← Back to R-Word Index

Methodology

How the R-Word Index is built, scored, and published.

What the Index Measures

The R-Word Index is a normalized daily score from 0 to 100 representing the relative level of recession-related concern in everyday online conversation and search behavior. It answers one question: how much are real people talking and searching about recession right now, and how worried do they sound?
It is not a GDP forecast, not a market signal, and not an official economic indicator. It is a recession pulse index — a daily measure of how frequently and intensely recession appears in human-generated online activity.

Data Sources

The index is built from multiple sources of everyday human online activity — community discussions, personal accounts, and search behavior. The emphasis is on authentic human expression rather than official news outlets or institutional commentary.
Community Discussions
Posts and comments from online communities covering economics, jobs, personal finance, housing, investing, small business, and general life discussion. These capture how ordinary people describe their own experience with economic stress, job loss, rising prices, and uncertainty.
Search Trends
Aggregated search interest for recession-related terms. When more people search for "recession," "layoffs," or "unemployment," it reflects a rising awareness and concern in the broader population — even among those who don't participate in online discussions.
Tech Community
Posts from tech-focused communities covering startups, hiring, and industry sentiment. The tech sector is often an early indicator of broader economic shifts — layoff announcements, hiring freezes, and funding slowdowns tend to appear here before mainstream discussion.

Scoring Pipeline (Powered by AI)

Each piece of collected content is analyzed by an AI model that reads the title and text and evaluates it on two dimensions:
1. Recession Relevance
The AI determines whether the content is actually about economic recession or downturn — not just tangentially mentioning the economy. It filters out historical references ("the 2008 recession was bad"), metaphorical uses, and general economic commentary that isn't about recession concern. Only content that genuinely discusses current or anticipated recession passes through with a high relevance score.
2. Concern Intensity
For relevant content, the AI assesses how worried or alarmed the author sounds — from neutral, analytical tone to extreme anxiety or panic. It understands context, sarcasm, and nuance: a cautious mention of "economic headwinds" scores differently from "I'm terrified we're heading into a recession."
3. Document Score
doc_score = relevance × concern_intensity
Where:

Daily Aggregation

For each UTC day, the system:
  1. Collects all scored content from that day across all enabled sources
  2. Filters to documents with relevance above the minimum threshold
  3. Computes per-source scores using top-40% averaging — only the highest-scoring documents per source contribute, reducing noise from low-quality matches
  4. Combines source scores using configured weights — community discussions (40%), search trends (35%), and tech community (25%)
  5. Applies a volume signal — if today's document count is significantly above or below the 30-day average, the score is adjusted accordingly
  6. Blends with recent days via momentum carry-over so recession anxiety accumulates rather than resetting daily
  7. Normalizes to a 0–100 scale using percentile rank against a rolling 365-day baseline
  8. Applies dynamic smoothing to reduce day-to-day noise while staying responsive to sudden shifts

State Labels

The daily index maps to five human-readable states:
0–19: Quiet20–39: Low Concern40–59: Elevated60–79: High Concern80–100: Alarmed

Update Frequency

The public index is published once per day. Collection and scoring run in the background, but the public value updates daily for consistency and clarity.

Continuous Improvement

We regularly refine how the R-Word Index is scored — tuning AI prompts, adjusting keyword weights, upgrading models, and fixing bugs we discover in the scoring logic.
When we make a meaningful change, we re-score the entire historical record so the chart reflects a single, consistent methodology from end to end. This means a value you saw for a past date may change when you revisit it. We consider this more honest than freezing an outdated number — the past should be measured with the best tool we have, not the tool we had six months ago.
Every daily index row is stamped with the methodology version (format: v{N}-{YYYY-MM-DD}) that produced it. A public methodology changelog lists every re-score event with the date, scope, and a summary of how much historical values moved.
On the daily archive pages, if a date's value has been updated under a new methodology, we show both the original reading (as posted that day) and the current re-scored estimate side-by-side, so you can see exactly what changed and why.

Limitations

The R-Word Index is an experimental public signal. It is designed to be transparent, restrained, and honest about what it can and cannot measure.