Thunderclass?
One place you should visit pretty often is Strange Maps, where I have gotten lost for hours at a time sifting back through the archives. I have also watched their visit stats skyrocket like a spaceship to utopia, so they must be doing something right. Case in point is this map of state-by-state obesity percentages:
What I wonder about this — and I’m sure there’s some empiricist or another who’s tackled the issue — is whether and in which states obesity correlates more closely to a particular economic class, or whether our fatness spreads out in a big equal blob. I can easily imagine fatness correlating with poverty in some areas but not others, and ditto middle classness. Maybe even certain segments of the rich. The easy assumption is that the underclass and the — oh let’s call it the thunderclass — overlap pretty significantly. But I bet at the within-state or even county-by-county level the story is far more revealing. To work, statisticians.
If you click my name, you will get a Google Doc with some hackwork statistics I generated by regressing the percentage of obese persons by state (2006, per CDC) against the percentage of persons living at or below 125% of the poverty line by state (2006, per Census). I then sorted the states based on how much the population of obese persons exceeds what the regression would predict.
Here are the top 5:
West Virginia 16% (i.e., 16% more obese than regression predicts)
Indiana 16%
South Carolina 15%
Alabama 15%
Michigan 13%
[Mississippi comes in at #12, at 9%]
… and the bottom 5:
New Mexico -15%
D.C. -16%
Massachusetts -18%
Montana -18%
Colorado -25% (i.e., 25% fewer obese than regression predicts)
I have no pride of authorship here so if anyone with a better knowledge of statistics can improve on this please do.
— alkali · Jul 8, 04:15 PM · #
Actually in all states (from what I’ve read), obesity correlates with poverty, race, and other factors (some studies have evaluated education levels, for example). As one would expect, these numbers are also influenced by medical comorbidities and genetics, which are not necessarily driven by socioeconomic factors.
However, the comorbidities causally related to obesity are essentially linked to the factors accounting for obesity (see: diabetes). This is troubling, since obesity is not a disease state (though it’s becoming one) but primarily a risk factor for other maladies. It’s not handled like a disease, nor treated like a disease – yet. You wouldn’t expect to go into the doctor’s office and hear something along the lines of “my diagnosis is that you are fat.” It’s more like “you are overweight, and if you don’t change that you are at risk for X, Y, Z.” Indeed, it’s similar to smoking.
Like smoking, there are cultural influences that affect obesity, but these are harder to pin down in epidemiological studies. You can plainly see on the map that the South is generally the “fattest” part of our country, and the Northeast and Midwest are the leanest. It’s known that both racial demographics and income levels affect this, but one could surmise that cultural tendencies (and, indeed, political tendencies) between these regions are strongly correlated to these numbers as well.
— mattc · Jul 8, 04:19 PM · #
@ alkali:
I updated your regression with racial demographics for whites, blacks, and hispanics, by state, and recomputed the expected fits. I’ve also included the regression coeffecients and the correlation statistics. You’ll see that this model fits the data better. A negative “Diff” means the state has more fat people than its poverty rates and racial makeup would suggest. Still, the goodness-of-fit is too low to say that these are the only factors that influence this rate (r^2 = 55%). As you’ll notice, though, the obesity rates are highly influenced by poverty rates and the % of Black people in the state.
State % obese % Poor White Black Hisp Exp % Diff
West Virginia 31.0 20.6 95.0 3.0 0.0 27.3 -3.7
Indiana 27.8 13.3 85.0 9.0 5.0 24.7 -3.1
Alaska 26.2 12.5 71.0 3.0 4.0 23.4 -2.8
Nevada 25.0 13.5 59.0 7.0 23.0 22.3 -2.7
Nebraska 26.9 14.3 83.0 4.0 8.0 24.2 -2.7
Ohio 28.4 16.7 83.0 11.0 3.0 26.3 -2.1
Michigan 28.8 17.6 79.0 14.0 4.0 26.7 -2.1
Wisconsin 26.6 13.9 86.0 5.0 4.0 24.6 -2.0
Texas 26.1 22.1 48.0 11.0 36.0 24.3 -1.8
Alabama 30.5 20.2 68.0 26.0 3.0 28.8 -1.7
Oklahoma 28.8 22.1 71.0 7.0 5.0 27.2 -1.6
Delaware 26.0 11.5 68.0 20.0 7.0 24.7 -1.3
California 23.3 17.4 44.0 6.0 36.0 22.0 -1.3
Missouri 27.2 15.6 83.0 11.0 3.0 25.9 -1.3
Iowa 25.7 14.4 89.0 2.0 5.0 24.4 -1.3
South Carolina 29.4 17.6 65.0 29.0 3.0 28.1 -1.3
Minnesota 24.7 11.4 86.0 4.0 4.0 23.6 -1.1
Washington 24.2 12.3 77.0 3.0 8.0 23.2 -1.0
Tennessee 28.8 20.4 77.0 17.0 4.0 28.0 -0.8
North Dakota 25.4 14.8 88.0 1.0 2.0 24.7 -0.7
Kansas 25.9 16.8 82.0 5.0 7.0 25.2 -0.7
New Mexico 22.9 21.4 44.0 2.0 42.0 22.4 -0.5
Virginia 25.1 11.7 68.0 19.0 7.0 24.6 -0.5
South Dakota 25.4 15.7 90.0 1.0 3.0 25.0 -0.4
Illinois 25.1 14.5 67.0 15.0 12.0 24.7 -0.4
Oregon 24.8 16.7 81.0 2.0 9.0 24.7 -0.1
Kentucky 28.0 22.2 89.0 7.0 2.0 28.0 0.0
Idaho 24.1 15.8 87.0 0.0 9.0 24.3 0.2
Georgia 27.1 17.0 59.0 29.0 8.0 27.3 0.2
Wyoming 23.3 12.7 89.0 1.0 7.0 23.5 0.2
New Hampshire 22.4 8.7 93.0 1.0 2.0 22.7 0.3
Arizona 22.9 19.7 57.0 3.0 32.0 23.2 0.3
New Jersey 22.6 11.6 62.0 13.0 17.0 22.9 0.3
Maryland 24.9 11.3 57.0 29.0 7.0 25.3 0.4
Mississippi 31.4 26.1 58.0 37.0 3.0 31.9 0.5
North Carolina 26.6 18.5 67.0 21.0 7.0 27.3 0.7
Hawaii 20.6 12.7 19.0 2.0 7.0 21.7 1.1
Florida 23.1 16.0 61.0 15.0 21.0 24.3 1.2
Pennsylvania 24.0 14.4 83.0 10.0 4.0 25.3 1.3
Utah 21.9 14.1 82.0 1.0 12.0 23.4 1.5
Maine 23.1 14.3 95.0 1.0 1.0 24.8 1.7
Arkansas 26.9 24.3 76.0 16.0 5.0 29.2 2.3
Connecticut 20.6 10.9 75.0 9.0 11.0 23.1 2.5
Rhode Island 21.4 13.8 79.0 6.0 11.0 23.9 2.5
New York 22.9 18.3 60.0 15.0 17.0 25.4 2.5
Vermont 21.2 11.6 95.0 1.0 1.0 23.8 2.6
Louisiana 27.1 22.4 64.0 31.0 2.0 30.1 3.0
Massachusetts 20.3 15.1 80.0 6.0 7.0 24.7 4.4
Colorado 18.2 13.9 73.0 4.0 20.0 22.7 4.5
Montana 21.2 18.7 90.0 0.0 2.0 26.0 4.8
Source DF Seq SS Adj SS Adj MS F P
% below Pov 1 120.225 79.116 79.116 18.46 0.000
White 1 16.461 3.790 3.790 0.88 0.352
Black 1 83.490 32.017 32.017 7.47 0.009
Hispanic 1 15.718 15.718 15.718 3.67 0.062
Error 45 192.854 192.854 4.286
Total 49 428.747
S = 2.07018 R-Sq = 55.02% R-Sq(adj) = 51.02%
Term Coef SE Coef T P
Constant 16.999 2.914 5.83 0.000
% below Pov 0.35597 0.08285 4.30 0.000
White 0.02816 0.02994 0.94 0.352
Black 0.11392 0.04168 2.73 0.009
Hispanic -0.08706 0.04546 -1.92 0.062
— mattc · Jul 8, 04:51 PM · #
My hat is off to you gentlemen.
— Matt Frost · Jul 8, 05:06 PM · #
@mattc:
Thanks. Am I reading your results right on the following points —
1) % hispanic population appears inversely correlated with obesity (albeit result not statistically significant at 0.05 level)?
[In English: more hispanics = fewer obese people, but the tie is not strong enough to meet the usual standard of statistical significance]
2) no obvious regional distribution to “DIFF”?
[In English: the state-by-state difference between the actual % of the population that is obese and what would you predict based on poverty and race doesn’t break down cleanly by region; e.g., some Southern states are fatter than what you would predict and some are skinnier, likewise New England]
— alkali · Jul 8, 05:14 PM · #
to answer your questsions:
1) Yes, in this model an increase in the % population of Hispanics correlates to a decrease in obesity rates, though this is only moderately statistically significant.
What might explain this? It’s a good question. First off, the Hispanic population is highly concentrated in this country into only a few states such as Texas, New Mexico, Arizona, and California, so the relationship between obesity and this demographic maybe inaccurate or undetectable on a state-to-state basis. I would say, though, that just by looking at the data it seems that Hispanics have a marginal affect on obesity rates at the state level. It does trend inversely, however.
2) Yes, it seems like there is no obvious regional distribution to “Diff.” For example, our top 5 states are West Virginia, Indiana, Alaska, Nevada, and Nebraska. Alaska is affected by its high Inuit population; a factor I did not control for but I know from research does affect obesity rates. The links between WV, IN, NV, and NE however are hard to discern. Maybe this is all part of the “Rust Belt” phenomenon?
3) This map had me thinking, so I went in and updated the regression AGAIN, to control for two separate variables: Marriage Rates by state and Out-of-Wedlock births by state. Neither factor came out significant, but that may be due to the time frame (i.e., I should have looked at Out-of-Wedlock births from the 1970’s or 80’s instead of 2005). The theory would be that states with low marriage rates and/or out-of-wedlock birthrates might show the “cultural” influence in the linkage between poverty and obesity…however this it was not the case in my model (the coefficient for out-of-wedlock was 0.087, p=.201).
— mattc · Jul 8, 05:52 PM · #
alright, now i’ve got something….
So the question is: why does poverty (and the regional, racial variations associated with it) correlate so strongly to obesity rates in our country? No more than a couple centuries ago, obesity was a sign of wealth and esteem, and the poor were malnurished and skinny (in most countries today, the still are). What happened to reverse this trend?
As the comments above have shown, poverty and race account for about 50-55% of the variation in obesity rates by state. What explains this? I tried Marriage Rates and Out-of-Wedlock births to see if I could find a correlation, and alas, I found one!
(http://www.statemaster.com/graph/lif_wal_sto_num_of_sup_percap-stores-number-supercenters-per-capita)
Thanks to the miracle of the internet and Google, I found a breakdown of the number of Walmarts per capita, BY STATE, and I included this into the previous obesity model with poverty rates, and racial makeup, and dabbled with the out-of-wedlock birth rates both in and out of the model again. The results are interesting:
So here is the regression table with Walmarts per capita (# of Walmarts per 1 million people), by state:
S = 2.01240 R-Sq = 58.44% R-Sq(adj) = 53.72%
Term Coef SE Coef T P
Constant 19.483 3.119 6.25 0.000
% below Pov 0.2178 0.1084 2.01 0.051
White 0.00997 0.03064 0.33 0.746
Black 0.10331 0.04090 2.53 0.015
Hispanic -0.08766 0.04419 -1.98 0.054
Walmarts 0.14094 0.07407 1.90 0.064
You’ll notice a slightly better model, but more importantly, though only moderately statistically significant…the more Walmarts there are in the state the higher the obesitry rate. The affect is not marginal, either, as an increase in 1 Walmart per million people is estimated to increase the overall state obesity rate by about 0.14%.
Obviously this is only an exploratory analysis, but it explains some of the regional variation in obesity rates…Walmarts are far more prevalent down South…these type of stores allow poor people to buy food (and usually unhealthy food) in bulk at a very cheap cost….put all that together with racial genetics and lifestyle choices – thats how you get fat people.
— mattc · Jul 8, 06:36 PM · #
I don’t think WalMart is the culprit here. There are plenty of counties in West Virginia that were only recently blessed with its presence, but where obesity is rampant. If you could check the magic WalMart inventory system, you might see West Virginians buying more old-fashioned staples, like brown beans and corn meal, than you think, but still suffering from obesity.
I think only a few states make good units of aggregation for this purpose, too. I’d love to see county-level data.
But carry on!
— Matt Frost · Jul 8, 07:23 PM · #
ok last post i promise – if anyone wants to try some of this out (it’s fun!)…scratch the Walmart influence….but instead look up the % of Roman Catholics by state… the first thing you’ll notice is that there are 2 states missing from the RC % list at statemaster.com….and they are Mississippi and West Virginia…
include RC % in a model with poverty rates, racial demographics, and out-of-wedlock birthrates, you get a 70% fit to obesity rates.
— mattc · Jul 8, 07:25 PM · #
What Matt Frost said. Wow!
— James · Jul 8, 07:27 PM · #
Why does poverty correlate with obesity? Unlike most places the price of food is very cheap in America. Usually, the cheaper the food, when it comes to fast food, junk food, and lots of pre-packaged food, is very unhealthy to eat. If you don’t have a lot of money it doesn’t necessarily mean you will eat this food, but it appears that this is the case. Lower-income neighborhoods are usually full of fast food restaurants and lower end grocery stores that don’t provide as many healthy options.
— Nick · Jul 8, 09:12 PM · #
The least obese states relative to their demographics tend to be ones settled heavily by the descendants of Massachusetts Puritans: Montana, Colorado, Massachusetts, Vermont, New York, Rhode Island, to name six of the top seven. (I’m assuming these Lousiana statistics are post-Katrina).
— Steve Sailer · Jul 9, 01:36 AM · #
Interesting multiple regression analysis, but your Hispanic example shows the weakness of the approach: Hispanics tend to be more obese on average (the lower Rio Grande Valley, for instance, has terrible problems with obesity and diabetes), but they tend to be attracted to states with slender upper middle class white people, such as California, rather than ones with fat lower class white people, such as West Virginia, because that’s where the jobs are.
A simpler approach would be to take the national average obesity levels for each racial group and calculate the expected overall obesity rate for each state based on its demographic makeup, then list the deviations from the expected.
For example, my observation is that Hispanics in LA, like whites in LA, tend to be less obese than their respective distant cousins in San Antonio.
— Steve Sailer · Jul 9, 01:45 AM · #