Two weeks ago, Alan I. Abramowitz pointed out a significant statistical discrepancy between the Gallup and Rasmussen daily tracking polls, and all other national polls on the presidential campaign:
Here's what I found. Since the beginning of May, over 74 days of polling, the Gallup tracking poll has shown Barack Obama with an average lead of 1.6 percentage points over John McCain. During the same time period, the Rasmussen tracking poll, over 76 days of polling, has shown Obama with an average lead of 1.8 percentage points. But during the exact same time period, 38 other national polls have shown Obama with an average lead of 5.2 percentage points.
Abramowitz's data was through July 21st. Given that this discrepancy continues to persist, I thought I would update it. From July 22nd through today, there have been forty-five national polls. Thirty-four of these polls were from Gallup and Rasmussen (although one was not a tracking poll), while eleven were from other polling firms. I found virtually the same discrepancy as Abramowitz (more in the extended entry):
I have no explanation for this difference. I also don't intend to pick sides, as there is an awful lot of data in both camps here. As per usual in my polling analysis, my feeling is that the truth probably lies somewhere in between, rather than on one end or the other.
However, I think this difference does actually explain something else. There is a long running debate in this presidential campaign as to whether it is a toss-up, or whether Obama holds a statistically significant (though not statistically overwhelming) lead. This polling discrepancy is, I believe, the source of the debate. The overwhelming number of polls put out by Gallup and Rasmussen are skewing the polling picture against Obama on at least an intuitive level. These two polls dominate the political junkie consciousness on the state of the campaign, and even dominate the averages in the campaign because they are so numerous. As such, people following the campaign closely are drawn to believe that we are more in the toss-up range. However, if all polling firms were weighted equally, Obama would hold a statistically significant advantage of around 3.5%-4.0%.across about a dozen polling firms, instead of the 2.0-3.0% lead he holds according to Pollster.com right now.
My thesis is that Gallup and Rasmussen spam are causing the campaign to appear about 1-2% closer than it probably actually is. While 1-2% isn't a very big difference in terms of actual numbers, it is the difference between a toss-up and a small, but statistically significant, lead for Obama. And that is actually a big difference.