Ask most campaigns how their constituency is performing and you will get a single number — a vote share, a swing, a sense of momentum. It will be an average. And like most averages taken across highly heterogeneous units, it will describe a place that does not exist.
A standard Assembly constituency in India contains somewhere between 250 and 350 polling booths. Those booths differ from one another in almost every way that matters: demography, turnout history, party loyalty, local grievance, the presence or absence of a functioning worker. Averaging across them does not summarise the constituency. It conceals it.
What the average hides
Consider a constituency showing a 4% swing toward your candidate between two cycles. That number invites a comfortable conclusion: things are moving in the right direction, keep doing what you are doing.
Now decompose it. The 4% is the net of booths swinging 15% toward you in the industrial periphery and booths swinging 8% away from you in the old town. Two entirely different political situations are occurring simultaneously, and the average has averaged them into a message that instructs you to change nothing.
The distribution is the finding. The average is the thing that stopped you from seeing it.
The data already exists
The obstacle to booth-level analysis is not data availability. Form 20 — the Final Result Sheet prescribed under the Conduct of Elections Rules — publishes the votes polled by each candidate at each polling station. It is a public document.
The obstacle is that Form 20 is published in formats that vary by state and by year, frequently as scanned images, sometimes with inconsistent booth numbering between cycles. Turning three cycles of it into a usable time series is genuinely laborious. That labour is the entire barrier, and it is a one-time cost per constituency.
What booth-level work makes possible
- Classification of every booth as reliably yours, reliably theirs, or genuinely contested — which immediately identifies where persuasion is even possible.
- Identification of the specific booths where the margin was decided in previous cycles, which is almost never the booths where campaigns concentrate their visible activity.
- Numeric vote targets per booth derived from the constituency margin requirement, so coverage becomes a measurable number rather than a claim.
- Detection of booths where turnout has historically underperformed relative to comparable areas, which is usually an organisational failure rather than a political one.
The uncomfortable finding
The most common result of a first booth-level analysis is that the campaign has been spending most of its resources in areas that were already voting for it. This is not stupidity. Supportive areas are pleasant to campaign in, the crowds are larger, the reception is warmer, and the candidate returns from the day feeling the campaign is going well.
The booths that decide the seat are usually the ones where the reception is coldest, which is precisely why they have been neglected.
Correcting this is not analytically difficult. It is politically difficult, because it requires telling a candidate that the rallies they enjoyed most were the least useful thing they did. That conversation is the actual work.
Where to start
- 01Obtain Form 20 for your constituency for the last three cycles. Budget real time for cleaning it.
- 02Build a booth-level table: votes per candidate, total polled, electors, turnout percentage, per cycle.
- 03Compute swing per booth between consecutive cycles and rank the booths by volatility.
- 04Overlay your current organisational presence. The gap between high-volatility booths and active committees is your first priority list.
None of this requires a consultancy. It requires the decision to work at the level where elections are actually decided, and the willingness to act on what you find.