2016 Postmortem
Related: About this forumWhy does Clinton do better in some polls and Sanders in others? House effect and bad methodology.
House effect is where some pollsters consistently favor one candidate by more than the margin of error for the other contemporaneous polls.
For example, the most recent poll that had Clinton way up in Iowa was by Loras, which has a HUGE pro-Clinton house effect.
Here is how the race in Iowa has developed over the past year according to Loras polling:
Here is how the race in Iowa has developed over the past year according to EVERY OTHER LIVE PHONE pollster:
Clinton and her supporters set the bar of expectations based on Loras polling at their own risk (remember the fate of those who have won Iowa but fell short of expectations -- Harkin '92, Gephardt '88, etc.)
Also, all campaign season there has been an odd phenomenon that robo-call polls have consistently shown a large pro-Trump and pro-Clinton effect (I have not seen a convincing explanation for this, but the effect is well documented).
This explains why you see most traditional polls according to well-proven polling methods show Sanders leads in Iowa and New Hampshire with contemporaneous robo-call polls from Gravis and Monmouth that show Clinton ahead in Iowa and a tighter race in New Hampshire.
If you do nothing other than exclude robo-call polls from the Pollster aggregator, Sanders is ahead in Iowa and Sanders is comfortably up by double-digits in New Hampshire:
It does Clinton no favors to set her expectations in Iowa based on robo-call polls because, historically, falling short of expectations is almost worse than losing in Iowa.
JackRiddler
(24,979 posts)What did David Brock say today? How communist atheist is Sanders anyway? Clinton speaks to Goldman Sachs for charity! Vote for her or you are sexist, racist and stupid! Why do you hate Obama? You're supporting Trump!
Attorney in Texas
(3,373 posts)stillwaiting
(3,795 posts)Especially with our computerized voting.
Together they can pack a 1-2 punch.
Polling shifts results x% from real votes. Computerized voting further shifts results y% from bad polling data. Allowing for (x+y)% to be stolen from actual results. If they wanted to. Of course, I'm sure they don't want to.
Gman
(24,780 posts)No one reads polls better than him.
litlbilly
(2,227 posts)Attorney in Texas
(3,373 posts)Second, Nate Silver would be the first to tell you that his projection is nothing more than a current assessment of the current data. As new data comes in, the 538 projection changes. Under the 538 model, new polls are weighed more heavily than older polls. As a result, the 538 projection is currently influenced by the crazy Loras poll which ridiculously gives Clinton an implausible 29% lead. Between now and the caucus, this extreme outlier Loras poll will phase out of the 538 model and the 538 projection will tighten.
Clinton is the PAC-funded favorite and Sanders is the grassroots-funded underdog. Of course she's the favorite. Sanders cannot celebrate the "upset" in Iowa unless Clinton and her supporters keep pushing the "inevitable" theme.
Gman
(24,780 posts)Gman
(24,780 posts)I don't presume to tell Silver he's wrong. This sounds reasonable too. HRC 50. BS 40
http://projects.fivethirtyeight.com/election-2016/primary-forecast/iowa-democratic/
Attorney in Texas
(3,373 posts)favorite depending on the model according to today's current 538 projection which will certainly change numerous times between now and the February 9 primary).
I have never disputed that Silver projects Clinton as the favorite in Iowa (so does betsmart, predictwise, and most of the pundit class).
Gman
(24,780 posts)But want to keep it as close as possible. She'll probably win IA. Then go with her strengths in the future states.
uponit7771
(90,370 posts)JackRiddler
(24,979 posts)do you have a case to make, like the intelligent OP? If not, what are you doing here?
Attorney in Texas
(3,373 posts)will learn the truth soon enough.
CajunBlazer
(5,648 posts)The first think that you can do is use averages of the latest polls - that balances out poll methodology problems and the effects of less than perfect samples.
You can see for the latest Iowa poll average on RealClearPolitics.com which currently shows Hillary with a 7.2% lead:
Real Clear Politics - Iowa
By the way I view the both Loras college and the CNN/ORC poll as outliers which balance each other out.
Better yet you can go to the Five Thirty Eight website where it takes into consideration all of the "house affects" and other poll characteristics by weighting the polls base on their historic tendencies and accuracies and then combines all of the recent polls to produce the odds of each candidate wining.
You can find the Five Thirty Eight Iowa site here where it shows that Hillary has a 82% chance of winning in Iowa:
Five Thirty Eight - Iowa
Attorney in Texas
(3,373 posts)currently factors in the following polls to generate its idea of the current aggregate lead:
CBS News/YouGov (showing Sanders +1)
CNN/ORC (showing Sanders +8)
KBUR (showing Clinton +9)
Loras College (showing Clinton +29)
Which of those polls shows a result WAY outside of the margin of error for every other poll?
The Loras polling is ridiculous, and Loras has been consistently out of the mainstream in all of its prior Iowa polls, and Loras has been consistently out of the mainstream by universally showing Clinton's numbers as way higher than the other pollster's numbers. That is the definition of a house effect. This house effect is not eliminated by the RCP (or Pollster) aggregation so long as the out-of-the-mainstream result is included in the aggregation; instead, the crazy number is averaged in with the plausible numbers.
But don't get me wrong -- I am ecstatic about Hillary setting her expectations in Iowa based on the Loras poll. I LOVE that idea.