The 2026 election was measured to plus or minus 0.6 points. The Italian-American vote was not.

On 28 October, la Repubblica Europe published a piece on Italian-American voters before the US midterms. It ends with a promise. Whether Italian Americans are turning away from the Republicans like Hispanic voters "remains unclear," and "the 3 November vote will help answer that question."

The Calandra Italian American Institute said the same thing in December 2024, in its Data Snapshot on Italian-American voting patterns: "Certainly, the midterm elections in 2026 will indicate the scope of the change on voting behavior."

Both sentences assume there is an instrument that can see the cell. I went looking. As far as I can find, there isn't one.

What the 2026 instrument actually is

The US networks' 2026 electorate survey is The Voter Poll by SSRS, run for ABC, the Associated Press, CBS, CNN, Fox News Media, MSNOW and NBC. Its methods statement, last updated 30 October 2026, describes roughly 100,000 interviews between 19 October and 3 November, producing a national sample above 75,000 and state samples for all 50 states. The national margin of sampling error is ±0.6 percentage points, assuming a design effect of 3.0. The AP's own write-up on 4 November calls it "an expansive survey of about 80,000 voters."

That is more precision than any election survey I have been able to inspect. Read the weighting table, though. The registered-voter weight includes self-reported age, gender, race/ethnicity, education and recalled 2024 vote, plus voter-file region, party registration, home tenure, a 2024 candidate-support score, turnout and early-vote status. Some states add modeled religion, and Texas adds Hispanic by age. New York, where Staten Island sits, adds race by educational attainment, plus age and gender by early-vote status.

The word "heritage" appears exactly once in the whole document:

Voter file: Muslim or Middle Eastern heritage (MI)

Not Italian. The plausible reading is that heritage gets a variable where it is part of the political story. In 2026, that is Michigan.

If the questionnaire asks about Italian ancestry, the published weighting does not use it, and I could not find a single released table that reports it. I checked the methods statement and the AP's 4 November summary of how key demographic groups voted. The cuts it reports are age, Hispanic voters, party identification, and issue priorities. No ancestry cut anywhere in it.

The one ancestry-resolved poll I could find

In October 2023, SocialSphere fielded a national survey of 544 Italian American registered voters for the Italian American Policy Center, and John Della Volpe signed the memo on 14 December 2023. The credibility interval is ±4.2 points.

Top lines from that memo: 41% Republican, 30% Democratic, 27% independent. 51% said they voted for Trump in 2020, 44% for Biden. In a 2023 head-to-head, Trump led Biden 46 to 34.

So the only instrument that can see the cell has 544 people in it, was commissioned by an advocacy organization, and last looked in October 2023. The instrument with 75,000 people cannot see the cell at all. That is not a gap that more interviews close. It is a gap that a question closes.

The county proxy, and why its precision doesn't help

The usual move is geographic: take a county with a lot of Italian ancestry and read its result.

Richmond County, Staten Island, is the county people reach for. The ACS 2020–2024 five-year estimates, table B04006, count 113,548 people reporting Italian ancestry out of 494,956 residents, or 22.9%, with a margin of error of ±4,522 people. Note the definition: B04006 counts people who report that ancestry, alone or in combination. Table B04004 asks a narrower question, single ancestry only: 65,214 people, or 13.2% of all residents. Two correct numbers, same county, same vintage, nine points apart. The share is a resolution choice before it is a fact.

Now the result. New York's 11th district, 100% of 377 election districts reporting: Nicole Malliotakis 101,719 votes, 56.1%; Mike DeCillis 79,370, 43.8%. The AP called it at 9:26 p.m. ET. Inside the district, Richmond County went 60.0% to 39.9% for the Republican; the Brooklyn part went 45.2% to 54.7% for the Democrat.

That is as precise as election data gets. A vote tally with every district reporting, not an estimate. And it says nothing about the question. 22.9% is a share of the county, not the county's vote. 60% is the county's vote, not the vote of that 22.9%. Multiplying them together would be a number with two decimal places and no meaning.

The past comparison is a collage of instruments

The line people draw back to is the New Deal. For New York City's Italian Americans, Roosevelt took 80.5% of their vote in 1932 and 78.7% in 1936, while La Guardia, one of their own, took 62.2% in 1933 and 62.6% in 1937. Those figures are quoted by Stefano Luconi in "Italian Americans and the New Deal Coalition" (Transatlantica, 2006), citing Bayor.

They come from historical scholarship, not from a survey series, and they are city-specific. The 2023 panel is a national sample of 544. The 2026 number is a county tally. Put them on one axis and the word "trend" is doing work no instrument supports.

What the absence means

This is not imprecise data. It is absent data. More interviews cannot fix a question that is not asked, and 0.6 points of precision cannot be spent on a cell nobody collected. Precision is a property of the answer you chose to collect, not of the instrument in general.

There are two honest readings of the absence, and both can be true.

The first is failure: nobody measured it, so any claim about the 2026 Italian-American vote is a claim without a receipt.

The second is information: the cell is missing because ancestry stopped predicting votes. Chris Stirewalt put the case in 2022, arguing about Hispanic voters by analogy: "there are no polls that even consider the question because no one would bother asking it." His point was that a group's heritage is not politically salient just because the group is.

The instrument supports the second reading. Heritage gets a variable in Michigan for Muslim and Middle Eastern voters, because there it changed a primary. Questionnaires are built from salience, not from completeness.

Which makes the two sentences I opened with strange. They promise an answer from an election that was never instrumented to give one. If the cell is not worth asking, it is not worth promising either.

What would fill it

One ancestry question on The Voter Poll, asked nationally, weighted like the rest. That is one line on a form that already interviews around 100,000 people. Until someone adds it, any 2026 "Italian-American vote" number in circulation is one of three things: a small panel, a precinct proxy, or invented.

Dead ends, left in

• Calandra's own snapshot says its method is "not a calculated correlation; it is a visual comparison," and it limits its claims to New York City. I did not turn its maps into a number.
• I could not find a 2026 release with an ancestry cell. Absence from the documents I could open is not proof of absence from every table that exists. That gap stays open.
• The IAPC memo reports top lines, not the questionnaire or the full cross-tabs. I quote the memo and leave the rest unread.

The vote on Tuesday answered a lot of questions. It could not answer this one. The question was never asked.

Amendment, 5 November: the two near-misses, opened and checked

Someone who follows polling sent back the answer to the question this piece ends on. The structural half is the useful half: The Voter Poll samples from state voter files, and voter files do not record ancestry. The group could not be oversampled or weighted even if a network wanted to. The cell is not missing by oversight. There is nothing to weight from.

The same answer pointed at two instruments that do reach ancestry. I opened both instead of quoting them.

The first is real, and I had missed it. America's New Majority Project fielded a Columbus Day survey on 1-5 October 2025 (McLaughlin & Associates with Gingrich360, 2,000 registered voters, online, +/-2.2 points). Its charts do carry an "Italian" column, next to White, Hispanic, Black and Asian. What the column covers is the point: it appears on the Columbus and historical-figures questions, and it does not appear on the 2026 generic-ballot slide. The survey's own summary says Italian Americans are only slightly more favourable toward Columbus than the public, and "not a statistically significant difference." The size of the Italian cell is not printed anywhere in the eleven pages.

The second is the FDU Poll in New Jersey (Dan Cassino), fielded 28 April to 6 May 2023, 817 adults, with an ancestry item as question D7A. It is the only systematic ancestry question in electoral polling I have now seen. It found that 17 per cent of New Jersey adults claim Italian ancestry, that Italian men are 57 per cent Republican against 45 per cent of non-Italian white men, and 59 against 45 on the 2020 Trump vote. Italian women show no gap at all. Seventeen per cent of 817 is about 139 adults, so a men-only cell is roughly 60 to 70, and the release does not print that n either. On my arithmetic (theirs is not published), a 12-point gap in a cell that size is about 1.8 standard errors, short of the conventional two. It is New Jersey, and it is 2023.

I looked at FDU's poll index on 5 November 2026 and no 2026 Italian release surfaced.

None of this moves the claim of the piece. It moves the gap. Yesterday the sentence I could not write was "no instrument reaches this." What I can write now is narrower and better: the instruments that do reach ancestry are a 2025 cultural question with no published cell size, and a 2023 state survey whose cell is too small to carry the gap it reports. Both sit under the resolution a claim about a change in a vote would need. The question is not only unasked. The two places it was half-asked were too small to hear an answer.