excess-entry

How many people contest a gram-panchayat head seat in India, and what that does to the vote. A descriptive study of candidate entry across four states, built on the in-rolls local-election data.

The name is the excess-entry theorem (Mankiw and Whinston, 1986): with free entry and a business-stealing externality, more competitors pile into a market than is efficient. Panchayat elections are non-partisan, so there is little party gatekeeping to consolidate the field, and the equilibrium number of candidates scales with the value of the seat over the cost of standing. The data show that scaling, and the welfare cost that comes with it: winners elected on a quarter of the vote.

Findings

All posts are gram-panchayat heads and comparable (sarpanch, mukhiya, pradhan).

Contestation scales across states. Mean candidates per seat: Uttarakhand 3.1-3.9, Rajasthan 6.0, UP 7.5, Bihar 12.0. The distributions barely overlap; Bihar's 25th percentile (8) beats UP's median (7) and Rajasthan's 75th (8).

state (year)	p25	p50	p75	mean	median effective N
Uttarakhand (2019)	2	3	4	3.1	3.0
Rajasthan (2020)	3	5	8	6.0	—
UP (2021)	5	7	9	7.5	3.5
Bihar (2016)	8	11	15	12.0	5.5

The votes split, and the mandate erodes. The typical Bihar mukhiya wins with 28% of the vote; 92% of winners fall short of a majority and 67% short of a third. In 92% of Bihar seats the result is "split-decided": the votes held by third-place-and-below candidates exceed the winner's margin over the runner-up, so the field could have flipped it. UP and Uttarakhand are milder (78% and 66% split-decided) but the pattern is monotonic everywhere: more candidates, weaker winners, more split-decided seats. Condorcet cycles are not observable here (no ranked ballots), but this is the condition under which the plurality winner is unlikely to be the majority's choice.

Reservation swaps the field, it does not shrink it. Women-reserved and open seats draw essentially the same number of candidates at every quartile, in every state. The caste dimension moves contestation and only in two states: SC-reserved seats are more contested than General in Bihar (14.9 vs 10.7) and UP (8.5 vs 7.2), flat in Rajasthan and Uttarakhand.

Namesake candidates run above chance. Two candidates share an identical name on 3.7% (Uttarakhand) to 7.0% (Bihar) of ballots, 1.6x to 7x more than random name assignment predicts, and almost all are genuinely different people (same name, different father). Clusters reach five identical names in Bihar, seven in UP. Consistent with engineered vote-splitting, though intent is not identified here.

Data

Clone the four source repos next to this one (or pass --root):

• local_elections_rajasthan — sarpanch 2020
• local_elections_bihar — mukhiya 2016
• local_elections_up — pradhan 2021
• local_elections_uttarakhand — pradhan 2008/2014/2019

Reproduce

pip install -r requirements.txt
ROOT=..                                   # dir holding the four source repos
python src/build.py        --root "$ROOT" --out tables/candidates.csv.gz
python src/contestation.py --cand tables/candidates.csv.gz
python src/mandates.py     --cand tables/candidates.csv.gz
python src/namesakes.py    --cand tables/candidates.csv.gz

Deterministic. build.py harmonizes the four datasets to one candidate-level frame (602,404 candidates, 87,838 GP-head seats); the three analysis scripts write CSVs to tables/ and PNGs to figures/.

Layout

src/build.py          harmonize the four states -> candidate-level frame
src/contestation.py   candidates per seat by reservation/caste/year, quartiles, effective N
src/mandates.py       winner vote share, sub-majority winners, split-decided share
src/namesakes.py      same-name candidates per seat vs a chance baseline
tables/               generated CSVs
figures/              generated PNGs

Caveats

• Effective N and mandates use per-candidate votes, available for Bihar, UP, and Uttarakhand but not Rajasthan (its file records only winner and runner-up votes).
• Cross-state levels partly reflect electorate size. Bigger panchayats mechanically draw more candidates, so the gradient mixes fragmentation with seat size. Candidates-per-thousand-electors is the fix and is not yet in the repo. Within-state cuts (Uttarakhand over time, reservation within a state) are cleaner.
• OBC labels. Bihar's EBC scheme and a UP spelling variant leave some backward-class seats in an "Other" caste bucket; those seats are real but uncategorized.
• "Split-decided" bounds the spoiler potential, it does not prove the plurality winner is not the Condorcet winner (that needs second preferences). "Namesake" is a form, not a proven intent; the within-state shuffle is a permissive chance baseline because names cluster geographically.

License / attribution

Analysis code under MIT. Underlying data are the in-rolls local-election datasets; cite those repos and their sources.