2016 Postmortem
Related: About this forumOne last polling primer: Some pollsters are about to miss badly in Iowa.
Well, my flight is delayed, and I'm sitting here bored off my ass, so why not do one last one of these before voting starts.
As always, my standard disclaimer: I am a Clinton supporter, but I am also a statical analyst in real life. When I do these primers, I do then intending to be nonpartisan.
Anyway, tomorrow is a huge day. Voting starts, we get to finally start awarding delegates and we can take out GDP bickering to a whole new level. But this year, much more than any other recent election, I can say with certainty that some pollsters have done f'd up their models.
At the heart of polling is something very few people are aware exists and even fewer ever get to see. Before any polling is even done, the pollster has to establish something called population parameters. These are what will dictate how samples are weighted and polling is conducted. These include what demographic breakdowns should look like in the final results, and they're unique to each pollster. They're also assumptions, generally based on history and what the pollster "feels" is right.
This year, there are three assumptions that pollsters have to account for that will influence the final results:
% of voters who are women
% of voters who are college aged
% of voters who are first timers
Pollsters have a choice to make: do they use the historical average for these numbers, the numbers from 2008 only or make some other assumption based off their feeling?
Now it should be obvious that historical average favors Clinton, as more women traditionally caucus then men, college aged is 15% or so and first timers are 1/3 the caucus.
2008 favors Bernie, primarily due to first timers accounting for a whopping 60% of turn out.
And own assumptions could be just about anything.
So, what is about to happen is we're putting those who use historical averages to establish parameters versus those who don't to the test. And remember, a margin of error is only valid for the population parameters of that poll, not for the actual real population. Come Tuesday, there will be pollsters frantically changing assumptions going into NH.
JudyM
(29,265 posts)are more likely accurate?
Godhumor
(6,437 posts)They're also the safest and easiest assumptions to make.
The Des Moines Register (Selzer) poll is considered the best poll in Iowa. It uses a historical base with sensible tweaks. They also use verified registration lists for their sample, which strengthens accuracy tremendously.
Paulie
(8,462 posts)Verified voter lists still a strengthener?
At some point behavioral can't be modeled very well. We humans do so many predictably unpredictable things!
Godhumor
(6,437 posts)Not the whole thing.
CoffeeCat
(24,411 posts)Selzer is the "gold standard" of Iowa Polling and I have tremendous respect for her.
So, that is your understand too--that Selzer begins with verified voter registration lists as her sample from which to poll people?
Godhumor
(6,437 posts)Though the percentage may have changed this cycle.
CoffeeCat
(24,411 posts)Godhumor
(6,437 posts)And it would be random from the registration list as well.
JudyM
(29,265 posts)How much would you discount the results because just live phone method?
Bearing in mind it is the very latest poll...
POLL UPDATE
2016 Iowa Democratic Presidential Caucus - Sanders 49%, Clinton 46% (Quinnipiac 1/25-1/31)
Population 919 Likely Voters - Democrat
Margin of Error ±3.2 percentage points
Polling Method Live Phone
Source Quinnipiac
This poll asked respondents 1 question tracked by HuffPost Pollster. Read our FAQ.
1) 2016 Iowa Democratic Presidential Caucus
Asked of 919 likely voters - democrat
Hillary Clinton (D) 46%
Martin O'Malley (D) 3%
Bernie Sanders (D) 49%
Undecided 2%
Poll chart and latest estimates for 2016 Iowa Democratic Presidential Caucus »
virtualobserver
(8,760 posts)Also interesting to see if the snow will affect turnout, and which groups will be affected.
Godhumor
(6,437 posts)People who have caucused before are more likely to put up with inconveniences to get out the vote than those who are considering doing it for the first time.
JackRiddler
(24,979 posts)have more value as motivating devices than they do as predictive instruments. Otherwise it's a game of chance, closer to a roulette table than a football game. Tomorrow night someone's going to think they're smart because they happened to bet on the right numbers, and the punditariat will act like it's the new conventional wisdom applicable to all cases henceforth.
Bjornsdotter
(6,123 posts)...I'll be looking forward to your post caucus analysis.
Skwmom
(12,685 posts)1. Voting irregularities. Is the Clinton campaign above that - no.
2. The Corp Media, Democratic Establishment and opposing side has been working overtime to piss off the women's vote to increase turnout (and to portray Clinton as a victim to get the sympathy vote).
3. The app to help O'Malley.
DanTex
(20,709 posts)Which pollsters are making which turnout assumptions?
Godhumor
(6,437 posts)For the others, a pretty easy rule of thumb:
If Sanders is ahead, they're assuming 2008 or favorable own assumptions
If Clinton is slightly ahead, they're using historical or tight assumptions
If Clinton is really, really ahead they're probably using historicals with some overly tight assumptions on top of that.
DanTex
(20,709 posts)I work with data professionally too, but not in a social scientific way.
SidDithers
(44,228 posts)I've enjoyed your posts immensely over the past few months.
Sid
ismnotwasm
(41,999 posts)It will be interesting
Z_California
(650 posts)I didn't understand this part of the polling process before.
One thing that suggests 2016 may be more like 2008 than historical would be the huge crowds showing up for Sanders events - seems similar to the excitement in 2008. I was there in 1984 (Mondale/Hart/Jackson) and while there was interest, there wasn't the excitement as in the last couple elections.
craigmatic
(4,510 posts)I'm just going to watch and not make predictions.