Finding 3 — BJP overperformed in reserved seats

In the 44 SC/ST reserved ACs, BJP gained +28.5 pp on median; in the 189 General ACs, +12.5 pp. The gap is unusual and worth a closer look.

The headline contrast

Party	Median swing — SC/ST Reserved	Median swing — General	Ratio
TVK	+33.2	+34.0	0.98× (essentially flat)
BJP	+28.5	+12.5	2.28× larger in reserved
AIADMK	−16.2	−13.3	Lost 1.2× more in reserved
PMK	−21.0	−14.0	Lost 1.5× more in reserved
DMK	−14.5	−15.6	Similar
INC	−11.8	−13.2	Similar
NTK	−2.6	−2.3	Similar

Sample-size caveat

BJP contested only 45 ACs in 2026 (NDA seat-sharing with AIADMK + PMK). Of those, 7 were reserved (SC/ST) and 38 were General. With n=7 the median is a small-sample summary — not a tight statistical estimate.

That said, the direction is consistent: in every SC reserved seat where BJP contested, the swing was meaningfully higher than the BJP average. The pattern is unlikely to be noise even at n=7.

Two compatible interpretations

Possibility A — Candidate-quality effect

NDA leadership often allocates BJP candidacy in reserved seats to high-profile community figures (former MPs, retired bureaucrats, community leaders). The candidate-quality lift could be doing the work rather than party-brand swing.

Possibility B — SC-vote consolidation

Some recent psephology suggests SC voters in TN are diversifying away from the historical DMK alliance. If BJP picked up displaced SC voters where it ran, this is consistent with the broader trend.

The data here can't distinguish A from B. Candidate-level information (party-symbol-versus-name) is in the dataset but caste tags per candidate are not.

The mirror: PMK lost more in reserved seats too

PMK ran in 31 ACs and lost a median 21 pp in the 3 reserved seats they contested, vs 14 pp in the 28 General seats. That's a 1.5× larger loss in reserved seats. Caste arithmetic: PMK's Vanniyar base is overrepresented in non-reserved seats; reserved seats are where PMK was relying on transferred SC support.

What this isn't

This is not evidence that BJP "won" reserved seats. They contested only 7, and the absolute final vote share matters more than the swing direction. We're seeing change, not standing.

For absolute 2026 vote share by party × reservation status, drill into the raw parquet:

SELECT party_norm, constituency_type, AVG(pct_2026) AS avg_pct
FROM read_parquet('s3://tnelection2026/insights/curated/swing_2021_to_2026.parquet')
GROUP BY 1, 2;

Method note

• Reservation status derived from the Constituency Type column in kracekumar/tn_elections 2021 data (values: General, SC, ST).
• 44 reserved ACs + 189 General ACs = 233 (one AC's reservation status was unmappable due to a join nuance).
• "Median swing" is across all (AC, party) rows; n = ACs where the party contested in either 2021 or 2026.