A Few Perspectives on Polls

Shout Bits Blog ^ | 09/11/2012 | Shout Bits

[Shout Bits]

The Old Media story this week is that Pres. Obama is well ahead of Gov. Romney in national polls, which is not untrue. However, not all polls are created the same, and as David Axelrod can attest, they are not objective truth. As many candidates who are about to lose say, 'the only poll that matters is the one at the ballot box,' but statistical polls are wonderful grist for the news. With the incentive to hype numbers, Shout Bits offers a few perspectives on polling.

Every poll has a margin of error, which in no way means what it says. Margin of error is a specific term used by statisticians to state the significance of a sample size relative to the general population. Left unsaid in poll reporting is the confidence in the margin of error – usually 95%. For example consider a poll with a margin of error of +- 3%. The poll's margin of error is saying that if the exact same question were asked to 100 different random samples of people selected in the same way, five of those samples would show a result outside of the +- 3% range. In short, margin of error has next to nothing to do with the accuracy or usefulness of a poll.

The margin of error also has nothing to do with the validity of the question being asked, or any errors in the selection of those being polled. In these ways, polling is an art, not a science. Most polling is done by phone and to people with old fashioned land lines because cell phone numbers are not published. A small, but significant number of voters do not have any kind of phone, and they skew Democrat heavily. Likewise, a growing number of people only have cell phones and are harder to reach, a group of younger people who also skew Democrat. Pollsters adjust their raw data to reflect this known GOP bias in their results. No doubt anyone but partisans like PPP makes a good faith effort to understand this selection bias and accurately adjust, but these adjustments vary from firm to firm.

Another polling bias is in the sample selection. Most pollsters report on the views of registered voters, which is a bias in favor of Dems. Dems and independents together greatly outnumber GOP registered voters, but their turnout is usually lower. This season, GOP enthusiasm is much higher than Dem, and independents in some demographics are leaning Romney. Therefore, measuring registered voters introduces a Dem bias. Rasmussen measures self-identified "likely" voters, and their numbers show a much tighter race. Rasmussen has proven to be the most accurate of the pollsters over the recent election cycle.

Even the wording of a poll affects the results. 'Do you favor repeal of Obamacare?' and 'Are you opposed to Obamacare?' will generate different answers from the same polling sample. For these reasons, there is always room to question tight polling. The Daily Caller suggested that when Obama advisor David Axelrod did not like a Gallup poll, he pressured them to change their bias adjustments in favor of Obama. Much as with S&P's downgrade of US Treasury debt, the Justice Department had uncanny timing in its later lawsuit against the company that had crossed cocktail swords with the Administration.

Is Romney in trouble? Very much so. Post DNC polling showed support increasing for Obamcare, so this weekend, Romney flip-flip-flip-flopped back to supporting key elements of Romney/Obamacare such as forcing private companies to cover people who wait until they are sick to apply for insurance. Multiple flopping on core issues stinks of desperation, Governor. Lay-followers of polling should know its limitations and biases. Too bad weather vane politicians like Romney are incapable of the same.

Shout Bits can be found on Facebook: http://www.facebook.com/#!/ShoutBits

[Vaduz]

Support increasing for Obamcare.
CNN,ABC and washington post bustrd when the rig own polls,see Breitbart.

MSM still up to same old BS.Obama pimping is a free fall.

[frogjerk]

Article was good until it lied at the end about what Romney said/did this weekend.