Two Questions for everyone

[Troy Kelley]

Hope everyone had fun in Europe while I was slaving away in the States!

I am attempting use ACT-R to predict interface errors made by someone using
a Helmet Mounted Display (HMD) while walking through the woods.  The HMD is
a fold-down type display that soldiers can wear on their helmets.

I have already collected the error data.  But I will not let myself see the
actual errors in the data, instead I would like to predict errors that were
made using ACT-R.

This creates a "model-fitting" problem because I cannot "tweak" my
parameters to fit my error data, because I do not know what my final errors
are going to be.

Question 1:  I was wondering if there is a list of standard defaults used
by other modelers for all the "tweakable" ACT-R parameters which I could
use to get an average from.  I am especially interested in Base Level
Learning, Activation Noise, Retrieval Threshold and the Base Level
Constant.  I have seen TABLE 8.4 on pg. 291 in Atomic Components of
Thought, but was wondering if there was something else.

Question 2:  My model has time between recall events where memory decay can
occur.  Most of the time between each recall event is measured in seconds.
Currently I have my cycle time set to 50 ms, and if my calculations are
correct, I would need 20,000 50ms cycles to equal 1 second.  This could
make for a very large model but it is possible to run this way.

However, it seems that even if my cycle time was 50ms or something larger,
that the general decay function would remain the same, or is this wrong?
In other words, stretching the X axis of the decay function by manipulating
certain parameters doesn't matter as long as the general shape of the
function remains the same.  This seems like what ACT-R allows you to do by
manipulation of the decay parameters, change the X axis of the decay
function.  However, this also seems to be counter-intuitive, especially
when I attempt to explain it to other people.  Any ideas on this?

Troy Kelley
ARL