Finding 2 — Religion didn't predict the swing

Pearson correlations between AC-level religion share and party swing are all |r| < 0.2. The TVK wave was not a communal realignment.

The headline test

For each of the 203 ACs where we could attach a 2011 Census religion mix (district level), we computed the Pearson correlation between religion% and the swing-from-2021 for each major party. The table below is the answer.

	DMK swing	AIADMK swing	TVK swing	BJP swing	INC swing	PMK swing	NTK swing
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

Reading this table

Correlations of ±0.1 are weak; ±0.2 is borderline; ±0.3+ would be meaningfully predictive. Every cell here is below the "meaningfully predictive" threshold.

What the cells say

• BJP × Muslim% is the largest negative (r = −0.13). Where Muslim share is higher, BJP gained less. Expected direction, modest magnitude.
• INC × Christian% is the largest negative (r = −0.19). Where Christian share is higher, Congress lost more. Surprising sign — Congress historically draws from minority blocks; possibly an alliance-arithmetic artefact in 2026.
• INC × Hindu% = +0.16 — the other side of the same coin. Where Hindu share is higher, Congress lost less. Same caveat.
• DMK swing × anything: all near-zero. DMK lost broadly, not concentrated in particular religion mixes.
• TVK swing × anything: all near-zero. The wave was not communal.

The Muslim-quintile cut

A finer view: group ACs by Muslim% into 5 quintiles, compute median swing per party in each.

Quintile (Muslim %)	n	DMK	AIADMK	TVK	BJP	PMK
Q1 (lowest Muslim%)	47	−15.1	−13.2	+32.8	+17.3	−16.3
Q2	39	−17.0	−14.7	+35.9	+1.4	−15.6
Q3	44	−11.1	−15.9	+33.7	+19.3	−13.2
Q4	33	−17.8	−12.6	+32.8	+17.0	—
Q5 (highest Muslim%)	40	−15.9	−13.1	+36.0	−10.0	−12.5

The visible signal is in Q5 BJP: when you reach the most Muslim-share ACs, BJP's swing goes from +17 pp to −10 pp. BJP went backwards in Muslim-heavy ACs.

But notice — TVK's swing is essentially flat at +32-36 across all quintiles. The TVK wave doesn't care about Muslim share.

What this means

The 2026 result wasn't a communal realignment in TN. The data is more consistent with:

1. Generalised incumbent fatigue — voters rotated AWAY from both DMK and AIADMK at roughly equal rates.
2. A celebrity-led alternative — Vijay's TVK was the available off-ramp; voters took it.
3. Geographic concentration in Chennai metro (see Finding 4) — not religious-block concentration.

The one religious signal that did appear is small but real: BJP underperformed in Muslim-heavy ACs. That's expected and arithmetically obvious; it's not a new political story.

What this analysis cannot see

This is an AC-level analysis using district-level religion data — the Census 2011 sub-district-to-AC mapping wasn't done at finer granularity (SHRUG is auth-walled and we used a coarser fallback). So:

• Within-AC variation is invisible. If Mylapore (within Chennai South) voted differently from T. Nagar, this analysis can't show it.
• District-level religion averages mask intra-district variation. Two ACs in the same district share the same religion mix in this dataset.

Booth-level analysis (Path B) would change both. We have the pipeline; we don't have the OCR done.

Method note

• Religion source: Census 2011 C-01 table (DDW33C-01) at district level.
• AC → district join uses the district name from kracekumar/tn_elections 2021 results, normalised against Census district names (manual aliases for Tuticorin/Thoothukudi, Kanniyakumari/Kanyakumari, etc.).
• 31 ACs (in districts created post-2011 Census — Tirupathur, Chengalpattu, Kallakurichi, Mayiladuthurai, Ranipet, Tenkasi) have no religion attached. Pearson correlations are over the 203 ACs with both fields populated.
• Pearson correlations computed in pure Python (no numpy dependency) — see pipelines/path_a_build.py.