hidden family factor be?
# 模型:标准化变量下的遗漏变量偏误公式 naive = beta + gamma*rho,其中
# rho 为等待与遗漏因素的相关,gamma 为遗漏因素在完整模型中对成就的系数。
# 缺口取 Watts 等 (2018) 0.236 - 0.081 = 0.155(出处),以及把 0.081
# 再压到 0 所需的 0.081。另按 Oster (2019) 的近似
# beta* = b_ctrl - delta*(b_raw - b_ctrl)*(Rmax - R_ctrl)/(R_ctrl - R_raw)
# 给出 delta = 1、Rmax = 1.3*R_ctrl 时的界;R_ctrl 未见原文报告,为示意。
B_RAW, B_EARLY = 0.236, 0.081
def main():
gap = B_RAW - B_EARLY
print("gap explained by early controls: {:.3f}".format(gap))
print(" rho gamma for gap gamma to erase 0.081")
for rho in (0.20, 0.30, 0.40, 0.50, 0.60):
print(" {:.2f} {:>12.3f} {:>18.3f}".format(
rho, gap / rho, B_EARLY / rho))
r_raw = B_RAW ** 2
print("bivariate R2 = 0.236^2 = {:.4f}".format(r_raw))
print(" R_ctrl Rmax Oster beta* (delta = 1)")
for r_ctrl in (0.30, 0.40, 0.50):
rmax = 1.3 * r_ctrl
bstar = B_EARLY - gap * (rmax - r_ctrl) / (r_ctrl - r_raw)
print(" {:.2f} {:.2f} {:>8.3f}".format(r_ctrl, rmax, bstar))
if __name__ == "__main__":
main()Single source and citation
Fan, Y. (2026). The Shrinking Marshmallow Effect, Family-Resource Confounding and the Limits of Child Self-Control Interventions (Version 1.0.0), Chapter 4: Why the Coefficient Shrinks: Omitted Variables, Measurement Error and Bad Controls. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-203 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026203,
author = {Fan, Yafu},
title = {The Shrinking Marshmallow Effect, Family-Resource Confounding and the Limits of Child Self-Control Interventions},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · Two Thirds Gone: the Marshmallow Coefficient Meets Family Background
- Chapter 2 · Rational Snacking: Not Waiting under a Prior on Environmental Reliability
- Chapter 3 · How Long Is It Rational to Persist? Waiting-Time Distributions and the Optimal Give-Up Point
- Chapter 5 · The Twenty-Second Threshold: Category Coefficients, Single-Trial Reliability and the Seven-Minute Ceiling
- Chapter 6 · The Ceiling on Intervention: Fadeout, an Optimistic Upper Bound and Two Levers
- Chapter 7 · Classics East and West: the Father's Fault in the Three Character Classic and Locke on Self-Denial
- Chapter 8 · Closing the Loop: the Toll-Booth Scenario Capsule, the Family Promise Ledger and Pre-Registration