There are really two basic ways to sort out the likely, from the unlikely.
The first is to base it on past voting behavior; the other is to base it upon current level of voter interest, or you can do both.
In reality you MUST do both.
Out of a sample of 1000, about 825 +/- are registered, and of that 825 +/- about 550 will actually vote.
A good likely voter screen has to deal with 4 groups. Clearly this in not perfect "model", and there are certainly grey areas, but it generally
works pretty well. The boundaries between the 4 groups % wise will change a few percent election to election, but historically these are pretty close.
Firstly, there are the "partisans" - people strongly associated with a party or a cause. They make up "about" 44% of the population, and of the 44% about 34% out of that 44% actually vote.
These people show up in EVERY poll, registered, likely, super likely, hotdog, whatever.. From a polling perspective, these guys are NOT the problem. You can't design a screen these people will NOT get through.
The next group of voters is the "Good Citizens" these people are weakly, if at all, associated with a party or cause. They vote because they are, well, simply "Good Citizens" and vote because it is the right thing to do.
They make up about 10% of the population, and 80% of them (ie 8% of the 54% or so who actually vote)
Because they very regularly vote, a "likely voter" poll that questions about past voting behavior will include them, but a poll which ONLY asks about "Are you paying attention" or "How much have you thought about the election" etc will chronically under represent this block.
The next group is the intermittent voters. These folks vote 40ish % of the time.
This is the group you have to sort out by asking "Who much attention are you paying…" etc….
The last group, the Unlikely, rarely vote but about 1/3rd are actually registered. (Motor Voter)
A few % of this group stagger to the polls each election, but in terms of screening you can just about write them off. Unless they score HUGE in the voter interest questions you never count these folks as likely.
Voters who are 22 or less you basically ignore the previous voting behavior questions.
If somebody is 18 and actually took the time to get registered (ie went tot he courthouse, etc NOT just Motor voter) , you count them as likely unless they prettty much tell you they won't vote.
About 2/3rd of 18 year olds who get registereed actually vote (over all voting is still olly 1/3 or so of actual 18 year oilds however), but the rate drops into the 30s for 19, 20, and 21 year olds.
There are 7 firms I would single out as doing a really good job on "likely" voter screening:
Democracy Corps (D)
Public Opinion Strategies (R)
Zogby (yes - I am saying something nice about Zogby)
There is a downside to screening for both however.
If you are still a long way out from the election, depending on if Candidate X or Y had a good week, the level of enthusiasm of their supporters will go up and down - thus moving them in and/or out of the likely voter category.
Because of this, polls like Gallup (actually especially Gallup) will show very large swings in the electorate when you are many months away from the actual election date. (Gallup has swung from Kerrry +12, to Bush +6, to Kerry +6 in the last 4 months)
I want to know who has more fans - the St. Louis Rams or the Green Bay Packers. I define a "likely" fan as somebody who is "enthusiastic" about their team.
The Rams just beat the Packers 63-7 in week 2 of the regular season. Needless to say, the Packer fans are less enthusiastic, and thus less likely, while the opposite is true of Rams fans. - A poll of "likely" football fans will show a huge swing to the Rams.
Fast forward to the day before the Superbowl between the Rams and the Packers.. here the "likely" Football fan screen would likely work very well….
This is just a quickie answer… more to come later