Poll of polls — seat range projections
All 11 agencies plotted against the 118-seat majority mark (234 total seats)
DMK+ Poll of Polls
128–142
comfortable majority
AIADMK+ Poll of Polls
72–85
strong opposition
TVK Poll of Polls
19–26
limited debut
Axis outlier (TVK)
98–120
vs consensus 19–26
The consensus story: 9 of 11 agencies agree DMK+ wins comfortably above 118. The sole structural outlier is Axis My India, which predicts a TVK landslide (98–120) and drops DMK+ below majority — a completely different political narrative, not a margin difference.
Vote share: Axis vs consensus
Both Axis and the consensus agree on DMK’s ~35% vote share — yet translate it into completely opposite seat outcomes
Consensus (Poll of Polls) Axis My India
The vote share gap is the key paradox. Axis gives TVK 35% — identical to DMK — yet converts it into 98–120 seats vs DMK’s 92–110. This implies Axis is modelling a dramatically different geographic concentration of TVK votes, with virtually no evidential basis. There is also a structural survey bias worth naming: TVK supporters, energised by a debut campaign and a celebrity leader, are far more likely to openly declare their vote than supporters of the established parties. DMK and AIADMK voters — particularly in rural areas where social pressure and patron-client relationships run deep — are more likely to give guarded or non-committal answers. An exit poll that does not correct for this differential candour will systematically over-represent TVK’s actual support.
Axis demographic audit — four critical errors
The Axis survey is not just numerically wrong — it misrepresents the very fabric of Tamil Nadu’s electorate
Categories listed by Axis
69%
of electorate classified
Unaccounted voters
31%
~1.8 crore missing
Brahmin: Axis vs actual
9% vs <3%
3× inflation
OBC/MBC: Axis vs actual
28% vs 51%
nearly halved
Axis survey Actual estimates Unaccounted (Axis only)
You cannot claim a party has 35% vote share when your survey ignores nearly one-third of the electorate. The figure is calculated over 69% of voters — not 100%. SC (20%) + ST (1%) + OBC/MBC (28%) + Muslim (6%) + Christian (5%) + Brahmin (9%) = 69%. The remaining 31% (~1.8 crore voters) simply vanishes from the survey universe.
Axis survey Actual estimates
The reservation test: Tamil Nadu’s 69% reservation policy is legally anchored in SC/ST (~20%) + OBC/MBC (~51%) = ~71% of population. Axis places these combined at only 49% (20+1+28). If Axis’s demographics were real, the State’s reservation framework would have no constitutional basis — the same one upheld by the Supreme Court under the Ninth Schedule. The fact that it does stand in court is proof that Axis’s sampling frame is fundamentally broken.
TVK swing claimed vs 2021
TVK (Tamilaga Vettri Kazhagam) was founded in February 2024 and did not contest the 2021 Tamil Nadu elections. The “swing” figures of +25 to +47 points are phantom baselines — there is no 2021 TVK result to swing from. Arithmetic change from a denominator of zero is statistically indefensible and arithmetically meaningless. These numbers should not have been published under the label “swing vs AE 2021.”
Sampling frame & selection protocol
No multi-stage stratified design described. No household or respondent selection rule. That 16 Innovas covered 16,447 km tells us about fuel consumption — it says nothing about who was sampled, how they were selected, or whether the sample represents the electorate. A travel log is not a methodology.
Weighting & post-stratification
No weighting matrix disclosed. Brahmin over-sample (9% vs <3%) and OBC under-sample (28% vs 51%) feed directly into final percentages with zero correction applied.
Margin of error & design effect
No confidence intervals published. Cluster sampling inflates MoE by 1.5–3× the theoretical ±0.5% for n=44,460. True MoE could reach ±1.5% — making DMK and TVK at 35% each statistically indistinguishable.
Non-response adjustment
No non-response rate or adjustment protocol given. Exit polls are structurally vulnerable — voters who decline to respond may differ systematically from those who agree.
Quality control mechanisms
96 surveyors across 234 constituencies = 0.41 surveyors per AC. No back-checking, validation, interviewer supervision, or inter-coder reliability testing is mentioned anywhere.
Data collection instrument
Electronic device use is a positive step — reduces interviewer transcription error. But “face-to-face interview + 44,460 sample” describes inputs only. How inputs became outputs is entirely missing.
Real methodology details a precise multi-stage stratified random sampling frame, randomisation at household and respondent level, interviewer training and inter-coder reliability, back-checking and data validation, a weighting matrix with non-response adjustment, and design effects with calculated margins of error. Science deserves better.
What the exit polls could be missing
Structural factors that even well-designed exit polls struggle to capture
The 85% turnout anomaly. Tamil Nadu recorded 84.69% voter turnout — 11% higher than 2021 and the highest ever in a state assembly election. High turnout at this scale has no historical baseline to model against. It could mean TVK enthusiasm drawing first-time youth voters, or DMK’s welfare machinery mobilising rural women (who turned out at 85.76%). Exit polls cannot reliably distinguish these sources.
Of 57.3 million registered voters: 83.57% male turnout · 85.76% female turnout · 60.49% third gender turnout — the female turnout surge is especially notable for welfare programme beneficiaries.
TVK as vote-splitter vs seat-winner. The structural trap for a debut party in first-past-the-post: TVK getting 20–25% of votes spread uniformly across 234 constituencies could produce near-zero seats. Geographic concentration matters far more than aggregate vote share, and exit polls model this poorly for parties with no electoral history.
AIADMK’s BJP baggage. When AIADMK joined the NDA, many predicted heavy losses — with minorities and Dalits consolidating toward DMK. Exit polls may be underestimating the cost to AIADMK in minority-heavy constituencies.
The TVK candour premium. TVK supporters are markedly more willing to openly declare their vote than supporters of either established party. DMK and AIADMK voters — particularly in rural constituencies where patron-client relationships and social pressure still govern daily life — frequently give guarded or evasive answers to surveyors. An exit poll without a differential candour correction will consistently over-report TVK’s share and under-report the combined Dravidian bloc. This is not a minor adjustment — it is a structural inflation that compounds across every caste sub-group Axis tabulates.
Welfare delivery and incumbency. DMK’s high-profile welfare schemes (Kalaignar Magalir Urimai, free bus travel for women) have strong last-mile delivery through party machinery. Exit polls structurally underweight the behavioural impact of welfare dependency on voting — beneficiaries frequently vote for continuity regardless of expressed sympathy for alternatives.
2021 exit poll accuracy — the baseline
How agencies performed last time, as a guide to credibility weighting for 2026
Exit poll midpoint Actual result
In 2021, even the most credible agencies significantly over-predicted DMK seats (projecting 175–195, actual: 159) and under-predicted AIADMK (projecting 38–54, actual: 75). The directional call was right, but the magnitude was wrong — a pattern that consistently recurs with Tamil Nadu polling.
Summary verdict
What the data says, what remains unknowable until May 4
Most credible DMK+ range
125–142
adjusted for 2021 over-prediction
Most credible AIADMK+ range
75–90
historically under-predicted
Most credible TVK range
10–30
debut party conversion discount
Axis TVK prediction credibility
Low
broken demographic base
The consensus of 9 out of 11 agencies pointing to a comfortable DMK+ win is the statistically safer position. The Axis My India prediction is not just an outlier — it is an outlier built on a broken sampling frame where OBC/MBC voters are cut in half, Brahmins are tripled, and 31% of the electorate does not exist.
A large sample built on a broken demographic universe does not improve with scale — it generates false precision. 44,460 interviews stratified on incorrect population weights produces confident answers to the wrong question.
The real wild card is geographic. If TVK’s vote is concentrated in 40–50 urban and semi-urban constituencies, they could disrupt both alliances meaningfully. If spread thin, even a 20% vote share produces close to zero seats under first-past-the-post. May 4 counting will be a definitive empirical test — and the clearest accountability moment for Indian exit polling in years.
