[From Rick Marken (950122.2200)]
Bruce Abbott said:
In other words, it still doesn't matter, unless c=d+h+t, which it does,
so it does matter. Rick, when I'm right, I'm right. (;->
And I replied, plaintively:
But if c = d+h+t then t IS KNOWN; you added it!! There is nothing to
"estimate"
And Bruce Abbott (950122.1950 EST) retorts challengingly:
O.K., so YOU compute it.
Does the analyst need to know t in order to calculate d? Answer:
yes.
So I reply meekly:
I know that you have to know t and c and h in order to reconstruct d.
The problem was this: if you knew this then WHY did you ask the
question that set this all off, which was:
Do the PCT-analysis programs you have written estimate the
reference value from the data?
This question is puzzling even if we substitute "target value" for
"reference value". Why in the world would you imagine that we
"estimate" the value of t when we have it right there in the program?
The goal of reconstructing d is simply to save storage space so we know,
right off the bat, that if the program displays c at d+h+t, then we must
save c,h, and t in order to reconstruct d later. When you asked
whether we "estimate" t , Tom and I (independently) were surprised
because it suggested that you thought there was some "theoretical"
component to the reconstruction of d. The only alternative
explanation I can think of for your question is that, at the time you
asked it, you did not know that c = d+h+t.
How are my colleagues going to deduce that r is close to t if they are
not told the value of t? To steal a quote from Rick (and the National
Enquirer), "inquiring minds want to know."
The best way to deduce r is by looking at a time plot of c, d and h. If c
stays nearly constant while d and h mirror each other, then c is being
held at a constant reference value. If you label this graph in screen
units, your colleagues will be able to deduce the screen location that
corresponds to r. Of course, you must have t to properly scale d if you
plan to deduce that r (in screen units) is the intercept of the regression
equation relating d to h.
But, of course, we don't "estimate" t; we simply know what it is from
inspection of the program code. After inspecting that code, I discovered
that t = maxx/2. So it's half whatever the maximum pixal width of
your screen happens to be.
By the way, do I ever get to hear what you think these analyses of the
THREECV1 program data (particularly the multiple regression of
analysis of the proportion of variance in H accounted for by C1, C2 and
C3 when C3 is controlled) imply about the experimental and statistical
methods used in psychology? I've asked three times now. You're not
avoiding answering, are you?
Not only would it be nice to hear the opinion of an expert in psychological
research methods and statitics; it would be particularly nice to hear it
from someone who's not wrong when he's right;-)
Best
Rick