Wednesday, September 26, 2012

On the Credibility of Polls, A Brief Hiatus

Polls, Polls, Polls
As you know, I watch poll releases pretty closely and the state-wide polling numbers released by Quinnipiac today were startling to me.  The race has clearly been trending towards President Obama since the DNC and he is a solid favorite, trending towards a highly probable favorite at this stage of the gain.  So that Quinnipiac would release polls showing him leading in Florida and Ohio was no shocker - virtually every poll since the DNC in those 2 states has shown the President in the lead.

What was shocking was the margins - 9 points in Florida and 10 points in Ohio.  Just for perspective, these are significantly larger margins than the President's margin of victory in 2008, and nothing in the national polling picture suggests that he will replicate his 2008 victory.

So what the heck is going on here and how credible are these polls?

Conservative commentator John Nolte notes that the sampling calibration in those two polls seems out of whack.  First, a brief explanation for those new to polling.  All pollsters calibrate their poll results based on a projection of the composition of the electorate.  This is because they are unable to truly randomly reach people as certain demographic groups may be more likely to answer the phone and speak to a pollsters than others.  So, for instance, if a pollster conducts a poll but finds that 70% of the respondents are men, but knows that in virtually every election, women comprise 51 to 55% of the actual vote, he will adjust the weighting of the responses to reflect that electoral reality.

In the case of the two Quinnipiac polls in question, Nolte notes that they show 9% more Democrats than Republicans in both Ohio and Florida, whereas the actual 2008 election results showed only 3% more in Florida and 8% more in Ohio.  If you were to normalize based on 2008 election results, it would shrink Obama's lead to 9% in Ohio and 3% in Florida.  If you were to normalize based on the mid-point between the 2004 and 2008 elections, Florida would be a dead heat and Obama would lead in Ohio by 1%.

Other polls released in the past 3 days show the President up by 3-5% in Florida and anywhere from 1%-8% in Ohio.

I suspect the reality is probably closer to a 3% lead for the President in Florida and a 5 or 6% edge in Ohio.  The Quinnipiac polls would appear to me to be outliers, just as the national Rasmussen poll, which Republicans like to cite, is an outlier for Romney, showing the national race a dead heat.

A Brief Hiatus
It is the worst time of the political season to be doing this sort of thing, but I will be taking a break for the next 10 days from blogging.  It is an intensely personal decision to support the faltering European economy by drinking large quantities of beer in Munich during Octoberfest.

I hope the first debate finds you well and I look forward to keeping you informed on all the twists and turns of the race down the home stretch upon my return.

No comments: