Why You Must Question The Pollsters, The Media And The Republicans!

DEMOCRATS: Remember The Media, Pollsters And Republicans In 2012 That Romney Was Going To Win! Remember Eric Cantor Was Going To Win The Primary By 35% And He Lost By 25%! Do Your Part And Vote And We Will Maintain Control Of The US Senate.

 

Polls Are More Misleading Today Than Any Time Before.  Pollsters Are Polling Land-Lines And We Know Factually, Senior Citizens Are Mostly The Only Ones Still Using Land Lines And They Do Always Vote Republican. It would be Another Five Years Plus, Before Polls Would Be Able To Incorporate Cell Phones In Surveys In Order To Again Become A True Scientific Measure.

 

 

When It Make Sense To Question The Polls [NYT]

 

I wrote an article this week headlined “Why Polls Tend to Undercount Democrats.” Reaction was fierce.

 

A number of readers compared the article’s argument to the “unskewed” polls phenomenon before the 2012 presidential election, when many commentators argued, mainly based on their instinct about the likely composition of the electorate, that the polls were missing Republican-leaning voters.

 

Other than the observation that the polls might be off, the similarities end there.

 

The unskewers peered into the crosstabs of the polls, saw party-identification figures that they didn’t believe, and said that their candidate would lead if the polls matched their assumptions.

 

As a general rule, this is not a good idea. There are instances when it is fine to criticize the composition of the polls, but usually only when it’s a metric where we have very good knowledge about the “truth” of the target population, as when the pool of adults differs from the census or when a poll from the voter file looks a lot different than the voter file.

 

 

The “unskewers” were relying more on their intuition about the likely electorate rather than any data. There are no census figures, for instance, on the partisan self-identification of likely voters. The figures usually used by unskewers, like the exit polls or the Voting and Registration Supplement to the Census Current Population Survey, are not nearly good enough for this sort of analysis.

 

To take an example from this year: We really have no idea how many Colorado voters were Hispanic in 2010 or 2012. It might have been 7 percent; it might have been 12 percent. We really have no idea whether they voted for President Obama by 30 points or 60 points. So I don’t embrace the view, argued by many on Twitter, that we know the Colorado polls are biased if they show Mark Udall only up by a bit with a seemingly small Hispanic share of the electorate.

 

When is it a good idea to question the polls? When there are good reasons to believe that the polls are missing or screening out certain kinds of voters.

 

Research suggests that polls need to do the following things to produce a representative sample: call enough cellphones; sample voters with out-of-state area codes; weight to recent population parameters and hit your targets; weight by population density or appropriate geographic areas; conduct interviews in Spanish; call back nonrespondents over multiple nights of interviews. We could go on.

 

If you believe that these are best practices, then you should also believe that a majority of the polls out there have Republican-leaning samples, even after weighting, because many or most of the polls aren’t embracing some or all of these practices. Or put differently: If you think I’m unskewing the polls, then you must also believe that the top pollsters are unskewing as well.

 

It’s hard to say how large the accumulated bias might be, but it’s fairly indisputable: If we had up-to-date population targets for states, every pollster would weight to them — and it would make every poll more Democratic. If pollsters could call out-of-state area codes with random digit dialing, they would — and it would probably make polls more Democratic.

 

  All that aside, I really don’t think we should expect huge polling errors on Tuesday. Young and nonwhite voters won’t turn out in large numbers. There aren’t many Hispanic voters in the battlegrounds. The states tend to have smaller urban-rural divides. The generational divide is smaller, as well, with Democrats running far better than President Obama among older voters and behind among younger voters.

 

Any modest bias can easily be overwhelmed by movement among undecided voters or in screening for likely voters. Over all, the Republicans are undoubtedly favorites to take the Senate.

 

Twitter @Sheriffali