Game time! What percent of the sample did Hispanics make up in that Georgia Poll?

[Godhumor]

1.0%

Now, it is a weighted poll which means fractional respondents are possible. So out of a weighted sample of 413 people, 1% is 4.13.

Sanders has the support of 1.79 Hispanics versus Clinton's 1.44 Hispanic voters. That's almost a lead of half a person!

Truthfully, I have no idea why the pollster even bothered to indicate level of Hispanic support in his press release. The Hispanic subsample is so miniscule any results derived are meaningless.

Hat tip to DSB for actually looking into the sample breakouts in the poll.

[AtomicKitten]

1. It's only meaningless because Bernie's ahead

.... with the Hispanic vote and some aren't having it. The Clinton campaign drones (Luis Guitierrez, Joaquin Castro, etc.) have tried so hard to push the "Bernie doesn't care about brown people" meme, and this poll exposes that meme for its fraudulence.

[Godhumor]

3. The poll itself is valid. The MOE on a subsample of only 4 people though is +/- 49%

There is no scientific way of deriving conclusions on a population from just 4 people. None.

[AtomicKitten]

6. optics aren't good for Clinton so her dutiful

... busy bees almost immediately start minimalizing it and ripping it apart

[Godhumor]

7. You think the Georgia Poll is bad news for Clinton?

Well, good for you.

Editing: more snarky than I wanted to be.

[teach me everything]

2. Excuses. Excuses. n/t

[SheilaT]

4. What's the percentage of Hispanic voters in Georgia?

That's the important number.

[Godhumor]

5. Actually it isn't. You can't sample 4 people and reach conclusions

It doesn't matter if the voting population is 1% Hispanic. There is no problem with only having 4 Hispanics in the sample, if you truly believe the population parameter for Hispanics is only 1%. But it is irresponsible of the pollster to try and make inferences from just those 4 people for the entire Hispanic voting population.

[SheilaT]

13. If you've chosen the sample correctly, and the 4 Hispanics polled

reflect the percentage of likely Hispanic voters, then those 4 are enough. Now, if they only polled 4 Hispanic voters in Georgia and are trying to say that's valid for the rest of the United States, those 4 are not enough.

When I first took statistics I was quite amazed at how small a sample could give good data. I took that class in the fall of 1976. and every time I read a new poll in the newspapers, if they gave me enough information (and usually they did) I'd double check the math about margin of error and such. Somewhat to my astonishment, the numbers always checked out. Keep in mind that back then the electorate was being so finely divided as it is now.

But just saying that a sample of four of a specific group, when within a much larger sample, isn't enough to go on, probably isn't correct.

[Godhumor]

14. Margin of error on 4 people is +/- 49% for an absolute MOE of 98%

You want to sample Hispanic voters as a group when they're 1% of a population then you run a poll targeting the Hispanic population. If you want to include just 4 in the sample as a whole to keep your population parameters in line, that is fine. But you don't assume those 4 are representative. I work big data analysis every single day. No matter how random, a sample of 4 is far too small, as the risk of a clustered result is enormous.

Edit: Went and did the math on MOE for a sample of 4. I was off by 2% points.

[dsc]

16. That is so filled with wrongness

To say one thing you do have right, it is possible the percents in the poll are close to the percents in real life. But as for the rest. I don't care how randomly you select your sample of 4, a sample of 4 is never going to be a statistically valid sample of a population no matter what size that population is. 30 is considered the absolute minimum for any of the assumptions used to make statistical calculations to be valid and even then the MOE would be so high the results would be close to meaningless. Without doing the math on a sample of 30 I would think the MOE would be in the neighborhood of +/- 15 percent.

[SheilaT]

17. If I recall my first statistics class correctly,

there are situations in which a sample of one is actually sufficient to accept or reject the null hypothesis (to use statistical language). So to simply assert that a sample of 4 is never going to be sufficient under any circumstances, suggests only a passing acquaintance with how statistics works.

I am not an expert. I do not claim to be. But I know that there are times when surprisingly small samples are good enough.

[dsc]

19. you recall very, very wrong

A sample of one is honestly nothing short of absurd. No, you can't reject a null hypothesis under any circumstances whatsoever with a sample of one. I honestly can't even fathom what you might be confusing this with.

[SheilaT]

22. I just pulled out my old statistics book. And while I can't completely follow all the reasoningq

some forty years later, there is an entire chapter called: Statistical Tests of Hypotheses: The One Sample Case. You might need to take a statistics class or two.

Granted, it wasn't talking about election polling, but there are times when one sample is sufficient. And I am still going to hold firm to my hypothesis that if the sample is chosen correctly, 4 Hispanics could well correctly represent all the Hispanics in the larger population.

[dsc]

23. that isn't what that means

not even close to what it means. It is not saying a sample of with an N of 1, they are talking about one sample as opposed to two samples. For example, you work on an assembly line in quality control, you take a sample of the product and compare it to a standard. As opposed to taking a sample from two different days and comparing them to each other. This isn't saying taking a sample with an N of 1.

[Godhumor]

21. A sample of 4 for any sizable population is an MOE of +/-49%

That indicates total movement possible of 98%.

For a population, a sample of 4 isn't an actual sample. It is inconsequential.

[Godhumor]

20. Your guess of +/- 15 is surprisingly close

In the case of this poll, the calculated MOE would be +/- 17.9% for sample size you suggested.

[dsc]

24. I used your 49

and then used the square root of 8 to estimate. So I couldn't have done it without your calculation.

[Glitterati]

8. In 2014 and 2012 1.8%

According to the Georgia Secretary of State Elections Division, voter registration statistics as of October 1, 2014 show that 92,000 Latinos are registered to vote statewide. Overall, Latinos make up only 1.8% of the state’s 5.1 million registered voters.

http://www.pewhispanic.org/fact-sheets/2014-state-election-fact-sheets/latinos-in-the-2014-election-georgia/

[Godhumor]

9. Right, so 4 Hispanics in a sample for a population makes sense

But that is not enough to make any claims about who Hispanic voters as a whole are supporting.

[SheilaT]

18. Once again, if that sample is properly selected

and matches the percentage of Hispanic voters in the population, it would be a sufficient number. I have no idea if the sample was properly selected, and the percentage of Hispanic voters sampled is smaller than the percentage of Hispanic voters in the population, so there is a bit of a problem, but it's not as vast as people here think. Sampling two or three more properly selected Hispanic voters, would have been sufficient.

[Godhumor]

25. No, it simply isn't. It just isn't.

There is no such thing as a valid sample of 4 for a sizable population.

[Vattel]

10. Lol, obviously you are right that the sample of hispanics is way too small to be meaningful.

The same thing applied in the poll cited by some HRC supporters that found that, among those likely democratic primary voters polled, more (self-identified) very conservative voters favored Sanders. There were only 19 voters who self-identified as very conservative. So the results were meaningless there too.

[Godhumor]

12. Right, exactly

It is why I tend to ignore subsamples in polls unless they are big enough to have a respectable MOE.

[Vattel]

15. good policy

[ericson00]

11. thinking Bernie is winning much is fuzzy math

n/t