Experiments With Human Subjects

Robin Hanson (hanson@hss.caltech.edu)
Sun, 22 Sep 96 16:06:22 PDT


Ira Brodsky writes:
>By "objective," I mean that most reasonable people would agree the
>experiment is fair from the outset. By "repeatable," I mean any
>disinterested party could recreate the experiment and come up with the
>same results.

I'd say these criteria are met by most of the experiments in the
tradition I'm involved with (experimental economics).

>Nope, there is plenty of experimentation in astronomy and geology. Didn't
>astronomers prove that gravity can bend light waves through an experiment?

Um, no.

>In the physical sciences, you can usually test one variable at a time. I
>don't dispute your claim that often you need to vary more than one thing at
>a time. The point is you shouldn't start doing that until after you have a
>firm understanding (verified through experimentation) of what happens when
>you vary each thing alone.
>
>The problem with the social sciences is that there are external forces
>varying many things *all* of the time. It's virtually impossible to
>isolate any single factor. For example, how do you design a rigorous
>experiment to determine whether giving welfare to unwed, teenage mothers
>will reduce, increase, or have no impact on criminal activity among their
>children when they grow up?

You can bring a small number of people into a lab and run an
experiment on them, and then do it again tommorow and get similar
results. You can also vary just one aspect of your experiment.

Of course the new people are different, and even the same
people would be in different moods, etc. But you're going to say you
can't do social science because of that, you'll have to dump medical
science too. Same people, same variance, just asking different questions.

How do you run a rigorous experiment to say what any specific welfare
bill would do in any specific state? Maybe you can't. But that
doesn't mean you can't do other things.

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