If you look down at the bottom of Rasumussen, you find this bit:

The telephone survey of 500 Likely Voters was conducted by Rasmussen Reports April 4, 2006. The margin of sampling error for the survey is +/- 4.5 percentage points at the midpoint with a 95% level of confidence

Rasmussen is saying that if they held the election immediately after running the poll, each answer would be within 9% of where they report it be - at least in 95 out of 100 cases. The other 5 times, it would be even further off than that.

So, Rick Perry has the support of 31-49%, Bell is 8-25%, Strayhorn is 10-28% and Kinky is 6-24%.

Of course, the actual election could be one of those five out of a hundred that are totally screwed.

It isn't a particularly good method, from a polling point of view. With 200 more responses, they could knock it down to +/- 3% at 99% confidence. That's why both DMN and Zogby have different numbers - they interview more people and eliminate a lot of the error.

I know this is terribly anal of me to get into in a comment, but I teach Political Science and polling is a hot button of mine.

Okay, but really, Bell has the charisma of a rock most of the time. He's not exactly a "rousing the troops" kind of guy. He seems not to be able to pull off the kind of campaign you predict will be needed to win. That's the Dem's real problem - well, that, and having pissed away a lot of their trial lawyer $ on Carole already.

Thanks for posting this - and yeah, based on the poli sci prof's analysis, it appears to be a pretty crappy poll.

Umm, +/- 4.5% is a nine point range, not 9 points each direction. And going from 95% to 99% confidence takes about 70% more responses. Going from 500 to 700 responses moves your confidence from 95% to 98% at the same MOE.

Zogby was 779 responses, so 3.5% MOE at 95% confidence. DMN was 1482 voters, for 2.5% at 95%. 3% at 99% would need many more responses.

Of course, that is on pure random selection. Throw Rasmussen's selection profile into the mix and you can poll 100,000 and still get skewed results.

It's just the kind of PM--Perception Management--that Thelma Bush and the Louise GOP love to use in polls, science, climatology and elections (and more) instead of reality-based evidence.

Thanks for the corrections, George. This is evidence that you should not cruise the blogs after drinking too much Scotch (but really, is there such a thing as "too much Scotch"?).

The thing to keep in mind is that it is still +/-4.5% at the 95% confidence level. Generally, that is the lowest level of confidence that will be considered as being worthwhile. Still, it wouldn't need any more responses to raise it to the 99% confidence level, it would simply increase the spread - from 4.5% to just under 6% MoE.

To get a 5% MoE at 99% confidence would require only 664 responses - not really that many more than the 500 they get. Closing that range becomes exponentially more difficult, though. A 4% MoE would need 1,037 and 3 % would need 1,843.

It's always good to remember that this sort of statistics was created to estimate the amount of live yeast in beer - not necessarily political preferences. To the extent that political preferences mirror the life cycle of a yeast colony, it is spot on. However, people are not yeast and they ocassionally throw a monkey wrench in pollsters concepts.

XT