ENBIS-8 in Athens

21 – 25 September 2008 Abstract submission: 14 March – 11 August 2008

Latent Class Models for Marketing Strategies: an Application to Promotion in the Pharmaceutical Sector

22 September 2008, 10:35 – 10:55

Abstract

Submitted by
Francesca Bassi
Authors
Francesca Bassi
Affiliation
University of Padova, Italy
Abstract
Some extensions of the latent class (LC) approach are applied to analyze the Italian pharmaceutical market. Available data were collected in a survey on Italian general practitioners. All doctors were asked to express a judgment on various aspects regarding promotional activity – on a seven-point scale - organized by the pharmaceutical industries with which they were in contact and to declare the percentage of drugs produced by each firm which they usually prescribe.
LC models for multilevel data were applied in order to identify market segments, i.e., groups of doctors with similar attitudes toward pharmaceutical representatives’ activities. A second aim of the paper is to verify which aspects of firms’ promotional activity may be determinant in influencing doctors’ prescriptions. LC regression models were estimated for this purpose.
Traditional LC models assume that observations are independent. However, in our case this assumption was violated since doctors judged more than one pharmaceutical industry; multilevel LC models make it possible to modify the above assumption. A multilevel LC model consists of a mixture model equation for level-1 and level-2 units, in which a group-level discrete variable is introduced so that parameters are allowed to differ across latent classes or groups. Our level-1 units were judgments expressed by doctors on the seven aspects of the promotional activity; our level-2 units were doctors. We were interested in defining clusters of doctors (classes).
LC regression models estimate a linear relation between a dependent variable and a set of explanatory variables, accounting for the fact that observations may arise from a number of unknown heterogeneous groups in which regression coefficients differ. LC regression models can be viewed as random-coefficient models that, like multilevel or hierarchical models, can take into account dependencies between observations. This extends the application of LC regression models to situations with repeated measurements. In our case, the dependent variable was the percentage of drugs produced by a certain pharmaceutical industry prescribed by practitioners, and predictors were the judgments expressed by doctors on the seven aspects of promotional activity.

Return to programme