2016 Postmortem
Related: About this forumMy first poll post in forever! The race has been insanely stable from Oct on (No movement at all)
As always, statistical analyst in real life, blah, blah, blah.
Anyway, as much fun as Pollster and RCP are to look at every time a new poll comes in, there is still a significant amount of noise when it comes to real long term trends. You might think Pollster gives a suite of tools to correct for this, but all most people do is either render the data meaningless by being overly selective, by under-loading the Pollster regression with too few data points or by using more or less smoothing without any real understanding of whether that is useful or not. In other words, all they do is create more noise while thinking they're reducing it.
Now, trending is an interesting topic. The time frames you use will dictate how noisy your roll up will look (Too short and your trends will be all over the place, too long and real movement gets lost).
So I decided to trend month over month. I think that is short enough that real events (Biden, the Surge, etc) are captured as changing opinion while being long enough that most wobble is eliminated.
I used RCP data for this, and I will admit right upfront I would have done this slightly different if I were on my work computer instead of my phone. My preference would be to split out the sample for each poll and add up the total "votes" each candidate gets for the month, as this would prevent weighing all polls, regardless of sample size, the same. However, this data isn't easily accessible at RCP, and, since, again, I did this on my phone, I ended up just averaging the poll percentages over a month (You might ask why I didn't do this with Pollster. The rason is primarily that RCP is a lot more selective with what polls are included in its aggregate. Since I'm doing a straight average and not a regression, I don't want to bother with untested polls.). Finally, I defined which month a poll belonged to by the month it finished polling not the month it started (So a 9/29 - 10/2 poll would be in October).
So the numbers are:
Month HRC - BS
Jun 62 - 13
Jul 57 - 18
Aug 51 - 23
Sep 42 - 26
Oct 48 - 26
Nov 56 - 32
Dec 55 - 31
You can clearly see the Bernie surge and the effect of Biden both before and after not declaring. And you might be wondering why I said the race is stable considering the numbers have changed each month. Well, the answer to that lies in the spread:
Month HRC's lead
Jun +49
Jul +39
Aug +28
Sep +16
Oct +22
Nov +24
Dec +24
Since Biden and Benghazi committee, Clinton has maintained a 22 or 24 point lead every month. The simple fact is that this race has not had any kind of meaningful change since the Biden fade. Certainly, the debates have done exactly jackshit to move the needle, as have the myriad daily GAME CHANGERS we read about here.
The race hasn't changed. And my guess is that it won't anytime before Iowa.
daleanime
(17,796 posts)to losing more ground in congress and if we lucky(?) barely holding on to the White House.
So excuse me for hoping that you are wrong.
MeNMyVolt
(1,095 posts)And you've been an incredible resource on this board, my dear Godhumor.
Godhumor
(6,437 posts)Sanders does not win with steady.
HereSince1628
(36,063 posts)that could be seen as reflecting an important change. That wouldn't fit the narrative, at all.
Godhumor
(6,437 posts)Yeah, no it doesn't matter a bit. Not until every other poll shows that same movement.
That poll is included in December, by the way.
HereSince1628
(36,063 posts)I understand.
dsc
(52,166 posts)DemocratSinceBirth
(99,710 posts)Overall, Clinton tops Sanders among registered voters who are Democrats and Democratic-leaning independents 50% to 34%. That's a slightly tighter margin than in late-November, when Clinton led 58% to 30% over Sanders:
But those overall results mask a shift back toward Clinton following the Democratic debate on Saturday night. In interviews conducted before the debate, Sanders ran closer to Clinton, with 37% support to Clinton's 45%. Among those interviewed after the debate, Clinton's lead grew to 60% vs. Sanders' 27%.
http://www.cnn.com/specials/politics/national-politics
It is reasonable to infer the non-white demographics moved accordingly post debate.
Alfresco
(1,698 posts)NCTraveler
(30,481 posts)The party is going to unify behind Clinton even more than they currently have. I know that seems out there considering how commanding her lead is, but the writing is on the wall. Sanders campaign sees it clear as day.
mmonk
(52,589 posts)but we'll still get a better clue once it all starts. Polls are not as reliable these days due to differences in technology and venues.
NCTraveler
(30,481 posts)Great op.
PotatoChip
(3,186 posts)I came upon it when searching for something unrelated, but found it helpful in better understanding how to look at these polls.
Still, I'd like to know if (in your professional opinion) you would consider the info to be accurate? If not, why not? For example, is he putting too much emphasis on some details over others, or is this a pretty good summary overall?
A PRIMER ON PRE-ELECTION POLLS:
Why Different Election Polls Sometimes Have Different Results
Cliff Zukin,
Professor of Public Policy and Political Science Bloustein School of Planning and Policy
& Eagleton Institute of Politics, Rutgers University
The views expressed here are solely those of the author
Primary Election Version
December, 2015
INTRODUCTION
Election polls are a special breed among public opinion surveys. They call for more judgmentsthe art rather than the science of the crafton the part of the pollster than other types of polls. And this brings into play a host of other reasons why the estimatesof well-established and well done pre-election polls may differ from one another, even when these polls are conducted at a similar point in time.
Also, the polling landscape has changed dramatically, even as recently as two presidential elections ago in 2008. Because of the growth of cell phones, declining response rates and increasing difficulty in determining likely voters, I have argued here that election polling has gotten harder to do well, and that 2016 will present election polls with a very challenging environment. This primer is meant to be a guide for journalists, academics, and anyone closely following polls in the 2016 election.
http://www.aapor.org/AAPORKentico/Standards-Ethics/AAPOR-Code-of-Ethics/Election-Polling-AAPOR-2015-primary_cz120215-FINAL.aspx
He goes into the subjects of:
-SAMPLING AND REAL SAMPLING ERROR
-MODES OF SURVEYS
a)Telephone Surveys
b)Cell Phones
c)IVR SurveysRobo Calling
d)Internet/On-line Polls
-TIMING AND FIELD PROCEDURES
-QUESTION AND WORD ORDERING
-POST-SURVEY ADJUSTMENTS
a)Weighting
b)Likely (Probable) Voters