CA-SurveyUSA: Obama leads McCain by 8

[Tender Branson]

Obama: 49%
McCain: 41%

What might be interesting:

28% of the sample is Hispanic and McCain leads 49-42 ...

http://www.surveyusa.com/client/PollReport.aspx?g=f25df9e5-868d-428e-9395-359684cc7901

[Eraserhead]

hmm.... the hispanic numbers are certainly a little... alarming.

But then again, Obama is leading among whites. Kerry didn't carry whites in '04, did he?

[Alcon]

hmm.... the hispanic numbers are certainly a little... alarming.

But then again, Obama is leading among whites. Kerry didn't carry whites in '04, did he?

He lost whitey by 4

[Tender Branson]

hmm.... the hispanic numbers are certainly a little... alarming.

But then again, Obama is leading among whites. Kerry didn't carry whites in '04, did he?

Outlier or downward trend ? We'll see next month. I´m going with outlier.

[Sam Spade]

Yes, McCain is running the same in PA as he is in CA

[War on Want]

hmm.... the hispanic numbers are certainly a little... alarming.

But then again, Obama is leading among whites. Kerry didn't carry whites in '04, did he?

Outlier or downward trend ? We'll see next month. I´m going with outlier.

[Fmr. Pres. Duke]

Outlier is thrown around this forum too much. This poll is not much further off that what we've seen coming out of California recently. In fact, it's right in line. The consensus is showing about Obama +8 right now.

[War on Want]

Outlier is thrown around this forum too much. This poll is not much further off that what we've seen coming out of California recently. In fact, it's right in line. The consensus is showing about Obama +8 right now.

He decreased his numbers with Hispanics 17 points in one or two weeks. How is that possible?

[Alcon]

He decreased his numbers with Hispanics 17 points in one or two weeks. How is that possible?

The MoE on Hispanics will be much higher, and the 1-in-20 rule (95% confidence rate) applies to subsamples too.  That is, in a perfectly-conducted poll with 20 subsamples, an average of one will be off in each poll.

[Gustaf]

He decreased his numbers with Hispanics 17 points in one or two weeks. How is that possible?

The MoE on Hispanics will be much higher, and the 1-in-20 rule (95% confidence rate) applies to subsamples too.  That is, in a perfectly-conducted poll with 20 subsamples, an average of one will be off in each poll.

Erm..sort of and not really. 1 in 20 will fall somewhere outside the MoE, probably just outside the MoE.

[Alcon]

He decreased his numbers with Hispanics 17 points in one or two weeks. How is that possible?

The MoE on Hispanics will be much higher, and the 1-in-20 rule (95% confidence rate) applies to subsamples too.  That is, in a perfectly-conducted poll with 20 subsamples, an average of one will be off in each poll.

Erm..sort of and not really. 1 in 20 will fall somewhere outside the MoE, probably just outside the MoE.

I thought the MoE was the range in which things were supposed to fall 19/20 times (95% confidence rate)?  It's about time I learned this concept.

[Gustaf]

He decreased his numbers with Hispanics 17 points in one or two weeks. How is that possible?

The MoE on Hispanics will be much higher, and the 1-in-20 rule (95% confidence rate) applies to subsamples too.  That is, in a perfectly-conducted poll with 20 subsamples, an average of one will be off in each poll.

Erm..sort of and not really. 1 in 20 will fall somewhere outside the MoE, probably just outside the MoE.

I thought the MoE was the range in which things were supposed to fall 19/20 times (95% confidence rate)?  It's about time I learned this concept.

Basically, yes. The thing is, when you make a poll the probability of the real result deviating from the poll result is lower the larger the deviation. You usually want a 95% confidence interval, and given the usual number of reposndents that usually means a MoE of about 3-5%. So, let's say the MoE is 3%. If we expand the MoE to, say, 4% or 5% we may increase our confidence interval to 96% or 97%. And so on. But the probability of the deviation being really big (10% or 20%) is incredibly small. Much smaller than 5%. This is why the "1 in 20" concept is a bit misleading. Most of those 1 in 20 polls will deviate by 3.1% or 3.2% (assuming a MoE of 3%) and not incredibly large figures.

Again, this, of course, assumes that the pollster is competent. And that the sample is reasonably large, since the MoE otherwise becomes pretty big.