[From Bruce Abbott (950212.1740 EST)]

Although no one has asked whatever happened to ol' what's-his-name, I
haven't gone away. My wife's mom suffered a coronary a week ago Saturday
and, as you might imagine, things have been a little topsy-turvy since then.
Balloon angioplasty saved her; it was fascinating to watch the x-ray tape of
the procedure. We hope to have her home near the end of this week if there
are no further complications.

I drop out for a while and CSG-L degenerates into philosophical debate--and
I missed all the fun... I hope nobody minds if I discuss a bit of
PCT-oriented research instead. (;->

In the few spare moments I've had I've managed to get a PCT version of the
inverted-t illusion working, with some interesting results. The screen
shows an inverted T whose horizontal line (the reference length) expands and
contracts symmetrically from its center. The participant's job is to keep
the vertical line (equivalent to the cursor in more typical PCT displays)
the same length as the reference. What typically happens with the
inverted-t illusion is that people make the vertical line too short, and
that is what happens here. It would appear that the error is a percentage
of the reference line length. The standard model with integrating output
function tends to yield a good correlation but terrible RMS error because of
the effect of the illusion. Linear regression does better, usually cutting
RMS error in half; this can be improved further by lagging the reference

I suspect that doing continuous adjustments as needed to track the varying
reference length just adds noise to the assessment of the illusion; in
addition to the perceptual error there are dynamic tracking errors that
factor in. The usual procedure is to start with the adjustable line
randomly longer or shorter than the reference and give the participant all
the time he or she wishes to complete the adjustment. Errors are averaged
over a number of trials.

On the other hand, the tracking procedure does allow you to collect data
over a wide range of target line lengths rather quickly. Perhaps what is
needed is to reduce the dynamic error by using a more slowly varying reference.

How about a PCT model that will give an improved account of performance on
this task? Among other results, it should yield the size of the perceptual