as to sample sizes… we’ve only go 50 states to work with. THAT in itself is a “too small” sample.

But as Donnie Rumsfeld would say “You work with the sample size you have, not the sample size you wish you had” ;-)

The reason I make all my data available is to allow people to actually DISPUTE it, not just say “well, there are other factors involved, blah, blah, blah — its the same reason I put up EVERY law, regulation, and policy I found regarding Bush’s military records. If there is something wrong about what I’ve written, people have the tools to prove it, because I’d rather be proven wrong that sit around being wrong and not know it.

So please don’t hit me with this “you haven’t considered all the relevant factors” crap. I didn’t. NO ONE CAN.

And, btw, I’m not looking at “non-blacks”, only Whites. If I was looking at “non-blacks” i’d say I was looking at ‘non-blacks’, because the data is broken down into four separate ethnic/racial categories.

Paul, if you don’t want people to raise very real issues about statistical significance of reducing small samples into sub-samples then you’ve made you point. I personally don’t have the statistical tools or capability to evaluate it…but I know that it is a real effect. It doesn’t mean your analysis is incorrect, simply incomplete to SAY whether the phenomnon is real or not at the sample size you are using.

That’s why you need to take this to a non-parametric statistician to see if you have a real phenomenon. You say you want criticism, well that’s positive criticism. I’m not asserting that what yopu have found is crap.

Apparently my advice is to your ears…so I have no more to say.

It was I (not Selise) that was concerned about whether this result implied a “negative for Obama” (and showed the pattern in the largest balck % States) because you said

…but even in the high percentage states, while the increase in black voter participation results in better numbers for Hillary, the impact is marginal if not negative for Obama

My head is spinning about this scenario that you are suggesting and trying to understand why the trend would diminish in the states with the largest black electorates and at the other extreme. I’m trying to think if the diminishing white population would have a countervailing impact, but that doesn’t make any sense, because then you’d have something of a bell-shaped curve with the peak in the middle of the “Mid-Sized black states”…not some dramatic lessening.

And what is the scale of the shift, in any case? Are we talking about a 10% jump of White males (that supported Obama) going to McCain if Hillary is the nominee and a 15% jump of females (that supported Clinton) if Obama is the nominee? Or is it on the order of 3-5%?

In Tennessee, which is on the upper end of your effect, there is an 8% shift from Clinton to McCain if Obama is the nominee. You break that down into male and female differences, which likely have higher sampling errors, but with that size of an original shift you might obtain real gender differences. But if the average shift is only 3%…is that statistically real? And if you find a 75:25 ratio within that small shift? That might seem statistically significant…but could be a gremlin of sampling.

Personally, I think if the statistical analysis is robust you could easily throiw out all the “home states” (Arkansas, Illinois, New York, Hawaii, Arizona) and should find the same trend. One would want to know if a “home state effect” was creating outliers that were biasing the slope of the trend line.

I did do that. small changes in the angle of trend lines appeared, but nothing major happened. And if you average the data from all five, the totals wind up looking pretty much like the averages from the other 45 states.

But when you start throwing out the outliers, where do you stop? WV is an outlier. So is MA. And the more outliers you throw out, the more states that look “normal” when the outliers are there start to look like outliers.

But as you parcel this data into smaller and smaller subsets it becomes critical to ask if some of the relationships you find are chance, and which are simply outliers and whether the trend lines you find have statistical significance.

I saw something when I first looked at the SUSA data… huge gender gaps for Clinton, smaller ones for Obama, with the gaps for both (but especially for Clinton) weighted more toward the males (i.e. it might have been a 15 point gap for Clinton, but it was caused 40% of males, and 55% of females, supporting Clinton.)

So I gathered the data, and looked at it more closely, and found really obvious patterns. And I found some surprises — like I expected the gender gap to be largest in heavily Republican states — and it was, except for the south, where it was the lowest. And when I looked at how the “white shift” operated, I expected to see a greater change among women than men, assuming that the racist men were already voting against Hillary because they are sexist as well, so there would be that much of a change due to race becoming a factor among males. And I was wrong — there’s not much difference in who switches from Clinton to McCain in terms of gender.

The reason I make all my data available is to allow people to actually DISPUTE it, not just say “well, there are other factors involved, blah, blah, blah — its the same reason I put up EVERY law, regulation, and policy I found regarding Bush’s military records. If there is something wrong about what I’ve written, people have the tools to prove it, because I’d rather be proven wrong that sit around being wrong and not know it.

So please don’t hit me with this “you haven’t considered all the relevant factors” crap. I didn’t. NO ONE CAN.

And, btw, I’m not looking at “non-blacks”, only Whites. If I was looking at “non-blacks” i’d say I was looking at ‘non-blacks’, because the data is broken down into four separate ethnic/racial categories.

As some have pointed out…science is unbiased…but the questions that scientists ASK should be relevant to issues IN THE WORLD. You are asking questions about sex, race, political choice…and distribution of race over geography. Part of that “choice” of questions may be the cards you were handed (i.e. the data provided by SUSA)…rather than, say, socio-economic or educational status.

But as you parcel this data into smaller and smaller subsets it becomes critical to ask if some of the relationships you find are chance, and which are simply outliers and whether the trend lines you find have statistical significance.

You are only looking at Whites (or, actually, non-Blacks) and examining that small number that shift from one Democratic candidate (rather than stay with the eventual nominee) to the Republican candidate. Then you are contrasting the differences in gender for this percentage. That could be a very small sample size that you are contrasting. Then you are making comparisons of that ratio across States to get a trend line. You need to speak with a non-parametric statistician to see if it’s allowable to use non-statistically significant data in developing a trend line, and then what criteria one uses to determine if the slope itself is significant from random or null.

And as you’ve already pointed out the trend lines don’t apply well to the extremes (States with large or small numbers of African American voters). So why did you decide to evaluate the classes where you did, excluding the samples that were “high” or “low” at those particular points. Arkansas and Tennessee are “extreme”, certainly… and would give a good trend line…but that would be softened if one included Delaware and Virginia.

Personally, I think if the statistical analysis is robust you could easily throiw out all the “home states” (Arkansas, Illinois, New York, Hawaii, Arizona) and should find the same trend. One would want to know if a “home state effect” was creating outliers that were biasing the slope of the trend line.

p.s. i didn’t make any comment about electability – maybe that was someone else?

Sorry selise, I keep forgetting that your acting like a data junkie here — when I see people talking about individual states, my “oh christ, s/he wants to tell me that Obama/Clinton is really more electable and is using individual state data to argue a point I don’t want to discuss” alarm goes off! ;)

excellent! but i don’t see the link to it…. am i blind?

no, you’re not blind. I’m not making that available to the public yet. But write to me at plukasiak AT comcast DOT net , and I’ll send you the whole massive mess. :)

Okay. So it’s not actually a situation where the States with the GREATEST Black voter proportion has the greatest “white shift”. What you are saying is it’s in the States that have moderate Black %ages where the effect is seen. That’s a different point than your “summary”.

you’re right… but that’s why its a summary of trends, and why i go into far more wonky detail later in the piece.