« Mourning Detroit | Main | Does the End of Growth Mean the Rise of Inequality? »


Feed You can follow this conversation by subscribing to the comment feed for this post.

Surely the cost comes from people not being able to contribute as much when they are ill and smoking makes you ill younger.

What I am saying is that even if the health care cost of dying from smoking equals the cost of dying from something else, if smoking ruins people lives before they retire. they will not be contributing to society.

The cost per son of health care may not change but the integrated lifetime productivity surely will.

Is this even close to sensible?

Chris J - "if smoking ruins people lives before they retire. they will not be contributing to society."

In the diagram with the blue circles, that would be shown as a smaller arrow going from smoking to income. That might be an additional reason why the relationship between smoking and health care expenditures looks like it does - when smoking goes down, national income goes up, which in turn causes health care spending to rise, because the only limit on health care spending is the amount of money available to spend.

It would be interesting to see how richer countries compare to countries in which tobacco is self-produced, and hence income should be largely uncorrelated to tobacco use. For example, in rural India and Bangladesh, people chew betel leaf, which is basically a large leaf wrapped around some home-grown tobacco. This is seen as a very "rural" or "villager" thing to do, so as incomes increase, people choose cigarettes as a more upscale alternative.

That might be one way to disambiguate.

RP = "countries in which tobacco is self-produced"

The original smoking data are here http://data.worldbank.org/indicator/SH.PRV.SMOK.MA.

Even in countries where tobacco can be self-produced, smoking will be a function of income. If tobacco is grown as a cash crop, then the farmer who smokes his own is smoking money. Even if tobacco is grown purely for own use, it has an opportunity cost: land devoted to tobacco can't be devoted to food crops.

I don't know which climates are suitable for growing tobacco. It's possible to grow tobacco in Canada, which makes me think it must be a crop that can be grown in a wide range of climates.

Chris J: "Surely the cost comes from people not being able to contribute as much when they are ill and smoking makes you ill younger."

True, but that's largely a private cost, borne by the individual smoker (through lower lifetime consumption brought about by early death). The "zombie" agument that Auld (and others) attack is that smoking imposes a social cost on people other than the smoker (i.e., through higher publicly funded health care costs). Non-smokers can legitimately complain about costs that smokers impose on non-smokers, but not about costs that smokers impose on themselves.

I suppose there might be a social cost through lower tax revenue brought about by the loss of smoker's earning capacity, but that's mitigated to the extent that (a) smokers die early, but still at the end of their working life (i.e., at the age of 65, say, rather than 80), so the foregone tax revenue from lost productive capacity is minimal, (b) smokers pay hefty taxes for the privilege of smoking (as they do in most developed countries), and/or (c) their early deaths reduce other government expenditures (CPP, old age security, etc.). I think the studies on this issue take that into account.

FR Woolley: "It's possible to grow tobacco in Canada, which makes me think it must be a crop that can be grown in a wide range of climates."

"Hey Tom, You ever been to Tillsonburg?/ Tillsonburg? My back still aches when I hear that word!" - Stompin' Tom Connors

Prof. Woolley, betel leaf is an important part of the South Asian economy: http://en.wikipedia.org/wiki/Betel#Economics. It has its own properties as a stimulant, but the people I know who chew it mix it with tobacco. India is the 2nd largest producer of tobacco on the world, and they grow it in West Bengal, which shares a border with Bangladesh.

I guess what I'm suggesting is that if we broaden the analysis to smokeless tobacco, we may be able to tease out some better results that are less dependent on income.

RP - yes, you're right, I don't see any good reason for excluding smokeless tobacco either. This shows the importance of diving down into the data.

Bob "smokers pay hefty taxes for the privilege of smoking"

This makes me wonder if there's an indirect linkage between health care expenditures on smoking, i.e. when governments spend more on health care there is greater need for revenue thus greater taxes on cigarettes (because cigarettes are dead easy to tax)thus lower rates of smoking.

"because cigarettes are dead easy to tax"

It isn't clear that that's always true (witness the experience in Ontario and Qubec with cigarette smuggling, which lead to sharp declines in the federal and provincial cigarette tax rates in the mid-1990s). Cigarettes are easy to tax, but they're also easy to smuggle, so it's a question of which is easier. Still, for some range of cigarette taxes, the effect you suggest might occur.

There's also the fact that, cross-nationally and within a country, these are very different goods. Cigarettes (or smokeless tobacco) are a simple consumer good, which a small number of people even make themselves. The change in demand as an economy grows is the like the change in demand for wine or watches. Healthcare as a good is organized by large systems that may start as small and local, but as an economy grows, healthcare becomes more complicated and acquires additional layers of regulatory costs, rent-seeking, and symbolic value as a status good. There are no cigarette payment plans. With a growing economy you may also get larger and more complex institutions (universities, regulators, professional associations, etc) that may be positive but which also encase the service in inflexible structures that make it less likely demand or costs will drop. And of course each consumer's individual choices and behavior doesn't affect healthcare as an industry in the same way that individual behavior may affect the demand for (and price of) cigarettes.

This is a perfect example that statistical significance does not entail explanatory power.

To follow on from Migeru above, I'm not sure that there's any justification for drawing that line through that graph of data.
Sure, you can retro-fit a curve to anything, but somewhere you have to ask if it's just a fitting exercise?

Metatone -"Sure, you can retro-fit a curve to anything, but somewhere you have to ask if it's just a fitting exercise?"

That's the point I'm making here: correlation does not imply causation, and this is the danger with doing anything with cross-country data. Spurious correlations abound.

The moral of the story? Don't conclude anything from regressions with poor explanatory power (R-squared of 11%). This has nothing to do with whether the regression is on country-level data.

Cigarettes are politically easy to tax because many people dislike smokers and want to stick it to them by making them stand around outside in the cold or suchlike.

Migeru - the r-squared is a very imperfect measure of how good a regression is. If I took time series data on health care expenditures and smoking rates - i.e. looked at trends in both in Canada over the past 30 years - I would get a much higher r-squared, but the regression would be no more informative than this cross-sectional one.

Cigarettes are politically easy to tax because many people dislike smokers and want to stick it to them by making them stand around outside in the cold or suchlike.

But they're hard to actually tax (at least above certain levels), since they're easy to smuggle and sell illegally.

You need both high significance and high explanatory power, is the point.

"Statistical significance alone, however, does not necessarily imply that test results are of much practical importance. The distinction between significant and meaningful results is crucial to a proper appreciation of hypothesis testing and should be emphasized." (source)

Significance measures Type I error, R-squared measures Type II error. You need to take both into account.

Migeru: Hang on, power measures type II error. Now, there is certainly a relationship between R^2, p-values and power that we can exploit. But, if we care about type II errors then shouldn't we go straight to the horses mouth, so to speak?

Furthermore, I think Frances' point is that even with a high R^2 and a low p-value, we still don't know anything about whether the relationship is causal or not.

The onset of diabetes from overweight and bad diet can also be used as an example. Avoidable behavior that degrades health has a significant cost to families, and in the case of poorer ones, definitely limits spending that might otherwise improve a family situation. Here's a WHO factsheet on the diabetes epidemic. http://www.who.int/mediacentre/factsheets/fs236/en/
Here's a brief excerpt.

"What are the costs of diabetes?

Because of its chronic nature, the severity of its complications and the means required to control them, diabetes is a costly disease, not only for the affected individual and his/her family, but also for the health authorities.
Studies in India estimate that, for a low-income Indian family with an adult with diabetes, as much as 25% of family income may be devoted to diabetes care. For families in the USA with a child who has diabetes, the corresponding figure is 10%.
The total health care costs of a person with diabetes in the USA are between twice and three times those for people without the condition. It was calculated, for example, that the cost of treating diabetes in the USA in 1997 was US$ 44 billion..."

If you want to talk about "correlation is not causation" you should choose an instance where there is correlation. R-squared of 11% is, well... not much of a correlation.

Perhaps we can have a discussion of the correlation/causation relationship between Frances' post and the attack of the killer spam bots? What's the trigger word? Is it health related discussions?

Bob - smoking.

Pretty ironic to use diabetes/overweight/bad diet in a discussion of "correlation does not imply causation" and "don't trust country-level statistics". There is a lot of evidence that the causality for diabetes is the other way: insulin resistance (to some degree genetic, certainly worsened by the carb-heavy Canada Food Guide in susceptible individuals) -> overweight -> diabetes and not the other way around. And the country-level statistics that were used to "prove" the lipid hypothesis (fat, with blame later redirected to saturated fat, causes heart disease) are about as good as those smoking ones plotted above. If you've done any amount of reading on the subject, and the politics behind it, you should throw up your hands in disgust.

For families in the USA with a child who has diabetes, the corresponding figure is 10%.

Fail. Type I Diabetes is NOT the same disease as Type II Diabetes; Type I is an autoimmune disease, Type II is a genetic/lifestyle disease. I have Type I Diabetes. The two diseases are very different under the hood.

Looking at Frances' post, the question comes back to WHY draw the line? There are only a handful of reasons to do any data analysis:
1) information gathering (Blind data mining and trying to guess if smokers in high-health expenditure regions are terrorists.)
2) forecasting (we don't care why cutting down the Amazon forest impacts shoe prices, but the data says it is a leading indicator.)
3) cause-and-effect changes (We can't directly impact B but we can impact A)

Depending on the reason, an R^2 of 11% is actually very good. If the only variable we can change accounts for 11% of the effect, that's worth knowing. (What effect does driving with the windows open have on fuel economy?)
If per capita health spending was only a function of smoking and completely random noise, then it would be good to know the true relationship that smoking had. (Policy makers could encourage people to smoke more because it eventually lowers health care expenditures.) However, there are good theoretical reasons to think that the random noise isn't random and health expenditures are tied to income.

In the absence of any other information whatsoever, looking at smoking rates tells you that up to a point, increases in the smoking are positively correlated and then negative. It contributes to #1. Knowing that, I am now curious to know if it holds within an smaller economy; do impoverished regions of Canada have both lower health care and lower smoking rates but high regions are higher?

Smoking is hateful to the nose, harmful to the brain, and dangerous to the lungs.


The comments to this entry are closed.

Search this site

  • Google

Blog powered by Typepad