Regarding the forces that influence the polls, isn't it a bit odd that all of these sensational, unsubstantiated tabloid allegations against DT are surfacing at the eleventh hour of the campaign? The main focus now is clearly NOT on any of the urgent, substantive issues (which the average, lazy-minded voter would be hard-pressed to seriously consider anyway), but on what DT might have said or done to some hapless, innocent gal decades ago!
HRC said during the last debate, "when they go low, we go high", alluding obviously to her supposed "moral superiority" in the campaign. No, I'm not a "conspiracy theory" type person. But having said that, responsible voters should at least step back and consider the timing, source, motivation and credibility of any such "news" or "information" that's being fed to them at the eleventh hour. Probably too much to ask, given the nature of the electorate.
As the old adage goes, “We see what we WANT to see, and believe what we WANT to believe”. Sad...very sad, indeed!
HRC said during the last debate, "when they go low, we go high", alluding obviously to her supposed "moral superiority" in the campaign. No, I'm not a "conspiracy theory" type person. But having said that, responsible voters should at least step back and consider the timing, source, motivation and credibility of any such "news" or "information" that's being fed to them at the eleventh hour. Probably too much to ask, given the nature of the electorate.
As the old adage goes, “We see what we WANT to see, and believe what we WANT to believe”. Sad...very sad, indeed!
Protest voting can be used for one of two things. One, get a third party into office...two, force the winning contender to recognize the force of the third party voters and what their desires are. Now, there are a limited number of voters. In the case of only two contenders, votes go towards one or other while at the end, the one with the most votes wins. In the case of the third party vote, the vote goes towards the third party but then removes the vote from one of the two other contenders. Here is where protest votes become a problem with those who believe the conscientious vote was an imperative. It leaves open the likelihood that the contestant who wins is actually the contestant that you disliked the most. You have removed a vote from the actual political contest that would support either one or the other primary contenders that are competing.
Rather than protest this way, it is best to vote an allied contender into office and stay involved in the political process, keep who allies best with your political perceptions on their toes. In the case of Hillary, Bernie Sander's has already done that. Hillary who has historically been a bit more moderate...is moving more left. Protest voting has already been done through the primaries and has affected the outcome of the political base. The primaries are done, now vote the allied politician into office. If Bernie had won, Hillary voters would have needed to do the same. NO Trump.
Rather than protest this way, it is best to vote an allied contender into office and stay involved in the political process, keep who allies best with your political perceptions on their toes. In the case of Hillary, Bernie Sander's has already done that. Hillary who has historically been a bit more moderate...is moving more left. Protest voting has already been done through the primaries and has affected the outcome of the political base. The primaries are done, now vote the allied politician into office. If Bernie had won, Hillary voters would have needed to do the same. NO Trump.
1
To collect information, which is questionably valid is one thing...but why post in the news stream? Here is why polls are injected into the news stream...they cause drama, they effect emotions. Sure they have cherry picked facts, but at the same time they lack truth and that is where the problem lies, information without proper emotions leaves people open to manipulation....in simpler terms, they are specifically being used to manipulate the masses into doing something.
As the recent release of polls from various sources has shown, polls can be structured and reported to pretty much say whatever the pollsters want them to report. The most recent polls have been influential in that they are getting the GOP to push for the repeal the 19th amendment. Past polls had gotten the conservatives and teaparty to believe that they were winning the presidential race...not just winning but overwhelmingly winning...but they lost by tragic numbers. The interesting thing about that is that it was conservative based polls that were telling the conservatives they were winning.
As the recent release of polls from various sources has shown, polls can be structured and reported to pretty much say whatever the pollsters want them to report. The most recent polls have been influential in that they are getting the GOP to push for the repeal the 19th amendment. Past polls had gotten the conservatives and teaparty to believe that they were winning the presidential race...not just winning but overwhelmingly winning...but they lost by tragic numbers. The interesting thing about that is that it was conservative based polls that were telling the conservatives they were winning.
1
Like other commenters here, I never answer my landline if my CID shows "unknown" or "private" or a number that I don't recognize. As soon as the caller gets my answering machine requesting name, number, and that they leave a short message, they always hang up. Over the last 12 months I have had 10 to 20 calls a day. How do pollsters allow for people like me?
3
Much as I liked this article, the *best* part is Case Jernigan's artwork! Let's hear it for America's Likely Voters!
1
To avoid robo-calls by spammers I never answer my landline or cellphone unless it's a known number or the caller leaves a message. Everyone I know acts the same way. How do pollsters account for people like me?
4
I am an admitted "Poll Junkie". You would have to pry the 538 app from my cold, dead iPhone. I actually make microwave popcorn before I log in to the Thursday morning polls.
I pore through the aggregate polls on various sites and the NYTs has been driving me crazy with Rasmussen! Every week they release a poll with Trump up by at least one percent and often quite a bit more. Isn't it time to drop such an obvious outlier?
I pore through the aggregate polls on various sites and the NYTs has been driving me crazy with Rasmussen! Every week they release a poll with Trump up by at least one percent and often quite a bit more. Isn't it time to drop such an obvious outlier?
3
Any way you slice the polls "The Donald" is headed for a resounding shellacking in a few weeks. We're looking forward to his concession speech, after which we can start referring to him as "The Loser." #MAGA
Given that in fact four candidates are on the ballot in all or nearly all states, two way results are purely hypothetical. They may be of academic interest, but don't tell us who is likely to win.
Many have advocated "instant runoff" voting so that voters who favor a minor candidate don't have to profess endorsement of one of the two major candidates. This would take a long time to implement at the voting booth, but could easily be implemented as an additional question, in place of the two-way question, in the polls as early as next week. The results would be very informative, particularly during the primary season. Those who want the answer to the hypoethetical two-way question could easily infer it from the instant runoff data.
Many have advocated "instant runoff" voting so that voters who favor a minor candidate don't have to profess endorsement of one of the two major candidates. This would take a long time to implement at the voting booth, but could easily be implemented as an additional question, in place of the two-way question, in the polls as early as next week. The results would be very informative, particularly during the primary season. Those who want the answer to the hypoethetical two-way question could easily infer it from the instant runoff data.
The polls are not going to reflect the outcome until intentionally adjusted a week before the election to reflect that its propagators weren't manipulating the electorate all along, which of course was what they were doing. Two weeks after Trump is elected pollsters will turn - the chances of a war with Russia - and this junk science piece, like the one about two women who once met Trump, will be relegated to the 'related articles' list.
3
Right and Romney will win in a landslide.
I hope you'll be here on November 9th.
I hope you'll be here on November 9th.
6
Air Marshal of Blowviana speaks for all those relegated to the couch pre-Romney. Wrong about nearly everything and designed to provide Trump with the false reality field that was used to trick him into declaring and outing the GOP for what it really is, the anti-woman party, once and for all.
4
I couldn’t help but notice, perhaps because I’m a woman, that there are very few women represented in the whimsical illustrations accompanying this article.
In the first one, out of 12 characters (including mega masculine icon Superman, a personal fave of Mr. Trump) only 1 female is represented. 1 out of 12!
Where is Wonder Woman? Her DC teammates are present. Don’t you think she might be interested in the first presidential election with a major party female candidate? (And if a Kryptonian can vote, surely an Amazon is eligible.)
The illustration showing ‘likely’ and ‘unlikely’ voters is also skewed male. Out of 14 ‘likely voters’, at most 5 are female. And in the lower group of 12 to 17 ‘unlikely’ voters, it looks like only 2 are women.
I’m sure many people think this is trivial. But when Trump supporters are tweeting #repealthe19th because the majority of women poll respondents support Hillary, I feel it’s time to draw attention to the common ways women are dismissed and underrepresented in the media.
There are more women than men in the US and women typically vote more than men do. In 2012, 53 percent of voters were women. But since it appears that Mr. Jernigan is one those artists who ‘doesn’t do’ female characters, the subtext of this article is that the average voter is male, and the few annoying women who want a say in our political process should stay where they belong, behind the cool monsters and superhero men, at the back of the line.
In the first one, out of 12 characters (including mega masculine icon Superman, a personal fave of Mr. Trump) only 1 female is represented. 1 out of 12!
Where is Wonder Woman? Her DC teammates are present. Don’t you think she might be interested in the first presidential election with a major party female candidate? (And if a Kryptonian can vote, surely an Amazon is eligible.)
The illustration showing ‘likely’ and ‘unlikely’ voters is also skewed male. Out of 14 ‘likely voters’, at most 5 are female. And in the lower group of 12 to 17 ‘unlikely’ voters, it looks like only 2 are women.
I’m sure many people think this is trivial. But when Trump supporters are tweeting #repealthe19th because the majority of women poll respondents support Hillary, I feel it’s time to draw attention to the common ways women are dismissed and underrepresented in the media.
There are more women than men in the US and women typically vote more than men do. In 2012, 53 percent of voters were women. But since it appears that Mr. Jernigan is one those artists who ‘doesn’t do’ female characters, the subtext of this article is that the average voter is male, and the few annoying women who want a say in our political process should stay where they belong, behind the cool monsters and superhero men, at the back of the line.
12
This election is all about women. Nate Silver reveals it but fails to adequately understand the depth of this issue. That is why he joined much of the media in cynically expecting the Boys on the Bus to blow over instead of explode. The story that Trump is a pig has been around for decades. Why should a hot mic make it explode? The reasons are complex, but, simple at their core. He has been abusing women for 50 years. He is a psychopathic liar, who lies knowingly and uncaringly. His fury is excited if his lie isn't going over and he goes further and further into aggression until it is. He has been abusing a woman a day for have a century. At 365 days a year, at least 18,500 women have been personally scarred by him in some way. He challenged all of them to come forward and risk his abuse, and the abuse of his Twitter mob. So far several have with stories that vet back to the time of their telling through third parties. No surprise. He's over. He's a meat peddler. His family has a history in the skin trade going back to his draft-dodging great-grandfather. The obvious is rendered abstract and obscure by polls and statistics and men. This is visceral. Poll that.
9
Voters who want to see HRC sworn in as our 45th president need to be very afraid of the trump team and stay vigilant.
Trump is very dangerous and has Brietbart on his team. They will stop at nothing to upset this election and throw it into chaos if they can.
Help others get to to polls. Vote early when possible.
Trump is very dangerous and has Brietbart on his team. They will stop at nothing to upset this election and throw it into chaos if they can.
Help others get to to polls. Vote early when possible.
13
Every poll that fails to enumerate the number of individuals who refuse to participate is not worth considering.
2
A few observations based on my analysis of the four-way polls from the Huffington Post general election polling archive from May 1st through today.
IVR polling: There are only two organizations that use this method, the Republican-oriented Rasmussen Reports and the Democratic-oriented PPP. Rasmussen's polls have averaged some 3.5 percentage points more Republican than the consensus (p<0.001), while PPP polls do not diverge significantly (p<0.3). Rather than throwing out the method entirely, it seems more likely that Rasmussen's polls have a Republican bias, as they did in 2012 (http://www.politicsbythenumbers.org/2012/11/05/trends-pollsters-and-meth....
Clinton has actually done about one point better in likely voter polls this season though the result is only marginally significant (p<0.10) .
Online polling shows no significant difference from polls conducted over the phone by live interviewers. In 2012, Obama's results in online polls ran about one point higher than the consensus. However the consensus underestimated his margin of victory over Romney, so the Internet polling might have been more accurate four years ago. This year they do not diverge at all.
The two conventions had substantial effects on the margin between the candidates. Trump's support grew about five points after the RNC; Clinton saw her support increase by a bit under four points after the DNC.
The AccessHollywood tape appears to have moved the polls some four points in Clinton's favor.
IVR polling: There are only two organizations that use this method, the Republican-oriented Rasmussen Reports and the Democratic-oriented PPP. Rasmussen's polls have averaged some 3.5 percentage points more Republican than the consensus (p<0.001), while PPP polls do not diverge significantly (p<0.3). Rather than throwing out the method entirely, it seems more likely that Rasmussen's polls have a Republican bias, as they did in 2012 (http://www.politicsbythenumbers.org/2012/11/05/trends-pollsters-and-meth....
Clinton has actually done about one point better in likely voter polls this season though the result is only marginally significant (p<0.10) .
Online polling shows no significant difference from polls conducted over the phone by live interviewers. In 2012, Obama's results in online polls ran about one point higher than the consensus. However the consensus underestimated his margin of victory over Romney, so the Internet polling might have been more accurate four years ago. This year they do not diverge at all.
The two conventions had substantial effects on the margin between the candidates. Trump's support grew about five points after the RNC; Clinton saw her support increase by a bit under four points after the DNC.
The AccessHollywood tape appears to have moved the polls some four points in Clinton's favor.
4
I hope that election junkie and iconic expatriate computer scientist Andrew S. Tanenbaum won't mind me culturally assimilating his famous quote and applying it to polling:
"The nice thing about polling standards is that there are so many to choose from; furthermore, if you do not like any of the results, you can just wait for next pollster's model or results."
"The nice thing about polling standards is that there are so many to choose from; furthermore, if you do not like any of the results, you can just wait for next pollster's model or results."
8
Really interesting summary of all the aspects of polls to be taken account of before drawing any conclusions. But the list is so daunting it seems unlikely that anyone who doesn't analyze polls for a living could actually evaluate them well. A good reason to look at the Upshot's own summary (which I do daily),
6
Terrific column. It would be helpful if you could name some of those polls that you have more confidence in and those that you give less credence to. And do you weight your results knowing this information?
5
Nate Silver publishes a "report card" on all the polls at http://projects.fivethirtyeight.com/pollster-ratings/.
3
I would like to compliment Case Jernigan's beautiful graphics for this article. They are humorous and yet represent the subject matter really effectively. Although I know little about design or visual art, it's easy to see that this is really good work by Jernigan, whoever he or she is.
20
Taking that all into account, it doesn’t sound like the results of different polls ought to be averaged.
1
Thank you. That was truly informative. I'll never go that deeply into an analysis of a poll, but I will now be even more confident in the analyses that you provide.
2
Only 28 more days (day day after election) till the pollsters start explaining how and why their polls were right or wrong.
I'm trying to hibernate till this too long nightmare is over. Haven't picked up the phone till I know who is calling for way too long.
I'm trying to hibernate till this too long nightmare is over. Haven't picked up the phone till I know who is calling for way too long.
7
The phone thing is huge, to use the word of this election cycle. Land llnes are more likely to be the phone choice of older voters, who are more likely to be republican.. This is how Gallup went so wrong in 2012.
2
No serious poll relies only on landlines any more. Every live phone poll includes a mix of landlines and cell phones. The automated pollsters ("IVR") draw their samples from lists of registered voters to avoid the FCC's restriction on randomly calling cell phones. PPP explains their method here: http://www.publicpolicypolling.com/aboutPPP/about-us.html.
1
Thanks. I check RCP's poll average weekly. Nice to know it's not a bad metric.
Now, how would you like to explain the betting odds?
Now, how would you like to explain the betting odds?
3
Hear, hear!! I, too, want to understand the odds on the betting-markets where people (apparently legally) gamble on the predicted outcomes. Are these historically accurate, and if so, how closely compared to polls and to election results, do they correlate??
Good advice. The truth is that it is actually very difficult for laypeople to get an accurate sense of the reliability of any individual poll. I think that is especially true this year, when nothing makes sense and no one really knows what is going on. That's why I stick to FiveThirtyEight and the NYT's own election forecast.
2
I would add that national polls may be good indicators of national sentiment, but presidential candidates are elected by states, so the Upshot's tracking of state polls and electoral college results is the only polling that I scrutinize carefully.
4
Polling is the confluence of statistical analysis, psychology, communications, and interpretive skills. The questions asked, how they're asked, and myriad externalities influence responses. Creating a most appropriate, representative sample from a target population is itself very challenging. Interpretation of collected data introduces more subjectivity. All of these moving objects and the perceived implications of the data summary also have high perishability, with outcomes changing within moments of initial response collection...simply re-asking the poll questions immediately after the first collection will render a different set of answers across the sample.
Long story short, poll results should be only presented as data points and should not be confused with actual information.
Long story short, poll results should be only presented as data points and should not be confused with actual information.
2
I always smile a bit when people criticize polls -- or pols who base their responses upon polls, rather than "doing what's right" or "voting their conscience." (eg, the red-state GOP or male politicians who are less critical of Trump's remarks than the purple-state or female leaders). After all, we live in a democracy. We expect leaders to listen to their constituents and respond to them. One of the ways of doing that is looking at polls and responding to the views of constituents -- who might be more "deplorable" than The Donald. If we lived in a dictatorship (or a leadership completely bought-and-sold by the elite), polls and constituents wouldn't matter, would they?
The illustrations for this article are whimsical charming. Thank you, Case Jernigan!
3
In other words, you prefer to crowdsource the polls. It's not the same principle, but it reminds me that wagering sites, where the odds are set by those who are financially invested, have earned considerable credibility in the prediction market.
1. Know the better and more accurate poll results. Five years ago, I automatically ignored a couple due to inaccurate predictions.
2. Yes, immediately after say a debate, CNN puts out poll results and they're online so people can vote multiple times. I see the results and have to wait until calling polls come out the next day or so.
2. Yes, immediately after say a debate, CNN puts out poll results and they're online so people can vote multiple times. I see the results and have to wait until calling polls come out the next day or so.
No, the CNN post-debate polls use live phone calling to a sample of people recruited before the debate. The recruitment method probably makes the respondents more attentive, and thus less representative, of the overall public. Also one has to wonder about the comparative refusal rates among supporters for the two candidates. If, say, Trump supporters thought their man didn't fare well, they be less willing to answer the phone when the pollster calls which would bias the result in favor of Clinton.
CNN uses a "representative sample," meaning that those who agree to take part agree to pick up the phone when called.
Agreeing in advance is no guarantee that they will answer the phone when called.
1
This article contains wise advice to the reader, but also journalists. Too often the topping result of the latest poll is trumpeted by the media, often the sponsor of the poll, without any qualifier whatsoever. Iif the representative sample is correct, why even correct for party affiliation? It seems from this report that the past is not necessary prologue for this election. That would put a huge qualifier on al,it's all poll analysis.
The biggest mistake I see among commentators and prognosticators is that they do not understand how the margin of error ("MOE") works.
A poll that shows an MOE of 3.5% does NOT mean that the two candidates are within 3.5 percentage points of the reported results. It means that they COULD BE as much as 7 percentage points from the reported results.
MOE is determined individually for each candidate. So, a poll with an MOE of 3.5% showing Secretary Clinton had by 4 points 48% to Mr. Trump's 44% means her actual results could be as high as 51.5% (48%+3.5%) and that Mr. Trump could be as low as 40.5% (44%-40.5%).
Alternatively, by reversing the same calculation, the actual results could be as narrow as 0.5%, with Clinton having 44.5% and Trump having 47.5%
Pew has a good write-up explaining MOE, here:
http://www.pewresearch.org/fact-tank/2016/09/08/understanding-the-margin...
A poll that shows an MOE of 3.5% does NOT mean that the two candidates are within 3.5 percentage points of the reported results. It means that they COULD BE as much as 7 percentage points from the reported results.
MOE is determined individually for each candidate. So, a poll with an MOE of 3.5% showing Secretary Clinton had by 4 points 48% to Mr. Trump's 44% means her actual results could be as high as 51.5% (48%+3.5%) and that Mr. Trump could be as low as 40.5% (44%-40.5%).
Alternatively, by reversing the same calculation, the actual results could be as narrow as 0.5%, with Clinton having 44.5% and Trump having 47.5%
Pew has a good write-up explaining MOE, here:
http://www.pewresearch.org/fact-tank/2016/09/08/understanding-the-margin...
4
My regrets. The above has typos (never comment from a smartphone if you wear bifocals.) The Pew link will explain the point of MOE I was hoping to make.
In addition the errors on each candidate's support are inversely correlated. If a poll overestimates candidate A's support, it's likely to underestimate candidate B's support, since they have to add to one hundreds (including undecideds and others). So the error on the lead is even larger than simply doubling the MOE on the individual candidate estimates.
Great post!
Well thanks for the breakdown, Mr. Cohn. Most of the time though, I can't be bothered to go through all this and I'm quite happy to have the Nates - Nate Cohen and Nate Silver of 538 do all that work for me so I can just look at the weighted averages.
9
With so many polls, how come I (or anyone I know) have never been polled?
2
You don't know me but we've received the call. We've seen plenty of elections and never participated in a poll. But, we didn't answer the phone especially if there was no ID. In this election, we're getting spammed more but we do answer the phone. It makes a difference.
1
Just keep it simple:
Go to RealClearPolitics.com and read the polls and stats in the Upper Right Hand Corner of the Home Page.
Takes about 90 seconds and you are done for the day.
Even simpler:
Go NYTimes Daily Forecast and read the % for the candidates.
Takes 15 seconds.
Go to RealClearPolitics.com and read the polls and stats in the Upper Right Hand Corner of the Home Page.
Takes about 90 seconds and you are done for the day.
Even simpler:
Go NYTimes Daily Forecast and read the % for the candidates.
Takes 15 seconds.
How about this: look at the candidates and vote your conscience. I am sick and tired of polls, talking about polls, cable news networks that feed off polls, politicians that only care about polls. This is why we have the situation we have, too much information based on irrelevant surveys of people. Inform the public with facts, leave your opinions and polls to yourselves..the end.
3
Polls are useless. They are used to justify the exorbitant sums politicians pay the polling companies. In the 1970s, 90% of people answered polls. Today, it is 10%. Young people don't have landlines. Mostly older people do. Many don't answer them because of telemarketing and scams. Young people block cellphone numbers they don't recognize.
So if you're spending days trying to get someone to answer his/her phone, it's most likely an older person who answers a landline or cell phone because he or she is retired. That means older white people. That's why it appears the election is close---it isn't. After Trump's recording, women, and Christians are running the other way. You'll never find that out with polls.
So if you're spending days trying to get someone to answer his/her phone, it's most likely an older person who answers a landline or cell phone because he or she is retired. That means older white people. That's why it appears the election is close---it isn't. After Trump's recording, women, and Christians are running the other way. You'll never find that out with polls.
2
You can do all that work, or you can go to fivethirtyeight.com and check out Nate Silver's projections. His projections are based on sophisticated computer models that take account of each polling firm's methodology and record for accuracy.
Nate has called the last two presidential elections on the nose, and not just the national results. Silver has correctly called each state's presidential and senate votes slightly more than 95 percent of the time.
politicsbyeccehomo.wordpress.com
Nate has called the last two presidential elections on the nose, and not just the national results. Silver has correctly called each state's presidential and senate votes slightly more than 95 percent of the time.
politicsbyeccehomo.wordpress.com
6
Fine discussion, Nate. You show the role that devilish uncertainty plays in statistical models. The margins of error that pollsters come up with should be tied to some model of the uncertainty. I think your dynamic tree charts of the possible election outcomes are a great way to visualize the effect of uncertainty, given the uncertainty in the key states, by selecting various initial conditions.
The outcome tree also helps handle intuition. Probably most of us, including a fair number of Republicans, feel that Clinton is going to win this one, given the inordinate personality of Donald Trump. But Clinton herself has a poor popularity rating that deters votes from going to her. Thus its up to a particular segment of the undecided population, those who are courageous enough to venture a vote on November 8. Who knows, turnout could be unusually low, even by normal election standards. So the outcome tree lets us plug and play outcomes, given the uncertainty.
Another use of polling is to modulate the feeling that we have about the results. Saul Kripke once gave an argument by saying, "I feel it in my bones." Maybe only Kripke could get away with that; we feel in our bones that this election will swing a certain way, but that intuition needs to be backed by measurement.
For example, Florida is odd to me for being tied in the polls. It's difficult for me to imagine Trump's winning Florida. But he could win, given this unpredictable outcome.
The outcome tree also helps handle intuition. Probably most of us, including a fair number of Republicans, feel that Clinton is going to win this one, given the inordinate personality of Donald Trump. But Clinton herself has a poor popularity rating that deters votes from going to her. Thus its up to a particular segment of the undecided population, those who are courageous enough to venture a vote on November 8. Who knows, turnout could be unusually low, even by normal election standards. So the outcome tree lets us plug and play outcomes, given the uncertainty.
Another use of polling is to modulate the feeling that we have about the results. Saul Kripke once gave an argument by saying, "I feel it in my bones." Maybe only Kripke could get away with that; we feel in our bones that this election will swing a certain way, but that intuition needs to be backed by measurement.
For example, Florida is odd to me for being tied in the polls. It's difficult for me to imagine Trump's winning Florida. But he could win, given this unpredictable outcome.
Republicans are not scientists.
Polling uses scientific analysis.
Therefore Republicans can not understand or concur with polls.
They also do not believe that giving women access to affordable birth control
will decrease the rate of abortions.
They believe that allowing multi-national corporations unlimited access to tax havens and loop-holes will increase jobs.
They do not believe the survey of the scientific community that global warming is here and is worsened by man-made activities.
If you can't believe in facts, it is difficult to discern the relevance of projections.
Polling uses scientific analysis.
Therefore Republicans can not understand or concur with polls.
They also do not believe that giving women access to affordable birth control
will decrease the rate of abortions.
They believe that allowing multi-national corporations unlimited access to tax havens and loop-holes will increase jobs.
They do not believe the survey of the scientific community that global warming is here and is worsened by man-made activities.
If you can't believe in facts, it is difficult to discern the relevance of projections.
19
In national elections we know that less than 60% of eligible voters actually vote. How many of the REGISTERED voters actually vote?
1
Don't believe IVR polls. People hate robocalls for a reason. Last night during dinner I was a 98 yo woman who had early voted for Gary Johnson. Interrupt my dinner again and who knows what answers you'll get. The only poll that counts is November 8th ( or November 28th if you're a Trump supporter in Florida).
Good rationalizations. For many of us, as you mention, both Rubio and McCain's recent primaries, if I have it right, the polls were wrong. Then for many of us who have interest in world affairs, as relates to the Clintons forthcoming foreign policy, the Brexit vote surprised all the ruling class in Britain, especially the western Central Bankers.
The savvy person ignores the polls
1
Dear Mr. Cohn: Thank you for your information. Now a question: in RealClear Politics, L.A. Time's polls are the only which every day shows Mr. Trump winning the election. Is it possible to know why?
3
LA Times is a tracking poll which uses poll data over time. That poll has shown Trump's tracking lead has gone down consistently over the last two weeks.
The NY Times did an interesting article on the LA Times poll:
http://www.nytimes.com/2016/10/13/upshot/how-one-19-year-old-illinois-ma...
The LA Times responded: http://www.latimes.com/politics/la-na-pol-daybreak-poll-questions-201610...
http://www.nytimes.com/2016/10/13/upshot/how-one-19-year-old-illinois-ma...
The LA Times responded: http://www.latimes.com/politics/la-na-pol-daybreak-poll-questions-201610...
POLL history on the difference between Republican and Democrat governing. Google, then choose.
PoliticsThatWork.com Change in Unemployment Rate by Party of President- Since 1945
Each party has held the presidency for the same number of years since 1945. During those years, the unemployment rate has risen 11.8% under Republican presidents and has fallen 7.2% under Democratic presidents. Unemployment has fallen during the overwhelming majority of Democratic years since 1949. Unemployment rose steadily under Republicans up until 1982, then fell during the remaining Reagan years, and then rose again under both Bush Presidents.
PoliticsThatWork.com Dow Jones Performance by the Party of the President
During the most recent 15 years during which Republicans have held the presidency, the value of the Dow has increased by 42%. During the Democratic presidencies, it has increased by 609%- 14.5 times faster. The average growth in the value of the Dow under Democrats during this period has been 14.75% and under Republicans it has been 5.11%.
PoliticsThatWork.com Change in Disposable Income Since 1930 by the Party of the President
In the 44 years that we have had Democratic presidents since 1930, the real per-capita disposable income has increased 271%. During the 40 years during which we have had Republican presidents, it has increased 44%. On average, it has increased 3.1% (after adjusting for inflation) under Democratic presidents and 1% under Republican presidents.
PoliticsThatWork.com Change in Unemployment Rate by Party of President- Since 1945
Each party has held the presidency for the same number of years since 1945. During those years, the unemployment rate has risen 11.8% under Republican presidents and has fallen 7.2% under Democratic presidents. Unemployment has fallen during the overwhelming majority of Democratic years since 1949. Unemployment rose steadily under Republicans up until 1982, then fell during the remaining Reagan years, and then rose again under both Bush Presidents.
PoliticsThatWork.com Dow Jones Performance by the Party of the President
During the most recent 15 years during which Republicans have held the presidency, the value of the Dow has increased by 42%. During the Democratic presidencies, it has increased by 609%- 14.5 times faster. The average growth in the value of the Dow under Democrats during this period has been 14.75% and under Republicans it has been 5.11%.
PoliticsThatWork.com Change in Disposable Income Since 1930 by the Party of the President
In the 44 years that we have had Democratic presidents since 1930, the real per-capita disposable income has increased 271%. During the 40 years during which we have had Republican presidents, it has increased 44%. On average, it has increased 3.1% (after adjusting for inflation) under Democratic presidents and 1% under Republican presidents.
18
Good review. As someone who used statistics throughout my career, this article is better than most. However, the mathematical illiteracy of the general public is legendary and I suspect much of this story went over the heads of the public. The most recent egregious example was the so called polling that happened the day after the debates. Trump supporters claimed victory from the on line polls, most of which had the statistical validity of an anecdote and kept insisting that on line polls were valid.
6
Even some serious professional polling organizations sometimes act in puzzling ways. A CNN/ORC poll in early September got a lot of attention because it purported to show a movement in Trump's direction. However they chose to conduct the interviewing over the Labor Day weekend when many people were away on vacation. I have the same issue with the NBC/WSJ poll conducted last weekend, though I suspect fewer people are away from home over Columbus Day weekend. When I was doing polling I would never have conducted a poll over a holiday weekend because we simply don't know which people might or might not be unavailable at that time.
Excellent graphics!
12
Never mind the polls. A savvy person knows that in the end, on election day, most Americans will realize that Trump can never be in the White House.
3
I think a savvy person acts as if Trump could very well end up in the White House and does everything possible up to and including election day to make sure that he doesn't. This election could depend on turnout. So be afraid. Very afraid. Fear is an effective motivator. "Knowing" what most Americans will do is a recipe for complacency ... and reduced voter turnout.
49
"...most Americans will realize that Trump can never be in the White House."
And that is why voters will NOT be complacent but will make sure they vote.
And that is why voters will NOT be complacent but will make sure they vote.
5
Thanks, Nate Cohn and the NYT, for the best polling and stats class I have ever taken in one sitting.
19
Worthwhile info.
2
Wonderful art work!
21
Thanks for the insights. The bottom line for me is how can any poll be showing anyone still voting for Donald Trump and what do those polls tell us about America.
5
Nate:
This might be out of your zone of expertise, but could you do a statistical analysis of the possibility of voter fraud in a national election? My understanding is that it is fantastically low. But Trump is now fanning the flames of facism, in my opinion, and alleging that if he loses it is proof that the election was rigged and there was widespread voter fraud. Someone needs to call him out on this. Thank you.
This might be out of your zone of expertise, but could you do a statistical analysis of the possibility of voter fraud in a national election? My understanding is that it is fantastically low. But Trump is now fanning the flames of facism, in my opinion, and alleging that if he loses it is proof that the election was rigged and there was widespread voter fraud. Someone needs to call him out on this. Thank you.
68
Urgently, and thoroughly!
Unfortunately, he and some of his supporters are not data driven, are distrustful of science, the media, government, and are given to impulse. I have to say that I'm a little worried!!
Unfortunately, he and some of his supporters are not data driven, are distrustful of science, the media, government, and are given to impulse. I have to say that I'm a little worried!!
1
Personally, I would love to see such an analysis. But it won't help your fanning the flames scenario, because the flames he's fanning are the last segment of society to even come into contact with a statistical analysis, much less begin to comprehend one.
Trump has been "called out" highly effectively on virtually every aspect of policy and politics; none of it penetrates the impervious shell that separates those who will vote for Trump from reality.
Trump has been "called out" highly effectively on virtually every aspect of policy and politics; none of it penetrates the impervious shell that separates those who will vote for Trump from reality.
1
Love the illustrations!
10
WRT to 4 way or 2 way polls, there is a simple mathematical point that Nate ignores. First of all 2 way pools are not realistic. Stein and Johnson are on the ballot in every state. So we have to look at 4 way polls.
But in a 4 way poll, you have to consider that votes for the 3rd party candidates are wasted; they are never going to take a single state. So what you have to do is to normalize, i.e. change the figures so that the votes for Trump or Clinton represent 100% of the votes , not some smaller figure.
Example, suppose Clinton gets 45% and Trump 35% and the remaining 20% go to the other candidates. Clinton lead is not 10%. Her share of the votes that count is 45/0.8 = 56.25% and Trump's share is 35/0.8 = 43.75% so her lead is actually 12.5% (if I have done the arithmetic right).
Of course, a state by state analysis is even better, but the same argument applies to each state.
But in a 4 way poll, you have to consider that votes for the 3rd party candidates are wasted; they are never going to take a single state. So what you have to do is to normalize, i.e. change the figures so that the votes for Trump or Clinton represent 100% of the votes , not some smaller figure.
Example, suppose Clinton gets 45% and Trump 35% and the remaining 20% go to the other candidates. Clinton lead is not 10%. Her share of the votes that count is 45/0.8 = 56.25% and Trump's share is 35/0.8 = 43.75% so her lead is actually 12.5% (if I have done the arithmetic right).
Of course, a state by state analysis is even better, but the same argument applies to each state.
18
I suspect I'm not alone in my obsession with the polls during these crucial final weeks, reacting with fear and dread each time Clinton's chance of winning shifts downward a fraction of a point, and feeling giddily confident when there's an uptick. Others have told me that the polls now shape their mood and how they function during the day.
Millions of us are bone-tired and practically traumatized, knowing there's more at stake this time around than ever before. We just want it to end. Along with watching the polls, I'm now crossing off each day on the calendar, willing November 8th to arrive already. How much more of this can we take?
Millions of us are bone-tired and practically traumatized, knowing there's more at stake this time around than ever before. We just want it to end. Along with watching the polls, I'm now crossing off each day on the calendar, willing November 8th to arrive already. How much more of this can we take?
56
Oh man, I'm so there with you!
Interesting article, of course one is brought back to that moment in 1948. The Chicago Tribune was so certain that "Dewey Defeats Truman " that they ran it on the front page of their newspaper on November 3, 1948. Opps, bad poll.
The polls are meant to excite one party or the other and to emphasis where more work may be needed. Also keep in mind the pollsters make their living by fear.
Yet we all know there is only one poll that matters, the election. Many supporters of a controversial candidate might be reluctant to declare for that candidate when asked, either through fear of rejection or peer pressure.
I look at who is supporting a candidate. Do the military leaders, or the civic leaders back this candidate or the other.
Donald Trump is supported by many police groups, border guards etc. A Reuters Poll in June of this year revealed that many of Clintons supporters are racist, they feel blacks are more violent than whites (33%), they also feel blacks are more criminally inclined (33%).
So while polling is important it is also subjective, depending on the structure of the question.
Further, polling is fun. It generates a ton of discussion and then becomes entertainment in its own way.
3
"The Chicago Tribune was so certain that "Dewey Defeats Truman " that they ran it on the front page of their newspaper on November 3, 1948. Opps, bad poll."
Rubbish.
The entire point of that famous flub is that the media DIDN'T poll. They were so certain Dewey's lead was insurmountable that they decided to save money and stop polling (much more expensive in the days when polling was door-to-door) - and thus missed the massive late movement to Truman.
This tendency of undecideds and learners to make final decisions late is exactly why tracking polling has become so extensive in the last few weeks of a campaign.
Rubbish.
The entire point of that famous flub is that the media DIDN'T poll. They were so certain Dewey's lead was insurmountable that they decided to save money and stop polling (much more expensive in the days when polling was door-to-door) - and thus missed the massive late movement to Truman.
This tendency of undecideds and learners to make final decisions late is exactly why tracking polling has become so extensive in the last few weeks of a campaign.
4
In spite of all the known inaccuracies of polling techniques, the Commission on Presidential Debates, a private organization run by Republican and Democratic party officials, is still using polls as an excuse to keep viable Presidential candidates from being heard in the Presidential Debates. Silencing political opposition should not be tolerated in our representative democracy.
10
I don't think "viable" means what you think it means in the context of politics.
Dream on - neither Stein nor Johnson are viable candidates; and both are such poor candidates that participating in the debates would probably hurt them anyway.
The bar for participation in the debates is very low - well outside any kind of polling error that would exclude then if the were, in fact, viable.
Let's focus on the reality we live in and not some alternative dimension where every John, Dick, and Harriet is viewed as a 'viable'.
The bar for participation in the debates is very low - well outside any kind of polling error that would exclude then if the were, in fact, viable.
Let's focus on the reality we live in and not some alternative dimension where every John, Dick, and Harriet is viewed as a 'viable'.
2
I note that the upshot has HRC at 88%. In the data world, where nothing is certain, it makes me wonder what the upshot would have predicted for Obama in 12? The tv MSM playbook of turning this into a horse race has caused an increase in my blood pressure. As a former political science student who took classes in polling and data before personal computers and cell phones, being able to dig into the data is strangely calming. That 37% of the electorate thinks Trunp is an acceptable candidate makes me despair.
44
Paying too much attention to polls during an election season is akin to a person watching the daily ups and downs of the stock market: it puts you too much into the forest so that you can't see the trees.
The best, or at least the most sane way to understand polls (and markets) is to look for trends on a minimum of a week, or a month, based on cumulative averages. Pollsters and professional investors know how to do this so that they can make more informed decisions about future directions in voter sentiment and market prices. Of course, this also assumes that you are paying attention to external events that can impact poll results and stock prices, like Trump's disgusting video or the literal meltdown of Samsung.
The best, or at least the most sane way to understand polls (and markets) is to look for trends on a minimum of a week, or a month, based on cumulative averages. Pollsters and professional investors know how to do this so that they can make more informed decisions about future directions in voter sentiment and market prices. Of course, this also assumes that you are paying attention to external events that can impact poll results and stock prices, like Trump's disgusting video or the literal meltdown of Samsung.
24
Republican strategists need to read this article so they can understand how political campaigns should be run. Late polling in 2012 clearly indicated Mitt Romney would lose but he and his team were completely caught off guard and stunned when they lost (by virtually the exact spread predicted by the reliable polls). Through tears, Ann Romney expressed her utter bewilderment at losing because they had encountered so many enthusiastic people at rallies across the nation (as if "emotion" should have translated to victory).
Perhaps because a larger percentage of Republicans are religious they tend to give themselves over to euphoria and fantasy rather than compiling and analyzing hard facts that may lead to a different conclusion than the one fitting their distorted view of reality and circumstances. It's a degree of naivete and ignorance that shouldn't come into play with professionally run campaigns -- especially at the presidential level.
In 2008, John McCain was reading polls and well aware he was on thin ice toward the end of the campaign. He made a couple of significant strategic moves to try to shore up support but they weren't effective and Mr. McCain wasn't stunned when he lost because he didn't ignore the accuracy of higher quality polls that were predictive of the outcome.
According to polls, Trump losing has really never been much in doubt, but some people will believe whatever makes them feel warm and fuzzy and buttress their fantasy of getting what they want.
Perhaps because a larger percentage of Republicans are religious they tend to give themselves over to euphoria and fantasy rather than compiling and analyzing hard facts that may lead to a different conclusion than the one fitting their distorted view of reality and circumstances. It's a degree of naivete and ignorance that shouldn't come into play with professionally run campaigns -- especially at the presidential level.
In 2008, John McCain was reading polls and well aware he was on thin ice toward the end of the campaign. He made a couple of significant strategic moves to try to shore up support but they weren't effective and Mr. McCain wasn't stunned when he lost because he didn't ignore the accuracy of higher quality polls that were predictive of the outcome.
According to polls, Trump losing has really never been much in doubt, but some people will believe whatever makes them feel warm and fuzzy and buttress their fantasy of getting what they want.
82
"Republican strategists need to read this article so they can understand how political campaigns should be run. Late polling in 2012 clearly indicated Mitt Romney would lose but he and his team were completely caught off guard and stunned when they lost..."
No they don't because no they weren't. Politics is essentially about the denial of reality for personal gain. The earth is rapidly overheating due to human-caused pollution with disastrous consequences. It's physics. But politicians believe that if one doesn't believe in the "laws" of physics, one can simply deny they exist, or vote to suspend or amend them. This is also the mentality of economists and lawyers.
No they don't because no they weren't. Politics is essentially about the denial of reality for personal gain. The earth is rapidly overheating due to human-caused pollution with disastrous consequences. It's physics. But politicians believe that if one doesn't believe in the "laws" of physics, one can simply deny they exist, or vote to suspend or amend them. This is also the mentality of economists and lawyers.
4
Actually, the problem for the Romney campaign wasn't an inability to read the polls, it was that they made starting assumptions about the electorate - about who would show up to vote - that were wrong. They based their assumptions on historical data and misunderstood the participation changes wrought by demographic shifts and the Information Age.
While their insistent commitment to their assumptions was a classic case of being wedded to comforting interpretations, any poll and any campaign is subject to the same error when they project the likely electorate. It is precisely this factor that has kept the Clinton partisans up at night, the "what if a lot of Obama voters - African-American, "millennials,"etc. - don't come out at the rates we expect?; what if non-college whites come out way heavier than ever before?" etc., etc.
While their insistent commitment to their assumptions was a classic case of being wedded to comforting interpretations, any poll and any campaign is subject to the same error when they project the likely electorate. It is precisely this factor that has kept the Clinton partisans up at night, the "what if a lot of Obama voters - African-American, "millennials,"etc. - don't come out at the rates we expect?; what if non-college whites come out way heavier than ever before?" etc., etc.
7
@Bill, I completely agree with your comment.
I would like to add that this time around Trump is stirring the pot by suggesting that the system is "rigged". He even used Bernie Sanders to support this theory. The DNC might have been rigged, or not, eitherway informed voters of which Trumps supporters are not, know that that has nothing to do with the general election.
Now with Trumps supporters believing the system is rigged and the media is against him, if God willing Trumps loses the election aftermath will not be the same as when Ann Romney cried. I think we will see violence and a lot of disruption.
Let's not also forget Trump suggested that if his followers don't like the outcome of the Supreme Court picks, his second amendment people could take matters into their own hands.
This could get very ugly very fast, and for a long time.
I would like to add that this time around Trump is stirring the pot by suggesting that the system is "rigged". He even used Bernie Sanders to support this theory. The DNC might have been rigged, or not, eitherway informed voters of which Trumps supporters are not, know that that has nothing to do with the general election.
Now with Trumps supporters believing the system is rigged and the media is against him, if God willing Trumps loses the election aftermath will not be the same as when Ann Romney cried. I think we will see violence and a lot of disruption.
Let's not also forget Trump suggested that if his followers don't like the outcome of the Supreme Court picks, his second amendment people could take matters into their own hands.
This could get very ugly very fast, and for a long time.
3
Basic Concept: The reciprocal of the square root of the sample size (n).
Basic Numeracy on BC: For larger samples, smaller values of 1 / sqrt(n).
Referring to one of Mr. Cohn's bulleted points under "Scan the Methodology", we could see the following:
Using the (very) good approximation for the margin of sampling error,
MoE = 1 / sqrt(n), let's look at such for the three sample sizes he explicitly mentioned:
n = 400: MoE = 1 / sqrt (400) = 1 / 20 = 0.05, or 5 percentage points.
n = 800: MoE = 1 / sqrt (800) = 0.035, or about 3.5 pp.
n = 1200: MoE = 1 / sqrt (1200) = 0.029, or about 2.9 pp (typically rounded to 3 pp in the NYT and elsewhere).
Now consider Mr. C's comment that "The gains from big samples are smaller than you might think, ..."
Notice as n went from 400 to 800 to 1200, the MoE decreased about 1.5 pp and 0.6 pp respectively.
Let's try n = 1600: MoE = 1 / sqrt (1600) = 1 / 40 = 0.025, or 2.5 pp, and so the decrease is now down to about 0.4 pp
More generally, an algebraic expression for the MoE is
MoE = 2 sqrt (p(1-p) / n)
where p represents the current proportion of the vote for the leader.
[Thus, p = 1/2 = 0.5 gives that 1 / sqrt(n), as used above.]
For more sophisticated sampling designs (such as those used), the above expression is more involved. Nevertheless, there is Always that sqrt(n) in the Denominator.
HTH