16 Florida Agricultural Experiment Stations
ANALYTICAL PROCEDURE
The structure of the general model adopted to explain the
demand for concentrate was such that techniques of multiple
covariance could be readily applied as an analytical tool.14 In
fact, the model was constructed with covariance analysis in mind.
Consequent computational requirements of the chosen frame of
reference and statistical procedure involved deriving least-squares
estimates of the model parameters and determining the "best"
specific model form that would summarize the demand for con-
centrate for the given data. Decisions regarding the choice of a
specific model were to be made by testing certain hypotheses
about the model parameters. Composite hypotheses to be tested
were:
(1) P1 =2 =3 = 0, (2) P4 = 5 6 = o
i.e., do the regressions contain only a linear component or both
a linear and quadratic component; and, do the individual age
regressions differ significantly?
(3) ai = 0, 61 = 0, V"k- = 0, 'ii 0,
i.e., do the specified "class constants" constitute significant
sources of variation?
ORGANIZATION OF THE DATA
Before the formal analysis could be undertaken, however,
the matter of performing necessary corrections and properly
organizing the data required consideration. Inspection of the
data and a knowledge of events which occurred during the course
of the study led to the rejection of the quantity data from one
store for the first two weeks.15 To obtain estimates of the omit-
ted observations and preserve the computational simplicity of
the original model, "missing data" techniques were applied.16
"14 R. L. Anderson and T. A. Bancroft, Statistical Theory in Research
(New York: McGraw-Hill, 1952), Chapter 21.
W. G. Cochran and G. M. Cox, Experimental Designs, (New York: John
Wiley & Sons, 1950), pp. 75 ff.
A. M. Mood, Introduction to the Theory of Statistics (New York:
McGraw-Hill, 1950), pp. 350 ff.
M. G. Kendall, The Advanced Theory of Statistics (3rd ed.; London:
Charles Griffin and Company, Ltd., 1948), II, pp. 337 ff.
"1 For a portion of week one, in store nine only, the company's private
brand of concentrate was not available. It seemed reasonable to assume
that the missing brand might affect the data from this store for both
weeks one and two.
16 M. S. Bartlett, Some Examples of Research in Agriculture and Applied
Biology, Supplement to the Journal of the Royal Statistical Society, IV,
No. 2 (1937).