206 The Marketing Book
responsibility. Some middle grades also tend to
consider it the responsibility of cashiers,
investment officers or everyone (all).
3 Managerial grades believe that this function is
mainly their own responsibility. These beliefs of
various role players within the branch are, of
course, of paramount importance, as it might
lead to under-training for certain function(s) at
grade levels, where customer contact is higher.
Therefore, this kind of study could focus the
training efforts and needs for specific selling
functions and certain grade levels/roles.
The six multivariate methods described above
are summarized in Table 9.1.
Regression and forecasting techniques
Multiple regression
Regression analysis attempts to investigate the
nature (and strength) of relationships, if any,
between two or more variables in marketing
phenomena. It can be used, for example, to
establish the nature and form of association
between sales and, say, the number of custom-
ers, the nature of competitive activity, the
amount of resources spent on advertising, etc.
The association between Y(sales) – which is the
dependent variable – and the independent
variables affecting sales are usually expressed
in a mathematical function of the type:
Y=f( 1 , 2 , 3 ,.. ., n)The purpose of regression is to make
predictions about scores on the dependent
variable based upon knowledge of independ-
ent variable scores (Speed, 1994).
Regression provides measures of associa-
tion, not causation; yet regression (and correla-
tion analysis) could assist marketing managers
in better understanding the implicit relation-
ships among various independent and depend-
ent variables (for example, age, income, educa-
tion and amount of credit card usage, or various
forms of salespeople’s incentives and their sales
calls, or the number of new orders obtained,
etc.).Generalized linear models
Generalized linear models (GLMs) have a
number of advantages over more traditional
‘hypothesis testing’ statistics: They are constituents of a unified theory of
data analysis (this makes the whole process of
choosing a test and understanding the analyses
easier).
They model rather than hypothesis test (they
can therefore be used for both description and
prediction).
They provide significant advantages over the
more traditional bivariate tests and in many
cases can replace them.
GLMs can be used to analyse most of the data
we are likely to collect in the social sciences.It is these advantages which make GLMs
perhaps the most important set of statistical
tests in the social sciences.What are GLMs?
Basically, they are methods which model data
using linear relationships. Linear relationships
are important due to the ease with which they
can be described mathematically. However, not
all relationships between variables are linear
and able to be adequately described in this way.
In such cases we can still take advantage of
linear methods by modelling the relationship
between variables using a linear model, but
including transformations which approximate
non-linear relationships to linear ones.
The techniques that fall under the GLM
umbrella include a number of popular
techniques: Ordinary least-squares (OLS) regression.
Logistic regression.
Log-linear modelling.