What? No it isn’t. First-past-the-post voting, no proportional representation, gerrymandered districts… the system is terrible.
Granted, the UK is even worse off, possibly because they’re using FPTP voting even though they have more than two parties.
For me, I’m a big believer in a 50-state strategy, and have been at least since Dean called it that, so I call dibs :).
It feels like a relevant comparison would be UK districts / Canadian ridings, versus U.S. Congressional districts, of which we have 435. No doubt, many of those Congressional districts are polled for purposes of projecting those Congressional races, but as so many of those districts aren’t competitive in the general election these days (due to gerrymandering creating safe GOP or Democratic seats), many districts probably see very little polling for the Congressional race, much less the Presidential race.
And, except for the couple of states where Electoral College seats are partially divided by Congressional district (Maine and Nebraska, IIRC), there’s no incentive for pollsters to survey at the district level. For that matter, smaller states that are considered to be safe in the Presidential race (e.g., Alabama, Wyoming, Vermont) aren’t polled often, either.
You and me and millions of other left-of-center types who are still pissed, even now, that the Dems abandoned the 50-state strategy after Obama was elected.
My recollection (which I’m too lazy to Google for accuracy) is that it came down to Rahm hating Dean’s guts, and killing the 50-state strategy just because of that.
An observation and a question about 538:
Observation: It’s depressing to me how, in the wake of Trump’s outing himself as a sexual predator, the polls are nonetheless now inching back in his favor. For instance:
Oct 11-13, Clinton by only 1 Nationally.
One might think polls from those days, immediately following the revelation of the Trump/Bush conversation, would show more evidence that the most decent fringe of Trump supporters and undecideds had abandoned him. But it seems not; 538’s percentages have been going down for Clinton over the past 48 hours. Ugh.
Question: It seems that 538’s numbers only change after there have been new polls. But shouldn’t the number be changing anyway as the number of days before the election grows smaller? I’d think a 5-point lead tomorrow would be more predicative of victory than a 5-point lead last week or last month, and that an identical lead a week from now would be even more so. But Silver’s victory percentages seem only to move with new polls, and not with time passing. I’m sure I’m missing something, but I’m not certain what.
I think it’s that as the election draws nearer, he makes less adjustment and allows for less variabilty in the poll numbers but you still need new polls. Just because we get closer to the election doesn’t make the week old poll more accurate though, which is what you sort of imply.
Ah, yes, that’s an excellent point. Thanks!
What we should expect to see then, I guess, is that if each day a new poll comes out identical to the previous ones, the percentages should go up for the leader even though the poll margins are the same.
Clinton is starting to stretch her lead.
I think we can all agree that the Rasmussen poll leans right. Polls that ‘lean’ aren’t really credible. Realistically, Clinton’s poll numbers are probably somewhere between the recent ABC/WaPo poll, which shows her only winning by 4 points and the Wall St Journal poll, which shows her winning by 11. I’m guessing the real edge is probably around 7-8%, which what Fox News had in their last poll. I don’t think the polls are done moving, either. It’s been a historically horrendous week for Trump, who really did nothing to defend himself except try to capitalize off of Wiki dumps that hardly got noticed.
We’re approaching where we were in early to mid-August. I still think the race will tighten a bit, but realistically, even if you look at Trump’s two previous comebacks, it has taken him about a good month to pull nearly even, which isn’t actually ‘even’. In short, if Trump were to win, it would have to be the greatest political upset in the modern history of American elections. The final numbers will probably fall in line with the long-term polling trends that have been consistently showing throughout the race. At this point my guess is that Clinton will end up winning by 3-5 points on election day – at least based on what exists now. But who knows in this bizarre campaign.
Ah. The ole dismiss the polls that you think lean against you trick.
Polling houses almost all have leans of one direction or the other. Toss out those that lean and you are left with none.
Either you calculate taking those various leans into effect and correcting for them (the 538 method, corrects the Rasmussen result by 2) or you consider all recent polls and presume the leans cancel each other out (RCP), perhaps throwing out an outlier in each direction (PEC’s method).
Tossing out the poll(s) you think leans against your preferred direction only is extremely silly.
FWIW by the different actual methodical approaches lead to 6.6 in 538’s PollsOnly and 6.7 NowCast; 5.5 in RCP’s four-way; and 4.7 in PEC’s meta-margin. Note that 538 and PEC also use state results to deduce the national number to varying degrees (PEC exclusively state based and thus lagging) and that RCP does not.
You don’t understand how polls work. Polls that lean can absolutely be more credible than polls that don’t. Bias and reliability are not the same thing.
Suppose you and I own polling agencies, AsahiPoll and RickSurvey. AsahiPoll runs extremely broad, carefull polls, but your demographic sampling is flawed in such a way that you always overstate the Democratic vote by two percent. We can safely say that if Clinton wins 47-43 on Election Day, your polls will have all been about 49-41, within a 3-point MOE, the week prior. That also holds for your polling of Senate races and whatnot.
RickSurvey, meanwhile, does a much better job of demographic sampling, so my company has no bias. On average, a RickSurvey poll is as likely to favour R as D. But it’s otherwise poorly designd and so is wildly varying - my survey could be up to ten points wrong on either side.
Which survey is more reliable?
I’m aware of that argument and I agree that a house effect doesn’t necessarily imply bias, which is probably a word I should have avoided using in the first place. But it does seem to be an outlier, and if 538’s rating system is to be believed, Rasmussen’s rating has gone down in recent years (C+). I actually remember following Rasmussen in 2008 and recall it being an accurate poll at one time.
Is “house effect” different from the strict statistical definition of “bias”? Because it seems like the same thing to me: a systematic error in sampling, estimation, modeling, etc.
Or is everyone just trying to avoid the term “bias” because the common language definition implies some sort of intentional manipulation on behalf of a certain candidate?
It’s certainly possible that Ramussen is a relatively imprecise poll, in which case it should be weighted lighter than the other polls in the sample. But their house effect is not evidence of that.
Everyone’s looking at Arizona, and wondering about Georgia, and hoping for a blue Texas, but y’all should cast your eyes north (or to the lower left on 538’s page) and take a gander-see at Alaska, which is moving rapidly into toss-up territory.
Now I don’t actually think a Clinton victory in Alaska is likely, but if the bottom drops out even more for Trump, I think perhaps Alaska may be the next to follow Arizona into the blue column, not Georgia.
It could be, but as I pointed out earlier, while the odds might be slightly better for Alaska, Georgia has been more volatile. Clinton has never led in Alaska, while she did briefly in Georgia back in August.
If Georgia and Alaska are even in the conversation in ten days, the election is over. If there is any doubt about Georgia, that makes, by my count, 358 other electoral votes as blue as Grover’s ass.
I think they’re the same thing.
Nate Silver comes from a sports stats background, and I think he took the baseball term “park effect” (how much a particular ballpark tends to increase/decrease runs scored) and modified it for pollsters.
+1, as the kids say. (To the whole post, not just this part. Excellent explanation.)
We’re 21 days out. Clinton’s race to lose. No data to back it up but it seems that Trump has most likely sustained permanent damage among the less partisan independent voters, especially women. Before his wave of bad press his ceiling was probably 45 percent. Now he’ll probably be lucky to make it past 40, though it’s very unlikely he drops below 35. Thus, the only conceivable way for Clinton to lose at this point would be for Clinton to drop below her floor.
That’s not very likely but still possible. While the national polls are indeed favorable for Clinton, the electoral college math still gives Trump a glimmer of hope. He’s within striking distance in North Carolina, Florida, and New Hampshire, tied in Nevada, and actually appears to have a slight advantage in Ohio. Pennsylvania is arguably the most difficult of the swing states for Trump to capture but Trump is running an unorthodox campaign that is wildly popular in some parts of the state.
If Trump has a good debate performance, and if Clinton’s email problems grow, the race could get interesting again - I think it will actually. Even so, it would probably require something dramatic for Clinton to lose. In my mind the worst thing that could possibly happen would be something that would cause her own base of support to crack at the last minute. Other than that, I suspect three weeks of Roger Stone dirty tricks intended to suppress voter enthusiasm in the key states. The problem for Trump is that the map is so wide now.
I agree. (I definitely think only women’s votes should count on issues like abortion.)
That Trump would be dramatically crushing Hillary if only men’s votes counted should be astounding and very sad. Pew Research (August 18) shows Trump winning 45 - 33 among men, and winning by the same 45 - 33 margin among non-Hispanic whites!