A couple months ago, I talked about the questionable likely voter model the USA Today/Gallup poll was using, giving McCain a 7% bounce from registered voters to likely voters in their July poll. Well, after a month absence, it appears that the wacked out likely voter model not only has returned, but has actually gotten worse.
Update
It has been pointed out that I misread the article and that it is 1,022 adults and 959 registered voters, with no number of likely voters.
As a result, my numbers aren't going to quite be correct. I'll update them if any likely voter number actually comes out. (why they bothered listing how many adults they sampled, I don't know).
So I'll go through the process for this poll like I did for the last poll:
Total registered voters in this poll numbered 1,022, and the result was McCain 50%, Obama 46%
McCain received between 50.4% and 49.5% of registered voters - or between 506 and 515 people.
Obama received between 46.4% and 45.5% of registered voters - or between 465 and 474 people.
Total registered voters in this poll numbered 959, and the result was McCain 50%, Obama 46%.
McCain received between 50.4% and 49.5% of registered voters - or between 475 and 483 people.
Obama received between 46.4% and 45.5 of registered voters - or between 436 and 445 people.
Now, the total likely voters in this poll numbered 959, and the result was McCain 54%, Obama 44%.
McCain received between 54.4% an and 53.5% of likely voters - or between 513 and 522 people.
Obama received between 44.4% and 43.5% of likely voters - or between 417 and 426.
So what does this tell us? We know that McCain's registered voter percent can't me less than 50.2%, and his likely voter percentage can't be better than 53.7% given his absolute minimum number of voters is 513 and the absolute maximum is 515.
In short, at maximum John McCain loses 2 people to the likely vote filter. There is a very good chance that every single McCain registered voter was counted as a likely voter.
What about Obama? At best, he dropped 39 people from the registered voter model to the likely voter model, assuming he had the lowest registered voter percent possible and the highest likely voter percent possible. It's likely the number is closer to 50.
In any case, at the absolute minimum, Obama registered voters were 19.5 times as likely to be dropped from the likely voter model than McCain voters were.
People can check my math, but I'm pretty sure this is correct.