What does The Fewest Questions conclude, in numbers, about “Fisher Information and the Cramér-Rao Bound: How Precise Can a Questionnaire Be?”?
Answer: Fisher information is a ruler for how much a question tells you; the information of independent questions adds up, and the Cramér-Rao bound says the variance of an unbiased estimator cannot fall below its reciprocal. Program p11 shows that going below the bound is possible only through bias and prior, and p12 shows that a yes/no question concentrates its information near its threshold, so a five-step ladder is precise only in the middle. The next chapter removes the premise of independence and shows how correlated questions drive information into a ceiling.
Scope of the numbers: Marked boxes separate published evidence, runnable Python calculations (19 programs), classical sources and the author's hypotheses.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python only
| n | bound 1/nI | var(mean) | var(shrunk) | MSE(shrunk) | MSE if th=2 |
| 1 | 1.0000 | 0.9985 | 0.2496 | 0.3113 | 1.2500 |
| 3 | 0.3333 | 0.3300 | 0.1856 | 0.2019 | 0.4375 |
| 5 | 0.2000 | 0.1997 | 0.1386 | 0.1454 | 0.2500 |
| 10 | 0.1000 | 0.0994 | 0.0821 | 0.0841 | 0.1157 |
| replications per row: 40,000; true theta = 0.5 |
Show the full program
# Chapter 1. p11: the Cramer-Rao bound is a floor for unbiased estimators
# only. Fixed-seed Monte Carlo, all parameters illustrative.
# 模型:n 个独立回答 Y_i = theta + e_i,e_i ~ N(0, 1)。
# 单题费希尔信息 I = 1,n 题合计 n,无偏估计方差下界 1/n。
# 样本均值无偏且恰好达到下界;收缩估计 c * 均值(c = n/(n+1))
# 方差低于 1/n,但它有偏,均方误差 MSE 取决于真值 theta。
import math
import random
THETA, REPS = 0.5, 40000
print("n bound 1/nI var(mean) var(shrunk) MSE(shrunk) MSE if th=2")
for n in (1, 3, 5, 10):
rng = random.Random(20261113 + n)
means = []
for _ in range(REPS):
ys = [THETA + rng.gauss(0.0, 1.0) for _ in range(n)]
means.append(math.fsum(ys) / n)
c = n / (n + 1.0)
mu = math.fsum(means) / REPS
var_m = math.fsum((m - mu) ** 2 for m in means) / (REPS - 1)
var_s = c * c * var_m
mse_s = math.fsum((c * m - THETA) ** 2 for m in means) / REPS
mse_far = c * c / n + (1.0 - c) ** 2 * 4.0
print("%-3d %.4f %.4f %.4f %.4f %.4f"
% (n, 1.0 / n, var_m, var_s, mse_s, mse_far))
print("replications per row: {:,}; true theta = {}".format(REPS, THETA))Single source and citation
Fan, Y. (2026). Fisher Information Matrices and the Cramér-Rao Bound: the Fewest Questions Needed to Estimate a Budget or a Birth Time (Version 1.0.0), Chapter 1: Fisher Information and the Cramér-Rao Bound: How Precise Can a Questionnaire Be?. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-235 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026235,
author = {Fan, Yafu},
title = {Fisher Information Matrices and the Cramér-Rao Bound: the Fewest Questions Needed to Estimate a Budget or a Birth Time},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 2 · The Information Ceiling of Correlated Questions: Why the Sixth Question Is Nearly Wasted
- Chapter 3 · An Item Response Model for Budget Questions: How Tightly Do Five Ladder Answers Pin Down a Budget?
- Chapter 4 · Adaptive Selection: the Maximum-Information Rule and an Information-Theoretic Floor on the Number of Questions
- Chapter 5 · The Fisher Information Matrix: Yes-Saying, Reverse-Keyed Questions and Optimal Design
- Chapter 6 · Estimation Theory for a Birth Time: Rounding Error, Sequential Tests and "It Matched Afterwards"
- Chapter 7 · Classical Sources East and West: "Knocking at Both Ends" and Fisher's Amount of Information
- Chapter 8 · Closing the Loop: a Five-Question Pre-Sale Budget Questionnaire, a Read-Only Question Order and a Pre-Registered Test