CBS News / YouGov: Braun +3 (IN), Sinema +3 (AZ), Tied in FL

[DataGuy]

YouGov is a terrible pollster. Don’t take seriously

It seems to me that pretty much every pollster is "terrible" according to this forum.

In reality, YouGov is B-rated pollster with a mean-reverted bias of only D+0.3 over 375 polls analyzed. That is not bad at all.

The fact is that pollsters aren't necessarily bad just because they don't come up with your desired outcome. They aren't necessarily bad just because they don't show the same results the herd is showing. Oftentimes, a so-called "outlier" poll ends up being the one that nails the result.

And don't forget the margin of error. In Indiana, which I presume is the least popular result of the three, the MOE is +/- 3.7%. So even though it's Braun +3 on the surface, it's actually telling you that it considers plausible any result from Donnelly +4.4 to Braun +10.4. If, in the end, Donnelly ekes out a narrow victory, you cannot call this poll "wrong," because it anticipated that result within its range of likely outcomes. A poll is only truly wrong if the result ends up being outside its margin of error, like Donnelly +6.

In fact, people should stop obsessing over topline results and instead think of polls as presenting a range of possibilities. Had that been done in 2016, we would have seen that Hillary was not quite as inevitable as she appeared on the surface.

[DataGuy]

Are we really dissecting YouGov crosstabs guys

Yes, we should not find their flaws within their methology. We must indulge into our own confirmation bias! I want to believe.

Honestly, this is the only sense I can make out of a poll with 8 % undecided voters 9 days before a pivotal election. I hate this political spining loaded with false equivalencies and What-Aboutisms

It's poor practice to attempt to discredit polls based on their crosstabs. The crosstabs always have much larger margins of error than the poll in general and were never really meant to considered accurate in the first place. You might be correct in saying that a certain crosstab is wrong, but that really proves nothing about the poll overall because errors in the other crosstabs might counteract it and balance things out.

[DataGuy]

It's not really fair to immediately trash 538's model just because Braun has taken a small lead in a handful of recent polls. Data-driven models are intentionally cautious in reacting to new polling data. They are meant to have an element of stability based on other factors independent of polling, such as fundamentals (although it's fair to debate what should actually count as a fundamental).

A very clear trend over several credible polls is required to counteract those fundamentals. It's a feature, not a bug. If your probabilities swing wildly with each new poll, you don't really have a model. You have just a regular polling average, and a very noisy one at that. Even most ordinary polling averages collect polls over an extended period of time to minimize statistical noise.

YouGov is the first non-partisan pollster in a long time to show Braun ahead, so any expectation that the models will swing dramatically just of because of that is absolutely unrealistic and uninformed.