2016 Postmortem
Related: About this forumIt looks like Nate Silver is not factoring in the Florida early voting poll from yesterday
http://projects.fivethirtyeight.com/2016-election-forecast/?ex_cid=rrpromovdogg
(1,384 posts)geek tragedy
(68,868 posts)Neither does anyone, because we don't know who will show up on election day.
The 28% of Republicans voting for Hillary is one of those "too good to be true" things
still_one
(92,216 posts)affirms what you just said
That 28% of republican cross overs in those who have already voted has some credibility because within those 28% said they also voted for Rubio, which kind of collaborates with what one would think from a republican who couldn't vote for trump.
Whether the three million plus early voting patterns actually translate to those who vote on November 8th, I guess it remains to be seen
Cosmocat
(14,565 posts)I am skeptical of that, too.
The only thing that might make sense, and the talked about it on Lawrence O'Donnell, is that it possibly is Cuban voters. Trump has the pissed off royal and they mostly are republicans.
The guy they had on who was behind the poll was talking and while they did not break down the hispanic voters by nationality, he seemed to indicate this was a big part of it.
And, to lend just a little bit more possible credibility to the poll, while there is that astounding percentage of Rs who voted Hill, there is not near the same number who voted Murphy in the senate race.
That lends some plausibility to it - hate Trump, voting Hill, but staying behind Marco.
geek tragedy
(68,868 posts)poll."
Sample size for the survey was 311 early voters, of which about 127 were registered Republicans which means they got about 36 Republicans who said they voted Clinton.
36 respondents to a survey is not a story.
Cosmocat
(14,565 posts)but, yeah he was definitely real solicitous about it ...
GWC58
(2,678 posts)they like having a senator get paid for doing nothing. Yeah, okay. Makes perfect sense to me. 😝🤗
Dem2
(8,168 posts)geo1
(34 posts)In the meantime, take a look at the details
[link:https://www.scribd.com/document/329698329/TargetSmart-William-Mary-Florida-Poll-of-Early-and-Likely-Voters|
rumdude
(448 posts)Divine Discontent
(21,056 posts)I find it hard to believe over a quarter of FL GOP'ers would vote Hillary. It's a great wish, but not happening. At most, she might get 21-22, but more likely 14-15, and even that high will result in her winning. They only did a little over 300 Exit Poll interviews. So, that means about 100 were Republican voters, and of that 100, about 30 said they voted for Hillary. You can't base a lot off 30 voters. I am pleased with the result not being low, say just 10-12, but, I'm guessing that may be why it's a C-?
geo1
(34 posts)Just checked his current page of Florida polls
Farmgirl1961
(1,493 posts)has Hillary at 69.5 and Trump at 30.4, with an EV of 299.9. Where does he get his information from? I wonder what these numbers will look like come election day...and why his numbers are so far out there when everyone else's are relatively stable.
I see at Daily Kos, Hillary is down to 91%, but PEC has her trending around 98/99 with an EV of 317. What gives?
jcgoldie
(11,631 posts)You can look up his answers if you read the articles. Basically as I recall, he says there are two main reasons why his statistical model is more conservative regarding Clinton's win chances. First, he gives higher weight to the large number of undecideds that remain in most of the polling even at this late date. In previous elections up to 95% were decided in response to polls whereas this year many polls have both major candidates in the low 40s or even less meaning a 6 point lead this year is much less stable than in previous elections because the undecided vote is more substantial. Secondly, he has mentioned that he assigns a greater weight to the fact that various state polls seem to be linked. The more independently the states vote the more remote Trump's chances would be because his path to 270 is so daunting he needs so many things to go right. But Silver argues that everything going right for Trump (ie winning all the swing states which he needs to do) is not as unlikely as it would be if these were mostly independent events because in reality if somehow Trump pulls out a NH, PA, or CO chances are pretty good that won't be a remote event but rather part of a wave. I know all of the models take this into consideration, but I believe his model weights this possibility greater thereby lowering the overall certainty.
edit to add: I've also read him expressing greater doubt in the efficacy of the polls as a whole given recent international polling meltdowns related to Brexit and the FARC.
Farmgirl1961
(1,493 posts)So -- in the end, would they be deem more reliable or less reliable. I go from being totally optimistic to feeling more than a bit stressed.
I know that all that matters is GOTV...GOTV...GOTV... but still, it sure would be nice to have a better sense if 98% is realistic, 91% is realistic or 69.5% is realistic (hoping that this is way, way, way too conservative!).
I would have thought that with more people involved in early voting that these fluctuations would have stabilized, or started to swing back the other way.
jcgoldie
(11,631 posts)I have a hard time believing that Clinton's chances are 98% although I would love it to be true. To me a valid criticism of Silver's model this cycle seems to be that it fluctuates wildly in response to relatively few new polls... I suspect that has to do with the fact that he weighs polls from other states so heavily in determining the percentage of who wins any individual state, but it sort of defeats the purpose of a predictive model if its completely reactionary from one day to the next.
warpigs72
(31 posts)Look at the bottom of the page and you can see the trend line adjustment. Right now, Nationally it is 1.6% (1.7 - 0.1) for Trump. This is added to the current spread. So if a few polls come in higher than previous the trend line gets adjusted. For Florida it is also 1.7% for Trump. That's why the probability changes so quickly in 538 model compared to others. In other words, 538 increases the weighted average difference by this trend line. I'm not sure how accurate this is, but it appears to be large so close to the election. If you would take out the 1.7% trend the probability would be a lot difference.
tableturner
(1,683 posts)men and especially women, THAT is how 28% of the GOP votes for Hillary in Florida.