poly: Why we die?

rcs >The study about the causes of death contains enough things
>that contradict "well-known" health information that I am
>wary of accepting it as accurate: At the minimum, the authors
>and I must differ on vocabulary.

robin>What contradictions do you have in mind? (I don't see them.)

Sex Male Female
           1.0 .41

We know that women live longer than men, but the death ratio of 2.5
is hard to believe: What averaging method was used to get this number?

Income 30K$+ 10-29K$ <10K$
            1.0 2.14 2.77

Is this a raw statistic? Where do social security recipients fall in
the distribution? How is drawdown of an IRA counted -- is it income?
What about resources supplied by children?
We know that serious illness can be financially ruinous -- does the
above table merely reflect this fact?

Smoking Never current former
            1.0 1.26 1.28

In the conventional wisdom, quitting smoking improves your health.
The above table sez it doesn't matter.

Body Mass Normal Underweight Overweight
             1.0 2.03 .94

Much as I would like to, I don't believe that being overweight lowers
risk of death.

rcs >[If taken at face value, it implies that any other study
>about causes-of-death must *exactly* match control & target
>populations for wealth, to avoid swamping the more subtle
>effect being studied.]

robin>It depends on the strength of the effect of interest.
    Random clinical trials, which seek relatively strong effects,
    can typically implicitly control for a wide range of
    not-too-strong effects such as wealth. When looking for
    weaker effects, researchers *have* regularly controlled for
    wealth.

>From the table, low-income has a relative risk of 2.77. Only age
is much more important, with exercise and sex comparable. But
smoking, drinking, education, weight, rurality, and race are all
less important. If I'm trying to determine how smoking affects
longevity, I have to untangle the income/smoking correlation
well enough to discover the relative smoking risk of 1.26.

rcs >The various factors studied are mostly strongly correlated,
>(wealth & age being an obvious example), and even defining
>single variable effects is tricky.

robin>I think you will find that these researchers are well aware
    of these issues, and have done a decent job addressing them.

Maybe it's unfair of me to attack a summary table, but a simple
reading would seem to say that I could calculate my relative risk
by multiplying 3.46 * 1.0 * 1.0 * 1.52 * 1.0 * 1.0 * 1.0 * 1.13 *
 .94 * 2.91. I doubt that's the researcher's intent, but what's
the point of a table like this? If I increase my income, but
give up exercise, will I live longer?

Rich Schroeppel rcs@cs.arizona.edu
Received on Mon Jun 29 21:36:31 1998