Exit poll trends NH-NV-SC: Hillary improving % of Blacks, Obama declining percentage of Whiites-will

[papau]

Ted Kennedy be able to reverse trend?

Obama has gotten the following percentage of the White vote: New Hampshire, 36%, Nevada 34%, and South Carolina 25%.

Hillary has gotten the following percentage of the black vote: Nevada 14%, and South Carolina 19%. (I do not have a NH number)




SC exit polls also show more than half of the voters think both are playing unfair:

When asked if Hillary unfairly attached Obama unfairly, 70% answered affirmatively.

When asked if Obama attacked Hillary unfairly, 57% answered affirmatively.

A majority, 51% responded that BOTH attacked unfairly.

[Occam Bandage]

1. Those states do have comparable demographics and cultural dynamics.

I can see how one would be able to deduce a trend among an insanely multivariable equation from n=3.

[Bicoastal]

3. Exactly--if we measured and unemployment was 70% in Zaire, 45% in Albania, 12% in France,

and finally 4.5% in the US...

Would we say that worldwide unemployment was dropping precipitously?

[papau]

5. Many variable true - but an interesting "trend" - :-)

[lapfog_1]

2. Fun with statistics...

but completely erroneous interpretation.

The racial makeup of the states selected is so radically different that sample size starts to be a big part of the equation.

Comparing whites in Iowa that voted for Obama V. whites in South Carolina that voted for Obama and coming to some sort of conclusion there is just really really silly.

[papau]

4. Why is it "silly" if the main driver is "being white" - are we saying more conservative

whites vote less for Obama than Eastern New England whites?

Or that Hillary has more pull with blacks that are buried beneath the load of a southern culture.

And there could be, indeed are, other variables, of course.

but it is an interesting "trend"

[lapfog_1]

6. It's silly because you don't understand statistics.

The sample size of black participants in Iowa is not statistically significant. Remember "margin of error" that comes with every poll (even exit polls). That's because the poll doesn't question EACH AND EVERY member of a group. They question just a very few people. Depending on the sample size and the ESTIMATE of the number of total participants of a given group that you want to characterize, you also have to calculate the margin of error. Given the total population of African Americans in Iowa and the sample size used to determine that groups preference. But let's say the exit poll snags, say, 2000 people to interview (you'd be surprised at how small the samples really are, I doubt anywhere near 2000 people were interviewed). Given that Iowa has something like 3 percent African American population, that's like 60 people. Oops. That group
is way too small to determine anything about racial preference. They could have easily gotten 30 people that voted for Hillary or 0 people that voted for Hillary. Sample size of African Americans (given that they select 2000 people again) is going to be, what 1000, 1200? Now you can get a sample size that means something.

But to determine a trend based on that is bogus.

If you want trends, you need to sample the same demographics over a period of time. Not DIFFERENT demographics over a period of time.

One could just as easily make up different trends not based on race but on occupation or income or just about anything. But because you are using widely differing demographics, those trends are simply bogus as well.

[papau]

7. In my world - actuarial stats - 25 responses is indeed a minimum - but a "best estimate" is still a

best estimate.

You want to control for other variables - and you should - but a trend is a trend -

and indeed it can be explained away as you have - quite correctly

but I come back to "a best estimate" is a best estimate