0 
1 
0 
The mailbag
Gerrymandering makes for interesting mail! Here are some excerpts from activists, a journalist, political scientist, and a few redistricters. ...
Service notice (RESOLVED)
I’ve been lying low. It’s a busy time here at Princeton: undergraduate advising, graduate teaching, and it’s midterms week! I figured since the Presidential race is basically over, the urgency of posting was reduced. Also, I’m working on a newspaper piece. Anyway, I will be back soon with more bloggy goodness.
For now, please ignore the Presidential calculations. The Huffington Post is using a new format for many-state poll drops from UPI/CVOTER and Ipsos/Reuters. This has gummed us up. No, Trump is not ahead in Colorado; that state is actually Clinton +5%. This artificially affects the EV count and Meta-Margin, though the win probability is still 99% since Clinton does not need Colorado. We’re working on fixing it.
10:04pm: A temporary fix is in place. The displayed calculations are correct.
1:14am: Walker Davis, PEC developer and troubleshooter, has fixed things now. All state polls from UPI and Ipsos are now going to the right place.
Topics:
I perfectly understand if you’re not interested, but Silver just posted an article about why his odds for Trump are so much higher than anyone else’s. He gives four main reasons, the biggest apparently in that he assumes errors in state polls will be correlated. Do you make that assumption? I couldn’t quite tell by reading your FAQ. Anyway, thought you might find it interesting to see his justification for being so cautious.
I think Sam accounts for correlated errors in his prediction model by using fat-tailed distributions. The question is just how fat the tails should be.
I wonder though, if Professor Wang would really be willing to take bets on Trump at 50:1 or 100:1 right now. I’m not saying 538 model is correct either, but the high confidence of the PEC model seems unwarranted given small number of elections to compare to.
Yes, though I would not take bets at the exact probability, because that would be a wash. The probability is 99% now, so I would take 50-1, which is equivalent to a 98% probability. To spell that out, that is $20 of someone else’s money against $1000 of my money.
Generally I might consider taking a bet at odds that were intermediate between mine and those of another public prognosticator: even odds against Norpoth, 20-1 against Silver or The Upshot, and so on. I would not make a bet with any of you guys mainly because I think recreational betting is a vice, but also because you can make up odds without recrimination.
Isn’t that essentially the position of an insurance company? It’s a bet you take if you have deep pockets. A large swath of the economy is built on that.
Mostly not interested, but I suppose I could write about it.
I get that u dont want us all piling on Silver.
But this creeping pandering isnt new.
Remember in 2012 he disappeared a post at the Times on why the mythology of dem oversampling in polls was junk analysis.
Yeah, he did that.
When people pass that article around, I’m torn between snarking on Silver and thinking it’s good for GOTV. On the other hand, the comments on John Scalzi’s article about how this campaign has affected him make me think that further obsessing about the possibility of a Trump win is actually doing people psychological harm.
Matt:
it is definitely doing a lot of persistant harm to the GOP base
it seems like Nate is compounding the harm– disbelief in polls, conspiracy theory, he couldnt even stand up to the over-representation/over-sampling myth in 2012
read this– i agree with barro 100% !!
http://uk.businessinsider.com/conservative-media-gop-trump-2016-10?r=US&IR=T
Silver is pandering to the GOP base for clicks–
isnt it better to just let the GOP base have full frontal math then let them believe the polls are “lying” or “underestimating” Trumps win chance?
As to the psychological impacts, it really says something that just the drop from 98-96% in the random drift probability is enough for a brief fight or flight reflex to kick in! That 96% isn’t quite enough
Slate announced that it will live blog projected election RESULTS while voting is taking place on Election day.
“From the moment that polls open on the East Coast, Slate readers will know as much about how many votes Hillary Clinton and Donald Trump have accumulated across the battleground states as Hillary and Donald and their teams do.”
Slate will use a combination of analytics and active tracking of turnout across preselected precincts to produce rolling projections of how many votes they have won as the ballots are cast.
http://www.slate.com/articles/news_and_politics/politics/2016/09/slate_and_votecastr_are_teaming_up_to_publish_real_time_projections_on_election.html
I think the two big questions remaining are:
What down-ticket price will the Republican Party be paying? A Democratic controlled Senate (likely); a flipped House (no longer unimaginable); loss of governorships or state legislatures (haven’t heard much about this yet)…
What is the GOP’s long-term clean-up costs after embracing Agent Orange?
Reportedly, the name “Trump” has been removed from Trump Hotels as a consequence of a 57 % reduction in bookings. Obviously his brand name has taken a huge hit. But it’s not so easy to remove “Republican” from “Republican Party” and pretend nothing has changed.
The Republicans don’t seem to be paying much of a price so far. Some of the Senate races seemed to have moved towards the Democrats, in particular I noticed Rubio is having problems despite that the national Democrats left the state (which doesn’t seem wise from the outside), and Ayotte & Heck seem to have paid a price. But overall, it doesn’t seem like there’s a uniform effect, or at least a big one. Trump’s effect seems to depend on the particular race.
But it’s not so easy to remove “Republican” from “Republican Party”
That’s what they did in 2010 with the “Tea Party”. There was a lot of denial early on that they were simply rebranded conservative Republicans.
Coattail effects have been real but weaker than in past cycles, so far. But in the past few days, the Clinton campaign seems to have switched to a concerted effort to push down-ticket candidates. Rallying in NH, for instance, when Hillary is in no real danger there–that’s for Maggie Hassan. It’ll be interesting to see what the effect is, if any.
Trump Hotels are remaining Trump. A new brand line targeting younger travelers (or perhaps at least those less enthused by gilded toilets) is opening as Scion. (similarly but on alarger scale, Marriott has over 30 brand lines, including, for instance, Aloft and Meridien).
Also, though I would love it to be true, the 57% booking plunge was specifically and solely for a niche booking app called Hipmunk. I’m guessing that — in a world of Trivago, Orbitz, Kayak, Travelocity, Expedia, Hotels.com, et al — Hipmunk is a very niche player with a very small data set and probably not a whole lotta users in the first place who skew more Trumpward than to W or Kimpton digs. Props, though, to Hipmunk, for getting their name out there big league by releasing this data.
It was fun to see the distribution starting to fuzz out on the high end because Texas had gone a little pink, but those days seem to be gone…
Yeah, Clinton got a couple bad polls. Looks like the closeness was an illusion. Of course, the bad polls are from UPI/CVOTER & Ipsos, which have remained at their levels.
But Georgia is down to a 2 point Trump lead! It’s a tossup!!!
Polls are one thing. Worse are the continuing shenanigans in many Republican-controlled states – including Texas:
http://www.politico.com/magazine/story/2016/10/the-real-voting-problem-in-the-2016-election-214386
Meanwhile, national polls look like Clinton’s lead might be increasing again. I was thinking we were in for some fractional tightening, just enough to get Joe Scarborough really excited.
A bunch of operators we haven’t seen before are starting to jump into the poll mix now that we’re getting into the last days. With luck it means that aggregation gets more accurate even if it generates more weird outliers for shock-poll news releases.
By the way, the Dems have released internal polling showing that people may be willing to vote for Dems downticket so Hillary has the means to end gridlock in DC:
https://t.co/cVJMMhGcK2
Interesting. I was wondering if a lot of #NeverTrump Republicans were going to vote for Hillary on a split ticket specifically TO create gridlock. Some Republican House and Senate candidates seem to be specifically trying to encourage that: they’re dissociating themselves from Trump but portraying themselves as the firebreak for Clinton administration liberalism.
When I hear the words internal polling I reach for my cognitive delete key. If the polling has shown the opposite, would they have released it?
I agree with bks in that desperate campaigns cite internal polling all the time before losing so why should we believe this now?
They are releasing the internal poll because some of the Democratic base is divided. Steve Schmidt said in MSNBC it is a mistake to focus on down ballot for Hillary. It is better to reduce turn out of the Republicans than to rile them up with Sanders and Warren. So the base is divided,
A: Should Hillary lock in the win of WH, and leave Senate and House to fly under the radar?
B: WH is done, locked in. Go for down ballot.
They are releasing internal polls to defend their move. It is meant for just partisan consumption and calm down the ruffled feathers.
Sam–can we help support the students who work on the PEC site in some way? I’m sure MANY of us would be glad to do something–could there be donation site set up at Princeton for us to send funds to?
I owe my sanity (what little there is left of it) to this site for the last six months!!
Marty
@Marlene Snyder // Oct 25, 2016 at 4:57 pm
(I was one of the PIGS once (Poor Indian Grad Students), this is my take, but Princeton might be different, I don’t know. )
I am not sure giving cash would sit well with Univ administrators and I am not sure how many hurdles the prof has to jump through to make it all kosher…
Most grad students would appreciate internship opportunities, job opening tips etc. Travel grants to help them attend conferences, donation of lab equipment etc would be welcomed by all grad schools. But again, even if there is money coming in for free, they have tons of rules and regulations to comply with.
Who is the mysterious Walker?
Walker Davis, web developer. He replaces Mark Tengi, who replaced PEC co-founder Andrew Ferguson. Without these Princeton students (two now alumni), PEC would not exist.
Thank you Princeton students and alumni for the outstanding work you do on this site.
Yes… I have to agree, thank you so much to the Princeton University Students and the entire staff at the Consortium for all of their amazing work!
I realized Walker must be one of the people doing the heavy lifting behind the scenes. Thanks Walker and many others who toil behind the scenes.
Trump has dropped ca. 3 % at electionbettingodds.com this morning. That’s a sizable change. Any ideas why? -I’m not seeing anything special in the news.
A one-touchdown lead is worth more with 14 minutes to play than with 15 minutes to play. The clock is running …
At this point, every day that passes is a significant loss of time remaining for a major change in the situation.
Similar to the narrowing range of the 1 sigma “strike zone” on the Median EV estimator as we inexorably march toward the pointy end of the triangle. 😉
Not only that, voting is well under way (I have already voted).
What accounts for the differences in the huffington pollster vs RCP averages? how did each compare as far as accuracy in last two elections?
Here is an example from today with respect to huffington pollster vs RCP averages in Iowa.
RCP IA Trump +3.7
Dem Moines Register 10/3-10/6 T+4
Loras 9/20-9/22 Tie
Quinnipiac 9/13-9/21 T+7
Huffington Post IA T+1 has 8 polls that are more recent than the Q poll RPC is currently using ranging from T+7 to C+4.
If you compare the two I think you would find a common theme of RCP being more selective in the polls they use in my opinion to put their thumb on the scale to slant more Republican.
A different example from much earlier in this election cycle was that RCP had their electoral map showing NY as dark blue but their Electoral College total did not include New York’s 29 EV’s. This miss match lasted over a month. I think RCP did this on purpose so that their readers would see that on their front page a EV count that was closer between the two parties than it really was even based on RCP own math.
I understand the difference in their methods but which site has shown more cred?
Thanks, Sam. Thanks, Walker. I know you do your best, even when time is even more scarce than usual.
House redistricting has resulted in a smaller no of democratic seats with 80% votes and large no of republican seats with 55% votes. This has an unintended consequences. If the votes swing 6%, democrats will see 100% seats and republicans will see 0% seats. This is similar to the collapse of Long term capital mgmt in terms of wrong assumptions about the probability distribution. Professor, do you have any thoughts on the likely outsized gains to democrats in house seats due to gerrymandering?
Interesting, the urban v rural divide portrayed on the map @ http://cookpolitical.com/house/charts/race-ratings is truly breathtaking. Compounded by partisan gerrymandering severly handicaps Dems attempt to take control of the House.
Many thanks to Walker Davis and all of the PEC team. I know from fist- hand experience that keeping up with format changes to third party web sites can be a Herculean task.
Not going to speak on Sam’s behalf but I’m pretty sure most of the optimally gerrymandered districts are actually 60% R instead of 55%. Hence, why many people think Dems need at least an 8-10% advantage in the generic ballot to swing the House.
I see Remington Research did a poll dump today heavily weighted towards Trump which brought things down. Do they poll often? I don’t even remember them from the past several elections, but I could just be overlooking them.
I looked at the crosstabs for their NC poll (Trump +3, but also the only trump lead in over a month in any poll) – they have Trump getting 20% of the black vote in the state. I just don’t think so…
RCP has them as a (R) pollster – but of course, fair to keep them as it all evens out. But that’s my guess for the small dip in the MM
Could poll aggregation web sites be targeted for manipulation by well timed data dumps? And is there any strategic advantage to be had by this, say for fund raising?
I am definitely not implying that in this case, but more as a hypothetical situation.
GQR just dropped a national house race poll with Republican’s up +1 but 23 percent undecided. 23 percent undecided? That has got to be next to useless.
Perhaps a technical issue. I see this from time to time. Not often.
Washington Post/Survey Monkey
Survey Monkey
Does the software catch this as a double count?
http://elections.huffingtonpost.com/pollster/2016-iowa-president-trump-vs-clinton
It would count those polls as coming from separate pollsters. Yuck.
The address for sending feedback to the Huffington Post pollster operation is pollster-feedback@huffingtonpost.com.
I did send them an email. I suggested at minimum moving the pollster name to the beginning and keeping the ‘/’ separator. Even I could write the code to truncate the results on that one. 😉
The relevant post in the google group is https://groups.google.com/forum/#!topic/pollster-users/CfDeLEmsQ1U
This gummed me up as well…
The right thinks that the ACA premiums are going to shake up the polls for the next two weeks. I don’t think there’s any issue that would move them enough for Trump.
Does anyone have the 2012 MM and/or EV chart for the end of October?
I couldn’t find a feedback link, so I am posting here. Feel free to delete after reading.
This page: http://election.princeton.edu/faq/ has a link to the histogram CSV file, http://election.princeton.edu/code/matlab/EV_histogram.csv . However that link sends me to the “Easy, tiger. This is a 404 page.”
Where can I find the current histogram data? Perusing the code directory, I found http://election.princeton.edu/code/data/EV_histogram.csv , which looks like it. In any case, you should consider fixing the link. Thank you for your most awesome work and site.
Also, shouldn’t EV_histogram.csv have 539 rows, for 0 to 538 votes? It only has 538 rows.
What can I say, I like my indices to be convenient. If it were Python it would be a different story. Anyway, when there’s a plausible 0 EV outcome, we can revisit this.