Thursday, December 11, 2008

Red State, Blue State, Rich State, Poor State needs additional "Light State, Dark State"?

++Addition++Professor Gelman responds by referencing RS,BS. He does not appear especially inclined to look beyond the graphical representations of his data included in the book, and the data are not publicly available, so I'm still unable to compare the white PMD in '00 and '04 to '08 with any precision.

He also makes a point that deserves emphasis:
Even if certain differences were entirely "explainable" by race--meaning that these patterns occur in the whole population but not when considering whites, blacks, and others separately--this doesn't mean these differences aren't real. As we know from studying public opinion and voting, economic issues are huge, and the correlation between economics and ethnicity (in the U.S. and in many other places) doesn't mean that economics doesn't matter. When trying to understand the differences between Democrats and Republicans on economic issues, I think it can be helpful to understand the economic positions of their voters--conditional on race and also in aggregate. Both analyses are relevant.

Andrew Gelman, co-author of Red State, Blue State, Rich State, Poor State has pointed out that the '08 Presidential election shows a political money divide similar to the '04 and '00 elections--the poorer the state, the larger the partisan gap between low income voters and high income voters is. Using voters earning less than $50,000 a year and voters earning more than $100,000 a year, there is a correlation of .49 (p=0) between the political money divide (McCain's $100k+ support minus his $50k- support) and a state's median income. Gelman performs a related analysis using the $50k- and $50k+ categories. He doesn't give an r-value, but a reworking of the same yields a nearly identical correlation of .47 (p=0).

One of Gelman's readers (who is a Princeton professor) points out the economic explanation appears to be a product of state differences in racial demographics:

Larry Bartels comments that if you only look at whites, the rich voter, poor voter pattern is similar in rich and in poor states. So one of our main findings from the Red State, Blue State book from the 2000 and 2004 elections did not persist in 2008.
That's a perspicacious observation on Bartels' part. Exit polling for '08 breaks down income by race for whites dichotomously, into under $50k and over $50k categories. The relationship between the political money divide for whites and a state's median income do not approach statistical significance (r-value=.14, p=.33).

After presenting the graph above based on Bartels' comments, Gelman closes by writing "Lots of interesting patterns here." But the only pattern the graph reveals is that blue states tend to have higher median incomes than red states do. The color-coding creates the impression that there is a pattern to be observed in this graph, but there is no meaningful relationship between the x- and y-axes, which is the concern of the post. Presenting a color-neutral graph (click to view more clearly) demonstrates that the state distribution looks kind of like an inverted shotgun discharge, not a relationship of any statistical significance. (Gelman apparently uses per capita median income, whereas I use median household income, but the lack of a relationship exists either way).

Gelman's rich state, poor state divide appears to be proxying a racial split more than it is revealing a split along class lines in poor states. In presenting the class hypothesis, he uses three states as examples: Mississippi (poor), Ohio (middling), and Connecticut (wealthy). But these states might be viewed in another way: Mississippi (much blacker than average), Ohio (average in blackness), and Connecticut (less black than average).

Mississippi provides an excellent example of why Gelman's analysis of the political money divide is substantially incomplete in not taking race into consideration. It is the poorest state and also the state with the largest political money divide. Among those making over $50,000, 74% voted for McCain. Among those earning less than $50,000, only 25% did. That's a political money divide of 49. But that gap is almost entirely accounted for by the presence of blacks. Among whites, 84% of $50k- voters backed McCain, while 91% of $50k+ whites did--a PMD of only 7.

On average (mean, weighted), McCain does 6.7 points better among whites making more than $50k than he does among whites making less than $50k. Among all voters, the gap is twice as wide. McCain does 13.4 points better among $50k+ voters than he does among $50k- voters. Half of the PMD used as the basis for explaining the differences in electoral behavior between the rich and the poor is a product of race--it disappears when non-whites are taken out of the equation.

As mentioned previously, the relationship between the white PMD and median income is almost nonexistent. If only the top 25 blackest states and DC are considered, it attenuates to nothing (r-value=.08, p=.72). A tenuous relationship between median income and white PMD does emerge when race isn't a salient factor. In the 25 whitest states, there is a correlation of .29 (p=.16) between white PMD and median income.

Whites simply don't split much along economic lines. This is especially true as the presence of non-whites (specifically blacks, as they are the electorally largest and most 'contentious' minority) increases. Race appears to be more important than class is in determining how people are likely to vote.

I'm skeptical about Gelman's claim that in '00 and '04, whites followed the larger PMD trend that he says did not persist in '08. As he points out elsewhere, state voting patterns changed little between the '04 and '08 elections. Some realignment of the white vote did occur, though. For Gelman's claim to hold, this would have to have consisted of poor whites in the McCain belt turning sharply more Republican than they already had been, and wealthy whites on the West Coast and Northeast becoming more Democratic than they had previously been.

Unfortunately, verification isn't currently feasible. Public exit polls for the '04 Presidential election do not have data broken down by income and by race. For comparative purposes, I've left a request with Gelman to make his data available on an easily accessible medium like Swivel.

Relatedly, I'm also skeptical about one of the book's major selling points:
Myth: Class divisions in voting are less in America than in European countries, which are sharply divided between left and right.
Fact: Rich and poor differ more strongly in their voting pattern in the United States than in most European countries.
Tautologically, I do not doubt it. But I suspect a comparison of white Americans and white Europeans would actually validate the myth. In refusing to consider the enormous importance of race in comparative geographic analyses, we inevitably obfuscate our view of what is going on.

Data are here and here.


Stopped Clock said...

I should have known that it was mostly about race the first time I saw that chart, but for some reason I was content at the time to believe the lower classes of Southern whites had gone over to Obama's side ... which is probably just what every Democrat wants to believe.

A few other comments:

Your first chart is titled "Political money divide by state, whites only" but it seems to be the all-races version. Is this a mistake? I notice your other chart has "for real" after it, as if the Swivel program wasn't allowing you to change the name of the first chart.

A couple of minor errors: the cost of living for Louisiana seems to be too high (127.5, but the source says only 95), and Maine's is too low (but I guess the source agrees with you there). And on the second chart you have Pennsylvania identified as the wealthiest state, which it isn't.

I've been told there are a lot of rich ex-Yankees in the Atlanta area; if so I suspect the inverted income profile in GA is due largely to these immigrants.

DC is hilarious.

On the topic of humor, I wonder how many Economist readers we could fool if we erased the legend and told people that this was an IQ map.

The Undiscovered Jew said...

I don't (yet) know enough about statistics to give a precise analysis, but it seems to me the size of a state's Hispanic population is not nearly important in terms on affecting white voting patterns as compared the size of a state's black population.

For example, the Southwest and Florida have, needless to say, experienced the most rapid increase in the absolute number of Hispancis, yet voting patterns have not been affected much by their presence.

At the presidential level Colorado, Florida, and New Mexico are all swing states, California is rock solid Democrat, and Arizona and Texas are safe Republican.

The only state that shifted left or right over the past 15 years was California, but not because of the rise of the immigrant population but because whites started voting more to the left (whites are still 75% of the voting population). In the the Clinton 90's, white California began its shift to the left.

There isn't much evidence a growing Hispanic population causes a change in the voting behaviour of whites one way or another.

Audacious Epigone said...


I'll reload both those charts. I'm not sure how to edit the name, and the error is just as you suspect. I just got sloppy in Excel, once I realized I wasn't going to use SoL in the post. I sorted data several times without pulling the right section along with the rest, so it's off (notice how in the notes I say PA is the wealthiest, but in the median income category it's not). Everything used in the post itself is correct, however.


Right, at least not yet. Blacks are much more 'potent'.

Audacious Epigone said...


The ACCRA index (which I felt too problematic to use) doesn't give a CoL for Maine. Apparently it's the only state that doesn't have any participating cities.

Audacious Epigone said...


Nevermind the reupping. The only thing that is off is the 'notes' column, which is meaningless. "Poor whites more Republican" applies to Wyoming and Georgia. "Pretty wide for wealthiest state" applies to Connecticut, and the long comment (which I used in the post) applies to Mississippi.

Ron Guhname said...

It's a safe rule of thumb: Don't do any social class analysis without taking race into account.

Some researchers think it's unethical to focus on race ("Blacks are not black, they're poor") but it's simply sticking your head in the sand (which is unethical itself).

agnostic said...

In the book and his public talks, he says that in the data from '00 and '04, half of the effect that state wealth has on the rich-poor divide goes away when race is controlled for.

But it's still there, and the book has graphs for whites only.

BGC said...

Agnostic: "half of the effect that state wealth has on the rich-poor divide goes away when race is controlled for. But it's still there, and the book has graphs for whites only."

A general methodological point: when adding a control (such as race) removes a lot of, but not all of, a putative cause being studied, this may be due to residual confounding.

In other words, race differentials are a proxy measure for some putative psychological cause of voting - and because race statistics are not a perfect measure of psychological differences; they will never be able to explain all the variance, even if race is a correct or better explanation of voting than class/ wealth.

To take matters further usually requires more-controlled studies, usually either experiments set up to test the causal hypothesis - or else highly selective (therefore much smaller) observational studies.

But to design such studies requires sharpening the psychological hypotheses which 'class/ wealth' and 'race' measures somewhat-crudely represent.

In other words, one needs hypothetically to define more precisely and specifically what it is about 'race' or 'class/wealth' that affects voting - in an explicit causal pathway; and then test the steps in that causal pathway.

Audacious Epigone said...




While the political money gap is only have as large for whites only, there is no statistically significant relationship between median income and the PMD for the same (at least not in '08)--race doesn't just attenuate the relationship, it makes it virtually disappear. So that is presumably a big change from '00 and '04.

Jay Fink said...

I have often heard Dems mention that more welfare money and social services per capita are spent in red states than blue. They claim red state voters are hypocritical and ignorant to vote Rep since they are such big recipiants of the welfare state. They never consider the huge role race plays in this, especially the large black population in the Deep South who in fact loyally vote Democrat.

Audacious Epigone said...


Nor do they take a sober look at some of the reasons why whites correspondingly vote overwhelmingly Republican in the South. They see welfare as a confiscation of white income for transfer to a chronically underachieving black population.