On Mon, 18 Sep 2000 17:01:29 +0200, "Andrew Harvey (EEC)"
<[EMAIL PROTECTED]> wrote:
> I would appreciate any advice offered on a question of analysis method
> for an experiment we are planning.
> Having recently acquired eye-tracking equipment (point-of-regard on a pc
> screen), we want to test whether operators (in this case Air Traffic
> Controllers) are looking at their cursor when they click with the mouse.
- I think it would be more interesting to know what the fractions
are. How often does person P look at his cursor, when doing D?
If the fraction is high, it makes your followup problem an easy one of
continuously adjusting the calibration....
> We can record the position of the cursor on the screen each time the
> mouse is clicked. Thus for each operator we can collect a time history
> of (x,y) values for the cursor and (X,Y) values for the point of regard.
>
> Normally I would simply plan to test that mean(x-X) and mean(y-Y) were
> zero. However, there is a documented tendency for this sort of equipment
> to drift over time i.e. the precision of X,Y degrades.
I would expect "drift" to be a bias that can be corrected. That is
not the same as degraded precision -- measurement with an increasing
amount of noise, not bias.
> Is some sort of analysis of covariance with time as the covariant
> appropriate? Presumabley this can only work if the drift function is
> linear in time?
> Or am I barking up the wrong tree completely?
>
If you can re-anchor your calibration, anchoring is what you should
do. If your error-of-measurement is truly getting worse over time,
you might need some sort of weighted analysis, but I think you want to
avoid that, if you can.
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
