This is the problem with momentum polls

You skew the top line by oversampling the side that is excited

So the London Times (YouGov) poll comes out showing Harris leading by four points (49-45). This is a flip from their July poll that had Trump leading by a two points. However, if you dig into the cross-tabs, you see something completely counterintuitive to such a result. Let me show you the difference when you take a peak at the cross tabs and modeling.

The first column of numbers is GOP voters, the second is Democrats, and the third is independents. The number of votes and then the turnout prediction.

Back in July, the cross-tab modeling suggested that Trump should have been up by 0.74% while the poll itself put him up by 2. This was a modeling adjustment of 1.26% in the favor of Harris (slanted in favor of Trump). Now in September, that same modeling would put Trump up by approximately 1% while the poll now shows him down by four. This is a modeling adjustment of 5% in favor of Trump (slanted in favor of Harris).

Looking into the details, you see Trump gaining with Republicans and Democrat, while holding steady with Independents. His share of the vote increase by 1.44%. Harris loses a point with Republicans, gains a point with Democrats, and gains four points with Independents, and increases her voter share by 1.18%. Overall, Trump models shows that Trump gained a quarter of a percentage since July, but the topline will show that Harris gained six points, which is what everyone will see.

This topline is accomplished in one way and one way only. By changing the demographics of who you believe will turn out. In this case, the modeling used is different from the standard modeling I am using (which probably slants a bit towards the Democrats) by a factor of five points, and by a difference of over six points from the previous poll they released. In other words, the demographic breakdown of the September poll is over six points different from the demographic breakdown they used in July, and that change is entirely to the benefit of Harris and the Democrats. This is the sole reason that their September poll moved so much in favor of Harris over Trump.

While this is not always the case, it seems to be more often than not when you see wide swings. Sometimes you do see changes (especially in the independents), and when you do, then you are seeing a real swing. But with these sorts of polls, you are seeing little more than noise.