CNN's most recent "poll" was rigged out the wazoo for Hillary Clinton.

[mhatrw]

First, the entire Democratic portion of the poll consists of just 280 "likely voters."



Second, CNN did not disclose the demographic breakdown of the "likely voters" they polled in the polling methodology. Hmmm. I wonder why!

Interestingly enough, Sanders leads by 14% among white voters. Sanders also leads among all voters under 50 and all men, but CNN did not see fit to publish this information.

Unweighing the averages, it is clear that the number of non-white voters CNN polled outnumbered the white voters polled by over a 2 to 1 margin. For the 56% of the voters who preferred Clinton supporters, the ratio of non-white voters for white voters was roughly 2.3 to 1. For the 38% of voters who prefered Sanders, the ratio of non-white voters to white voters was roughly 1.8 to 1. That works out to roughly 2.1 non-white voter polled for every white voter poll.

Therefore, this poll's "model" predicts that fewer 1 in 3 of the SC Democratic primary voters will be white. Furthermore, since the white voter category was placed within the margin 8.5% margin of error cutoff and men and the under 50 voter categories were not, this poll's "model" also predicts that fewer than 1 in 3 of the SC Democratic primary voters will be men and fewer than 1 on 3 of the SC Democratic primary voters will be under 50.

[LondonReign2]

1. If true, that is excellent

A surprise loss is far more disastrous that an expected one.

[mhatrw]

9. That's true. But if CNN can get people to think a Clinton landslide in SC inevitable, then

it will sap the energy of Sanders volunteers and depress the number of first time Democratic voters in SC and maybe even the number of first time caucus goers in NV.

[thereismore]

2. If 50+ has almost the same sampling error as Total, we know the numbers come heavily from 50+ people

Yes this poll is skewed towards older people. And probably women.

[Ferd Berfel]

3. For a few months I schlogged through these kinds of details

`of previous polls. I came up with a lot of the same crap.

MOre power to ya for doing it. I can't any more

[Schema Thing]

4. I know nothing about polling science...

How large should the sample size be in a place like SC? I know 280 seems laughably small, but what would be respectable? And I suppose it's not just the sample size, but how those people are distributed geographically and demographically?

[HereSince1628]

7. Depends on what result you want. Figures don't lie but liars figure.

[thomservo]

5. Unbelievable.

Good work.

[Persondem]

6. Could you explain this part of your calculation please?

"For the 56% of the voters who preferred Clinton supporters, the ratio of non-white voters for white voters was roughly 2.3 to 1."

Specifically how did you get 2.3 to 1?

[mhatrw]

13. 16 divided by 7 = 2.28

For Clinton supporters, the final weighted average of 56% was only 7% points away from the average for non-white voters while it was a full 16% away from the average for white voters. That means that among Clinton supporters the number of non-whites polled must have been about 2.28 times the number of whites polled.

For Sanders supporters, the final weighted average of 38% was only 9% points away from the average for non-white voters while it was a full 16% away from the average for white voters. That means that among Clinton supporters the number of non-whites polled must have been about 1.78 times the number of whites polled.

Since about 59.6% of those polled who said they would vote for for either Sanders or Clinton chose Clinton, that means that the number of non-whites polled was at least double the number of whites polled. And the ratio must have been even higher than 2 to 1 for women vs. men and over 50 voters vs. under 50 voters.

[Persondem]

15. It's not a weighted average. It's a percent. 56 out of 100 poll respondents chose Clinton.

There is no way you can get numbers of respondents of those categories that way. The #'s are NOT averages.

[mhatrw]

17. Weighted averages work for percentages exactly as they work for numbers

when all you are trying to do is figure out ratios/percentages. I realize that you are trying to confuse this issue for the mathphobic among our audience, but frankly you are 100% full of shit on this.

Let's do a thought experiment so you can understand. Suppose I were to administer a math test to only men and women.

Suppose that 75% of the class is women and these women on average get 90% of the questions right. Also suppose that 25% of the class is men and these men on average get 50% of the questions right.

That means that the average for the class is .75 * .9 + .25 * .5 = 67.5 + 12.5 = 80%. That is simply how weighted averages work. Because there are 3 times as many women as men, the final average is 3 times close to the average for women as it is to the average for men.

Now let us imagine a different math problem is which I all I know is that the average for the entire class is 80%, the average for the women in the class in 90% and average for the men in the class is 60%. The math works in BOTH DIRECTIONS!. Since 80% is 3 times closer to 90% than it is to 50%, we know the women in the class must outnumber the men in the class by a 3 to 1 ratio.

If you question this conceptually, well you are simply ignorant of the basic mathematical principles of weighted averaging.

[demosincebirth]

8. Quite obvious to you. Every poll that show HC ahead is either rigged or missed some

segment of the electorate.

[mhatrw]

14. Find a flaw in my analysis. nt

[demosincebirth]

18. It's all flawed with wishful thinking.

[MohRokTah]

10. Unskew them polls, it worked for Romney!!!

[riversedge]

11. The term to use now is -RIGGED-everying is Rigged.

[quantumjunkie]

12. K&R

[jfern]

16. There's clearly something very flawed with the numbers

Hillary has 56% overall. 40% with whites. 63% with non whites. That implies about 30% white.
The MOE overall is 6% or a sample of 278
For whites, the MOE is 8.5% or a sample of 138 or 50%
For non-whites the MOE is 8.0% or a sample of 156 or 56%