[From Rick Marken (2013.09.21.0950)]
John Kirkland (20130921)--
Note that I have changed the subject line because I think your post is
relevant to the question of the usefulness of "conventional"
psychological research to understanding organisms (particularly
humans) as purposeful (control) systems, a topic that we might discuss
when we come to Ch 16, on Experimental Methods. This question has been
the focus of my own work on PCT for the last 30+ years because I come
to PCT out of a "conventional" research tradition. One of my latest
papers (Richard S. Marken (2013) TAKING PURPOSE INTO ACCOUNT IN
EXPERIMENTAL PSYCHOLOGY: TESTING FOR CONTROLLED VARIABLES.
Psychological Reports,112,184-201) describes my current perspective on
the relationship between PCT and conventional experimental research in
psychology. My interview gives a good summary of my views:
Your post is interesting, John, because the "conventional" methods you
are discussing are not the experimental methods I've been tilting
against for the last 30 years. They are multidimensional scaling
methods, with which I am fairly familiar (having been trained in
perception and psychophysics), but which may be less familiar to
others in the "course". My quick summary of these methods is this: MDS
consists of mathematical procedures that are used to place measures of
the judged "distance" between pairs of items (words, pictures, sounds)
into a multidimensional space with as few dimensions as possible. For
example, you can ask a person to judge the distance between pairs of
states in the US. Then you can use multidimensional scaling methods to
fit these distances into as low a dimensional space as possible; with
distances between states you probably need only 2 dimensions to
account for all the distances and what you will see is a literal "map"
that could be thought of as representing the psychological map of the
US that exists in the subject's head.
So what's the problem with this approach, from a PCT perspective. This
would take a whole paper in itself but let me give you the start of an
answer by replying to your post. You say:
JK: Here's what I'm pondering over.
When we aggregate sorting data there is a consistent MDS map obtained,
usually rendered into 3D...
RM: This is a very general statement but my first question is " why is
this an interesting finding"? Did you have a reason to expect that the
MDS maps would be inconsistent; have more or less than 3 dimensions?
The problem here is that I don't know what model this research is
testing; research makes sense to me when it is aimed at testing a
model of the phenomenon under study. I don't understand what model
(of perception, I presume) is being tested by this research. Another
problem I have is that the analysis is being done on aggregate data --
averaging over people. PCT is a model aimed at finding general laws
that apply to all individuals, not to the average individual. I know
that MDS can be done on one person at a time; I would find that data
more interesting than the aggregate data, but only if it were clearly
testing a model of perception.
JK: Here's a factoid. These maps tend to be highly consistent.
RM: Yes, you said that. But why is this surprising? Is there some
model that predicts inconsistency? Have you looked at what MDS
solutions look like for randomly generated measures of distance? Does
such a test need as many dimensions as there are items?
In order to determine the relationship between the results of MDS
analyses and the PCT model of perception I think you would have to
determine how the PCT model would behave in the kind of tasks used to
obtain the data that goes into an MDS analysis. That is, you would
have to show how the PCT model would judge, for example, the distance
between states from memory. If you could do that, then you could
compare the results you get from an MDS analysis of the human data to
those you get from an MDS analysis of the PCT model generated data. Of
course, this would have to be done with individual subjects but MDS
could, then, be used to test the PCT model of perception.
Anyway, there are a lot of complex issues here. But I think the
fundamental issue here is MODELING! From a PCT perspective, the
meaning of research results depends on how they compare to those
expected from a model -- best of all a working model. MDS is not,
itself, a working model, by the way. It is a way of _describing_ data;
it is really a fancy form of what I call _curve fitting_, which is
often thought of as a kind of modeling, but it's not the kind of
modeling we do in PCT. Curve fitting is a way of _describing_
characteristics of overt behavior. In other words, it's a way of
_looking at_ behavior -- what a person is doing. And as we will find
out as we continue wending our way through B:CP, you can't tell what
people (control systems) are _doing_ by just looking at what they are
doing -- even if you look at what people are doing through the
seductive lens of MDS.
Richard S. Marken PhD