A pharmaceutical company wishes to study the consistency of
the amount of active ingredient in one of its prescription liquid
drugs (e.g. intravenous antibiotics). This drug is produced by two
manufacturing plants, one in Connecticut and one in Virginia. At
both plants, the liquid is manufactured in batches of the size to
fill large vats. The vats are then shipped to the same packaging
plant in Pennsylvania where they are subdivided and stored in
barrels. From the barrels, the liquid is put into intravenous drip
bags and sealed. Each drip bag is stamped with a code indicating
from which plant, which vat, and which barrel it came.

To examine the consistency of the amount of the active ingredient,
an analyst randomly selected four different batches from each of the
two plants, and then three barrels per batch. Three drip bags were
taken from each barrel, and the strength of the active ingredient
was measured (in parts per million, ppm). There are 72 observations
total.

This is a completely made up scenario, but the data (I am pretty sure)
came from the exercises in:

Neter, Kutner, Nachtsheim, and Wasserman (1996). Applied Linear
 Statistical Models, 4th Edition, Irwin Press, ISBN 0-256-11736-5.