Quantitative methods in marketing 213
order to be able to predict the dependent
variables. It splits the sample of observations
into two groups on a sequential routine, trying
to keep the subgroups that emerge as homoge-
neous as possible, relative to the dependent
variable. The homogeneity is measured by
minimizing the sum-of-square deviations of
each subgroup member from its subgroup
mean. AID is used in marketing for market
segments analysis, analysing the effect of
advertising levels on retail sales, predicting
consumption/sales and brand loyalty.
The method is not as powerful as regres-
sion analysis and, since the minimum subgroup
size should be no less than 30, the original
sample of objects required must be fairly large
(1000 or more).
Three of the regression and forecasting
techniques described above are summarized in
Table 9.7.
Forecasting methods
Forecasting methods are mainly applied in
forecasting sales and market demand. Cham-
bers et al. (1979) classify them into three
categories: qualitative techniques, time-series
analysis, and causal models. In each category
there is a series of models; some are suitable for
forecasting initial sales and others for forecast-
ing repeat purchases. Consequently, one should
make clear the differentiation between diffu-
sion and adoption models, although, unfortu-
nately, the space available here is not sufficient
for a detailed presentation.
Probably the most well-known forecasting
techniques are the time-series methods. These
rely on historical data and, by definition, are of
limited application to the forecasting of new
product sales.
In order to select a forecasting technique for
new products, the first principle is to match the
methodology with the situation. The degree of
newness of the product, for example, is crucial,
as are product and market characteristics, the
forecaster’s ability, the cost, the urgency and the
purpose for which the forecast is needed.
The second principle is that at least two
methods should be used and one of these
should always be the subjective judgement of
the forecaster, who must override the formal
technique decision when information coming
from outside the model clearly shows that the
technique’s forecast may be at fault. There are
powerful arguments for combining forecasts by
different techniques. Methods are selective in
the information they use, so that a combination
of methods would incorporate more informa-
tion and improve accuracy. Doyle and Fenwick
(1976) advocate this and produce evidence of
improved accuracy.Simulation methods
The cost, the time involved and other problems
associated with field experimentation often
preclude a method as a source of information
for particular situations. In such instances it is
often desirable to construct a model of an
operational situation and obtain relevant infor-
mation through the manipulation of this model.
This manipulation, called simulation, describes
the act of creating a complex model to resemble
a real process or system and experimenting
with this model in the hope of learning some-
thing about the real system.
Simulations represent a general technique
which is useful for studying marketing systems
and is one of the most flexible methods in terms
of application. Simulation models have been
formulated to serve two management
functions:1 Planning.
2 Monitoring and controlling operations.Marketing simulations can be conveniently
divided into three classes (Doyle and Fenwick,
1976). The first deals with computer models of
the behaviour of marketing system compo-
nents, the second with computer models on
the effect of different marketing instruments
on demand, and the third with marketing
games.