Political analysis becomes dangerous the moment it starts telling us what we want to hear.
Every constituency tempts us with an easy story.
“This is a safe seat.”
“This community votes for that party.”
“This leader has a personal connection here.”
“Turnout increased, so that must benefit X.”
“The party won here last time, therefore its organisation is strong.”
Sometimes these statements contain truth. The problem begins when we confuse a piece of evidence with the whole explanation.
A constituency is not a coloured shape on an election map. It is not one vote-share number. It is not a demographic pie chart. It is certainly not whatever conclusion makes your preferred political theory look intelligent.
If you want to understand a constituency, your first job is not predicting it.
Your first job is trying to disprove yourself.
And few recent examples demonstrate why better than Bhabanipur.
The Mamata Banerjee Example
For years, Bhabanipur in south Kolkata was closely associated with Mamata Banerjee and the Trinamool Congress.
In the 2016 West Bengal Assembly election, Banerjee won Bhabanipur with 65,520 votes and 47.67% of the vote, defeating Congress candidate Deepa Dasmunshi by 25,301 votes. (Janadesh Online)
Then came 2021.
Banerjee chose to contest Nandigram rather than Bhabanipur. There, BJP candidate Suvendu Adhikari received 110,764 votes, compared with Banerjee’s 108,808, a margin of just 1,956 votes. (Election Results)
Meanwhile, TMC’s Sobhandeb Chattopadhyay contested Bhabanipur in the regular Assembly election and won 73,505 votes, or about 58.4%, against BJP candidate Rudranil Ghosh’s 44,786. The margin was 28,719 votes. (Election Results)
Chattopadhyay subsequently vacated the constituency, allowing Banerjee to contest the September 2021 by-election.
She won decisively.
Banerjee received 85,263 votes, or 71.9%, while BJP candidate Priyanka Tibrewal received 26,428, or about 22.3%. The margin was 58,835 votes. (The Indian Express)
If you were building a constituency model from that result alone, the conclusion would have seemed obvious:
Bhabanipur = Mamata Banerjee stronghold.
And that is precisely where good political research should become suspicious.
Because by 2024, the picture was already changing.
During the Lok Sabha election, Bhabanipur formed one Assembly segment of Kolkata Dakshin. TMC won the parliamentary constituency overall, but its lead over the BJP within the Bhabanipur Assembly segment was only 8,297 votes. According to booth-level reporting based on Election Commission data, the BJP led in 149 of the segment’s 269 booths and five of its eight municipal wards. (Telegraph India)
Then came the 2026 Assembly election.
The Election Commission’s official result records BJP candidate Suvendu Adhikari receiving 73,917 votes, or 53.02%, while Mamata Banerjee received 58,812, or 42.19%. The margin was 15,105 votes. (Election Commission of India Results)
That does not mean the earlier description of Bhabanipur as favourable territory for Banerjee was fabricated.
It means that political geography is not permanent.
And that distinction is the beginning of serious constituency research.
Lesson One: Never Confuse the Last Election With the Next Electorate
The easiest mistake in constituency analysis is this:
Candidate A received 60% last time, therefore Candidate A begins with 60% this time.
No.
The previous result tells you what happened under a particular combination of candidates, turnout, alliances, issues, electoral rolls, campaign conditions and political context.
Change those variables and you are studying a different contest.
Look at Bhabanipur’s numbers:
| Election | TMC/Candidate | Main rival | Result |
|---|---|---|---|
| 2016 Assembly | Mamata Banerjee: 47.67% | Deepa Dasmunshi | TMC +25,301 |
| 2021 Assembly | Sobhandeb Chattopadhyay: 58.4% | Rudranil Ghosh | TMC +28,719 |
| 2021 bypoll | Mamata Banerjee: 71.9% | Priyanka Tibrewal | TMC +58,835 |
| 2024 LS, Bhabanipur segment | TMC led BJP | BJP | TMC +8,297 |
| 2026 Assembly | Mamata Banerjee: 42.19% | Suvendu Adhikari: 53.02% | BJP +15,105 |
The underlying figures come from official election results and reporting based on booth-level Election Commission data. (Janadesh Online)
The table is more useful than simply labelling the constituency “safe” or “competitive”.
It shows movement.
And movement is what a researcher should be hunting for.
Lesson Two: Read Booths, Not Just Constituencies
A constituency result compresses thousands of political decisions into one number.
That compression is useful, but brutal.
Imagine Candidate A gets 52% and Candidate B gets 45%.
You know who received more votes.
You still don’t know where the election was made.
Bhabanipur’s 2026 booth-level picture demonstrates the point beautifully. Reporting based on the Election Commission’s Form 20 found enormous differences between polling stations. One booth reportedly gave Banerjee 936 votes and Adhikari 18. At another, Adhikari received 699 and Banerjee 15. (India Today)
That is not one political environment.
That is multiple political environments living inside the same constituency boundary.
Reporting on the final booth data found Adhikari leading across roughly 205 of 267 polling stations, while Banerjee led in around 61, with minor discrepancies between published counts depending on treatment of booth records. (India Today)
Now your analysis becomes more interesting.
Instead of asking:
“Why did Bhabanipur vote this way?”
Ask:
Which parts of Bhabanipur voted differently, and why?
That is a much better research question.
Lesson Three: Demography Is Context, Not Destiny
Another seductive shortcut is demographic determinism.
Find the caste.
Find the religion.
Find the language.
Apply presumed party preference.
Congratulations, you have built a spreadsheet that looks sophisticated and may understand very little.
Bhabanipur is particularly useful because it is unusually diverse. Reporting around the 2021 bypoll described roughly 40% of the electorate as Gujarati, Marwari, Sikh or Bihari, around 20% Muslim and roughly 40% Bengali, although such community estimates should be treated as approximations rather than official electoral classifications. (ThePrint)
The constituency contains eight Kolkata Municipal Corporation wards: 63, 70, 71, 72, 73, 74, 77 and 82. (ThePrint)
But knowing the demographic composition is only the beginning.
You still need to ask:
Did different groups turn out at the same rate?
Are political preferences uniform within those groups?
Did those preferences change?
Are age, occupation, locality and class interacting with identity?
Are voters responding to the candidate, state government, national government or local issue?
Demography can tell you who lives somewhere.
It cannot automatically tell you why they voted.
Lesson Four: Turnout Is Not a Directional Signal
Political commentary loves turnout.
“High turnout means anger.”
“Low turnout helps the incumbent.”
“Women are turning out, therefore…”
Stop.
Turnout tells you participation changed.
It does not independently tell you the political direction of that change.
Bhabanipur’s September 2021 bypoll recorded turnout of only about 53.3%, substantially below the regular Assembly election earlier that year. Yet Banerjee won with 71.9% of the vote. (Hindustan Times)
That alone should destroy the habit of translating turnout directly into partisan conclusions.
The proper questions are more granular:
Where did turnout increase?
Where did it decrease?
Compared with which election?
Among which polling stations?
Did areas previously favourable to one candidate participate differently?
Turnout is a clue.
It is not an answer.
Lesson Five: Separate Candidate, Party and Election
The same constituency can behave differently in a municipal election, Assembly election and Lok Sabha election.
Why?
Because voters are capable of understanding that they are voting in different elections.
A person may have one opinion about their councillor, another about their MLA and another about who should govern India.
This is why Bhabanipur’s 2021 Assembly bypoll cannot simply be compared mechanically with its 2024 Lok Sabha segment result or its 2026 Assembly result.
The electoral contexts were different.
The candidates were different.
The stakes were different.
The opposition configurations were different.
And the broader political environments were different.
The 2024 Kolkata Dakshin parliamentary result illustrates this nicely. TMC candidate Mala Roy won the parliamentary constituency with 615,274 votes and 49.48%, while BJP candidate Debasree Chaudhuri received 428,043 and 34.42%. (The Times of India)
Yet inside the Bhabanipur Assembly segment, TMC’s advantage was far narrower. (Telegraph India)
The parliamentary constituency and its individual Assembly segment were telling different stories simultaneously.
That is not contradictory.
That is politics.
Lesson Six: Look for the Trend Before the Shock
Election analysis is often written backwards.
Something surprising happens.
Then everyone suddenly discovers ten reasons it was “obvious”.
That is storytelling, not research.
The more useful question is:
What evidence existed before the result that challenged the dominant assumption?
With Bhabanipur, 2024 mattered.
A constituency where Banerjee had won the 2021 bypoll by 58,835 votes subsequently produced only an 8,297-vote TMC advantage at Assembly-segment level in the 2024 Lok Sabha election. (The Indian Express)
That did not prove what would happen in 2026.
It did tell researchers that treating 2021 as a permanent baseline would be dangerous.
That is the difference between identifying a signal and making a prediction.
Good research notices the signal.
Bad research turns the signal into certainty.
Lesson Seven: Read Absolute Votes Alongside Vote Share
Percentages are elegant.
People vote in absolute numbers.
Suppose a party’s vote share increases from 40% to 45%.
Great.
But what if total turnout collapsed?
What if its actual number of votes barely moved?
What if the opposition fragmented?
What if the electorate changed?
What if one party’s vote remained stable while another party’s supporters simply did not turn out?
Always keep these four numbers together:
Registered electors → turnout → absolute votes → vote share.
Then add margin.
Only then start telling stories.
Otherwise percentages can become political optical illusions.
Lesson Eight: Electoral Rolls Are Part of the Analysis
The denominator matters.
Bhabanipur has also experienced changes in its registered electorate over time. Published constituency data put the electorate at 205,713 in 2016, while Election Commission documentation for the 2021 bypoll listed 206,456 electors. (Janadesh Online)
Changes to electoral rolls became politically contested ahead of the 2026 election, including allegations by Banerjee concerning deletions during the Special Intensive Revision process. Those claims and disputes should be treated as claims requiring institutional and legal examination, rather than simply folded into an explanation of the result. (Telegraph India)
That distinction matters enormously.
A researcher should record:
What the data establishes.
What a political actor alleges.
What authorities respond.
What remains disputed.
Never mix those four columns.
Lesson Nine: Your Fieldwork Should Be Able to Prove You Wrong
Suppose your dashboard says a party is gaining.
You visit the constituency.
You talk only to people who agree.
You return and announce:
“The ground confirms the data.”
Congratulations. You have conducted a sightseeing tour of your own hypothesis.
Real field research should be adversarial toward your assumptions.
If you believe Party A is gaining, deliberately visit booths where it historically performs badly.
Talk to:
shopkeepers,
students,
women,
older residents,
first-time voters,
housing-society residents,
informal workers,
local journalists,
party workers from multiple sides,
and people who say they do not care about politics.
Then ask open questions.
Not:
“Are people angry about X?”
Ask:
“What matters most to you here?”
The first question manufactures evidence.
The second allows evidence to surprise you.
Build a Constituency Dashboard That Can Disagree With You
A serious constituency file should contain at least five layers.
Electoral: booth-level results, turnout, margins, NOTA and historical elections.
Demographic: population patterns, age, gender, migration, occupation and community composition, with clear sourcing and caution where official data are unavailable.
Geographic: wards, villages, colonies, transport links, markets, institutions and physical barriers.
Governance: roads, water, electricity, housing, schools, hospitals, welfare delivery and recurring complaints.
Political: candidate history, party organisation, alliances, local leadership and campaign context.
Then add the sixth layer that spreadsheets hate:
people.
Interviews.
Conversations.
Contradictions.
Stories that do not fit your model.
Because those contradictions are often where the useful research begins.
The Real Lesson From Bhabanipur
Bhabanipur should not be used to prove that one politician was permanently strong, permanently weak, destined to win or destined to lose.
That would simply create another convenient narrative.
Its value as a case study is methodological.
In 2016, Mamata Banerjee won there.
In the regular 2021 election, TMC retained it.
Months later, Banerjee won the bypoll with 71.9%.
In 2024, TMC’s parliamentary lead within the Assembly segment narrowed substantially.
In 2026, the constituency produced a different Assembly result, with Suvendu Adhikari receiving 53.02% against Banerjee’s 42.19%. (Janadesh Online)
That sequence should make every political researcher slightly uncomfortable.
Good.
Comfort is dangerous in research.
The constituency you think you understand may already be changing underneath your spreadsheet.
The booth you ignored may contain the beginning of that change.
The community you treated as a single bloc may contain five different political conversations.
The turnout number you celebrated may mean something completely different once you open the booth data.
And the “safe seat” in your presentation may simply be a historical description wearing the costume of a prediction.
So the next time someone hands you a constituency map and asks:
“Who is strong here?”
Don’t rush to answer.
Open the previous results.
Open the booth data.
Compare elections.
Check the electorate.
Walk the wards.
Talk to people who disagree with your hypothesis.
Separate facts from claims.
And keep one question permanently sitting at the top of your research sheet:
What evidence would prove me wrong?
Because the hardest part of reading a constituency isn’t finding data.
It is resisting the temptation to make the data say what you already wanted to believe.
Leave a Reply