Unskewing the polls?

[Denny Crane]

As I have posted before, polling data are complicated things. There's a big difference between a poll of all americans vs. a poll of registered voters vs. a poll of likely voters.

The polling firms have very different algorithms for weighting the raw data to get their results. Rasmussen is the only firm I know of that weights based upon party affiliation, and they happen to be the most accurate poll of 2008.

On the matter of weighting polls, the Washington Examiner writes:

http://www.examiner.com/article/is-the-latest-washington-post-abc-poll-skewed-for-obama

The latest Washington Post/ABC poll released today shows the race between President Obama and Governor Romney tied at 47 percent each. The mainstream media typically skews these polls by sampling registered voters (which means less enthused supporters of the Democrats are counted even though they are less likely to vote) instead of likely voters, and Democrats are over-sampled.

The survey is skewed and an “un-skewed” analysis of their numbers showed Romney likely has a 50 to 43 lead right now. This difference of a skewed-result tie as opposed to an “un-skewed” 7 point lead for Romney has serious implications for all the conclusions drawn from the survey results.

In this survey, the sampling is based on only 24 percent responding as Republicans, while Democrats were 33 percent, Independents were 36 and the remaining 7 percent responded otherwise. No serious observer of American politics or pollster believes that Republicans make up only 24 percent of the population. Is it precisely this under-sampling of Republicans, and proportional over-sampling of Democrats, that skews this survey.

Rasmussen in its Sept. 1 poll wrote:

During August, 37.6% of Americans considered themselves Republicans. That’s up from 34.9% in July and 35.4% in June. It’s also the largest number of Republicans ever recorded by Rasmussen Report since monthly tracking began in November 2002.

So what happens to the polling data if you use a 37.6% figure as the republican make up of the electorate?

This site has taken the raw data and applied the 37.6% figure to the weighting methods of the polls:

http://www.unskewedpolls.com/

(Romney leads in all the poll by significant margins, average of +7.8, lowest is +3 in the Fox News poll)

And yeah, the site is clearly partisan right wing. Attack the messenger. Or be intellectually honest and curious and explain why their polling data is actually wrong.

 

[barfo]

As I have posted before, polling data are complicated things. There's a big difference between a poll of all americans vs. a poll of registered voters vs. a poll of likely voters.

The polling firms have very different algorithms for weighting the raw data to get their results. Rasmussen is the only firm I know of that weights based upon party affiliation, and they happen to be the most accurate poll of 2008.

On the matter of weighting polls, the Washington Examiner writes:

http://www.examiner.com/article/is-the-latest-washington-post-abc-poll-skewed-for-obama



Rasmussen in its Sept. 1 poll wrote:



So what happens to the polling data if you use a 37.6% figure as the republican make up of the electorate?

This site has taken the raw data and applied the 37.6% figure to the weighting methods of the polls:

http://www.unskewedpolls.com/

(Romney leads in all the poll by significant margins, average of +7.8, lowest is +3 in the Fox News poll)

And yeah, the site is clearly partisan right wing. Attack the messenger. Or be intellectually honest and curious and explain why their polling data is actually wrong.

So, was Rasmussen's poll of party affiliation 'unskewed', or are we just picking and choosing which poll to 'unskew' here?

If you do a similar 'unskewing' of the 2008 polls, what happens? Does Rasmussen still nail it? Or does the 'unskewed' poll in 2008 show McCain winning? Can't have it both ways, Denny. Either skewing is bad, or Rasmussen was right in 2008. Not both.

By the way, 37.6% republican is absurd - it would mean the electorate identifies much more with R's than they did at the time of the 2010 election. Not very likely.

barfo

 

[DaLincolnJones]

Denny, let the left keep thinking that they are sooo far ahead in all the polls. Hell, I want them over confident, I want them to keep thinking that this race is in the bag. I want that small percent of complacent Dems to not even bother to show up at the booth.

 

[crandc]

Nate Silver, who has a very good track record, provides a stat heavy analysis.

Methodology means a lot. For example, likely vs. registered voters. Also, some polls, like ABC, use only land line. About 25% of the population has only cell, and they are generally lean left voters.

A national poll can be very misleading because we don't have a national election. We have 51 elections where some count more than others. California, New York, Illinois, Pennsylvania and Texas are the biggest. Obama is taking CA, NY & Ill., Romney taking Texas. Pennsylvania officially swing. So you can say "both candidates at 47%" nationwide and it doesn't say a lot. Looking at the swing states, frankly, it's not good for Romney. If he gives up on Pennsylvania, which the legislature promised to him by blocking 750,000 voters, he has to win Ohio and Florida. Neither of which looks good. Winning Wyoming and Alaska can't make up for it.

 

[Denny Crane]

So, was Rasmussen's poll of party affiliation 'unskewed', or are we just picking and choosing which poll to 'unskew' here?

If you do a similar 'unskewing' of the 2008 polls, what happens? Does Rasmussen still nail it? Or does the 'unskewed' poll in 2008 show McCain winning? Can't have it both ways, Denny. Either skewing is bad, or Rasmussen was right in 2008. Not both.

By the way, 37.6% republican is absurd - it would mean the electorate identifies much more with R's than they did at the time of the 2010 election. Not very likely.

barfo

I would think you ask 1000 people, "are you republican or democrat" and not skew the results of that question whatsoever. And collect that over a large series of polls, so it isn't just 1000 people but a much bigger sample.

Rasmussen nailed the 2008 polls, where the number of registered republicans was, in fact, much lower.

So here's a CBS News / NY Times poll.

It reports 93% of democrats support Obama (5% don't). It reports 90% of republicans support Romney (7% don't), and 51% of the rest support Romney to 40% for Obama.

How is Obama in the lead? Think about it.

 

[Denny Crane]

BTW

http://www.rasmussenreports.com/pub.../partisan_trends/summary_of_party_affiliation

 

[Denny Crane]

Two Democratic pollsters discuss the skewing issue here:



Pat Caddell worked for McGovern in '72, Carter in '76 and '80, Gary Hart in '84, Biden in '88, and Jerry Brown in '92.

Douglas Schoen worked on numerous Democratic campaigns, and was a pollster for the Clinton white house. He worked for Hillary's senate campaign.

 

[The_Lillard_King]

NFL poll says Romney wins!

If Mitt Romney wins the election, you will have read it here first. Going in to last weekend's games, Romney led President Barack Obama four games in two. In other words, of the six rules that involve a game in the first two weeks of the season, four point to a Republican victory while two point to a Democratic victory

Three more rules were up for evaluation on Sunday:

The Eagles Rule: If Philadelphia scores at least one offensive touchdown per nine first downs in its second away game, the Democrat wins the White House. Otherwise, the Republican wins.
The Jets Rule: If the New York Jets win their second away game, the Republican wins the White House. Otherwise, the Democrat wins.
The Texans Rule: If the Houston Texans win their second away game, the Republican wins the White House. Otherwise, the Democrat wins.

All three point to a Romney victory: Philadelphia never scored a touchdown, and both the Jets and the Texans won. That brings the score, after Week 3, to 7-2 in the former Massachusetts governor's favor.

 

[BlazerCaravan]

I see your NFL and raise you a NASCAR... NASCAR viewers prefer Obama

http://www.csmonitor.com/USA/Politi...ney-with-NASCAR-voters-Another-straw-the-wind

 

[BrianFromWA]

I see your NFL and raise you a NASCAR... NASCAR viewers prefer Obama

http://www.csmonitor.com/USA/Politi...ney-with-NASCAR-voters-Another-straw-the-wind

Pshaw. Everyone knows that NASCAR only goes left.

 

[ppilot]

Nate Silver, who has a very good track record, provides a stat heavy analysis.

Methodology means a lot. For example, likely vs. registered voters. Also, some polls, like ABC, use only land line. About 25% of the population has only cell, and they are generally lean left voters.

A national poll can be very misleading because we don't have a national election. We have 51 elections where some count more than others. California, New York, Illinois, Pennsylvania and Texas are the biggest. Obama is taking CA, NY & Ill., Romney taking Texas. Pennsylvania officially swing. So you can say "both candidates at 47%" nationwide and it doesn't say a lot. Looking at the swing states, frankly, it's not good for Romney. If he gives up on Pennsylvania, which the legislature promised to him by blocking 750,000 voters, he has to win Ohio and Florida. Neither of which looks good. Winning Wyoming and Alaska can't make up for it.

Silver is definitely one of the most insightful and useful reads at this point in the campaign as nobody does analysis better than him.

 

[barfo]

I would think you ask 1000 people, "are you republican or democrat" and not skew the results of that question whatsoever.

You would think that, but do you KNOW that? Or are you just saying it's that way because that's what you think it should be?

Rasmussen nailed the 2008 polls, where the number of registered republicans was, in fact, much lower.

And you ignored my point. If you are claiming the polls are biased and must be 'unskewed', then we have to 'unskew' Rasmussen in 2008. Otherwise you are being inconsistent.

So here's a CBS News / NY Times poll.

It reports 93% of democrats support Obama (5% don't). It reports 90% of republicans support Romney (7% don't), and 51% of the rest support Romney to 40% for Obama.

How is Obama in the lead? Think about it.

Hard to say, since you didn't provide any link to the poll.

There's no doubt that polling is a complex business. There is quite a bit of doubt that either you or the blogger in your OP link know more about polling than the pollsters do, however.

barfo

 

[BlazerCaravan]

Silver is definitely one of the insightful and useful reads at this point in the campaign as nobody does analysis better than him.

And Denny was quoting his work back when it looked like Obama was fading.

 

[barfo]

Just by coincidence, I came across this fun skewering of the unskewing.

barfo

 

[julius]

Just by coincidence, I came across this fun skewering of the unskewing.

barfo

interesting read.

And from that read, this one was pretty funny too.

 

[Denny Crane]

You would think that, but do you KNOW that? Or are you just saying it's that way because that's what you think it should be?




And you ignored my point. If you are claiming the polls are biased and must be 'unskewed', then we have to 'unskew' Rasmussen in 2008. Otherwise you are being inconsistent.



Hard to say, since you didn't provide any link to the poll.

There's no doubt that polling is a complex business. There is quite a bit of doubt that either you or the blogger in your OP link know more about polling than the pollsters do, however.

barfo

Turns out Rasmussen's party allegiance data is based upon survey of hundreds of thousands of people.

You missed the point. Rasmussen's 2008 poll WAS unskewed, which is why it was the most accurate.

As far as what the pollsters know and do, check out the video I posted with an interview of two top Democratic pollsters and their view on all this.

 

[BlazerCaravan]

Good in 2008, not so much in 2004, 2006, or 2010...

http://fivethirtyeight.blogs.nytime...rate-quinnipiac-surveyusa-performed-strongly/

 

[barfo]

You missed the point. Rasmussen's 2008 poll WAS unskewed, which is why it was the most accurate.

Uhm, are you sure about that? I don't believe it. I don't believe any major pollster presents the numbers without any adjustment. To do so is to surrender to sampling bias.

As far as what the pollsters know and do, check out the video I posted with an interview of two top Democratic pollsters and their view on all this.

Yeah, I watched that, it isn't news. Different pollsters make different assumptions about things. Rasmussen assumes that everyone is a Republican just like him. So naturally you think he's great. Some pollsters have a similar bias towards Democrats. Some are all over the map. Some pollsters cheat. Some pollsters have bad methodology. Some pollsters go "wee wee wee" all the way home.

barfo

 

[Denny Crane]

Good in 2008, not so much in 2004, 2006, or 2010...

http://fivethirtyeight.blogs.nytime...rate-quinnipiac-surveyusa-performed-strongly/

Rasmussen’s polls — after a poor debut in 2000 in which they picked the wrong winner in 7 key states in that year’s Presidential race — nevertheless had performed quite strongly in in 2004 and 2006. And they were about average in 2008. But their polls were poor this year.

(That's from your own link)

 

[Denny Crane]

Uhm, are you sure about that? I don't believe it. I don't believe any major pollster presents the numbers without any adjustment. To do so is to surrender to sampling bias.



Yeah, I watched that, it isn't news. Different pollsters make different assumptions about things. Rasmussen assumes that everyone is a Republican just like him. So naturally you think he's great. Some pollsters have a similar bias towards Democrats. Some are all over the map. Some pollsters cheat. Some pollsters have bad methodology. Some pollsters go "wee wee wee" all the way home.

barfo

I don't think Rasmussen is great because he tilts republican. I think his methodology is sound and he's been quite accurate more than the other pollsters.

I'm having trouble figuring out how you don't get it.

Unskewed means "use today's ratio of republicans to democrats vs. using the exit poll results from 2008." You know that due to Democratic Party voter suppression that republicans didn't turn out that year.

 

[barfo]

I don't think Rasmussen is great because he tilts republican. I think his methodology is sound and he's been quite accurate more than the other pollsters.

His final poll for president in 2008 was good. He sucked in 2010 and was average in 2008 for non-presidential races.

Unskewed means "use today's ratio of republicans to democrats vs. using the exit poll results from 2008."

Well, that's not what unskewed means, but maybe that's what it means to you. In any case, that's not what's going on here. It's more like "use the best case number for Republicans, which is Rasmussen's, and apply it to other pollster's polls". Truth is, you don't know what today's ratio of R vs. D is. You like Rasmussen's number because his is highest. And there is a serious lack of science in taking the raw numbers from a poll and applying only that weighting.

You know that due to Democratic Party voter suppression that republicans didn't turn out that year.

I didn't hear about that. Do tell!

barfo

 

[Denny Crane]

I like Rasmussen's number for republicans and democrats because it fits the results we've seen (for the most part) and the obscure fact that he's surved hundreds of thousands of people to arrive at his number.

What results have we seen? Near 50-50 elections in 2004, 2000, 1996, 1992...

As far as taking raw numbers from the polls and applying a truer ratio of democrats to republicans goes, that's not what they're doing.

 

[barfo]

I like Rasmussen's number for republicans and democrats because it fits the results we've seen (for the most part) and the obscure fact that he's surved hundreds of thousands of people to arrive at his number.

What results have we seen? Near 50-50 elections in 2004, 2000, 1996, 1992...

Yes... but there are independents too.

As far as taking raw numbers from the polls and applying a truer ratio of democrats to republicans goes, that's not what they're doing.

Depends on who 'they' is. The unskewing people, that's exactly what they are doing, except for the truer part.

barfo

 

[MARIS61]

Nate Silver, who has a very good track record, provides a stat heavy analysis.

Methodology means a lot. For example, likely vs. registered voters. Also, some polls, like ABC, use only land line. About 25% of the population has only cell, and they are generally lean left voters.

A national poll can be very misleading because we don't have a national election. We have 51 elections where some count more than others. California, New York, Illinois, Pennsylvania and Texas are the biggest. Obama is taking CA, NY & Ill., Romney taking Texas. Pennsylvania officially swing. So you can say "both candidates at 47%" nationwide and it doesn't say a lot. Looking at the swing states, frankly, it's not good for Romney. If he gives up on Pennsylvania, which the legislature promised to him by blocking 750,000 voters, he has to win Ohio and Florida. Neither of which looks good. Winning Wyoming and Alaska can't make up for it.

Land lines are owned mainly by seniors or people who need it for internet access.

 

[barfo]

Land lines are owned mainly by seniors or people who need it for internet access.

Even an ancient person like me doesn't have a land line anymore. A poll that depends upon land lines is going to have very serious sampling bias. Now, it's possible to correct for that... but it's a fact that the polls that include cell phones get different results (even after corrections) than the polls that don't.

barfo

 

[Denny Crane]

Yes... but there are independents too.



Depends on who 'they' is. The unskewing people, that's exactly what they are doing, except for the truer part.

barfo

So, 40% democrats, 20% republicans, 40% independents and 3/4th of the independents vote republican? (I don't think so).

The unskewing people are not taking raw numbers from the polls and manipulating those in any way. They're taking the final results as manipulated by the pollsters (skewed) and removing the big oversampling of democrats from the weighting.

 

[Denny Crane]

Land lines are owned mainly by seniors or people who need it for internet access.

Or by the parents of college grads who can't find a job and have to live at home with their parents.

 

[barfo]

So, 40% democrats, 20% republicans, 40% independents and 3/4th of the independents vote republican? (I don't think so).

I don't think so either. Where did you get 40/20/40?

The unskewing people are not taking raw numbers from the polls and manipulating those in any way. They're taking the final results as manipulated by the pollsters (skewed) and removing the big oversampling of democrats from the weighting.

That's not accurate. They are imposing Rasmussen's very high percentage of R's on the data. It's totally manipulating the data, in exactly the same sense as the pollsters manipulated the data, but done purely for political purposes. Even Rasmussen, while biased, has some science; this guy is just trying to make the numbers look good for Romney, science be damned.

barfo

 

[BlazerCaravan]

Rasmussen’s polls — after a poor debut in 2000 in which they picked the wrong winner in 7 key states in that year’s Presidential race — nevertheless had performed quite strongly in in 2004 and 2006. And they were about average in 2008. But their polls were poor this year.

(That's from your own link)

My bad; I was posting from memory and my phone.

 

[Denny Crane]

I don't think so either. Where did you get 40/20/40?

From the actual presidential election results. Even Bob Dole got 41% of the vote, and that was with a strong independent 3rd party candidate giving independent voters a solid alternative choice.

That's not accurate. They are imposing Rasmussen's very high percentage of R's on the data. It's totally manipulating the data, in exactly the same sense as the pollsters manipulated the data, but done purely for political purposes. Even Rasmussen, while biased, has some science; this guy is just trying to make the numbers look good for Romney, science be damned.

barfo

You have it backwards. All the polls but Rasmussen are biased if they're weighted as if there are just 24% republicans. Science be damned, indeed.

They're also not manipulating the data, they're using the pollster's final data and all the science that goes into it.

Poll says 93% of democrats support Obama and 90% of republicans support Romney. Achieved via the poll's methodology, science, weights, etc.

For example, take 1000 people. 37.6% will be republican, or 376, but only 90% will vote for Romney. That's 338 votes. 33.3% will be democrats or 333, but only 93% will vote for Obama. That's 310 votes. Of the rest, (1000 - 376 - 338 = 286), 51% vote for Romney, 40% for Obama. 51% of 286 is 146. 40% of 286 is 114.

So Obama gets 114+310 = 424, and Romney gets 146+338 = 484. 6 point advantage Romney.

I mentioned voter turnout suppression in a previous post. I live in California. Though I'm going to vote for Gary Johnson, my vote otherwise would not matter because Obama's clearly going to win the state by a huge margin. Republicans may as well stay at home. Hell, democrats may as well stay at home, too. By reporting bad polling numbers for Romney, don't you think many republicans will figure it's not going to matter, so they won't go vote?

So 25% of republicans did not show up at the polls in 2008. Bush fatigue, overwhelming media failure to look at Obama's resume, McCain not so popular a candidate with his own party, big money discrepancy, etc.

When they did turn out at the polls in 2010, it was a miserable election for Democrats. What pollsters other than Rasmussen do is they are using the exit polling data from 2008 in their weighting formulae. That exit polling data reflects the 25% who did not show up at the polls, and results in a massive undersampling of republican voters in their polls.

The media is a 24/7 Obama commercial, for the most part. Reporting the election is over, may as well stay home republicans, is a form of voter suppression, no?

I should also note that I'm not seeing a flaw in the methodology. But you can't use Rasmussen's national party affiliation data to unskew state polls.