Friday, December 05, 2008

Political money divide by state in '08 Presidential election

++Addition2++Agnostic calculated the Spearman correlation. Using the 'standard' Pearson correlation (which I default to unless otherwise noted), and median family income for '06-'07, a correlation of .48 between the political money divide and a state's median income. So I was wrong in asserting that Republican states show more of a political money divide than poor states do.

++Addition++Agnostic shows that the political money divide correlates as strongly (slightly more so, actually) with median income (.38) as it does with McCain's total support in the state. Race plays an important role in both of these relationships. Fortunately, the exit polling provides data for whites by income, although not as detailed as the numbers for all races. I will see how the money divide looks if only whites are considered.


[Thanks to MWC for catching an elementary mistake on my part that originally had Utah's figure calculated incorrectly. It's been fixed, and I've removed the embarassing gaffe, but it's still traceable in the comments for the saddists among you!]

Intrigued by Razib's review of Red State, Blue State, Rich State, Poor State, but having decided to put it on the Christmas list, I might be flying blindly here. Razib writes:

To a first approximation the rich tend to vote Republican, the poor tend to vote Democratic, the old stereotype holds. But there is also the reality that the wealthier states tend to vote Democratic and the poorer ones tend to vote Republican. ...

The rich in Connecticut, the wealthiest state per capita, are not much more Republican than the poor. In contrast, the rich in Mississippi, the poorest state, are much more Republican than the poor. Ohio, a middle income state, is somewhere in the middle. What Gelman et al. are showing here is that looking just at states removes critical information; class is a much better predictor of political orientation in poor states than it is in rich states. It isn't that rich states are blue because they are rich, it is that in rich states income doesn't matter much in relation to politics.
The data used are from the '08 election, so if anything it at least offers a look at the extent of continuity in a trend identified in the book, looked at from a bit of a different angle. Honestly, I'm not sure where to go with what follows, but since it's been gathered and might be found useful to some people, a table is presented below.

The larger the political money divide was, the more strongly the state on the whole backed McCain. The correlation among the two is a moderate .35 (p=.01).

As the book predicts, wealthier states see less of a partisan economic split than poorer states do. Per capita gross state product and the political money divide correlate at .26 (p=.06). But that's without taking cost of living into account, which would surely attenuate the already weak correlation, as blue states are generally more expensive places to live than red states are. So more than poorer states seeing an accentuated political money divide, it is Republican states that do.

The political money divide is computed by taking McCain's proportional support among those earning more than $100,000 a year and subtracting his proportional support among those earning less than $50,000 a year. This was done in part because more precise income brackets lacked sufficient sample sizes in several states, while said breakdowns were available for every state. So if McCain earned 68% of the $100k+ vote and 47% of the $50k- vote, the political divide is 21 (68-47). That's the portrait of a state (North Dakota) divided pretty sharply along economic lines. States like these tended to go for McCain. Oregon represents the only state in which the political money divide is negative (that is, those making $50k- were very slightly more likely to vote for McCain than those making $100k+ were). Whiter states like these tended to back Obama.

This surprises me a bit because egalitarianism at the state level is marginally good for Republicans (the correlation between the income ratio of a state's top 5% of earners and its bottom 20% of earners and McCain's level of support is an inverse .12, p=.21).

Following is a ranking of states based on the political money divide as previously defined:

State$ divide
South Carolina23
North Dakota21
New York17
West Virginia16
South Dakota16
New Jersey15
North Carolina14
New Mexico13
Rhode Island10
District of Columbia5
New Hampshire4

My first thought is that this proxies fairly well for a state's level of social conservatism. Parenthetically, if anyone is aware of a decent state ranking by level of social conservatism, please alert me to it. Race--specifically the relative size of a state's black voting population--is also important (the two correlate at .45, p=0), a point that is hardly startling when it is recalled how black folks 'push' whites towards the GOP.

* Utah is the sole exception to the earlier assertion that data by the aforementioned income brackets are available for all states. The $100k+ breakdown is not reported, but by working backwards from the $50k- and $50k-$100k yields 81% support for McCain among the 17% of Utah voters bringing in more than $100,000 per year.


Stopped Clock said...

It does give a figure of 67% for those making >50K in general, which accounted for 60% of voters. If the lower 43 of those 60 voted 66% for McCain, then to make the group as a whole get 67%, the top 17 (i.e. those >100K) would have had to have voted about 70% for McCain. Of course, this would lead to a figure of about 60% McCain for the overall vote, but as you said, sometimes exit polls get messed up. Still, if we're only interested in the money-divide variable, I think it would make more sense to use 18 (70 minus 52) rather than 48.

Even with a score of 18, Utah stands out from its neighbors, but not by nearly as much as it would with 48.

mwc said...

Interesting stuff. I'm confused by your last paragraph, however. 0.52*0.4 + 0.66*0.43 = 0.49, meaning McCain could have garnered 63% of the vote while taking only 80% of the richest 17% (.8*.17 = 0.14). How exactly did you come up with the 123% number?

Audacious Epigone said...

Wow, really sloppy math. I did (.40+.43)*.52=.4316, apparently. Then .17x=.2, I guess. That was in my head. No more of that, I promise. All excel or forget (that's what I get for playing with Agnostic's stat site--I leave simple calculations for my own head).

agnostic said...

You should follow the blog for that book, as well as his personal blog.

Turns out you've been scooped!


And here

Greater state income predicts smaller rich vs. poor gap in Republicanism.

From what I've heard, the state version of GDP isn't really too sound as a measure of state wealth, though median income is. It's too hard to keep track of what's made and consumed within vs. across state boundaries.

Audacious Epigone said...


I will. Fortunately, it's complementary, not repetitious. He looks at $50k+ and $50k-. Certainly worth looking at for it's own sake, but the $50k+ encompasses almost 2/3s (62%) of all voters (nationally)--it's not really getting at the 'rich'. Also, the point made here that income split runs stronger along the Rep/Dem divide than it does along the rich/poor divide is unique (at least to the two posts you link to). Further, I wonder why Andrew doesn't give us r values.

agnostic said...

The Spearman rank correlation between your index of rich poor Republicanism gap and median state income is -0.38 (p = 0.008). So it's as strong as the Repub / Dem variable.

Gelman is a stats prof at Columbia, so maybe he puts most of his r's in papers, and treats the blogs as a hobby where he just presents the plots to get the gist across. (Plots are always better for getting the gist than r's and p's.) I dunno, really.

Stopped Clock said...

I just realized that the exit pollers were measuring household income, not individual income. That means that singles making $45000 a year are going to show up as "below 50K" but a family of four minimum wage employees will be between 50-100K.

But I think something like this came up a few weeks ago and you found that there was only a small difference between Republicans and Democrats in average family size, so that should make only a slight difference here as well, right?

I also suspect that a lot of voters ... well, maybe a few ... misunderstand the question when asked about family income and give them only their own. It seems hard for me to believe that half of voters are from families with <50K income, unless, I suppose, they're mostly seniors ... I dont know. I am probably only saying things you already know. I do wish the exit polls had asked for individual income instead of family income though.

Audacious Epigone said...


Using responses from 2004-2006, the average household size by party affiliation according to the GSS:

Democrat - 2.70 (n=3097)
Republican - 2.81 (n=2599)

Not a big difference.

Further, as long as there is not a pronounced systematic difference across states, it's not going to have a material impact on computing an index like this.

With exit polling, generalization is always an enemy. The more cross-tabbing you can do, the better.