[From Bill Powers (980216.1440 MST)]
Martin Taylor 980216 13:10--
Now, I _think_ I understand what you are
trying to say, but to check it, let me provide a paraphrase of the part of
your message that follows the disagreement about the historical root of
the topic of psychophysics.
Point 1. Psychophysical experiments cannot tell you the nature of the
perception being tested.
Point 2. The Test cannot tell you the nature of the perception being
tested, for the same reasons that apply to the psychophysical experiments.
So far, so good. All that we can say about the perception is in the form of
a model.
Point 3. The above invalidates any data obtained from psychphysical
experiments, at least insofar as they might be claimed to say something
about the subject, because not all the relevant variables might have
been considered the first time.
This is not what I have said. Some kinds of psychophysical data have been
obtained in circumstances where the subject was in control of the
perception, so whether the experimenter had a good model of what was being
perceived or not, the _subject_ certainly did, and we can be sure that the
controlled perception was in the state the subject intended.
Some psychophysical experiments satisfy most of the conditions of the Test,
and we can be reasonably sure that the results are valid. Those that did
not satisfy those conditions need to be reviewed.
To repeat, I have never said that all psychophysical experiments are invalid.
That is your interpretation. What I have said is that since we don't know
whether a valid procedure was carried out, in most cases, we have to look
at them all again before accepting any of them. Regardless of how many
thousands of papers have been written.
4. The above argues that the Test is a valid way to determine
something useful about the subject, because it can be done again if
later data suggest that not all the relevant variables had been considered
the first time.
That is not the only reason the Test is useful. The Test can reveal when
the subject is NOT perceiving what you thought was being perceived, and
that is something that open-loop experimenter manipulations can't do. If
you apply a stimulus open-loop and you get a response, all you can assume
is that the stimulus was perceived by the subject. There is no way to show
that some other stimulus was in fact affected by the first one, so it was
the disturbance of the second stimulus that actually led to the response.
Nor, in the usual approach, is there any indication that the response (even
yes/no) might be affecting something that is disturbed by the apparent
stimulus and stabilized by the response.
This is only to be expected. An experimenter who knew nothing about control
theory couldn't have been thinking in terms of controlled perceptions.
Control theory brings a new set of terms into the equation, and this
necessarily calls for some re-examination of old results.
I'm not sure why the idea of re-doing earlier experiments in the light of
new theoretical considerations evokes such strenuous opposition. Are we so
sure that these studies will stand up to a review that even to suggest
looking at them again is insulting? What IS the problem, anyway?
Best,
Bill P.
ยทยทยท
Have I got that right? I think I understand what you mean. And perhaps
why you say it. But I don't understand the logic used to construct the
argument.
Martin