That’s kind of nonsensical, frankly. Polls do work so the casual observer has a sense of who’s likely to win by observing polls. And the current king of poll reading, Nate Silver, uses the idea that previous results and polling affects future results. The only reasonable thing I can take from your post is that “polls aren’t everything/perfect” which qualifies for a “no shit, sherlock”.
No, actually it is not. Mind you I had expected it to be but surprisingly polling aggregation has continued to do fairly well. Primaries to be sure are smaller pools less densely sampled compared to national general election polling, and caucuses even tougher … but overall a surprisingly good set of results. Certainly no reason to conclude the lowest in 50 years.
Yes it is of course true that polls in particular right now, mid-June and through July are historically of only a little predictive value for the final score and aggregated large sample polls just before the election is more predictive. Third inning up by two has only a certain value at predicting the final score of a baseball game, less predictive of the outcome than up by five going into the ninth … but that does not mean that it is of no value. It is how the contest is going right now. No hitter so far does not assure that next inning won’t let up five runs … but one suspect that your pitcher is likely to continue to have a good day and the fans care.
Nate Silver is undoubtedly the most unbiased person to cite on the accuracy of polling. What possible motivation could he have for putting the best spin on the issue?
Look, I’m a Silver fan. But his article stretches the wonderfulness of modern polling. Maybe we should check a couple of other opinions.
http://www.bloomberg.com/politics/articles/2016-05-03/the-2016-guide-to-political-predictions-which-matter-and-who-was-most-underestimated
But I guess in the good old days polls were 100% accurate.
Gee, if I strip out all the other misses, I’d get 100% accurate every time. But why should you strip out any misses?
And look at that table column marked “538 grade.” Those marks are not very high.
Here’s the real issue, and it won’t be found in that article. Most elections aren’t that hard to call. The House regularly re-elects around 95% of incumbents. The Senate is far more competitive, but 80% of incumbents win every year. Governors are somewhere in between. You’d have to be very, extremely, deliberately incompetent at your job to go below an 80% success rate, and I’d say 90% is a minimum baseline for anyone who wants to get hired a second time.
Of course, when no incumbents are in the race it gets a little harder. That’s also true for races in which fewer people vote. Primaries are harder yet. And caucuses are outliers.
From your link.
Any article that argues that some absolute percentage of correctness allows you to gauge polling is not thinking clearly or is deliberately trying to slip something by you. What I want to know is how good the polls are for races that aren’t obvious. But nobody can tell me that.
And none of that addresses the even deeper issue. Polling is in crisis. Pollsters know that. Election polls get all the headlines, but they’re a small fraction of total polling. Most polls are taken on subjects where you don’t have a 95% success rate baked in. They are getting harder to do. You might as well deny global warming as deny global polling failure.
I was called last night by a pollster who asked me questions about ONE proposition on the upcoming November ballot in California and the poll took almost half an hour. She had so many questions, and a lot of the questions were repetitive. She read EVERY argument pro and con, one at a time, and then went over all of the options "Is this extremely believeable, partially believeable, partially unbelieveable, not at all believeable. And she repeated every choice for every question after every argument. And then when she had read the arguments pro, she asked me if I had changed my mind from my original belief. She then read all of the arguments for the con, one at a time, with all of the options after each question, then asked me at the end of the cons if I had changed my mind. She then read a list of supporters, and asked me how I felt about them, and a list of the opponents and how I felt about them, then once more asked me how I felt about the proposition and what my opinion was. She then read the ballot language on the proposition and asked me once again how I felt. About 20 minutes in, I was about to hang up. I can’t imagine having to go through that for all of the propositions.
86% success without stripping out the misses. And one of those misses had an obvious in hindsight flaw.
Yes, but more a shift than a crisis. As noted d in the article I linked, SurveyMonkey is an example of a polling company that has both embraced the new reality and is accurate. And one of their claims to fame is being right in the widely mispredicted last UK election.
I kinda thought my whole point was that cherrypicking single polls was absolutely the wrong thing. But I guess I’ll repeat that as often as necessary until it sinks in.
Polls in general/aggregate within a month of a contest are accurate predictors. SurveyMonkey, as a company, has gained a reputation for being accurate over multiple polls. Are you simply denying this?
Polls taken so far don’t offer evidence that Johnson hurts Trump more than Clinton. In the RCP avg of polls which don’t include Johnson, Clinton leads by 5.8%. In the average of those w/ Johnson her lead is 5.3%. If you go to RCP and look at particular polling outfit’s polls with and without, same thing, Clinton’s lead tends to slightly shrink when adding Johnson. It’s all subject to sampling error, other poll problems and ‘it’s early’. Just saying, the assumption Johnson would pull more from Trump doesn’t seem to be supported by most actual poll results right now.
Jill Stein would definitely pull more from Clinton if she’s a non-negligible factor by November but this is a caveat also applying to Johnson. Unless he really breaks out, which means getting to the 15% threshold to be allowed in the debates, then performs very impressively in those debates, his share will likely shrink dramatically by November, and Stein’s probably basically disappear. Now a lot of respondents are using those names to register how they don’t like Trump and Clinton. Many fewer will stick to throwing their vote against Clinton or Trump away by voting for a no-hoper when the election actually comes. Again assuming Johnson hasn’t broken out to much higher level.
It’s also plausible to think that if Johnson does go to a higher level, more of that vote might be from Trump than Clinton. But right now that doesn’t appear to be true. Right now Clinton is an advantageous position with or without Johnson, but slightly less with.
Or maybe we should check how aggregation did this cycle so far, as I did in another thread discussing the Indiana primary. Not a complete analysis I admit but bottom line was
Again, this was not the performance I was expecting this season. In September I created a thread specifically called “Problems with polls” bemoaning exactly what your links bemoan, and quoted Silver’s concerns too. There had been some major misses: Israel, mid-terms, Scotland, the U.K. … But bumped it to report back that things may not be so dire afterall.
Now I appreciate your input in that thread, polling has contributed to an excessive focus on the horse race over the issues, so maybe good riddance. But reports of the death of polling have been greatly exaggerated, and the aggregation method as exemplified by the confidence wndows around results given by 538, have been frighteningly good, being wrong pretty much exactly how often they should be wrong if their numbers are crunched right. And that was with primaries.
Again aggregated polls done in June are not so good at predicting what results will be in November. They are instead a picture of the horse race at this moment, subject to change, to stumbles, to being great on the stretch, so on.
And polling techniques are evolving to meet the challenges of modern times.
They actually do tell us a lot.
We’re talking past one another. Polling is much larger than mere election result predictions. Defining an accurate prediction as how many times you predict who will win is sidestepping whether polls add any value. Needing to wait until a month before an election is not what most people talk about as prediction. I don’t have any idea how accurate SurveyMonkey is over the entirety of their polls of all sorts and I’m positive you don’t either. Nor does it matter, since that’s not really what I’m addressing.
Yes, I would hope that given the millions of dollars flooding in, polling techniques will adapt to modern conditions. I’m emphasizing will because I don’t think they have yet.
Even if you limit the subject to election polling, about all we can posit is that polls give a fairly accurate picture of what opinion is like right now. That’s a very different thing than getting the answer to what opinion will be like many months in the future.
Almost every post on polling here on the Dope is wrong, mostly because people don’t understand that important point and so stretch the meaning of the data beyond any usefulness. Worse, people extrapolate linearly. For almost a full year every bump in the polls had people screaming about brokered conventions. I kept saying they weren’t going to happen. But the polls said otherwise. Um, they didn’t. They didn’t say anything about the future at all.
I’d much rather get good discussion going than come back to say I told you so. The best way to foster good discussion is to ignore the polls. Really. There’s other stuff to talk about.
Certainly it seems we are talking past each other.
OF COURSE what polls tell us, to some degree of accuracy (and understanding the limits of that accuracy is important) is what opinions are right now. And “right now” has a certain degree of predictive value, relatively less now actually than a month or so ago historically, and more as we enter September and beyond.
But yes when I cheer for a sports team I do care about the score right now, early in the game, even though it has no absolute correlation with the final score. How the teams are doing may result in various tactical adjustments.
Um actually the best number crunchers of the polls did not say “brokered convention” - Wang in particular was belittling that talk. 538 called it unlikely too. It was pundit talk not polling talk.
Issues this time? Only so much to go on about?
I meant here at the Dope. There’s a dent in the wall where my head kept meeting it.
I would think a far more accurate metric would be “what percentage of the time were those polls within their MOE?” (and, related, was each MOE out of line with other polls?)
So let us be clear. The problem you have is not the polls but how some posters misunderstood and/or misused the polls.
In fact the polls contributed only a little data to the speculations regarding a brokered convention and pathways to it or not. Mostly they were speculations made in the absence of much actual data.
Does “right now” tell us much about how Johnson will do? I do not believe it tells us much. Pretty much all most know about him right now is that he is a generic “none of the above” option. If he had more exposure maybe he could attract more support, or drive away the support he has. The importance of national head-to-heads with him listed is huge as he only gets on the debate stage if he hits the 15% or greater threshold in five national surveys before the scheduled debate. If not then not even a kiddie table.
For those who to see him get exposure and thus votes in this election (for whatever reason), or even just care about it, his inching closer to that threshold is a thing that matters. And keeping Stein off the polls list matters to them too as he has gotten a few 11s and a 12 in a three-way choice but no more than 4 in a four-way choice poll.
Looking at the race right now I don’t think he’ll make that threshold. And my not polling-based speculation is that the votes he will get on election day will be mostly voters who would have not bothered voting for president if there was not a “none of the above” protest option to select. I highly doubt his being a choice will alter the outcome in any state and certainly doubt that it will have an impact on the outcome.
No, the misunderstandings go all the way down. The media always does horse race journalism. You aren’t overtly partisan if you report on facts, however dubious those facts might be. That’s much easier to cover than policy. Print media used to be able to do long, often thoughtful examinations of policy. Cable media virtually never do. They bring in pairs of opposed talking heads to yell at each other. And when viewers eyes glaze over, they trot out the interactive maps and go back to poling numbers.
There are also more polling firms these days, and each of them have an incentive to get their name into the stories about polls. And there’s that new thing, the poll agglutinator that puts each day’s polls into a running total that goes back to the media for their daily article.
The entire polling culture is as fault from top to bottom. It’s rotten all the way through.
Well continuing the hijack then (note that I did try to bring this back to Johnson) …
Right now, Clinton versus Trump, what policy issues do you miss being discussed in detail?
Seriously is there much to actually discuss? It is not like Trump actually has a coherent set of policies.
And (trying again to bring this at least somewhat back to the op of the thread) it is not like most who currently would say they would vote for Johnson are actually making a choice based on a preference for any policy positions that he articulates. Let’s face it, most American voters don’t even know what the word “libertarian” even means. The might know that Johnson and Weld are formerly of the GOP. They know they are outsiders to the current system. They know that they don’t know them well enough to dislike them yet and do know enough about Trump and Clinton that some dislike them both. Would many an issues based Sanders supporter really want to go with someone who is against all income redistribution, wants to abolish the IRS, and who is not sure that humans are responsible for global warming? Would many a Cruz or even a Bush or Kasich supporter really want to support someone who ran a marijuana products shop, wants to legalize drugs, is pro-choice, and is for gay marriage equality? He’s not going to be a major factor and highly likely most of his votes on election day won’t be coming from either of the two major party candidates but from people who otherwise would have stayed home. He might actually help the GOP downticket by getting them in the booth though.
Anyway, you choose what media and fora you frequent. Want a policy debate? Start one, maybe here maybe in GD. We have lots of 'em with no reference to polling, about trade policy, living wage, funding college, immigration, the hollowing out of the middle class … Personally I don’t watch the cable or network news hardly ever … even for horserace analysis it is typically inane at best. But if I turn it on I have no one else but myself to blame that I am being exposed to it.
But yes the Elections forum is mostly about the horse race, about the contests, and issues as they impact the contests more than for their own sake. Complaining that people reference polls, which often is the only actual data we’ve got, and critically dissect them here is really a bit silly.
I will state that it is a major mistake for the politicians themselves to be publicly following polls as if they are worthy major news. Trump of course does that and Sanders was too.
I watched the Samantha Bee bit on Johnson. Yeah he was a little goofy in the interview bits, but come on he was on a comedy show. I loved when they showed the Q&A at rthe LP convention and asked “Driver’s licenses - yes or no?”. He so knew he was going to get booed for saying some basic competency testing is a good idea. Lol.