In a study designed to determine the reliability of size
measurements of tumors in cancer patients, two researchers recruited
26 oncologists to measure simulated tumors. The simulated tumors
were made of one of two materials and were one of three shapes,
which were chosen to physically resemble the texture and size of
tumors which are found in cancer patients. The small tumors were
spherical with a diameter of 31.5 mm, the large tumors were
spherical with a diameter of 38.3 mm, and the oblong tumors were elliptical 
with diameters 31.5 mm, 31.5 mm, and 38.3 mm.

Two copies of each simulated tumor were made; all were placed randomly
in rows on a folded blanket and then covered with a sheet of half-inch
foam. The oncologists then independently measured each tumor with their usual
equipment (ruler and calipers) and recorded the size obtained. ``Size'' is cross-sectional
area, which they define as the product of the longest dimension and
the shortest dimension of a tumor. By this measure, the sizes of the
simulated tumors are thus (31.5)^2=992.25 mm^2 for the small
tumor, (38.3)^2=1466.89 mm^2 for the large tumor, and 
31.5 x 38.3=1206.45 mm^2 for the oblong tumor. (Their actual cross-sectional 
areas were 980 mm^2, 1467 mm^2, and 1199 mm^2.)

The data set contains the following information:
 -- the response variable, size, in mm^2
 -- oncologist (numbered 1 through 26)
 -- model material (1=cork or 2=rubber)
 -- model shape (1=small, 2=oblong, or 3=large)
 -- replicate (two of each model type/shape combination).

These data appeared in the book:

Ramsey and Schafer (1997). The Statistical Sleuth: A course in methods 
of data analysis, Duxbury Press.