[extropy-chat] Are vaccinations useless?

[Rafal Smigrodzki]

On 3/18/06, Robin Hanson <rhanson at gmu.edu> wrote:
>
> Rafal Smigrodzki  wrote:
> >BTW, the Rand study he quotes is junk.
>
> The RAND study is the single most informative study we have about the
> overall (marginal) health value of medicine in rich nations today.  I know
> Rafal has complaints about it, but one can find imperfections in any
> study.  I challenge Rafal to point to another study he thinks is more
> informative.   We could then compare flaws.

### Well, exactly, the Rand study is the single most informative study
we have on the effects of free health insurance on certain health
outcomes late in the last century. It is also a piece of junk.

My second most important point in this thread (aside from our pretty
extensive discussion of smallpox prevalence and vaccinations, which
you didn't include in the precis above) is that you do not have
sufficient aggregate information to judge the value of medicine today.
If you have to admit that the Rand study is the best you have, then
you are not justified to make any claims at all.

The main form of well-validated data with implications for medicine we
have are RCTs looking at specific interventions, which may be then
aggregated with information about prevalences - and these decisively
point to significant usefulness of  medicine (i.e. a difference of 3
to 5 years of life gained), if practiced based on the RCTs, as so
called evidence-based medicine. Yet, it appears that you discount
these data in favor of uncontrolled aggregate studies. Not
surprisingly, you arrive at estimates (life years gained =0 ) that are
clearly out of the mainstream (it's enough to read some of the
references on Robin's site to appreciate this fact).

It boggles my mind why you dismiss RCTs, while having a fondness for
aggregate correlational studies.

Rafal

PS. A brief and biased dictionary of terms:

RCTs - randomized controlled trials, the most trustworthy form of
medical (and generally scientific) evidence, because they allow the
exclusion of confounding variables. For example, you take 10,000
patients, randomly assign five thousand to take a sugar pill, give a
medicine to the rest, and count the ones that drop dead in each group.

Correlational studies - studies where only a correlation between
events is observed, and confounding variables are either ignored, or
controlled for post-hoc. For example, you count the number of pills
popped and check if the ones who pop more often die sooner - but of
course, you don't really know if the pills kill them, or do they pop
pills because they are sicker to begin with.

Aggregate correlational studies - you take two variables very far from
the individual events that make up life, like the amount of money
spent per capita on medical insurance, and the average lifespan in
various countries, and try to make your biases look inconspicous while
improvising explanations for the correlations you find.