Experimental Pricing As an Approach to Demand Analysis 17
The particular technique utilized to estimate the missing
observations required the introduction of two new variables (and
two regression parameters) into the model. Quantities for store
nine, weeks one and two, were replaced by zeros and variables
defined as follows:
xk-i_= -1 for store nine, week one, 0 otherwise
xik-i= -1 for store nine, week two, 0 otherwise
Upon introduction of these new variables, the general model
for testing and estimation assumed the revised form:
YIk-j- = 0 + a + 8i + k + i+ 1iik-I- + P2X2k-i- + P31ik-i-
+ X4kk-l- + PXik-l- + P6X6ik-- + PTXik-i- + B8ki-+ 'ik-j-
In organizing the data preparatory to analysis, a certain
amount of difficulty was encountered with the matter of price
"age" classification. To begin with, the market price of 16.5t
per 6-ounce can of Brand B actually became effective one week
before the experiment began. From a temporal standpoint, then,
this price should have been identified with "age two." However,
because changes in the pricing unit, displays and availability of
the product were made at the beginning of the experiment, it
was decided to designate the first week of the experiment as a
component of "age one."
Unfortunately, the pricing design was the source of addi-
tional difficulty in delineating price "age" categories in that the
pricing arrangement prevented the specification of age classes
that were completely free from ambiguity.17 Despite this di-
lemma, the following "age" classifications appeared to be the
most logical: weeks one, two, five and seven were assigned to
"age one"; weeks three, six and eight, to "age two"; and weeks
four and nine, to "age three." is
Aside from the problem of "age" classification, the identifica-
tion of other categorical (or class) sources of variation was
"1 Hindsight allows the conclusion that the price effective in weeks one,
five and six should not appear in the pattern for weeks two, three and four,
nor in the pattern for weeks seven, eight and nine.
Apparently, notwithstanding the difficulties cited, this classification
is consistent with the notion that the demand function might possibly have
rotated about the existing market price as purchasers adjusted to the
induced price changes. It is only when demand curves are thought of as
shifting under the impact of price changes that the ambiguity of the "age"
classes becomes obvious.