Experimental Pricing As an Approach to Demand Analysis 25
difference between Remainder (5) and Remainder (3). The me-
chanics of the test are given in Table 6.
TABLE 6.- Testof H4 : 6 = 0.
Source D.F. SS MS F
Remainder (5)
(omitting 6: ) -------------- 59 0.927910
Remainder (3)
(including s) --..........--- 50 0.107511 0.002150
Additional reduction due
to fitting store
constants .--.............. 9 0.820399 0.091155 42.398
Similar tests for each set of class constants are presented
in Tables 7 through 9.
An examination of F ratios beginning with Table 6 indicates
that the inclusion of stores, weeks and ages in each case resulted
in a significant reduction in the remainder term. On the other
hand, as shown by Table 9, the additional reduction from fitting
store x age constants was non-significant.
Since the store x age constants did not materially contribute
to an explanation of the variation in concentrate purchases,
this set of constants was removed from the model. Adoption of
the revised model [form (C) with x i omitted] necessitated the
computation of new estimates of the regression coefficients.24
"2 From the standpoint of formal completeness, the exclusion of the inter-
action constants called for a re-examination of the model form. In
particular, it became necessary to re-open the question concerning the
distinctness of the separate age regressions, i.e., to re-test the hypothesis,
I = P2 = 3, P4 = P5 = P6. This amounted to testing the appropriateness
of a modified version of form (D), consisting of form (D) with the A;',
omitted.
Modified form (D) was rejected as a satisfactory model on the basis
of the outcome of the following test:
Factor D.F. Sum Squares Mean Sq. F
Remainder (1)
+ Store X Age
(Line G, Table 2, App. III) .. 72 6.555333
Additional reduction
due to modified (C) ............... 4 6.406250
Additional reduction
due to modified (D) ............... 4 0.007919 0.001980 <1
Remainder (9) .......................... 64 0.141162 0.002206