[from Jeff Vancouver 990930.0950 EST]
Marc,
That's up to you. You need to come back with _3_ things.
Nicely laid out - easy to respond to.
1) An assumed CV
2) 3 levels ( take your pick ) and aspects of the CV that reside at each
level.
I think you are actually asking for at least 3 CV's: one for each level.
And if a level has more than one unit, it may have more CVs relevant to that
level. However, I think 2 levels will due for this model, so I will specify
the CVs I hypothesize and how to test for them.
The higher-level CV is "cost of schedule." I have already conducted a study
which tests for this CV by disturbing cost, blocking cost, and observing
results. The jist of this study has been posted to this net, but I am happy
to send you the article I have under review about it.
The lower-level CV is "deleted time block." In order to get the cost of
schedule below the goal I assign participants, they must delete time blocks
for the RNs scheduled (assuming the schedule is over budget, and they have
accepted and are controlling for the cost of schedule CV at the level
assigned [i.e., the cost CV is disturbed]). To delete a time block, they
right-click on it, which evokes a dialog box asking if they are sure they
want to delete the time block. I planned to not always evoke that dialog
box as a method of blocking the deleted time block CV (I was going to record
the right-click with a time stamp). I also thought that I would ask if they
right-clicked (on some trials) so I could show that they knew they did the
behavior, but were not satisfied until they got the response (deleted time
block).
3) The combination of modes ( as proposed by Bill or yourself ) that
account
for the "information" flow ( perceptions and reference levels ) between
levels.
Okay, this is clearly the tricky part. What I am thinking is that we can
model (and test) two of the modes. First, what I call the assumption mode
(it most corresponds to Bill's automatic mode, but not quite). The idea is
that I might not have a signal from the environment that represents
accurately the current state (or that represents it, but does not take into
account that the actions I have taken take time to be realized because of
lag in the environment). But I know in memory that the actions, which are
really the perceptions of lower-level systems (e.g., perceptions of deleted
time blocks), result in changing the higher level perception (e.g., cost of
schedule).
So, for example, if we tell participants that the scheduling program does
not always function well in that sometimes when you delete a time block, it
does not update the cost of schedule information, but that you can be
assured that the deletion, and hence real cost of schedule will be
appropriately affected, then participants would evoke this mode of control
(phew, long sentence). They would act as if cost was being effected by
actions even though they could not see the results on the higher-order
system, only the lower-order system. To model this, we simply need to have
a connection between the perception of deleted time block (the lower-level
perception) to the input function of the higher-level unit, which currently
has a direct connection to the environment. Seems pretty easy to me. What
I mean by "assume memory is in place" is that the weight attached to the
connection between the perception coming from the lower-level input function
to the higher-level input function is not zero. The tricky part is
controling the weight of that connection. The tricky part I eliminated is
accounting for it in the first place.
The other mode is imagination. We would simple ask them to tell us what
happens to the cost of schedule if they delete a time block. To model this,
we need a connection from the output function of the higher (or lower?) unit
to the input function of the higher (or lower?) unit. Again, the issue of
the weight of that connection and when it is not zero is the issue.
I do not expect that I have answered all your concerns (I have not answered
all mine). This is an iterative process.