Problem with Daily Kos Tracking Poll?

If you're among the poll obsessed like me and click for the latest polls several times a day at Pollster.com and Real Clear Politics, you may have noticed that the Daily Kos tracking poll appears the most favorable to Obama.

Today, the tracking poll, conducted by Research 2000, has Obama leading McCain 49-43, or by 6 points. In contrast Gallup Tracking has Obama up by four points, as does Diageo/the Hotline (.pdf file). The Rasmussan tracking poll has the race as a tie. One tracking poll even has McCain up by 2 points, though I can't tell if this is just for the Battleground states or not.

One might be quick to say that the difference between the Kos/R2K poll and the others is within the margin of error--and you'd be right. On the other hand, the Kos/R2K poll has consistently favored Obama over the past few days.

Kos laudably releases the internals of the poll which show the demographic composition of the poll. I was surprised to see that Latinos comprise 13 percent of the sample. Now, Latinos compose around 13 percent of the population, so you might say that makes sense.

However, many Latinos are non-citizens. The share of non-citizens is especially high among the voting-age population. A much higher share of Latinos not of voting age are citizens because they were born in the U.S., and thus are citizens by right under our Constitution even if their parents are not.

Latinos skew around 2-1 for Obama, so including too high a share of Latinos might skew the survey in favor of Obama and against McCain. I don't know what share Latinos compose in many other polls--it's 8% in the tracking poll which favors McCain. I'd be curious to hear from people with more thoughts or information on this point.

Update: Two top experts in Latino politics have told me that Latinos will probably compose around 8 percent of the electorate, confirming my sense that the share of Latinos in the Daily Kos poll is too high.