While we did extraordinarily well in 2006, there were several House races left on the table.  Candidates like Eric Massa and Larry Kissell came very close, and with DCCC support, could have won their districts.  In the case of Kissell, I've talked to two people in high level party positions - one local to North Carolina and one in DC - who told me the same thing about why they didn't put more into that race.  Polling.  They did polls one or more weeks before the election, and it just looked out of reach by six or more points.  In a case where you are moving resources around the country, it's hard to make a call to support someone like Kissell when your data says otherwise.  So I want to circle back to the wireless only problem I highlighted yesterday. It looks like that population is growing by 3-3.5 percent a year, which means by 2008 the wireless only population could hit 20%.  This is a population that cannot be polled by traditional phone techniques.

Meelar in the comments added some interesting thoughts:

Right now, the default assumption is to simply assume that excluding them won't matter, due to some combination of a) The wireless-only people are more or less similar to the population with landlines that we can reach, b) The wireless only people are less likely to vote and c) There aren't enough wireless-only people to be important.  Moreover, wireless-only is only one of the problems that cause sampling to not be truly random; others include people who work nights, low response rates, call-screening, and so on.

The problem is that all these other problems are getting worse as well (response rates are low and dropping, call-screening is on the rise, etc.)  So over the next, say, 5 years, there's going to be a lot of development in the field.  The place it's farthest along is Zogby, which currently uses email-based polling and heavily weights the data it gets based on a ton of demographics.  This sort of approach will probably get more and more common. Of course this introduces its own problems...we live in interesting times.

I don't trust Zogby, but there are plenty of commercial sector market research firms doing and using internet research.  This cycle, I hope we begin to see more of it in the political space.  Internet polling can test rich video and sophisticated messaging, and it can use psychological symbols unavailable to phone bankers.  It may also reduce dishonesty in responses, since people are less likely to lie to a computer than another human being.

The problem with traditional polling is compounded by a changing electorate.  There are going to be some really interesting GOTV tools rolled out this cycle, and I expect that the youth voting surge is going to increase.  Already, last cycle the polling on individual House races was so unreliable it led to some poor decision-making, underestimating turnout in grassroots campaigns like those of Massa and Kissell.  While you can compensate this by averaging across different polls on a national level and extrapolate support, in most Congressional and even Senate races there aren't multiple polling firms in the field with public results.

We need to figure out new metrics for receiving party support aside from money and polling.  Perhaps opt-in email addresses acquired?  Friends on MySpace?  Newly registered voters (I like this one)?  Chatter across blogs using sites such as Blogpulse?

I'm not sure, but the whole landscape of politics is shifting.  It's like an entirely new grammar is emerging, but we're not there yet.

Update [2007-5-16 13:41:0 by Matt Stoller]:: David Kowalski makes the interesting point in the comments that Republicans assumed that Jim Leach received no help from the NRCC and seemed 'safe'.