If you're like me (and who isn't?), you enjoy obsessing over polls and models, and 538 is especially fun for that. Granted, their polling data hasn't been unskewed yet (Romney 2012!), so you'll have to do that yourself. http://projects.fivethirtyeight.com/2016-election-forecast/ If you're interested in the nuts and bolts of their model, they've got ya covered.
I don't know statistical modeling like 538, but from anecdotal evidence I wouldn't be at all surprised to see AZ go Trump, and if that happens his chance of winning the presidency goes up to 45% http://projects.fivethirtyeight.com/2016-election-forecast/arizona/ Too close for my liking.
It's late, so I might be misreading either your post or 538, but I don't see where getting AZ raises Trumps chance to 45%. AZ after all was a Romney state, Trump can win AZ and still lose in a landslide. barfo
Yeah, I don't see anything that implies Arizona materially changes either candidate's odds of winning right now. As things stand, 538 projects Clinton to win ~352 electoral votes. Arizona has 11...even if Trump won those, Clinton would still have an electoral college landslide. Right now, Arizona (and North Carolina) are "cherry on top" type states for Clinton...meaningless in terms of winning the election, but helpful for winning big enough to claim a "mandate," (as politicians love to do). Of course, if Trump turns things around and starts to win in states like Ohio and Pennsylvania, then Clinton having a shot in Arizona and North Carolina could be a lot more important in terms of just winning the election.
538 isn't simply a poll aggregator like RCP or Huffington Post. They use a model, which adjusts based on factors that they've found to be predictive from past elections. Their methodology is laid out at the link I gave in my original post.
2012 proves that Republicans and other conservatives don't believe in these liberally-slanted polls; their President, Mitt Romney, has been great for them since he won their polling in 2012.
It sure looks like their polls are either out of date or they're ignoring the 5 most recent polls in PA.
As I recall, they use the most recent poll from a given agency (but previous polls from the same agency matter for trend lines). Of the four most recent polls in Pennsylvania, they come from the same two agencies (Quinnipiac and PPP) and 538 has the most recent from each of those two polling agencies.
That poll also has by far the biggest sample size and it has the best pollster grade (compared to the more recent polls). I agree that it's strange that such an old poll has the highest weight, but recency isn't the only factor they use for weighting. The real issue is that there still isn't a ton of polling in Pennsylvania. You're stuck with using smaller polls or an older poll.
Why do you think sample size matters more than a) outlier, b) how old the poll is? At that point in time, Ted Cruz was still running and getting votes in the republican primaries, Sanders in the democratic ones (he's still running, apparently).
Larger sample size improves reliability a lot. I can poll myself, with a sample size of one, but the result will be highly volatile compared with another person's self-poll (sample size: 1). The fact that such a large poll showed Clinton that far ahead (and, yes, Cruz and Sanders were technically still in it, but in both parties voters were coalescing around Trump and Clinton) is significant. More significant than more recent, smaller polls? I have no idea, off the top of my head.
How many more attacks by the followers of Islam will it take to flush these polls down down the tube?
I don't have a quibble with the sample size. It's how old the survey is and that it's way out of line with every other poll and that they chose to weight it 1.5x more than the polls less favorable to their preferred candidate. Cruz didn't drop out until about a month later. The history is undeniable. And as MarAzul points out, quite a few potentially opinion changing events have occurred since (not limited to the others dropping out of the race).
They don't choose the weightings personally, the computer model does. You can continue believing, as you did in 2008 and 2012, that 538 biases their model to "make it favor their preferred candidate." Or intervenes when their robot buddy goes off message. This cycle, unfortunately, there doesn't appear to be a good site for unskewed polls. You can try this Twitter feed, though.
Since you like 538.com so much: http://fivethirtyeight.com/features/the-polls-were-skewed-toward-democrats/ And: http://thewire.in/47165/renowned-po...hillary-clinton-in-us-presidential-elections/ However, despite Silver’s successes in the last two elections, many – such as New York Magazine‘s Ed Kilgore – are skeptical about his predictions. This is in large part because of a slew of huge misses in the last two years. Salon reportedsome “high-profile” upsets for Silver after the 2014 midterm elections, both in senate and gubernatorial results. He had abjectly failed to anticipate the decisive influx of Republicans, and his models consistently portrayed an uncertainty driven by an assumed close race. In response, Silver published a piece in his defence that detailed ways in which the polls on which he based his models were skewed towards Democrats. Even more damningly, Silver severely underestimated the allure of Trump, and in August 2015 gave him a 2% chance to win the Republican nomination. It was only by mid-February that his FiveThirtyEight foretold a 45-50% likelihood of a Trump nomination. In a piece titled ‘How I Acted Like a Pundit and Screwed Up on Donald Trump‘, Silver attempts to explain the miss by citing a lack of statistical models to track Trump due to there being no precedent for his rise, and an over-reliance on gut feelings and pundit-esque “subjective odds”. “When Trump came around, I’d turn out to be the overconfident expert,” he said, “making pretty much exactly the mistakes I’d accused my critics of four years earlier”. But this professed ‘unforeseeability’ of the Trump phenomenon doesn’t explain why, this primary season, he predicted a 99% chance of Clinton winning the North Carolina Democratic primary, which senator Bernie Sanders won by half a point, or a 90% chance of winning Indiana which, again, Sanders won. Clinton is the archetypal establishment candidate for the democratic party and there should have been no dearth of data to build an analysis model on.
This is damning of Silver's personal ability to pick'em from his gut but says absolutely nothing about his forecast model. His "2%" prediction was just him eyeballing precedent and political science theory--it wasn't based on a statistical model. Once they began forecasting the primaries, using the poll numbers, the model had a strong success rate. The polls, in fact, didn't underestimate Trump--they generally got his results correct and occasionally overestimated his strength. Link If your point is that his models sometimes get it wrong and aren't perfect, you've successfully defeated that strawman argument! Good work! Your earlier comment was that the statisticians at 538 are purposefully biased, adjusting things to force a result for "their preferred candidate." That's what you also said in 2008 and 2012. Reality proved to be a bit less easy to argue with, though. Oh, and about your "evidence" that 538 is putting their thumb on the scale--a single Pennsylvania poll that isn't the most recent one yet has the highest weight: 538 officially cooking the books for Trump, their preferred candidate. Clinton needs to beat Trump and the 538 media machine.
National polls mean what, exactly? The PA polls are relevant to the Electoral College, and 538 says Trump has like a 2.5% chance to win (ok, I exaggerate deliberately to mock Silver).
