So Polls Really Are Skewed.....and This is Exactly How Much.

[WelfareQueen]

This is an excellent and impartial analysis by Nate Silver on historic trends in polling data State by State.

Bottom line: It looks like a worse night for Dems than the currents polls would indicate.

As you will see....some States polling data skews heavily Democrat (Alaska +7.2%)....some skew heavily to the GOP (Hawaii GOP + 10.8%).


But here is the key issue....Every State in which there are tight Senate Races...the polls historical have skewed in favor the the Democrats. The only exceptions are Colorado and Iowa.


So Kansas...Kentucky...Louisiana....Georgia...North Carolina...New Hampshire....all polling data for many elections cycles have skewed in favor of the Democrats. Most of those States the Dem bias is around 1.5%. A few States (North Carolina and New Hampshire it's as little as .2%

We're talking a difference of around 50,000 votes in many of these States which is significant.

Here is the data.

 

[Zander]

What a glorious waste of time......

 

[Mac1958]

.

I dunno.

I remember hearing this kind of stuff from the GOP in 2012.

.

 

[WelfareQueen]

.

I dunno.

I remember hearing this kind of stuff from the GOP in 2012.

.

Mac....The reality is no poll is 100% accurate. I wasn't one of the unskewed poll nutters in 2012. But Nate Silver is a very respected guy. Most of the States where there are tight Senate races have historically tended to skew in favor of the Dems. This is not by a lot. We're talking about 1.5%.

In a State where 3,000,000 votes are cast that means roughly a 50,000 vote difference one way or another.

In close elections that can mean a lot.

 

[Statistikhengst]

This is an excellent and impartial analysis by Nate Silver on historic trends in polling data State by State.

Bottom line: It looks like a worse night for Dems than the currents polls would indicate.

As you will see....some States polling data skews heavily Democrat (Alaska +7.2%)....some skew heavily to the GOP (Hawaii GOP + 10.8%).


But here is the key issue....Every State in which there are tight Senate Races...the polls historical have skewed in favor the the Democrats. The only exceptions are Colorado and Iowa.


So Kansas...Kentucky...Louisiana....Georgia...North Carolina...New Hampshire....all polling data for many elections cycles have skewed in favor of the Democrats. Most of those States the Dem bias is around 1.5%. A few States (North Carolina and New Hampshire it's as little as .2%

We're talking a difference of around 50,000 votes in many of these States which is significant.

Here is the data.

You were kind enough to send me this also in PM and so I am going to repeat my response here, a factual argument:

I am aware of the work. And I already read it. Thanks for making this thread.

However, I have some strong disagreements with part of Nate's research in this case and will show you my logic.

First, I disagree with him using a three week time frame. The end-polling, and only the end-polling, should be used to compare accuracy with actual election results, for the very same reason that many voters want to wait to the last second to make a decision: the right to be able to change their mind. Pollsters who show what we could now interpret as "skewed" results 3 weeks before an election may or may not have done so, for we know how fickle the American electorate can be. The only way to be fair about this is compare the end-polls, meaning the very last poll conducted by a pollster for a specific race, and that usually happens in the last week, maximum the last 10 days. In fact, the majority of end-polls happen within the last 5 days.

Fact is that in the last 4 cycles (2006, 2008, 2010 and 2012), the MATHEMATICAL bias based on end-polls, as it should be, has been decidedly to the Right, and more so in the SW of our Union. There have been some very weird, almost unexplainable exceptions to this, for instance, Montana in the 2012 presidential (where even Rasmussen was 9 points to the LEFT) or North Dakota in the 2008 presidential election, but those are small blips on a very large, very visible screen.

This is definitely one point where Nate Silver and I will not completely agree, for I strongly disagree with his time-frame.

Not only that, going back to 1998 also makes no sense at all: there are very few pollsters from 1998 who are still putting out a lot of polls, or any at all. There was a large changeover in pollsters starting in 2003 with Rasmussen and in 2014, of the five pollsters who produce most of the public polling, four of them did not exist in 1998. And virtually all of the partisan pollsters who operate for the one party or the other are all very new on the scene, most of them since 2008. And the oldest firm still putting out results, Gallup, is the most discredited of them all, based on very terrible polling from 2010 and 2012.

Also, comparing poll bias (and I mean, mathematical poll bias) between mid-term cycles and presidential cycles is also probably not such a great idea, either, for we all know that two different types of electorates appear for those two very different types of elections.

My suggestion to you and myself is to wait out the results and then we will know.

One thing is for sure: the generic aggregate in 2014 is very, very different than it was in 2010, although virtually the same pollsters have conducted generics in both of those mid-term cycles.

I am not saying that Nate is wrong, per se. Mathematical poll bias exists. I am saying that I disagree with a cornerstone of his methodology in this case.

When the final cavasses are in, we will all know how good or bad the polls were.

 

[Statistikhengst]

.

I dunno.

I remember hearing this kind of stuff from the GOP in 2012.

.

Mac....The reality is no poll is 100% accurate. I wasn't one of the unskewed poll nutters in 2012. But Nate Silver is a very respected guy. Most of the States where there are tight Senate races have historically tended to skew in favor of the Dems. This is not by a lot. We're talking about 1.5%.

In a State where 3,000,000 votes are cast that means roughly a 50,000 vote difference one way or another.

In close elections that can mean a lot.

That statement is absolutely spot-on the mark.

This is why the aggregate is usually far more close to "the truth" than one individual poll. That being said, virtually every pollster, both from the Left and the Right, had a verifiable mathematical bias to the RIGHT in both the 2010 mid-terms and the 2012 presidential election AND most of the senatorial elections of 2012.

The posting above this one will explain the difference, for a real comparison can only be made with end-polls to actual results. A three week time frame is, well, bullshit.

 

[asaratis]

I spoke to God...it's gonna be a slaughter...

 

[Statistikhengst]

I spoke to God...it's gonna be a slaughter...

Did he tell you the name of his chosen Shochet as well?

Hmmmmm???