12 Florida Agricultural Experiment Stations

the time, it was deemed unwise to run the risk of impairing
relations with company management by insisting upon testing
two prices higher than the current market price. The issue
was settled by testing one price above and three below the market
price. The actual pricing pattern imposed upon the sample stores
is presented in Table 2.
 All stores included in the sample handled three brands of
frozen orange concentrate in 6-ounce cans: a nationally adver-
tised brand (Brand A) selling at the highest retail price; a private
label (Brand B), 2 cents lower; and a packer's label (Brand C),
4 cents lower than the advertised brand. In addition, the co-
operating organization marketed Brand B concentrate in 12-
ounce containers.5

 THE STATISTICAL MODEL
 Intuitive appeal, computational efficiency and a desire to be
consistent with ex ante economic theorization about the prob-
lem led to the choice of a logarithmic "fixed unknown constants"
model for analyzing the data.6
 Along with price-stores, store traffic, weeks and price "age"
seemed to be the logical sources of variation that could be given
meaningful interpretation. However, it was assumed that the
necessity of considering store traffic as a source of variation
could be avoided by expressing purchases of concentrate on a
per customer basis.7

 "All frozen orange concentrate handled in the test stores was produced
in Florida.
 Seemingly, economic intuition would call for a rejection of the arithmetic
"fixed unknown constants" model, which would require that the demand
curves for the various stores differ by a fixed absolute value, i.e., be equi-
distant at all points. Suppose differences in concentrate sales among the
test stores could be attributed primarily to variations in the general income
status of the separate store clientele. Under these circumstances, the
assumption of a constant absolute difference in purchase rates among the
stores at any given price would imply that the income elasticity of demand
for concentrate could vary over an extreme range of values. For example,
 Aq I
the formula for estimating income elasticity, & -q, would become
c* Accordingly, income elasticity would vary inversely with
AIl q
quantity and directly with price along the demand curve.
 'It could be argued that the experiment would lead to biased estimates
of the demand parameters if customer choice of stores were affected by
the imposed variation in orange concentrate prices. However, a formal
test led to the acceptance of the hypothesis of zero correlation between
price of concentrate and store traffic (see Appendix II).