A single NC-3 poll narrows our margin, not our verdict

One new survey is on the board in North Carolina's 3rd District. Tulchin Research, polling 600 likely voters from July 27 to August 2, put Republican Rep. Greg Murphy at 44 percent and Democrat Raymond Smith Jr. at 41 percent. The payload lists no margin of error for the survey; at that sample size a three-point separation is close enough that the poll is best read as a close result rather than an advantage for either candidate.

Our forecast moved, slightly. The mean margin in NC-3 is now R+23.9, against R+25.8 in our previous run at 10:30 a.m. ET on Aug. 31. The 5th-to-95th percentile range narrowed correspondingly, from R+42.1 to R+9.7 before, to R+40.1 to R+7.7 now. Murphy's win probability is 99 percent, up from a rounded 100 percent, and Smith's is a 1 in 100 chance where it had rounded to zero. That is a district our model treats as safely Republican, with one poll pulling the central estimate about two points and nothing else in the payload pointing the same way.

The distance between the poll and the model is the story here. A single survey with a three-point gap sits far outside our simulated range, which still puts nearly all of its mass in double-digit Republican territory. Our model discounts an outlier of that size rather than following it, and 64 days out it has not changed how it reads this seat.

Nationally, our average of 22 generic-ballot polls over the past 45 days stands at D+6.7. Our model gives Democrats a 94 percent chance of reaching the 218 seats needed for House control, with a mean of 241.4 seats. One caveat is ours to own: our simulation correlates error nationally, by census division, by state and within each district, but we have no cluster data to link demographically similar districts across regions, so that House number is marginally firmer than a fuller model would make it.