So, the 538 polls only got it half right.

[Shermann]

Their final prediction for Trump was 46.8%, and he got 50.4% (error of 3.6%)

Their final prediction for Harris was 48.0%, and she got 48.0% (error of 0.0%).

538 seemed to overestimate the impact of third-party candidates as it all went to Trump. They are a bit vague about their margin of error (frankly I don't think it is knowable), but you hear the 3% number bandied about a lot. Well, they missed that for Trump. But they nailed the Harris result.

[TwilightZone]

1. You're forgetting undecided voters.

Last-minute polls still included undecideds, so polls-to-results often isn't a direct match.

Assuming, for the sake of simplicity, a 50/50 split:

Trump: 46.8% + 2.1 = 48.9

Harris: 48.0% + 2.1 = 50.1.

1% total went to Stein and RFK Jr.

Both would be within your hypothetical 3% MOE.

There are still several million votes left to process in CA, so the percentages will change slightly when the results are 100%.

[Shermann]

3. Are you sure those predictions don't omit undecided voters?

I've gone over their methodology, and it says nothing about that. Undecided voters are basically not participating in the poll, so it would seem to be nonsensical to include them in the core predictions. It would make sense to incorporate that data into the margin of error though.

You seem to be retroactively amending the prediction based on that data.

[TwilightZone]

4. The numbers don't add up to 100%. The rest (other than third-party) has to be undecideds by default.

Most polls showed Stein at under 1% and RFK under 0.5%. The total of 1% they received is about what was forecast in the polls.

The rest has to be undecideds.

Math:

100 - 1 for Stein/RFK Jr. = 99.

99 - 48 (Harris) - 46.8 (Trump) = 4.2.

Using my simplistic assumption of 50/50;

4.2/2 = 2.1. Add to each side for 99% + 1% Stein/RFK = 100%

"Undecided voters are basically not participating in the poll"

Choosing "undecided" is participating in the poll. If undecided wasn't an option, every poll would be 99% for Trump/Harris because of the 1% for TPs.

If it's not undecideds, what is it? 4.2% is missing and it's not third-party voters.

[Shermann]

5. How do you fairly grade the poll accuracy after the election then?

Again, splitting the undecided vote 50/50 and amending the predictions seems sketchy.

[TwilightZone]

6. Assuming the 3% MOE applies, the poll averages were pretty spot-on.

Individual polls had more variation, of course, but many of the latest had it 50/49 one way or the other, which is right in the ballpark. Harris is running about a 3/2 split in California, so the gap will also get a bit smaller.

I split undecideds 50/50 for two reasons:

1) It's easy for math purposes
2) We won't know the real split until all votes are in and someone like Pew does some post-election analysis.

I suspect that the real number is higher than 50% for Trump because it seemed like he had momentum, but undecideds are unpredictable.

[Silent Type]

2. Polls were much closer than the "polls are not predictive" folks said. But not sure we could

have changed anything had we known.

[brush]

7. 538 needs to hang it up. Have they gotten anything right lately?

Same with the rest of the polling industry.

But if the fix is in (trump and Johnson's little secret), no polls can ever calculate anythhig.

[EdmondDantes_]

10. Well they were pretty much spot on with this election

And with the 2020 election. And their polls only average was on in 2022 (the "deluxe" model that included some additional information was off).

[brush]

11. Half right is right now?

[FBaggins]

13. They were much closer than half right

If they got something wrong, it was their decision to exclude some of the pollsters from their averages (despite those pollsters often being closest in 2016)... and those pollsters were closest to correct again this year.

But those wanting to score 538 as missing the results should look at their predictions again. The single most likely outcome (by a significant margin) on their model was Trump 312 to Harris 226 - which is precisely what happened.

[dsc]

8. A polling average would have a lower MOE than the polls which make it up

but it was fairly close in any case.

[Sympthsical]

9. We should wait for final results before wondering about national numbers/polls

California takes for-ev-er. As of this writing, there's still about 4 million votes laying around that have yet to be added to the final total. Considering Harris is winning them by about 20 points - with no reason to suspect that will change - we'll no doubt see some of that 3.6% shrink.

Mix in all the other states that are still counting, we don't yet know final percentages.

This is why I loathe the "Where are the 15 million votes?!" stuff that started right away as a springboard for a thousand conspiracy theories.

Some states count faster than others. Deadlines for mail-in votes vary. State laws have different windows for when all this stuff gets done.

[iemanja]

12. The polls again underestimated Trump's support.