[From Bill Powers (950313.0915 MST)]
Samuel Saunders (950312:2340 EST) --
The data on Luis is interesting, but it would be more so without some of
the side-issues like color blindness! Why not go into the program and
change the cursor colors to two colors he can easily distinguish?
The low k-factors and the high RMS error suggest that Luis isn't
tracking very accurately. It might help if the score incremented about 5
times as fast on the average, and at a rate inversely proportional to
the error instead of just count/don't-count. If Luis gets the idea that
you get more credit for tracking closely, he might not allow so many big
errors.
It would be nice to disguise this experiment as some sort of game for
kids. Tracking for tracking's sake isn't very interesting except to
wierd PCT experimenters. Got any ideas?
I don't think the mouse acceleration will be very important in the
present preliminary investigations, but it's a real no-no when we're
taking data for posterity. The k-factor represents output sensitivity in
units of the rate of change of cursor position per unit error. The mouse
acceleration emphasizes fast rates of change (I think), moving the
cursor farther for fast movements than for slow movements (of equal
size). So it distorts the k-factor that we obtain.
However, the program contains its own mouse-reading program (very
elementary). I assume that the acceleration is a software feature, not
hardware. If so, there would be no acceleration when using our mouse.pas
unit. This, however, needs to be verified. If the acceleration is in
hardware or in the mouse driver we will have to eliminate it somehow.
You can check this quickly by seeing how far the cursor moves (in the
practice mode) when you move the mouse between two fixed stops (like
between two heavy books). The cursor movement should be the same whether
you move the mouse slowly or quickly.
In the interests of standardization, we need to have a mouse calibration
program. The way I've done this in the past is to use a routine that
asks the user to position the mouse on the zero of a ruler, hit space,
move the mouse exactly 6 inches, and hit space again. The resulting
count is stored in a tiny file; from then on the program tries to read
that file before asking for the calibration procedure. Then we can
specify a standard mouse sensitivity of 100 counts per inch, so that all
mouses used by everybody will be equivalent. I'll post that routine
later today, in a modified version of the mouse.pas unit.
I really think that the mouse is an inferior way of measuring behavior,
and that we should give some serious thought to going to some form of
joystick. The mouse slips now and then, and it measures movements in a
uncertain frame of reference -- relative to the hand instead of relative
to laboratory space. And now we have the problem of mouse "improvements"
like accelerators.
The worst aspect of a mouse is that people use it quite differently,
even within the same movement. The easy way is to let your hand rest on
the mouse pad and drag along as you move the mouse. Unfortunately, this
introduces unknown frictional forces. When you have a large movement to
make, you do it by rotating the upper arm, but for the final adjustments
or for correcting small errors you move the mouse from the wrist or just
with the fingers. Even within the same movement the output system can be
changing from one degree of freedom to another! This really makes it
difficult to do any physical analysis, maybe impossible.
A far better device is a crank mounted on a good-quality ceramic
potentiometer. Ideally, the pot would be mounted under the elbow with
the shaft vertical, and the crank would swing horizontally, with a
vertical handle on it grasped in the fist. This would measure a single
degree of freedom of movement (rotation of the upper arm about its long
axis) and external friction would be negligible. Alternatively, the
pivot can be horizontal, still located next to the elbow, with the
control movement being flexure at the elbow joint in a vertical plane.
While Tom Bourbon was employed, he was developing a sort of universal
system which would allow measurement of any single degree of freedom by
positioning the pot and crank relative to the body. Too bad.
Tom and I both have 4-channel analog-to-digital converter cards in our
computers which allow measurements with 12-bit accuracy. However, we
seem to be the only two who went that far. It is possible to plug a
couple of shielded wires from the potentiometer into two pins of a game
port and get quite linear and repeatable position measures of lower
resolution; this is what I recommend in practical terms, and I can send
instructions for how to set this up.
Now that we have a respectable number of programmer-experiments working
together all over the country, so we can't sit down at the same
apparatus, I think we should start paying attention to standardization
of equipment and calibration. If some student is going to replicate our
results 25 years from now, we have to specify the quantitative factors
involved in these experiments and demonstrations. The nearer we can come
to doing our experiments under exactly the same conditions, the more
meaningful will be any samenesses and differences we find.
Ideally, what we need is someone with access to a shop (or the required
personal skills) who can manufacture a bunch of cranks with long
shielded wires going to a connector ready to plug into a game port. The
mounting --- well, if there are any volunteers I'll send some drawings
of a setup that can be adjusted for direction of movement and length
from elbow to hand. I trust that we would all be willing to lay out 20
to 40 bucks for such a device, or whatever it takes. The cheaper the
better, of course. What say you all?
And what say you all to laying out $250 or so for A/D converter cards?
That's the best choice of all.
Rick, does a Mac have a game port that will accept a standard joystick?
ยทยทยท
-----------------------------------------------------------------------
Best,
Bill P.