Posted on 08/01/2016 8:27:01 PM PDT by LS
I am not a statistician. I'm not a mathematician. But I can read. And while I want to say from the outset that we should NOT ignore polls or automatically dismiss those we don't like, we do need to be reasonable about which ones are possibly trustworthy and which are not.
For example, in the last three weeks there have been two polls from VA with Trump leading, one from Hampton U. with him at .5% and another from an outfit I never heard of with him up 5. I don' think either of these is reliable. The second group at the same time they had Trump leading in VA by 5 had him losing nationally by 15!
But there are several things recently that are extremely troubling about the polls I'm seeing. Let's review what a reliable poll would look like: 1) it would have a D/R/I split close to that of 2012, or somewhere in the 35 or 36 D, 31 or 30 R, and 28 or so I. 2) it would have a 51-49 or 52-48 F/M split. 3) it would reasonably approximate different age groups. 4) it would be a four way race (because that is what we have) 5) it would have a good sized sample (at least 500) 6) Here's the kicker: it would be "likely" voters. As we have seen in the past, there is a D loss of about 1-2 from "registered" to "likely" voters. But if the poll uses "adults," usually you can subtract another 2 from the R side.
Now, here's where I think major caution is in order: recently I have been reading the methodologies---say, for the CNN/ORC poll today that has Cankles up 11. They had a sample of over 1000 (good) but of "adults" (horrible) but then said "800 [or so] of them were registered voters."
So, what do we know about who actually is being polled? Not much. Did CNN poll 600 "registered voters" and 200 "adults?" Or 799 registered voters and 1 adult? It makes a huge difference. If the former, you have tainted your sample by 25% to the left.
This unscientific mixing of groups of voters---without stipulating who is actually being polled---means that the real result could be 2-4 points off before even addressing the D/R/I and M/F splits.
We know about the Reuters "reweighting" to achieve the results they want---that's bias as blatant and obvious as can be imagined. But even then, their splits are off, skewing their polls even further. However, there is something else going on that I have not ever seen in polling before---so please inform me if you have seen this.
The Suffolk PA poll showing Cankles up 9 and the ABC national poll showing her up 3 had an interesting phrase at the bottom of their methodology. It was that they asked for the "youngest person in the household" (presumably able to vote, because they then said the sample was "registered voters"). Now, think about that: merely asking for the youngest voter means that de facto one is more likely to get liberal Republicans and extremely liberal Democrats, whereas asking for the "oldest person" in the home would de facto give you a more conservative group of respondents.
One person, trying to defend this practice, said that 73% of this group were 35-50 year olds. But that's a massive bias still. Give me a sample of 100% 50 year olds and I'll show you a much, much more Republican-leaning group than a group of 35 year olds.
So, to reiterate: this is not to ignore polls. In 2012, while most were off, most were off in the direction of Romney. Obama won by just under 4 points, the RCP average was Obama by .7, GWU/Battleground, CNN, and Monmouth were all ties (meaning they missed by 4); IDB Tipp was Obama +1, Rasmussen was Romney +1, and Pew and ABC actually came the closest, with Obama +3. All of these used "likely voter samples".
Part of the problem was that Gallup allocated all undecided voters (3%) equally, when in fact they broke heavy for Obama because "he cares about people like me"). Likewise, Battleground figured Romney was winning indies by 15, but in the key battleground states, his biggest margin was 11 (NV), with WI and CO at 9, and most of the other key states saw indies go to Romney only by 2 points.
So what does all this mean? 1) We won't know anything at all until they begin using "likely voter" samples, and even then, they have to be realistic splits. 2) Any poll asking for the "youngest person in the home" should be tossed, as should all Reuters polls. 3) Even the closest pollsters are probably off, but not necessarily in the D direction as they were in 2012. Rather, we saw that they simply could not figure out where "undecided" and independent voters would go. The two questions Rush Limbaugh points to a great deal---"Cares for someone like me" and blaming Bush for the economy---clearly drove these indies and undecideds toward Obama.
FWIW, as you know, minor swings in 10 states can mean a massive electoral college shift. I think the polls nationally are undercounting Trump/Republicans (as was seen in the primaries), and are not properly yet assigning indies and undecideds. I do NOT put much stock in the "Wilder" effect for Trump, where people are afraid to openly say they will vote for him (although there is some evidence of this according to pollsters who use phone vs. live methods.) We'll see. When the die is cast, far fewer people will be comfortable pulling the lever for Cankles. She is just too known a product. But, hey, I could be wrong.
Massive undersampling of Republicans.
No wonder Hillary is up by double digits!
No one believes that including CNN but it still helps them sell an agenda.
Whether the poll is true or not doesn’t matter.
This is good reading I need to sit down with it later. Love what I read so far.
Lots of good materials
Well done LS
Insert “the only poll that matters” trope here.
But you know, it’s true.
DAMN THE TORPEDOES!
it will come down to the debates..if Trump kicks Hillary’s ass he becomes president..if it’s vise versa she wins..that simple..tonight on Hannity Trump sounded very presidential..intelligent..good message..he continues talking like that he can win
When a poll is cited as having a ±3% MOE, it means nothing more than that you can say with 95% confidence that the mean answer for the sample population falls within ±%3 of the entire population of that cohort -- IF the sampling mehtodology is truly random. That is all it means.
It has no mathematically rigorous predictive ability, despite its being labelled "scientific" because it is nothing more than an indication of how that cohort would vote if they were voting today. It is of little practical predictive ability because the cohorts: adults ≠ registered voters ≠ likely voters ≠ people who will vote on election day ≠ people who will vote on election day and honestly answer the question of how they voted.
This is why Gallup has dropped out of the presidential polling business: in order to be useful, your model must be able to turn your target population into the actual voting population, and there is no mathematically, statistically, or even psychologically sound methodology for doing that. "Likely" generally "seems" better, but even that varies with the quality of a pollster's special sauce. The degree to which "likely" is properly modelled to extrapolate to voters-on-the-first-Tuesday-of-November is entirely a crap shoot which no polling organization has mastered for more than two elections in a row.
Nice clear assessment. It would be interesting to track non-respondents for all these polls.
One other thought, this is an election where both candidates have significant negatives. Which polls or polling methods work best under such conditions.
Do the math. Just the women and minority vote puts Hillary close to 45%. Hopefully that’s her ceiling.
Remember also that increasing the size of the sample only decreases the MOE as the square root of the number sampled. Sampling 5000 adults only reduces the MOE from ±3 to ±1.34, but increases the cost by a factor of 5.
And, remember the biggest error is not in the MOE, which is well known, but in determining how to decide which of your respondents are going to actually vote.
“I think Trump will win if he backs off this Kahn guy...”
Kahn will be a distant, nearly forgotten memory, long before November 8th.
He won’t even be be a small factor in the vote, by that time.
Yeah probably some computer games they play to make it seem legit.
I thought the same thing. Sean asked him a question (I can’t remember what it was) and he changed the subject and got on message. He needs to do that more often.
Agree with your post, and would add that there’s WAY too much that can happen between now and November to put any faith in polls at all. Wikileaked emails, terrorist attacks, social media kerfuffles, etc. At best, they are a sign of trends.
he was very presidential..but he needs to do that more often..not just when he talks to a neutral party..he is very capable of being presidential..and the Trump I saw tonight can beat Clinton in a debate
I remember when McCain was up 7 points..looking like he would win the election..the second the stock market tanked that was it..game over..no one knows what will happen between now and November anything can happen and change everything
Official Pile On/Eating Macaca Thread
http://www.freerepublic.com/focus/f-chat/2125687/posts
Nov 5, 2008
"Two weeks ago, if I had told you that deep red precincts would turn out at record levels, you would have thought that John McCain would not only win, but that he might win by a comfortable margin.
Well, didn't happen.
The "turnout model" of politics is officially dead---the one that says if you get "your people" to the polls, you win. The trouble is, they have to really be "your people."
I went downtown to count votes at 7:00. Our whole team thought not only had we won Dayton, but OH, and probably---with turnout rates like that---indicated McCain would do very well. I can't tell you how excited they were based on the turnout. In Miamisburg, two very Republican precincts had voted out 75% by 5:00 in the evening.
We got a report from a Trotwood Republican precinct that had voted at an 80% clip. Darke County called in saying they were at near-record 80%.
Finally---and I know I don't have credibility here (Freeper Captain Kirk was dead on)---the national numbers suggest that there was a "reverse Bradley," a guilt vote. This could account for some of the R defections.
I'm eating macaca. So pile on.
But you can trust me on this: we all have a lot more to worry about now than who was the best political prognosticator. God save our once-great nation."
I also think that there is a real possibility that there are a lot of Trump supporters that are unwilling to tell a pollster hiw they really feel. There has been so much negative press and so much shaming test people don’t want to answer.
And people have the impression that the NSA is monitoring all phone calls.....resulting in something like this....
Obamacare Navigator: ..."your voting record does nor conform to standards"
..."and your answers to polling questions prior to the last election are way out of line..."
I hope to God that Trump is aggressively courting the black vote. Trump has to be able to convince them that all those years of voting Democrat has brought them nothing but misery.
If there ever was a time for a GOP candidate to get a larger than normal share of the black vote, this would be the time. If Trump can get 15% to 20% of the black vote, this election is over as Trump would win big.
Go get those votes Trump!
Kahn is rapidly becoming a liability to Killary as the hits on him keep on coming. He is getting torn apart.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.