Lighthouse Research poll trims Kennedy's projected margin in Utah's 4th District
A single September survey nudged our UT-4 forecast toward Jonny Larsen, but Mike Kennedy remains a heavy favorite in our model.
Written by the newsroom · anthropic/claude-opus-5.5 · October 1, 2026 at 10:02 AM EDT
Correction: the House forecasts used in this article predate our redistricting correction. They used 2024 House results on earlier district lines. The current forecasts use presidential results recalculated on the 2026 lines for affected districts. Read the methodology.
Correction: the forecasts in this article predate a polling-data repair. Some imported polls omitted candidates who shared a party, and ranked-choice rounds were not consistently selected. The current forecasts use corrected results and recalculated pollster adjustments. See current forecasts.
A Lighthouse Research survey of 513 likely voters, sponsored by the Utah Debate Commission and fielded September 8-17, found Republican Mike Kennedy at 41.7% and Democrat Jonny Larsen at 32.9% in Utah's 4th District. That is an 8.8-point gap in the poll. Another 5.7% chose another option, and the remaining 19.7% of respondents did not pick any of those choices.
The poll moved our forecast slightly, without changing who is favored. Our model now gives Larsen a 1 in 100 chance of winning, up from 0% in the run published at 9:02 AM ET today. Kennedy's chance stands at 99%, down from 100%. Our projected margin narrowed from R+22.4 to R+20.3.
The new survey shows a closer race than our model expects. Its 8.8-point margin sits near the Democratic edge of our range, which runs from R+34.5 at the 5th percentile to R+6.1 at the 95th. Because the poll stands alone and leaves a large share of respondents outside the two-candidate result, it pulls our estimate only modestly. The earlier range ran from R+36.6 to R+8.2.
The district does not alter the national House picture. Our model gives Democrats a 98% chance of winning a House majority, with an average of 245.5 Democratic seats against the 218 needed. That figure comes from a simulation that links polling error nationally, by census division, by state and within each district. It does not yet tie together demographically similar districts in different regions, because we lack the data to do so, which makes the control probability somewhat firmer than it would otherwise be.
Polls this dispatch was written from
- Lighthouse Research · Sep 17, 2026 · n=513 · UT-4
Every constant behind these numbers is on the methodology page.