Schizophrenic Polls - Gallup

In the last two days, Gallup has released two polls. The Gallup Daily Tracking shows Obama up 6, the USAToday/Gallup shows McCain up 4. Two polls from the same organization with opposite results. How is that possible?

The answer is really quite simple - but it involves a little bit of math and an awful lot of guessing. Polls rely on the fact that a random sample of the whole can tell you within a margin of error, and with a certain degree of confidence, how the whole will behave.

In Presidential polling you'll see two kinds of polls - registered voters and likely voters. First, registered voters. This is, in simple math terms, the maximum total universe. If you aren't registered to vote, you won't be able to participate. Simple as that. The Gallup Daily Tracking poll is a registered voter poll.

The second kind of poll is likely voters. Likely voters are a sub-set of the registered voter universe - all likely voters are also registered voters, but not all registered voters are likely voters. So just who is a likely voter? From USA Today's analysis:

Gallup editor Frank Newport tells Jill that "registered voters are much more important at the moment," because Election Day is still 100 days away, but that the likely-voter result suggests that it may be possible for McCain to energize Republicans and turn them out this fall.

Who is a likely voter? In this poll, Frank says, that was determined by how much thought people have given to the election, how often they say they vote and whether they plan to vote in the election in November.
...
As for the difference between the tracking and USA TODAY/Gallup polls, Frank says not to read too much into it. "Statistical noise" may be largely to blame.

And herein lies the answer to not just why one polling organization can produce two polls with opposite results, but also to why polling has been so hazardous this election cycle. When it comes to speculating on who will turn out and actually vote, your (educated) guess is as good as mine, and maybe as good as Gallup. This year has blown away all turnout projections and assumptions.

Making this educated guess on likely voters is the "secret sauce" of political polling firms. Each one develops its own "turnout model" and runs with that. And they don't really like to share all that with the public - just the paying customers. Since you can't look under the hood, it's not really worth the test drive.

As Gallup's Newport points out, horserace polls this early really don't matter - no one really knows in July who is going to be a likely voter come election day in November. But it won't stop some in the media from talking breathlessly about it all day long. Next time they're talking about polls, pay attention to that little text at the bottom of the screen when they flash the numbers - if it's registered voters, pay attention, if it's likely voters, heed USA Today's warning. Personally, I'll be ignoring most of the likely voter polls until October.