Model Methodology
This document describes the analytical framework collaboratively developed for the Tamil Nadu 2026 Legislative Assembly Election projection model. The model combines actual ECI vote share data from 2016 and 2021 with a four-force swing model, a 13-point constituency deep analysis, and a constituency-type-aware turnout decomposition. It is designed to be transparent, falsifiable, and honest about uncertainty.
1. Data Foundation
The model's baseline is built on actual Election Commission of India (ECI) Form 20 results for all 233 Tamil Nadu constituencies across the 2016 and 2021 assembly elections. Vote shares are mapped to the 2026 alliance structure: the SPA (DMK + INC + VCK + MDMK) and NDA (AIADMK + PMK + BJP) alliances are constructed by aggregating the constituent parties' 2021 vote shares into a single alliance total per constituency. This is the most important correction from naive models that compare DMK 2021 vs AIADMK 2021 directly — the 2026 contest is an alliance contest, not a party contest.
Data Sources
ECI 2021 Results
opencity.in Form 20 dataset — 4,233 candidate rows across 233 constituencies
ECI 2016 Results
opencity.in Form 20 dataset — 2016 assembly election constituency-wise results
SIR 2026 Enrollment
Special Intensive Revision 2026 — 5.6M new voters added; constituency-level growth estimated from district SIR data
Poll Aggregator
IANS-Matrize (Mar 2026), News18 VoteVibe (Feb 2026), Lokpoll (Mar 2026) — credibility-weighted average
Caste Demographics
District-level caste census estimates from 2011 SECC + academic literature on TN caste geography
Anti-Incumbency Signals
Daily AI agent: RSS news scraping + LLM sentiment scoring + constituency tagging from Tamil media
2. Poll Credibility Weighting
Not all polls are created equal. The model weights each poll agency by its verified historical accuracy across past Indian state elections. The Agni Research poll is excluded entirely — it has no verifiable track record and its TN 2026 numbers are outliers relative to all other agencies.
| Agency | Credibility Score | Weight | Rationale |
|---|---|---|---|
| IANS-Matrize | 5.5/10 | 46% | Best track record in South India; correctly called 2021 TN within 8 seats |
| News18 VoteVibe | 4.0/10 | 33% | Consistent methodology; tends to over-estimate ruling coalition by 5-8 seats |
| Lokpoll | 2.5/10 | 21% | Two consecutive major misses (2019 LS, 2021 TN); included but down-weighted |
| Agni Research | Unverified | Excluded | No verifiable historical track record; outlier numbers; shown as reference only |
3. The Four-Force Swing Model
The core projection model applies four independent forces to each constituency's 2021 baseline vote share. These forces were identified through collaborative analysis of the structural shifts between 2021 and 2026. Each force has a direction (who loses, who gains) and a magnitude (percentage of the vote pool that moves).
DMK Anti-Incumbency (SPA → TVK + NDA)
Five years of DMK governance have generated measurable grievances: TASMAC liquor policy, sand mining corruption, price rise, infrastructure delays, and candidate non-accessibility. The anti-incumbency pool is estimated at 5-9% of SPA's 2021 vote share depending on the scenario. Critically, this pool flows primarily to TVK (55-68%) because Vijay's primary political message is anti-DMK, not anti-NDA. AIADMK absorbs 15-25% of the anti-inc pool (traditional opposition voters returning), and the remainder stays home or goes to NOTA.
AIADMK/EPS Cadre Recovery (NDA structural gain)
In 2021, AIADMK lost despite winning 33.5% of the vote — their cadre was demoralised after Edappadi Palaniswami's defeat and the OPS-EPS split. By 2026, EPS has consolidated the party, completed a statewide yatra, and formalised the PMK+BJP alliance. The PMK brings the Vanniyar bloc (est. 8-10% of TN voters, concentrated in North TN). BJP adds urban Hindu consolidation. The model adds 3-5% to NDA's 2021 base in AIADMK-strong constituencies, and 1-2% in mixed seats. This is the single largest structural shift from 2021.
NDA Hindutva Consolidation (BJP/temple belt premium)
BJP's presence in the NDA alliance adds a Hindutva premium in temple-belt constituencies (Madurai, Tirunelveli, Kancheepuram, Vellore districts) and urban Hindu-majority seats. This is modelled as a 0.3-0.5% additional NDA gain in temple-belt seats, and 0.5% in PMK-strong Vanniyar constituencies. The Hindutva premium is deliberately kept small because TN has historically resisted Hindutva politics — it is a marginal but real factor in specific geographies.
TVK Youth/Minority/Anti-DMK Surge
TVK's vote comes from three sources: (a) Anti-DMK protest voters who would previously have stayed home or voted NOTA — Vijay gives them a credible vehicle; (b) New SIR voters (18-25 age group) who lean 60% TVK due to Vijay's star power and youth-focused messaging; (c) Minority youth (Muslim + Christian + Dalit) who are disillusioned with DMK's governance but unwilling to vote AIADMK. TVK's primary draw is from SPA, not NDA — this is the most important calibration insight. Every vote TVK gains is predominantly a vote SPA loses.
4. Turnout Decomposition Model
Higher voter turnout in Tamil Nadu 2026 is not a neutral expansion of the electorate — it is a directional signal. The model decomposes turnout uplift into three constituency-type-specific drivers, each benefiting a different party.
Key Insight: The Marginal Voter is Anti-DMK
Every 1% of turnout above the 2021 baseline of 76.6% represents voters who did not vote in 2021 but will in 2026. These are not a random sample of the population — they are self-selected motivated voters. In TN 2026, the motivation is either Vijay excitement (TVK) or AIADMK cadre anger (NDA). DMK's booth machinery ensures their base votes regardless of turnout level. Therefore, higher turnout is net negative for SPA.
| Constituency Type | Turnout Driver | SPA / 1% uplift | NDA / 1% uplift | TVK / 1% uplift |
|---|---|---|---|---|
| Urban (Chennai/Coimbatore/Trichy) | Vijay excitement | -0.8% | +0.20% | +0.50% |
| Semi-urban / Mixed | Mixed | -0.8% | +0.35% | +0.35% |
| Rural / AIADMK-strong | EPS yatra + PMK anger | -0.8% | +0.55% | +0.15% |
5. Candidate Individual Strength Score (CISS)
The CISS is a 0-10 score measuring a candidate's personal electoral strength independent of their party. A candidate with a high CISS can outperform their party's base vote; a low-CISS candidate underperforms. The CISS feeds into the seat outcome model as a ±2% swing on the party's projected vote share.
CISS Formula (5 Sub-Components, 2 points each)
Social Media Reach (0–2)
Facebook page likes + Instagram followers + Twitter/X followers, normalised to TN candidate average. A candidate with 500K+ combined reach scores 2.0; below 10K scores 0.5.
News Coverage Frequency (0–2)
Local Tamil media (Dinamalar, Dinakaran, Vikatan, Sun TV) mentions in the 6 months prior to election. Proxy for name recognition and public visibility.
Years in Party / Public Life (0–2)
Seniority in the party + grassroots depth. A 20+ year party worker scores 2.0; a parachute candidate scores 0.5. Captures organisational loyalty and cadre support.
Past Electoral Performance (0–2)
Personal vote above or below party average in 2021. A candidate who won by 8% when their party average was 5% scores 2.0. A candidate who underperformed by 5% scores 0.5.
Local Governance Record (0–2)
For incumbents: MLA fund utilisation rate, constituency work visibility, accessibility score. For challengers: community standing, local organisation, welfare work.
Vote swing impact: CISS of 8+ = +1.5% to party base vote. CISS of 6-7 = +0.5%. CISS of 4-5 = neutral. CISS of 2-3 = -0.5%. CISS below 2 = -1.5%.
6. Scenario Ladder
The model produces six scenarios based on turnout level, using a systematic anti-incumbency activation ladder (base 0.50 + 0.035 per 1pp above 70%, capped at 0.81 for surge). The toss-up band (seats within 2% margin) is shown separately — these seats could go either way on election day.
| Scenario | Turnout | Anti-Inc Activation | SPA | NDA | TVK | Verdict |
|---|---|---|---|---|---|---|
| Conservative | 70% | 0.500 | 143 | 88 | 3 | SPA Majority |
| Base | 74.3% | 0.650 | 125 | 102 | 7 | SPA Majority |
| Moderate Low | 76% | 0.710 | 114 | 111 | 9 | Hung Assembly |
| Moderate ★ | 78% | 0.780 | 106 | 116 | 12 | Hung — NDA leads |
| Moderate High | 80% | 0.800 | 102 | 117 | 15 | Hung — NDA leads |
| Surge | 82% | 0.810 | 83 | 131 | 20 | NDA Clear Majority |
Note: Toss-up seats (30–38 depending on scenario) are seats where the projected margin is within 2%. These seats could flip with a 1pp shift in turnout or TVK vote share. The dashboard shows both the projected winner and the toss-up band for transparency.
7. Limitations and Honest Caveats
⚠ Candidate data is estimated
Official ECI candidate lists will only be published after nomination filing closes. All candidate names, caste, and CISS scores are estimated from historical patterns and will be updated once official lists are available.
⚠ Caste demographics are approximate
Tamil Nadu has not conducted a caste census since 1931. All caste percentages are estimates from the 2011 SECC, academic literature, and district-level surveys. They carry ±5-10% uncertainty.
⚠ TVK seat wins are highly uncertain
TVK is contesting its first election. There is no historical baseline to calibrate their performance. The 0-12 seat range reflects genuine uncertainty, not model precision.
⚠ The model does not predict last-mile factors
Booth management quality, last-minute defections, EVM logistics, weather on polling day, and viral social media events in the final 48 hours are not modelled. These can shift 5-10 seats.
⚠ Hung assembly is the most likely outcome at 76–80% turnout
Three of six scenarios (Moderate Low, Moderate, Moderate High) produce a hung assembly. At base turnout (74.3%), SPA holds a majority. At surge (82%), NDA achieves a clear majority. The model's honest conclusion is that TN 2026 is genuinely uncertain, and any single-number projection claiming certainty should be treated with scepticism.
This analytical framework was developed collaboratively through iterative refinement of model assumptions, data calibration, and political reasoning. The model is designed to be updated as new information becomes available — poll results, candidate announcements, and ground-level reports will all be incorporated before election day on April 23, 2026.
All projections are probabilistic estimates, not predictions. The goal is to provide a rigorous, transparent, and honest analytical framework for understanding the Tamil Nadu 2026 election — not to call a winner.