After the conventions, presumably, but how long after? I wasn’t paying enough attention last time until the last month or so before the election. How soon do they start this?
Also, I’m wondering if this year is weird enough to throw their algorithms off. As I understand their process, they rate all the polls that they use in their amalgamation based on previous accuracy, and then weight the results of those polls accordingly as they amalgamate them. It worked very well in 2012. Is it likely to be as accurate this time?
Pretty soon, if not during. Polling right now isn’t really too reliable but it gets better as we get closer to November. The conventions usually shake up the polls by quite a lot.
538’s model is also for two person races with not many undecideds. It’s never been tested when there are a lot of undecided voters, or a third party candidate taking up enough votes to make a difference in the outcome.
To cite one poll, what if it’s 39-29 Clinton? How do you make odds for something like that? That’s three times as many undecideds as her lead!
Then there’s Gravis, who put out a useless poll that showed Clinton 51-49 over Trump. Why would you put out a poll that forces people to choose between the two when most polls show 20%+ undecided or third party?
Last question first, by also allowing the option of “other” … which they did (but that Trump did not tweet) and which showed the same Clinton +5 that other RV polls have been showing (50 to 45).
Doubling back … not sure how you conclude that the 538 aggregation method only applies to two candidate races or is incapable of dealing with undecideds. Likely the method deals with that some by having an increased error bar in the predictions, even if that error bar never gets any attention by most. FWIW polls were not so far off on the shares for Nader or Perot (either time).
Right now any use of polls to make predictions has to have a huge error bar. Partly because of the huge number of undecideds, partly because no one knows whether or not either third party candidate will get any traction whatsoever, and mostly because polling right now is of only very limited predictive value for what the final result will be in any case… as I recall according to Wang a S.D. of 4 to 8 points depending on how far back you go. (And his state polling aggregation method, an excellent approach, has so little to use that it may be silly to even report it yet.)
I strongly expect that future LV polls, especially one that include the third party candidates, will have fewer “undecideds” than the current larger number of RV polls which have mostly been between the two choices alone. As the season goes on polling tends to move to those LV polls.
Another way to handle a large number of undecideds is to figure out how much they’d need to break by along with trying to figure out what fraction of them will actually come out to vote (one suspects undecideds are less likely voters than are decideds). Place that with some knowledge about their demographics and you can place some odds of what will happen. So imagine that 39/29 Clinton and call the “other” all as true undecided … 32% “undecided”. They’d need to all vote for one of the two and break for Trump 20 to 12 for Trump to pull ahead. What are the odds of that?
I think just that his method is not just any one poll or the same method of aggregating them that HuffPo uses. His is simply
HuffPo uses a tracking method that, for example, includes 11 polls for Ohio currently, and 10 for PA. How they weight them I don’t know. When polling is sparse and infrequent those methods may give quite different results.
The other bit is that Wang uses the median of those last three or of those done in the last week and I would be fairly sure that HuffPo’s model uses some variant of the mean. (And RCP does a rolling average.) He likes the median because it is outlier resistant. Probabilities are figured out by the estimated standard error of the median, converting that into z-scores. Thus last three Ohio polls are Clinton +6, Clinton +5, and Trump +4 (done from April to end of May) which makes the median Clinton +5, and last three PA are Clinton +1, Clinton +1 and Clinton +15, making median Clinton +1. Obviously the mean is different and a model that includes the last 11 polls going back to February in some formula weighting recency I am sure will give a different number.
Aha! Thank you. I’ve always assumed that PA and OH are very similar demographically, with PA always a bit bluer.
I did call PA as a key tipping state in earlier threads and was soundly ridiculed: “PA is lock Blue, not swing.”
It now appears that, to a close approximation, the election boils down to simply “Trump wins if and only if he gets 3 (or 4) of the four states FL, OH, PA, VA.” (Since FL and OH are redder than the other two, this can be further simplified to “Trump wins if and only if he gets PA or VA.”)
This is also a useful page to understand his approach, especially the difference between his “random drfit” and “Bayesian” probabilities, which I was unclear about.
As to the issue of tipping-point states the graphic on this page regarding 2012 may be useful.
Obviously the map could get “scrambled” with major shifts in demographic alliances … but the analyses so far fail to provide any evidence that there is any major such “scramble” or “realignment” in progress despite the atypicality that is Trump. That said every cycle brings some shifts.
PA could BTW both be a lock blue AND be the tipping-point state. The tipping-point may not be close to be swung in any particular election. I would not surprised if it does move into Colorado’s tipping-point place even if it is solidly Blue.
Do you have facts to back this up? Seriously, feel free to wait and do some research before you start typing.
Now, as for my wild ass guess : I’ve been amused by the lack of projections on 538. My guess is that Nate & Co. are shaking their heads at the sheer insanity of this race.
There’s also that Silver is not a pure polling based model and the farther away an election is the more they rely on “fundamentals” as part of their mix (the balance alluded to in this WaPo Monkey Cage article). Fundamentals though often veer into punditry as he admits to on Trump as linked above. Call them Bayesian priors if you want but there may not be many applicable priors to confidently add to the numbers. Wang is pure polls-based and just puts huge error bars up asa result. His Bayesian prior is not based on fundamentals but historic tendency to end up from where polling had been averaging nationally since January with an S.D. of 7%. And he reports with and without that Bayesian prior.
Silver right now is gun shy about incorporating any fundamentals and does not trust the numbers alone this early on, I think.
[QUOTE=GIGOBuster]
I told others in my circle that Trump was not going to make it because it was likely that a few candidates will drop of the race when the voting starts and the votes will then likely go to the less crazy candidates. But I wonder now. Finding that the current top three in the polls are from the crazy group makes me then wonder now if Trump has a shot to become the Republican candidate. (And still not likely to win in the general).
It looks to me that what would normally happen would be for a good number of the candidates to drop as soon their money ran out as it was one of the best yardsticks to figure out that there was not much support to continue anyhow. But this is not like previous elections, many candidates now have wealthy sugar daddies and groups that will make that yardstick not as limiting as before. With many big egos in the race and with money it is possible that we will see a lot of Republican primaries with no clear leader or with no “moderate” Republican candidate in the lead.
In past elections candidates found soon enough that they had no more support and then they closed shop; then they supported the moderate candidate that was more likely to win in the general election; but what would happen if the less reasonable or uncompromising the ones that are still leading? It is then possible that Trump would be the one to get the votes of Carson and Cruz once they finally drop of the race while the moderates remain in second and third dividing the vote as they also will see no clear reason why to drop of the race.
[/QUOTE]
Clear reason being that they had much more money in the recent primary to burn as other candidates had in previous races.
So yeah, back then I was having doubts that Trump was going to go down in the primaries, the polls were already too much in favor of the orange one. And I thought then that having so many candidates was not a sign of inclusivity for the Republicans, but a sign of weakness and division; however, all that money behind most candidates pointed to me then that that division was going to continue for a long time and Trump was going to benefit by that.
Seeing how Nate also noticed that the polls pointed to the trend and what took place in the primaries I base the idea that Clinton will win based on what was going on the matching polls and other factors. While Trump was winning the primaries, the same pollsters that showed Trump getting ahead by being more outrageous also showed at the same time that when matching Trump to Clinton or Sanders the Democrats lead was usually into double digits. (Now I know that early polling has been dismissed, but I do think that when there is a polarized electorate that the early trends are more robust than many are assuming nowadays)
Looking at the money factor, the latest news of Trump begging for money and demanding the GOP to set his presidential effort while Clinton is creaming him in the advertisement front in the battleground states right now tells me that Clinton has also that fundamental in her favor.
I also think that Trump dismissing the media that fed him 2 billion in free publicity does not help. IIUC not even Nixon banned press groups from covering him. So that free support will not be present as he had it in the primaries.
Then there are the electoral maps that continue to look dismal to Trump.
And finally for this post (I think that there are more factors, but these are the most important IMHO) I’m a Hispanic that will volunteer to make sure that Trump will eat his sorry idea that Hispanics will love him. When Trump is so deluded about the actual support he has in a group one should not expect that he will be reasonable about what he needs to do to win the support of that group in the general election.
: I’ve been amused by the lack of projections on 538. My guess is that Nate & Co. are shaking their heads at the sheer insanity of this race.