Indicator Definition: Childcare Discussion Density

Childcare Discussion Density measures the share of council speeches classified under the “Childcare & Education” category in each municipality. This page discloses the full methodology: definition, municipality selection criteria, classification model specification, known limitations, and data access.

Updated: 2026.05.16·spec_version: v2-tier1-500threshold-coverage30pct-truedensity·License: CC BY 4.0 (commercial use permitted)

Image for the Childcare Discussion Density methodology — PHOTO Illustration: Nursery school playground (AI-generated / Childcare Discussion Density methodology)

§1. Indicator Definition

Childcare Discussion Densityis defined as the percentage (%) of council speeches classified under the “Childcare & Education” category out of all speeches recorded for a given municipality.

The formula is as follows.

Childcare Discussion Density (%) = (speeches classified as “Childcare & Education” ÷ total speeches) × 100

1.1 12-Category Automatic Classification Model

KOUSOU 0 uses a machine-learning model that assigns each council speech to exactly one of the following 12 categories. Classification is single-label — a speech cannot belong to more than one category simultaneously.

Childcare & Education
Elderly Welfare
Infrastructure & Disaster Prevention
Industry & Employment
Healthcare & Public Health
Fiscal Affairs & Budget
Council Operations
Administrative Organization
Community & Neighborhoods
Tourism & Culture
Environment & Energy
Public Safety & Crime Prevention

1.2 Denominator Design

Using total speeches as the denominator eliminates the influence of council size — number of seats, session frequency — making small and large councils directly comparable on a single scale.

§2. Municipality Selection

This indicator covers 238 municipalities out of 1,788 nationwide. spec_version: v2-tier1-500threshold-coverage30pct-truedensity.

2.1 Selection Criteria (all three conditions must be met)

Condition 1: ≥ 500 total speeches indexed (since 2015)
Condition 2: Auto-classification coverage ≥ 30% (share of indexed speeches that received a category label)
Condition 3: Municipality attributes registered (population, aging rate, Fiscal Power Index, female council member ratio)

2.2 Municipalities Not Yet Covered

The remaining 1,335 municipalities are not yet included. Coverage will expand as classification progresses. The addition schedule is tracked on the Coverage page.

§3. Classification Model Specification

Classification is performed by a supervised machine-learning model. In the interest of methodological transparency, training data construction and inference procedure are disclosed below. Source code is not published, but the description is sufficient to allow independent reproduction.

3.1 Training Data

A sample of council speeches was manually annotated across the 12 categories and used as training data.

3.2 Inference

Each speech is assigned to exactly one category (single-label, not multi-label).

3.3 Current Accuracy

National average classification coverage: 24.8% (as of May 2026)
To ensure reliability, density is computed only for municipalities with coverage ≥ 30% (§2, Condition 2)

§4. Attribute Data Sources

All attribute values reflect fiscal year 2023 figures.

Population & aging rate: Ministry of Internal Affairs — “Population, Population Dynamics and Households Based on the Basic Resident Register” (as of January 1, 2023) + National Census.
Fiscal Power Index (FPI): Ministry of Internal Affairs — “Municipal Settlement Status Survey” (FY 2023). FPI measures fiscal self-sufficiency; a value of 1.0 means a municipality covers all standard expenditures from own-source revenue; values below 1.0 indicate dependence on central government equalization grants.
Female council member ratio: Council member rosters as collected by KOUSOU 0.

§5. Limitations

Density values are observed characteristics of councils and municipalities; they do not indicate quality or merit of councils.

Please read the following limitations before citing or reproducing this indicator.

Classification accuracy is not 100%. Misclassification will introduce noise into reported density values.
Municipalities that do not publish minutes are excluded.
Speeches substituted by written proceedings or paper-only sessions may not be captured.
Formulaic utterances — chair reports, roll-call votes, and similar procedural statements — are excluded.
Only speeches of 50 characters or more are included, to filter out short, pro-forma statements.
Outlier values (e.g., Monbetsu City at 0.31%) remain under ongoing data-validity review.