INTERLUDE
Now that weāve got through the preliminaries (overview of statistics, data and the retail industry) itās time to focus on what this book is really trying to do: give marketers insights into customer behaviour that will drive the business. Each chapter will start with a general marketing question and then provide an analytic answer to that question. Then a business case will detail the use and output involved in using marketing science to answer the marketing strategy question.
This means that, after the Chapter 2 introduction to regression and factor analysis, reading the book in order isnāt necessary. Generally, Iād suggest flipping through to find the marketing problem youāre dealing with and that chapter will give you an analytic answer to that problem. This means that each chapter is nearly stand-alone.
However, it is also true that the chapters on dependent variable techniques (eg regression modelling) are organized in increasing complexity: ordinary regression, logistic regression and Poisson regression, survival modelling and simultaneous equations. So if your marketing problem involved, say, multiple stages and the answer is simultaneous equations, while it is generally designed to āstand aloneā, you might want to review ordinary regression.
05
Understanding and estimating demand
Introduction
Business objective
Using ordinary regression to estimate demand
Properties of estimators
A note on time series data: autocorrelation
Dummy variables
Business case
Conclusion
MARKETING QUESTION
How can I estimate demand? What things impact demand?
ANALYTIC SOLUTION
Ordinary regression
Introduction
So our first question is a basic and very important one: how do you estimate demand? By ādemandā is typically meant (especially in retail circles) units or quantity. That is, the number of total items purchased. This is not the number of transactions or amount of net revenue but the count of items in the basket. Two packs of batteries count as two units even if they are only one stock-keeping unit (SKU).
Is there an issue in defining the dependent variable as units when one unit can include a set of AA batteries as well as a giant screen TV? No, not really. Regression is always āon averageā and a unit is a unit, on average. However the merchandizers construct the product hierarchy in defining units is what weāll use. Besides, what else can we do? Remember, a simplifying assumption always puts us on the right path.
(It is possible of course to add deeper granularity by making a demand model not just total units but units of...