What does Reasoning Trajectories and Outcome Luck conclude, in numbers, about “Four Process-Tracing Tests Written as Likelihood Ratios”?

Answer: With ρ equal to zero, the true ratio for seeing both clues is 4.000, matching naive multiplication, and the posterior is 0.800. At ρ of 0.25 the true ratio falls to 3.024 and the posterior to 0.752; at 0.5 the values are 2.519 and 0.716; at 0.75 they are 2.209 and 0.688. At ρ equal to one the second clue merely echoes the first, the true ratio returns to the single-clue value of 2.000, and the posterior is 0.667. Naive multiplication reports 0.800 throughout.

Scope of the numbers: After-the-fact explanations are often folk theories: in a synthetic experiment, retrospective self-reports correlate only -0.013 with the weights used but 0.826 with an observer's guess; concurrent thinking aloud keeps order and leaps. Parameters are illustrative and all calculations come from the book's programs.

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: 0.25, 0.5, 0.667, 0.688, 0.716, 0.75

two clues, both seen; prior P(H)0.5
rho true LR naive LR P(H) true P(H) naive eff. clues
0.004.000 4.000 0.800 0.800 2.00
0.253.024 4.000 0.752 0.800 1.60
0.502.519 4.000 0.716 0.800 1.33
0.752.209 4.000 0.688 0.800 1.14
1.002.000 4.000 0.667 0.800 1.00
k clues all seen, rho0.5
k true LR naive LR eff. clues
12.000 2.0 1.00
22.519 4.0 1.33
43.994 16.0 2.00
810.042 256.0 3.33
Show the full program
# Chapter 2. p23: correlated clues are counted twice by naive Bayes.
# Two clues with the same margins a = 0.70 under H and b = 0.35 under R
# (single LR = 2). Dependence model: with prob rho the second clue just
# copies the first (same preparation drives both); otherwise it is an
# independent draw. Then P(both | H) = a (rho + (1 - rho) a), likewise
# under R. Naive multiplication uses a^2 / b^2 = 4 whatever rho is.
# Effective clue count = log(true joint LR) / log(single LR).
# A second block repeats the clue k times with copy prob rho = 0.5.
import math

A, B = 0.70, 0.35


def joint(p, rho, k):
    # prob that k chained clues all show, each copies the previous one
    # with prob rho, else draws fresh with margin p
    return p * (rho + (1 - rho) * p) ** (k - 1)


def post(lr):
    return lr / (1 + lr)


def main():
    single = A / B
    print("two clues, both seen; prior P(H) = 0.5")
    print("rho   true LR  naive LR  P(H) true  P(H) naive  eff. clues")
    for rho in (0.0, 0.25, 0.5, 0.75, 1.0):
        lr = joint(A, rho, 2) / joint(B, rho, 2)
        naive = single ** 2
        eff = math.log(lr) / math.log(single)
        print("{:<5.2f}{:>8.3f}{:>10.3f}{:>11.3f}{:>12.3f}{:>11.2f}".format(
            rho, lr, naive, post(lr), post(naive), eff))
    print("k clues all seen, rho = 0.5")
    print("k   true LR   naive LR  eff. clues")
    for k in (1, 2, 4, 8):
        lr = joint(A, 0.5, k) / joint(B, 0.5, k)
        print("{:<3d}{:>8.3f}{:>11.1f}{:>10.2f}".format(
            k, lr, single ** k, math.log(lr) / math.log(single)))


main()

Single source and citation

Fan, Y. (2026). Reasoning-Trajectory Profiling, Process-Tracing Games and Separating Cognitive Structure from Stated Intentions and Outcome Luck (Version 1.0.0), Chapter 2: Four Process-Tracing Tests Written as Likelihood Ratios. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-385 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026385,
  author    = {Fan, Yafu},
  title     = {Reasoning-Trajectory Profiling, Process-Tracing Games and Separating Cognitive Structure from Stated Intentions and Outcome Luck},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book