General election polls: of value when?

Politics & Elections
1

Many have, when head to head polls against Trump were cited in support of candidate A or B, oft-repeated that general election head to head polls are not very useful too early on. True that.

But now we have a presumptive nominee and while the general is not right around the corner it starts to get close enough to make me wonder:

When DO general election head to head polls begin to have any predictive value? (Smart asses who want to say never can just not bother please.)

2

I’m assuming they’re of value when Nate Silver’s website has a general election model up and running.

3

I’d say after both conventions are over is a good starting point. An election year typically features party out of power convention, Olympics, party in power convention as summer draws to a close and Labor Day approaches. 2020 may be different as summer vacation season might not exist if there’s 25% unemployment.

4

According to this oft-quoted cite, the head-to-heads start having predictive value about 300 days before the general election. IOW, about the second week in January. By now, per the graph, they’ve got an r[sup]2[/sup] of about .5, which is of course way less than certainty, but still pretty good.

5

Thank you. That’s the article I was thinking of but unable to find!

So we are pretty much on a flat portion of the curve with a little correlation now that begins to gradually get better more beginning August on. Which makes sense.

Thanks again.

6

It might make sense mathematically…but , like the math of the stock market, past performance does not prove anything.

This year is unlike anything we have ever experienced in our lives. Right now, it’s still amusing…we trade stories about empty shelves at the store, and pretend that it will be temporary. …But in a few more months when we have Depression-level unemployment , people will be in a very different mood.

It would be interesting if the polls ask not only which candidate you support, but whether the pandemic affected your decision.

7

No, it’s not like that.

8

Here is a simplified look at the Electoral College. States are listed in order of D_vote minus R_vote from the 2016 election. The second percent shown is D+Green vote minus R+Libertarian vote.
various 205
-3.5% -6.3% AZ 11

-3.7% -6.1% NC 15
-1.2% -2.7% FL 29 *
-0.8% -3.3% WI 10 *
-0.7% -2.3% PA 20 *
-0.2% -2.8% MI 16 *
+0.4% -2.9% NH 4 *
+1.5% -1.1% MN 10
+3.0% -0.2% ME 2

+2.4% -0.9% NV 6
various 210
By assigning AZ to the Rs and NV to the Ds, we obtain an interesting symmetry: each side has 216 “almost certain” evs with 106 (in 8 states) undecided. (Maine has 4evs total but is shown with only 2 evs because they’ve voluntarily given up electoral power by awarding 1 each to D and R more-or-less automatically.)

Only five states (FL, WI, PA, MI, NH) are at all likely to be tipping states — only these states matter! If Trump wins Florida, the Democrat must sweep the (swingable) Rust Belt. If Biden wins Florida (unlikely?) he only needs to also get PA or MI or both WI and NH.