Available online at www.sciencedirect.

com

Journal of Interactive Marketing 25 (2011) 201 214 www.elsevier.com/locate/intmar

The Effect of Pricing and Advertising on Customer Retention in a Liberalizing Market

Yolanda Polo a , F. Javier Sese b, & Peter C. Verhoef c
c

Department of Marketing, Faculty of Economics and Business, University of Zaragoza, Gran Via 2, 50005, Zaragoza, Spain Department of Marketing, Faculty of Economics and Business, University of Zaragoza, Mara de Luna, s/n, 50018, Zaragoza, Spain Department of Marketing, Faculty of Economics and Business, University of Groningen, Office Duisenberg Building 329, P.O. Box 800, 9700 AV Groningen, The Netherlands
b

Available online 29 March 2011

Abstract This study investigates the drivers of customer retention in a liberalizing market. The authors address key retention issues that allow them to contribute to existing retention research in several critical ways. They (1) examine the effects of pricing and mass advertising, (2) account for (new entrants) competitors' actions, (3) investigate the dynamic impact of marketing tactics, and (4) study the proposed relationships in a market recently opened to competition. Using longitudinal data for a sample of 650 mobile phone consumers and a split-population hazard model that accounts for the notion that some customers are never at risk of defection, the authors show that both the focal firm's (incumbent) and the competitors' price and mass advertising exert a significant influence on the probability of terminating an existing incumbent relationship. They find that the relationships between marketing variables and retention are not static but vary over time. Price is generally less effective in the early stages of market liberalization, which suggests that customers become more price sensitive in later stages. Finally, the study findings can have important strategic implications on designing customer management and marketing resource allocation strategies, as well as on providing a competitive regulatory framework in liberalizing markets. 2011 Direct Marketing Educational Foundation, Inc. Published by Elsevier Inc. All rights reserved.
Keywords: Customer retention; Pricing; Mass advertising; Liberalizing market; Split-population hazard model

Introduction Customer retention and the building of long-standing relationships with customers are central elements in the creation of value (Berger and Nasr 1998; Gupta, Lehmann, and Stuart 2004). For this reason, both marketing research and practice show great interest in understanding the driving forces behind establishing, developing, and maintaining successful relational

Yolanda Polo and F. Javier Sese appreciate the financial support received from the following projects: SEC2008-04704 (MEyC, FEDER), PI 138/08, and S09-PM062 from the Spanish Regional Government of Aragn, as well as for their recognition as members of the Generes research group. The authors thank the Editor Edward Malthouse and the two anonymous reviewers for helpful comments during the review process. Corresponding author. E-mail addresses: ypolo@unizar.es (Y. Polo), javisese@unizar.es (F.J. Sese), p.c.verhoef@rug.nl (P.C. Verhoef).

exchanges (Morgan and Hunt 1994, p. 22). Various studies have investigated the antecedents of customer retention or customer relationship duration (Bolton 1998; Drze and Bonfrer 2008; Pfeifer and Farris 2004; Reinartz and Kumar 2003; Verhoef 2003). However, such studies often neglect the effect of mass marketing instruments (Verhoef, van Doorn, and Dorotic 2007), the influence of pricing (Blattberg, Malthouse, and Neslin 2009), and/or the impact of competitors' pricing and advertising strategies on a focal firm's relationships (Gupta and Zeithaml 2006). Recently, some studies have started to consider these influences. Prins and Verhoef (2007) note the impact of advertising expenditures by the firm and its competitors on customer adoption of a new e-service. Venkatesan, Reinartz, and Ravishanker (2009) investigate the impact of customer attitudes toward the focal firm and the competitors on customer behavior in the presence of CRM activities. Rust, Lemon, and Zeithaml (2004) examine the effect of brand advertising on

1094-9968/$ - see front matter 2011 Direct Marketing Educational Foundation, Inc. Published by Elsevier Inc. All rights reserved. doi:10.1016/j.intmar.2011.02.002

202

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214

retention and explicitly incorporate competition to project financial returns from marketing investments. Dawes (2009) investigates the impact of price levels and increases on customer retention and finds that a price increase leads to greater churn, though the effect differs across customer groups. Notably, studies in this research stream have only investigated liberalized markets (Wieringa and Verhoef 2007). The deregulation of strategic, utility goods markets (e.g., power, transport and telecommunications) is occurring in many countries (Roberts, Nelson, and Morrison 2005) and forces state-owned monopolists (i.e., incumbents) to compete with new entrants. Existing customer retention knowledge might, therefore, improve significantly with further empirical applications in liberalizing markets,1 and this knowledge may be of significant value to incumbents, who must understand the determinants of relational continuity and develop defensive strategies (Hauser and Shugan 1983), as well as to new entrants, who require an understanding of the factors that motivate existing customers to switch (Green, Barclay, and Ryans 1995). Research that examines defensive and offensive marketing strategies in these newly competitive markets remains limited. We respond to this gap by investigating the extent to which the incumbent's and competitors' price and advertising affect customer retention in a liberalizing market. We have the following research objectives and areas of contribution to the literature. First, we assess the impact of the incumbent's price and mass (or above-the-line) advertising efforts on customer retention and, thereby, identify specific actions that incumbents can develop to increase relationship duration and profitability (Bolton 1998). Second, we provide insights into the effects of pricing and mass advertising by competitors to help them identify the issues that must be addressed in order to attract customers from the incumbent firm. Third, we investigate whether these effects change over time in a liberalizing market and, thus, help clarify its changing dynamics. Fourth, we offer a methodological novelty and apply a split-population hazard model that takes a segment of customers who never leave the incumbent company into account and, thereby, offer insights into how incumbent firms and new entrants should formulate their strategies to allocate marketing resources optimally. This knowledge is relevant for public policy officials too because they have to design a fullycompetitive regulatory framework. We rely particularly on Bolton's (1998) research which examines the impact of price on customer retention. She finds that an increase in price decreases retention. In the context of the recently liberalized Spanish mobile phone market, we extend Bolton's research in several critical ways. We study the effect of advertising on the timing to customer defection, explicitly account for competitors' prices and advertising, and examine these dynamic relationships in a liberalizing market. Using a sample of 650 mobile phone customers and a split-population
1 We use the term liberalizing market to refer to a market previously dominated by a state-owned monopoly that has been opened to allow other firms to compete (Roberts, Nelson, and Morrison 2005; Wieringa and Verhoef 2007).

hazard model, our results reveal that the length of an incumbent relationship depends mainly on the incumbent's price and advertising, as well as the competitors' price, that the effects of pricing on retention strengthen over time, which implies that customers become more price sensitive, and that a segment of incumbent customers has a zero probability of terminating the relationship (and switching to a new provider). The remainder of this manuscript proceeds as follows: In the next section, we review previous research about the role of price and above-the-line advertising in customer retention and discuss their expected effects in the context of liberalizing markets. We then describe the study context and data set that we use to examine the drivers of retention and discuss the research methodology. Subsequently, we present the estimation results and conclude with a discussion of the main findings and contributions of our study. Theory and Hypotheses We consider two types of firm in a liberalizing market: the incumbent firm and competitors that enter the market. In our study, we are interested not only in the direct impact of price and mass advertising of both the incumbent firm and its competitors on the probability that incumbent firm's customers terminate their relationships and switch to the new entrants, but also their potential dynamic effects. To outline the role of pricing, we begin by hypothesizing about the effects of the incumbent firm's price before continuing with the effects of competitive pricing and, finally, discussing the potential dynamic effects of both incumbent and competitive pricing. We follow a similar pattern in our discussion of the role of advertising in retention. Price Incumbent Price and Customer Retention Price consistently appears as an important conceptual determinant of customer behavior (Bolton, Lemon, and Verhoef 2004). Pricing research in various disciplines, including microeconomics, psychology and marketing, concludes that price explains more variance in customer purchase decisions than other marketing mix variables do (Keaveney 1995; Winer 1986). Customer relationship management (CRM) studies have mainly focused on the role of price for acquiring new customers (e.g., Lewis 2006). However, researchers have also acknowledged that pricing strategies to acquire customers might also impact customer retention and customer lifetime value (e.g., Lewis 2006; Verhoef and Donkers 2005). Marketing modeling studies often consider the impact of price and price promotions on market performance metrics such as sales and market share (e.g., Bijmolt, van Heerde, and Pieters 2005; Nijs et al. 2001). Meta-analytical evidence suggests high price elasticities and, at the consumer level, microeconomic theory suggests that utility-maximizing consumers respond to changes in price by changing their product or service demand (Guadagni and Little 1983; Winer 1986). The consideration of price remains rather limited for CRM research. Some studies note the impact of price perceptions or payment equity, which

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214

203

refers to the consumer's perception of the fairness of the price paid for products and services (Bolton and Lemon 1999). Studies in these research domains indicate that better price perceptions induce longer customerfirm relationships (e.g., Bolton, Lemon, and Verhoef 2004). Keaveney (1995) categorizes price as a critical trigger of switching decisions. Bolton, Lemon, and Bramlett (2006) find that higher-priced options have a lower probability of being renewed. Similarly, Dawes (2009) shows that both high prices and price increases create churn. Therefore, the incumbent's price should relate negatively to customer retention, and we hypothesize: H1A. Incumbent price has a negative effect on incumbent firm's customer retention. Competitive Pricing and Customer Retention New firms enter liberalizing markets aggressively to undermine the dominant position of the incumbent and acquire its customers. To influence customer behavior, they frequently use price as a compelling marketing instrument (Hauser and Shugan 1983), yet CRM studies provide limited knowledge about the effect of such competitive pricing. Researchers consider how differences in payment equity levels might affect customer behavior (e.g., Verhoef, Franses, and Hoekstra 2001), though without conclusive results. Marketing literature also suggests that price decreases by competitors lower brand sales. Similarly, we expect competitive prices to influence the ability of the incumbent to retain its customers. The aggressive price policies of new entrants should encourage consumers to switch to the new supplier to obtain immediate cost savings. We hypothesize that: H1B. Competitors' prices have a positive effect on incumbent firm's customer retention. Dynamic Effects of Price on Customer Retention The relationship between price and retention varies over time (e.g. as a result of market developments) (Bolton, Lemon, and Verhoef 2004). Several arguments are critical to understand the dynamic effects of price. First, research on first-mover advantages indicates that switching costs develop over the course of a relationship (Klemperer 1995; Polo and Sese 2009). Switching costs require new entrants to invest additional resources and set their prices more aggressively than the incumbent to acquire customers (Chen 1997; Lieberman and Montgomery 1988), which suggests a weaker relationship between price and customer switching. Wieringa and Verhoef (2007) find a significant portion of inertial consumers in the Dutch energy market. Then, we expect that the effect of price should be weaker in the early than in the later phases of liberalization. Second, according to information economics, people use signals to infer unobservable product and service quality in the presence of information asymmetries (Kirmani and Rao 2000). A high price reflects better inputs and higher quality. Similarly, traditional behavioral paradigms suggest that price reflects a sacrifice that the consumer makes to obtain a product or service (Rao and Monroe 1988), so they are willing to pay high prices only to obtain a high-quality product. Information asymmetries tend to be significant in recently liberalized markets (Wieringa

and Verhoef 2007) because, although customers are familiar with the former monopolist, they lack information to evaluate the quality of a new entrant's products and services. A low-price entrance strategy, therefore, should be associated with lower perceived quality, which would discourage switching behavior among existing customers. Overall, in markets recently exposed to competition, we expect the effect of price to be complex and differ depending on the stage of liberalization. We anticipate weaker effects of the incumbent's and competitors' prices on retention in the early phase of liberalization and hypothesize that: H1C. The negative effect of the incumbent price on incumbent firm's customer retention is weaker in early stages of the liberalization process than in later stages. H1D. The positive effect of competitors' prices on incumbent firm's customer retention is weaker in early stages of the liberalization process than in later stages. Mass Advertising Incumbent Mass Advertising and Customer Retention Prior CRM research regards advertising as important for the acquisition of new customers. Thus, it should be particularly crucial in the early stages of the customer lifecycle. Advertising informs prospective users about the existence, characteristics and economic value of an offer, creates awareness and knowledge among consumers, shapes their preferences and affects their purchase behavior (Mehta, Chen, and Narasimhan 2008). Once the relationship is established, advertising might still play an important role by creating further brand preferences and positive brand associations (Vakratsas and Ambler 1999) that can strengthen the relationship over time. Moreover, advertising can influence how consumers evaluate and experience the product or service through what is referred to as the transformative effect of advertising (Mehta, Chen, and Narasimhan 2008). Therefore, advertising positively influences the customer's experience of the product or service, which should result in increased customer loyalty (e.g., Verhoef et al. 2009). Most CRM researchers focus on the impact of customerspecific marketing interventions, or below-the-line advertising (De Wulf, Odekerken-Schrder, and Iacobucci 2001; Verhoef 2003), on subsequent customer behavior, largely ignoring mass advertising efforts. In a notable exception, Rust, Zeithaml and Lemon (2000) reveal the positive impact of branding on customer repurchase intentions. Prins and Verhoef (2007) investigate the effect of mass advertising efforts on the adoption of a new service and find positive effects on adoption timing. We, therefore, expect that the incumbent's advertising efforts positively affect its customer retention. H2A. Incumbent mass advertising has a positive effect on incumbent firm's customer retention. Competitive Mass Advertising and Customer Retention In contrast, competitors' mass advertising should induce switching among consumers of the focal supplier. Advertising

204

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214

is a key source of product and service information that reduces the risk of the buying process and the perceived costs of switching brands (Klemperer 1995). Moreover, advertising that highlights the advantages and strengths of competitors' products and services may weaken the existing customer relationship with the incumbent firm by making customers aware of the potential benefits of switching. In support of these arguments, Shum (2004) finds that advertising by competitors in the breakfast cereal market overcomes brand loyalty and encourages switching. We hypothesize that: H2B. Competitors' mass advertising has a negative effect on incumbent firm's customer retention. Dynamic Effects of Mass Advertising on Customer Retention Markets recently opened to competition should reveal a complex and varying effect of advertising, depending on the status of the advertiser (incumbent versus new competitors). Information economics literature indicates that information asymmetry about product quality means that firms can use advertising to signal their true product quality (Kirmani and Rao 2000). Because consumers are more likely to repurchase highquality products, high-quality firms invest more in advertising to attract initial sales (Milgrom and Roberts 1986). However, the incumbent and new entrants have different motivations to advertise. The quality of the incumbent's products is already well known, so it advertises mainly to reinforce, rather than initiate, behavior (Tellis 1988). In contrast, customers have little knowledge about the quality of new market entrants (Wieringa and Verhoef 2007), so new competitors mostly use advertising as a quality signal, that is, as a means to induce trial among incumbent customers. Research on advertising dynamics also suggests that advertising elasticities decrease over a product's life cycle (Assmus, Farley, and Lehmann 1984; Tellis 1988). According to Chandy et al. (2001), in a young market, customers lack the knowledge and experience to assess product attributes and judge quality whereas, in older markets, customers have experience with the offer and have less motivation to use advertising information in their evaluations. Deregulated markets follow a monopoly scenario, so advertising efforts by both the incumbent and new entrants may be less effective in influencing customer behavior. However, at the brand level, advertising elasticities of new brands are greater than those for established brands (Hanssens, Parsons, and Schultz 2001), so communication efforts by the new competitors may be more effective in altering customer behavior than efforts by the incumbent. By the same token, consumer inertia and habitual behavior in deregulated markets require unknown brands to invest more heavily in advertising to induce trial (Tellis 1988). The order of entry also influences consumer preferences (Lieberman and Montgomery 1988). The first entrant usually receives disproportionate attention in the consumer's mind, along with favorable brand attitudes and preferences (Carpenter and Nakamoto 1989). Customers tend to associate the first firm to enter a market with superior product attributes and lower risk (Niedrich and Swain 2003) whereas competitors need to

advertise more to be noticed (Lieberman and Montgomery 1988). This theoretical reasoning receives empirical support from Tellis's (1988) finding that customer inertia and habitual behavior force unknown brands to invest more in advertising to affect customer behavior. These arguments suggest a complex relationship between competitive mass advertising and customer retention that changes over time. At the beginning of a liberalization period, competitive mass advertising may effectively induce switching behavior but, over time, this effect decreases. We expect a similar dynamic pattern for the effect of the incumbent's mass advertising. Therefore, H2C. The positive effect of incumbent mass advertising on incumbent firm's customer retention is stronger in early stages of the liberalization process than in later stages. H2D. The negative effect of competitors' mass advertising on incumbent firm's customer retention is stronger in early stages of the liberalization process than in later stages. Study Context and Data We study customer retention in the Spanish mobile phone industry. Europe has long been home to many regulated markets (Leeflang and van Raaij 1995) and, until recently, (mobile) communication services were provided in isolation by stateowned companies in most countries. Various EU directives established foundations for liberalizing the voice telephone service and its related infrastructure in the early 1990s. Since then, most national telecom authorities have begun to open their monopolies to competition. In Spain, the state-owned company (Movistar) became privately owned in 1997 and the market was officially deregulated in December 1998. However, it took some time until these directives were put into practice, including mobile number portability (MNP),2 and it was not until October 2000 that the market became truly liberalized. At that date, Movistar had been providing mobile services for more than 10 years and its customer base consisted of more than 10 million users. Vodafone (initially Airtel) and Orange (initially Amena) were the two new entrants that competed with the established firm for mobile communications consumers. The mobile phone industry offers an attractive setting to study retention for several reasons. First, this industry has only recently been opened to competition, and the incumbents' and new entrants' marketing strategies diverge. The incumbent aims to retain its existing customers and maintain its first-mover advantage. The new entrants focus on creating a customer base that will provide a foundation for their future profitability. The heterogeneity in these strategies provides a high degree of variation in the independent variables and enables us to test for the associations between price and mass advertising and
2 Mobile number portability is a regulatory policy that enables mobile phone users to keep their telephone number when they change networks. The absence of this measure has been identified as one of the most important barriers to competition in the mobile phone context (Buehler, Dewenter, and Haucap 2006).

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214

205

customer retention. Second, in the mobile phone industry, firms incur significant upfront investments to acquire customers (e.g., free handsets, MNP), which increases the damaging impact of customer switching. Thus, firm profitability depends heavily on the ability to build long-term relationships. Third, in the mobile phone industry, reducing churn can create important value (Gupta, Lehmann, and Stuart 2004; IDATE 2009). Multiple studies model churn in the mobile phone industry (e.g., Lemmens and Croux 2006; Neslin et al. 2006). In 2007, more than 4 million consumers switched mobile phone operators in the Spanish market, a more than 10% annual churn rate (CMT 2007). Churn rates in many other industries tend to be much lower on average (e.g., Risselada, Verhoef, and Bijmolt 2010). Fourth, industry development is of central importance to public policy officials in charge of providing a competitive regulatory framework. The efficient design of this framework requires sufficient market knowledge and an understanding of customerfirm relationships over time (e.g., to implement MNP; Maicas, Polo, and Sese, 2009). To study customer retention in a liberalizing market, we use a sample of 650 mobile phone customers who were randomly selected from a population of active Movistar mobile customers as of January 2001.3 For each of the 650 customers, the market research company provided longitudinal (monthly) information that describes the customerfirm relationship from January 2001 to December 2004, including the date the customer entered into a relationship with Movistar (the earliest entries occurred in February 1995), mobile phone spending and the acquisition of additional services (cross-buying). Fig. 1 displays the data structure and duration timeline for four customers. In this unbalanced data set, some customers appear throughout the whole timeframe (i.e., do not switch providers) but others switch before the end of the observation period to the new entrants. To test for the influence of the incumbent's and competitors' marketing efforts on retention, we collected monthly data about pricing and advertising efforts by the three companies that provided mobile phone services in the Spanish market during the study period. We merged these data sources to create a comprehensive database that combines customer-level information with firm-level marketing strategies. The measures of the explanatory variables that we introduce in the empirical application are shown in Table 1. We measure price as the average revenue per user (ARpU). In the mobile phone industry, prices are very complex and normally involve nonlinear tariffs for different call types. These schemes also imply different rates per usage and different charges for various services (e.g., calls, data transfer and text and image messaging), which increases the complexity of developing a single price measure. Given the lack of individual-level data, empirical research usually employs simpler, averaged measures that approximate the price charged by each mobile phone operator. Although ARpU, which appears in previous retention research (Shy 2002), is not independent of service usage, it
3 Coincidentally, we have the same number of observations as Bolton (1998). However, the data set used in this paper is different from that used in Bolton (1998).

Data structure
Customer 1 Customer 2 Customer 3 Customer 4

February 1995

January 2001

December 2004

Beginning of the relationship Legend:

Beginning of relationship Relationship termination

Observation window

Fig. 1. Data structure.

offers a good proxy for price in the mobile communications industry (McCloughan and Lyons 2006).4 We distinguish between service and brand advertising. Service advertising informs existing and prospective customers about service attributes and characteristics whereas brand advertising aims to create favorable brand associations and preferences (Prins and Verhoef 2007). We measure service advertising as all expenditures to communicate about the service and its characteristics and brand advertising as expenditures related to advertising the brand (not the service). Because of the potential for endogeneity in the marketing mix, we lag ARpU and the two advertising variables by one period in our empirical application to reduce potential correlations between marketing variables and error terms (Venkatesan and Kumar 2004). Consistent with prior research (Dover and Murthi 2006; Reinartz, Thomas, and Bascoul 2008), we introduce relationship characteristics (spending, cross-buying, type of contract) as an additional source of variance in customer retention. We also control for observed heterogeneity by introducing a set of customer demographic variables (gender, age and employment status). Econometric Model To investigate the drivers of customer retention in a liberalizing market, we use a hazard model. This modeling approach is uniquely suited to addressing the following issues that are present in our dataset (Chintagunta and Dong 2006; Helsen and Schmittlein 1993): (1) right censoring (customers for whom we do not observe the event of interestrelationship
4 Another price measure unaffected by the consumers' usage rates is revenue per minute (RpM), which has only been reported recently by Spanish mobile operators and is not available for our observation window. The high correlation between ARpU and RpM in the mobile communications markets in the European Union (N 0.7, Global Wireless Matrix Database 2008) suggests that the study results would not be significantly affected by the type of price measure.

206 Table 1 Measurement of explanatory variables. Variable Incumbent marketing instruments Movistar's ARpU Movistar's service advertising Movistar's brand advertising Competitors marketing instruments Vodafone's ARpU Vodafone's service advertising Vodafone's brand advertising Orange's ARpU Orange's service advertising Orange's brand advertising Relationship characteristics Mobile phone spending Cross-buying Type of contract Label

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214

Firm customer level Firm-level Firm-level Firm-level

Measure ARpU (average revenue per user) of Movistar in period t 1 Log of service advertising expenses by Movistar in period t 1 Log of brand advertising expenses by Movistar in period t 1 ARpU of Vodafone in period t 1 Log of service advertising expenses by Vodafone in period t 1 Log of brand advertising expenses by Vodafone in period t 1 ARpU of Orange in period t 1 Log of service advertising expenses by Orange in period t 1 Log of brand advertising expenses by Orange in period t 1

ARpUMt-1 SaMt 1 BaMt 1

ARpUVt 1 SaVt 1 BaVt 1 ARpUOt 1 SaOt 1 BaOt 1

Firm-level Firm-level Firm-level Firm-level Firm-level Firm-level

Spit Cbit Tcit

Customer-level Customer-level Customer-level

Mobile phone spending by customer i in period t. Dummy variable: 1 = customer i acquires an additional service from the focal company in period t; 0 = otherwise Dummy variable: 1 = customer i has a postpaid subscription in period t; 0 = customer i has a prepaid subscription in period t.

Demographics Gender Age Employment status

Gei Agit Esit

Customer-level Customer-level Customer-level

Dummy variable: 1 = female, 0 = male Age (in years) of customer i in period t. Dummy variable: 1 = customer i is employed; 0 = customer i is unemployed

termination and switching to a new entrantat the end of the study period), and (2) time-varying covariates. Specifically, we use a split-population hazard model. The standard hazard model assumes that all individuals eventually experience the event of interest. However, in markets recently opened to competition, a portion of customers may have a zero probability of switching providers, especially those who have been forced to purchase from the monopolist for a long period and have no experience with alternative options or switching (Wieringa and Verhoef 2007). In the marketing literature, this behavioral pattern has also been referred to as state-dependence or spurious loyalty (Seetharaman, Ainslie, and Chintagunta 1999). This aspect is important for managing customerfirm relationships and allocating marketing resources optimally because customers who have a zero probability of switching will not respond to new entrants' or the incumbent's marketing stimuli. This also has important implications from a modeling perspective because standard hazard models would not take into account the fact that the estimated parameters for price and advertising would be zero for this group of non-switching customers. We therefore follow Schmidt and Witte (1989) and extend the standard hazard model by assuming that a subset of customers has a zero probability of terminating the relationship. That is, the model segments the population into two subpopulations, one that will experience relationship termination and one that will never experience it. The split-population hazard model simultaneously estimates coefficients for both the event's timing, conditional on its occurrence, and the probability that the event will ever occur. Split-population hazard models have been used in multiple marketing contexts (Prins and Verhoef 2007; Schweidel, Fader, and Bradlow 2008;

Sinha and Chandrashekaran 1992; Srinivasan, Lilien, and Rangaswamy 2006). Formally, the model can be expressed as follows. Let Ti, the time to relationship termination (customer defection from the incumbent and switching to a new entrant) for individual i, be a random variable with a cumulative distribution function F(t) and a density function f(t) = F(t). The survivor function S(t) = 1 F(t) is the probability that defection has not yet occurred at time t. The hazard function hi(t) is the conditional probability that defection occurs at time t, given that it has not occurred yet. From our data, we observe relationship termination only for customers who defect before the end of the study period. Those who have not yet defected at that time (right-censored observations) will either defect some time after the end of the study period or never defect at all. Thus, we also need to explicitly model the probability of eventual defection for every individual customer. The splitpopulation hazard model that we develop facilitates the simultaneous estimation of (1) the timing to defection and (2) the probability of eventual defection. We model the timing to customer defection as a hazard function of time-varying and time-invariant covariates. For discrete time processes, where the information is measured at discrete time intervals (months), we observe a longitudinal series of binary outcomes denoting whether or not customer defection occurred at the observation point, which act as the response variable. We specify a baseline hazard function h0(t), which allows the intrinsic probability that the event occurs (the probability without any explanatory variable) to change over time, separate from the effects of the covariates (Chintagunta and Dong 2006). Similar to Prins and Verhoef (2007), the parametric form we use for the baseline hazard function is the

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214

207

complementary loglog. We represent the hazard part of our model, which includes customer-level variables (subscript i) and firm-level variables (subscript t), as follows: hi t = 1exp exp0 + 1 ARpUMt1 + 2 SaMt1 + 3 BaMt1 + 4 ARpUVt1 + 5 SaVt1 + 6 BaVt1 + 7 ARpUOt1 + 8 SaOt1 + 9 BaOt1 + 10 Usit + 11 Us2 it + 12 Cbit + 13 Tcit + 14 Gei + 15 Agit + 16 Esit 2 where ARpUM, ARpUV, and ARpUO are, respectively, ARpU for Movistar, Vodafone, and Orange; SaM, SaV, and SaO are, respectively, service advertising by Movistar, Vodafone, and Orange; BaM, BaV, and BaO are, respectively, brand advertising by Movistar, Vodafone, and Orange; Sp and Sp2 are mobile phone spending and its squared term, respectively; Cb is cross-buying behavior; Tc is the type of mobile contract; Ge is gender; Ag is age; Es is employment status; and is the parameter vector that captures the effect of the explanatory variables (see Table 1 for a more detailed description) on the hazard rate (hi(t)). Simultaneously, we model the unobserved probability of eventual defection pi as a function of customer-specific variables (customer relationship characteristics and customer demographics):
2 3 pi = 1 + expY 1 = 1 + exp 0 + 1 Usi + 2 Ui + 3 Cbi + ; + 4 Tci + 5 Gei + 6 Agi + 7 Esi

Estimation Results Of the 650 consumers in the sample, 173 experienced the event of interest (defection) during the study period (January 2001December 2004), a 26.6% defection rate, and switched to alternative service providers (i.e., Vodafone or Orange). The 173 switching consumers were heterogeneously captured by the two competitors: 43% (74 consumers) switched to Vodafone and 57% (99 consumers) moved to Orange. We first conducted analyses of the appropriateness of the split-population hazard model to determine whether it outperforms a standard hazard model in our study setting.5 We compared the split-population hazard model with a model that excludes the logit part (i.e., single-population model). According to the likelihood ratio test, the split-population model significantly improves the fit (2 = 139.18; b .01) and the Akaike information criterion (AIC) of the split-population model is smaller than the AIC of the single population model (3259.41 versus 3293.69), which indicates a better fit. Although the Bayesian information criterion (BIC) suggests a better fit of the standard hazard model (3454.51 versus 3440.00), the BIC penalizes complex models more heavily than the AIC does. This finding is consistent with the STATA program as well in that it is difficult for complex split-population models to deliver a better fit (Prins and Verhoef 2007). On the basis of these arguments, we conclude that the proposed split-population hazard model is appropriate in our study.6 To gain insights into the effects of price and advertising on the conditional probability of defection (hazard part), we compare a set of nested models. First, we estimate a base model that includes the main effects of the explanatory variables (main effects model): the incumbent's and competitors' marketing efforts, relationship variables and demographics. Second, we add a set of interaction terms between the marketing variables and the time since market liberalization to test for the dynamic effect of price and advertising on the customer decision to terminate the relationship (model with dynamic interactions). In the hazard part, a positive parameter coefficient indicates a positive effect of the explanatory variable on the hazard rate (the probability that customer i terminates the relationship in period t given that this event has not occurred before time t); in the logit part, a positive coefficient indicates a positive effect of the variable on the probability that an individual eventually defects. Main Effects Model The estimation results of the main effects model in Table 2 include the estimated coefficients along with the p-values and standard errors for the hazard part and the logit part of the splitpopulation hazard model. In the hazard part of the model, the results reveal that the incumbent's ARpU and advertising
5 Following Franses and Paap (2001), we conducted an analysis of the error terms to check the empirical adequacy of the proposed model. The analysis of the residuals indicates the proposed model provides an adequate fit to our data. 6 Furthermore, the estimation results of the single population hazard model are similar to those of the split-population hazard model.

In this equation, is the parameter vector that measures the impact of the explanatory factors (see Table 1) on the probability of eventual defection. Following Prins and Verhoef (2007), we derive the log-likelihood for the sample of N-independent observations as follows: L = Ci ln pi hi t Si t 1 + 1Ci ln1pi + pi Si t :
i=1 N

4 where N, the number of customers in our sample, equals 650; Ci is a censoring indicator (equals 1 if the event is observed and 0 if it is censored); pi is the probability of eventual defection (Equation 3); hi(t) is the hazard rate (Equation 2); and Si(t) is the survival function. The contribution of consumer i to the likelihood function is as follows. If the observation is completed (Ci = 1, first part of the expression), the contribution equals the probability of eventual defection (pi), multiplied by the conditional probability of terminating the relationship at time t, hi(t), and by the probability that the customer has not ended the relationship before t, Si(t 1). If the observation is not completed (Ci = 0, second part of the expression), the contribution to the likelihood is null for those who will never defect (pi = 0), and equals the probability of eventual defection, pi, multiplied by the survivor function Si(t), for those who will eventually defect. We use STATA 10.0 to estimate the split-hazard model.

208

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214

significantly influence the conditional probability of defection. Consistent with H1A, ARpU has a positive, significant effect on the hazard rate (1 = .190; b .01); it substantially shortens the customerfirm relationship. Advertising has the opposite effect.7 Both the service and brand advertising of the incumbent firm have a negative effect on the hazard rate. They substantially lengthen the time to relationship termination (2 = .369; b .01; 3 = .080; b .1), as H2A predicts, though we only find weak support for the effect of brand advertising. ARpU has a negative and significant effect on the hazard rate for both competitors (Vodafone and Orange), which indicates that a high (low) ARpU of new entrants reduces (increases) the hazard rate and lengthens (shortens) the customer-incumbent firm relationship (4 = .149; 7 = .100; b .01). Therefore, we find support for H1B.8 With regard to mass advertising, only one of the four competitive advertising coefficients is significant: Vodafone brand advertising has a negative and significant effect on the hazard rate (6 = .161; b .01) and lengthens the customer-incumbent firm relationship (as we discuss in the next subsection). We do not find support for H2B.9 The results related to the set of relationship characteristics and demographics indicate that spending is a significant predictor in the hazard part. Consistent with prior research (Reinartz and Kumar 2003), we introduce a squared spending term to test for the nonlinear effects of spending. We find a nonlinear, inverse U-shaped relationship between mobile phone spending and the hazard rate (10 = .001; 11 = .000; b .10), which implies that intermediate spending levels increase the conditional probability of terminating the relationship. However, we only find weak support for this effect. The type of contract (prepaid versus postpaid) is a positive and significant predictor of the hazard rate (13 = .249; b .01). Compared with prepaid customers, postpaid customers exhibit a higher probability of ending the relationship with the incumbent. None of the demographic variables included in the model has a significant effect on the hazard rate. In addition, the shape of the
We also estimated a model in which we included total advertising expenditures (service + brand), but it did not perform as well as our proposed model with separate service and brand advertising terms (2 = 18.8; b .01). The mass advertising parameters for the three companies were also negative (Movistar's and Orange's were statistically significant). Therefore, the distinction between service and brand advertising provides more insight into the impact of mass advertising on retention. 8 Noting the limitations associated with the use of ARpU to measure price in mobile communications (i.e., it combines price and usage), we perform additional analyses to address this issue in more depth. Specifically, we add an interaction term ARpU*spending to the model to evaluate whether the effect of ARpU differs among customers with different spending levels. The results of the analyses reveal that the effect of ARpU does not significantly differ among customers with different spending levels. The positive effect of Movistar's ARpU on the hazard rate decreases for customers with high mobile phone spending, but it is weakly significant (p b .1). With regard to the effect of competitors' prices, the negative effect of ARpU on the hazard rate for Vodafone and Orange is less pronounced for customers with higher spending levels. These results are not significant from a statistical point of view (p N .1) and are available on request. We thank an anonymous reviewer for this suggestion. 9 In our model, similar to Prins and Verhoef (2007), we assess the potential long-term effect of advertising, but the results do not support any cumulative advertising effect.
7

hazard monotonically increases with duration time (17 = .452; b .01), so the probability of relationship termination increases with the length of the customerfirm relationship.10 The logit part of the model provides some insight into the impact of customer relationship characteristics and demographics on the probability of eventual defection. Mobile phone spending and the probability of eventual defection are weakly related in a nonlinear, inverse U-shaped manner (1 = .055; b .1; 2 = .001; b .05). Customers with intermediate spending levels exhibit the highest probability of eventually becoming defectors. Cross-buying additional services from the incumbent exerts a negative effect on the probability that the customer eventually defects (3 = 1.975; b .05) whereas the type of contract has a positive and significant influence on the eventual defection probability (4 = 2.405; b .05). Model with Dynamic Effects In addition to examining the main effects of marketing practices on the hazard rate, we assess the interrelationships between marketing variables and the time since market liberalization and thereby test our conjectures about the dynamic effects of pricing and advertising on the conditional probability of relationship termination. We compute the time since market liberalization as the number of months from a fixed point in time (liberalization date). To examine the dynamic effect of price and advertising, we first estimate a model that includes the interactions between ARpU and time since liberalization (dynamic price model), and then another model with the dynamic effect of advertising alone (dynamic advertising model). Finally, we simultaneously introduce both dynamic interaction groups and examine their effects on the hazard rate (full dynamic model) (see Table 3 for all results (hazard part)). Compared with the main effects model, adding interaction terms significantly improves the model fit for the three dynamic models (2DPM = 33.62; 2DAM = 50.42; 2FDM = 55.02; b .01). This suggests that the effect of the marketing variables on the hazard rate changes over time. An examination of the model fit measures (bottom of Table 3) indicates that the dynamic price model performs better than its two counterparts. In our data set, only two of the six advertising interaction terms with time since liberalization are statistically significant in the full dynamic model (one in the dynamic advertising model), which offers rather weak support for the dynamic effect of advertising. In the dynamic price model, adding interaction terms with time since liberalization improves the model fit compared with the main effects model (AIC = 3140.88 versus 3259.41; BIC = 3368.48 versus 3454.51). This finding suggests that the effect of ARpU changes over time. By adding the interaction terms, we find that the main effects of the ARpU variable are no
10 We performed another analysis to examine the potential for segments and capture additional heterogeneity in our data by introducing the interaction term spending*type of contract. The results reveal a negative interaction term ( = 0.0032). Compared to that for prepaid customers, the association between spending and the hazard rate for postpaid users is less pronounced. However, this association is not significant (p N .1). We thank an anonymous reviewer for this suggestion.

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214 Table 2 Split-population hazard model estimation results: main effects model. Variable Label Hazard part Parameter Coefficient (p-values) Standard error Incumbent marketing instruments Movistar's ARpU ARpUMt 1 Movistar's service advertising Movistar's brand advertising SaMt 1 BaMt 1 1 2 3 0.1903 (0.001) 0.3694 (0.014) 0.0803 (0.093) 0.1497 (0.002) 0.0729 (0.386) 0.1612 (0.000) 0.1001 (0.012) 0.1158 (0.308) 0.0057 (0.806) 0.0002 (0.076) 0.0001 (0.054) 0.0127 (0.322) 0.2495 (0.042) 0.0146 (0.971) 0.0005 (0.961) 0.2211 (0.528) 0.4524 (0.000) 0.9837 (0.000) 0.0584 0.1538 0.0479 Logit part

209

Parameter Coefficient (p-value) Standard error

Competitors marketing instruments Vodafone's ARpU ARpUVt 1 Vodafone's service advertising SaVt 1 Vodafone's brand advertising Orange's ARpU Orange's service advertising Orange's brand advertising BaVt 1 ARpUOt 1 SaOt 1 BaOt1

4 5 6 7 8 9

0.0495 0.0841 0.0416 0.0416 0.1135 0.0231

Relationship characteristics and demographics Spending Spit 10 Spending2 Cross-buying Type of contract Gender Age Employment status Shape Constant Model fit Log-likelihood AIC BIC Wald chi2 Sp2it Cbit Tcit Gei Agit Esit 11 12 13 14 15 16 17 0

0.0691 0.0001 0.7875 0.1306 0.3969 0.0101 0.3507 0.0545 0.2399 1559.25 3259.41 3454.51 49.55

1 2 3 4 5 6 7

0.0546 (0.091) 0.0011 (0.046) 1.9753 (0.045) 2.4047 (0.034) 0.0028 (0.940) 0.0099 (0.561) 0.2090 (0.580) 0.0134 (0.762)

0.0339 0.0005 0.9355 1.2259 0.5613 0.0149 0.5632 1.3420

b .1. b .05. b .01.

longer significant.11 We do, however, find some significant interaction effects in the expected directions. They suggest that ARpU has no real effect in the early stages of liberalization but begins to influence the hazard rate as market liberalization progresses. In support of H1C, we also find that the effect of the incumbent's ARpU on the conditional probability of relationship termination strengthens over time as the market evolves
11 The main effects of advertising on the hazard rate do not change substantially, with the exception of Vodafone's brand advertising parameter, which becomes insignificant.

after liberalization (1 = .012; b .1). Therefore, the positive effect of ARpU on the hazard rate is higher in later phases of market liberalization. We find a similar pattern for the effect of competitors' ARpU: Vodafone's ARpU exerts a stronger effect on the hazard rate as market liberalization progresses (4 = .009; b .1). For Orange, we also find a negative coefficient, though it is not significant (7 = .001; N .1), in mixed support of H1D. The significant interaction term of Vodafone's ARpU with time since liberalization indicates that the effect of ARpU on the hazard rate is lower in the earlier phases of the market liberalization than in the later ones. The coefficients in the full

210

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214

Table 3 Split-population hazard model with dynamic interactions (hazard part). Variable Label Parameter Dynamic price model Coefficient (p-values) Incumbent marketing instruments Movistar's ARpU Movistar's service advertising Movistar's brand advertising Dynamic interactions : time since liberalization Movistar's ARpU Movistar's service adv Movistar's brand adv ARpUMt 1 SaMt 1 BaMt 1 1 2 3 0.0209 (0.781) 0.2217 (0.062) 0.0541 (0.096) 0.0122 (0.056) Dynamic advertising model Full dynamic model Standard error 0.1167 0.3139 0.1489

Standard Coefficient error (p-values) 0.0755 0.1190 0.0378 0.0999 (0.027) 0.0009 (0.997) 0.0690 (0.614) 0.0030 (0.661) 0.0019 (0.706) 0.0997 (0.013) 0.2046 (0.191) 0.0644 (0.442) 0.0254 (0.476) 0.0293 (0.883) 0.0307 (0.457) 0.0098 (0.099) 0.0052 (0.136) 0.0036 (0.624) 0.0014 (0.644) 0.0310 (0.026) 0.0003 (0.032) 0.0782 (0.767) 0.2278 (0.472) 0.2734 (0.345) 0.0021 (0.797) 0.2378 (0.286) 0.0071 (0.448) 0.3631 (0.000)

Standard Coefficient error (p-values) 0.0453 0.2447 0.1369 0.0692 (0.553) 0.2406 (0.443) 0.0941 (0.527) 0.0044 (0.004) 0.0031 (0.742) 0.0024 (0.661)

ARpUMt 1x TsL 1 SaMt 1x TsL BaMt 1x TsL 2 3

0.0063

0.0069 0.0049

0.0014 0.0093 0.0054

Competitors marketing instruments Vodafone's ARpU Vodafone's service advertising Vodafone's brand advertising Orange's ARpU Orange's service advertising Orange's brand advertising

ARpUVt 1 SaVt 1 BaVt 1 ARpUOt 1 SaOt 1 BaOt 1

4 5 6 7 8 9 4 5 6 7 8 9

0.0780 (0.524) 0.0832 (0.280) 0.0522 (0.131) 0.0051 (0.968) 0.0985 (0.240) 0.0182 (0.328) 0.0094 (0.091) 0.0011 (0.791)

0.1226 0.0769 0.0345 0.1272 0.0839 0.0186

0.0402 0.15656 0.0838 0.0356 0.1986 0.0413

0.1197 (0.562) 0.2740 (0.110) 0.0686 (0.486) 0.0352 (0.807) 0.0676 (0.739) 0.0322 (0.459) 0.0070 (0.006) 0.0151 (0.037) 0.0061 (0.039) 0.0002 (0.766) 0.0046 (0.542) 0.0011 (0.730) 0.0314 (0.038) 0.0003 (0.042) 0.0590 (0.831) 0.3765 (0.001) 0.2028 (0.505) 0.0027 (0.752) 0.2528 (0.301) 0.0081 (0.667) 0.3644 (0.000)

0.2064 0.1715 0.0984 0.1438 0.2031 0.4342

Dynamic interactions: time since liberalization Vodafone's ARpU ARpUVt1 x TsL Vodafone's service adv Vodafone's brand adv Orange's ARpU Orange's service adv Orange's brand adv SaV t 1 x TsL BaVt 1x TsL ARpUOt 1x TsL SaOt 1x TsL BaOt 1x TsL

0.0063 0.0041

0.0059 0.0034 0.0074 0.0031

0.0022 0.0072 0.0029 0.0037 0.0075 0.0033

Relationship characteristics and demographics Spending Spending2 Cross-buying Type of contract Gender Age Employment status Time since liberalization Shape

Spit Sp2it Cbit Tcit Gei Agit Esit TsL

10 11 12 13 14 15 16 10 17

0.0234 (0.097) 0.0003 (0.062) 0.0805 (0.683) 0.3929 (0.000) 0.0742 (0.730) 0.0092 (0.116) 0.2209 (0.195) 0.0219 (0.729) 0.4441 (0.000)

0.0141 0.0001 0.1972 0.1077 0.2151 0.0059 0.1703 0.0428 0.0283

0.0139 0.0001 0.2642 0.3164 0.2896 0.0081 0.2228 0.0035 0.0324

0.0152 0.0002 0.2757 0.1122 0.3042 0.0085 0.2442 0.0044 0.0371

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214 Table 3 (continued) Variable Label Parameter Dynamic price model Coefficient (p-values) Constant Model fit Log-likelihood AIC BIC Wald chi2 b .1. b .05. b .01. 0 Dynamic advertising model

211

Full dynamic model Standard error 6.8951

Standard Coefficient error (p-values) 7.0157 (0.000) 1534.04 3193.01 3428.74 91.13

Standard Coefficient error (p-values) 1.9815 5.8239 (0.000) 1531.74 3133.48 3426.05 91.92

13.6412 6.8951 (0.000) 1542.44 3140.88 3368.48 95.57

model remain stable and become more significant. Together, these results suggest that the effect of ARpU on the hazard rate changes with the time since market liberalization and customers become more price sensitive over time. In our data, we do not find support for the hypotheses about the dynamic effect of the incumbent's and competitors' mass advertising (H2C and H2D). However, the finding pertaining to Vodafone's brand advertising deserves special attention. In the main effects model, we find a significant negative association with the hazard rate whereas, in the full dynamic model, the main impact of Vodafone's brand advertising on the conditional probability of ending the relationship becomes positive but insignificant (6 =.069; N .1), with a negative, significant interaction term with time since liberalization (6 = .006; b .05). The positive impact of Vodafone's brand advertising on the hazard rate thus decreases progressively, eventually becoming negative in the latest stages of liberalization. At the end of 2001, Vodafone took over the Spanish mobile operator Airtel as part of its global expansion strategy, which suddenly confronted consumers with a new and unfamiliar brand. We suspect that unfamiliarity and lack of knowledge about the new brand might have caused some negative consumer reactions to brand advertising by Vodafone, which might partially explain this counterintuitive result. Collectively, we find strong support for the dynamic effect of price but no support for the dynamic effect of advertising. We summarize our hypothesis testing results in Table 4. Discussion and Implications We study customer retention in a liberalizing market and focus on the extent to which customers' decisions to terminate the relationship with the incumbent firm and switch to a new
Table 4 Hypothesis testing results. Marketing variable Price Effect Incumbent's main effect Competitors' main effect Incumbent's dynamic effect Competitors' dynamic effect Incumbent's main effect Competitors' main effect Incumbent's dynamic effect Competitors' dynamic effect Hypothesis H1A H1B H1C H1D H2A H2B H2C H2D

entrant depend on both the incumbent's and the new entrants' price and mass advertising. By combining several innovative research elements (actual customer behavior rather than intentions to repurchase, data that combine longitudinal information about pricing and mass marketing communication with data about customer purchase behavior, a study context only recently opened to competition and a split-population hazard model that accounts for the notion that some customers have a zero probability of defecting), we contribute to existing marketing literature in several critical ways. First, we contribute to retention literature by examining the effect of two frequently neglected marketing instruments, price and mass advertising (see Blattberg, Malthouse, and Neslin 2009). We find that our price measure, ARpU, is a significant variable not only at the beginning of the relationship, during the acquisition process, but also during the development of the relationship (Bolton, Lemon, and Verhoef 2004). In addition, by considering the impact of mass advertising, we provide new insights into the drivers of customer retention and extend Bolton's (1998) findings to a mobile communications context. Above-the-line (service and brand) advertising by the incumbent determines the timing to relationship termination and leads to longer relationships. Service advertising appears to have a greater impact so communicating about the service can help the incumbent firm retain its customers in this setting. Second, unlike most CRM literature, our study explicitly includes the impact of competitive marketing instruments on existing customers' behavior. Previous research establishes how competitive communications can predict customer behavior in the context of a new service adoption (Prins and Verhoef 2007). We show that customer retention is not so significantly affected by competitors' advertising campaigns. This interesting finding

Hypothesized effect on customer retention + Stronger over time Stronger over time + Weaker over time Weaker over time

Hypothesis testing Supported Supported Supported Mixed support (only for Vodafone) Supported Not supported Not supported Not supported

Mass advertising

212

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214

implies that different customer behaviors are not affected in the same way by firms' general marketing actions. Regarding ARpU, incumbent customers react to reductions in competitors' ARpU by significantly increasing the probability that they will end their existing relationship with the incumbent vendor. These results support the assertion that competition is an important driver of customer behavior and must be considered in CRM research. Third, we contribute to customer management literature by investigating customer retention in a liberalizing market. Current knowledge mainly comes from studies in liberalized, well-established industries but, considering their distinctive features, generalizing these findings to liberalizing markets might be dangerous (Wieringa and Verhoef 2007). The set of explanatory variables that we include in our model (specifically, the incumbent's and competitors' ARpU and advertising) helps to explain the timing to relationship termination in this context. Furthermore, we investigate the dynamic effects of these marketing tactics, which mainly apply to ARpU and not so much to advertising. Specifically, the effect of ARpU becomes stronger in later stages of market liberalization for all three companies, which suggests that customers become more value conscious and price sensitive as the liberalization process advances. This result is consistent with various research streams. In line with first-mover advantage research (Lieberman and Montgomery 1988), incumbent customers bear high switching costs and their behavior reflects their habits and inertia. Consistent with information economics research (Kirmani and Rao 2000), aggressive price policies by new entrants suggest low quality, which discourages switching behavior. However, as market liberalization progresses, customers gain knowledge about and confidence in competing suppliers, so they base their behavior on more rational aspects, making them more value conscious and price sensitive. Fourth and finally, we significantly advance knowledge about the drivers of customer retention with our use of advanced modeling techniques. The split-population hazard model accounts for the notion that a portion of the population will never experience the event of interest (defection). Consistent with Wieringa and Verhoef (2007), we confirm that inertia is a key determinant of relationship continuance for a significant group of consumers, some of whom will never defect and are, in that sense, inherently loyal. The proposed model also enables us to characterize this customer segment according to customerspecific variables, which might help both incumbents and new entrants to allocate their marketing resources optimally by targeting consumers with a non-zero probability of switching. Management and Public Policy Implications In recent years, various European governments have opened their utility goods markets to competition (Roberts, Nelson, and Morrison 2005). Deregulation prompts intense competition among privatized incumbents and newcomers. Understanding why existing incumbent consumers might want to end their current relationship and switch to new entrants is key for the incumbent to secure its dominant position and for new entrants

to build an initial customer base. This knowledge is also central for public policy officials, who must create a competitive regulatory framework. We suggest some insights that may help firms and public policy officials determine strategies and resource allocations. When market liberalization occurs, defensive and offensive marketing strategies come into conflict. Incumbents can use price and advertising to preserve their existing customer relationships and their advertising strategies should communicate about both the service and the brand to protect their current dominant status, though the effect of service advertising is stronger. Incumbents must also adapt their marketing strategies as market liberalization continues. Our results indicate that their customers become more price sensitive over time and incumbent price policies should adapt accordingly. In contrast, new entrants should rely heavily on aggressive price policies from the start to capture customers from the incumbent. Although conventional wisdom suggests that advertising is essential to acquire consumers, our results imply that it has a limited role in capturing incumbent customers; it may be more important for acquiring consumers who are new to the market. New entrants' strategies should also be subject to updating over the course of the market liberalization process to reflect customers' greater price sensitivity. With regard to resource allocations, our finding of a group of incumbent customers with zero probability of ending the relationship offers some important guidelines. These customers are irresponsive to any (incumbent or new entrant) marketing initiative. They also exhibit certain relationship characteristics and demographic variables that firms may use to identify them to avoid spending valuable marketing resources on attracting them. Finally, the public policy implications stem mainly from this same group of customers who will never defect from the incumbent. Regulatory goals might aim to give all consumers the option to switch providers, but public policy officials should be aware that market deregulation per se does not produce this outcome. Additional programs might be considered to inform this inherently loyal consumer group of the possibility and potential advantages of switching to alternative suppliers. In addition, our results indicate that, after market deregulation, the dynamic adaptations in consumer behavior lead to increased price sensitivity over time. Public policy officials who want more immediate outcomes will need to increase customer knowledge about and confidence in the new suppliers, perhaps through special programs prior to liberalization that inform consumers about the impending market situation. Study Limitations and Further Research Some limitations of the current study may provide a basis for further investigations. First, for theory testing purposes, we investigate the drivers of customer retention in the specific context of the mobile phone industry. This industry is increasingly important in developed economies and constitutes a good example of a market configuration recently opened to competition. Nevertheless, our results cannot be generalized to other contexts without caution. For example, among new entrants in

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214

213

this market, service advertising is not a significant predictor of the timing to relationship termination for incumbent customers, a result that might not hold in other contexts. Additional research, therefore, might examine the relative effect of price and advertising in other industries. Second, we used longitudinal behavioral data to test the proposed hypotheses. Although our research meets two criteria required to establish causal relationships (concomitant variation and time order of occurrence of variables), we cannot rule out alternative explanations. To assess causal relationships fully, experimental studies would be required. We also considered the impact of advertising spending levels, which could follow specific patterns and differ significantly over time. We ignored these patterns, so further research should address the effects of specific patterns and changes in advertising on retention. Third, we measure price as the average revenue per user (ARpU). Although frequently used by firms, public policy officials and researchers in a mobile communications context, largely because of the difficulties associated with obtaining customer-level price measures (Shy 2002), the ARpU measure suffers from some limitations that recommend the use of caution in generalizing our research findings to other contexts. Further research could develop a customer-level price measure that facilitates the analysis of pricing effects in mobile communications. Fourth, price and advertising are two key determinants of customer retention, but they are only a subset of the potential factors affecting relationship continuance. We do not consider, for example, the role of direct marketing communications. Fifth, and finally, it would be interesting to test for the effects of brand strength in the context of a liberalizing market. Brand strength likely plays a key role in affecting customer purchase decisions and may also exert a moderating influence on the impact of marketing activities (e.g. advertising) on retention. References
Assmus, Gert, John U. Farley, and Donald Lehmann (1984), How Advertising Affects Sales: Meta Analysis of Econometric Results, Journal of Marketing Research, 21, February, 6574. Berger, Paul D. and Nada I. Nasr (1998), Customer Lifetime Value: Marketing Models and Applications, Journal of Interactive Marketing, 12, 1, 1730. Bijmolt, Tammo H.A., Harald J. van Heerde, and Rik G.M. Pieters (2005), New Empirical Generalizations on the Determinants of Price Elasticity, Journal of Marketing Research, 42, May, 14156. Blattberg, Robert C., Edward C. Malthouse, and Scott A. Neslin (2009), Customer Lifetime Value: Empirical Generalizations and Some Conceptual Questions, Journal of Interactive Marketing, 23, 2, 15768. Bolton, Ruth N. (1998), A Dynamic Model of the Duration of the Customer's Relationship with a Continuous Service Provider: The Role of Satisfaction, Marketing Science, 17, 1, 4565. and Katherine N. Lemon (1999), A Dynamic Model of Customers' Usage of Services: Usage as an Antecedent and Consequence of Satisfaction, Journal of Marketing Research, 36, May, 17186. , Katherine N. Lemon, and Peter C. Verhoef (2004), The Theoretical Underpinnings of Customer Asset Management: A Framework and Propositions for Future Research, Journal of the Academy of Marketing Science, 32, Summer, 27192. , Katherine N. Lemon, and Matthew D. Bramlett (2006), The Effect of Service Experiences over Time on a Supplier's Retention of Business Customers, Management Science, 52, 12, 181123. Buehler, Stefan, Ralf Dewenter, and Justus Haucap (2006), Mobile Number Portability in Europe, Telecommunications Policy, 30, 7, 38599.

Carpenter, Gregory S. and Kent Nakamoto (1989), Consumer Preference Formation and Pioneering Advantage, Journal of Marketing Research, 26, August, 28598. Chandy, Rajesh K., Gerard J. Tellis, Deborah J. MacInnis, and Pattana Thaivanich (2001), What to Say When: Advertising Appeals in Evolving Markets, Journal of Marketing Research, 38, November, 399414. Chen, Yongmin (1997), Paying Customers to Switch, Journal of Economics & Management Strategy, 6, 4, 87797. Chintagunta, Pradeep K. and Xiaojing Dong (2006), Hazard/Survival Models in Marketing, The Handbook of Marketing Research. Newbury Park, CA: Sage Publications, 441 454. CMT [Comisin del Mercado de las Telecomunicaciones] (2007), Annual Report 2007, (available at: http://www.cmt.es/es/publicaciones/anexos/ Informe_anual_CMT_2007_web.pdf ). Dawes, John (2009), The Effect of Service Price Increases on Customer Retention: The Moderating Role of Customer Tenure and Relationship Breadth, Journal of Service Research, 11, 3, 23245. De Wulf, Kristof, Gaby Odekerken-Schrder, and Dawn Iacobucci (2001), Investments in Consumer Relationships: A Cross-Country and CrossIndustry Exploration, Journal of Marketing, 65, October, 3350. Dover, Howard F. and B.P.S. Murthi (2006), Asymmetric Effects of Dynamic Usage Behavior on Duration in Subscription-Based Online Service, Journal of Interactive Marketing, 20, 34, 515. Drze, Xavier and Andr Bonfrer (2008), An Empirical Investigation of the Impact of Communication Timing on Customer Equity, Journal of Interactive Marketing, 22, 1, 3650. Franses, Philip Hans and Richard Paap (2001), Quantitative Models in Marketing Research. Cambridge: Cambridge University Press. Global Wireless Matrix Database (2008). Merrill Lynch. Green, Donna H., Donald W. Barclay, and Adrian B. Ryans (1995), Entry Strategy and Long-Term Performance: Conceptualization and Empirical Examination, Journal of Marketing, 59, October, 116. Guadagni, Peter M. and John D.C. Little (1983), A Logit Model of Brand Choice Calibrated on Scanner Data, Marketing Science, 2, 3, 20338. Gupta, Sunil, Donald R. Lehmann, and Jennifer Ames Stuart (2004), Valuing Customer, Journal of Marketing Research, 41, February, 718. and Valarie Zeithaml (2006), Customer Metrics and Their Impact on Financial Performance, Marketing Science, 25, 6, 71839. Hanssens, Dominique M., Leonard J. Parsons, and Randall L. Schultz (2001), Market Response Models: Econometric and Time Series Analysis. 2nd ed. Boston: Kluwer Academic Publishers. Hauser, John R. and Steven M. Shugan (1983), Defensive Marketing Strategy, Marketing Science, 2, 4, 31960. Helsen, Kristiaan and David C. Schmittlein (1993), Analyzing Duration Times in Marketing: Evidence for the Effectiveness of Hazard Rate Models, Marketing Science, 11, 4, 395414. IDATE (2009), Mobile Churn Management: The Colour of Money, Mobile Industry Report. Keaveney, Susan M. (1995), Customer Switching Behavior in Service Industries: An Exploratory Study, Journal of Marketing, 59, April, 7182. Kirmani, Amna and Akshay R. Rao (2000), No Pain, No Gain: A Critical Review of the Literature on Signaling Unobservable Product Quality, Journal of Marketing, 64, April, 6679. Klemperer, Paul (1995), Competition when Consumers Have Switching Costs: An Overview with Applications to Industrial Organization, Macroeconomics, and International Trade, Review of Economic Studies, 62, 51539. Leeflang, Peter S.H. and W. Fred van Raaij (1995), The Changing Consumer in the European Union: A Meta-analysis, International Journal of Research in Marketing, 12, 5, 37387. Lemmens, Aurlie and Christophe Croux (2006), Bagging and Boosting Classification Trees to Predict Churn, Journal of Marketing Research, 43, May, 27686. Lewis, Michael (2006), Customer Acquisition Promotions and Customer Asset Value, Journal of Marketing Research, 43, May, 195203. Lieberman, Marvin B. and David B. Montgomery (1988), First-Mover Advantages, Strategic Management Journal, 9, Summer, 4158.

214

Y. Polo et al. / Journal of Interactive Marketing 25 (2011) 201214 , Katherine N. Lemon, and Valarie A. Zeithaml (2004), Return on Marketing: Using Customer Equity to Focus Marketing Strategy, Journal of Marketing, 68, January, 10927. Schmidt, Peter and Ann Dryden Witte (1989), Predicting Criminal Recidivism Using Split Population Survival Time Models, Journal of Econometrics, 40, January, 14159. Schweidel, David A., Peter S. Fader, and Eric T. Bradlow (2008), A Bivariate Timing Model of Customer Acquisition and Retention, Marketing Science, 27, 5, 82943. Seetharaman, P.B., Andrew Ainslie, and Pradeep K. Chintagunta (1999), Investigating Household State Dependence Effects across Categories, Journal of Marketing Research, 36, November, 488500. Shum, Matthew (2004), Does Advertising Overcome Brand Loyalty? Evidence from the Breakfast-Cereals Market, Journal of Economics and Management Strategy, 13, 2, 24172. Shy, Oz (2002), A Quick-and-Easy Method for Estimating Switching Costs, International Journal of Industrial Organization, 20, 1, 7187. Sinha, Rajiv K. and Murali Chandrashekaran (1992), A Split Hazard Model for Analyzing the Diffusion of Innovations, Journal of Marketing Research, 29, February, 11627. Srinivasan, Raji, Gary L. Lilien, and Arvind Rangaswamy (2006), The Emergence of Dominant Designs, Journal of Marketing, 70, April, 117. Tellis, Gerard J. (1988), Advertising Exposure. Loyalty and Brand Purchase: A Two-Stage Model, Journal of Marketing Research, 25, May, 13444. Vakratsas, Demetrios and Tim Ambler (1999), How Advertising Works: What Do We Really Know?, Journal of Marketing, 63, January, 2643. Venkatesan, Rajkumar and V. Kumar (2004), A Customer Lifetime Value Based Framework for Customer Selection and Resource Allocation Strategy, Journal of Marketing, 68, October, 10625. Venkatesan, Rajkumar, Werner J. Reinartz, and Nalini Ravishanker (2009), The Role of Customer Attitudes in CRM Activities, Working paper. Verhoef, Peter C., Philip Hans Franses, and Janny C. Hoekstra (2001), The Impact of Satisfaction and Payment Equity on Cross-Buying: A Dynamic Model for a Multi-Service Provider, Journal of Retailing, 77, 3, 35978. (2003), Understanding the Effect of Customer Relationship Management Efforts on Customer Retention and Customer Share Development, Journal of Marketing, 67, October, 3045. and Bas Donkers (2005), The Effect of Acquisition Channels on Customer Retention and Cross-Buying, Journal of Interactive Marketing, 19, 2, 3143. , Jenny van Doorn, and Mathilda Dorotic (2007), Customer Value Management: An Overview and Research Agenda, MarketingJournal of Research in Management, 2, 5169. , Katherine N. Lemon, A. Parasuraman, Anne Roggeveen, Leonard L. Schlessinger, and Michael A. Tsiros (2009), Customer Experience: Determinants, Dynamics and Management Strategies, Journal of Retailing, 85, 1, 3141. Wieringa, Jaap E. and Peter C. Verhoef (2007), Understanding Customer Switching Behavior in a Liberalizing Service Market, Journal of Service Research, 10, 2, 17486. Winer, Russell S. (1986), A Reference Price Model of Brand Choice for Frequently Purchased Products, Journal of Consumer Research, 13, September, 2506.

Maicas, Juan Pablo, Yolanda Polo, and F. Javier Sese (2009), Reducing the Level of Switching Costs in Mobile Communications: The Case of Mobile Number Portability, Telecommunications Policy, 33, 9, 54454. McCloughan, Patrick and Sean Lyons (2006), Accounting for ARPU: New Evidence from International Panel Data, Telecommunications Policy, 30, 52132. Mehta, Nitin, Xinlei (Jack) Chen, and Om Narasimhan (2008), Informing, Transforming, and Persuading: Disentangling the Multiple Effects of Advertising on Brand Choice Decisions, Marketing Science, 27, 3, 33455. Milgrom, Paul and John Roberts (1986), Price and Advertising Signals of Product Quality, Journal of Political Economy, 94, 4, 796821. Morgan, Robert M. and Shelby D. Hunt (1994), The CommitmentTrust Theory of Relationship Marketing, Journal of Marketing, 58, July, 2038. Neslin, Scott A., Sunil Gupta, Wagner Kamakura, Junxiang Lu, and Charlotte H. Mason (2006), Defection Detection: Measuring and Understanding the Predictive Accuracy of Customer Churn Models, Journal of Marketing Research, 43, May, 20411. Niedrich, Ronald W. and Scott D. Swain (2003), The Influence of Pioneer Status and Experience Order on Consumer Brand Preference: A MediatedEffects Model, Journal of the Academy of Marketing Science, 31, 4, 46880. Nijs, Vincent R., Marnik G. Dekimpe, Jan-Benedict E.M. Steenkamp, and Dominique M. Hanssens (2001), The Category-Demand Effects of Price Promotions, Marketing Science, 20, 1, 122. Pfeifer, Phillip E. and Paul W. Farris (2004), The Elasticity of Customer Value to Retention: The Duration of a Customer Relationship, Journal of Interactive Marketing, 18, 2, 2031. Polo, Yolanda and F. Javier Sese (2009), How to Make Switching Costly: The Role of Marketing and Relationship Characteristics, Journal of Service Research, 12, 2, 11937. Prins, Remco and Peter C. Verhoef (2007), Marketing Communication Drivers of Adoption Timing of a New E-service among Existing Customers, Journal of Marketing, 71, April, 16983. Rao, Akshay R. and Kent B. Monroe (1988), The Moderating Effect of Prior Knowledge on Cue Utilization in Product Evaluations, Journal of Consumer Research, 15, September, 25364. Reinartz, Werner J. and V. Kumar (2003), The Impact of Customer Relationship Characteristics on Profitable Lifetime Duration, Journal of Marketing, 67, January, 7799. S.Thomas, and Ganal Bascoul (2008), Investigating Cross-Buying and Customer Loyalty, Journal of Interactive Marketing, 22, 1, 520. Risselada, Hans, Peter C. Verhoef, and Tammo H.A. Bijmolt (2010), Staying Power of Churn Prediction Models, Journal of Interactive Marketing, 24, 3, 198208. Roberts, John H., Charles J. Nelson, and Pamela D. Morrison (2005), A Prelaunch Diffusion Model for Evaluating Market Defense Strategies, Marketing Science, 24, 1, 15064. Rust, Roland T., Valarie A. Zeithaml, and Katherine N. Lemon (2000), Driving Customer Equity: How Customer Lifetime Value Is Reshaping Corporate Strategy. New York: The Free Press.