2016 Postmortem
Related: About this forumA Primer in framing. Hillary was +6 in the ABC poll not the -4 reported in the press release
Disclaimer: I'm a statistical analyst. I do support Hillary, but I do these for more of an educational purposes than a partisan one. Yadda, yadda, yadda.
So, just a little while ago I lost it with the superPAC sponsored PPP debate poll for being nothing more than advertising for Clinton. I hope that has earned me a little leeway with some here, as I am about to go negative on a reported piece of very good news for Bernie from the pollster responsible for the Washington Post/ABC poll. Simply put, the pollster pulled a pretty shady trick in the press release to sell the story that Clinton lost 13 points of spread since the October poll. You can see it here:
http://abcnews.go.com/Politics/sanders-gop-steady-terrorism-worries-back-poll/story?id=35337895
And the money quote is this:
At 60-34 percent, Hillary Clinton is 26 percentage points ahead of Sanders, down from 39 points in October. He has gained significantly among registered Democrats and Democrat leaners under age 50, runs closely with Clinton among liberals and has closed the margin, to some extent, among nonwhites and women -- results that may give Clinton pause, if not palpitations.
------
Funny thing, though, those numbers, as reported by the pollster itself, weren't the actual poll numbers in October. What is reported in the press release is that Clinton was ahead 64 to 25 in October. However, in the actual poll from October she was ahead 54 to 23.
http://abcnews.go.com/Politics/clinton-rebounds-democratic-race-gaining-sanders-biden-alike/story?id=34580456
Yes, the pollster flipped a 6 point bump from October to November for Clinton into a four point loss.
But how did the pollster manage to get his own numbers wrong? He went back and reallocated Biden's supporters to their second choices. He also did that without stating in his November press release that he was doing so. I had to go into the October press release to figure out what he did.
This is not the correct way to report trending, at all. Second choice application may be fun for discussion purposes, but it cannot be used as a valid step in if someone drops out. There are many issues with why you can't just ignore your actual reported poll results by retconning what you previously showed. Let's touch on just one. In polling, the expectation is that people put very little thought into their second choices. As a result, second choice can actually benefit tremendously from name recognition. In other words, second choice answers tend to be, well, meaningless.
So why would the pollster decide to issue the press release with this framing (the actual poll for November uses the 54 not the 64 to report October results, by the way)? That is easy, to create a story--in this case by turning a positive poll for both candidates into a surge by one in particular. It is, in effect, dishonest.
The actual poll results tell the real story. Sanders picked up 11 points in a month by moving from 23 to 34, which is great. Hillary crested 60 for the first time by moving up 6 points from 54, herself. That is also great. The net change in spread is +5 to Sanders, but it didn't come at any expense to Hillary, who actually gained, as well.
Very different story than Hillary falling 4 points back to 60 while Sanders gained 9 for a net change of 13.
The poll is fine. The press release carried an agenda, which was achieved by some pretty dubious means.
Iliyah
(25,111 posts)karynnj
(59,504 posts)and that would be the appropriate comparison.
Clintons support for the nomination is more than double Sanders and triple the unannounced Bidens. Leaving Biden out of the equation, she has even more support, 64 percent, compared with 25 percent for Sanders, with others in the low single digits. Thats improved slightly for Clinton from a 56-28 percent race vs. Sanders in September.
They used the polls without Biden in their graph in the new link -- no error there.
Agnosticsherbet
(11,619 posts)karynnj
(59,504 posts)The numbers they took from October are the numbers without Biden.
Clintons support for the nomination is more than double Sanders and triple the unannounced Bidens. Leaving Biden out of the equation, she has even more support, 64 percent, compared with 25 percent for Sanders, with others in the low single digits. Thats improved slightly for Clinton from a 56-28 percent race vs. Sanders in September.
They used the polls without Biden in their graph in the new link -- no error there.
SunSeeker
(51,572 posts)delrem
(9,688 posts)I think the debate has just started.
NurseJackie
(42,862 posts)that the "M, dollar-sign, M" is in Hillary's camp. Looks like that's just another urban legend.
Godhumor
(6,437 posts)Just really bad form.
karynnj
(59,504 posts)Clintons support for the nomination is more than double Sanders and triple the unannounced Bidens. Leaving Biden out of the equation, she has even more support, 64 percent, compared with 25 percent for Sanders, with others in the low single digits. Thats improved slightly for Clinton from a 56-28 percent race vs. Sanders in September.
They used the polls without Biden in their graph in the new link -- no error there.
Godhumor
(6,437 posts)To remove Biden. My point is that the poll itself includes Biden and that, due to inherent issues with moving second choice to the front, it creates an incorrect picture. My other point is that the pollster flipped from the full poll as reported in October, which had Hillary at 54 to the total where Biden's supporters second choice was included at 64 points.
I stand 100% by my OP.
karynnj
(59,504 posts)There is no way that adjusting to remove Biden is a problem. Ignoring it actually just leads to a murky result. I worked for decades doing statistical analysis and operations research at Bell Labs -what ABC did was entirely reasonable. When Biden was no longer a choice, she immediately polled higher in other polls because more of the people preferring him went to HRC than to Bernie - just as the ABC without HRC poll indicated. It is a more accurate reflection of the relative strength of HRC and Bernie then ignoring the Biden numbers.
Godhumor
(6,437 posts)Because, again, of the bias in second choices. It is exactly why all other pollsters didn't run away from including Biden in its polls and notated the increase in support due to him not entering the race in their October polling.
I understand why they did what they did, I just vehemently disagree with the rationale.
Nice to meet another statistical analyst, by the way.
karynnj
(59,504 posts)There were many 2004 polls in 2003 that were with and without Al Gore. If the sample was constructed properly, there should be no bias based on this. (I do agree that if they created a strange hybrid that was who got more first plus second votes - it would be incredibly subject to game playing. Here, that is not what happened - they asked Biden people who they would vote for if Biden did not run.)
Godhumor
(6,437 posts)They don't reallocated support on a trending basis as people drop out. Even in the ABC November poll results they refer to the 54 point support from October and not the 64 point support given when second choice is assumed as first choice.
And there is bias in simply putting the second choice forward as first choice due to the known bias that people don't know enough alternatives to really understand who they would vote for if their chosen candidate dropped out.
The press release did not explain what they did and why; that is why I'm bugged by it.
karynnj
(59,504 posts)Godhumor
(6,437 posts)As in someone can say they'd support Hillary if Biden didn't run. When he doesn't announce the person follows the other candidates closer and actually decides to align with Bernie. Second choice answers have error associated with them if the first choice actually ends up being removed.
Godhumor
(6,437 posts)Funny how my post criticizing the HRC poll caught fire while this one hasn't...
Sheepshank
(12,504 posts)karynnj
(59,504 posts)Clintons support for the nomination is more than double Sanders and triple the unannounced Bidens. Leaving Biden out of the equation, she has even more support, 64 percent, compared with 25 percent for Sanders, with others in the low single digits. Thats improved slightly for Clinton from a 56-28 percent race vs. Sanders in September.
They used the polls without Biden in their graph in the new link -- no error there.