dirk morschett, thomas foscht, thomas rudolph, peter

Dirk Morschett, Thomas Foscht, Thomas Rudolph, Peter Schnedlitz, Hanna Schramm-Klein, Bernhard Swoboda (Eds.)

European Retail Research

GABLER RESEARCH EditorsDirk Morschett, University of Fribourg, Switzerland, [email protected] Foscht, University of Graz, Austria, [email protected] Rudolph, University of St. Gallen, Switzerland, [email protected] Schnedlitz, Vienna University of Economics and Business, Austria, [email protected] Schramm-Klein, Siegen University, Germany, [email protected] Swoboda, University of Trier, Germany, [email protected]

EDITORIAL ADVISORY BOARDIn the editorial advisory board, a number of distinguished experts in retail research from different countries support the editors:

– Steve Burt, University of Stirling, UK– Michael Cant, University of South Africa, South Africa– Gérard Cliquet, University of Rennes I, France– Enrico Colla, Negocia, France– Ulf Elg, Lund University, Sweden– Martin Fassnacht, WHU - Otto Beisheim School of Management, Germany– Marc Filser, University of Dijon, France– Juan Carlos Gázquez Abad, University of Almeria, Spain– Arieh Goldman, Hebrew University, Israel (†)– David Grant, University of Hull, UK– Andrea Gröppel-Klein, Saarland University, Germany– Herbert Kotzab, Copenhagen Business School, Denmark– Michael Levy, Babson College, USA– Cesar M. Maloles III, California State University, USA– Peter J. McGoldrick, Manchester Business School, Manchester University, UK– Richard Michon, Ryerson University, Canada– Dirk Möhlenbruch, University Halle-Wittenberg, Germany– Heli Paavola, University of Tampere, Finland– Luca Pellegrini, IULM University Milan, Italy– Barry Quinn, University of Ulster, Northern Ireland– Will Reijnders, Tilburg University, The Netherlands– Thomas Reutterer, Vienna University of Economics and Business, Austria– Jonathan Reynolds, Oxford, UK– Sharyn Rundle-Thiele, University of Southern Queensland, Australia– Brenda Sternquist, Michigan State University, USA– Gilbert Swinnen, Universiteit Hasselt, Belgium– Ikuo Takahashi, Keio University, Japan– Waldemar Toporowski, University of Goettingen, Germany– Volker Trommsdorff, Technical University Berlin, Germany– Gianfranco Walsh, Koblenz-Landau University, Germany– Barton Weitz, University of Florida, USA– Joachim Zentes, Saarland University, Germany

Dirk Morschett, Thomas Foscht, Thomas Rudolph, Peter Schnedlitz, Hanna Schramm-Klein, Bernhard Swoboda (Eds.)

European Retail Research 2011 I Volume 25 Issue I

RESEARCH

Bibliographic information published by the Deutsche Nationalbibliothek

The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografi e;

detailed bibliographic data are available in the Internet at http://dnb.d-nb.de.

”Jahrbücher zur Handelsforschung“ were fi rst published at:

Physica-Verlag (1986-1988)

Gabler Verlag (1989-1999/2000)

BBE-Verlag (2000/01-2004)

Kohlhammer Verlag (2005-2007)

The 25th Volume Issue I is sponsored by

1st Edition 2011

All rights reserved

© Gabler Verlag | Springer Fachmedien Wiesbaden GmbH 2011

Editorial Offi ce: Stefanie Brich | Sabine Schöller

Gabler Verlag is a brand of Springer Fachmedien.

Springer Fachmedien is part of Springer Science+Business Media.

www.gabler.de

No part of this publication may be reproduced, stored in a retrieval system

or transmitted, in any form or by any means, electronic, mechanical, photo-

copying, recording, or otherwise, without the prior written permission of the

copyright holder.

Registered and/or industrial names, trade names, trade descriptions etc. cited in this publica-

tion are part of the law for trade-mark protection and may not be used free in any form or by

any means even if this is not specifi cally marked.

Umschlaggestaltung: KünkelLopka Medienentwicklung, Heidelberg

Printed on acid-free paper

Printed in Germany

ISBN 978-3-8349-3093-4

V

Preface EUROPEAN RETAIL RESEARCH is a new bi-annual that is in the tradition of the reputable book series “Handelsforschung” (Retail Research) which has been published by Prof. Dr. Volker Trommsdorff in Germany for more than two decades. Since 2008, this publication is edited by a team of retail researchers from Austria, Germany, and Switzerland. With this issue, the initial team is complemented by Thomas Foscht from Austria.

The aim of this book series is to publish interesting and innovative manuscripts of high quality. The target audience consists of retail researchers, retail lecturers, retail students and retail execu-tives. Retail executives are an important part of the target group and the knowledge transfer be-tween retail research and retail management remains a crucial part of the publication’s concept.

EUROPEAN RETAIL RESEARCH is published in two books per year, Issue I in spring and Is-sue II in fall. The publication is in English. All manuscripts are double-blind reviewed and the book invites manuscripts from a wide regional context but with a focus on Europe. We respect the fact that for many topics, non-English literature may be useful to be referred to and that retail phe-nomena from areas different from the US may be highly interesting. The review process supports the authors in enhancing the quality of their work and offers the authors a refereed book as a pub-lication outlet. Part of the concept of EUROPEAN RETAIL RESEARCH is an only short delay between manuscript submission and final publication, so the book is – in the case of acceptance – a quick publication platform.

EUROPEAN RETAIL RESEARCH welcomes manuscripts on original theoretical or conceptual contributions as well as empirical research – based either on large-scale empirical data or on case study analysis. Following the state of the art in retail research, articles on any major issue that concerns the general field of retailing and distribution are welcome, e.g. - different institutions in the value chain, like customers, retailers, wholesalers, service compa-

nies (e.g. logistics service providers), but also manufacturers’ distribution networks; - different value chain processes, esp. marketing-orientated retail processes, supply chain proc-

esses (e.g. purchasing, logistics), organisational processes, informational, or financial man-agement processes;

- different aspects of retail management and retail marketing, e.g. retail corporate and competi-tive strategies, incl. internationalisation, retail formats, e-commerce, customer behaviour, branding and store image, retail location, assortment, pricing, service, communication, in-store marketing, human resource management;

- different aspects of distribution systems, e.g. strategies, sales management, key account man-agement, vertical integration, channel conflicts, power, and multichannel strategies.

VI Preface

Basically, we seek two types of papers for publication in the book: - Research articles should provide a relevant and significant contribution to theory and practice;

they are theoretically well grounded and methodologically on a high level. Purely theoretical papers are invited as well as studies based on large-scale empirical data or on case-study re-search.

- Manuscripts submitted as more practice-oriented articles show new concepts, questions, is-sues, solutions and contributions out of the retail practice. These papers are selected based on relevance and continuing importance to the future retail research community as well as origi-nality.

In addition, the editors will invite articles from specific authors, which will also be double blind reviewed, but address the retailing situation in a specific country.

Manuscripts are reviewed with the understanding that they are substantially new, have not been previously published in English and in whole, have not been previously accepted for publication, are not under consideration by any other publisher, and will not be submitted elsewhere until a decision is reached regarding their publication in EUROPEAN RETAIL RESEARCH. An excep-tion is given by papers in conference proceedings that we treat as work-in-progress.

Contributions should be submitted in English language in Microsoft Word format by e-mail to the current EUROPEAN RETAIL RESEARCH managing editor or to [email protected]. Questions or comments regarding this publication are very welcome. They may be sent to anyone of the editors or to the above mentioned e-mail-address. Full information for prospective contributors is available at http://www.european-retail-research.org. For ordering an issue please contact the German publisher “Gabler Research” (www.gabler.de) or a bookstore. We are very grateful for editorial assistance provided by Matthias Schu. Graz, St. Gallen, Siegen, Vienna, Trier and Fribourg, Spring 2011

Thomas Foscht, Thomas Rudolph, Hanna Schramm-Klein, Peter Schnedlitz, Bernhard Swoboda Dirk Morschett (managing editor for Volume 25 Issue I)

Contents Why Does Segmentation Matter? Using Mixed Methodology to Identify Market Segments ......................................................................................................................... 1 Jaime R.S. Fonseca RFID-Based Tracking of Shopping Behaviour at the Point of Sale – Possibilities and Limitations ..................................................................................................... 27 Günter Silberer and Stefan Friedemann Prospects for PoS Market Research with RFID Technology: Examination of Consumers’ In-Store Shopping Processes ................................................................................. 47 Thorsten Blecker, Carsten Rasch and Thorsten Teichert In-Store Logistics Processes in Austrian Retail Companies ..................................................... 63 Alexander Trautrims, David B. Grant and Peter Schnedlitz Ethical Sourcing – Choice of Sourcing Strategies and Impact on Performance of the Firm in German Retailing ............................................................................................... 85 Jonas Bastian and Joachim Zentes Country Reports Retailing in India – Background, Challenges, Prospects ........................................................ 107 Doreén Pick and Daniel Müller Retail in Poland New Challenges and New Strategies ........................................................ 141 Tomasz Doma ski

EUROPEANRETAIL

RESEARCHVol. 25, Issue I, 2011, pp. 1-180

Why does Segmentation Matter? Using Mixed Methodology to Identify Market Segments Jaime R.S. Fonseca Abstract The purpose of this chapter is to describe how markets can be segmented. In other words, it studies ways of grouping customers for the most effective targeting by means of a new con-ceptual model which combines the use of latent segment models with a mixed research scheme (merging qualitative and quantitative research methods). A particular retail market segmentation solution depends on both market segmentation base variables and a specific segmentation procedure providing a better understanding of the market. Knowledge of seg-ment structure is extremely important in marketing because of its managerial utility, particu-larly with regard to targeting and positioning. Companies that identify underserved segments can then outperform the competition by developing uniquely appealing products and services. This research begins with an overview of segmentation aspects and aims, and uses a mixed research scheme to present an application with a latent segment model (LSM) procedure for retail market segmentation and information criteria AIC3 and AICu for model selection, in order to uncover the segment structure underlying a dataset from retail chain customers. Keywords Market Segmentation, Base Segmentation Variables, Segmentation Methods, Latent Segment Models, Mixed Research Jaime R.S. Fonseca Chair for Data Analysis, School of Social and Political Sciences (ISCSP), Centre for Public Administration and Policies (CAPP), Technical University of Lisbon, Portugal (E-mail: [email protected]).

Received: September 28, 2010 Revised: February 7, 2011 Accepted: February 16, 2011

EUROPEANRETAIL

RESEARCHVol. 25, Issue I, 2011, pp. 1-25

D. Morschett et al (eds), European Retail Research, DOI 10.1007/978-3-8349-6235-5_1,© Gabler Verlag | Springer Fachmedien Wiesbaden GmbH 2011

2 European Retail Research Vol. 25, Issue I, pp. 1-25

1. Introduction and Objectives

Market segmentation is a theoretical marketing concept involving artificial groupings of con-sumers constructed to help managers design and target their strategies (Wedel/Kamakura 1998). Today, companies recognize that they cannot appeal to all customers in the market or at least not to all customers in the same way, because each customer is unique and they all come from different backgrounds, live in different areas and have different interests and goals. As a result, they are too varied in their needs and buying practices. Furthermore, companies themselves vary widely in their abilities to serve different segments of the market and rather than trying to compete in an entire market, each company must identify the parts of the mar-ket that it can serve best and most profitably (Sun 2009). Companies that identify segments efficiently can then outperform the competition by developing uniquely appealing products and services.

By dividing the market into relatively homogenous subgroups or target markets, both strategy design and tactical decision-making can be more effective and robust for successfully bridg-ing the gap between segmentation principles and successful application, which continues to be a major challenge for the marketing community.

Segmentation technique – identifying homogenous sub-populations within larger heterogene-ous populations – has emerged as an important marketing tool over the past half-century, as a response to the need to effectively communicate with and spur into action an increasingly diverse population of individuals, families and businesses who rely on a rapidly multiplying set of communication channels (Heuvel/Devasagayam 2004). It is well known that customer segmentation is most effective when a company tailors offerings to segments that are the most profitable and serves them with distinct competitive advantages. This prioritisation can help companies develop marketing campaigns and pricing strategies to extract maximum value from both low- and high-profit customers. By tailoring the product to different groups, com-panies are able to meet the needs of more customers more accurately and consequently to gain a higher overall share or profit from a market.

This article develops an overall framework that describes how markets can be segmented. In other words, the focus of this study is the way customers are grouped together for the most effective targeting. It uses a new conceptual scheme that combines latent segment models in mixed research (merging qualitative and quantitative research methods) and is expected to result in market segments that satisfy homogeneity within and heterogeneity across segments. Regardless of the tool used to segment the population, each segment must contain homogene-ous elements. The bases of these similarities should be easily interpretable and should provide useful guidelines for the promotion of products or services specific to each segment.

Fonseca, J. 3

It is planned to delve more deeply into the third part of the market segmentation scheme (Table 3), i.e. the best conceptual scheme for effective market segmentation. It is organised as follows. In section 2, we give an overview of the subject, while in section 3 we present our proposed market segmentation model and corresponding information criteria. In section 4 we report the results from a retail dataset and finally, in section 5, we make some concluding remarks.

2. Why Segmenting?

Consumer diversity is increasing rapidly and companies have long sought to differentiate their products from those of competitors, and this is where market segmentation comes in. Why segmenting? Because identifying segments where competitors see an undifferentiated mass market creates several opportunities for new marketing strategies based on a better knowledge of specific customers’ needs and preferences. It is generally agreed that the foundation of strategic marketing is market segmentation, target marketing and product positioning.

Nowadays, segmentation is a crucial marketing strategy, helping marketers to identify con-sumer needs and preferences and find new marketing opportunities. It also enables marketers to regulate marketing mixes to meet the needs of particular segments.

Several marketing researchers have responded to management needs by conducting market segmentation studies, for instance Assael/Roscoe (1976), Calantone/Sawyer, (1978), Punj/ Stewart (1983), Beane/Ennis (1987), Kamakura/Kim/Lee (1996), Lockshin/Spawton/Mac-intosh (1997), Cohen/Ramaswamy (1998), Dibb (1999), Kim/Srinivasan/Wilcox (1999), Bock/Uncles (2002), Palmer/Miller (2004), Sun (2009). The marketing planning process flows from the selection of target markets to the formulation of a specific marketing mix and positioning, the objective for each retail chain product. Segmentation theory suggests that groups of customers with similar needs and purchasing behaviours are likely to demonstrate a more homogeneous response to marketing programmes and the constitution of segments is essential to target marketing (Fonseca/Cardoso 2007b).

Segments are derived from the heterogeneity of customer wants. Smith (1956) defines market segmentation as a process that involves viewing a heterogeneous market as a number of smaller homogeneous markets, in response to differing preferences, attributable to the desires of consumers for more precise satisfaction of their varying wants. The definition of Kotler (1972) was conceptually consistent with Smith’s, and he defined it as the subdivision of a market into homogeneous subsets of customers, where any subset may conceivably be selected as a market target to be reached with a distinct marketing mix. For Dolnicar (2008) market segmentation is a strategic tool that accounts for heterogeneity among individuals by grouping them into market segments that include members similar to each other and dissimi-lar to members of other segments. According to Sun (2009), market segmentation is dividing


the whole market into meaningful, relatively small and identifiable market segments, which are groups of individuals or organisations with similar product needs.

In other words, market segmentation is the science of dividing an overall market into seg-ments whose members share similar characteristics and needs (member homogeneity).

A market segmentation solution is a function of the market segmentation base variables and of a specific segmentation (clustering) procedure, and it provides a better understanding of the market and, consequently, the means to develop more successful business strategies (Fonseca/ Cardoso 2005) by addressing the specific needs of the selected segments.

Because an organisation adopts either mass-market or market segmentation strategies, two essential questions must be addressed when a market segmentation decision is made: (1) which method is to be used to segment the market and (2) which segmentation base vari-ables to use.

Concerning methods, since the appearance of Smith’s now classic article (1956), market seg-mentation has become an important tool both in academic research and applied marketing (Punj/Stewart 1983), and the primary use of cluster analysis in marketing has been for market segmentation. Cluster analysis is a very weak analytical segmentation technique, but tradi-tionally it is perhaps the one used most for segmentation. We have therefore selected several uses of this tool in marketing (see Table 1) from 1967 to 2007.

Hierarchical cluster algorithms are among the most commonly used for clustering analysis in marketing research. However, users of these approaches tend to discard much of the detail found in the dendrogram (Arabie et al. 1981). Moreover, as is well known, the dendrogram does not constitute a unique solution, which is a disadvantage of hierarchical cluster analysis.

Quantitative segmentation tools can range from simple categorisation analysis, such as CART and CHAID regression tree analyses (McCarty/Hastak 2007; Thomas/Sullivan 2005; Chen 2003; Levin/Zahavi 2001), to more sophisticated clustering techniques, such as hierarchical cluster analysis, two-step cluster analysis, K-means (Lee/Lee/Wicks 2004; Hruschka/Natter 1999; Jedidi/Jagpal/DeSarbo 1997), conjoint analysis (DeSarbo/Ramaswamy/Cohen 1995; Green/Srinivasan 1990; Green/Krieger 1991), multidimensional scaling (Carroll/Green 1997; Biggadike 1981; Wind,/Douglas/Perlmutter 1973), discriminant analysis (Tsai/Chiu 2004; Harvey 1990; Moore 1980), or latent segment models (Cohen/Ramaswamy 1994; Fonseca 2010).

Fonseca, J. 5

Table 1: Use of Cluster Analysis

Authors Goal

Green/Robinson 1967 To identify matched cities for test marketing

Green/Carmone 1968 To identify similar computers in the computer market

Bass/Pessemier/Tigert 1969 To identify market segments with respect to media exposure

Montgomery/Silk 1971 To identify opinion leadership and consumer interest segments

Morrison/Sherman 1972 To determine how some individuals interpret sex appeal in advertising

Greeno/Sommers/Kernan 1973 To identify market segments with respect to personality variables and implicit behaviour patterns

Sexton 1974 To identify homogeneous groups of families with product and brand usage data

Anderson/Cox/Fulcher 1976 To identify the determinant attributes in bank selection decisions and use them to segment commercial bank customers

Calantone/Sawyer 1978 To study the stability of market segments in the retail banking market

Schaninger/Lessig/Panton 1980 To identify segments of consumers on the basis of product usage attributes

Kiel/Layton 1981 To develop consumer taxonomies of search behaviour in Australian new car buyers

Becker et al. 1985 To divide consumer markets by looking at a consumer’s personality

Jain 1993 To analyse markets through social, economic and special segmentation vari-ables such as brand loyalty and consumer attitude

Segal/Giacobbe 1994 To use cluster analysis to uncover four basic “naturals" demographic segments

DeSarbo et al. 1995 K-means cluster analysis for major packaged goods

Kotler 1997 Proposed that consumer markets should be divided according to geographic, demographic, psychographic (lifestyle and personality), and behavioural vari-ables

Dibb 1998 Cluster analysis to identify segments in 270 pregnant women, by using demo-graphic and satisfaction variables

Hruschka/Natter 1999a K-means using demographic and attitude variables

Hofstede/Steenkamp 1999 To develop an integrated methodology based on consumer means-end chains to identify segments in international markets

Baker/Burnham, 2001 To identify market segments based on a cluster analysis of respondents' brand and price preferences

Lin 2002 To consider demographic and psychographic variables

Dibb/Stern/Wensley 2002 A cluster analysis for measuring the impact on organisational performance

Kau/Tang/Ghose 2003 A cluster analysis for seeking patterns, motivations and concerns for online shopping

Lee et al. 2004 To segment the festival market based on motivation factors

Jayawardhena/Wright/Dennis 2007 Cluster analysis and K-means for stability


In this issue of data analysis, the proposed latent segment model approach to clustering, part of our conceptual scheme, offers some advantages when compared to other, more traditional techniques. For example, (1) it identifies market segments (Dillon/Kumar 1994), (2) it pro-vides the means to select the number of segments (McLachlan/Peel 2000), (3) it is able to deal with diverse types of data/different measurement levels (Vermunt/Magidson 2002), (4) it out-performs more traditional approaches (Vriens 2001), and (5) it is appropriate for dealing with covariates for a better understanding of customers (Fonseca/Cardoso 2007a).

Basically it enables to simultaneously optimise a research function (LSM and information criteria) and efficiently find segments of cases within that framework. It is therefore useful for a better understanding of market structures.

In order to be valuable to marketers, a market segmentation plan needs to be able to identify different segments of customers who have uniform, stable responses to a particular set of marketing variables, the segmentation base variables (see Table 2).

This is the second question we have to address, and several authors have conducted research into it, such as Sharma/Lambert (1994); Wedel/Kamakura (1998); Kim et al. (1999); Gonzá-lez-Benito/Greatorex/Muñoz-Galleg (2000); DeSarbo/Degeratu/Wedel/Saxton (2001); Vriens (2001); Heilman/Bowman (2002); Fennell et al (2003).

The greatest opportunity for creating a competitive advantage often comes from new ways of segmenting, because a company can meet buyer needs better than competitors or improve its relative cost position (Porter 1985). The identification of segmentation variables is therefore one of the most creative parts of the segmentation process.

Table 2: Some Segmentation Base Variables

Segmentation base Description

Demographics Consumers can be grouped on the basis of characteristics such as age or household

Socioeconomic Consumers can be grouped on the basis of characteristics such as income, occupation and education

Product usage Potential to use the firm’s product is behaviourally based segmen-tation, with attributes such as awareness, used in the past, would consider using

Psychographics Consumers can be grouped on the basis of personality, attitudes, opinions, and life styles

Generation Generation, or cohort, refers to people born in the same period of time: similar age, similar economic, cultural, and political influences in formative years

Fonseca, J. 7

Generally, a combination of psychographics (for understanding) and demographics (for tar-geting) will give good results. For instance, concerning demographic variables, Sharp/Romaniuk/Cierpicki (1998) and Lin (2002) have suggested that they are useful in seg-menting markets, though most of the evidence does not support this assertion (Fennell et al. 2003; Uncles/Lee 2006). Some studies have shown insignificant or no effect of demographics on consumer price responsiveness, such as Kim et al. (1999) and Scriven/Ehrenberg (2004). Granzin (1981) suggests a simple solution to the problem that links in with Sim-cock/Sudbury/Wright (2006) calling for more sophisticated segmentation: Choosing other variables to work alongside demographics. We argue that demographic variables are very important to a better understanding of segments, and can be used as covariates when estimat-ing latent segment models and not as being part of segmentation base variables.

3. Methodology

The process of identifying segments requires a thorough analysis of the entire market, not only focusing on customer’s needs and shopping habits but also providing knowledge of changing market conditions and competitive actions (Segal/Giacobbe 1994). From traditional market segmentation studies, including mixed research methods, we can enumerate six steps in the market segmentation process. They are summarised in Table 3.

As for step 3, selecting market research tools, we can use data collecting tools - varying from qualitative to quantitative. Market research design and staged design can be sustained by a mixed or pragmatism methodology, which can be defined as research using both qualitative and quantitative methods and by mixing the two methods when beneficial (Onwuegbu-zie/Leech 2005; Leech et al. 2010). In this methodology, both quantitative and qualitative approaches are about taking observations of the world (data) and presenting them within a framework (a model) (White 2002).

In order to design a market research questionnaire, we often start step 3 with qualitative research to define ways in which customers view product or service categories and the differ-ences between these views. We conduct preliminary focus groups or other qualitative methods, such as in-depth interviews, in order to achieve an insight into how consumers and business audiences feel about the product category and competitive brands, for instance, uncovering and refining our learning about customers to obtain a fuller picture and deeper understanding of the segments. Owing to its use of situation- and context-appropriate designs and methods, mixed method research seems particularly suited to action research (Vitale/Armenakis/Feild 2008). In questionnaire design, we can, for instance, use market segmentation dimensions such as behaviour, attitude or a combination of these to form psychographic segments, and another dimension, demographic for instance, as covariates, for a better characterisation of the


segments. In a developmental survey, we also use a questionnaire for collecting data, and then a quantitative method for analysing the obtained dataset.

Table 3: Segmentation Steps

Segmentation step Description

Step 1 Determine the market boundaries

Select a market or product category to study

Step 2 Segmentation base variables

Marketers must use their knowledge of the market to select a few relevant vari-ables in advance

Step 3 Selecting market research tools (mixed research process)

Select tools for collecting and analysing data. From the stages of social research, we notice that qualitative and quantitative can coexist in each researching process. (1) In the first phase we have research preparation, in which we determine the study subject (specification of problem, paper overview, research theory) and the research structure (test structure, measurement, sampling, ethics). (2) This is followed by research (direct observation, indirect interviews, life history, discussion group, content analysis, survey, secondary data, simulation). (3) Finally, an infor-mation analysis (data processing and analysis) is conducted. It would be very difficult to exclude one of the two methodologies in any of these three phases, but social scientists frequently do not manage the available information in statistical results, thus missing chances to present statistics that could result in a bigger clarification of research questions (King/Tomz/Wittenberg 2000).

Step 4 Profiling each market segment

Involves selecting those variables that are most closely related to consumers' actual buying behaviour

Step 5 Segment targeting

A marketer should look for opportunities that provide a good strategy. In step 3, selecting tools for collecting and analysing data, we introduce a mixed methodol-ogy, in order to test the solution, by using all the information obtained from the qualitative data collection tools, such as interviews, focus groups and participant observation, for exploring new topics, assisting theory building, providing context for quantitative data, and helping to explain or clarify quantitative findings (seg-ments). We think that we are finding out more about the needs and preferences of customers by merging knowledge and using qualitative (quantitative) conclusions to update quantitative (qualitative) conclusions. In step 2, one of the most important steps in segmentation schemes, there is a large array of possible segmentation bases - set of variables or attributes used to assign potential customers to homogeneous segments. For a review we can see (Wilkie, 1990) and (Wedel & Kamakura, 1998), for instance. Following the latter authors, “The identification of market segments is highly dependent on the vari-ables and methods used to define them.” This sentence stresses the great impor-tance of segmentation base variables and methods for analysing data. Table 2 summarises some examples of possible segmentation bases.

Step 6 Product positioning

This involves developing a product and marketing plan that will appeal to the selected market segment

In this study we focus more on steps 3 and 4, especially tools for analysing data and profiling our segments. But the market segments identified should mostly satisfy the three criteria that we show in Table 4.

These criteria are all met by using latent segment models, with the aforementioned advantages. It is a probabilistic/statistic clustering approach which assumes that observation of the vari-

Fonseca, J. 9

ables in a sample arises from different segments of unknown proportions. They are very good models for modelling complex phenomena, then synthesizing and extracting knowledge. The proposed conceptual segmentation scheme (1) provides internal homogeneity and external heterogeneity, (2) enables marketers to reach segments separately using observable character-istics of them and (3) because of sparing use of theoretical information criteria for model se-lection balances (fitting a model with a large number of components requires estimating a very large number of parameters and potential loss of accuracy in these estimates (Leroux/Pu-terman 1992) and complexity of models (which tends to improve the model fit to the data), the selected latent class model shows a trade-off between a good description of the data and the model number of parameters.

Table 4: Segment Criteria

Criterion Meaning

Internal Homogeneity/External Heterogeneity Customers within a segment should have similar responses to the marketing mix variable of interest but a different response to members of other segments

Parsimony The degree to which the segmentation makes every customer a unique target. That is, segmentation should identify a small set of groupings of substantial size

Accessibility The degree to which marketers can reach segments separately using their observable characteristics

The segmentation process is used to distinguish between customers and non-customers, where "customers" are extended to include buyers, payers, loyal customers, etc, and to understand their composition and characteristics Who they are? What do they look like? What are their attributes? Where do they live? This analysis supports a whole array of decisions, ranging from targeting decisions to determining efficient and cost effective marketing strategies or even evaluating market competition, (Levin & Zahavi, 2001). The three most relevant criteria for segments (Table 4) are always reached by this conceptual scheme, when segment structure really exists, which is not the case with other tools, such as cluster analysis models.

denotes the vector representing the scores of the ith case for the pth segmentation base variable (i = 1,…,n ; p = 1,…,P). We consider that the cases on which the attributes are measured arise from a population which we assume to be a mixture of S segments, in propor-tions s (mixing proportions or relative segment sizes), s = 1,…,S. The statistical probability density function of the vector, given that comes from segment s, is represented by,

, with representing the vector of unknown parameters associated with the spe-cific chosen probability density function. Then the population density can be represented as a finite mixture of the densities of S distinct segments, i.e.


)|(11

)|( siyP

psf

S

ssiyf

},,1{ , }1,,1{ with }, ,{ ,11

,0 ssS

sss

S

j if

sf

kj

kj

ksi

ks

is

j1

)()(

)()(

ˆ

)ˆ|y(ˆ

)ˆ|y(ˆ

isSs

Max ˆ,...,1

[1]

where i = 1,…,n,.

is the vector of all unknown parameters.

The LSM estimation problem simultaneously addresses the estimation of distributional pa-rameters and classification of cases into segments, yielding mixed probabilities. The estima-tion process is typically directed to maximum likelihood using the expectation-maximisation (EM) algorithm (Dempster/Laird/Rubin 1977; McLachlan/Peel 2000).

LSM naturally provides means for constituting a partition by means of assigning each case to the segment with the highest segment-membership probability, that is with where

[2]

In order to derive meaningful results from clustering, the mixture model must be identifiable, i.e. a single maximum likelihood solution should exist (Bozdogan 1994). One goal of tradi-tional LSM estimation is to determine the smallest number of latent segments S sufficient to explain the relationships between the segmentation base variables. If the baseline model (S = 1) provides a good fit to the data, no LSM is needed since there is no relationship be-tween the variables to be explained. Otherwise, a model with S = 2 segments is then fitted to the data. This process continues by fitting successive LSM to the data, adding another dimen-sion each time by incrementing the number of segments by 1, until a parsimonious model is found that provides an adequate fit. They are very good models for modelling complex phe-nomena and then synthesizing and extracting knowledge.

Concerning methods for the selection of the appropriate latent class model, we propose to use traditional information criteria Especially, because all the observed variables have similar measure, all of them categorical, we will use the AIC3 information criterion the best one for this situation (Fonseca 2010).

We can now answer the questions on page 5, concerning segmentation tool and segmentation base variables. Thus, as the best segmentation tool we propose latent segment models and for segmentation base variables we consider, with marketers, some store attributes and some cus-tomer attributes that interact between them.

Fonseca, J. 11

These variables are considered as manifest variables or indicators (USAGE FREQUENCY, …, INTERNET

USE), from which model parameters are estimated and some covariates used (SEX, ... , CLASS), which are only employed for a better understanding of segments and their members (see Table 5). Results from the estimation of these models are valid for all cases, products, branches, countries, services and all kinds of variables (categorical, continuous, or mixed), because they are probabilistic/statistic models.

Table 5: Variables and Covariates of the Dataset Variables used for a retail chain customers’ segmentation Segmentation base Usage frequency

Psychographic

Visit pattern Coming from Travel time Why shopping Monthly spending on purchases for the home Monthly spending in store Quality of fresh produce Store treatment Efficiency of Store’s staff variety of products Product presentation Store environment Cleanliness Shop Prices Product quality Private Label Internet use Covariates Sex

Demographic Age Family size Life cycle Income

Socioeconomic Education Occupation Class

4. Results from a Retailing Dataset and Discussion

We used two types of data collection in the research: Qualitative and quantitative. Here, we only report the quantitative data analysis, based on a dataset obtained from a questionnaire given to a retail chain's customers.

Table 6: AIC3 for Model Selection

Model LL AIC3

1-Cluster -45613.3 91484.666 2-Cluster -44022.5 88681.023 3-Cluster -43082.7 87179.382


After eliminating the questionnaires with non responses, we had a dataset with 1,449 custom-ers characterised by the segmentation base variables shown in Table 5.

Table 7: Parameter Estimates of Two-Class Latent Model

Cluster size Cluster 1 (61 %) Cluster 2 (39 %) Indicators Usage frequency Every day 0.2280 0.4265Two or three times a week 0.3754 0.2670Once a week 0.1842 0.1523Twice a month 0.0410 0.0278Once a month 0.0700 0.0586Occasionally 0.1013 0.0679Visit pattern During the week 0.3511 0.1898At the weekend 0.1838 0.1246Both 0.4652 0.6857Coming from Home 0.6468 0.8012Work 0.2643 0.1133Passing by 0.0598 0.0622Other 0.0291 0.0234Travel time Two minute walk 0.1587 0.2085Two to five minute walk 0.1963 0.2524Five to ten minute walk 0.1584 0.1521More than ten minutes' walk 0.0673 0.0770Five minutes or less by car 0.1467 0.1582Five to ten minutes by car 0.0858 0.0674Ten to fifteen minutes by car 0.0982 0.0425More than fifteen minutes by car 0.0886 0.0417Why shopping Near home 0.6015 0.6912Near work 0.0851 0.0436Passing by 0.1291 0.0419Low prices 0.0430 0.0424Variety of brands 0.0066 0.0216Variety of products in general 0.0196 0.0278Habit 0.0346 0.0432Quality products 0.0250 0.0353Quality of fresh produce 0.0060 0.0047Cleanliness / hygiene shop 0.0068 0.0070Fast service 0.0119 0.0061Friendly service 0.0072 0.0243Promotions 0.0034 0Opening hours 0.0077 0.0023Other 0.0011 0.0018Monthly spending on purchases for the home Mean 266.1543 380.0798 Monthly spending in store

Mean 89.3863 200.9802

By estimating these LSM from the baseline model (homogeneity model or non-structure seg-ments) to a three-class latent model, we selected a two-class latent model by using AIC3 and AICu (Fonseca 2010a) for model selection, because we had a mixed-mode dataset (Monthly spending on purchases for the home and monthly spending in store are continuous, the others categorical). These models automatically select the number of segments, 2-segment in this case, because the graph for AIC3 shows an elbow (see Table 6), by using an information crite-rion, which is an advantage when compared with cluster analysis.

dirk morschett, thomas foscht, thomas rudolph, peter

Documents