2016 Electoral College polling thread

[BugI02]

Why is it tweaking the data? If you want to know who is going to win a state, do you just randomly call 1000 people in the state? That poll wouldn't be very accurate because it wouldn't represent the true demographics. I think when you're only sampling 1000 or so, it's science mixed with an understanding of the electorate that determines that you're properly representing the people who will actually vote.

I know you know that there is a lot of solid mathematics behind extrapolating probability from a given sample size. Once you begin to weight your raw data you're on much shakier, more subjective ground relative to the math

 

[Cygnus]

Q: what was LA Times doing differently and who was running their poll compared to all the others?

Follows the same people over time and looks at how their views shift. They used 3k people and polled those same folks every so often.

 

[jamil]

I know you know that there is a lot of solid mathematics behind extrapolating probability from a given sample size. Once you begin to weight your raw data you're on much shakier, more subjective ground relative to the math

Right. That's the part that's where they use understanding of the electorate to do the weighting. It works when the electorate stays the same. And we saw what it does when the electorate doesn't follow Hoyle.

 

[printcraft]

Follows the same people over time and looks at how their views shift. They used 3k people and polled those same folks every so often.

Ah!

 

[hoosierdoc]

Follows the same people over time and looks at how their views shift. They used 3k people and polled those same folks every so often.

They polled dialy but only a subset

 

[T.Lex]

T. Lex, I think about something Nate Silver said a while back that kind of got lost in the noise. There are very few data points to work with vis a vis presidential race polling (to modern standards). The MoE has to be yuge and this particular election was an outlier. Not giving up on polling just yet, either

Yeah, and I think there was something to Trump's message resonating with union workers regardless who their bosses told them was the right candidate.

But I think once you start down that road you have an even bigger problem. Once you decide to tweak the raw data, how do you decide how much? You have many more options but even less data to guide you. Where do you get the statistical guidance? More polling?

I do think we'll see future polling trying for larger numbers, with more diverse sampling, to try to confirm trends.

Q: what was LA Times doing differently and who was running their poll compared to all the others?

Follows the same people over time and looks at how their views shift. They used 3k people and polled those same folks every so often.

They polled dialy but only a subset

So, my main problem with the LA poll is that it is a black box. We don't know who their sample is, how it was formed, or anything demographic about them. Then, it is hard to account for the randomness of the subset that they polled daily.

What we do know is that it did a better job of predicting the winner (but not necessarily the popular vote), unlike many/most of the more traditional polls. They may have found some sort of secret sauce that they won't share. We'll have to see how accurate it is on a go-forward basis.

pedoesta gave a list of WHO to target SPECIFICALLY.
If NOT a tactic why would podoesta try to manipulate the polls at all?

Let's break that down. Does he want to manipulate the polls to benefit his candidate? Absolutely. That's his job.

He sent a wish-list type request to either a single, or at the most a small pool of media that he considered friendly. That's pretty much his job, too - to leverage relationships to benefit his candidate.

We do not know if his request was honored. We do not know if it was treated like most of these kinds of requests that I've seen - with an eye roll and disdain. The people doing the polling understand that his job is to ask and their job is to dodge. Maintain the relationship, but not sacrifice credibility. (In retrospect, the entire polling industry took a hit.)

Clearly, his request was not to every media organization.

Could a single poll have been influenced by him? Sure. That's why it is not a good idea to trust any single poll. (Including the LA Times.)