[From Bruce Abbott (960329.1300 EST)]
Chris Cherpas (960328.1635 PT) --
1. Control Limits
It seems as though the 25 points you plotted were the same
ones used for calculating UCL and LCL; is that the case?
Yes.
I usually think of control charts as taking 25 _prior_ points
for calculating the control limits and center and _then_
plotting new sample data to analyze. Otherwise, I'm not
familiar with how you use the limits to interpret the process.
I just co-opted Minitab's MACHART facility as a quick way to compute and
plot running averages. I'm sure one could use the charts as you suggest,
but it's not how Minitab creates these charts. It bases the mean and
control limits on all the data points; those that fall beyond the control
limits are considered to indicate that the process has gone "out of control"
as a particular place in the chart if more than a couple appear there in
sequence -- or at least that's my understanding.
2. Dispersion before center?
Also, you might consider looking first at an R-like-chart
(e.g., plot the ranges of the moving averages) before trying
to interpret your x-bar charts. This may make no sense
in this context, but I recall an SPC rule of thumb is
to not expect control in the x-bar until you see it in the range
chart first (of course, this is based on using subgroups).
Minitab produces R-charts, too, but again, my use of a control chart for the
weight data really had nothing to do with control charts per se.
3. Samples
Sounds like a hassle to me, but what if you got weights
five times per day? In other words, if the behavior that
"controls weight" varies throughout the day, one might
need to sample throughout the day. Compromise position:
if rats are weighed before and after each FR session,
then it might be a nice control to weigh the rats that
aren't being run yet twice a day -- at the same times
you'd be doing the before- and after- session weighings.
This might be a good idea for a formal study, but it would require somewhat
different housing arrangments to do it well, not to mention more assistants
than I currently have been able to bamboozle into helping me out.
I'm weighing all rats only once each day. The FR rats' food consumption is
being monitored, but to accurately get consumption from change in weight
would require somehow preventing the rat from peeing during a session, and I
haven't been able to find a cork small enough. (;->
But thanks for the suggestion!
Regards,
Bruce