The Cold Verdict of “R”
仮説検証に用いた統計手法
本章では、物語本文とあわせて読解問題を掲載します。
作中で桜子と西野が用いたのは、推測統計の基本的手法です。
具体的には：
•群間比較（t検定）
•p値による有意性判断
•箱ひげ図による分布の可視化
です。
今回の検証は、「保管条件」というカテゴリ変数が、不良率や材料特性に統計的に有意な差をもたらしているかどうかを確認するものでした。
重要なのは、
•差があるように見えること
•それを主要因と断定できること
は同じではない、という点です。
本文を読みながら、
•仮説の立て方
•検定の意味
•有意でない結果の解釈
にも注意して問題に取り組んでください。
物語は感情で進みますが、
統計は境界を引きます。
Sakurako had placed her faith in the logic of engineering.
If improper storage was truly the root cause, then controlled comparison would expose it.
To test her hypothesis, she quietly enlisted Nishino from Maintenance. Though only in his third year, Nishino had specialized in statistics and handled data with disciplined precision.
“If we stratify properly,” he said, eyes steady, “and compare strictly controlled lots with the neglected ones, the difference should appear in the numbers.”
Under the cover of night, they conducted their verification.
•Group A (Strict Control): Lots stored in airtight containers inside a climate-controlled chamber.
•Group B (Neglected): Lots exposed to condensation and outside air near the loading dock.
They measured flowability and green density, compiling the results into a dataset.
Nishino opened R.
# Comparing flowability between groups
t.test(flowability ~ storage_group, data=test_data)
# Comparing density variation
t.test(density ~ storage_group, data=test_data)
# Visualizing distribution
boxplot(defect_rate ~ storage_group, data=test_data,
main="Defect Rate by Storage Condition",
xlab="Storage Group", ylab="Defect Rate")
Blue text scrolled across the console.
Nishino’s expression changed.
“…p-value is 0.28 for flowability.
Density difference—also not significant.
And defect rate… no statistically significant separation.”
Sakurako stared at the screen.
The boxplots showed overlap—substantial overlap.
Not identical.
But not decisive.
“The variance exists,” Nishino said quietly, “but it’s within statistical noise. Based on this sample size, storage condition does not explain the spike in defects.”
Silence settled in the small room.
“That means,” Sakurako said slowly, “even if we enforced humidity control perfectly…”
“…we might reduce variability slightly,” Nishino finished, “but it would not eliminate the current defect surge.”
Not the main factor.
Not the dominant driver.
The hypothesis—so elegant, so reasonable—had failed.
⸻
The Web Tightens
If it was not storage,
and not method compliance,
then what was generating this invisible instability?
Sakurako felt something shift inside her.
Had Yamashita known?
Or worse—
had he known that this would not be enough?
She remembered the way he had closed the distance in the QC office. Not intimidation. Not anger.
Certainty.
The stillness of someone who understood the structure better than anyone else in the room.
A spider does not chase.
It waits.
“…Sakurako-san?” Nishino’s voice sounded faint.
She realized her hands were trembling.
The logic had not betrayed her.
The data had.
And the web, whatever it was, was larger than storage conditions.
Reading Comprehension Questions
The Cold Verdict of “R”
⸻
Question 1
What was the primary purpose of Sakurako and Nishino’s experiment?
(A) To prove Yamashita had manipulated the data
(B) To determine whether storage conditions were the dominant cause of defects
(C) To demonstrate Nishino’s statistical expertise
(D) To compare humidity levels between warehouses
⸻
Question 2
Why did Nishino use a t-test instead of a correlation coefficient?
(A) Because both variables were continuous
(B) Because storage condition was treated as a categorical grouping variable
(C) Because correlation analysis is less accurate
(D) Because R software cannot calculate correlations
⸻
Question 3
What does the p-value of 0.28 most strongly imply in this context?
(A) The storage method is definitively unrelated to defects
(B) The sample size was too small to calculate variance
(C) The observed difference is not statistically significant under conventional thresholds
(D) The regression model was incorrectly specified
⸻
Question 4
What is suggested by the phrase, “Not identical. But not decisive.”?
(A) The data perfectly confirmed the hypothesis
(B) There was some variation, but it was insufficient to establish a dominant causal relationship
(C) The experiment failed due to equipment malfunction
(D) The results were fabricated
⸻
Question 5
What does the “spider at the center of a web” metaphor most likely represent?
(A) Yamashita’s fear of exposure
(B) A random coincidence in statistical analysis
(C) The possibility that Sakurako’s logical path was anticipated or constrained within a larger structure
(D) Nishino’s manipulation of the dataset
非有意という結論
本章では、物語と読解問題を通して、
一つの仮説が統計的に支持されなかった場面を扱いました。
t検定の結果は、有意差なし。
これは、
•「保管条件は絶対に無関係である」
という意味ではありません。
示されたのは、
•今回のデータでは、主要因と断定できない
という限定的な結論です。
統計は白黒を断言する道具ではありません。
仮説を絞り込むための境界を示す道具です。
物語の中で崩れたのは、
データそのものではなく、
「それが原因であるはずだ」という確信でした。
設問を通して、
•有意差の意味
•仮説棄却の解釈
•因果と証拠の違い
をどのように感じたでしょうか。
論理が否定されたとき、
感情はどこへ向かうのか。
その揺れも含めて、ぜひ感想を共有してください。
