The word “other” is doing a lot of work
When a poll names two candidates and buckets the rest, it is not simplifying the race. It is changing the question.
Integrity Polling research staff · Methods
· 3 min read
Sample data. This article is a demonstration of our analysis format. The figures it discusses are illustrative unless a source is cited.
Open a typical horse-race poll and look at the ballot question. It usually reads something like: “If the election were held today, would you vote for the Democrat, the Republican, or someone else?” Then the results come back 47–45–8, the eight is labeled “other/undecided,” and the write-up is about the two-point gap.
That eight is not a rounding error. In a close race it is the whole race.
What “other” erases
Three things disappear when you collapse the ballot to two names.
The candidate’s name. A voter who intends to vote Libertarian is being asked whether they prefer the Democrat, the Republican, or an abstraction. Some will pick the abstraction. Many will pick the major-party candidate they dislike least, because that is what the question invited them to do. The poll then overstates both major candidates and understates the third.
The size of the bloc. In our demonstration gubernatorial release, 9 percent of likely voters chose a named candidate who was not the Democrat or the Republican. That is more than twice the gap between the leaders. Lumped into “other/undecided,” those voters would have been indistinguishable from the genuinely undecided 6 percent, and the single most useful fact about the race, that it probably turns on where Park’s voters land, would not have been visible.
The accountability. After the election, you cannot grade a poll on a candidate it never measured. If the Libertarian takes 6 percent and the poll had “other” at 8, was that a hit? Nobody can say. Naming the candidate makes the poll falsifiable, which is the entire point of publishing one.
The same race, two ways of asking
Demonstration data. Left column is our actual demonstration ballot test; right column is a simulation of the same respondents answering a two-candidate question with “someone else” as the only alternative.
Integrity Polling demonstration release, Sept. 14–18, 2026 (illustrative). Two-candidate simulation reallocates minor-party voters 55/30/15 to major candidates/other/undecided, in line with reallocation patterns in published forced-choice experiments.
The usual defense, and why it fails
Pollsters who collapse the ballot argue that minor-party support is “always overstated” in polls and fades by Election Day, so naming the candidates inflates them. There is some truth in the first half. Third-party candidates frequently underperform their September polling.
But the remedy for a measurement that tends to run high is to publish it and then publish how high it ran. It is not to stop measuring. A pollster who names every candidate and then grades the minor-party error after the election is doing science. A pollster who removes the candidates from the question is doing something else.
The fade also is not universal. Independents and third-party candidates have won statewide offices and finished second in others. A method that structurally cannot see those outcomes coming is not conservative. It is blind in one direction.
What we do instead
Every ballot-qualified candidate is read by name, in rotated order, in every horse-race question we field. “Another candidate” and “undecided” are accepted but never read. We report each named candidate’s share in the topline chart and the crosstabs, and after the election each named candidate gets a line on the Scorecard with the signed error.
If that produces a number for a minor candidate that turns out too high, we will publish the miss. That is a better outcome than never having asked.