Key numbers (corrected)

Every number here is from docs/insights/path_a_summary.json. This page replaced an earlier version that contained methodology errors — see we-got-it-wrong for details.

Seats won, 2026 (the ground truth)

Party	Seats won	% of 234
TVK	108	46.2%
DMK	59	25.2%
AIADMK	47	20.1%
INC	5	2.1%
PMK	4	1.7%
CPI(M)	4	1.7%
IUML	2	0.9%
VCK	2	0.9%
AMMK	1	0.4%
BJP	1	0.4%
DMDK	1	0.4%

Majority threshold is 118 (half of 234 + 1). TVK fell 10 seats short.

For each party, the median % of votes they won across the ACs they contested:

Party	Median 2026 %	Mean 2026 %	# ACs contested
TVK	33.73%	34.88%	233 (all but 1)
DMK	31.80%	32.49%	176
AIADMK	28.84%	28.50%	172
INC	28.19%	28.60%	28
PMK	28.16%	27.05%	18
BJP	21.98%	22.59%	33
NTK	3.49%	4.00%	234 (all)

Why TVK won 108 with only 33% median share

Tamil Nadu had 3-4 strong parties splitting the vote. TVK's 33% was enough to win pluralities in 108 ACs. DMK and AIADMK each ran near 29-32%; first-past-the-post + multi-party splits favoured the new entrant.

True swing 2021 → 2026 (only where party contested both years)

Party	Median true swing (pp)	# ACs both years
DMK	−15.0	163
AIADMK	−12.6	151
INC	−14.0	18
PMK	−10.2	10
BJP	−8.8	8 (small sample)
NTK	−2.4	234

Three things to notice:

DMK and AIADMK losses are similar magnitude — about a 5th of their previous vote share went elsewhere.
BJP actually lost ground (−8.8 pp) in the 8 seats where they contested in BOTH years. The earlier version of this site claimed BJP gained because it included BJP's brand-new NDA-allocated seats as "swings". Corrected here.
TVK and others not listed: TVK didn't contest in 2021 (0 ACs both years), so no true-swing number exists.

Religion vs party — Pearson r at AC level

Correlations between AC-level religion share (Census 2011) and 2026 vote share:

	DMK	AIADMK	TVK	BJP	INC	PMK	NTK
Hindu%	+0.03	+0.08	−0.06	+0.03	+0.16	−0.09	+0.05
Muslim%	+0.03	−0.00	+0.05	−0.13	+0.12	+0.06	−0.02
Christian%	−0.04	−0.09	+0.02	−0.00	−0.19	+0.09	−0.03

All values |r| < 0.2 — religion mix at AC level is not a strong predictor of any party's swing.

Caveat

We use district-level religion mix (Census 2011, 32 distinct values across 203 ACs ≈ 6 ACs per identical religion vector). Within-cluster variation is invisible — this biases correlations toward weak. Genuine sub-district religion data would tighten the test. See religion.md for the caveat in detail.

Seats won by reservation status

44 SC/ST reserved + 190 General = 234.

TVK won reserved AND general

TVK: 23 of 44 reserved (52%) + 85 of 190 general (45%). Slightly stronger in reserved. The actual Dalit-vote story.

Reserved seats (44)

Party	Seats won
TVK	23 (52%)
DMK	9
AIADMK	9
VCK	2
CPI(M)	1

General seats (190)

Party	Seats won
TVK	85 (45%)
DMK	50
AIADMK	38
INC	5
PMK	4
CPI(M)	3
IUML	2
DMDK · BJP · AMMK	1 each

Entry-type breakdown (so you can audit the "swing" numbers)

For each party: how many ACs they contested only in 2021, only in 2026, or both. This is the column that distinguishes a real swing from an alliance-reshuffle artefact.

Party	Both years	Only 2021 (withdrew)	Only 2026 (new entry)
DMK	163	25	13
AIADMK	151	40	21
TVK	0	0	233
BJP	8	12	25
INC	18	7	10
PMK	10	13	8
NTK	234	0	0

The TVK row says it all: there's no such thing as a "TVK swing" — every TVK number is a 2026 level.

Data coverage

Layer	Coverage	Gap
2026 candidate results	234 / 234 ACs	—
2021 candidate results	234 / 234 ACs	—
Religion mix attached	203 / 234 ACs	31 in post-2011 new districts
Caste granular (jati)	0 / 234 ACs	not in open data
Booth-level swing	0 / ~80,000 booths	image PDFs; needs OCR (Path B)

Key numbers (corrected) ​

Seats won, 2026 (the ground truth) ​

Median 2026 vote share (level — NOT a swing) ​

True swing 2021 → 2026 (only where party contested both years) ​

Religion vs party — Pearson r at AC level ​

Seats won by reservation status ​

Reserved seats (44) ​

General seats (190) ​

Entry-type breakdown (so you can audit the "swing" numbers) ​

Data coverage ​