From Field Notes to Interaction: Structure Emerging from Complexity
KJ法は「結論」ではなく「視界」をつくる
本章では、川喜田二郎のKJ法と、統計的交互作用の概念を通じて、複雑な現象の捉え方を整理します。
KJ法の有用性は、答えを出すことにあるのではありません。
•情報を断片のまま放置しない
•無理に数値化できない観察を整理する
•仮説の種を浮かび上がらせる
そこにあります。
製造現場では、最初から数式で解ける問題ばかりではありません。
違和感、ばらつき、例外。
それらを構造として見える形にするのがKJ法の役割です。
桜子がKJ法を用いたのも、結論を急ぐためではなく、
問いを正確に立て直すためでした。
その問いが、やがて交互作用という視点へとつながります。
Jiro Kawakita (1920–2009): The Scientist of the Field
Jiro Kawakita was a Japanese geographer and anthropologist whose work was grounded in intensive field research, particularly in the Himalayan region during the 1950s.
1. Fieldwork and the Birth of the KJ Method
In the Himalayas, Kawakita encountered an overwhelming abundance of fragmented information: meteorological patterns, geological formations, plant distributions, and the cultural practices of local communities. These were not neatly tabulated datasets but living, interconnected observations.
He faced a fundamental question:
How can meaning emerge from scattered fragments?
To address this, Kawakita began writing individual observations on cards and grouping them according to natural affinities. Rather than forcing categories in advance, he allowed patterns to emerge through iterative clustering.
This process became known as the KJ Method.
It was not a statistical test.
It was a way of revealing structure within complexity.
⸻
2. “Wild Science” and the Philosophy of Going to the Site
Kawakita advocated what he later called “Wild Science” (Yasei no Kagaku): the idea that understanding begins not in abstraction, but in direct engagement with reality.
He encouraged students to gather information firsthand and integrate it through disciplined reflection. His “Mobile University” programs brought learners into rural communities to apply this integrative method to real-world issues.
Over time, elements of the KJ Method were incorporated into Japanese quality control practices and became part of what are often called the “Seven New QC Tools,” particularly for organizing qualitative information in the early stages of problem-solving.
⸻
3. Connection to the Story
Sakurako uses the KJ Method not to reach a conclusion, but to clarify uncertainty.
• Kawakita sought structural coherence in fragmented Himalayan data.
• Sakurako seeks structural coherence in fragmented factory observations.
In both cases, the method does not produce an answer.
It produces a sharper question.
⸻
Interaction: When Causes Do Not Act Alone
The concept of interaction in statistics reflects a fundamental reality: causes often do not operate independently.
A key architect of the mathematical framework for analyzing such complexity was Sir Ronald A. Fisher, whose work in experimental design and analysis of variance profoundly shaped modern statistical practice.
⸻
1. What Is an Interaction Effect?
An interaction occurs when the effect of one factor depends on the level of another factor.
For example:
• Factor A (High humidity) may slightly reduce quality.
• Factor B (Older material) may also slightly reduce quality.
If these effects were independent, their combined impact would simply add.
However, in some systems, the presence of both factors may produce a much larger—or qualitatively different—effect than either alone.
This is interaction.
It is not “1 + 1 = 2.”
It is conditional influence.
⸻
2. Fisher and the Design of Experiments
In the early 20th century, a common approach in experimentation was OFAT (One Factor At a Time): changing a single variable while holding others constant.
While useful, this approach cannot reveal interactions.
At the Rothamsted Experimental Station in England, Fisher confronted agricultural data where crop yield depended not only on fertilizer type, but also on soil conditions and rainfall.
He recognized that analyzing factors in isolation was insufficient.
Through the development of Analysis of Variance (ANOVA) and systematic experimental design, Fisher provided tools to:
•Estimate main effects
•Detect interaction effects
•Quantify uncertainty through probability models
This framework allowed researchers to evaluate whether observed differences were likely due to structured influences rather than random variation.
It did not “prove” truth.
It evaluated evidence.
⸻
3. Connection to the Story: A Shift in Perspective
Sakurako’s initial t-test examined a single main effect: humidity.
The result suggested that humidity alone did not sufficiently explain the defect surge.
However, when she considered the possibility that humidity and delivery timing might interact, the structure changed.
The interaction plot revealed that:
•Most delivery periods were stable across humidity levels.
•One specific period behaved differently under high humidity.
The system was not uniformly fragile.
It was conditionally fragile.
This shift—from asking “Is humidity the cause?” to asking “Under what conditions does humidity matter?”—reflects a Fisherian perspective.
The “thorn” in the interaction plot was not dramatic proof.
It was structural asymmetry.
And structural asymmetry demands further investigation.
統計は遠い理論ではない
交互作用や分散分析という言葉は、専門的に聞こえるかもしれません。
しかし統計は研究室の中だけの学問ではありません。
たとえば、コンビニエンスストアのPOS（Point of Sale）データ。
•曜日
•天候
•気温
•時間帯
•商品配置
これらの要因は単独で売上を決めているわけではありません。
「暑い日 × 週末 × 店頭陳列」
といった組み合わせが、特定商品の売上を急増させることがあります。
これはまさに交互作用です。
現代の小売業、物流、製造、広告配信に至るまで、
統計は条件の組み合わせを読み解くために使われています。
それは特別な研究者だけの道具ではありません。
構造を理解しようとする人のための言語です。
桜子が見つめているのは、犯人ではなく構造です。
そしてその姿勢は、ヒマラヤでも、農場でも、
コンビニのバックヤードでも変わりません。
