2016 Postmortem
Related: About this forumPrinceton Election Consortium's poll analysis: Hillary has a 90% chance of winning.
Sept. 13, 2016
http://election.princeton.edu
It is good for a cheap laugh to flay the media for its obsession with horserace. This weeks ongoing ruckus with Phlegm-ghazi confirms that reporters cannot get out of their mental rut of some older storyline. In this case, the storyline is Clinton is secretive. Let us pause for a moment. She was concealing her pneumonia because the press would make a big deal out of it. And, wait for it
the press made a big deal out of it.
You, Dear Reader, are complicit in this. I notice that more of you click Presidential links than on the nifty Competitive Congressional District Finder. You like the Presidential horserace. My reason for generating the best prediction I can is to reduce the noise of campaign news. I thought it would clear mental space for thinking about policies, or downticket issues.
The Presidential forecast [methods] takes a low-noise snapshot of state polls, then adds possible drift based on recent elections and this year. Because of intense polarization, few voters are movable. The calculation says that Clintons win probability is 90%. The Senate forecast does the same [methods], but also factors in Presidential-year or midterm-year bias. It says that Democrats+Independents probability of taking control is 72%, which is in the 20-80% range, meaning that things could really go either way. Other forecasts tend to count uncertainties twice, or to overestimate how movable voters are. Other forecasts are also under commercial pressure to attract eyeballs.
Still, the comment section is still peppered with anxious questions about Clintons chances. Honestly, some liberals can be total ninnies. You dont see the conservatives in hysterics
though actually, here is their version of a meltdown. I take it back. You go.
SNIP
oasis
(49,410 posts)geek tragedy
(68,868 posts)Loki Liesmith
(4,602 posts)Because the probability only represents the confidence of the model. It doesn't tell you the real probability of a win in november because that's a one off event that hasn't happened yet. We can never know that number.
Wang's model is resistant to changes because it doesn't use averages in polls and doesn't estimate trends. It just looks at the medians of recent polls.
that said, it will move down tonight. My estimate a 3 point decline in Wang's number.
geek tragedy
(68,868 posts)If that assumption proves false (and it's certainly looking questionable these days), then his entire model begins to look iffy.
PEC shows Clinton as 72% chance of winning Nevada, even though the polls show Trump ahead.
Loki Liesmith
(4,602 posts)until we get data to update our priors.
If, for example, the polling averages all snap back in the next week or so (plausible given the oscillating trend in Trumps topline numbers at Huffington Post) then Wang's model LOOKS better than its competitors. Assuming that one of the criteria we want in a model is that is reliable as well as accurate. (anyone can be right the night before an election...but 2 months out?)
geek tragedy
(68,868 posts)Final result is that we got wiped out, and fell below Wang's "reasonably assured" number.
Loki Liesmith
(4,602 posts)Just not much depth of polling per state. Too many possible outcomes with impoverished data. Silver has had a bad run with mid terms a couple of times.
Not sure what i'm debating. Wang's model will adjust itself to the "correct" values if that's ultimately warranted. Silver's models have a fast time constant and Wang's models a slow one. We wont' know which was appropriate till the end.
Loki Liesmith
(4,602 posts)OH polls will move the median Ohio value.