How many of the 329 dummies actually mattered

Three weeks before polling, I published a list of 329 suspect candidate pairs: major-alliance nominees who shared a name, exactly or near-exactly, with another candidate in the same constituency. Almost always an independent. The thesis was straightforward. At least some of these are deliberate vote-splitters, fielded to confuse voters at the EVM and bleed the major candidate’s tally.

Today the votes are in. So: did it work?

The honest answer is the kind that does not write its own headline. In the cleanest test, the strategy did not move a single seat. In the noisiest test, two razor-thin TVK losses had enough flagged-namesake votes to plausibly cover the margin. Both can be true at once, and which one you treat as the “real” answer depends on how generously you classify a dummy.

What is unambiguous: this was a wave year, and waves drown small mechanics.

The strategy did not fail in 2026. It ran into a year where margins were too wide for it to matter.

The headline number, with two definitions

Of the 329 flagged pairs, 263 had both the major and the suspect actually contest. They were spread across 221 distinct major candidates. 152 of those majors lost their seats.

How many of those losses were “consequential,” meaning the dummy votes were at least equal to the margin of defeat?

The strict test is the one the headline pipeline uses, and the one I think the data supports for any causal claim. EXACT-tier and NEAR_FULL-tier dummies are the least likely to be coincidence. By that definition, no major candidate lost a seat where engineered name-confusion plausibly covered the gap.

The inclusive test adds WORD_MATCH dummies: cases where one significant name fragment is shared, often a common Tamil first name like Saravanan or Murugan. That produces two cases. Both are TVK losses.

I would not stake a causal claim on either. WORD_MATCH dummies are the noisiest signal in the dataset. A constituency with common Tamil name patterns will throw up a few of these by chance. But they are the only two seats in the entire 263-pair tally where the arithmetic comes close to working, and both being narrow TVK losses in a wave year is worth flagging. Both have margins under half a percent of polled votes, which is the bottom 1% of the statewide distribution.

Why the answer is small, and why that is the finding

The median dummy in the dataset polled about 200 votes (174 for EXACT, 210 for NEAR_FULL, 197 for WORD_MATCH). The median margin of loss across the 152 lost seats was 27,002 votes. The 25th-percentile margin was 9,554 votes.

These are not seats where 200 to 1,000 confused voters tip outcomes. They are seats where the wave carried the winner home by tens of thousands, or where the constituency’s traditional base voted as it always has.

The strategy needs a close election to bite. 2026 did not have many.

To put a number on it: in only 3 of 152 lost seats did the major lose by under 1,000 votes. In only 15 did they lose by under 5,000 votes. The rest ran into a winner who was not winning narrowly.

The dummy-candidate strategy did not “fail” in 2026. It ran into a year where margins were too wide for it to matter. In 2016, when AIADMK won several seats by under 1,000 votes, the same 329 pairs would almost certainly have produced a non-zero strict-test count.

By tier

EXACT-tier dummies (identical name after normalising prefixes and initials) are the most likely to register as engineered confusion. They polled the fewest votes per pair on average. The ones that look most engineered drew the smallest crowds. That is a useful corrective to the assumption that brazen dummies do more damage.

By alliance

The ruling DMK was the most-targeted alliance, the predictable pattern: spoilers cluster on incumbents. TVK attracted 61 dummy pairings despite being a debut party. The strategy was deployed defensively against the wave, not just by the usual operators. Total votes across all 263 pairings: roughly 79,823, small per seat, zero in the column that matters.

The closest the math came to working

Ratio = combined dummy votes divided by margin. Above 1.0 means the dummies polled more than the gap. Four of the top five near-misses are TVK losses, not by design but by distribution: the wave seats TVK won were not close, so TVK’s narrow losses are the only TVK results that show up here at all.

The TIRUKKOYILUR case: TVK’s Vijay R Baranibalaaji lost to AIADMK’s Palanisamy S by 285 votes (0.13% of polled). A flagged independent polled 571. The door is open. The data does not show anyone walking through it.

The full near-miss table, sortable by ratio and margin percent, is in the interactive dashboard.

What “3 crore votes that elected nobody” looks like, three weeks later

In the earlier post, NTK’s strategy of running everywhere was the clearest illustration of how Tamil Nadu’s elections work: 19,72,537 votes, 4.00% share, zero seats. The dummy-pair data is the same arithmetic at lower magnitude. The roughly 80,000 flagged-dummy votes are a third example of votes that were cast, were counted, and changed nothing. It is the structural feature of first-past-the-post in a fragmented field.

The mechanism is waiting for a year where margins are narrow. 2026 was not it.

Explore the data