Methodology Changelog

Formula details

• Attendance weight: 25% — sourced from PRS India (MP_LOK) and TN Vidhan Sabha (MLA).
• Fund Utilization weight: 25% — sourced from MPLADS.gov.in portal.
• Legislative Activity weight: 20% — questions asked, debates participated, bills co-sponsored, from PRS India.
• Integrity weight: 30% — ECI affidavit criminal case count and category severity scoring.
• Missing data default: 50 (neutral) for any unavailable component.
• Final Reality Score normalised 0–100.

Reason for version

First public release. Weights chosen based on publicly available Tamil Nadu accountability research and precedents from ADR India scoring models.

Formula details

• Sentiment: positive/negative ratio from Tamil news media over 30 days. Recency weighting: last 7 days count 2× vs earlier days.
• Volume: article count vs Tamil Nadu 30-day average (z-score normalised).
• Coverage Quality: NLP classifier assigns higher scores to substantive policy/legislative coverage vs controversy, trivial or ceremony-only coverage.
• Sources monitored: Dinamalar, Vikatan, Dinakaran, The Hindu Tamil, Puthiya Thalaimurai digital editions.
• Final Perception Score normalised 0–100.

Reason for version

First public release. Sentiment model trained on Tamil public domain news corpus. Coverage quality classifier manually validated on 200 labelled articles per label class.

Formula details

• Gap Score = Perception Score − Reality Score.
• Positive gap (≥ +20): politician is media-overrated relative to verified performance.
• Negative gap (≤ −10): politician is media-underrated — works more than covered.
• Near-zero (−9 to +19): balanced media-to-performance ratio.
• No cap or floor — gap can theoretically range from −100 to +100.

Reason for version

Simple linear difference chosen over ratio to preserve directionality and make the score human-interpretable without transformation.