Aside from the joy of getting to write the word "balls" in the thread title, the reason I am eager to separate this thread from the other thread is to make this one completely about analysis of the election, predictive models that try to forecast the election, and their subsequent success/failure. In this regard, I'm hopeful the mods will agree to its merits and the posters will agree to limit substantive discussion of the issues to the other thread. That said, we are now ready to begin forecasting the 2016 Presidential election. I will post occasionally here with updates from all the major online polling aggregators (probably as often as I update my MLS East v. West thread), and other posters will pick up the slack when I become derelict in duty. First, I'll introduce the aggregators (if anyone knows others, I'm happy to start including them): Electoral-vote.com: this website has been with me since 2004 and has done a fairly good job predicting election results, although the blogger is insanely left-of-center and his analysis of polling isn't terribly accurate (see him and Zogby about 2004's election). He won't be in these early posts because he's not going to start running his aggregator until after the convention - I believe he still thinks Trump won't emerge as the nominee. He also does Senate predictions. Election Projection: This website was also with me since 2004 and I consider it, with all deference to Nate Silver, the absolute best place for a liberal to get election predictions. The website organizer is as socially conservative as they get and has no qualms about personally discussing his aggregator with anyone (myself included). If you want your beliefs challenged, go here. He also includes predictions of the Senate and House. Pollster.com: I picked this website up in 2008 on the recommendations of a professor, and have not been led astray. They don't do Electoral Vote forecasts, but do report trend lines on individual state and national vote indication polling averages. FiveThirtyEight: The one, the only. This year, they're more interactive with the forecast page. Go here once a day. No more. More frequently, and you start to read a signal in the noise. Sam Wang: An actual political scientist. He's not as well-known, but he does just as well as Silver. RealClearPolitics: This is where lazy people get what Pollster.com offers. C'est le vie. Okay, I will split predictions into Popular Vote and Electoral Vote columns, and then average them to create an aggregate of the aggregators. In July of 2012, this approach led me to conclude that Obama would receive 51.7% of the vote and 332 electoral votes, so I'm willing to give it another go. One last note: I am including Johnson vote share but not Jill Stein. Only RCP includes Stein-in options. Pollster.com includes "Other," which combines Johnson but not Stein, and both RCP and 538 do Johnson-in only variants. So to save time, for now Johnson is in, Stein is out. If there is movement toward including both Stein and Johnson, or dropping both of them, among these aggregators, I will follow suit. Electoral Vote Totals: Electoral-Vote: N/A ElectionProjection: 349, 189 FiveThirtyEight: 348.9, 188.3, 0.8 Sam Wang: 330, 208 Average: 342.6, 195.1, 0.3 Popular Vote Totals: Electoral-Vote: N/A ElectionProjection: 51.9, 45.1 Pollster: 45.5, 38.6, 5.1 FiveThirtyEight: 48.7, 41.8, 8.1 RCP: 41.9, 36.6, 7.9 Average: 47.0, 40.5, 7.0 These initial forecasts indicate a Hillary Clinton victory on par with that of Barack Obama in 2008. However, even if all Johnson voters were to vote for the Republican candidate instead of Johnson (something we know would not happen), Hillary would barely lose in a nail-biter.
'Nother thing about electoral-vote.com is that they're pretty dead set on the election only being an electoral college vote and nothing more. I think others have shown that once you hit 52+%, that electoral map really starts to reflect the popular vote. Swing states are swing states for a reason: they reflect subtle demographics in the nation as a whole, so when the nation shifts, the swing states reflect that. Don't get me wrong. The Dems have a hefty federal election advantage, and I've posted at length about that.
You're not wrong. I wish all the forecasters provided both their popular vote guesses and their electoral vote guesses. Wang has a margin for the leading candidate instead of true popular vote. It's a little inconsistent.
Well . . . If a layperson is checking 538 on a daily basis this far out, that might be a sign of OCB. Or that their MLS and/or MLB team is struggling.
You must be checking twice a day! Eastern Conference Pos Team Pts P W D L GS GA +/- 1 Philadelphia Union 26 17 7 5 5 29 25 +4 2 New York City 24 17 6 6 5 27 31 -4 3 New York RB 23 17 7 2 8 28 23 +5 4 DC United 21 17 5 6 6 17 17 +0 5 Montreal Impact 21 15 5 6 4 24 22 +2 6 Orlando City 20 15 4 8 3 28 25 +3 7 Toronto 19 15 5 4 6 17 18 -1 8 New England 19 16 4 7 5 21 28 -7 9 Columbus Crew 16 15 3 7 5 19 22 -3 10 Chicago Fire 14 15 3 5 7 15 20 -5
Looks like Drumpf polls really bad everywhere. Can you say landslide? http://fivethirtyeight.com/features...-national-polls-basically-say-the-same-thing/ All of this yields some slightly complicated conclusions. On the one hand, according to our models, Clinton’s state polls tell a stronger story for her than the national polls do. On the other hand, a lot of that advantage is concentrated in traditionally red states. If Trump underperforms in states such as Texas and Mississippi, that will hurt his position in the popular vote without compromising his Electoral College math — provided, of course, that he doesn’t actually lose them. Hence, our models conclude that Trump is more likely to win the Electoral College despite losing the popular vote than the other way around, although either possibility is unlikely.
Okay, here's a great poll we can talk about that explains why you do averages of ALL polls: First, the topline: Yes, in an election expected to be close in Florida, there is a good chance that any given poll will show either Trump or Clinton with a narrow lead. Second, the average still favors Hillary. FiveThirtyEight and Pollster.com both have her winning the state by 4-5 point margins, even including this poll. If OnMessage is the best pollster of 2016, we'll never know until the election is over. Not including them risks biasing the result. If the CLT is correct - and trust me, it is - excluding this poll for being slightly off is much worse than including them with a low weight, as Nate Silver does. Third, any group that has a partisan funder or donor is going to have means of using legitimate polling techniques to nudge the scales in their favor. See below: A few more things to keep in mind: 1) Likely Voters in July are probably going to be less accurate than Likely Voters in October, but probably more accurate than Registered Voters in October. 2) This poll doesn't show demographic breakdowns, meaning that OnMessage probably oversampled Republicans and white Floridians. AGAIN, that's not nearly as big a deal as if the poll wasn't ever conducted. 3) Historic voter trends - this is the real problem. If this poll excludes people who voted in 2008 and 2012 but not 2010 and 2014, it will oversample Trump voters. If this poll excludes Gary Johnson as an option, it will oversample Trump voters. Any poll not doing three-person questions given this election cycle should be weighted lower. 4) These poll results came out a full two weeks after they were commissioned - the stratification part seems like post-stratification, and coupled with no demographic breakdown report, indicates that the pollsters got a more pro-Clinton initial outcome and decided to tinker a bit. Nevertheless, average it in. Trump's probably a little closer in Florida than the 5-point lead the aggregators had, anyway.
The real question is why Rick Scott's people bothered to fund this poll. He's not up for reelection. If he was, I'd expect to see a poll like this circulated if they were having trouble raising money for reelection. You need to have something -- anything -- to tell your donors. "No, really, Trump won't devastate the down-ballot races. In fact, our polling has Trump in the lead. " I've never heard of OnMessage. Maybe they're tied to Rubio too, or to some other down-ballot candidates. Bottom line -- you don't need to delve deeply to justify throwing all partisan polling into the trash bin. Genuine campaign polling will never see the light of day. But partisan polls like this one? They aren't conducted to gauge reality. They're done to twist it.
No! This is the opposite of what I wanted people to take away from this. Yes, all polling firms with partisan leanings have potential for substantial bias. But at the end of the day, even the partisan firms have an obligation to get it right, and their credibility suffers for getting it wrong. Don't throw out the partisan polls. Especially the ones you disagree with.
Actually, no. You're confusing two different kinds of things here. When I say partisan poll, I don't mean outfits Rasmussen -- guys with partisan leanings, whose polls tend to have a partisan bias, but who are generally (plus or minus) straight up, professional pollsters. I'm talking about polls by "pollsters" who are hired by campaign or party professionals to do a poll that leads to a specific outcome for use in their campaign and fundraising materials. These aren't polls in the way you're thinking. They're straight up propaganda. I believe that's what this poll is. Look at the PDF. It tells you nothing. They tell you nothing about their screen. They tell you nothing about other questions they may have asked. (For all we know this was a push poll.) You only got it because it was floated to the local media who was dumb enough to publish it. So, no, polls of this sort should be sent straight into the trash. They aren't legitimate because they were never designed to be legitimate. But if you want to argue this, then show me the evidence that their poll is on the level. Meanwhile, I'll just note that they guy who runs this organization and the organization itself is rather suspect: http://www.floridadems.org/news/flo...ections-complaint-against-rick-scott-campaign Let's Get To Work is an electioneering organization (and maybe not an entirely legal one). In fact, it's supposed to have closed shop. I'm guessing OnMessage is its new incarnation. So, no, I see no reason to endow a poll by these guys with any credibility whatsoever.
Interestingly, another poll of Florida came out today with Trump ahead. http://winwithjmc.com/wp-content/uploads/2016/07/Florida-Executive-Summary.pdf It's pretty crappy -- A 72% white electorate in FL isn't going to happen, let alone Trump+15 among Hispanics. https://t.co/CpaGl1YmXC— Nate Cohn (@Nate_Cohn) July 11, 2016 -- but it doesn't send up all the red flags that the OnMessage poll does. For one thing, it actually shows all its data. Still, clearly something went wildly wrong with that data, as Cohn suggests. If @American Brummie wants to use a poll as an example of why polling averages are better, this is the better example. That said, for both of these polls, I expect they'd never see the light of day had they found the opposite result. It's just that in the first, I'm pretty sure the poll was designed to get the result it got. In the second, I suspect he's just a terrible pollster. (He probably did massage the likely voter screen to get a desired result here too, but at least he's up front about the demographics and shows his data.) In any event, according to 538, both outfits are pretty small time (very few polls) and have c-range ratings.
Interesting discussion. Although I agree with @Knave's take, I also see @american Brumie's point. I'd dare to say that even if you know a poll is certified crap, probably it is better to add the subset of data with a lower weight than just completely dismiss it, at least from the mathematical POV.
My worry is that, when you make posts, you give others the impression that your opinion carries weight. You start throwing polls away for bad reasons, and soon you throw ones away for slightly less objective reasons. Give these polls a low weight. Give them a crappy weight, even. Base it off of prior accuracy. But don't throw it away. The Rick Scott poll could be as you say - it truly could. It could have been that way for years and now thinks it can get broader market share by being credible. It could have been taken over by a bunch of newbs who just forgot to release the needed demographic crosstabs. It could be even more cynical than we think - a push poll. You just don't know. And until you know, the safest thing to do in polling is weight and average. It literally cannot hurt your estimates enough to matter.
Oh, good lord ... If that's how you're going to handle this, then: My worry is that when you make posts, you give others the impression that you're a pedant who assumes he's here to educate the rubes and idiots. When we don't bow before you, or worse disagree with or express skepticism toward your ideas, you get upset. (Oh, sorry, you don't like it when I suggest you're terribly insecure.) It's completely obnoxious. If all you want to do is talk down to people about politics, I suggest you get yourself some sycophant students.
The funny thing is, all I did was say his example poll is a crappy, trashy poll -- and quite possibly a push poll. And he agrees with this! He said straight up that it should be given very little weight in a polling average, and even agrees that my suspicions about the poll may very well be true. In other words, the disagreement is basically between whether this poll is worthless or just next to worthless. Do we discount the poll entirely, or do we discount it almost entirely. For this divergence he went and slammed me as someone whose opinion should hold no weight. Good lord ... At any rate, I'm glad you're enjoying your popcorn.
Between 538 and RealClearPolitics, I think my projection needs are fulfilled. Anyways, on Florida, I think Trump will lose Florida, and potentially get pummeled badly; the Hispanic vote will probably be too much to overcome. I generally subscribe to the theory that Trump's primary path to victory runs through the Rust Belt as opposed to Florida.
Three things: The difference between the two (low weight versus discard) is Nate Silver against Dean Chambers. The "slippery slope" argument is used far too much, but in the world of election forecasting it's very easy to start down the road of dismissing every poll you don't like. First, the partisan pollsters. Then, the ones where the demographics don't 100% match your expectations of the electorate. Then so on... 2) We don't know if this poll is "bad," or is actually bad, or is an outlier, or is right on the mark. You don't know. There are priors to inform us about its likely place in the world, but those priors are weak. For example, that poll made its way to Taegan Goddard. He's reputable. FiveThirtyEight has OnMessage as a C+, with a GOP +1.1% "bias." You know who also has a C+? Rasmussen, the Columbus Dispatch, a ton of university pollsters, Dan Jones out in Utah, and ARG, all of whom have storied histories as pollsters. Nate Silver has included the poll, given it a weight of 0.78, and if we were to exclude it, the basic average of Florida polls since the beginning of June would change inconsequentially. 3) Perhaps you don't know this about me, but I'm very much against people who would discard information to fit a narrative - ANY narrative. You may have forgotten how this discussion started: What an interesting claim. What a simplistic way to sort the world. It must be very easy for you. Public Policy Polling has a 538 B+, but they're a partisan pollster, so off they go. McLaughlin Group has a C-, but conducts Internet polling which a lot of pollsters ignore - crap, they're Republicans, so into the trash bin. Gravis (R) - they've got a B-, but dammit all to hell, need to toss them too! Democracy Corps/QDR (D) - They have a B- as well, and they call cell phones. Into the heap. Now that we're just left with the nonpartisan pollsters, what about the RDD pollsters? Surely, they don't capture reality, so toss them! Oh, and remove the ones that don't report to AAPOR, because a) I'm sure you know what AAPOR is and b) if there's no accreditation, there's no point! And maybe we can remove those where they were wrong in 2012 or use 2014-based likely-voter models. Way to go, Dean Chambers.