India’s parliamentary constituencies are very large by global standards: the median Lok Sabha
Parliamentray Constituency (PC) electorate reached 1.82 million registered electors in 2024. The
conventional reading of Indian turnout has been that very large PCs suppress voting; we show this
“size penalty” is now a compositional artefact. Across 2,171 PC-elections in 2009, 2014, 2019 and
2024, the unconditional small-vs-large decile turnout gap halved from +22.86 to +12.03 percentage
points (pp), yet the conditional 1 M-vs-2 M turnout gap crossed from +1.42 pp (95% CI: −2.85, +5.69)
in 2009 to −6.16 pp (95% CI: −10.30, −2.02) in 2024 once urban, SC, ST shares, Esteban–Ray
linguistic polarisation, Shannon linguistic diversity (122-language grain) and polling-station density
are controlled. Urban share is the largest single compositional channel associated with female
turnout at 52.4 % (95% CI: 44.6, 60.2), followed by linguistic diversity at 28.2 % (95% CI: 21.4, 35.1)
and linguistic polarisation at 20.6 % (95% CI: 14.3, 27.0); linguistic diversity also displays an
opposite-sign gender split (men: −8.2 %). A turnout-maximising delimitation counterfactual that
splits 543 PCs into 824 (59 two-way, 111 three-way) holding parent-PC compositional covariates
fixed is predicted to raise national turnout by +2.32 pp (95% CI: +1.43, +3.21) under our preferred
specification, with substantial sensitivity to alternative specifications: re-running the same exercise
under three alternative specifications returns +1.42 pp, +1.17 pp and +0.30 pp respectively, giving a
defensible range of +0.3 to +2.3 pp. The plan delivers a small female-favourable tilt of +0.21 pp (95%
CI: −0.14, +0.56).
