Agree or disagree: no race should ever be rated anything beyond 'tilt'...

[Karpatsky]

...unless one candidate is regularly polling above 50%. Something I've noticed in a lot of races where the final result went against the poll aggregates (see e.g. IA-GOV, OH-GOV, PA-PRES, MI-PRES) is that the predicted winner would reliably score in the upper 40s and polls would be published as "Candidate +4" or so, without mentioning that they still do not have a winning coalition. This leads to a lot of 'upsets' when undecideds break the other way when really noone had any business predicting the race with any confidence one way or another.

Thoughts on this rule?

[mileslunn]

I think it is true if no candidate is over 50%, it should not be beyond tilt, but I don't think that is true across the board.  There is absolutely zero chance of Trump winning in California in 2020 and likewise zero chance of Democrats winning Alabama in 2020.  Now yes long shots can occur, such as Democrats winning Texas in 2020 or Trump picking up Colorado, but that is about the furthest I would go.  Probably the biggest surprise in recent memory is Obama winning Indiana in 2008 as that went for Bush by 20 points four years earlier.  But since it has swung back to solidly GOP looks like more a blip than trend whereas him winning Virginia that was a long term trend in shifting political allegiances which there were already strong signs of before 2008.

[Karpatsky]

Absolutely not. NM-Sen was clearly always Safe D even though Heinrich was polling sub-50 against a split field.

Fair enough, strong third parties mess up the formula, but in general - unless the candidate regularly polls high enough to guarantee a win regardless of where the undecideds go.

[mileslunn]

Absolutely not. NM-Sen was clearly always Safe D even though Heinrich was polling sub-50 against a split field.

Fair enough, strong third parties mess up the formula, but in general - unless the candidate regularly polls high enough to guarantee a win regardless of where the undecideds go.

Exactly.  Back in the summer of 2016, most polls showed Clinton in the upper 40s in Michigan and Pennsylvania while Trump only upper 30s.  Reason Trump won is most undecided as well as the Libertarians who were around 8% swung over to him while Clinton remained in the high 40s.  If consistently over 50% then pretty safe.

That being said in multi-party elections outside the US, being over 50% is not usually necessary, you can look at polls relative to historical data and make an educated guess based on that as actually outside the US (excluding countries with run off elections and only for final round) it is very rare for the winner to get over 50%.  US is the only G7 country on first round where the winner routinely tops the 50% (France does too but on a runoffs not first round while extremely rare in Canada, never in Germany or UK, sometimes for slates in Japan and Italy, but almost never for a single party).