So, what happened to the Romney campaign?

[gebobs]

What polls specifically? National, State, or Local level? Who were they done by? Professional pollsters or news agencies?

I was only concerned with the presidential race. State and local races here in Georgia are easy to predict. Candidates typically run unopposed.

Accuracy is measured in a number of ways, simply saying Obama was going to win doesn't mean they were accurate from a Political Science stand point, which is the standard they are supposed to be judged at.

Great. If you're asking me, as a statistical professional well-versed in all of this, if I did an in-depth analysis, then the answer is no. As I said, it was my impression.

[Historian]

[gebobs]

For example let's take Ohio;
Obama won by +1.9

Only three polls that were conducted after 10/30 had Obama winning and by two or less. They were Columbus Dispatch, Gravis, and Univ of Cincy. The other five that had Obama winning were 3 or more, and Rasmussen had them tied. So out of 9 polls, six were not correct (and only one was within the margin of error). That's not accuracy.

First of all, you obviously know that election polls are imperfect samples. They are samples of what people report they will do in the future, not what they actually do at the time. As such, there is an essential disconnect right off the bat. Then there are the sampling issues. You have to try and match today what you think the demographics of the population at the polls will be in the future. It's an woefully imperfect science. Just ask Thomas Dewey.

Anyhoo...what was the margin of error in each? How were the polls trending?

Is it your gambit that the polls are kludged to try and be a kind of self-fulfilled prophecy?

[Historian]

[DraftBoy]

First of all, you obviously know that election polls are imperfect samples. They are samples of what people report they will do in the future, not what they actually do at the time. As such, there is an essential disconnect right off the bat. Then there are the sampling issues. You have to try and match today what you think the demographics of the population at the polls will be in the future. It's an woefully imperfect science. Just ask Thomas Dewey.

Anyhoo...what was the margin of error in each? How were the polls trending?

Is it your gambit that the polls are kludged to try and be a kind of self-fulfilled prophecy?

Absolutely, they all should be around +/- 1, unless its a ****ty poll.

No it is not.

[gebobs]

Absolutely, they all should be around +/- 1, unless its a ****ty poll.

Not quite. As I said, unlike say sampling red balls from a bag of red and white balls, an election poll's sample and the population it intends to predict are two different things.

The sample determines the margin of error. Not all polls can afford the same samples. Take a look at the sample sizes. Some are larger than others. You can qualify one or the other as ****ty, but they do publish the margins of error for a reason. Take for example the Washington Post/ABC poll from 11/4. It's as well funded as any. They sampled 3205 people among whom they use the responses from 2345 likely voters. From this, they try to predict the national population of 120 million voters with its myriad intricacies.

So what's the margin of error for that poll? ± 2.5%

And what does that mean exactly? Does it mean that the election results should likely fall within this range? Well, technically, no. It means that if you had called everyone up in the nation during the actual poll and counted the results of likely voters, that those results would likely fall within that range. Two different things. Though less likely, the actual results could fall outside that range. The demarcation point, the -2.5% on the one side and the +2.5% on the other, are not lines in the sand. They just indicate that point on probability distribution where the likely results fall within one standard deviation of the mean. Technically.

Less technically, a political poll is like trying to tell the weather tomorrow by looking at a photo of your backyard the day before.

Your ±1% bar is not reasonable.

[DraftBoy]

Not quite. As I said, unlike say sampling red balls from a bag of red and white balls, an election poll's sample and the population it intends to predict are two different things.

The sample determines the margin of error. Not all polls can afford the same samples. Take a look at the sample sizes. Some are larger than others. You can qualify one or the other as ****ty, but they do publish the margins of error for a reason. Take for example the Washington Post/ABC poll from 11/4. It's as well funded as any. They sampled 3205 people among whom they use the responses from 2345 likely voters. From this, they try to predict the national population of 120 million voters with its myriad intricacies.

So what's the margin of error for that poll? ± 2.5%

And what does that mean exactly? Does it mean that the election results should likely fall within this range? Well, technically, no. It means that if you had called everyone up in the nation during the actual poll and counted the results of likely voters, that those results would likely fall within that range. Two different things. Though less likely, the actual results could fall outside that range. The demarcation point, the -2.5% on the one side and the +2.5% on the other, are not lines in the sand. They just indicate that point on probability distribution where the likely results fall within one standard deviation of the mean. Technically.

Less technically, a political poll is like trying to tell the weather tomorrow by looking at a photo of your backyard the day before.

Your ±1% bar is not reasonable.

See your point on margin of error issue.

[Historian]

[Night Train]

They ran Romney for President...that's what happened.