A paper company produces cardboard tubes for holding spools of
paper.  In order to test the strength of the tubes, each shipment
is separated into three roughly equal parts based on time of 
production (early shift, middle shift, late shift); these strata
typically correspond to a truckload of tubes.  Five tubes from
each shift are randomly selected and then 3 rings are sawn from the
end of each tube and tested for strength.  The sample mean and 
standard error are then computed.

Is this an experimental or classification study?  Was selection of
tubes randomized properly?  Was selection of rings properly randomized?
Suggest alternatives if you're unhappy with any aspects of the design.

Since the tubes aren't randomly assigned to shift, this is essentially
a classification study.  Randomization of the tubes seems reasonable.
Note that if we made measurements directly from the tubes, the
analysis of effects would be the same as the analysis for a CRD.
Since the rings are simply cut off from one end of the spool, selection
of the rings is not properly randomized. Ideally, three non-overlapping
locations would be selected on each spool and rings would be cut from
each location.  Or the spool would be cut into rings and three rings
would be randomly sampled.  In this context, we have tubes randomly
sampled from shift and rings randomly sampled from spool.  They are
both (nested) random effects and the rings are said to be "sub-sampled"
from the tubes.  We'll examine designs like these a couple weeks hence.

Some students wanted to randomly select one ring from each of 15 
tubes--a reasonable idea, though it might be too expensive in practice.


Since manufacture of the tubes is a continuous process, samples could
occasionally be cut for testing.

Computation of a mean and standard error should be carried out as
for a stratified sample.