Hillary’s Support Among Nonwhite Voters Has Collapsed

[Pharaoh]

On February 27th, Hillary Clinton led Bernie Sanders among African-American voters by 52 points.

By March 26th, she led Sanders among African-Americans by just nine points.

And today, Public Policy Polling, a widely respected polling organization, released a poll showing that Sanders leads Clinton among African-American voters in Wisconsin by 11 points.

https://www.laprogressive.com/hillary-fading/

[geek tragedy]

1. Sample size: 72 voters.

Those who know anything about statistics understand how dumb it is to extrapolate a national--or local for that matter--trend based on a sample size of 72

[NWCorona]

2. My zip code is 30% black and 32% Asian with a population of 43k

Bernie took it easily

[CalvinballPro]

5. Anecdotal evidence proves nothing. It's a story you make up and post in a comment, nothing more. nt

[NWCorona]

6. Yeah I made up the votes

[CalvinballPro]

7. Yeah, that's what I'm saying. You completely made them up. Now just 2.6 million more times...

...and your candidate might have a chance!

[NWCorona]

8. Ok

[CorporatistNation]

33. Hillary, "It Ain't Over Yet!"

&list=RDUcGrzbxRaxs&index=1

[RichVRichV]

3. A sample size of 72 gives a margin of error of around 11.7 percent.

That means there's a 95% confidence interval that they could be anywhere from tied to Bernie having around a 23 point lead.


Yes it's a small sample size. Still a pretty extraordinary change of events even assuming best case scenario for Hillary. However one poll does not a trend make. We need more results to verify if it's true or an outlier.

[geek tragedy]

4. No, it's so small no one would attempt to extrapolate from it.

Margin of error of that size means it's noise, not signal.

The pollster did not treat it as relevant.

[CajunBlazer]

9. ^^^^^^ True this!

A sample size of 72 might be okay for a population of 100 but nothing bigger. Any idiot who uses a sample size of 72 has no clue what he/she is doing. Heaven knows how else the poll was screwed up. Some people know nothing about statistics and are just fricking dangerous.

[geek tragedy]

11. Nothing wrong with the top line numbers

But when you do a poll of a state the cross tabs will get wonky for smaller demographics.

[CajunBlazer]

16. True, but with a sample size that small...

...a poll will produce margins of error at the 95 percent confidence rate which would negate the usefulness of the poll.

[RichVRichV]

12. Funny how demographic breakdowns were never noise before now.

There have been lots of polls leading up to now that had small demographic breakdowns taken as valid. It's funny how it wasn't an issue before. What is considered a valid margin of error for you? 8%? 5%? 3%?

According to the 2008 CNN exit polls, African Americans made up 8% of the Wisconsin primary voters. We'll go with this polls sampling of 10% instead.


In order to get an 8% moe you would need ~155 African Americans polled. At 10% of the voters it would need to be a Democratic sample size of 1550 people.

In order to get a 5% moe you would need ~400 African Americans polled. At 10% of the voters it would need to be a Democratic sample size of 4000 people.

In order to get a 3% moe you would need ~1110 African Americans polled. At 10% of the voters it would need to be a Democratic sample size of 11100 people.


You almost never see polls with total sample sizes over 2000 people for one party (even those are very rare). So to stick to correct sampling for the state you're never going to get an moe of better than ~8% at best on such a small demographic. You would need a specialized poll specifically at sampling a small subset of the total electorate to get anything more meaningful.



2008 Wisconsin exit polls -CNN

[geek tragedy]

13. Exit polls have (a) higher sample sizes and (b) higher response rates

Small sample sizes combined with low response rates are not a recipe for meaningful data extraction.

[RichVRichV]

14. What low response rate?

You have a direct line to ppp I'm unaware of? I've read the results and the abstract from the poll. There is nothing indicating a response rate in this poll lower than any other phone based poll.


At no point does it say it weighted African Americans. It clearly states in question 9 that 10% responded were African American which is in line with demographic breakdowns for the state. That means the poll reached the correct percentage of African Americans to make it valid for their demographics for the state.


The demographic breakdowns are valid for all like polls or invalid for all like polls. Up to this point they have been generally accepted as valid. Therefor this one is valid. The fact that you don't personally accept it is irrelevant.



The simple truth is this is one data point, which will either be corroborated or refuted by future data points. The fact it has a high moe makes it less reliable by itself but does not invalidate it. I'm sorry if it doesn't favor your candidate.

[geek tragedy]

15. Response rates affect the likelihood of getting a random sample

The margin of error % only applies if you have a completely random sample.

Low response rates--which are a chronic problem with telephone polls--make it much less likely that the sample set is truly random. Because people who respond to surveys often differ from those who don't.

So when you have a tiny sample size, and a low response rate, even the smallest bias due to response rates can blow the whole thing up.

[RichVRichV]

17. I understand how it works.

That doesn't change the fact that demographic breakdowns on phone polls have been accepted up to this point (even when the demographics we're relatively small). I've seen some polls that cut off at 10% (which this one falls a little outside of) and others that don't cut off at all (such as this particular poll).


Again, what moe do you accept as a valid data point? And how do you expect to reach that acceptable moe on 8% of the total voting population?

[cui bono]

19. You're arguing with someone who pretended to be a Bernie supporter for many months

and while doing so would write OPs and posts chastising Bernie supporters for being mean and nasty.

Don't expect them to be reasonable.

.

[RichVRichV]

20. It doesn't matter.

As an IT specialist who's major was engineering I love debating logic, math, and statistics. These types of threads are right in my wheelhouse. It doesn't matter to me if the other side is debating in good faith or not. I'm enjoying myself regardless.

[geek tragedy]

22. Generally for those small cross tabs I really don't pay the numbers that much heed.

Especially if it's a pollster who's not familiar with the state.

Also note that while 72 people might, if you squint hard enough, maybe provide some inkling of insight into black voters in Wisconsin, in no possible way is it a random sample of black voters nationwide, let alone New York, Philadelphia, and Maryland.

Also point to consider--95% confidence interval on a pristine sample still means 1/20 will be off. Given all the polls showing Clinton with large leads amongst African-Americans, this begins to look like an outlier.

Similarly, if a poll came in showing Clinton leading amongst white Millennials in a small sample size, it would be suspected of being an outlier.

TL; DR treat small sample size findings with extreme skepticism if they seem like an outlier compared to other polls

[thesquanderer]

24. People are over-stating the author's reliance on those 72 voters

re:

while 72 people might, if you squint hard enough, maybe provide some inkling of insight into black voters in Wisconsin, in no possible way is it a random sample of black voters nationwide, let alone New York, Philadelphia, and Maryland.

The point about those 72 voters was about Wisconsin. He used other points to talk about why Clinton's POC support could be seen as dropping elsewhere. He wasn't saying that the Wisconsin poll result was applicable to other areas; rather, he said it was part of a trend along with other information from elsewhere.

[geek tragedy]

26. Based on what, caucuses with zero demographic information available?

Ohio
Florida
North Carolina
Arizona
Illinois
Missouri

were the last six primaries.

[thesquanderer]

28. Based on lots of things. Read the article at the link, no need for me to re-write it here. (n/t)

[geek tragedy]

29. The article is innumerate gibberish. As is everything

that guy writes.

[RichVRichV]

25. It wouldn't matter if it was 720 instead of 72,

a single state is not representative of the whole country, or states in other regions. The sample size has nothing to do with that fact. Taken by itself it is an outlier and irrelevant nationally. However it can be used to prove a trend if other states (especially in other regions) begin agreeing with it.

Hillary has won states in the south with large non-white votes by 50 points. Bernie has won states in the west with large non-white votes by 50 points. Regional influence is showing to have an even greater effect than racial influence.

You can choose to ignore statistically relevant data all you want, and yes 72 is statistically relevant, I'll treat it for what it is. It's a lone data point that is largely meaningless by itself but can be shown as proof in corroborating other data points.

[geek tragedy]

27. Caucus states can't be extrapolated to primary states.

We'll see on Tuesday whether that # is an outlier.

And yes increasing the sample size by 10 fold does change its value added.

[RichVRichV]

31. First off Caucus versus primary is irrelivent here.

You can't use a primary vote and compare it to a caucus vote, and vice versa. it has no bearing on the polls for each. A poll is a poll is a poll. The methodology used for polling a caucus state is the same as the methodology used for polling a primary state. Therefor polls for either type are valid for comparison.


And obviously increasing a sample size by a factor makes it more reliable, and hence more usable data. That point was never under contention


Now let's talk noise. That's an interesting word you chose to use when describing 72 people being polled in a nationally recognized poll. I work in a wireless field. I understand noise all too well. Noise is ambient interference that obfuscates valid data. A small sample size isn't noise, it isn't invalid. An example of noise in this case would instead be online polls. Except there's another thing interesting about noise. One objects noise is another objects data.


Online polls is an interesting case in point. The Hillary crowd has immediately and vehemently disregarded online polls as being unscientific, self selecting, and irrelevant because they are not reliable at determining with accuracy how people will vote. The only statistical significance to online polls is that they have been demonstrably repeatable. Beyond that they have had no statistical significance. The same goes for Bernie's dominance on social media. That is also demonstrably repeatable, and has no statistical significance. But as I said one object's noise is another object's data. If you look beyond statistics and look at online polls and social media from a logical point of view, they both indicate Bernie has been dominating among the online crowd. The online crowd is predominantly young. Hence these useless online polls and social media's were an early indicator that Bernie would do well with the young crowd. That's the law of associative property at work. The online polls weren't invalid for determining who would win the race because they were self selecting, they were invalid because they only cover a small subset of the total voter base (those that spend a lot of time online).


Another point dismissed by Hillary supporters is Bernie's large crowd sizes. Those don't translate to how the population at large will vote. That point is true. The crowd's have been predominately young, and the early crowds in particular were largely white. There simply wasn't enough diversity for it to translate to voting patterns. However that doesn't mean the crowd sizes were useless as indicators. The large crowds are a correlation and causation of the grassroots activism that has made Bernie a viable candidate, as well as a driving force behind his mass of small donations. There is relevance to them beyond determining how the voting will go.


Another point that was ignored by Bernie supporters was the African American group here on DU. From early on it was staunchly Hillary in spite of DU going dominantly for Bernie, to the point even African American Bernie supporters were driven out. Many in the Bernie crowd saw it just as an overzealous group not representative of the African American base as whole. I saw it as disconcerting to be so far outside the norm of the rest of the site. It was later confirmed by the polling. I was hopeful Bernie would make a late turnaround the way Obama did in 2008, but saw the writing on the wall from the beginning.



The difference between you and I is that you have been taught to accept only what is considered valid data and reject the rest as "noise". I have been taught that all demonstrably repeatable data is valid until proven otherwise. Even if it doesn't prove what I am looking for. One thing I have learned from years of programming experience. If you ignore data just because it isn't what you're looking for, you miss out on a lot.

[geek tragedy]

32. a lot to go through here

1) a good online poll (e.g yougov) is certainly worth considering. it's a relatively new field, and the technology is changing all the time, but certainly it should be considered a viable option, provided the methodology checks out. However, polls that are self-selecting or Freepable are junk.

2) social media metrics and search engine metrics are not irrelevant--but it's a lot more art than science at this point trying to extract meaning from them. Bernie Sanders dominated online discussions, but so did Ron Paul at one point. Things turned out better for Bernie, but we're still aways from figuring out what's up with what.

3) Crowd sizes are good predictors for caucus states, pretty lousy indicators for primary and general elections.

4) Polling caucus states is a fool's errand except for Iowa.

5) I didn't put so much into the DU interactions--as much respect and admiration as I have for the AA posters here, small sample size and I don't want to extrapolate anything from DU. But the freakout about the BLM protesting of Sanders--not just here but all across social media-- showed to me that the Sanders movement had limited appeal beyond its Vermontish demographic base. It did lead to one of the best twitter tags--Berniesoblack.

5) back to the initial poll in discussion, another poll from fox news with a 9% MOE shows her up with all nonwhite voter 51-36. NY state, much bigger sample size, shows more than 2:1 Clinton advantage. Until further notice, I think this particular quirky subdata is going to prove to be an outlier.

[RichVRichV]

35. Let's go down the list.

1) a good online poll (e.g yougov) is certainly worth considering. it's a relatively new field, and the technology is changing all the time, but certainly it should be considered a viable option, provided the methodology checks out. However, polls that are self-selecting or Freepable are junk.

This is exactly what I'm talking about. Your calling online click polls junk shows just how closed minded you are on this subject. It has become common acceptance through force of will here that they have no valid measurement. That's nonsense. You try to narrow everything down to how will it determine the vote, when even politics are much more dimensional then that. The fact they are self selecting has no bearing on whether they provide useful data, only on their relevance to voting.

And by freeping I assume you're implying there is a small contingent of Bernie supporters using bots to artificially inflate them. That is complete nonsense. In the 8 months Bernie has dominated online polls, there hasn't been one shred of proof of anyone running up the counts. Someone would have come forward in that time. Nothing stays a secret on the internet that long. Further, online polls that require registration and tracking (such as requiring a facebook or twitter account to vote) have shown the same margins as general online polls. And you call us conspiracy theorists. There isn't one bit of difference between saying the online polls are being manipulated and Republicans saying there is vast voter fraud. Neither has had one shred of actual proof showing widespread use.

No, online polls are not a good indicator of how an election will go. However that is a far cry from the data being useless or junk.

2) social media metrics and search engine metrics are not irrelevant--but it's a lot more art than science at this point trying to extract meaning from them. Bernie Sanders dominated online discussions, but so did Ron Paul at one point. Things turned out better for Bernie, but we're still aways from figuring out what's up with what.

Once again you equate everything down to an immediate vote. You try to use Ron Paul as proof that social media popularity is meaningless because he lost in 2008. However you ignore the fact that his popularity and platform had a direct link to the formation of the tea party, a group that has been a thorn in our side ever since and has all but co-opted the Republican party. I would not call that a failure on his part, regardless of how the vote went. Again, you're so focused on one metric you ignore the larger picture. I think the biggest thing Hillary supporters are failing to see is that even if Bernie loses the movement he is championing isn't going away.

3) Crowd sizes are good predictors for caucus states, pretty lousy indicators for primary and general elections.

Caucus states have lower turnout, and hence enthusiasm to vote are more important there. However all elections, caucus, primary, and general are contingent on get out the vote. If they weren't then we wouldn't do last minute pushes for all of them. Oh, and they're not a predictor at all. You can't predict the future! You can only measure the here and now and make educated guesses of the future. That's also true of polling. That's a nit I like to pick.

4) Polling caucus states is a fool's errand except for Iowa.

There is not one bit of difference between Iowa and the other caucus states except Iowa goes first, hence more resources were devoted to it. More data points are always more accurate than less. The same can be done for any state, primary or caucus.

)5) I didn't put so much into the DU interactions--as much respect and admiration as I have for the AA posters here, small sample size and I don't want to extrapolate anything from DU. But the freakout about the BLM protesting of Sanders--not just here but all across social media-- showed to me that the Sanders movement had limited appeal beyond its Vermontish demographic base. It did lead to one of the best twitter tags--Berniesoblack.

Once again you show closed mindedness by automatically rejecting anecdotal evidence. Anecdotal evidence is not reliable (by itself), that is not the same as wrong. Even anecdotal evidence alone can become a proof when there is enough of it and it is consistent.

Let me give you an example, home remedies. For hundreds and even thousands of years people have used various home remedies to treat all forms of issues. Over time people told others of their results and the most promising anecdotal methods were adopted. There was no science to prove this. Now that we have science, some home remedies are being shown as false. Many more are being proven right. Here's two I know of well. For hundreds of years American Indians used an herb called Echinacea to treat all forms of illness. For thousands of years the Chinese used an herb called Astragalus to treat illness. Neither culture had science to prove it, only generations of anecdotal evidence. Fast forward to today and we do have science, which has shown that Echinacea promotes T-Cell production and Astragalus promotes NK-Cell production. Both are key cells in the immune system.

What's the point of this story? Even anecdotal evidence has value. No don't trust a single data point of anecdotal evidence, that would be foolish. Skepticism is a good thing. But also don't dismiss it out of hand. All repeatable data has merit, even if you can't always see what it is.

5) back to the initial poll in discussion, another poll from fox news with a 9% MOE shows her up with all nonwhite voter 51-36. NY state, much bigger sample size, shows more than 2:1 Clinton advantage. Until further notice, I think this particular quirky subdata is going to prove to be an outlier.

Once again, you can't take a state from one region and expect it's results to translate to a state in another region. Wisconsin very well could be an outlier from national results. But that doesn't automatically make the poll data wrong. Outlier means outside the norm, not wrong. It's completely conceivable that Hillary could win the black vote in New York by 20 points and lose the black vote in Wisconsin by 20 points.



You're trying to win an argument. I'm trying to get you to look beyond just cheering on your candidate when analyzing data. Data doesn't have bias. Data doesn't have an agenda. Data is just data. It's all in how it is interpreted, and sadly it is difficult for humans to interpret without inserting personal bias. We're all too often guilty of it, myself included.

[geek tragedy]

37. TL: DR we'll see soon enough nt

[uponit7771]

21. +1

[rhett o rick]

10. LOL. Grasping for straws. Face it, the People are going to throw your corrupt friends out

of Washington the DC. There has got to be a word that describes the worship of wealth. Mammon?

[ucrdem]

18. "... in Wisconsin."

Details, details.

[Wilms]

23. ^

[LettuceSea]

30. Take the church and the Pa$tor out of the equation and the vote gets interesting

I would love to see some sort of polling that split the vote between "churchgoer" and "non" etc.

Would be interesting to see the effect the church has on the AA vote, or lack thereof.

[tabasco]

34. LOL. Just posted the same article and self-deleted.

K&R.

[snowy owl]

36. so, are black regretting early votes as Sanders becoming better known?

anybody know