Site Meter

Friday, November 09, 2012

Another Polling Rant

At Steven Shepard discusses the night of truth which settled the partisan debate about the skewed samples of polls.  

I tried to post two comments.  One notes that the article presents the debate over whether all major pollsters but Rasmussen and Gravis were statistically biased.  It notes the evidence is that those two are biased and the MSM pollsters aren't.  I objected that this was not an open question before the 2012 election, since Rasmussen's polls had a 3.8% Republican minus Democrat bias in 2010.  

By shouting loudly, Republicans managed to make the debate over alleged bias in pollsters with good records and not the bias in a pollster with a very bad record.  People who looked at the data  from 2010 were not surprised by the results in 2012.  I have no doubt that there will be an article in 2014 on the Republicans' argument that all polls but Rasmussen Gravis and new imitators are biased and, after the election, another  noting the new discovery that Rasmussen is biased and that the Gallup likely voter filter introduces bias.  I am confident that, two years from now, journalistic conventions will imply that two year old facts must be considered irrelevant again.

My more serious concern with the generally good article is that the choice of one word repeats an unsupported claim made by the unskewers in an article noting that they're predictions were wrong.

The unfortunate choice of word in context (my emphasis)

 Groups such as Project New America and Americans United for Change commissioned multiple surveys in battleground states over the final weeks of the campaign, in some cases releasing more than one poll in a state per week, in large part as an effort to combat what one prominent Democratic strategist involved in the effort called the "negative storyline" formed by automated polls and other surveys that understated Obama's vote share. These were live-caller polls, conducted using landlines and cell phones, and in most cases they reflected Democrats' view that the demographic composition of the electorate would be more favorable to Obama.

My comment (edited a bit)

Excellent article.  

I want to object to "in most cases they reflected Democrats' view that the demographic composition of the electorate would be more favorable to Obama."  This is an assertion of causation (the effect reflects the cause).  It is only an accurate statement if the Democrats belifs caused the Grove results.  This can happen -- weights can be used to impose a guess about turnout on the rasw data.  However youpresent no evidnece that Grove did any such thing.  You could have made the other guess and written that Grove's results tended to support the Democrats' view etc assuming that the causation was from what respondents said over the phone to Democrats' views.  

In the whole sad unskweing debate, the unskewers (who you note turned out to be totally wrong) asserted without evidence that Democrat commissioned and MSM commissioned polls were unreliable because the stated results were caused by assumptions about turnout.  All pollsters who responded asserted that they chose weights based on census data and likely voter subsamples based on responses to questions.

In spite of the evidence that Grove and PPP polls were reality based since they were accurate, you casually assert causation from Democrats' beliefs to the data their contractors present.  It is known that at least some Republican internal polls reflected assumptions about turnout (no links but many Republicans have explained why their predictions were off this way).  You just assume that if Republicans do something Democrats must do it too.  This fits a rule of conventional journalism, but, in this case lead you to casually make an assertion about the direction of causation based on no evidence at all.

I think you should ask Grove if Democrats' beliefs about turnout affected who they called, their interview script or their processing of the raw data.  Just asserting this without even asking is not optimal journalism.  You will guess that I guess that Grove will  deny that any such causation occurred (and I guess accurately deny it).  I think there is a fundamental difference between Democratic and Republican approaches to polling as Republicans impose their guesses on the data and Democrats don't.  I think this is part of a general pattern in which Republicans are much less reality based thanDemocrats.  I thinkjust assuming that things must be symmetric is not a good approach to reporting on the two major parties we actually have.

No comments: