Last night I got a bit flippant about likely voter screening, motivating one correspondent to point me to an extended analysis of the subject done during the 2004 Kerry-Bush presidential election on mysterypollster.com. I found it educational, and you may, too.
My correspondent was quick to point out that a key feature of likely voter screens for most polling organizations is that they are based on multiple questions. Some give each respondent a score on each likely-voter question, e.g., one point for either of the top two answers on a likely-to-vote question, one point for a yes on a do-you-know-where-you-vote question, and so on, and then filter out those with low total scores.
Gallup has been very open about its model, including the fact that it adjusts the cutoff to get a ratio of likely voters to registered voters equal to a historical average of 55%. It is probably that ratio on which their heavy and light turnout models are based.
I'd happily take the Gallup low-turnout model's 13% lead for Republicans, since that would be the biggest wave election in the last century and a complete repudiation of the progressivist programs.
