Applications in Market Research

During the past decade we have done substantial work on several projects where statistical models of market research have been developed. The projects have been supported by Procter & Gamble and Nielsen/BASES. The main activity was and still is concentrated around the following topics.

Mixed Poisson models of consumer purchase behaviour

Market ResearchWe have done a large study which demonstrates the applicability of Mixed Poisson models for describing consumer purchase behaviour (such as purchase incidence and brand choice) in markets. Research in the present direction includes numerical and theoretical study of the dynamical models leading to these schemes, study of probability distributions arising under different conditioning arguments, asymptotic properties of various statistical procedures concerning the parameters of the negative binomial and Dirichlet distributions. A large consumer panel containing the data about all purchase occasions of about 50000 households for 3 years is the main data set we use to test our models.

Price elasticity

Several models for studying price elasticity have been developed and applied for computing elasticities and cross-elasticities for a variety of household brands. The technique is based on a suitable data transformation and the use of robust regression techniques. The technique has been successfully applied to a variety of sales data.

Analysis of sales dynamics

The objective is to analyze and forecast time series related to sale quantities of certain products. The main difficulties of the problem are due to nonstationarity and small number of observations. The working methodology is an application of the Singular Spectrum Analysis and suitable dynamical models to multivariate trends to model interactions between different products.

Advertisement testing

Assume there is a list of M products from which some number N of respondents select the most preferable one before and after the advertisement of one of these products. The problem is to test the hypothesis that the advertisement is good enough to increase sales of the product up to a certain level. This seemingly easy problem can be approached via quite sophisticated statistical models describing the mechanism of consumer behaviour.

Multivariate statistical analysis of questionaires

The objective is development of a methodology of exploratory analysis of typical questionary data, construction of suitable parametric models and making statistical inference in them. The parametric statistical problems arising include inference in a mixed regression model.

Cardiff investigators