Hats Off to Pollsters FiveThirtyEight, Rasmussen and Pew

(MoneyWatch)  One dynamic sector of online media deserves a special call out as we evaluate the web's overall performance in the 2008 election cycle. That is a cluster of new political pollsters -- especially FiveThirtyEight, Rasmussen Reports and Pew Research -- all of which called the popular vote breakdown perfectly -- Obama 52% and McCain 46%.

With ever more powerful data analysis tools, pollsters like these have been learning how to "bake in" anomalies in poll data collected in the traditional manner, e.g., interviews conducted over land lines. This election cycle, they had to evaluate the impact of the growing number of Americans, especially young people, who don't even bother with land lines any more.

Since there is no established methodology for polling cell-only voters yet, this was no small problem to overcome.

Barack Obama's presence on the ticket as an African-American also raised the dreaded spectre of the "Bradley Effect," whereby white support for a minority candidate often proved to be overstated in the past when compared with actual voter behavior in the voting booth.
Greater computing power facilitates greater ability to crunch demographic data. As government databases have migrated online, voter registration figures are far more accessible now, so journalists, social scientists, and pollsters can uncover new patterns indicating shifts in the electorate -- as was correctly noted by many polls in Virginia, Indiana, Ohio, Nevada, New Mexico, and Florida, for example.

The reason companies like FiveThirtyEight, Rasmussen, and Pew can now be classified as media companies is that they publish much more than just poll numbers on a daily basis. They host blogs, analytical pieces, prediction markets, comments, and other tools for tracking many subjects -- like economics, sports, media, etc.

This group of three have long been on my personal radar. I've been checking them on a daily basis for the better part of a year. RealClearPolitics is another favorite, though perhaps due to weaknesses in some of the specific polls it chooses to aggregate, it missed the final outcome this time around. RCP included Gallup (+11 Obama), Reuters (+11 Obama), NBC (+8), CBS (+9), and ABC (+9) partnered polls, none of which seem quite as state-of-the-art as the best web start-ups. Without knowing RCP's weighting factors, my hunch is that this probably accounts for the site's relatively inaccurate (+7.6) final poll -- too much reliance on the old-time pollsters.

As the post-election euphoria winds down, stat-heads everywhere will begin to study where they were right and where they went wrong. You'll see them calibrating their methodologies further. And two years from now, in the mid-term elections, they'll get another chance in the national spotlight to continue gaining market share over the television network-sponsored operations, which simply did not perform all that well this time around.

***

Disclosure: Rasmussen often posts prediction widgets from Predictify, where I also work as editor-in-chief.