What sense does it make to include only likely voters for approval ratings

This makes no sense to me. I mean, I can understand it if it were used in a political contest. But fucking approval ratings?

First they don't include non-registered voters. Who are mostly poor or immigrants and less likely to support Republicans. Then they disregard "registered" voters. Numbers would likely drop by 5% to 10% for Bush if they included "registered" voters and ignored all Americans. What these pollsters decide who will vote and therefore decide that Bush's approval rating should be 50% instead of 40 to 45% where it actually is.

This is unbelievable.

So you're scoring high with people who won't help you win your next election, and scoring low with people who will vote for your opponent. The likely outcome of the election is that you'll lose.

These are suppose to be surveys of how Americans feel about Bush. Or any President.

Not how "likely" voters (which by the way is very subjective) see Bush.

Take that new poll from PA. 44% of PA residents gave Bush a good approval rating. But the poll says that Bush's approval rating is at 51% among "likely voters."

So headline reads, Bush Approval is at 51% in PA.

Well no, it's at 44%.

understand what youre saying, but a politician can have all sorts of people like him or hate him, but its only the people that cast ballots that count. This method of polling best gauges support, but approval ratings can be very different than re-elect numbers.

The constant use of "likely voters", rather than registered voters, is rather recent and seems to have been done out of pressure from the right-wing, which for years asserted that national polls underestimated the Republican vote. As you say, the use of "likely voters" does make sense during a presidential campaign but there's no point to it when you're two, three, four years ahead of the next election. Gallup is one of those polling organizations that has switched to this formula; they used to publish polls of registereds until the last week of the presidential race when they would switch to presenting the "likely" numbers. But in recent years, they've turned to using a highly debateable methodology of determing who is likely to vote - more than a year in advance of the next election! - that has produced very erratic numbers that are sometimes at odds with the rest of the media polls. Such as this week.

On the specific question of presidential approval among likely voters, that's never made sense to me either. I guess we could look back, but I seem to recall presidential approval surveys in the past were conducted among all Americans, not just those registered to vote.

The normal way to do it is to poll the general public for approval ratings, and then ask likely or registered voters who they will vote for. Sometimes they do opinion ratings amongst registered voters or likely voters only, but generally that is to measure a president's electability, and is often done in conjunction with re-election questions. For instance, they will ask likely voters who they plan to vote for, who they would vote for if the election were held today, and then what is their general impression of the candidates and the president. It's to get an idea of how people are making their decisions, such as are they supporting a candidate they give low marks to, which might indicate weak support for that candidate.

Other polls have different purposes, and measure the approval rating amongst the general populace. The problem is the media just sort of spurts out the poll numbers and doesn't bother to explain who is doing the poll and for what reasons. So you have to read carefully to find out whether a poll is tracking the approval of a president or is measuring his re-electability (sic). I doubt many in the media even take the time to learn what they are reporting nowadays. They all remind me of undergrads spouting out the shortest answer they can crib from a textbook, rather than doing any research to learn what they are reporting. (um, sorry for the rant).

I admit I haven't been watching many polls lately, so I don't know if that has changed. Did you have a specific poll in mind?

They are trying to look at a hypothetical what if we voted tomorrow?

The problem is that the "likely voter" model applied to the raw results may be wrong and actually increase the error substantially.

"Likely voter" models had GWB* winning Florida by 6 percent 24 hours before the election in 2000. The math model applied underestimated black and democratic turnout substantially.