Strange things are afoot at the Marist Circle-K.
On our earlier post on polling and trendlines, a couple of commenters pointed out that the RCP average lead for Kamala Harris increased slightly today after a couple of new national polls dropped. One of these, Marquette Law, shows the national race a dead heat at 50/50. The other, however, came from Marist and shows Harris with a five-point lead and a majority at 52/47. Their previous two iterations put Harris up by 2 at the end of September, and by one at the beginning of the same month:
Harris (52%) leads Trump (47%) among likely voters nationally, including those who are undecided yet leaning toward a candidate. Earlier this month, two points separated Harris (50%) and Trump (48%) among likely voters.
That does look like momentum. Except for this one small problem, that is:
#New General election poll – Independents
🔴 Trump 54% (+10)
🔵 Harris 44%Last poll – 🔴 Trump +4
Marist #A+ – 363 LV – 10/10
— Political Polls (@PpollingNumbers) October 16, 2024
Trump (54%) leads Harris (44%) among independents who are likely to vote, widening the 4-point edge Trump (50%) had against Harris (46%) previously.
Er … wut? Trump gained six points among independents, and yet lost three points in the topline gap? How is that possible?
Answer: It’s not. Something very strange is definitely afoot in this latest iteration. Is it the sample? Could be. Today’s D/R/I split among likely voters is an eye-popping D+7 (40/33/26), which is pretty far off from the R+3 that Gallup sees this cycle (48/45, but with leaners). The previous survey’s LV split was 37/34/28, but then again, the early September iteration’s LV split was 40/35/24. Marist suggests that they are weighting for demographic imbalance, but if so, perhaps not well enough.
We won’t bother unskewing the poll, which is a fool’s errand. Suffice it to say that an LV model in this cycle that uses 40% Democrats is hardly reliable for predictive value. The result of that appears to be an overstatement of Democrat voter intent and a significant understatement of independent voter impact on the poll. The statistically significant shift of support for Trump in even the relatively decent sample of independents (363 respondents) provides a much clearer picture of the trend in this cycle, and that matches up well with what we see in other polls.
One final note: Marist will likely have a final entry in the poll market in two weeks. The sample on that survey will be very interesting to see, especially with their final iteration serving as the measure of their reliability in future polling.
