[ACT-R-users] Modeling ideas

[Kelley, Troy]

Wolfgang,

Thanks for your reply.


-That depends on the reason why the two types of stimuli are differently
recognizable. If it is perceptual conspicuity -----(such as number of
discriminating basic features), you either have to use PM or estimate
time and probability of success ---for the basic perceptual processes.
If it is differences in complexity (such as number of parts),
differences 
-can be modeled in the cognitive layer. Categorizing more complex
stimuli may take more steps (production cycles).

These are two different types of targets that soldiers are shooting at,
one is friendly and one is enemy.  There are 12 friendly and 12 enemy so
they are equally distributed.  There are two types of friendly, one with
a large round white circle in the middle of the target, and the other
friendlies are marked with a large gray circle in the middle of the
target.  The enemy is a flat green color with no circle marker.  There
is only one type of friendly target presented in each condition (either
white or gray).

 From the data it is clear the the gray friendly targets are harder to
recognize than the white friendly targets and they produce more friendly
fire errors.

I would like to use your last suggestion, that complex stimuli take more
steps to recognize, more production cycles, that would get me away from
using PM right now.  Not that I don't want to use PM but I am under a
time crunch for this model so the less programming the better ;-)
 
> Also, I have some data that shows a workload effect on performance 
> after a certain critical workload level has been reached, which then 
> causes performance to level off after the critical point has been 
> reached.  So, I have 3 conditions (easy, medium and hard) and 
> performance is good during the easy condition, but it changes to be 
> poor performance for both the medium and hard conditions.  So the 
> workload reaches a critical point during the medium condition, which 
> negatively effects performance, and essentially stays that way for the

> hard condition as well.  Any ideas on how to model this - "change in 
> performance after a certain critical point has been reached" - would 
> be appreciated.

-That looks like a threshold effect, which can easily be modeled using
the dynamics of source spread and retrieval -theshold. More workload -
-usually means there are more chunks sources of activation and
therefore, each of them spreads less activation into the ----network. So
with growing workload the potency for being a retrieval cue goes down
for the 
-activation sources. Together with the retrieval threshold this can 
-produce the "critical point" effect.

Yes, I have spoken with Christian before about modeling workload as
spreading activation, but I have never actually done it.  I was also
having trouble understanding how this would be a thresshold effect as
well.  Seems as if it were spreading activation, it would effect the
performance in a linear fashion, with no drop off point.

Troy