A marine biologist is testing growth of flatfish.  Five flatfish are
placed in each of 10 tanks; the 10 tanks are randomly assigned to one 
of five temperature gradients (two tanks per temperature).  The fish
are allowed to acclimate for 17 days then their length and weight are
measured.  The length and weight are measured four more times at 12-day
intervals.

Identify a reasonable design for this experiment.  What would you 
choose for a response variable?  In actual practice, all of the fish
in one tank died after the second measurement period due to a hydroid
invasion.  In addition, three fish in another tank died after the
first measurement and were replaced by three other fish who were
measured on the second, third, fourth and fifth time periods.  How
do these changes affect your analysis?

Most students recognized this as a repeated measures design, with
time as the repeated measures factor.  The design should have tank
nested in treatment, and fish nested in tank and treatment.

Suggested responses included weight, length, density, BMI, and relative
weight gain.  Relative weight gain could be measured from the onset
of the experiment, or from the beginning of each new 12-day study 
period. Wet weight over length^2 is sometimes used as a measure of
growth.

Some of our balanced methods won't work because of the missing data.
We use this example as an opportunity to talk about Type III analysis
for unbalanced designs.