Outsider's Question of KA Elections

[Sgt Raven]

KA? Please...

Really, I could possibly see PRK, but not KA.

[trashman]

Don't count Carly out. The polls are too inconsistent to be trusted.

[snip]

Turnout is key - and I don't think anyone knows for sure what it will be.

Valid point, certainly true...but I think that the chance of victory, in California, has much more to do with depressed Democratic turnout than increased Republican or conservative turnout.

--Neill

[Sgt Raven]

Valid point, certainly true...but I think it the chance of victory, in California, hase much more to do with depressed Democratic turnout than increased Republican or conservative turnout.

--Neill

Well Bradley lost primarily because of prop 15 was on the ballot and he came out against it.

[Window_Seat]

I'm with CL's predictions above with the exception that I still think Cooley can pull it out as it's a neck and neck race. Note that that was a tightening from a Cooley lead.

-Gene

And then be afraid, very afraid...

Erik.

[meg]

The Field poll was apparently done over 12 days and thus may not accurately reflect the current opinion of the voters. Note also that the margin of error to which they admit was +- 3% - which means the race may be substantially closer than they are saying (or it could be a blowout).

Rasmussen admits a +- 4% margin of error and is actually calling it a toss-up. Heck, Fiorina might actually win if the vote were today.

/Statistician_hat_on

The margin of error is affected by the sample size. Bigger sample size == smaller MOE. 3 or 4% as a MOE is pretty standard for election polling. That MOE can be gotten with a reasonable sized (read: not atrociously expensive) sample size, and yet is tight enough to call all but the closest races.

Believe it or not, it's actually problematic to shoot for *too* small a MOE: it's called oversampling and creates issues with the validity of the method of calculating said MOE. You always want some margin of error; allows for people to change their mind, previously undecideds to decide, people who intended to vote to wind up not voting and vice versa, etc...

/Statistician_hat_off

Oh hell, like the statistican hat is ever off.

[trashman]

Well Bradley lost primarily because of prop 15 was on the ballot and he came out against it.

That certainly is true, but I don't see anything on the ballot that Boxer has endorsed that makes me think she'll implode this weekend.

We'll see - I certainly hope Fiorina squeaks out a win - but we're in the final spin-cycle of this election: news outlets report that Boxer seems likely to win, which history has shown will dampen turnout for her opponent...

--Neill

[OleCuss]

/Statistician_hat_on

The margin of error is affected by the sample size. Bigger sample size == smaller MOE. 3 or 4% as a MOE is pretty standard for election polling. That MOE can be gotten with a reasonable sized (read: not atrociously expensive) sample size, and yet is tight enough to call all but the closest races.

Believe it or not, it's actually problematic to shoot for *too* small a MOE: it's called oversampling and creates issues with the validity of the method of calculating said MOE. You always want some margin of error; allows for people to change their mind, previously undecideds to decide, people who intended to vote to wind up not voting and vice versa, etc...

/Statistician_hat_off

Oh hell, like the statistician hat is ever off.

Thank you.

But to complicate things further, when it comes to election polling it is even more complex than sample size (as I strongly suspect you already know).

As I understand it, they typically choose their sampling based on various factors.

So you can survey the population in general on a particular issue or race - and you won't likely get anything very predictive of the outcome of a race.

If you survey Registered Voters you'll get a better idea of how a race is going.

If you survey Likely Voters you may get results that are very indicative of where things are going - or you may not. To do it well you have to take a look at a number of factors in choosing your sample of voters - things like the percentages of Democrats and Republicans and likely turnout. One can further look at age and certain other demographic information (youngsters may have the vote, but they don't vote as much as the oldsters).

The upshot is that when it comes to political polling, even if you choose a sample size large enough to know with great certainty what the general population (or the Registered Voters) (or the Likely Voters) thinks about an issue or wants done - but if your statistical model predicting who is actually going to vote is incorrect or is not accurately reflected in the sample you actually surveyed - it's still garbage.

Net effect is that you can have competent pollsters using different statistical models and polling on the same race in about the same time frame and get significantly different results.

Then, to make it worse, there are some of us who just generally won't talk to pollsters. So is your sample off because conservatives are inclined not to talk to you (or is it liberals who don't talk to pollsters?)?

On average, it is my understanding that Rasmussen has typically been better at this than most and I am thus inclined to put more stock in what they are saying. But this year feels strange enough to me that it may be that no one will have it right.

Oh, if they are saying that there is a 20 point difference between the candidates or positions on a Proposition - the predicted winner is likely to win. But at this time, I consider anything with spread of less than 10 points to have a potential to go either way. My gut says that the conservative turnout is underestimated to where conservative issues are going to do about 3-4 points better at the ballot box than they've done in polling (but it could actually be the opposite - my gut does not have a solid record for statistical genius).

But one thing I think is clear is that any incumbent who has not clearly gotten to above 50% in the polling is in trouble. That means Boxer is in trouble and Brown may be as well (not incumbent but as a current public office holder and former governor I think he's in statistical trouble).

FWIW from a non-statistician.