Data & Predictions (was: "The Fourth Turning" - A Must Read)

Robin Hanson (hanson@hss.caltech.edu)
Thu, 6 Mar 1997 17:33:49 -0800 (PST)


Lee Daniel Crocker wrote:
>"data" ... "not quantitative" is a contradiction. Data are numbers.
>Measurements. Surveys. Experimental results. If you can't measure
>it, there's no way to know whether your vague feeling of connectedness
>is something real and testable, or if it's just the wanderings of a
>random pattern-matching machine run amok, finding patterns caused by
>pure chance and baselessly extrapolating them into the future.

I responded:
> How do you ever walk into a new building and find room 225? After
> all, your eyes and ears aren't giving your numbers, so there is no
> way to know if your vague feeling that you should head toward that
> thing that looks like up stairs is something real and testable, rather
> than pure chance patterns. Do you really start writing down numbers?

Lee Daniel Crocker answered:
>If I wanted to make specific predictions about them, I'd have to have
>some way to test whether my actual results matched my descriptions ...
>What makes it specifically quantitative is that I can rigorously
>define what the measurement is, and I can repeat the experiment
>enough times to calculate a t-test or other measure of the probability
>that the measurements are chance or that they correlate to some cause.
>99% of the measurements you make are subconscious, so you're right
>that I don't have to count every step or label every room to get to
>where I'm going. But to make predictions, or to explain things to
>someone else, I'd better make them more explicit: ...
>I'm not discounting the value of human pattern-matching machinery.
>That's what gives us conjectures to test, and it is a very valuable
>skill. But until they are defined, measured, and tested, it is a
>very dangerous game to make predictions on pattern alone.

Now I'm confused. Are you saying that you can go about your life,
such as walking through new buildings or driving on new streets
without making any predictions based on all those subconscious
measurements you make? All it gives you are "conjectures to test"?
You don't predict "Hmm.. I think going up those stairs would probably
get me to the second floor"? Or are you saying you do predict but
that it is "dangerous"? (relative to what?)

And when I said that Strauss & Howe don't give quantitiative data, I
meant that they refer to relevant historical facts as evidence, but
that they don't map them onto numbers nor organize them in a way that
would make a statistical analysis convenient. This sounds like data
just as much as your visual perceptions of a building.

Robin D. Hanson hanson@hss.caltech.edu http://hss.caltech.edu/~hanson/