What I Want to See from the Polling Averages

There's no question that the polling average first pioneered by RealClearPolitics has demonstrably improved media reporting on polls. Media outlets have been forced to walk back on the inherent fraud of reporting on their poll as the most important and/or only recent poll. Outliers are no longer reported on as Holy Writ. The result is a much more stable and accurate picture of the race.

Still, there's still something wanting about the overall state of polling today. Despite the tremendous technological advantages allowing us to reach more respondents cheaper either through robocalling or the Internet, we still don't have a real-time picture of the election.

The current RCP average includes interviews taken from September 9th, 10 days ago. Is this really a reflection of where the race is today? Do we really have to wait until the Tuesday after a convention to assess what the bounce was?

The Daily Kos/Research 2000 tracking poll points the way in some respects. They are releasing results from individual nights. And these results are well within the 5.1% margin of error, and don't evince much of the jumpiness pollsters warn us about in one night samples. The last three nights were Obama +5, Obama +8, and Obama +8. I may think the poll overall is 2-3 points too pro-Democratic, but even skewed polls usually get the trendline right.

What if all pollsters would release similar one-night figures and we were able to aggregate these into an RCP or Pollster-style average? Wouldn't that eliminate a lot of the uncertainty of low sample sizes in the same way that RCP eliminated the uncertainty of CNN hawking its relatively high MOE poll while NBC/WSJ did the same?

This at least has the advantage of being an apples to apples comparison. The current polling averages encompass polls taken on different dates, often not overlapping.

We'd still have rolling samples, but as a political operative, I want to see the immediate impact of the financial crisis on the polls. Pollsters have told us that these one-night results are supposedly invalid, because of 1) low sample sizes, and 2) swings caused by one day news stories.

Averaging solves the first part, and -- excuse me -- aren't the one-day swings relevant in that they are possibly leading indicators of lasting shifts? If the media's time horizon for covering the campaign is a single news cycle, isn't it only fair that we measure the campaign that way too? And aside from AB/EV, Election Day is a single day event -- not a three-day composite.

Second, I'd like to see an RCP-style average for demographic groups. Though I don't deny there has probably been an Obama shift the last few days, have white women really swung eleven points in a week? I don't think we can make that contention without seeing the RCP/Pollster average for white women. Because pollsters can't hawk their topline data as a unique product anymore, they focus on the closely guarded crosstabs, which are even less accurate than a single poll. An RCP style average for key groups would give us a real (and fascinating) picture of how different groups have really shifted over time.