The Poll That Moved a Race
Six weeks before the mayoral election, the Riverbend Ledger's site posts 'Holloway Surges Into Lead in Mayor's Race.' The story rests on a poll by the state university's polling center: Holloway 47%, Ruiz 44%, undecided 4%, n=402 likely voters, margin of sampling error plus or minus 4.9 points, conducted Sept. 12-16 by live interviewer on landline and cellphone, and the release includes the topline and demographic tables. The second paragraph says Holloway 'has momentum,' citing a July poll that showed Ruiz ahead 46-44. The story does not mention that the university center was paid $6,000 by a nonprofit whose board includes Holloway's former campaign treasurer. The story has been shared 4,000 times when Ruiz's manager emails, demanding equal treatment, and attaches an internal campaign poll (n=800, margin plus or minus 3.5) showing Ruiz 49, Holloway 45.
1. Does the data support 'surges into lead'? Walk through the arithmetic and state the accurate characterization.
Read the worked answer
No. Three separate problems, in ascending order of seriousness.
The lead is inside the error band. A 47-44 difference is three points with a margin of error of 4.9. The reported leader's share could plausibly be anywhere from about 42 to 52; the race is a statistical tie and the story should say so.
Comparing two candidates compounds error. The margin of error on a difference between two candidates in the same sample is larger than the single-candidate figure - commonly approximated as roughly twice the margin for each share, though it is smaller than that when candidate preferences are correlated, as they usually are. Either way, three points is nowhere near decisive.
'Momentum' requires two comparable polls. The July poll is cited with no method, sample, sponsor, or population. The movement - Ruiz 46-44 to Holloway 47-44 - is a handful of points and sits inside both surveys' error bands. Unless the two polls used the same question wording, population, and mode, a shift between them cannot be called a trend. 'Surges' and 'momentum' are causal claims the data cannot carry.
Accurate framing: 'Holloway 47%, Ruiz 44% among likely voters; race is a statistical tie.' Also worth reporting: 4% undecided is unusually low six weeks out, and the sample of 402 produces large subgroups with much wider error bands - any crosstab claim about, say, independents or a ward should be treated as noise.
2. What should the story have disclosed, and how should the newsroom handle the campaign's internal poll?
Read the worked answer
Disclosure the story owed readers: who conducted the poll, who paid for it, field dates, population screened (likely voters, and how 'likely' was defined), sample size, margin of sampling error, mode, question wording, and whether the full topline was released. The $6,000 from a nonprofit whose board includes Holloway's former campaign treasurer is a material funding connection and belongs in the story - not as an accusation, but as a fact a reader needs to weigh the number. A university's name is not itself a disclosure.
On the internal poll:
- It is a partisan source, released by the campaign that benefits, at a moment chosen to shape coverage.
- The email supplies n and margin but no field house, dates, question wording, or full topline.
- Treat it as a claim, not a finding: 'A poll released by the Ruiz campaign put Ruiz ahead 49-45, a difference within that poll's margin of error of plus or minus 3.5 points.' Attribute in the sentence, in the headline if it is covered at all.
- Ask for the full instrument and methodology, and say in the story if the campaign declines.
- Do not commission matching coverage, and do not present it as equal in weight to an independent public poll - but do not suppress it either; readers can handle a labeled partisan number.
Do not let the complaint drive the correction. The error is the paper's, not the campaign's.
3. Write a corrected lede and state a rule the newsroom should apply to every future poll story.
Read the worked answer
Corrected lede:
Holloway leads Ruiz 47% to 44% among likely voters six weeks before the mayoral election, according to a poll released Monday by the State University Polling Center - a difference within the survey's margin of sampling error of plus or minus 4.9 percentage points, which means the race is statistically tied.
The poll of 402 likely voters was conducted Sept. 12-16 by live interviewers calling landlines and cellphones. It was paid for with $6,000 from the Riverbend Civic Trust, a nonprofit whose board includes Holloway's former campaign treasurer. The center released the full topline and demographic tables.
A July poll showed Ruiz ahead 46-44; that survey's methodology was not reported, and the movement between the two is within the error bands of both.
The rule, stated as a desk checklist:
- No verb - surge, lead, momentum, collapse, gain - unless the change exceeds the margin of error on the difference.
- Every poll story carries: sponsor, funder, field house, dates, population, n, margin of error, mode, and question wording or a link to it.
- Compare polls only when method, population, and wording are comparable; otherwise report each separately.
- Label partisan and advocacy polls in the sentence and the headline.
- Say 'statistical tie' when the gap is inside the error band, and explain in plain language what that means.
- A correction on a poll story gets the same prominence as the original, plus a note on what changed.