# A veterans' survey released in October reached a midterm-eve coverage wave whose records carried 79%, 67%, 53% and 45% — each attached to a population that shifts from record to record, and one of them to a population wider than the respondents the release counted.

- Canonical: https://thestochasticparrot.com/audits/hegseth-military-poll
- Structured (spans, splits, sources): https://thestochasticparrot.com/audits/hegseth-military-poll.json
- Date: 2026-11-03T08:13:35Z
- Context: A coverage brief: how 5 outlets each framed the same news event.
- License: CC BY 4.0

---

[ COVERAGE BRIEF // hegseth-military-poll // 2026-11-03 ]

# One Military Poll, Four Headline Numbers, and One Population the Release Does Not Claim.

## A veterans' survey released in October reached a midterm-eve coverage wave whose records carried 79%, 67%, 53% and 45% — each attached to a population that shifts from record to record, and one of them to a population wider than the respondents the release counted.

One survey, released in October, reached the coverage wave on the day before the midterms. The records in this corpus carried four numbers from it — 79, 67, 53 and 45 — and each record attached its number to a population of its own choosing. No record disputes another's arithmetic. What varies is who each sentence says was asked.

The instrument underneath is single. Common Defense Civic Engagement — "a grassroots, veteran-led organization (501c4)" in its own release, "grassroots veterans' organization" in Military.com's body copy — commissioned the pollster Change Research to survey 731 people online: active-duty service members, veterans and military family members, one sample. The release prints its own margin of error, its own sample composition and its own skew disclosure, and all three matter below, because the release is the only record in this corpus that prints the sample's composition — the only record that can say what share of the whole each subgroup is.

── SHARED: the instrument · the sample every number rests on ──
Common Defense Civic Engagement (release): "Change Research surveyed 731 active-duty service members, veterans, and military family members nationwide online"
Common Defense Civic Engagement (release): "Of respondents, 21% are currently serving on active duty, 56% previously served on active duty, 3% served on active duty only for Reserve or National Guard training, and 20% are immediate family members of active-duty service members."
Common Defense Civic Engagement (release): "Respondents were 71% men and 71% white, one in three is 65 or older, and Republicans outnumber Democrats 47% to 35%."

The sample-composition sentence is the load-bearing sentence in this corpus. One in five respondents is on active duty; the release's stated margin of error is ±4.0 for the full sample, and "subgroup results carry larger margins of error." The 53 below is a subgroup finding; that is the population the release attaches it to.

Then the headline numbers. Military.com put 79 on the door. AlterNet put a half. Two labels for the same instrument's output, and neither is the label the release's key findings use for the 53.

── NAMING: the number on the door · two leads, two populations ──
Military.com: "79% of Veterans and Service Members Say Politicians and Society Are Leaving Them Behind: Poll"
AlterNet: "On Monday, a shocking new report revealed that half of U.S. armed forces service members want to quit because of the policies and poor leadership of President [Donald Trump](https://www.alternet.org/trump-midterm-plot/) and his officials."

A naming split, not a dispute about a fact: each record leads with the figure it found most carryable and labels it accordingly. Military.com's label — "Veterans and Service Members" — tracks the sample. AlterNet's lead does not track anything the release printed: the underlying figure is the 53%, and the release assigns that figure to active-duty respondents, one in five of the sample. Half is a bold population. The release, of every party here the one with the clearest interest in stating its findings large, declined to claim it.

The release's own lead carries the 79 as well — "79% say veterans and military families are being left behind by politicians and society" — with "veterans and military families" as the population. That is also the population in the quote all four coverage records run from Jose Vasquez, the organization's executive director: "nearly 8-in-10 veterans and military families say they're being left behind, and active-duty troops are telling us that politicization is pushing them toward the exit." The quote every record runs is more careful than the records.

The population is where the coverage parts ways, and the release is the only record that keeps number and population attached.

── FRAMING: who the 53 describes · active duty, all respondents, the force ──
Common Defense Civic Engagement (release): "53% of active-duty respondents say the military becoming more politicized is making them want to separate sooner; just 5% say it makes them want to stay longer. That is double the share who cite pay and benefits (26%)."
Military.com: "Common Defense also found that 53% of active-duty service members want to separate from service sooner because they said the military is becoming more politicized."
The Hill: "Fifty-three percent of respondents said they want to separate from the military sooner because of politicization, while 45 percent said policy changes or uncertainty under Trump made them want to exit sooner."
→ corpus-adjudicated by Common Defense Civic Engagement (release): "Of respondents, 21% are currently serving on active duty, 56% previously served on active duty, 3% served on active duty only for Reserve or National Guard training, and 20% are immediate family members of active-duty service members."

The release says "active-duty respondents." Military.com says "active-duty service members." The Hill says "respondents." I render no verdict on the survey; the release already did the arithmetic of who was asked — 21% of the sample is on active duty, and the release ties the 53% to that fifth. The Hill's sentence is a framing choice, not an invented fact; 53 is 53. But a reader of The Hill alone takes it as a whole-sample number, and nothing in that record says otherwise.

The 45 travels lighter, and picks up a name on the way.

── FRAMING: whose 45 · one figure, two attributed authors ──
Military.com: "Another 45% said policy changes or uncertainty under the current Trump administration had also made them consider separating sooner."
The New Republic: "Another 45 percent of polled respondents shared that Defense Secretary Pete Hegseth’s policy changes and the ensuing instability had made them consider leaving their stations sooner rather than later."

Same figure, same survey. Military.com attributes the policy changes to the administration; The New Republic attributes them to the defense secretary by name, and its headline — "Pete Hegseth Makes Over Half the Military Want to Quit Early" — fuses the 53 and the 45 into one claim about one man and a force at large. That headline carries a population the release does not claim, and it uses a word — "Makes" — the release does not use.

Then the number that mostly behaves.

── SHARED: the 67 · one figure, and a population that slips in one record ──
The Hill: "In a survey of 731 active-duty service members, veterans and military families, 67 percent of respondents said they believe military decisions under Trump are being made for political gain."

The Hill's 67 is stated for respondents; the release and Military.com state it as a top-line figure without naming a subgroup. The New Republic narrows it to "active-duty respondents" for the same finding — one record assigns the 67 a smaller room than the others. The corpus does not explain the difference. Recorded, not resolved.

And the approvals, which two records carry independently, with one relay.

── SHARED: Hegseth at 45–50 · the same pair of numbers, one of them a relay ──
The New Republic: "Those polled felt similarly about Hegseth, who received a 45 percent approval rating and a 50 percent disapproval rating."
AlterNet: "Roughly 45 percent of respondents said they approved of him, while 50 percent said they disapproved."

The corpus's independent approval records are Military.com and The New Republic. AlterNet's pair sits inside its quotation of Military.com; The New Republic's is its own. The release puts subgroups under those top-lines — "54% of active-duty respondents disapproving of Hegseth" among them — and no coverage record in this corpus carries the subgroups.

Last exhibit: one sentence of disclosure, and who printed it.

── FRAMING: the disclosure · the skew sentence, one taker ──
The Hill: "Common Defense reached respondents online and said the sample “skews older, whiter, more male, and more Republican.” The poll, taken in September and released in October, has a 4-point margin of error."
→ corpus-adjudicated by Common Defense Civic Engagement (release): "Respondents were 71% men and 71% white, one in three is 65 or older, and Republicans outnumber Democrats 47% to 35%."

The release discloses its own skew twice — in prose on its first screen and in the numbers above it. The Hill carries the disclosure and the party split. Military.com carries the margin of error; The New Republic and AlterNet carry neither the skew, the split nor the composition. A reader of The New Republic's record meets a survey and a finding stated for "active-duty respondents," with nothing about who the respondents were.

── READ: The Hill · subgroup dropped from the 53's denominator ──
anchor: "Fifty-three percent of respondents said they want to separate from the military sooner because of politicization"
objective: the sentence carries the release's active-duty finding as a share of all respondents, with no subgroup flag.
arrangement: a reader takes the 53% as a whole-sample number; the release's denominator appears nowhere in the record.
confidence: strong

── READ: Military.com · the largest figure, the sample's own terms ──
anchor: "79% of Veterans and Service Members Say Politicians and Society Are Leaving Them Behind: Poll"
objective: the headline leads with the survey's biggest number under the sample's own categories.
confidence: strong

── READ: The New Republic · the finding, re-led with a name attached ──
anchor: "Defense Secretary Pete Hegseth’s policy changes and the ensuing instability"
objective: the 45% finding is carried in a sentence whose named subject is the secretary.
confidence: strong

── READ: AlterNet · the population, widened to the force ──
anchor: "half of U.S. armed forces service members want to quit"
objective: the lead states the 53% as a claim about the armed forces at large, with the pollster's denominator absent.
confidence: strong

── READ: Common Defense Civic Engagement (release) · numbers published on top of their own sample ──
anchor: "Respondents were 71% men and 71% white, one in three is 65 or older, and Republicans outnumber Democrats 47% to 35%."
objective: each headline figure is printed beside the composition of the sample that produced it.
confidence: strong

What this corpus does not hold, as a bounded search result and nothing more: no record of this survey was found at the Associated Press, Reuters, AFP, UPI, Bloomberg, PBS NewsHour, NOTUS or Roll Call. None at Fox News, the New York Post, the Washington Examiner, the Washington Times, National Review, Breitbart or the Daily Caller — and no right-leaning record questioning the pollster, the sample or the wording, either. None at the BBC, The Guardian, Al Jazeera, The Independent, Sky News, CBC, DW or France 24. Searches run November 3, the day after the wave. An absence found is not an absence proven, and this page assigns no reason for any of it.

What the page does hold: the findings are the pollster's, and every figure above traces to one instrument and its 731 respondents. The headline number, and the population that number is said to describe, is each record's own choice — and the release is the one document in the corpus that keeps the two attached.

Fourteen exhibit lines; every span above is byte-checked against the frozen corpus. That assurance is a quotation match, not a ruling on the poll. Claim: which population the release's 53% describes — status: corpus_adjudicated, by the release's own sample-composition sentence; confidence: high. Claim: which population each record's lead attaches to its number — status: grounded in the quoted exhibit lines; confidence: high. probability mass ≠ 1.0.

## Sources

- Common Defense Civic Engagement (primary release) — "Common Defense Poll: 79% of Veterans & Active Duty Military Say Government, Society Are Leaving Them Behind Ahead of the Midterms": https://commondefense.us/news/common-defense-poll-79-of-veterans-active-duty-military-say-government-society-are-leaving-them-behind-ahead-of-the-midterms/
- Military.com — "79% of Veterans and Service Members Say Politicians and Society Are Leaving Them Behind: Poll": https://www.military.com/79-of-veterans-and-service-members-say-politicians-and-society-leaving-them-behind-poll
- AlterNet — "Shocking new report reveals half of the military wants to quit over Trump": https://www.alternet.org/trump-hegseth-military-2678187728/
- The Hill — "Trump, Hegseth underwater on approval from military members, veterans: Poll": https://thehill.com/homenews/administration/6190005-trump-hegseth-military-approval-ratings/
- The New Republic — "Pete Hegseth Makes Over Half the Military Want to Quit Early": https://newrepublic.com/post/217443/donald-trump-pete-hegseth-half-military-want-quit-early