Approval

[jasonxctf]

Bush's approval rating is now 34%

 

www.drudgereport.com

[mr_genius]

Bush's approval rating is now 34%

 

www.drudgereport.com

 

 

probably has something to do with Iraq being a complete mess and a deficit that is way out of control.

[KipWellsFan]

probably has something to do with Iraq being a complete mess and a deficit that is way out of control.

 

Isn't the deficit supposed to shrink?

[Balta1701]

Already threw this one into the Dems only thread. It's a CBS news poll. Here's the link to someone other than Drudge. Here are my comments:

 

I think this is clearly an outlier on the negative side, but it's nice to see the Port deal doing some dragging in this number. And the approval rating for Iraq has dropped hugely from their poll in January.

 

The internals of this poll seem really skewed. They veer from Bush being at 35 in October up to 42 in January and back down to 34 in February. That's a lot of movement. I bet there's something funny in the way they're adjusting their sample which gives extra weight to Republicans polled...where a few ticks upwards in Republican approval gives a larger tick upwards in overall approval because they're a minority in the sample.

[Balta1701]

Isn't the deficit supposed to shrink?

They've actually found a statistical trick to make that happen (Hi Addisonst!). Instead of using any actual deficit that they ran, they've lately been talking in terms of the predicted deficit as a percentage of GDP in 2004 (the election year). Bush said that he would cut the deficit in half by 09. Their plan to do that is to use that number, which is higher than any other deficit we ran, and because of the fact that the total deficit could remain constant at the same time as the percentage deficit decreases if the economy grows.

[mr_genius]

Already threw this one into the Dems only thread.  It's a CBS news poll.  Here's  the link to someone other than Drudge.  Here are my comments:

 

 

i must be reading that wrong. it looks like you're saying CBS is skewing polls to favor a Republican president.

[Balta1701]

 

i must be reading that wrong.  it looks like you're saying CBS is skewing polls to favor a Republican president.

That wasn't my intention. I'm saying that CBS is somehow polling a sample that contains more Democrats than it should if it was a totally random sample (their raw numbers seemed to confirm this to my eyes). This just means that for whatever reason, they're calling more Democrats than Republicans. Maybe it's a state thing, maybe it's a race thing, who knows, but for some reason the're calling more Democrats than they should.

 

So what they wind up having to do is come up with a rational estimate of how biased their sample is, and then normalize those responses to that estimate. This is a trick that is used in a lot of polls...you try to come up with an estimate of how wrong you are in order to correct in a rational way.

 

But the other feature is that in any statistical operation, there is going to be random variation. That's why polls are usually given as plus or minus 3 percent...if you did the poll 100 times with a truly random sample, and the true approval rating was 40%, 95 times it would come up as somewhere between 37-43 just entirely based on random variation. The other 5 would be even farhter away.

 

Now, the other thing to note is that as your sample size gets smaller, the chance for errors increases. So, if we suddenly start talking about "all republicans" instead of all respondents, we have a much smaller group with a much higher margin of error. Say 10%. So if Bush's approval among Republicans was 70%, then 95 times out of 100 in a truly random sample of Republicans, it would come out between 60% and 80% entirely by random chance.

 

Now, the last thing is the bias adjustment. So we have a sample which is already varying on a much larger range than the overall sample...now we're multiplying that by some factor in order to make up for the projected defiency in Republicans in our total sample. So what that winds up doing is causing much more variation in the poll than their should be...we've taken a random variation across 20% and multiplied it by something like 1.5, so suddenly we've got a range of 30% in what number the polled Republicans are giving him.

 

So, the net result of this is that the poll swings quite wildly based on the fact that the sample of Republicans is too small, and small variations in either the sample set of Republicans polled or in Bush's approval among Republicans are magnified in the final poll, causing the poll to skew much more wildly than it should.

 

Bottom line, I think that the poll in January showing a 42% approval from this same group was probably a little too high, and this one is probably a little bit too low. Their sample sizes back that conclusion up, and a consequence of it is Bush's numbers swinging by 8% in a month.

 

He's probably dropped a little bit this month, but until I see 3-4 different polls all saying he dropped by 8%, this one should be treated as an outlier.

[Balta1701]

So today, we have 2 new polls out, 1 of which comes from a company which for some reason always has Bush polling higher than almost anyone else.

 

Fox News:

Bush Approval: 39%

 

Gallup:

Bush Approval: 38%.

 

Fox News shows a 5 percentage point drop for Mr. Bush's overall approval since their last poll in early Feb. Gallup only has a small tick downwards in approval, but a statistically significant uptick in disapproval. This is only the 2nd time Bush has fallen below 40% in the Fox Poll.

 

So, I think it's safe to say that Bush has fallen even farther in Feb., but not as far as CBS's poll had him falling. CBS currently looks like an outlier, especially compared with these polls. In fact, their January poll probably was an outlier too (on the positive side), so they're just bouncing around back and forth due to something screwy with their sampling technique.

[Balta1701]

LA Times shows the same thing...Bush at 38, Down 5 from January.

 

Quinnipac has Bush at 36, in a poll which talked to 1837 people (almost twice the average, for a margin of error of 2.3%).

 

This is what you call clustering. CBS is an outlier by about 2 points. Bush somewhere between 36-38, down about 5 points from January.