Spatial patterns in
consumer preferences in
the Netherlands
1 Master thesis MSc Marketing, Marketing Intelligence track
“Spatial patterns in consumer preferences in the Netherlands” Jonas Braadbaart, graduate student
Department of Marketing, Faculty of Economics and Business University of Groningen
Nettelbosje 2
9747AE, Groningen The Netherlands
Completed on the 20th of August of 2014 Academic supervisors
Niels Holtrop, MSc Hans Risselada, PhD Student number author: s2448092
Contact information author + 31 6 27584328 braadbaart@hotmail.ch Hamseweg 52a
2
Executive summary
3
Preface
Having lived in six countries, I have always had an active interest in differences in and between consumer populations. When pursuing courses in the MSc Marketing I was therefore very much frustrated to find that demographic data was often rather useless when segmenting or profiling firm customers. This kindled my interest in marketing applications of location profiles as an alternative to demographic profiles. It was thought that by identifying spatial patterns in consumer preferences, the efficiency, reach, and targetability of consumer segments derived from usage patterns, social interaction, and other types of data could be improved.
4
Contents
Executive summary ... 2 Preface ... 3 Introduction ... 6 Conceptual framework ... 8The measurement of consumer preferences ... 9
Population homogeneity ... 11 Data ... 13 Methodology ... 15 Interdependence ... 15 Service quality ... 17 Results ... 18
Stability and validity of the findings ... 22
General discussion ... 24
Applicability to marketing practice ... 25
Limitations ... 26
Conclusion ... 28
References ... 29
Appendix A: Dimensions extracted from the Healthwise survey ... 37
5
“People, not markets, respond to the actions of the retailers and manufacturers.”
6
Introduction
This thesis investigates spatial patterns in consumer evaluations of service quality. The decision to focus on consumer services rather than consumer products was made as a result of the growing economic importance of services. From 1975 to 2010, the services share of world GDP has increased from 55% to 70% (World Bank WDI). This has led to a number of developments in marketing practice, including – but not limited to – the rise of the customer-centric firm, a services orientation adopted by a number of traditional manufacturing firms, and a changing technology and manufacturing landscape (Libai, Muller, & Peres 2009; Jiao, Simpson, & Siddique 2007; McKinsey 2014; Ulaga & Reinartz 2011). Of all these changes, the rise of the customer-centric firm in particular stands out. The availability of more and more customer data in firm databases has led to an emphasis in marketing research on modeling social interactions and customer relationships (Achrol & Kotler 1999; Rust & Chung 2006). However, this usage of firm data has also left services firms in the blind when trying to identify market opportunities among consumers that are not yet customers of the firm. Even today, situations therefore arise in which a retailer bases its market entry decision on “industry wisdom”, and follows a competitor into a new market instead of using available marketing information (Gielens & Dekimpe 2007).
The problem is that marketing information with regards to services is very hard to quantify and apply across firm- and industry boundaries. This corresponds with the common view in marketing that services are “heterogeneous, intangible, and perishable, [with] production inseparable from consumption” (Hollensen 2013: 461-462). It should therefore come as no surprise that in service delivery, price and advertising have been shown to interact dynamically (Polo, Sese, & Verhoef 2011). This means that if marketing opportunities in service delivery are to be identified, a departure is needed from frameworks used to quantify consumer demand for products in marketing (Bass 1969; Ataman, Van Heerde, & Mela 2010). In this thesis, such a framework is developed and estimated using location data. By investigating regional differences in consumer service quality evaluations, two interrelated research questions will be answered:
1. Can spatial patterns in consumer preferences be identified?
2. Could these spatial patterns be used to identify marketing opportunities?
7 Figure 1: Estimated number of pharmaceutical prescriptions per 1000 inhabitants in 2012 (Source: NPCF-VAAM 2014).
ability to service or target a local consumer population. Furthermore, by establishing cross-regional segments in a national consumer population, hitherto untapped economies of distribution in service delivery could be identified. This would in turn allow for a more effective use of the location data widely available in firm databases.
Empirical evidence will be provided using a nationwide survey of customers of Dutch pharmacies in 2012 and 2013. Because healthcare insurers reimbursed 96% of the €4,398,000,000 in pharmaceuticals sold through 1,981 Dutch pharmacies in 2012 (SFK 2013), service quality is expected to play a very large role in consumer evaluations. Nevertheless, as can be seen in figure 1, pharmacies also operate under different local market demand. The figure shows the annual number of pharmaceutical prescriptions per 1,000 inhabitants as estimated by NPCF-VAAM for the 40 Dutch NUTS3 regions. The NUTS acronym stands for the EU “Nomenclature des Unités Térritoriales Statistiques”.
NUTS3 regions are the primary unit of spatial analysis used in this thesis, and usually have between 150,000 and 800,000 inhabitants. In the Netherlands in 2012, total consumer disposable income in a NUTS3 region varied between roughly €2 billion and €28 billion (CBS StatLine 2014). There were 1,342 NUTS3 regions in the European Union as of 2014 (Eurostat 2014).
8
Conceptual framework
At its core, this thesis investigates the benchmarking of consumer preferences in a country using location data. This can be formulated such that for customer of store in region in country in year , certain measurements will be available that need to be developed into meaningful and applicable marketing information. This could be done by looking at sales volumes, or estimating the utility of customer visiting store relative to store . Such a linear or logistic approach however does not take into account regional differences in consumer populations. These regional differences are illustrated in figures 2a and 2b below using the statistics of a survey item also used in the empirical analysis. This item asks respondents to award an overall grade to their pharmacy on a ten point scale ranging from 1 (very bad) to 10 (excellent). The item, Grade For Service (GFS), is here aggregated at the NUTS3 level. As can be seen regional statistics are often quite different from national statistics . A manager will therefore experience difficulties when evaluating the service quality of store in region in year , and will instead need to rely on hard metrics such as sales data or profit margins for his or her evaluation of the firm channel. This also leaves little room for action other than hiring or firing employees or managers. The issue is further complicated by changes over time in consumer preferences and consumer markets (Guadagni & Little 1983; Russel & Bolton 1988; Rutz & Sonnier 2011).
9 Nevertheless a cluster of five NUTS3 regions can be identified with relatively high GFS values ( ) and low variance ( ), indicated with orange circles in figures 2a and 2b. This raises the question of whether the pharmacies in these five Eastern regions indeed do offer better service than their counterparts in other regions, or whether inherent differences exist in the way consumers evaluate the service offered by their local pharmacy, such as consistently high grading or more uniform responses in a population (Fox 2010). If these spatial patterns in consumer preferences are indeed due to some kind of inherent traits or characteristics in a consumer population, then at least in theory, regional clusters could be established to improve geo-targeting, servicing, and store benchmarking. In order to derive marketing information from location data reliable enough to allow a manager to establish service regions, the relationship between geographical location and consumer preferences needs to be understood.
The measurement of consumer preferences
There are two ways to model this relationship – through market share, or through customer satisfaction measurements. A number of recent marketing publications have for instance shown spatial autocorrelations between brand market shares in regional product-markets (Ataman, Mela, & Van Heerde 2007; Bronnenberg & Sismeiro 2002; Bronnenberg, Dhar, & Dubé 2007, 2011). Brand market share is thus used as an instrument to measure consumer preferences (Berry, Levinsohn, & Pakes 1995; Guadagni & Little 1983; Dubé, Fox, & Su 2012). While this approach can be used to model FMCG and CPG markets, for services markets it poses a number of problems. First of all, a large number of services firms enter into long-term relationships with their customers, making the measurement of market share dynamics dependent on years or even decades of data (Neslin, Grewal, Shankar, Teerling, Thomas, & Verhoef 2006). Second, services markets are often considered ‘glocal’ as a result of network, substitution and tipping effects (Chen & Steckel 2012; Dubé, Hitsch, & Chintagunta 2010; Libai et al. 2009; Wang & Qie 2011). And third, because services firms operate as two-sided markets when they actively seek to connect supply and demand through service delivery, the analysis of either the demand or the supply side is less informative than it would be in the case of marketing a product (Athey & Ellison 2011; Rochet & Tirole 2006; Sridhar, Mantrala, Naik, & Thorson 2011).
10 populations. This is done in a modelling framework known as the ‘interdependence of consumer preferences’ (Granovetter & Soong 1986; Yang & Allenby 2003). The basic premise of this modeling framework is that consumer preferences are partially defined by an individual consumer being part of a local consumer population (Granovetter 1978). In the past, several causal mechanisms have been proposed for the interdependence of consumer preferences, from word-of-mouth to the endogeneity of supply and demand (Lantané 1996; Yang, Chen, & Allenby 2003). Because empirical evidence for any of these mechanisms invariably links to systems theory, sociology, and macroeconomics, these mechanisms will not be treated here (Cavalli-Sforza & Feldman 1981; Grover & Srinivasan 1987; Helbing 2010; Helpman & Krugman 1989; Meyers 2012; Pesaran, Schuermann, & Weiner 2004; Porter 1990).
Nevertheless, as is illustrated in figures 3a and 3b below for car dealerships in the San Diego (CA) area, the customer base of a services firm is often geographically determined. This means that the relative utility for a large number of consumer services will include a spatial component (Farag, Krizek, & Dijst 2006). As can be seen from the pattern of spatial concentration in the dealership’s customers in figure 3b below, it is very probable that this spatial component in the utility of services will at least in part run in parallel with the density-dependent process of agglomeration (Carroll & Hannan 2000). Agglomeration – the spatial concentration of consumer populations – is known to affect consumption both quantitatively and qualitatively (Fujita, Krugman, & Venables 1999; Porter 1990).
11
consumer
population
consumer preferences region services firmFigure 4: Venn diagram of the relationship between services firm, population, and region. The question thus arises if a marketer can use location data to improve consumer profiles, consumer segments, or customer acquisition (Ataman et al. 2007; Bronnenberg et al. 2007; Steenkamp & De Jong 2010; Ter Hofstede, Wedel, & Steenkamp 2002). By answering this question, marketing research could also move beyond the default application of customer satisfaction measurements as a means to increase customer loyalty (Bolton & Lemon 1999). While increasing customer loyalty has been shown to have a positive effect on share-of-wallet in some services industries but not in others, the role of customer satisfaction in customer acquisition is far from understood (Cooil, Keiningham, Aksoy, & Hsu 2007; Mägi 2003; Mittal & Kamakura 2001; Rust & Chung 2006).
Population homogeneity
The relationship between consumer populations and consumer preferences could provide part of the answer. The spatial dimension of this relationship is modeled schematically in the Venn diagram in figure 4. It incorporates the spatial homogeneity (overlap) and heterogeneity (no overlap) of a consumer population in country (Allenby & Rossi 1999;
12 Figure 5: Population density per km2 in 2012
(Source: CBS StatLine 2014).
Nevertheless the preferred way to analyse consumer populations is at the highest level of aggregation available in the data – firm or national. The common approach in LSEs is to identify meta-level tiers in consumer populations, and segment on the basis of demographic data or lifestyle and consumption patterns (Ter Hofstede et al. 2002). This will undoubtedly help tailor the firm’s offering to existing customers, but it will not allow for geo-targeting, nor will it allow for the identification of regional marketing opportunities in customer acquisition. Because recent advancements in information technology and mobile communications have opened up a whole new range of possibilities, what is currently missing in the service marketers’ toolbox is a method to identify and estimate local consumer demand for services using location data (Melton & Hartline 2010; Miceli, Raimondo, & Farace 2013; Rust & Chung 2006).
13 preferences are known to be highly dynamic and volatile as a result of consumer utility maximization and social interaction processes (Banks, Blundell, & Lewbel 1997; Blundell, Pashardes, & Weber 1993; Geroski 2003; Hartmann 2010; Helbing 2010; Risselada, Verhoef, & Bijmolt 2010). It could therefore well be that the increased costs and complexity of including dynamic interdependencies in a model or decision system outweigh the economic value of the marketing information derived from location data (Ter Hofstede et al. 2002).
Data
The empirical data are taken from an ongoing nationwide customer satisfaction survey of Dutch pharmacies conducted by the Healthwise Center of Expertise for Healthcare Management and Economics of the University of Groningen in the Netherlands. Survey questionnaires collected over 2012 and 2013 were used. The Healthwise survey in those two years consisted of 55 ratio- and nominal-scaled multiple response questions supplemented by a number of open questions. Survey participation was on a voluntary basis. Respondents were coded on year, pharmacy, and zip code, the latter enabling the derivation of spatial cross-sections from the data using the standard EU NUTS hierarchical geocoding, so that the survey population could be aggregated on NUTS3, NUTS1, and national levels (Goodman 1998; Leeflang, Wittink, Wedel, & Naert 2000).
Survey items with too many non-responses were removed. Because the present investigation is concerned with the benchmarking of service performance evaluations, survey items inquiring about a respondent’s personal health were also removed. To enhance the sample representativeness of the sample population for a regional consumer population, NUTS3 regions with less than 150 respondents were excluded from the analysis. Pharmacy outliers in terms of negative service performance were removed as a further step to enhance sample representativeness. The data was then scanned to ensure that no pharmacy accounted for more than 33% of the respondents in a given region. After these measures, the Healthwise survey data consisted of 25 survey items answered by 14.206 customers of 291 Dutch pharmacies in 24 NUTS3 regions in 2012, and by 10.572 customers of 166 Dutch pharmacies in 15 NUTS3 regions in 2013, with and respondents per pharmacy per NUTS3 region.
14 On these 25 survey items a principal component factor analysis was performed in SPSS. Six dimensions with eigenvalues of were extracted from 24.778 questionnaires (KMO = 0.901, Bartlett’s test of sphericity with 300 df., explaining 65.03% of inter-item variance). These six dimensions and the corresponding survey items are given in Appendix A. However, two dimensions extracted from the survey consisted of items that would allow a respondent to select N/A. Because over two-thirds of the survey respondents had opted to select N/A in at least one of the four survey items scored on these two dimensions, it was decided to drop these two dimensions from subsequent analyses. The statistics of the four remaining dimensions are given in table 1. Together, these four dimensions explain 55.98% of the variance in the 25 Healthwise survey items on service performance. As expected, service is by far the most important component, and has an eigenvalue of 7.117 explaining 28.47% of all inter-item variance (blue box in table 1). It can also be seen that the service dimension is very uniform ( ). The prescriptions and communication dimensions show considerably larger variance – almost 4 and 7 times larger than the service dimension, respectively (red box in table 1).
Extraction Service Prescriptions Communication Facilities
Eigenvalue 7.117 3.455 1.920 1.501 % of variance 28.47% 13.82% 7.68% 6.01% Statistics Mean 3.789 1.486 3.099 3.047 SE 0.003 0.005 0.007 0.004 Variance 0.167 0.633 1.155 0.351
Table 1: Dimensions and national aggregate statistics of the extracted factors.
15
Methodology
As has been discussed above, the interdependence of consumer preferences framework relates consumer preferences to both geographical location and consumer populations. Even though a heterogeneous consumer population calls for a fully individualized classification of consumers, a services firm also needs to be able to identify sizeable segments in a target population to allow for economies of distribution in service delivery. This means that a middle ground needs to be found between model fit and the applicability of the statistical estimates (Leeflang et al. 2000; Liang, Paulo, Molina, Clyde, & Berger 2008). Further complicating the modelling effort is the limited data that invariably hampers any statistical analysis (Chen & Steckel 2012; Pischke 1995; Schweidel, Fader, & Bradlow 2008). As a consequence, the methodology used to identify spatial patterns in consumer preferences (1) needs to makes use of reliable statistical techniques, (2) take into account that only limited data will be available for a given consumer population, and (3) deal with the fact that consumer populations are heterogeneous by nature.
Interdependence
Here this is done by assuming that for a location in region in country with utility for customer , the regional subpopulations comprising of customers of locations are independent draws from a country-specific population . As such, are
thought to be more similar than dissimilar in terms of their probability distribution and probability density than regional populations drawn from another country . While this assumption will not always hold, it is nevertheless useful to make, both from data collection and implementation perspectives. This assumption of equal variances is referred to as the associativity of means in technical literature, and interdependence or endogeneity in economic research (Franses 2005; Grabisch, Marichal, Mesiar, & Pap 2009; Yang & Allenby 2003). In the present analysis, it allows for the comparison of regional subpopulations relative to each other and to the national aggregate, by assuming that all populations are drawn from the same (empirical) distribution. As such, the estimates of non-nested models can be compared to one another because the population samples are assumed to be drawn from the same distribution.
16 The objective of the cluster analyses is to identify the relative homogeneity of service quality evaluations in a regional survey population. Survey respondents will be clustered on two dimensions extracted from the survey: service and communication (see Appendix A and table 1). These two dimensions offer a good representation of individual consumer attitudes towards service quality, and are composed of a total of nine multiple-choice survey items. It could be seen in table 1 that they are also the two survey dimensions with respectively smallest and largest degree of variance and uniformity across the survey population.
The EM algorithm used to cluster survey respondents consists of an iterative many-to-one mapping of observations onto a parameter space of unobserved variables , where is the complete data specification only known through . In the present analysis, is the regional population, and the observed part of this regional population measured by the survey data. Because the functional form of the relationship between and cannot be known at the start of an empirical investigation, the EM algorithm assumes this relationship consists of a likelihood function (Dempster et al. 1977). To be able to say anything about , the EM algorithm needs to make use of a sufficiency statistic which allows the algorithm to infer the statistical properties of the complete data specification using estimated sampling densities (Barndorff-Nielsen 1978; Cox 2006; Dempster et al. 1977). These density-driven inferences are necessary because with limited data available, even a survey population of 14,206 respondents will correspond to a sample of less than 0.1% of the actual regional population. While any likelihood function could be used depending on topologies found in the data, it was here decided to use a Gaussian normal mixture of probabilities (Carlsson 2009; Hartmann 2010). As such, the likelihood of a statistical manifold with dimensions and clusters can be specified as
( ) ∏ ∑
for components, their relative probabilities , and independent observations of (Dempster et al. 1977; Fraley & Raftery 2002). This likelihood is evaluated iteratively through a series of eigenvector decompositions with Cholesky factorization (Fraley & Raftery 2002: 612). For data points of unobserved data estimated from the sampling densities of , the mixture probability ̂ ̅̅̅̅̅̅̅̅̅̅̅̅ of a cluster can be inferred from the joint density of the complete data specification through ∏ ̂
(Fraley & Raftery 2002:
17 Having estimated the cluster solution for the sample population in a region , the mixture probability ̂ of the largest cluster in the cluster solution is used to index the relative homogeneity of service quality evaluations in a regional population . The ̂ measure of homogeneity of response is thus based on the variance and probability density of a two-dimensional statistical manifold, and is therefore closely related to the notion of uniformity in preference measurement (Fox 2010). While these statistical estimates are naturally subject to sampling and measurement bias, they are nevertheless expected to allow for the identification of spatial patterns in the homogeneity of consumer preferences with regards to service quality.
Service quality
However, estimates of the relative degree of homogeneity of consumer preferences alone will provide no insight into the role service quality plays the overall evaluation of a pharmacy by a customer in region . To determine whether the estimated relative homogeneity of consumer preferences in a region could be linked to regional differences in the impact of service quality on customer satisfaction, linear regressions were estimated at three levels of aggregation – NUTS3, NUTS1, and national aggregate. The dependent variable in these linear regressions is the Grade For Service (GFS) statistic described on page 8. As independent variables in the linear regression the four service performance dimensions extracted from the survey will be used – service, communication, prescriptions and facilities. By using GFS as a dependent variable, consumer evaluations of service quality can be embedded in the overall evaluation of the pharmacy without concern for model nesting or data compatibility. With Dutch healthcare insurers reimbursing 96% of all consumer purchases in the pharmacy channel in 2012, it is expected that pharmacy evaluations are based primarily on the quality of the service offered by the pharmacy staff. This assumption was confirmed both by the results of the factor analysis and by prior empirical research identifying service quality as the single most important dimension in store evaluations in the Netherlands (Ter Hofstede et al. 2002: 171-174).
18 dimension are multiplied with the measures of model fit. This yields an indication of the magnitude of the linear relationship of service quality to overall customer satisfaction in a region or country ,
̂
It is not known beforehand whether the impact of service quality on customer satisfaction will be different across regions of country .
Results
To investigate spatial patterns in the homogeneity of consumer preferences, a bivariate cluster analysis was performed using the service and communication dimensions. Cluster solutions were found using the EM algorithm to iteratively estimate the Maximum Likelihood of Gaussian probability mixtures with up to clusters. The algorithm was implemented in the mclust R package version 4.3 (Fraley, Raftery, Murphy, & Scrucca 2012). The optimal numbers of clusters for a survey population in region or country were determined through unsupervised learning (Baudry, Raftery, Celleux, Lo, & Gottardo 2010). Presented in table 2 on the next page are the log likelihood, number of clusters, and mixture probabilities of the largest service/communication cluster for the optimal cluster solution. It can be seen that as more data is available, the log-likelihood value increases exponentially. This is a result of the uniform distribution of the input data, which was already visible in the relatively small variance in the service dimension in table 1 (Clauset, Shalizi, & Newman 2009; Cox 2006). A further result can be seen in the upward bias of the estimates, such that at the NUTS3 level the largest cluster was almost invariably the cluster in which respondents had awarded the pharmacy the highest possible rating on the four-scale service dimension ( ), while the average value of the
communication dimension in varies from region to region.
19 seen from the ̂ values in table 2, mixture probabilities of vary considerably across NUTS3 regions, indicating large regional differences in the homogeneity of consumer preferences.
Table 2: Bivariate cluster analyses
Region ̂ ̂ ̂ 1 Oost-Groningen 201 238.6 33 78 4.00 3.92 39.11% - - 3 Overig Groningen 505 464.2 28 207 4.00 3.68 34.43% - - 4 Noord-Friesland 243 -73.4 20 82 4.00 3.92 34.96% - - 7 Noord-Drenthe 291 106.4 25 113 4.00 3.82 30.69% 36.67% +19.49% 8 Zuidoost-Drenthe 219 70.8 15 111 4.00 3.92 51.63% 48.93% -5.23% 10 N-Overijssel 598 961.0 27 291 4.00 3.93 48.93% 49.65% +1.47% 11 ZW-Overijssel 477 841.7 19 291 4.00 3.89 56.20% - - 12 Twente 533 870.4 25 277 4.00 3.90 53.59% - - 13 Veluwe 1006 998.0 47 381 4.00 4.00 37.82% 53.08% +40.35% 15 Arnhem/Nijmegen 436 40.3 20 181 3.97 4.00 41.50% 51.00% +22.89% 17 Utrecht 1970 3706.1 47 810 4.00 3.91 41.13% 32.23% -21.64% 19 Alkmaar e.o. 376 213.0 21 175 4.00 3.87 38.32% 48.22% +25.84% 21 Aggl. Haarlem 474 335.0 26 201 4.00 3.91 42.90% 46.68% +8.81% 23 Groot-Amsterdam 1659 2035.8 50 641 4.00 3.90 38.50% 53.03% +37.74% 24 Gooi en Vechtstr. 283 257.3 19 136 4.00 3.92 49.23% - -
25 Leiden & Bollenstr. 1256 2004.8 47 509 4.00 3.92 40.70% 36.18% -11.11%
26 Aggl.'s-Gravenhage 386 430.1 23 189 4.00 3.88 47.71% 50.71% +6.29% 28 Oost-Zuid-Holland 398 636.5 20 219 4.00 3.82 47.96% - - 29 Groot-Rijnmond 712 -592.5 17 367 4.00 3.69 40.10% 30.41% -24.16% 33 West-N-Brabant 619 16.2 44 224 4.00 3.93 37.02% - - 34 Midden-N-Brabant 266 137.6 22 102 4.00 3.87 38.28% - - 35 NO-N-Brabant 527 431.2 34 229 4.00 3.90 39.72% - - 36 ZO-N-Brabant 379 57.3 28 155 4.00 3.94 41.47% 50.58% +21.97% 40 Flevoland 392 262.5 24 177 4.00 3.87 45.63% 43.83% -3.94% NL1 North. Netherlands 1459 2549 43 589 4.00 3.90 36.01% 41.04% +13.97% NL2 East. Netherlands 3442 7343 91 1321 4.00 4.00 38.38% 33.53% -12.64% NL3 West. Netherlands 7514 12337 51 3386 4.00 3.90 41.96% 31.14% -25.79% NL4 South. Netherlands 1791 3150 43 881 4.00 3.78 47.11% 50.58% +7.37%
Nat. National aggregate 14206 378931 34 5543 3.97 4.00 38.59% 31.96% -17.18% Legend: Third block of columns from the left: Log likelihood of the optimal cluster solution with clusters. Fourth block of columns from the left: number of respondents in the largest cluster , and in-cluster averages of service and communication dimension ( and ). Fifth block of columns from the left: mixture probabilities of for 2012 and 2013 data ( ̂ and ̂ ), along with the absolute change in mixture probability of from 2012 to 2013 ̂ ; a dash (-) indicates no data was available for 2013.
20 statistically significant parameter estimates are given in table 3 below. Parameter estimates that were not statistically significant were omitted from the table. As can be seen from table 3, regional differences in the extent of the linear relationship between consumer evaluations on these four dimensions and customer satisfaction confirm what had already been found for mixture probabilities ̂ in table 2.
Table 3: Linear regressions
Region ( ) 1 Oost-Groningen 201 0.560 0.615 **0.095 0.198 34.44% - - 3 Overig Groningen 505 0.527 0.526 0.137 0.227 27.72% - - 4 Noord-Friesland 243 0.520 0.588 **0.088 0.212 30.58% - - 7 Noord-Drenthe 291 0.659 0.621 0.103 0.232 40.92% 45.31% +10.73% 8 Zuidoost-Drenthe 219 0.415 0.432 0.164 0.156 *0.158 17.93% 20.41% +13.83% 10 N-Overijssel 598 0.374 0.415 0.152 0.103 0.167 15.52% 21.45% +38.21% 11 ZW-Overijssel 477 0.295 0.352 0.168 0.248 10.38% - - 12 Twente 533 0.485 0.451 0.153 0.122 0.216 21.87% - - 13 Veluwe 1006 0.470 0.484 0.079 0.141 0.223 22.75% 24.75% +8.79% 15 Arnhem/Nijmegen 436 0.574 0.528 0.139 0.245 30.31% 34.78% +14.75% 17 Utrecht 1970 0.547 0.555 0.112 0.066 0.234 30.36% 29.59% -2.54% 19 Alkmaar e.o. 376 0.499 0.497 *0.084 0.125 0.221 24.80% 17.15% -30.85% 21 Aggl. Haarlem 474 0.463 0.534 *0.079 0.129 0.123 24.72% 16.74% -32.28% 23 Groot-Amsterdam 1659 0.572 0.501 0.090 0.151 0.230 28.66% 35.40% +23.52% 24 Gooi en Vechtstr. 283 0.472 0.463 **0.090 0.220 0.149 21.85% - -
25 Leiden & Bollenstr. 1256 0.459 0.464 0.120 0.088 0.253 21.30% 20.71% -2.77%
26 Aggl.'s-Gravenhage 386 0.540 0.513 **0.074 **0.066 0.290 27.70% 11.85% -57.22% 28 Oost-Zuid-Holland 398 0.498 0.491 0.145 *0.088 0.239 24.45% - - 29 Groot-Rijnmond 712 0.539 0.467 0.143 0.164 0.192 25.17% 35.25% +40.05% 33 West-N-Brabant 619 0.573 0.598 0.078 0.216 34.27% - - 34 Midden-N-Brabant 266 0.489 0.551 0.183 0.183 26.94% - - 35 NO-N-Brabant 527 0.627 0.567 0.152 0.099 0.210 35.55% - - 36 ZO-N-Brabant 379 0.531 0.532 *0.088 0.255 28.25% 19.55% -30.80% 40 Flevoland 392 0.543 0.603 0.100 0.194 32.74% 32.95% +0.64% NL1 North. Netherlands 1459 0.558 0.561 0.126 *0.044 0.216 31.30% 37,39% +19.46% NL2 East. Netherlands 3442 0.482 0.502 0.119 0.078 0.216 24.20% 28.41% +17.40% NL3 West. Netherlands 7514 0.528 0.513 0.105 0.110 0.224 27.09% 30.41% +12.26% NL4 South. Netherlands 1791 0.569 0.566 0.112 0.069 0.219 32.21% 19.55% -39.30%
Nat. National aggregate 14206 0.532 0.526 0.112 0.089 0.222 27.98% 30.14% +7.72% *) statistically significant at the level; **) significant at the level.
21 Figure 6: Mixture probabilities of in 2012 (Source:
Healthwise; see table 2).
Figure 7: Relationship of service quality to customer satisfaction in 2012 (Source: Healthwise; see table 3). The results given in table 3 also show that the relationship of service quality to customer satisfaction can be highly non-linear. For region 11, “ZW-Overijssel”, a standardized beta of service and poor model fit with an score of show that only of the variation in consumer pharmacy evaluations can be linearly related to service quality. This means that in region 11, the impact of service quality on customer satisfaction is minimal, so that for instance long-term relationship management might benefit the pharmacy. The results also show that in region 7, “Noord-Drenthe”, on average 40.38% of the variation in GFS scores is linearly related to service quality evaluations. This means that in NUTS3 regions only 100 kilometers apart, consumer perceptions of service quality will on average be a four times more important for their overall evaluation of the pharmacy. This is an important finding, because it shows that service quality evaluations have a spatial component that has hitherto been ignored in marketing research. It could also provide a partial explanation as to why large-scale national aggregate level marketing studies on customer satisfaction such as Mittal & Kamakura (2001) often find high levels of response bias in customer satisfaction ratings.
22 As could be seen from figure 6 on the previous page, the dominant pattern in ̂ is that of regional cluttering. Indicated by black circles are some of the ̂ values that fall within percentage points of each other in neighbouring NUTS3 regions, evidencing cross-regional patterns in consumer populations in support of the interdependence of consumer preferences hypothesis. On the contrary, in figure 7 the dominant pattern is random dispersion. This can for instance be seen in a linear relationship score that is 4 times larger in the “Noord-Drenthe” region (white circle) than in the “ZW-Overijssel” region located only 100 kilometres South (black circle). Nevertheless, consist with the spatial patterns in the GFS statistic given on page 8, orange circles in figures 6 and 7 highlight regions with high ̂ values and low values, demonstrating stronger-than-average homogeneity in service quality evaluations but a lower-than-average linear relation of service quality to customer satisfaction. This means that other factors external to the survey will have a bearing on customer evaluations of a pharmacy. However, as could already be seen in tables 2 and 3, the two indices are also not stable over time, with ̂ and values that could change as much as 57% over the 24 month period of data collection.
Stability and validity of the findings
23 StatLine 2014). Given that the maximum yearly change in consumer preferences identified was a 57% change in the linear relationship between service quality and customer satisfaction, this would imply that consumer preferences can change at rates one hundred times faster than consumer populations.
There are two possible explanations that could account for the lack of stability over time of the spatial patterns. The first is that consumer preferences cannot be directly related to consumer populations. This means that a basic premise of the interdependence of consumer preferences hypothesis is false – consumer preferences are not geographically interdependent. Instead, they are dynamically interdependent. This is consistent with current marketing research on social interactions at the individual consumer level (Garber, Goldenberg, Libai, & Muller 2004; Iyengar, van den Bulte, & Valente 2011). The second explanation is that of sampling bias. In this explanation, in the absence of more conclusive empirical evidence, pharmacy service performance is simply a much stronger determinant of consumer service evaluations than the interdependence of consumer preferences. While something could be said for both explanations, neither explains why cross-regional spatial patterns could be identified in 2012 data. Unfortunately, because of data limitations, neither can be investigated in more detail – panel data for a bigger geographical area over a longer period of time would be required.
As a result, only a scrutiny of the internal validity of the methodology can be given. A Chow test was performed to check if pooling the data for the national-aggregate level was allowed (Leeflang, Bijmolt, Pauwels, & Wieringa 2014). -statistics of 3.30 and 2.34 at the NUTS3 and NUTS1 level respectively demonstrate that at the level, pooling the data is not allowed. These finding corresponds with the parameter estimates given in table 3 on page 20, where at the regional level widely divergent estimates were found. These results provide statistical evidence that in the case of consumer preferences, analyses at disaggregate regional levels should be preferred over analyses at the national aggregate level.
Level of analysis -stat. df. ̅̅̅̅̅̅ ̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅̅ ̅̅̅̅ ̅ ̅̅̅ ̂
NUTS3 24 14,206 3.30 14,109 0.514 0.510 26.50% 1174 35 41.75%
NUTS1 4 14,206 2.34 14,189 0.525 0.522 27.47% 8964 59 41.13%
24 Also included in table 4 on the previous page are the weighted arithmetic means of the 2012 estimation results (Grabisch et al. 2009). As can be seen from these figures, for both the linear and non-linear model measures of model fit go up as the sample size increases, demonstrating the presence of aggregation bias in both the regression and the cluster analyses. This effect is nevertheless much more pronounced for the MLE method. It can therefore be concluded that the Gaussian MLE mixture model estimated using the EM algorithm is much more sensitive to sampling bias (Clauset et al. 2009; Deville, Hosseini, & Deville 2011). This strong source dependency of the MLE method can also be seen in the classification results given in table 5 below. The table shows that in addition to , the bivariate cluster solutions on any given level of aggregation tend to include between 8 and 85 small clusters with of the survey population. These small clusters sum up to between 20% and 30% of the total survey population in a region, and could be considered the heterogeneous part of the population. Likewise, 4 to 8 larger clusters with between and of the survey population were identified in the bivariate cluster analyses. These larger clusters classify between 24% and 45% of the survey population on any given level of analysis, and could therefore potentially be used for a targeting strategy in combination with the cluster identified as .
Level of analysis ̅ ̅ Cum. Cum. Cum.
NUTS3 591 35 1 ( ) 44% 4-8 79% 8-45 100%
NUTS1 3,552 59 1 ( ) 43% 3-5 70% 37-85 100%
National 14,206 34 1 ( ) 39% 4 77% 29 100%
Table 5: Ranges of the 2012 cluster solutions at three levels of analysis and aggregation.
Table 5 nevertheless also demonstrates that the latent class cluster analyses applied in this thesis to identify homogeneity in consumer preferences might be too restrictive for practical purposes. Whether or not this is invalidates the empirical results and renders them useless for marketing practice will be discussed in the general discussion section that follows.
General discussion
25 parallel. This means that while marketing opportunities can be identified, trends in consumer preferences cannot, and as such the results argue against basing strategic marketing decisions on location data. This is not surprising given the status quo in spatial economics that “… market potential analysis offers a tantalizing hint of how it might be possible to think in terms of continuous space rather than prespecified regions” (Fujita et al. 1999: 33). So that while firm databases, mobile devices and clickstream data are generating a wealth of location data, the exponential increase in complexity of multilevel models that include both a temporal and a spatial dimension have so far hampered the application of location data in marketing practice (Bronnenberg 2005; Elhorst 2014).
Applicability to marketing practice
The findings nevertheless do call into question if the national aggregate level is the optimal level of analysis for consumer preferences. Spatial patterns found in the data indicate that cross-regional geo-targeting can sometimes be a better option. Furthermore, the results show that in a consumer population, service quality evaluations change dynamically over time. While this will certainly be to some extent due to the subjective and situational nature of customer satisfaction measurements, the changes over time in the spatial patterns also challenge earlier versions of the interdependence of consumer preferences hypothesis that suggest that interdependence is primarily a result of geographical proximity (Granovetter 1978; Lantané 1996). The empirical results presented in this thesis on the contrary indicate that the interdependence of consumer preferences in the 21st century is a dynamic rather than a geographic phenomenon. This would also mean that the best way to model consumer preferences is to model individual social interactions as a dynamic system, rather than through models estimated at the population level (Bosma et al. 2011; Helbing 2010; Iyengar et al. 2011). Location data could then be used to introduce a geographical component in the dynamic system (Garber et al. 2004). While such a model will in theory perform better, in practice its implementation also requires data that will rarely if ever be available to a services firm.
26 however also show that, like consumer preferences in product markets, any application in marketing practice of the interdependence of consumer preferences framework requires both continuous measurement and real-time updating and monitoring of the spatial patterns (Athey & Ellison 2011; Guadagni & Little 1983). This also means that similar to a weather system, the identification of long-term marketing opportunities for service delivery in local consumer populations remains impossible with existing services marketing models. While a number of marketing opportunities could potentially be identified using the spatial patterns indexed through the homogeneity of consumer preferences or the degree of linearity of the relationship between service quality and customer satisfaction, consumer demand for services remains the black box it has been for the past fifty years. It is suspected that at least part of the answer to these rather restrictive findings lies in the limitations in scale and scope of the empirical research presented in this thesis.
Limitations
27 on a weekly or monthly basis would provide a much better view of the dynamics of consumer preferences than data aggregated on a yearly basis.
28
Conclusion
Perhaps more surprising than finding spatial patterns in consumer evaluations of service quality is that these patterns were observed in 24 Dutch NUTS3 regions similar in size and population to the US state of Massachusetts. This is an important finding. It in fact forces services managers to rethink the value of marketing strategies on the national level, and look at regional or even sub-regional marketing strategies for services marketing. For some firms, this might not be feasible. Nevertheless, for larger SMEs and the LSEs already operating at the national level, the spatial patterns in consumer evaluations show that regional differences in consumer preferences can be identified that might provide valuable additional insights for customer relationship management and customer acquisition strategies.
29
References
Achrol, R.S. & Kotler, P. 1999. Marketing in the Network Economy. Journal of
Marketing, 63 (Fundamental Directions in Marketing Issue): 146-163
Albuquerque, P. & Bronnenberg, B.J. 2009. Estimating Demand Heterogeneity Using Aggregated Data: An Application to the Frozen Pizza Category. Marketing Science, 28(2): 356-372
Albuquerque, P. & Bronnenberg, B.J. 2012. Measuring the Impact of Negative Demand Shocks on Car Dealer Networks. Marketing Science, 31(1): 4-23
Allenby, G.M. & Rossi, P.E. 1999. Marketing models of consumer heterogeneity.
Journal of Econometrics, 89: 57-78
Amari, S.-I. 1995. Information Geometry of the EM and em Algorithms for Neural Networks. Neural Networks, 8(9): 1379-1408
Andrews, D.W.L. 1993. Tests for Parameter Instability and Structural Change With Unknown Change Point. Econometrica, 61(4): 821-856
Ansari, A., Mela, C.F. & Neslin, S.N. 2008. Customer Channel Migration. Journal of
Marketing Research, 45(1): 60-76
Ataman, M. B., Mela, C.F. & Van Heerde, H.J. 2007. Consumer Packaged Goods in France: National Brands, Regional Chains, and Local Branding. Journal of Marketing
Research, 44(1): 14–20
Ataman, M.B., Van Heerde, H.J. & Mela, C.F. 2010. The long-term effect of marketing strategy on brand sales. Journal of Marketing Research, 47(5): 866-882
Athey, S. & Ellison, G. 2011. Position Auctions with Consumer Search. The Quarterly
Journal of Economics, 126 : 1213–1270
Avery, J., Steenburgh, T.J., Deighton, J. & Caravella, M. 2012. Adding Bricks to Clicks: Predicting the Patterns of Cross-Channel Elasticities Over Time. Journal of Marketing, 76(3): 96–111
Banks, J., Blundell, R. & Lewbel, A. 1997. Quadratic Engel Curves and Consumer Demand. The Review of Economics and Statistics, 79(4): 527-539
Barndorff-Nielsen, O. 1978. Information and Exponential Families. Chicester, UK: John Wiley & Sons, Ltd.
Bass, F.M. 1969. A New Product Growth for Model Consumer Durables. Management
Science, 15 (5): 215-27
Bass, F.M. 1980. The Relationship Between Diffusion Rates, Experience Curves, and Demand Elasticities for Consumer Durable Technological Innovations. The Journal of
Business, 53(3) Part 2: S51-S67
Baudry, J.-P., Raftery, A.E., Celleux, G., Lo, K. & Gottardo, R. 2010. Combining Mixture Components for Clustering. J Comput Graph Stat., 9(2): 332–353
Berry, S.T., Levinsohn, J. & Pakes, A. 1995. Automobile prices in market equilibrium.
Econometrica, 63(4): 841–890
Bilbao-Osorio, B; Dutta, S. and Lanvin, B. (Eds.). 2013. The Global Information
Technology Report 2013: Growth and Jobs in a Hyperconnected World. The World
30
Blattberg, R.C., Kim, B.-D. & Neslin, S.A. 2008. Database Marketing: Analyzing and
Managing Customers. New York, NY: Springer Science+Business Media, LLC
Bloemer, J., De Ruyter, K. & Wetzels, M. 1999. Linking perceived service quality and service loyalty: a multi-dimensional perspective. European Journal of Marketing, 33(11-12): 1082-1106
Blundell, R., Pashardes, P. & Weber, G. 1993. What do We Learn About Consumer Demand Patterns from Micro Data? The American Economic Review, 83(3): 570-597
Bobrowski, O. & Mukherjee, S. 2014. The Topology of Probability Distributions on Manifolds. Probability Theory and Related Fields
Bogomolova, S. 2011. Service quality perceptions of solely loyal customers.
International Journal of Market Research, 53(6): 793-810
Bolton, R.N. & Lemon, K.N. 1999. A Dynamic Model of Customers' Usage of Services: Usage as an Antecedent and Consequence of Satisfaction. Journal of Marketing
Research, 36(3): 171-186
Bosma, N., Stam, E. & Schutjes, V. 2011. Creative destruction and regional productivity growth: evidence from the Dutch manufacturing and services industries. Small Bus Econ, 36: 401–418
Bronnenberg, B.J. & Sismeiro, C. 2002. Using Multimarket Data to Predict Brand Performance in Markets for Which No or Poor Data Exist. Journal of Marketing
Research, 39(1): 1–17
Bronnenberg, B.J. 2005. Spatial models in marketing research and practice. Appl.
Stochastic Models Bus. Ind., 21: 335–343
Bronnenberg, B.J., Dhar, S.K. & Dubé, J.-P. 2007 Consumer Packaged Goods in the United States: National Brands, Local Branding. Journal of Marketing Research, 44(1): 4–13
Bronnenberg, B.J., Dhar, S.K. & Dubé, J.-P.H. 2011. Endogenous sunk costs and the geographic differences in the market structures of CPG categories. Quant Mark Econ, (9): 1–23
Brynjolfsson, E. & McAfee, A. 2014. The Second Machine Age: Work, Progress, and
Prosperity in a Time of Brilliant Technologies. New York, NY: W.W. Norton &
Company, Inc.
Bucklin, R.E. & Sismeiro, C. 2009. Click here for Internet Insight: Advances in Clickstream Data Analysis in Marketing. Journal of Interactive Marketing, 23(1): 35
Burnham, K.P. & Anderson, D.R. 2002. Model Selection and Multimodel Inference: A
Practical Information-Theoretic Approach, Second Edition. New York, NY: Springer
Verlag New York, Inc.
Carlsson, G., Ishkhanov, T., de Silva, V. & Zomorodian, A. 2008. On the Local Behavior of Spaces of Natural Images. Int J Comput Vis, 76: 1–12
Carlsson, G. 2009. Topology and Data. Bull. Amer. Math. Soc., 46(2): 255–308
31
Cavalli-Sforza, L. L. & Feldman, M. M. 1981. Cultural Transmission and Evolution: A
Quantitative Approach. Princeton, NJ: Princeton University Press Monographs in
Population Biology 16
CBS StatLine. 2014. Centraal Bureau voor de Statistiek. Retrieved from www.cbs.nl, May 2014
Chen, Y. & Steckel, J.H. 2012. Modeling Credit Card Share of Wallet: Solving the Incomplete Information Problem. Journal of Marketing Research, 49(5): 655-669
Chu, J., Chintagunta, P.K. & Vilcassim, N.J. 2007. Assessing the Economic Value of Distribution Channels: An Application to the Personal Computer Industry. Journal of
Marketing Research, 44(1): 29-41
Cirera, X. & Masset, E. 2010. Income distribution trends and future food demand. Phil.
Trans. R. Soc. B, 365: 2821–2834
Clauset, A., Shalizi, C.R. & Newman, M.E.J. 2009. Power-Law Distributions in Empirical Data. SIAM Review, 51(4): 661–703
Cooil, B., Keiningham, T.L., Aksoy, L. & Hsu, M. 2007. A Longitudinal Analysis of Customer Satisfaction and Share of Wallet: Investigating the Moderating Effect of Customer Characteristics. Journal of Marketing, 71(1): 67–83
Conley, T.G. & Topa, G. 2002. Socio-Economic Distance and Spatial Patterns in Unemployment. J. Appl. Econ., 17: 303–327
Cox, D.R. 2006. Principles of Statistical Inference. Cambridge, UK: Cambridge University Press
De Graaf-Ruizendaal, W.A., Kenens, R.J. & De Bakker, D.H. 2012. Vraag Aanbod
Analyse Monitor: Verantwoording rekenmodellen versie 3.1. Utrecht, NL: NIVEL
Delre, S.A., Jager, W., Bijmolt, T.H.A. & Jansen, M.A. 2010. Will It Spread or Not? The Effects of Social Influences and Network Topology on Innovation Diffusion. Journal of
Product Innovation Management, 27: 267-282
Dempster, A.P., Laird, N.M. & Rubin, D.B. 1977. Maximum Likelihood from Incomplete Data via the EM Algorithm. Journal of the Royal Statistical Society, Series B
(Methodological), 39(1): 1-38
Deville, Y., Hosseini, S. & Deville, A. 2011. Effect of indirect dependencies on maximum likelihood and information theoretic blind source separation for nonlinear mixtures.
Signal Processing, 91: 793–800
Dixit, A.K. & Stiglitz, J.E. 1977. Monopolistic Competition and Optimum Product Diversity. The American Economic Review, 67(3): 297-308
Dubé, J.-P.H., Hitsch, G.J. & Chintagunta, P.K. 2010. Tipping and Concentration in Markets with Indirect Network Effects. Marketing Science, 29(2): 216–249
Dubé, J.-P.H., Fox, J.T. & Su, C.-L. 2012. Improving the Numerical Performance of Static and Dynamic Aggregate Discrete Choice Random Coefficients Demand Estimation.
Econometrica, 80(5): 2231–2267
Edwards, Y.D. & Allenby, G. 2003. Multivariate Analysis of Multiple Response Data.
Journal of Marketing Research, 40(2): 321–334
Eichengreen, B. & Gupta, P. 2013. The two waves of service-sector growth. Oxford
32
Einav, L. & Levin, J.D. 2013. The Data Revolution and Economic Analysis. NBER Working Paper 19035
Elhorst, J.P. 2014. Spatial Econometrics: From Cross-Sectional Data to Spatial Panels. Heidelberg, DE: Springer
Eurostat. 2014. NUTS classification. Retrieved from http://epp.eurostat.ec.europa.eu, July 2014
Farag, S., Krizek, K.J. & Dijst, M. 2006. E‐Shopping and its Relationship with In‐store Shopping: Empirical Evidence from the Netherlands and the USA. Transport Reviews: A
Transnational Transdisciplinary Journal, 26(1): 43-61
Feit, E.M.; Wang, P., Bradlow, E.T. & Fader, P.S. 2013. Fusing Aggregate and Disaggregate Data with an Application to Multiplatform Media Consumption. Journal of
Marketing Research, 50(3): 348–364
Figueiredo, M.A.T. & Jain, A.K. 2002. Unsupervised Learning of Finite Mixture Models.
IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(3): 381-396
Fox, J.-P. 2010. Bayesian Item Response Modeling: Theory and Applications. New York, NY: Springer Science+Business Media, LLC Statistics for Social and Behavioral Sciences
Fraley, C. & Raftery, A.E. 2002. Model-Based Clustering, Discriminant Analysis, and Density Estimation. Journal of the American Statistical Association, 97(458): 611-631
Fraley, C., Raftery, A.E., Murphy, T.B. & Scrucca, L. 2012. mclust Version 4 for R:
Normal Mixture Modeling for Model-Based Clustering, Classification, and Density Estimation. Department of Statistics, University of Washington: Technical Report No.
597
Franses, P.H. 2005. On the Use of Econometric Models for Policy Simulation in Marketing. Journal of Marketing Research, 42(1): 4-14
Fujita, M., Krugman, P. & Venables, A.J. 1999. The Spatial Economy: Cities, Regions,
and International Trade. Cambridge, MA: The MIT Press
Garber, T., Goldenberg, J., Libai, B. & Muller, E. 2004. From Density to Destiny: Using Spatial Dimension of Sales Data for Early Prediction of New Product Success. Marketing
Science, 23(3): 419-428
Geroski, P. A. 2003. the Evolution of New Markets. Oxford, UK: Oxford University Press
Gielens, K. & Dekimpe, M.G. 2007. The Entry Strategy of Retail Firms into Transition Economies. Journal of Marketing, 71(2): 196-212
Goodman Jr., J.L. 1998. Aggregation of Local Housing Markets. Journal of Real Estate
Finance and Economics, 16(1): 43-53
Grabisch, M. & Perney, P. 2002. Agrégation Multicritère. Downloaded from the author’s website, May 2014.
Grabisch, M., Marichal, J.-L., Mesiar, R. & Pap, E. 2009. Aggregation Functions. Cambridge, UK: Cambridge University Press, Encyclopedia of Mathematics and Its Applications 127
Granovetter, M. 1978. Threshold Models of Collective Behavior. American Journal of
Sociology, 83(6): 1420-1443
33
Grover, R. & Srinivasan, V. 1987. A Simultaneous Approach to Market Segmentation and Market Structuring. Journal of Marketing Research, 24(2): 139-153
Guadagni, P.M. & Little, J.D.C. 1983. A Logit Model of Brand Choice Calibrated on Scanner Data. Marketing Science, 2(3): 203-238
Hartley, H.O. 1958. Maximum Likelihood Estimation from Incomplete Data. Biometrics, 14(2): 174-194
Hartmann, W.R. 2010. Demand Estimation with Social Interactions and the Implications for Targeted Marketing. Marketing Science, 29(4): 585-601
Helbing, D. 2010. Quantitative Sociodynamics: Stochastic Methods and Models of
Social Interaction Processes, Second Edition. Berlin, DE: Springer-Verlag
Helpman, E. & Krugman, P.R. 1989. Trade Policy and Market Structure. Cambridge, MA: The MIT Press
Hollensen, S. 2011. Global Marketing: A Decision-Oriented Approach. Harlow, UK: Pearson Education Limited
Hotelling, H. 1931. The Economics of Exhaustible Resources. The Journal of Political
Economy, 39(2): 137-175
Huff, D.L. 1964. Defining and Estimating a Trading Area. Journal of Marketing, 28(3): 34-38
Iyengar, R., Ansari, A. & Gupta, S. 2007. A Model of Consumer Learning for Service Quality and Usage. Journal of Marketing Research, 44(4): 529–544
Iyengar, R., van den Bulte, C. & Valente, T.W. 2011. Opinion Leadership and Social Contagion in New Product Diffusion. Marketing Science, 30(2) 195–212
Jiao, J., Simpson, T.W., & Siddique, Z. 2007. Product family design and platform-based product development: a state-of-the-art review. J Intell Manuf, 18:5–29
Kahneman, D. & Tversky, A. 1979. Prospect Theory: An Analysis of Decision under Risk.
Econometrica, 47(2): 263-292
Kareev, I. 2013. Lower Bounds for Average Sample Size and Efficiency of Sequential Selection Procedures. Theory Probab. Appl., 57(2): 227–242
Kim, J.B., Albuquerque, P. & Bronnenberg, B.J. 2011. Mapping Online Consumer Search.
Journal of Marketing Research, 48(1): 13 –27
Kleinberg, J. 2002. An Impossibility Theorem for Clustering. In Becker, S., Thrun, S. & Obermayer, K. (eds.). 2002. Advances in Neural Information Processing Systems 15
Lafferty, J. & Lebanon, G. 2005. Diffusion Kernels on Statistical Manifolds. Journal of
Machine Learning Research, 6: 129–163
Lantané, B. 1996. Dynamic Social Impact: The Creation of Culture by Communication.
Journal of Communication, 46(4): 13-25
Larivière, B. & Van Den Poel, D. 2005. Predicting customer retention and profitability by using random forests and regression forests techniques. Expert Systems with Applications, 29: 472–484
Leeflang, P.S.H. & Reuyl, J.C. 1986. Estimating the parameters of market share models at different levels of aggregation with examples from the West German cigarette market.
European Journal of Operational Research, 23: 14-24
Leeflang, P.S.H., Wittink, D.R., Wedel, M. & Naert, P.A. 2000. Building Models for
34
Leeflang, P.S.H., Bijmolt, T.H.A., Pauwels, K.H. & Wieringa, J.E. 2014. Modeling
Markets. Unpublished, forthcoming (Springer).
Li, Y., Wang, N. & Carroll, R.J. 2013. Selecting the Number of Principal Components in Functional Data. Journal of the American Statistical Association, 108(504): 1284-1294
Liang, F., Paulo, R., Molina, G., Clyde, M. A. & Berger, J. O. 2008. Mixtures of g-priors for Bayesian Variable Selection. Journal of the American Statistical Association, 103: 410-423
Libai, B., Muller, E. & Peres, R. 2009. The Diffusion of Services. Journal of Marketing
Research, 46(2): 163–175
Liski, E.P. & Nummi, T. 1990. Prediction in growth curve models using the EM algorithm. Computational Statistics & Data Analysis, 10: 99-108
Mägi, A.W. 2003. Share of wallet in retailing: the effects of customer satisfaction, loyalty cards and shopper characteristics. Journal of Retailing, 79: 97–106
McFadden, D., Talvitie, A.P., et al. 1977. Demand Model Estimation and Validation. The Institute of Transportation Studies, University of California at Berkley and Irvine, Urban Travel Demand Forecasting Project Phase 1 Final Report Series, Vol. V
McKinsey & Company. 2014. McKinsey Quarterly 2014 Number 1: Shaping the future
of manufacturing. New York, NY: McKinsey & Company
McLachlan, G.J. & Krishnan, T. 1997. The EM Algorithm and Extensions. New York, NY: John Wiley & Sons, Inc
Melton, H.L. & Hartline, M.D. 2010. Customer and Frontline Employee Influence on New Service Development Performance. Journal of Service Research, 13: 411-425
Meyers, R.A. (ed.) 2012. Mathematics of Complexity and Dynamic Systems. New York,
NY: SpringerScience+Business Media, LLC.
Miceli, G., Raimondo, M.A. & Farace, S. 2013. Customer Attitude and Dispositions Towards Customized Products: The Interaction Between Customization Model and Brand.
Journal of Interactive Marketing, 27: 209–225
Mittal, V. & Kamakura, W.A. 2001. Satisfaction, Repurchase Intent, and Repurchase Behavior: Investigating the Moderating Effect of Customer Characteristics. Journal of
Marketing Research, 38(1) : 31-142
Narayanan, S., Desiraju, R., & Chintagunta, P.K. 2004. Return on Investment Implications for Pharmaceutical Promotional Expenditures: The Role of Marketing-Mix Interactions. Journal of Marketing, 68(4): 90-105
Neslin, S.A., Grewal, D.L, Shankar, V., Teerling, M.L., Thomas, J.S. & Verhoef, P.C. 2006. Challenges and Opportunities in Multichannel Customer Management. Journal of
Service Research, 9(2): 95-112
Nitzan, I. & Libai, B. 2011. Social Effects on Customer Retention. Journal of Marketing, 75(6): 24-38
Niyogi, P., Smale, S. & Weinberger, S. 2011. A Topological View of Unsupervised Learning from Noisy Data. SIAM J. Comput., 40(3): 646–663
NPCF-VAAM. 2014. Vraag Aanbod Analyse Monitor. Retrieved from www.npcf.nl, June 2014
35
Pesaran, M.H., Schuermann, T. & Weiner, S.M. 2004. Modeling Regional Interdependencies Using a Global Error-Correcting Macroeconometric Model. Journal of
Business & Economic Statistics, 22(2): 129-162
Pischke, J.-S. 1995. Individual Income, Incomplete Information, and Aggregate Consumption. Econometrica, 63(4): 805-840
Polo, Y., Sese, F.J. & Verhoef, P.C. 2011. The Effect of Pricing and Advertising on Customer Retention in a Liberalizing Market. Journal of Interactive Marketing, 25: 201-214
Porter, M. E. 1990. The Competitive Advantage of Nations. New York, NY: The Free Press
Risselada, H., Verhoef, P.C. & Bijmolt, T.H.A. 2010. Staying Power of Churn Prediction Models. Journal of Interactive Marketing, 24: 198–208
Rochet, J.-C. & Tirole, J. 2006. Two-Sided Markets: A Progress Report. The RAND
Journal of Economics, 37(3): 645-667
Rogers, E. M. 2003. Diffusion of Innovations (fifth edition). New York, NY: Free Press
Russel, G.J. & Bolton, R.N. 1988. Implications of Market Structure for Elasticity Structure. Journal of Marketing Research, 25(3): 229-41
Rust, R.T. & Chung, T.S. 2006. Marketing Models of Service and Relationships.
Marketing Science, 25(6): 560-580
Rutz, O.J. & Sonnier, G.P. 2011. The Evolution of Internal Market Structure. Marketing
Science, 30(2):274-289
Schennach, S.M. 2014. Entropic Latent Variable Integration Via Simulation.
Econometrica, 82(1) : 345–385
Schwarz, G. 1978. Estimating the Dimension of a Model. The Annals of Statistics, 6(2): 461-464
Schweidel, D.A., Fader, P.S. & Bradlow, E.T. 2008. Understanding Service Retention Within and Across Cohorts Using Limited Information. Journal of Marketing, 72 (1): 82–94
Seetharaman, P.B., Ainslie, A. & Chintagunta, P.K. 1999. Investigating Household State Dependence Effects across Categories. Journal of Marketing Research, 36(4): 488-500
SFK. 2012. Compiled by Griens, A.M.G.F., Janssen, J.M., Kroon, J.D.L., Lukaart, J.S. & van der Vaart, R.J. 2012. SFK Data en feiten 2012: Het jaar 2011 in cijfers. Den Haag, NL: Stichting Farmaceutische Kengetallen
SFK. 2013. Compiled by Griens, A.M.G.F., Janssen, J.M., Kroon, J.D.L., Lukaart, J.S. & van der Vaart, R.J. 2013. SFK Data en feiten 2013: Het jaar 2012 in cijfers. Den Haag, NL: Stichting Farmaceutische Kengetallen
Shimogawa, S., Shinno, M., & Saito, H. 2012. Structure of S-shaped growth in innovation diffusion. Physical Review E, 85: 05612
Sieling, H. 2013. Statistical Multiscale Segmentation: Inference, Algorithms and
Applications. Göttingen, DE: Dissertation zur Erlangung des
